03 Applied Superconductivity

BIOLOGICAL EFFECTS OF ELECTROMAGNETIC
FIELDS WITH EMPHASIS ON HEALTH AND

SAFETY
INTRODUCTION
The biological effects of electromagnetic elds is a large
and heterogeneous subject that spills over into biophysics,
medicine, engineering, and, more recently, epidemiology,
risk assessment, tort law, and public policy. Within this
literature, the biological effects of electromagnetic elds
comes up in a variety of contexts. More than 100 epidemiology studies have searched for possible associations between exposure to electromagnetic elds at either powerline or radiofrequencies and various health effects. Hundreds of other studies have, broadly, a toxicological focus
(i.e., they were designed to identify effects of electromagnetic elds that may have some bearing on human health
effects). Other studies have examined biological endpoints
with potential signicance to therapeutic applications of
electromagnetic elds. Still other studies are basic in orientation and involve biological endpoints that have no direct relevance for health and safety. This vast literature,
which includes hundreds of reports of biological effects of
electromagnetic elds, is far too extensive to review here.
(An Appendix will present selected references, available
online, to recent reports of expert groups on the subject.)
EXPOSURE AND DOSE
In toxicology, exposure is the concentration in air, water, or
food, or it is the intensity of a physical agent (an electromagnetic eld, in this case) in the external environment of
a subject. Dose is the amount of the agent delivered to the
organ that is the target of toxicity. Appropriate measures
of exposure include the following:
External eld strength, measured in volts/m (electric

eld) or amps/m (magnetic eld). Most investigators
cite the magnetic ux density in Tesla instead. Typically, eld strengths outside of the exposed body (the
external eld Eo ) are reported.
Time dependence of the eld, including frequency (Hz)
and modulation parameters (for alternating elds) or
pulsewidth, slew rate, and other parameters (pulsed
elds). The frequency biological responses depend
on the frequency or temporal characteristics of eld
pulses, both because of their intrinsic time responses,
and because of the frequency-dependent coupling of
external elds into the body.
Incident power density (intensity), in W/m2 .
Typical measures of dose include the following:
Internal eld, Ei (i.e., the electric eld induced within

body tissues).
Induced current density in A/m2 . The current density

induced within the body is related to Ei by Ohms Law
Ji = Ei
where is the conductivity (S/m) of the tissue. The
current density within the tissue can be a better predictor of biological effects than the internal or external
eld strength.
Specic absorption rate (SAR). For many effects reported from exposure to high frequency elds, the relevant measure of exposure is the SAR, dened as the
rate of heat generation in watts per kilogram of tissue,
in terms of the electric eld Ei in the tissue,
SAR = Ei2 /
(1)
where is the tissue density (kg/m3 ).

Sources of Exposure and Typical Exposure Levels
The interactions between electromagnetic elds and the
body can be considered separately in three broad frequency
ranges: static elds (0 Hz), low frequency elds (<3 kHz),
and high frequency elds (3 kHz <f <300 GHz). The boundaries between these ranges are arbitrary. However, 3 kHz
corresponds roughly to the upper frequency for nerve excitation: above 1kHz, the thresholds for nerve excitation
increase rapidly with frequency. Below about 13 kHz, the
most conspicuous hazards to humans from electromagnetic
elds are associated with shock, and at higher frequencies
they are associated with excessive heating of body tissues.
Static Fields. Technologies producing static magnetic
elds include magnetic levitation trains, magnetic resonance imaging devices used in medicine, industrial processes using direct electric currents for materials processing in industry, and particle accelerators used in high energy physics. Static electric elds are commonplace in the
environment because of naturally occurring electrostatic
processes, and they are produced in some industrial processes. Human exposure to static magnetic elds varies
widely. The Earths magnetic eld is approximately 50 T.
Some workers are exposed for prolonged times to static
magnetic elds of tens of mT [aluminum workers (1)] to
1 T or more [physicists working on particle accelerators
(2)]. Patients undergoing MRI imaging are briey exposed
to magnetic ux densities up to about 4 T. Static electric
elds in ordinary room environments are several kV/m.
Low Frequency Fields (<3 kHz). Power distribution facilities and electrical equipment are ubiquitous sources of
powerfrequency (5060 Hz) electric and magnetic elds
in the environment. Other sources of strong low frequency
elds include some industrial induction heaters and some
specialized communications facilities.
In ordinary room environments, typical 5060-Hz elds
are about 0.10.2 T (magnetic ux density) and 10100
V/m (electric eld strength). The strongest 5060-Hz electric eld strengths commonly found in the environment
exist beneath high voltage transmission lines, where elds
can exceed 12 kV/m at ground level [for a 765-kV transmission line located 12 m above ground (3)]. Field strengths of
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright 2007 John Wiley & Sons, Inc.
Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety
250 V/m have been measured 30 cm from an electric blanket. By contrast, the strongest powerfrequency magnetic
elds found in ordinary environments are associated with
appliances. For example, the magnetic ux density near arc
welders, soldering guns, and other high current devices can
exceed 1 mT. Magnetic ux densities beneath high voltage
power lines rarely exceeds 2 T (all of these values are from
Reference 4).
High Frequency Fields (3 kHz to 300 GHz). Many technologies produce electromagnetic elds in this frequency
range, including industrial heating, radar, and communications applications.
Exposures to radiofrequency elds vary widely, depending on the proximity to transmitters. An Environmental
Protection Agency survey published in 1982 reported a median exposure level of 50 W/m2 time-averaged power density for the population of the 12 cities, with approximately
1% of the population studied, or about 380,000 persons,
potentially exposed to levels greater than 104 W/m2 (5).
Sources of potentially hazardous radiofrequency elds
are common in household and industrial environments.
Devices operating in the industrial-scientic-medical
bands at 27 MHz, 915 MHz, and 2450 MHz include microwave ovens and industrial heaters. Fields close to or
within such devices can be clearly hazardous, although the
systems are normally designed to prevent human exposure
at levels above recommended limits. Medical diathermy
units expose patients to radiofrequency elds (typically at
915 MHz or 2450 MHz) for purposes of heating.
Coupling Between External and Internal Fields
Static Fields. Body tissues are essentially nonmagnetic,
and at low frequencies the magnetic eld inside the body
is essentially the same as that immediately outside it.
However, body tissues are conductive, with high permittivity at low frequencies (Table 1) (6), which means that an
external DC electric eld will not induce an electric eld
within the body in the absence of body contact to a conductor. The external eld, however, will induce charges on the
outer surface of the body, which can result in strong electric
elds in the air near body edges (e.g., at the tips of hairs).
Low Frequency Fields. At low frequencies, the internal
and external magnetic elds are virtually identical, because of the nonmagnetic nature of tissues.
However, alternating electric and magnetic elds will
both induce electric elds within the body, which from a
perspective of health and safety can be signicant. These
induced elds depend on numerous parameters including
the frequency of exposure, geometry of the body, and so on.
The magnetically induced electric elds are determined
by Faradays Law
B/t = E
(2)
(i.e., the induced electric eld is proportional to the time

rate of change of the magnetic eld).
For a circular loop of tissue with radius R and an alternating magnetic eld with (RMS) ux density B0 , the
magnetically induced electric eld E is given by

E = f B0 R
(3)
where f is the frequency (Hz).

Standards-setting organizations have used ellipsoids of
revolution to model the body. If the ellipsoid is oriented
with its longest semi-major axis along the z direction,
which coincides with the orientation of the magnetic eld,
the induced current density J (RMS) is given by
2 f B0 / 2 4 2
1/2
J=
(b x + a4 y2 )
(4)
a2 + b 2
where a and b are the two shorter semi-axes, which are assumed to be directed along the x and y axes (7). Thus, the
magnetically induced electric eld in this model is proportional to the distance from its center.
The electrically induced electric elds within the body
depend on the shape of the body and whether it is
grounded. For an ungrounded spherical object in air whose
dielectric properties are typical of those of tissue, the electrically induced electric eld E can be found by solution of
Laplaces equation to be (8)
E=
3 f Eo o
(5)
where o is the permittivity of space (8.85 1012 F/m) (all

elds are RMS). For typical tissues, 0.2 S/m at 60 Hz,
and the electrically induced electric eld is about 8 orders
of magnitude weaker than Eo . The total internal electric
eld Ei is the sum of the electrically and magnetically induced electric elds.
These calculations can be extended to more realistically
shaped models of the body and to grounded objects. Figure
1 shows the induced elds in a grounded person standing
erect in a vertical 60-Hz electric eld. In these objects, the
induced elds are greater than simple estimates based on
a spherical model by about two orders of magnitude, a consequence of the shape of the body. The current density in
the ankles is 10 times larger than in the trunk, because of
the smaller cross-sectional area of the trunk and leg compared with the torso (9). Contact with ground results in an
approximate doubling of the electrically induced electric
elds in the body.
To summarize, inside the body of a person who is not in
contact with the ground and exposed to an electric eld in
the air, at low frequencies:
1. The internal and external magnetic elds are essentially the same.
2. The electrically induced electric eld in the body is proportional to the frequency and inversely proportional to
the conductivity of tissue. At 50/60 Hz, it is about eight
orders of magnitude smaller than Eo . These elds vary
greatly throughout the body because of local variations
in the electrical properties of tissue.
3. The magnetically induced electric eld in the body is
proportional to the frequency and to the size of the body.
These electric elds are largest near the periphery of the
body, and are nominally zero at its center.
Table 1. Permittivity (Relative to Vacuum) and Conductivity (S/m) of Tissues at 37 C (Adapted from Reference 6)a
Tissue
10 kHz
bone
fat
blood
muscle (perpendicular to bers)b
muscle (parallel to bers)
640
30000
2800
70000
80000
1 MHz
0.01
0.02.07
0.7
0.085
0.55
87
2040
12105
100 MHz
0.02
.02.07
0.7
0.60.9
23
4.57
67
6770
10 GHz
0.06
02.07
1.0
0.91.0
50
47
45
4042
0.51.7
0.30.4
910
89
a
Considerable variability exists in the dielectric properties of tissue. Listed values are either typical values or ranges of
reported data.
b
No data exist for oriented muscle at high frequencies. However, anisotropy is less pronounced at high frequencies.
Table 2. Induced Current Density Associated with Biological Effects of Alternating Magnetic Fields
Induced Current Density (mA/m2 )
1000
Order
of
magnitude
ux density
(50/60
Hz)
needed to induce current
density
in
human body
>
100 mT
1001000
10100 mT
10100
110 mT
110
011 mT
<1
<
0.1 mT
Effects (50/60 Hz)
stimulation of motor cortex in brain and

peripheral nerves; possibility of ventricular brillation denite hazards
stimulation of excitable tissue has
been reported, other effects reported
(changes in central nervous system excitability, reversal of visual evoked responses).
magnetophosphenes, possible nervous
system effects; reported enhancement
of bone healing.
endogenous current density; minor biological effects reported
absence of established effects
Adapted from References 15 and 16.

Table 3. Thresholds for Perception of Microwave Energy (50)
Frequency, GHz
2.45
10
35
94
Approximate energy penetration depth, m

.02
.004
7 104
2 104
Threshold for perception (10 s exposure, 0.024 m2 exposure area),W/m2

660
200
90
50
Table 4. Exposure Standards for 60-Hz Electric and Magnetic Fields

Occupational
Electric eld, kV/m
Magnetic ux density, mT
Contact current, mA
General public
Electric eld, kV/m
Magnetic ux density, mT
Contact current, mA
IEEE 2002
ICNIRP 1998
ACGIH 2000
20
2.7
1.5
8.3
0.4
1.0
25
1.0
1.0
5
0.9
0.5
4.2
0.08
0.5
n/a
n/a
n/a
Figure 2. Whole-body average SAR in a prolate spheroidal model

of man in a plane wave, as a function of frequency of the incident
wave. The major semi-axis is 0.875 m, minor semi-axis 0.138 m,
and volume 0.07 m3 . The SAR is given in W/kg, normalized to
an incident power density of 10 W/m2 (from Reference 10 with
permission). Three curves are shown, for E (electric eld vertical),
H (magnetic eld vertical), and K (wave propagating from the top)
polarizations.
Figure 1. Fields induced in a grounded man standing in a kV/m60 Hz electric eld. The current entering the body (from capacitive
coupling with the eld) and current densities in the body are indicated. (From Reference 3 with permission).
4. A buildup of charge exists on the surface of the body, as

with DC elds, which will lead to an enhancement of the
eld in the air near sharp objects such as tips of hair.
High Frequency Fields. At frequencies above tens of MHz
(for bodies the size of an adult human), the simple calculations presented above become increasingly in error because
of wave propagation effects. One must then calculate the
induced elds in the body using the full set of Maxwells
equations, usually by numerical techniques (10). Reference
10 provides an extensive discussion of experimental and
theoretical dosimetry.
Figure 2 shows the whole-body average SAR in an ellipsoid modeling the body, subject to plane-wave irradiation
at an incident power density of 10 W/m2 . At low frequencies, the SAR is proportional to the square of the frequency,
as expected from equations 3 and 5. The whole body SAR
exhibits a maximum near 70 MHz (for an ellipsoid modeling an adult human in a vertically polarized eld) because
of an electrical resonance resulting from the properties of
the body as an antenna.
When irradiated with waves whose frequency is near or
above the resonant frequency, the SAR within the body becomes nonuniform. Close to the resonant peak, the SAR is
maximum in the center of the object. At frequencies above
about 10 GHz, the exposure comes to resemble plane-wave
irradiation of a tissue plane. In that case, the energy penetration depth (6) is given by
=
3.38 105
f
(6)
1 (tan())2
where f is the frequency in Hz and

tan() = tan(
)
2 f o
(7)
is the loss tangent of the material. The SAR as a function

of depth x into the tissue is given by
SAR =
Io ex/
(8)
Figure 3. Maximum permissible exposure for persons in controlled (dashed line) and uncontrolled (solid line) environments.
These correspond approximately to occupational and general public exposure limits.
At frequencies above about 10 GHz, the energy penetration

depth of the incident wave is less than a few mm and the
energy is absorbed very close to the surface.
Mechanisms of Interaction
Fields in the frequency range considered here are nonionizing (i.e., their photon energy is too weak by orders of
magnitude, for the elds considered here to disrupt chemical bonds), which is in sharp contrast to ionizing radiation such as X-rays, which damage tissue by disrupting
chemical bonds and creating highly reactive free radicals.
It should also be noted that powerfrequency elds are not
properly considered to be radiation, because the exposure
occurs in the near elds of the sources that for all practical
purposed do not radiate.
Numerous mechanisms exist by which nonionizing electromagnetic elds can interact with biological systems, (for
reviews, see References 1214. Each mechanism is characterized by an intensity and frequency threshold for producing observable effects. These mechanisms can be classied as thermal or nonthermal, depending on whether the
mechanism is directly associated with tissue heating or
some other mechanism.1 They can be further classied as
rst or second order in the eld strength. This distinction
is important because mechanisms that are quadratic in
eld strength are generally weaker than rst-order mechanisms, but can be effective at higher frequencies.
1 Thermal and nonthermal are used inconsistently in the bioeffects literature, at times referring to mechanism for producing
an effect, and at times (even in the same paper) referring to the
presence or absence of what the investigator considers to be a signicant change in temperature. Some investigators use the term
athermal to indicate a biological effect that occurs in the absence
of noticeable heating, but which is still produced by thermal mechanisms. An example of an athermal effect might be a physiological response caused by thermoregulation, although no noticeable
change in body temperature occurs.
Thermal Effects. Thermal effects are produced by Joule

heating (a quadratic effect in electric eld), as given by the
SAR. A useful benchmark for assessing the threshold for
thermal effects is the basal metabolic rate of man, about 1
watt per kg of body mass. Wholebody average or local SAR
values substantially higher than 1 watt/kg are likely to be
thermally signicant, either because of the total thermal
load on the body or because of localized temperature rise.
With typical values for the conductivity of tissue (Table 2
(15, 16)), it corresponds to internal electric eld strengths
ranging from 100 V/m (50/60 Hz) to 30 V/m (1 GHz). Thus,
internal eld strengths above 30100 V/m will cause signicant heating if the exposure persists long enough.
Thermal effects can be observed from microwave elds
(typically pulsed elds) in the absence of large (or even
measurable) temperature increases. Pulsed microwaves of
high peak but low average power can elicit auditory sensations in people by generating sounds in their heads. Each
microsecond pulse raises the tissue temperature by a few
microdegrees, but the acoustic waves that are produced
(by thermal expansion of tissue uids) can exceed 100 dB
sound pressure (17) The sensations arise from perception of
these acoustic transients by normal hearing mechanisms.
At a higher level of exposure, a variety of responses can
be elicited by brief exposure to intense microwave elds.
Mice exposed in the head to intense microwave pulses, sufcient to heat their brain temperatures by a few tenths
of a degree within one second, exhibit a variety of involuntary body movements and other stun phenomenon (18).
The effect is related to the high rate of heating of the brain
tissue, which results in changes in membrane potentials
(19), not to the magnitude of the temperature rise itself.
Such exposures require specialized equipment (associated
with military weapons systems) and do not occur in normal
environments.
Nonthermal Mechanisms. Numerous nonthermal mechanisms are known, although in many cases the required
eld strengths are too high to result in observable biological effects under realistic exposure conditions. These effects include:
Membrane Excitation. Low frequency electric elds can
excite membranes, causing shock or other effects. At 50/60
Hz, the threshold current density in tissue for producing
shock is very roughly 10 A/m2 , which corresponds to an
electric eld strength of about 100 V/m in the tissue. Thus,
at 50/60 Hz, the internal electric eld strength needed to
produce shock is comparable with that needed to produce
signicant heating.
As the frequency is increased above about 1 kHz, the
threshold eld strength needed to excite cellular membranes increases rapidly. By contrast, the internal eld
strength to produce signicant heating increases slowly
with frequency, being determined instead by the frequency
dependence of the conductivity. Thus, at low frequencies
(<1 kHz), the obvious hazards are generally related to
shock or other membrane excitation phenomena, whereas
at higher frequencies, the obvious hazards generally are
a result of burns or other effects of heating body tissues.
This distinction is not absolute; for example, low frequency
currents can produce serious burns, which might be more

serious, under some hazard scenarios, than electric shock.
Electroporation. Electric elds can create pores in cell
membranes (electroporation) by inducing electrical breakdown, which requires potential differences across the membrane of 0.11 V, which in turn requires high eld strengths
in the surrounding medium (typically 104 105 V/m). However, electroporation may occur in the absence of thermal
damage if the exposure is brief enough, and it may be a
factor in electrical injury (20).
Perception of Weak Fields. At an opposite extreme, some
animals are sensitive to very small electrical elds; for example, some sharks and rays can detect electric elds in
the surrounding sea water below 1 V/m (21). This extraordinary sensitivity involves specialized electrical sense organs (the ampullae of Lorenzini) that depend on the salt
water environment of the animal, and is unrelated to biological effects in humans.
Electric FieldCharge Interactions. DC currents introduced into the body can produce electrochemical changes
through iontophoresis and electro-osmosis, which can lead
to hazards including electrochemical burns (22). These effects require DC currents to be introduced into the body,
which in turn requires contact with external conductors.
Electric FieldPermanent Dipole Interactions (Force Linear in E). Electric elds can exert torques on permanent
dipoles, which range from dipolar molecules to cells and
larger tissue structures. However, the rotational response
is damped by viscous forces from the surrounding media,
and the alignment is disrupted by thermal agitation. A
tradeoff also exists between the threshold eld strengths
needed to produce a signicant alignment and the cutoff
frequency of the response. Very high elds (above 106 V/m)
are needed to produce a signicant alignment of globular
protein molecules, but the cutoff frequencies are high (
1 MHz). Larger structures, such as cells, can be oriented
by weaker elds, but the cutoff frequencies are lower. To
produce a biologically signicant effect by such a mechanism would normally require eld levels that are thermally
damaging.
Electric FieldInduced Dipole Interactions (Force
Quadratic in E). An electric eld can exert mechanical
forces on cells and other structures through its interaction with induced electric dipoles, a phenomenon called
dielectrophoresis. This effect has important applications
in biotechnology, such as manipulating cells or to studying
their electrical properties (23). Such effects require high
elds, typically thousands of volts per meter or more.
Effects of Magnetic Fields.
Magnetic Force Effects. With few exceptions, biological
tissues are diamagnetic (i.e., their magnetic susceptibilities are very close to that of vacuum). Nevertheless, it
is possible under some circumstances to detect magnetically induced forces on biological materials, either through
magnetic eld-permanent dipole or magnetic eld-induced

dipole interactions. These mechanisms can be rst or second order with the eld.
Magnetic elds will induce a torque on a magnetic
dipole, which is a linear function of the eld strength. One
biological effect developing from such forces is the orientation of magnetotactic bacteria by the Earths magnetic
eld (50 T) (24). These bacteria contain chains of small
magnetite particles, and it is the interaction between the
Earths eld and these particles that aligns a bacterium.
Similar particles have been found in other species, including human brain tissue, and some investigators have suggested that it might provide a mechanism for magnetic eld
sensitivity in some animals (25).
Much smaller forces can be exerted by inhomogeneous
magnetic elds on ordinary biological materials, which is
a quadratic eld effect analogous to dielectrophoresis. Investigators have levitated (magnetically suspending in air)
saline droplets containing fertilized frog eggs, using inhomogeneous magnetic elds of about 13 T (26). Very strong
elds (generally > 10 T) can orient cells and lipid membranes (27).
Other Effects of Magnetic Fields. Various subtle effects
can be observed in humans and animals exposed to strong
DC magnetic elds. At ux densities generally > 1 T,
changes can be observed in the electrocardiogram, typically an increase in the amplitude of the T wave. These
changes result from electric potentials that are induced in
blood as it ows through the aorta, which give rise to an
additional voltage superimposed on the normal electrocardiogram (28).
Effects of DC magnetic elds on rates of bimolecular
reactions with nonzero-spin intermediates are well documented, typically at levels of at least 1 mT, and generally
over 10 mT (29). Some investigators have proposed that
low frequency magnetic elds can cause biological effects
through a similar mechanism (30). However, other scientists have argued that such mechanisms are unlikely to
result in biological effects from 50/60-Hz magnetic elds
at typical exposure levels (31).
BIOLOGICAL EFFECTS OF ELECTROMAGNETIC FIELDS
IN HUMANS
This subject is large, and particularly as it relates to possible chronic health effects from low levels of exposure, very
contentious. We will rst consider the effects that underlie
major exposure standards for electromagnetic elds, and
then comment about possible effects of chronic exposure to
electromagnetic elds in humans.
Static Fields
The strongest DC magnetic eld that a human is likely to
encounter is in the low-Tesla range (e.g., during imaging
by MRI). No biophysical reason exists to anticipate pronounced effects in humans at such levels, and limited testing with animals has disclosed only subtle effects. One investigator noted mild sensory effects in humans exposed
to 4 T elds in an MRI system (32). One group reported a
change in the permeability of the bloodbrain barrier permeability in rats from exposure to 1.5-T magnetic elds
(33). This effect (which is so far unconrmed by independent studies) remains poorly understood and its implications for health and safety are unclear. Blondin et al. reported the threshold for human perception of DC electric
elds to be 4045 kV/m (34).
Low Frequency Fields

The threshold for human perception of 50/60-Hz electric
elds is in the range of 2 to 10 kV/m, because of the movement of the hairs on the subjects skin. The threshold for
perceiving spark discharges corresponds to eld strengths
outside the body of about 3 kV/m, when a person can perceive mild shocks when touching a conductive object located in the eld. At about 810 kV/m, painful shocks can
be felt by nger contact with vehicles or other large objects
in the eld (35). Above about 20 kV/m, annoying shocks
are felt at the shoe level (36). Exposure to 50/60-Hz electric elds at still higher levels is very unpleasant because of
transient shocks. These effects are a result of currents that
are passed into the body by touching a conductive object;
the elds induced within the body from such exposures, in
the absence of contact with conductive objects, is generally
far below the thresholds for inducing shock (Fig. 1).
Bone Repair Stimulated by Pulsed Magnetic Fields

An often-reported effect of magnetic elds that remains
poorly understood is the stimulation of fracture repair in
bones. Magnetic bone stimulators have been approved for
sale in the United States since the late 1970s. These devices
employ a variety of pulsed elds, and they induce electric
elds in the body with peak levels of the order of 0.1 V/m
(46). The corresponding peak current densities are of the
orders of tens of mA/m2 in soft tissue, which are above the
level of naturally occurring (endogenous) elds but are well
below anticipated thresholds for eliciting action potentials
in excitable tissue.
Magnetophosphenes
A well-established effect, rst observed by dArsonval in
1896, is the production of visual sensations, called magnetophosphenes, when the head is exposed to alternating
magnetic elds (1020 mT, 50 Hz). These effects are caused
by small currents (20200 mA/cm2 ) that are induced in
the retina of the eye (47,48), and are the most sensitive
responses observed in humans from exposure to low frequency electric or magnetic elds. The effect has not been
considered to be a hazard by the expert committees that
have examined the literature.
Endogenous Electric Fields
Cardiac Excitation and Fibrillation

Various effects have been reported or can be anticipated
from low frequency alternating magnetic elds at sufcient
ux densities, which are associated with induced electric
currents (Table 2) (3739). One potentially lethal effect is
cardiac brillation from induced electric elds in the chest.
Reilly estimated that the threshold current density in the
human heart for causing electrical stimulation is about 1
A/m2 (60 Hz), with thresholds for inducing brillation being about 23 times higher. Current densities in excess of 2
amp/m2 (60 Hz), if applied directly to the heart, can induce
brillation in dogs (40). The National Radiological Protection Board (NRPB) estimated the threshold induced current that would produce cardiac brillation in the human
to be 3 A/m2 (41). To produce such currents in the human
chest would require magnetic ux densities above 100 mT
at 60 Hz (42).
Naturally occurring (endogenous) electric elds exist in the

body because of bioelectric phenomena such as the electrocardiogram, electromyogram, and electroencephalogram.
Such elds generally have most of their spectral density
below 100 Hz and correspond to current densities of 110
mA/m2 ; but current densities of 1000 mA/m2 can develop
during brief periods of electrical activity (action potentials)
on the surface of nerve or muscle cells (49). The ubiquitous
presence of such elds in body tissues would seem to place
a lower limit on the strength of externally induced elds
that would be biologically signicant.
High-Frequency Fields. As the frequency of the applied
eld is increased above 13 kHz, cell membranes become
progressively less sensitive to stimulation, and thermal
phenomena predominate. A variety of effects, some hazardous, have been reported from high frequency electromagnetic elds (mostly at radiofrequencies and above).
Nerve Stimulation by Pulsed Magnetic Fields
Perception and Pain
Some clinical devices use pulsed magnetic elds to stimulate nerve and muscle tissues (43, 44), which typically
employ pulsed magnetic elds, usually of millisecond duration and with magnetic slew rates (dB/dt) of tens of T/s,
and peak magnetic elds ranging from several hundred mT
to several T.
Other effects such as headache and general discomfort
(45) have been reported in humans at ux densities > 60
mT (50/60 Hz) and may involve neurological effects. However, the literature on these effects is sparse and variable
in quality, and the mechanisms for the effects are not well
understood.
The thresholds for perception of microwave energy have

recently been measured by Blick et al. for brief (10-second)
exposure to microwave energy over an area of 0.024 m2 on
the backs of human volunteers (Table 3) (50). A thermal
model, which takes into account the reection of energy
from the skin and heat conduction, shows that the temperature rise at the skin surface at the threshold for perception
is about 0.07 C over this entire frequency range (51).
The threshold for perception decreases with increasing
frequency, because of the shorter penetration depth into
tissue (and corresponding increase in SAR near the skin
surface), modied by thermal conduction effects.
The threshold for painful stimulation under similar exposure conditions would be 50100 times higher for these
exposure conditions. However, the thresholds for perception and pain are likely to vary greatly, depending on the
location of the exposed surface, duration of exposure, and
other variables.
Other Thermal Effects

A wide variety of thermal effects have been demonstrated
in animals that can be anticipated to occur in humans as
well. Some of the more signicant include the following:
Cataract is an established hazard of microwave enBurns

At still higher exposure levels, or for longer exposure times
at these exposure levels, burns may it result. Despite the
many high power sources of radiofrequency and microwave
energy that exist in modern society, the incidence of serious burns from radiofrequency energy is low (a handful of
incidents have been reported of burns from defective microwave ovens, some of them disputed (52), and occasional
industrial accidents occur that involve exposure of workers
to high levels of radiofrequency elds).
The thresholds for producing thermal damage in tissue
depend on the magnitude and duration of the increase in
temperature. Tissues can tolerate heating to temperatures
below about 43 C almost indenitely, whereas higher temperatures will lead to thermal damage at progressively
shorter times at higher temperatures (53). Thus, the microwave exposure needed to produce thermal damage depends on several factors, including the time over which irradiation persists, the size of the heated region, and the
rate at which heat is transported from the heated region.
Despite the fact that burns are an unequivocal potential
hazard from exposure to RF energy, the incidence of such
injuries is very low. Reasons include the general inaccessibility of high powered sources of RF energy to the general population and the fact that overexposure will lead
to severe pain before thermal damage occurs, forcing the
subject to withdraw from exposure. Incidents of serious
injury to exposure to RF energy are typically associated
with industrial accidents (for example, one case involved a
tower climber who was suspended in front of a high powered broadcast transmitter and could not escape) or in military environments in which a person comes into contact
with very intense RF beams.
Thermoregulatory Effects
If the heat load to the body is comparable with, or even
below, rates of heat generation by metabolism, biological
effects can occur as a result of the normal operation of the
thermoregulatory mechanisms. Characteristic thermoregulatory changes include alterations in blood ow, respiration, sweating, and many more subtle physiological responses. Various subtle physiological effects observed in
animals after moderate exposure to microwaves can be
interpreted as normal thermoregulatory responses, and
various reported synergistic interactions of drugs and microwave exposure might likewise have a thermoregulatory
component (54). The basal metabolic rate in man is about
1 W/kg of body mass, and whole-body exposures somewhat
above and below this level can be expected to produce thermoregulatory responses.
ergy, which has been reported occasionally in humans

exposed to high intensity microwaves. Cataracts can
be produced in animals at high exposure levels, generally > 1000 W/m2 , and are associated with temperature increases of several degrees in the eye. Exposures
at such levels would be acutely hazardous to personnel because of the likelihood of thermal injury to the
body (and not just in the eye). Occasional claims of
cataract produced by low level microwave exposure
have not been substantiated by animal studies (55).
Reproductive effects (birth defects and other adverse
effects) are well established in mammals exposed to
microwave energy sufcient to raise body temperature [e.g., (56)]. The exposure required to produce such
effects is quite high (tens of W/kg), and the effect can
be presumed to be thermal in nature. High exposures,
far above present exposure safety limits, would be required to produce such effects.
Behavioral disruption has been observed in several
species of animals exposed to microwaves at wholebody exposures of 46 W/kg, at several frequencies
above 100 MHz (57). In such studies, the animals
are trained to carry out a task (for example, pressing a lever to obtain food pellets) and exposed to microwave energy. At some exposure level, the animals
stop performing the assigned task and begin a different behavior, typically one associated with thermoregulation (for example, spreading saliva on the tail in
rats). Major standards-setting committees have considered behavioral disruption to be the effect of radiofrequency energy that has been reliably demonstrated at the lowest whole-body exposure level. Behavioral disruption can be interpreted as a normal
response of the animal to the excessive heat load to
its body, and is not, strictly speaking, an adverse effect at all; however the thermal burden to the body in
such cases is undoubtedly close to hazardous levels.
Consequences of Long-Term Exposures

All of the scientic results discussed so far pertain to effects produced by short-term exposures to electromagnetic
elds, usually at high levels. A very different body of literature has emerged since the late 1970s pertaining to possible health effects from long-term exposures to environmental elds, either at power line (50/60 Hz) or radiofrequencies, based on epidemiological studies that focus on possible statistical correlations between exposure and a health
endpoint.
The epidemiology literature is far too large to review
here. The following paragraphs provide citations to major
recent reviews and indicate the nature of their conclusions.
Powerline Fields. Most epidemiology studies related to

power line elds have focused on a possible link between exposure to 50/60-Hz magnetic elds and childhood
leukemia, rst suggested in a 1979 paper by Wertheimer
and Leeper (58). A variety of other health endpoints have
been examined as well.
By now, more than 100 epidemiology studies have been
reported related to power frequency elds and human
health; for recent reviews, see Kheifets and Shimkhada
(59) and Feychting et al. (60). A denitive review of the
literature, focusing primarily on epidemiology literature
but considering relevant animal studies, was published in
2002 by the International Agency for Research on Cancer
(IARC) (61). The IARC review classied power frequency
magnetic elds as possible human carcinogens (Group 2B
in its classication scheme), based on limited epidemiological evidence together with limited or inadequate animal
evidence (other 2B carcinogens, as determined by IARC, include coffee, automobile exhaust, and pickled vegetables).
IARC judged the evidence for other health effects to be inadequate to draw conclusions.
Radiofrequency Fields. Public concerns that use of mobile telephones might cause brain cancer were triggered
by a story broadcast on a United States television show in
1993 by a man whose wife had used a mobile telephone and
subsequently developed brain cancer, which he attributed
to the effects of the phone. Such observations have obvious weaknesses for documenting cause-and-effect relationships: many millions of people use mobile telephones and
an incidence of the disease in the United States is 1520
new cases per hundred thousand people per year, and consequently many users of mobile phones will develop brain
cancer every year even if no causal link to the phones exist. In the intervening decade, more than a dozen largescale epidemiology studies, and numerous animal studies,
have been undertaken (for recent reviews, see Ahlbom et
al. (62) and Moulder et al. (63). Several massive epidemiology studies are nearly completed and much more data will
become available on this issue in coming years; present
results are generally negative but insufcient to persuade
most health agencies that no hazards (cancer or otherwise)
are associated with use of mobile telephones. The problem
is complicated by the largely unknown cause of brain cancer in its various forms, by the long latency (time between
initiation of the tumor and its clinical detection in a patient), and by the rapidly changing technology of wireless
communications.
EXPOSURE LIMITS
Several government and nongovernment agencies have established guidelines for human exposure to electromagnetic elds. The standards are complex, and the reader is
referred to the original documents for authoritative information about the standards and how they would be applied.
DC Fields
Few standards exist for exposure to DC electric and
magnetic elds. The International Commission on NonIonizing Radiation Protection (ICNIRP) set a limit for occupational exposure of 2T, with a time-weighted-average
of 0.2 T over a workday (64). Similar limits have also been
set by the National Radiation Protection Board (NRPB) in
the United Kingdom and the American Conference of Governmental Industrial Hygienists (ACGIH), as well as other
agencies. Exposure limits set by various agencies for static
electric elds are in the range of 2560 kV/m. It should be
noted that these limits are designed to protect against possible health effects of the elds; protection of patients with
implantable devices undergoing MRI examination (for example) is a different matter entirely.
Low Frequency Electric and Magnetic Fields
Several agencies have established exposure standards for
both powerfrequency electric and magnetic elds. Most
are designed to limit magnetically induced current densities within the body to 10 mA/m2 at 50/60 Hz, on the assumption that induced currents below the levels of endogenous currents are unlikely to be hazardous. Other limits
apply to contact current, which is introduced into the body
by contact with an external conductor in a eld. Exposure
limits set by U.S. and European agencies are in the range
of 0.57 mT for magnetic ux density and 1030 kV/m for
electric elds of 50/60 Hz (Table 4).
Radiofrequency Energy
The exposure standard for RF elds that is most widely
adopted around the world is that of the International Commission for Non-Ionizing Radiation Protection (ICNIRP);
another and quite similar standard is IEEE C95.1-2005
(65).
The ICNIRP and IEEE C95.1 exposure guidelines provide two sets of limits: basic restrictions and reference levels. Basic restrictions are limits based on health effects
and specify the maximum elds or SAR that exist within
the body from exposure to elds. As the induced eld or
SAR within the body is not practical to determine in most
practical settings, both ICNIRP and IEEE C95.1 also provide a set of reference levels. These levels are the strengths
of the elds outside the body, or incident power density,
that would ensure compliance with the basic restrictions.
The guidelines allow exposures above the reference levels, but in that case, a detailed exposure assessment would
be needed to ensure that the basic restrictions are satised. In addition, ICNIRP and IEEE C95.1 have two tiers
of limits that correspond to limits for occupational groups
and the general population (ICNIRP) or for persons in
controlled and uncontrolled environments (IEEE
C95.1). A controlled environment in the IEEE limits is
one in which occupancy of a person is subject to control and
accountability as established by an RF safety program.
Both the ICNIRP and IEEE C95.1 limits have been criticized at times as being thermal and not protective against
athermal hazards. Indeed, they are designed to protect
against hazards of RF energy (e.g., burns, shock, excessive
10
heat load to the body), which are characteristically associated with short-term exposures to elds at high levels. Both
ICNIRP and IEEE C95.1 acknowledge that athermal biological effects have been reported by some scientists from
low level exposures to RF energy, but the standards-setting
committees considered the evidence that is presently available about athermal effects to be insufcient to use to develop exposure guidelines. Thus, overall, the literature on
athermal effects of AM electromagnetic elds is so complex,
the validity of reported effects so poorly established, and
the relevance of the effects to human health is so uncertain, that it is impossible to use this body of information
as a basis for setting limits on human exposure to these
elds (ICNIRP).
Controversies and Unresolved Issues
The biological effects of RF and microwave energy has been
remarkably contentious for many years, which is a common
situation with environmental health issues of all sorts, in
part because of the inevitable gaps and ambiguities in the
data, and in part because of different perceptions by different stakeholders about the interpretation of the data. In
common with other environmental health issues, the scientic literature in this eld includes many studies that are
highly variable in relevance to health and scientic quality. People who are inclined to worry can pick and choose
data to support their fears even as health agencies review
the broader literature and fail to nd persuasive evidence
for a health problem.
Particularly intense controversies have swirled for
decades over concerns expressed by many people that exposure to electromagnetic elds, of some sort, might be
linked to some form of cancer. Specic concerns have developed about the following: power frequency magnetic elds
and childhood leukemia or brain cancer, police radar detectors and testicular cancer, mobile telephone handsets and
benign or malignant brain tumors, and living near radio
transmitters and childhood leukemia. Despite these concerns, careful reviews by health agencies (some of which
are cited in the Appendix and are readily available on the
Internet) have consistently failed to nd persuasive evidence for any health effects from exposure to electromagnetic elds below IEEE or ICNIRP exposure limits.
DISCUSSION
In virtually all ordinary environments in modem society,
the eld levels that are present are very far below recommended (ICNIRP or IEEE) exposure limits. Limits, for
radiofrequency energy, are generally exceeded only very
close to high powered transmitters, which are hardly ever
present in ordinary environments but may occur in some
occupational settings. Few injuries are reported from exposure to radiofrequency elds, and still fewer from exposure
to low frequency elds. The electrical accidents that are reported more commonly involve contact with charged conductors. For example, construction workers near AM broadcasting facilities can suffer electrical bums from radiofrequency currents induced in cranes or other large metal objects near the transmitters. In the United States, acciden-
tal electrocution claims the lives of about 1000 individuals

every year. No doubt exists that electricity is dangerous,
but it seems that exposures to elds is not a major source
of injury.
APPENDIX:
Recent Assessments of Possible Health Risks of Electromagnetic Fields by National Expert Groups (Most of these
focus on radiofrequency elds; others include discussion of
possible health effects of power frequency elds. Several of
these agencies release yearly updates; only the most recent
report is included below.)
National Radiation Protection Board (UK)
Documents of the National Radiation Protection Board
(UK): Volume 14, No. 2 Health Effects from Radiofrequency Electromagnetic Fields: Report of an independent Advisory Group on Non-ionising Radiation.
http://www.hpa.org.uk/radiation/publications/documents
of nrpb/abstracts/absdl4-2.htm
UK Independent Expert Group On Mobile Phones
Mobile Phones and Health, UK. May 2000. http://www.
iegmp.org.uk/index.htm
Health Council of the Netherlands
Elektromagnetische Velden: Jaarbericht 2005 / Electromagnetic Fields: Annual Update 2005. The Hague:
Health Council of the Netherlands, 2001, 2001 / 14.
http://www.gr.nl/pdf.php?ID=1281&p=l
New Zealand Ministry for the Environment
National guidelines for managing the effects of radiofrequency transmitters, December 2000. http://www.mfe.
govt.nz/publications/rma/radio-freq-guidelines-dec00.pdf
The Royal Society of Canada
Expert Panel on Potential Health Risks of Radiofrequency Fields from Wireless Telecommunication Devices
2004 update report: Recent Advances In Research
On Radiofrequency Fields And Health: 20012003.
http://www.rsc.ca//les/publications/expert panels/RF//
expert panel radiofrequency update2.pdf
Swedish Radiation Protection Authority
Recent Research on Mobile Telephony and Health
Risks Recent Research on EMF and Health Risks.
Third annual - report from SSIs Independent Expert Group on - Electromagnetic Fields. - SSIs Independent Group on Electromagnetic Fields, 2005.
http://www.ssi.se/PdiUpload/SSI EMF 2005.pdf#search=
electromagnetic+elds
BIBLIOGRAPHY
1. Stuchly, M. A. Human Exposure to Static and Time-varying
Magnetic Fields. Health Phys. 51, 1986, pp 215225.
2. Budinger, T. F. et al.Biological Effects of Static Magnetic
Fields. Proc. 3rd Annual Meeting of the Society for Magnetic
Resonance in Medicine:Society for Magnetic Resonance in
Medicine, Berkeley, 1984, pp. 113114.
3. Carstensen, E. L. Biological Effects of Transmission Line
Fields. Elsevier: New York, 1987, p 12.
4. ibid.
5. Tell, R. A.; Mantiply, E. D. Population Exposure to VHF and
UHF Broadcast Radiation in the United States. Radio Sci.
1982, 17, (5S), 39S47S.
6. Foster, K. R.; Schwan, H. P. Dielectric Properties of Tissues.
In C., Polk; E., Postow, Eds.; Handbook of Biological Effects of
Electromagnetic Fields; 2nd Ed.; CRC Press: Boca Raton, FL,
1996, pp 27106.
7. Tenforde, T. S. Interaction of ELF Magnetic Fields with Living
systems. In Polk, C.; Postow, E., Eds.; Handbook of Biological
Effects of Electromagnetic Fields; 2nd ed.; CRC Press: Boca
Raton, FL, 1996, p. 189.
8. Foster, K. R.; Schwan, H. P. Dielectric Properties of Tissues.
In C., Polk; E., Postow, Eds.; Handbook of Biological Effects of
Electromagnetic Fields; 2nd Ed.; CRC Press: Boca Raton, FL,
1996, pp 27106.
9. Guy, A. W.; Davidow, S.; Yang, G.-Y.; Chou, C.-K. Determination
of Electric Current Distributions in Animals and Humans Exposed to a Uniform 60 Hz High Intensity Electric Field. Bioelectromagnetics, 1982, 3, pp 4771.
10. Durney, C. H.; Massoudi, H.; Iskander, M. F. Radiofrequency Radiation Dosimetry Handbook; 4th ed.; Brooks Air
Force Base: San Antonio, TX, Report USAFSAM-TR-85-73.
http:www.brooks.af.mil/AL/OE/OER/handbook/cover.htm.
11. ibid.
12. Schwan, H. P.; Foster, K. R. RF-eld Interactions with Biological Systems: Electrical Properties and Biophysical Mechanisms. Proc. IEEE 1980, 68, pp 104113.
13. Valberg, P. A.; Kavet, R.; Rafferty, C. N. Can low-level 50/60 Hz
Electric and Magnetic Fields Cause Biological Effects?, Rad.
Res. 1997, 148, pp 221.
14. Tenforde, T. S. Biological Interactions and Potential Health Effects of Extremely Low Frequency Magnetic Fields from Power
Lines and Other Common Sources. Ann. Rev Public Health.
1992, 13, 173196.
15. World Health Organization. Environmental Health Criteria.
69. Experimental Data on the Biological Effects of Static Magnetic Fields. World Health Organization: Geneva, 1987.
16. International Non-Ionizing Radiation Committee of the International Radiation Protection Association. Interim Guidelines on Limits of Exposure to 50/60 Hz Electric and Magnetic
Fields. Health Phys. 1990, 58, pp 113122.
17. Foster, K. R.; Finch, E. D. Microwave Hearing: Evidence for
Thermoacoustic Auditory Stimulation by Pulsed Microwaves,
Science 1974, 185, 256258.
18. Wachtel, H.; Brown, D.; Bassen, H. Critical Durations of
Pulse Microwave Exposures that Evoke Body Movements, Bioelectromagnetics Society 12th Annual Meeting San Antonio,
TX,1014 June, 1990, p 55.
19. Barnes, F. S. Cell Membrane Temperature Rate Sensitivity Predicted from the Nernst Equation, Bioelectromagnetics
1984, 5, pp 113115.
11
20. Lee, R. C.; Gaylor, D. C.; Bhatt, D.; Israel, D. A.; Role of Cell
Membrane Rupture in the Pathogenesis of Electrical Trauma.
J. Surg. Res. 1988, 44, pp 709719.
21. Kalmijn, A. J. Electric and Magnetic Field Detection in Elasmobranch Fishes. Science 1982, 218, pp 916917.
22. Geddes, L. A. Ed. Handbook of Electrical Hazards and Accidents. CRC Press: Boca Raton, FL, 1995, pp 4448.
23. Foster, K. R., Sauer, F. A., Schwan, H. P. Electrorotation and
Levitation of Cells and Colloidal Particles, Biophys. J. 1992,
63, pp 180190.
24. Frankel, R. B.; Blakemore, R. P.; Wolfe, R. S. Magnetite in
Freshwater Magnetotactic Bacteria. Science 1979, 203, pp
13551356.
25. Kirschvink, J. L.; Kobayaski-Kirschvink, A.; Diaz-Ricci, J. C.;
Kirschvink, S. J.; Magnetite in Human Tissues: A Mechanism
for the Biological Effects of Weak ELF Magnetic Fields. Bioelectromagnetics, Suppl. 1992, 1, pp 101113.
26. Vales, J. M. Jr.; Lin, K.; Deingre, J. M.; Mowry, K. L., Stable
Magnetic Field Gradient Levitation of Xenopus Laevis:Toward
Low-Gravity Simulation. Biophys. J. 1997, 73, pp 11301133.
27. Tenforde, T. S. Interaction Mechanisms and Biological Effects
of Static Magnetic Fields. Automedica 1992, 14, pp 271293.
28. Tenforde, T. S.; Gaffey, C. T.; Moyer, B. R.; Budinger, T. F., Cardiovascular Alterations in Macaca Monkeys Exposed to Stationary Magnetic Fields: Experimental Observations and Theoretical Analysis. Bioelectromagnetics 1983, 4, pp 19.
29. Schulten, K. Magnetic Field Effects in Chemistry and Biology.
Adv. Solid State Phys. 1982, 22, pp 6183.
30. McLauchlan, K. Are Environmental Magnetic Fields Dangerous? Physics World 1992, 5, pp 4145.
31. Scaiano, J. C.; Mohtat, N.; Cozens, F. L.; McLean, J.; Thansandote, A. Application of the Radical Pair Mechanism to Free
Radicals in Organized Systems: Can the Effects of 60 Hz be
Predicted from Studies Under Static Fields? Bioelectromagnetics 1994, 15, pp 549554.
32. Schenck, J. F. Health and Physiological Effects of Human
Exposure to Whole-body Four-Tesla Magnetic Fields During
MRI. Ann. NY Acad. Sci. 1992, 649, 285301.
33. Prato, F. S.; Wills, J. M.; Frappier, J. R. H.; Drost, D. J.; Lee, T.
-Y.; Shivers, R. R.; Zabel, P. Blood-brain Barrier Permeability
in Rats is Altered by Exposure to Magnetic Fields Associated
with Magnetic Resonance Imaging at 1.5 T. Micro. Res. Tech.
1994, 27, pp 528534.
34. Blondin, J. P.; Nguyen, D. -C.; Sbeghen, J.; Goulet, D.; Cardinal,
C.; Maruvada, P. S.; Plante, M.; Bailey, W. H. Human Perception of Electric Fields and Ion Currents Associated with Highvoltage DC Transmission Lines. Bioelectromagnetics 1996, 17,
pp 230241.
35. Zaffanella, L. E.; Deno, D. W. Electrostatic and Electromagnetic Effects of Ultra-high-voltage Transmission
Lines.Electric Power Research Institute: Palo Alto, CA, Final
Report EL-802, 1978.
36. Cabanes, J.; Gary, C. La Perception Directe du Champ Electrique. CIGRE Report 233-08, 1981.
37. Bernhardt, J. H. The Establishment of Frequency Dependent
Limits for Electric and Magnetic Fields and Evaluation of
Indirect Effects. Radiat. Environ. Biophys. 1988, 27, pp 1
27.
38. Jalinous, R.; Barker, A. T.; Freeston, I. L. The Design, Construction, and Performance of a Magnetic Nerve Stimulator. IEEE
Int. Conf. Electric Magnetic Fields in Medicine and Biology,
London, December 45, 1985, pp 5963
12
39. Barker, A. T.; Jalinous, R.; Freeston, I. L. Non-invasive Magnetic Stimulation of Human Motor Cortex. Lancet 1985, 1, pp
11061107.
40. Roy, O. Z., Summary of Cardiac Fibrillation Thresholds for 60
Hz Currents and Voltages Applied Directly to the Heart. Med.
Biol. Eng. Comp. 1980, 18, pp 657659.
41. National Radiological Protection Board (NRPB). Advice on Acceptable Limits of Exposure to Nuclear Magnetic Resonance
Imaging. Her Majestys Stationery Ofce (HMSO): Chilton,
Didcot, 1984.
42. Reilly, J. P., Electrical Stimulation and Electropathology. Cambridge University Press: Cambridge, 1992.
43. Irnich, W.; Schmitt, F. Magnetostimulation in MRI. Magn. Res.
Med. 1995, 33, 619623.
44. Silny, J. Stimulation Thresholds of the Time-varying Magnetic
Field in the Human Organism Bioelectromagnetics Society 8th
Annual Meeting, Madison, WI, June 15, 1986, p. 38.
45. Silny, J. Risks of Electromagnetic Fields for Humans. Versicherungsmedizin 1991, 43, pp 142148.
46. Polk, C. Electric and Magnetic Fields for Bone and Soft Tissue
Repair. In Handbook of Biological Effects of Electromagnetic,
Polk, C.;Postow, E. Eds.; CRC Press: Boca Raton, FL, 1996, p.
241.
47. Lovsund, P.; Oberg, P. A.; Nilsson, S. E. Quantitative Determination of Thresholds of Magnetophosphenes. Radio Sci. 1979,
14 (6S),pp 199200.
48. Seidel, D.; Knoll, M.; Eichmeier, J. Stimulation of Subjective
Appearance of Light (Phosphenes) in Man by Alternating
Magnetic Fields. Pugers Archiv. 1968, 299, pp 11.
49. Wachtel, H. Comparison of Endogenous Currents in and
Around Cells with those Induced by Exogenous Extremely
Low Frequency Magnetic Fields. In Advances in Chemistry,
Series 250; Blank, M., Ed., American Chemical Society: Washington, DC, 1995, pp 99107.
50. Blick, D. W.; Adair, E. R.; Hurt, W. D.; Sherry, C. J.; Walters,
T. J.; Merritt, J. H. Thresholds of Microwave-evoked Warmth
Sensations in Human Skin. Bioelectromagnetics. 1997, 18, pp.
403409.
51. Riu, P. J.; Foster, K. R.; Blick, D. W.; Adair, E. R. A Thermal
Model for Human Thresholds of Microwave-evoked Warmth
Sensations. Bioelectromagnetics. 1997, 18, pp. 578583.
52. Budd, R., Burns Associated with the Use of Microwave Ovens,
J. Microw. Power Electromagn. Energy, 1992, 27, pp 160
163.
53. Welsh, A. J. Heat Transfer in Medicine and Biology. In Laser
Irradiation of Tissue, Schitzer, A.;Eberhart, R. C. Eds.;Vol. 2.
Plenum Press: New York, 1985, p 167.
54. Adair, E. R. Thermoregulation in the Presence of Microwave
Fields. In Handbook of Biological Effects of Electromagnetic.
Polk, C.;Postow, E. Eds.; CRC Press: Boca Raton, FL, 1996, pp
403433.
55. Cleary, S. F. Microwave Cataractogenesis. Proc. IEEE 1980,
68, pp 4955.
56. Nawrot, P. S.; McRee, D. I.; Staples, R. E. Effects of 2.45 GHz
Microwave Radiation on Embryofetal Development in Mice.
Teratology 1981, 24, pp 303314.
57. DAndrea, J. A. Microwave Radiation Absorption: Behavioral
Effects. Health Phys. 1991, 61, 2940.
58. Wertheimer, N.; Leeper, E. Electrical Wiring Congurations and Childhood Cancer. Am.J.Epidemiol. 1979, 109, pp
273284.
59. Kheifets, L.; Shimkhada, R. Childhood Leukemia and EMF:

Review of the Epidemiologic Evidence. Bioelectromagnetics
2005, Suppl7, pp 19.
60. Feychting, M.; Ahlbom, A.; Kheifets, L. EMF and Health. Anu.
Rev. Public Health 2005, 26, pp 165189.
61. IARC Monographs on the evaluation of carcinogenic risks to
humans. Non-ionizing Radiation: I, Static and Extremely Lowfrequency (ELF) Electric and Magnetic Fields. Vol. 80. Lyon
France, IARC: 2002.
62. Ahlbom, A.; Green, A.; Kheifets, L.; Savitz, D.; Swerdlow, A.;
Epidemiology of Health Effects of Radiofrequency Exposure.
Environ. Health Perspect, 2004, 1/2, pp 17411754.
63. Moulder, J. E.; Foster, K. R.; Erdreich, L. S.; McNamee, J. P.
Mobile Phones, Mobile Phone Base Stations and Cancer: A
Review. Int. J. Radiat. Biol. 2005, 81, pp 189203.
64. International Commission on Non-ionizing Radiation Protection(lCNIRP).Guidelines on Limits of Exposure to Static Magnetic Fields. Health Phys. 1994, 66, PP 101106.
65. Institute of Electrical and Electronics Engineers (IEEE).
IEEE Standard for Safety Levels wtih Respect to Human Exposure to Radio Frequency Electromagnetic Fields, 3 kHz to
300 GHz. IEEE: New York, 2005.
KENNETH R. FOSTER
Department of Bioengineering,
University of Pennsylvania,
Philadelphia, Pennsylvania
CORPORATE MODELING WITH DSS AND ERP

SYSTEMS
Model Management Module
In recent years, corporate modeling within business organizations has moved from standalone database applications to integrated and exible enterprise wide information systems. Computer modeling and planning tools such
as management information systems (MISs) and decision
support systems (DSSs) are widely used by corporate management to select alternatives based on key performance
measures. The ability of DSS to model the real world systems and to predict system performance relies heavily on
the capabilities of the model base of the DSS.
Each DSS is built primarily to optimize a particular
business problem such as inventory ordering policy, job
scheduling, or production planning. A typical supply chain
system might involve more than one business operation
and therefore require the development of multiple DSSs.
Often, no link exists between the corporate DSS applications used by different departments. Data within each of
these applications are conned to departmental or functional boundaries, rather than being integrated to be used
by multiple users for multiple purposes and at multiple
places (1). This lack of data collaboration and integration
between business processes and functions has limited the
organizations capability to evaluate the operational problems globally, which affects the competitiveness of the organization in the marketplace.
Strategic Models
DECISION SUPPORT SYSTEMS (DSS)

Information system applications can be broadly classied
into transaction processing and decision support. Transaction processing systems are used in almost all companies to store information about their daily activities. Using
these systems, companies have accumulated huge amounts
of data that can be used for making business decisions. A
DSS deals with the design and use of computerized systems to assist managers in making more effective decisions
concerning different tasks based on the accumulated data,
such as inventory management. It is not practical to manually perform the classications each time a decision has
to be made. Therefore, a DSS is needed that has the ability
to perform the calculations on its own whenever a change
occurs.
Fundamentals of DSS
The basic components of a DSS include (1) Database Management Module, 2) Model Management Module, and 3)
User Interface Module. The schematic representation of a
DSS is shown in Fig. 1 along with direction of the data ow.
The database is the most important part of a decision
support system, as all models developed use these data to
generate decisions. The database can be accessed using a
user interface module that is usually in the form of screens
or forms. A database management system (DBMS) consists
of a database and a set of programs to access the data in
the database.
The model management module consists of algorithms
that need to be executed to perform the analysis. The ad-
Tactical Models
Database
Operational Models
DBMS
Functions:
-Creation-GenerationMaintenance-UpdateManipulation - Use
Figure 1. Components of DSS.
vantage of using a DSS is that it is capable of integrating

data access and decision models. The database acts as an
integration and communication mechanism between models. Both the model management module and the user interface module have access to the database. The results
generated by the model management module can be viewed
using the user interface module.
Material Requirement Planning (MRP) and Manufacturing
Resource Planning (MRP II) Applications
Early business applications of DSS focus on MRP and MRP
II. MRP is an inventory control technique that uses bills of
materials, inventory, expected receipts, and a master production schedule to determine material requirements. The
rst MRP system was introduced in the 1960s. It brought
innovational revolution to manufacturing and inventory
management. In the 1970s, modern MRP systems emerged
and strengthened several functions such as Master Production Scheduling (MPS) and Material and Capacity Planning (MCP), which are used to decide what, when, and
where to make products based on forecasted demand and
current inventory.
In the 1980s, the scope of MRP was expanded to integrate MRP with order processing, shop oor scheduling,
and resource planning, including personnel and machine
utilization analyses. These newer systems are called MRP
II. Although the basic method used to generate MRP is the
same, a typical MRP II system can also convert information
from MRP into specic work schedules on the shop oor,
evaluate workloads and capacity utilization, and generate
shipping documents. The framework of an MRP II system
is shown in Fig. 2.
CHARACTERISTICS OF ENTERPRISE RESOURCE
PLANNING (ERP) SYSTEMS
The ERP system is a new form of corporate modeling that
does the same thing as traditional DSS, but it is not conned to the boundaries of business applications. The intro-
Corporate Modeling with Dss and Erp Systems
Resource
Planning
Product
Planning
Rough-Cut
Capacity
Planning
Master
Production
Scheduling
Detailed
Capacity
Planning
Material
Requirement
Planning
Demand
Management
Front End
Engine
Material
and Capacity
Planning
Order
Release
Purchasing
Back End
Shop-floor
Scheduling and
Control(SFC)
Venders
Scheduling and
Follow-up
Figure 2. Framework of MRP II.
duction of ERP systems in the early 1990s stemmed from

the continuous enhancements made to MRP and MRP II
systems between 1960 and 1980 (2).
One study of mid-size to large companies conducted by
AMR Research (3) found that 67% of the surveyed companies are implementing some form of ERP, whereas another
21% are evaluating potential ERP systems solutions. Independent surveys conducted by Morgan Stanley (4) and
Deloitte & Touche/IDG Services Group (5) show that ERP
system implementation and upgrades are identied as one
of the top priorities among global Chief Information Ofcers as more companies are beginning to adopt ERP systems. A survey conducted by Deloitte Consulting in association with Benchmarking Partners, Inc. shows that 70%
of the respondents out of the 164 individuals from 62 Fortune 500 rms interviewed expected that their ERP system
would provide them with improved quality and visibility of
information (6).
Integration of Business Functions
The rise of ERP systems has been the center of discussion
in the software industry since the 1990s (7, 8). They are
strategic tools that allow companies to gain a competitive
edge in todays dynamic business environment by integrating all business processes and optimizing the resources
available. ERP systems are not purely software systems;

they affect how a business operates (9, 10). The enterprise
must be closely linked to manufacturers, suppliers, and
customers to produce goods tailored to customer requirements in the shortest possible time (1). To achieve this improved efciency and effectiveness, ERP systems equip the
enterprise with the necessary capabilities to integrate and
synchronize the isolated functions into streamlined business processes.
Although the acronym ERP originated in the manufacturing and distribution industries (2), ERP systems typically attempt to cover all basic functions of an organization
regardless of the type of industry. ERP is originally designed as a system that integrates all data and processes of
an organization into a single unied system. A typical ERP
system will use multiple components of computer hardware and software to achieve the integration. The key ingredient of ERP systems is the use of a single centralized
database to store data for various system modules. To be
considered an ERP system, a software package would need
to replace two or more independent applications and eliminate the need for interfaces previously required between
applications.
System Maintenance
ERP software is distinct from traditional in-house software
in several ways (11, 12). Unique attributes of ERP software are as follows: (1) It is bought from a vendor versus
built in-house; (2) helpdesk and maintenance support are
available from the vendor versus being entirely internallysupported maintenance activities; and (3) the installed version is replaced by choosing from readily available versions
versus reengineering or rewriting the whole system internally (11). ERP maintenance and upgrade are not purely
internal issues nor are they 100% external matter controlled entirely by the vendor or a third party outsourcer,
although the ERP software vendor has signicant inuence on ERP client maintenance and upgrade activities.
Carney et al. (13) discusses the contractual details
on the vendors long-term responsibility for maintenance,
emergency upgrades (in case of patches to repair bugs),
and policy of expected upward compatibility of future releases of the software. Contractual agreements between
ERP vendors and organizations allows the organization to
spell out all the benets and costs of a maintenance project
and reduce the risks involved in maintenance projects.
can be reviewed and evaluated both internally and externally. This type of data sharing has caused vendors to shift
focus from developing an all-in-one solution toward developing best-of-breed applications based on different types of
industry.
ERP vendors began breaking up and componentizing
their suites to make them easier to integrate with each
other and with legacy systems. The most popular applications within ERP II that benet companies include customer relationship management (CRM) and supply chain
management (SCM). The relationships between the business functions are shown in Fig. 3.
Most of the leading ERP vendors offer Web enabled
packages and Web services for companies to provide realtime information collected from these applications to their
external suppliers and end-users. Deployment of ERP II
systems takes place through a single vendor with a variety
of best-of-breed application vendors. The integrated ERP
II systems usually have a base system that enables the installation of additional application modules either from its
own product line or from different application vendors.
INFRASTRUCTURE OF ERP SYSTEMS
ERP I Systems
Early ERP systems (ERP I) have focused on optimization
of the business operations and have attempted to integrate
functions from all departments into a single computer system. Each system uses a single database so that the various
departments can easily share information and communicate with each other (1, 2). To successfully implement ERP,
all processes in a company must conform to the models describing the best practices in the industry (9). It involves
reengineering the existing business processes to the best
business process standard (14). One weakness of ERP I systems is that they do not share corporate information with
external parties. Firms face a variety of demands from an
increasingly competitive business environment, and these
demands are forcing decision makers to share information
beyond the connes of their internal environment.
Firms are looking more into multi-enterprise collaboration with their external partners across their supply
chains. This shift in focus and needs is not only from the
perspective of the rm itself but also from all its associated partners. Shaei and Sundaram (2) consider ERP II
as an application and as a strategy to include participation of all relevant partners of the rm across its supply
chain, enabling collaborative commerce. The main reason
why ERP II came into existence was the need to look at a
way to give customers and partners access to scheduling,
delivery, inventory, manufacturing, invoicing, and planning
information. ERP II is further empowered by new technologies like the Internet and Web services, and this has made
usage convenient for users in remote locations (15, 16).
ERP II Systems
ERP II emphasizes open architecture and vertical-specic
functionality. Lin et al. (17) report that most ERP I solutions provided by vendors in 2000 were proprietary and
differ from each other signicantly. In ERP II systems, data
From the viewpoint of information technology infrastructure, ERP systems have three major components: enterprise-wide database, application modules, and
client/server system (19). The enterprise-wide database in
a particular ERP solution varies from one ERP vendor
to another depending on the data structure and database
schema. Commonly used databases include IBM DB2, Microsoft SQL Server, Oracle Database, and mySQL AB.
Database repositories check for redundancies of data collected from different applications throughout the enterprise and store them in a standardized format readable
and accessible by each of these applications. Application
modules may include SCM, CRM, PLM, and many others.
The user interface is one characteristic of the client/server
model and is usually graphical and interactive, which allows end-users to retrieve data from the back-end and to
perform reporting services.
Open Source Development
Open source development in information technology has
evolved from traditional client/server computing to Webbased Internet platforms. It is a practice that promotes
access to an end products source materials, typically the
source code. It gained popularity with the growth of the
Internet by enabling diverse production models, communication paths, and interactive communities. This shift of
focus has changed the ERP system architecture to a 4tiers Web-based architecture (17): Client, Web server, application server, and database server. Software vendors are
adapting to this architecture by developing Web services.
Typical Web services consist of plug-in modules that can
be installed on any Web portal using the same programming language. Some of these programming languages
such as PHP and J2EE are open source languages licensed
under the GNU General Public License (16). Many open
source software companies began developing ERP software

Customer Relationship Management
Finance/
Accounting
Invoice
Sales Order
Shipping
Accounts
Receivable
Master
Production
Schedule
General
Ledger
Accounts
Payable
Inventory
Management
Bill of
Materials
Work
Order
Payroll
Purchasing
and Lead Times
Routings
and Lead Times
Supply Chain M anagement
Figure 3. Applications of the ERP system (source: Reference 18).
built on these open source platforms. The increased adaptability and decreased reliance on a single supplier and reduced costs are the main reasons open source ERP has been
developed.
demand network design (20). The presented solution is

a Web-based software that provides on-demand and realtime availability of information to all members of a manufacturing system, enabling them to react quickly and efciently.
Web-Based Systems
Web services are the most preferred ERP modules developed today both commercially and through the open source
community. The usage of Web-based ERP portals provides
transparency of usage to suppliers and end-users alike
without having to install additional software packages in
their machines. This portal is accessible anywhere within
the local area network (LAN) using an Internet browser as
long as the end-users are granted the appropriate access
to the portal. The same Web portal can be congured to
be publicly accessible from the Internet, thus giving external suppliers and end-users the ability to modify and view
real-time corporate data using an Internet browser or an
enabled mobile device.
The efciency of an enterprise depends on the quick
ow of information across the complete supply chain, from
the manufacturers to the suppliers. One instance of an
ERP solution that achieves high efciency is an integrated
Web-based logistics management system for agile supply
CONCLUSIONS
An ERP system should be sufciently versatile to support different manufacturing environments like make-tostock, assemble-to-order, and engineer-to-order (20). Wellplanned implementations of ERP systems have rich functionality across all areas like sales, account receivables,
engineering, planning, inventory management, production,
purchase, accounts payable, quality management, production, job scheduling, and distribution planning.
It is expensive and time consuming for companies to implement ERP systems; companies can take many years to
implement ERP systems, and the costs ranges from $10
million for a moderate size company to over $100 million
for large international enterprises (22, 23). Product selection for ERP packages is difcult because of the continuous
evolvement of ERPs technology and functionality. Statistical analysis techniques such as regression analysis, logis-
tic models, discriminate analysis, and data envelopment

analysis can be used to analyze empirical data from an
enterprise survey to provide important insights into ERP
implementations (24).
There are many different approaches exist to implementing the proper ERP solution and customizing the system to match the organizations business requirements.
For example, see References (2, 8, 10, 14, 25, 26). The interoperability of the ERP systems armed with a wide selection
of open-source software and commercial packages has enabled organizations to use this form of corporate modeling
to improve long term productivity and efciently manage
its business operations in the coming years.
BIBLIOGRAPHY
1. Y. John; K. Yim, A Study on an Environment of ERP Introduction, Proc. 2001 International Conferences on Info-tech and
Info-net, Beijing, China, Oct. 29Nov. 1, 2001.
2. F., Shaei; D. Sundaram, Multi-Enterprise Collaborative Enterprise Resource Planning and Decision Support Systems,
Proceedings of the. of the 37th Hawaii International Conference on System Sciences (HICCS 04):Hawaii, U.S.A, 2004.
3. J. Shepard; B. Locke; M. DAquila; K. Carter, Market Analytix Report: The Enterprise Resource Planning Report,
20042009. AMR Research Report, 2005.
4. D. M. Togut; E. Bloomberg, Morgan Stanley CIO Survey Series Release 4.5. Morgan Stanley Research Report, 2003.
5. Achieving, Measuring, and Communicating IT Value.Deloitte
Touche and IDG Research Services Group Report, 2002.
6. S. Baldwin, ERPs Second Wave: Maximizing the Value of
ERPEnabled Processes. Deloitte Consulting: Atlanta, U.S.A,
1998.
7. D. Oliver; C. Romm,Enterprise Resource Planning Systems:
An Outline Model Of Adoption; Proceedings of the GITM World
Conference:Memphis, TN, USA, June 1113, 2000.
8. C.J Stefanou, Supply Chain Management (SCM) and Organizational Key Factors for Successful Implementations of Enterprise Resource Planning (ERP) Systems; Proceedings of
AMCIS: Milwaukee, WI, USA, August 1315, 1999.
9. Y.F. Jarrar; A. Al-Mudimigh; M. Zairi ERP Implementation
Critical Success FactorsThe Role and Impact of Business
Process Management, Proceedings of the 2000 IEEE International Conference on Management of Innovation and Technology (ICMIT 2000): Singapore, Nov. 1215, 2000.
10. P. Bingi M.K Sharma J.K. Godla Critical Issues Affecting
an ERP Implementation, Information Systems Management,
Summer 1999.
11. C.S.P Ng, G. G. Cable, T. Chan, An ERP Maintenance Model,
Proceedings Proc. of the 36th Hawaii International Conference
on System Sciences (HICCS 03),Hawaii, U.S.A, 2003.
12. L. Markus; C. Tanis, The Enterprise Systems
ExperienceFrom Adoption to Success. In Framing the
Domains of IT Management: Projecting the Future Through
the Past, R. W. Zmud and M. F. Price, Eds. Pinnaex
Educational Resources: Cincinnati, OH, 1999; p. 173207.
13. D. Carney, S. A. Hissam, D. Plakosh, Complex COTSbased
Software System: Practical Steps for Their Maintenance,
Journal of Software. MaintenanceMaint.: Research Res. and
Practice, 2002, p 357376.
14. N. Gibson, C. Holland, B. Light, A Case Study of a Fast Track

SAP R/3 Implementation at Guilbert. Electronic Markets,
1999, Vol.9.
15. W. Li, L. Peng Upgrade ERP from C/S to B/S Based on Web
Service, Proc. of the Second IEEE International Conference
on Services System and Services Management 2005 (ICSSSM
05): Chongqing University, China.
16. S. Nicolas,
S. Jose Maria, Open Source Software ERPs: A New
Alternative for an Old Need. IEEE Software. May/Jun 2006,
Vol.23, p. 9497.
17. H. Lin, P. Hsu, J. Leu, W. Tsai An Analysis of ERP Systems
Based on Ntier Architecture, Proc. 26th International Conference of Information Technology Interfaces 2004(ITI 2004):
Cavtat, Croatia, 2004.
18. J. Herizer, B. Render, Principles of Operations Management,
5th ed., New York: Prentice Hall, 2004.
19. T. Anderegg, ERP: A-Z Implementers Guide for Success, Resource Publishing, 2000.
20. P. Helo,Y. Xiao, J.R. Jiao, A Web-Based Logistics Management
System for Agile Supply Demand Network Design. Journal of Manufact.uring Technology Management, 2006, v17, pp
10581077.
21. A.L. Symeonidis, D. Kehagias, A. Koumpis, A. Vontas, Open
Source Supply Chains, Proceedings of the 10th ISPE International Conference on Concurrent Engineering Research and
Application, Enhanced Interoperable Systems: Madeira, Portugal, July 2630, 2003.
22. T. H. Davenport, Putting the Enterprise into the Enterprise
System. Harvard Business Review. 1998, 76(4):p 121131.
23. V. A. Mabert, A. Soni, M. A. Venkatarmanan, Enterprise Resource Planning Survey of U.S Manufacturing Firms. Production Inventory Manage. 2000, 41(2):p 5258.
24. M. Vincent, S. Ashok, Model Based Interpretation of Survey Data: A Case Study of Enterprise Resource Planning Implementations. Mathematical. & Computer. Modeling; July
2006, Vol.44, pp 1629.
25. L. Frank, Architecture for Distributed ERP Systems. Proceedings Proc. of the 2005 VLDB Workshop on Trends in Enterprise Application Architecture (TEAA 2005): Trondheim, Norway, August 28, 2005.
26. M.A. Janson, A. Subramaniam, Packaged Software: Selection and Implementation Policies. INFORM. 1996; 34(2):p
133151.
JIM LEE
CHUN HONG LAU
University of Louisiana at
Lafayette, Lafayette,
Louisiana
DISTRIBUTED PARAMETER SYSTEMS

Many physical systems are naturally modeled by partial
differential equations (PDEs): Chemical reactions, uid
ow, vibrations of exible structures, and acoustic elds
are some examples that occur in engineering applications.
Typical control problems might be to reduce the vibrations
in a exible structure (such as a robot arm or a satellite
antenna), to reduce the noise levels in an aircraft, to damp
out undesirable eddy currents in a uid, or to attain a uniform temperature distribution for processing steel bars.
Current engineering practice is to obtain a mathematically
simpler linear ordinary differential equation model by approximating the original nonlinear PDE model based on
the physics of the process. Then, applying the formidable
arsenal of known nite-dimensional control theory and design methodology, a control scheme is formulated and subsequently tested and modied based on exhaustive computer simulation experiments. In the 1960s, certain mathematicians envisioned an alternative approach: Retain the
more accurate PDE model, and formulate the control problem for this model directly (1). The obvious advantage is
that the physical interpretation is retained at all stages of
the control design process. However, there is also the considerable disadvantage of introducing the very complicated
mathematics involved in manipulating PDEs. Moreover,
the feedback couplings, which are an inevitable feature of
control, introduce new types of PDEs that were not well
understood in the 1960s. This has provided the motivation
for intensive research into modeling and control of systems
described by partial differential equations.
Although it is useful to subdivide the area of control of
distributed parameter systems into
Modeling as a controlled PDE, including wellposedness issues and identication of parameters
Extension of nite-dimensional linear systems theory

and control concepts to deterministic and stochastic
PDE systems
Nonlinear systems
Implementation of the theory to controller design for
real physical systems
We stress that all four areas are closely interrelated.
MODELING
For example, whereas the modeling aspect draws on a vast
body of known theory of physics and mathematics of PDEs,
the control action must be incorporated into the model. Although distributed control (the control acts over a whole
region of the system) is fairly straightforward to model,
control action on the boundary requires a careful analysis. The closed-loop system with boundary control action
results in a nonstandard PDE, and much research has
gone into developing appropriate mathematical formulations, using both PDE and semigroup approaches (24).
Similar remarks apply to the modeling of the observation
via sensors that act on the boundary or at points in the
medium. More recently, the introduction of a new genera-
tion of advanced sensors and actuators (smart materials)

has necessitated formulating new composite PDE models
in which the actuators and sensors form an intrinsic part
of this new type of PDE model (5, 6). With ever continuing
technological advances in materials science and other areas of physics, we can expect a steady stream of new modeling problems for composite congurations of controlled
PDEs. A consequence of physical modeling is the inclusion
in the PDE model of several physical quantities (constants
or operators) that are not known precisely. To complete
the modeling step, these need to be estimated from measurement data. This important step has developed into a
subspecialty called parameter identication (7). The techniques can be described roughly as a sophisticated deterministic least-squares data-tting procedure in which emphasis is laid on the appropriate choice of a numerical
approximation scheme to suit the PDE under consideration. This is in contrast to nite-dimensional identication
techniques that use a stochastic algorithmic approach. Although a complete mathematical theory of stochastic evolution equations exists (8) and there is some stochastic
identication theory in a PDE setting, these theories have
not become current practice. The philosophy behind the estimation/identication step for PDE models is to exploit the
given PDE structure fully, as is done in the closely related
area of inverse problems.
EXTENSION OF FINITE-DIMENSIONAL THEORY

Most of the literature concerns the extension of known
nite-dimensional theory in systems and control. There
are two main approaches, the rst using a PDE description and the second using a semigroup one. The advantage
of the rst approach is that it is directly applicable to PDE
systems, and by using the power of PDE estimates applying
to a specic type of system (for example, parabolic or hyperbolic), very sharp results on controllability or stabilizability
can be obtained. For PDE systems, there are many possible
concepts of controllability; the main two are exact controllability (the ability to steer exactly to a given state of the
system) and the weaker concept of approximate controllability (steering arbitrarily closely to a given state). The rst
concept implies stabilizability of the system a feedback controller makes the system stable) as for nite-dimensional
systems, but the second concept does not. For this reason,
stabilizability replaces controllability as the key property
needed for control design. Similar remarks hold for the
dual concepts of observability and detectability. The literature on establishing that these properties hold for particular PDE systems is vast and continues unabated from the
1960s; the reason is that every new conguration of coupled
PDEs with assorted control action leads to a new, very difcult mathematical problem requiring sophisticated PDE
techniques (9). The other main problems studied from a
PDE viewpoint are stabilization by boundary control feedback and the linear quadratic control problem and its associated operator Riccati equation (3, 10). The latter problem
is the key to the main control design for distributed parameter systems. Recently, extensions of this problem to
the so-called minmax versions have received attention.
Distributed Parameter Systems
On the other hand, the semigroup approach has the advantage that several different types of PDEs and delay
equations can be included in the same theoretical formulation, and this formulation closely resembles that for ordinary differential equations (11, 12). The basic assumption is that the uncontrolled system can be modeled as a
strongly continuous semigroup. The attractive feature of
this approach is that it is more accessible to engineers (13)
and the theory naturally includes frequency-domain descriptions that are so useful in robust control design. A
wide range of control topics have been covered using this
description: linear control, dynamic compensators, linear
quadratic Gaussian and H-innity control, Kalman ltering, model reduction, servo problems, observer theory, P.I.
controllers, and adaptive control, to name just a few. The
best results have been obtained for linear systems with distributed control and observation; that is, there is sensing
and control distributed over the physical system. For systems that allow sensing and control at interior points or
on the boundary, the mathematical technicalities increase
dramatically. During the past decades, a theory for such a
class of well-posed linear systems has matured (14). The
key property of this class of well-posed linear systems is
that they are closed under composite congurations of cascade, parallel, and closed-loop connections. Many classic
control problems such as linear quadratic Gaussian and Hinnity control (15), (Riccati-) balanced realizations, tracking problems, passivity (16), and certain stabilization problems (17, 18) have been solved for this class of systems. The
price one has to pay for such a broad coverage is that the
step from the original PDE formulation to a semigroup one
is nontrivial and the results obtained for a particular PDE
example are not always the sharpest possible with dedicated PDE techniques. More recently, research has begun
on a wider classes of systems, where more general types of
semigroups are studied (19).
ods. This very high-order system is then approximated by

nite-dimensional techniques (20), and the controller design is based on this reduced order model. A disadvantage
of this approach is the lack of an adequate error analysis. Alternatively, one can rst do a PDE design and then
approximate the controller. For example, the H2 design
that comprises a linear quadratic controller coupled with
a deterministic observer with output gain from a dual Riccati equation. A fairly complete theory exists of numerical
approximations of the operator Riccati equations involved
and the effect of the approximating controller on the original PDE (10). Experience has shown that an appropriate
choice of the numerical approximation is crucial: Modal
approximations rarely give adequate results. An excellent
overview of this design methodology and its successes in
nontrivial applications to vibration control of a plate and
noise attenuation in two- and three-dimensional cavities
can be found in (5). Although the H2 control design usually
yields good results, in some applications, a minmax modication is to be preferred. A closer analysis of the full PDE
controller can often lead to useful information on the optimum placement of sensors and actuators and to choices for
low-order suboptimal controllers. The above H2 methodology is now well established, but it is a linear theory. If the
nonlinearities are not great, then a linearization of the nonlinear model is appropriate. However, the control of highly
nonlinear systems like controlled uid ow remains a challenge. A promising approach called Proper Orthogonal
Decomposition exploits this new theory on local low-order
approximations of PDEs to design low-order nonlinear controllers. In this approach, physical experimentation and
extensive computer simulations go hand in hand with
a theoretical analysis of the PDEs and their numerical
approximations (see (26)).
NONLINEAR PDES
1. Butkovskiy, A. G. Theory of Optimal Control for Distributed

Parameter Systems American Elsevien: New York, 1969.
2. Lions, J. L. Optimal Control of Systems Described by Partial Differential Equations, (transl.); Springer-Verlag: Berlin,
1971.
3. Bensoussan, A. et al., Representation and Control of InniteDimensional Systems, Vols. 1 and 2; Birkhauser:
Boston,
1992.
4. Lagnese, J. E.; Leugering, G.; Schmidt, E. J. P. G. Modeling,
Analysis and Control of Dynamic Elastic Multi-Link Structures; Birkhauser:
Boston, 1994.
5. Banks, H. T.; Smith, R. C.; Wang, Y. Smart Material Structures,
Modeling, Estimation and Control; Wiley: New York, 1996.
6. Smith, R.C. Smart Materials: Model Development, Vol 32 of
Frontiers in Applied Mathematics; SIAM: Philadelphia, 2005.
7. Banks, H. T.; Kunisch, K. Estimation Techniques for Distributed Parameter Systems; Birkhauser:
Boston, 1989.
8. Da Prato, G.; Zabczyk, J. Stochastic Equations in Innite Dimensions; Cambridge University Press: New York, 1992.
9. Lions, J. L. Controlabilite exacte, perturbations et stabilisation
de systmes distribues, Vols. 13; Masson: Paris, 1988.
10. Lasiecka, I.; Triggiani, R. Control Theory for Partial Differential Equations: Continuous and Approximation Theories, vol.I;
Abstract Parabolic Systems, Vol. II; Abstract Hyperbolic Like
Of course, many systems are nonlinear and PDE models

of physical systems often entail nonlinear damping effects
through the boundary conditions. Considerable literature
exists on the stability analysis of nonlinear PDEs using a
Lyapunov approach (20). More recently, research has been
done on proving stabilization by feedback control implemented on the boundary for certain nonlinear PDE models,
but as yet there is no general theory (21).
A more traditional topic is optimal control of nonlinear PDEs based on the philosophy of Pontryagins maximum principle (22, 23). Existence results are hard to prove,
but necessary conditions can be obtained and numerical
schemes for implementation are available (24). Of course,
the controllers are open loop, but in some applications, this
sufces, for example, shape optimization (25).
IMPLEMENTATION
The implementation of controllers for PDE systems inevitably involves approximation. The usual approach is to
rst approximate the PDE by various nite element meth-
BIBLIOGRAPHY
Distributed Parameter Systems
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
Systems over a Finite Time Horizon; Vols. 74,75 of the Encyclopedia of Mathematics and Applications; Cambridge University
Press: Boston, 2000.
Balakrishnan, A.V. Applied Functional Analysis; SpringerVerlag: Philadelphia, 1976.
Curtain, R. F.; Pritchard, A. J. Innite-Dimensional Linear
Systems Theory Vol. 8 in Lecture Notes in Control and Information Sciences; Springer-Verlag: Berlin, 1978.
Curtain, R. F.; Zwart, H. J. An Introduction to InniteDimensional Linear Systems Theory; Springer-Verlag: New
York, 1995.
Staffans, O. J. Well-Posed Linear Systems, Vol. 103 of the Encyclopedia of Mathematics and Applications; Cambridge University Press: Boston, 2005.
Mikkola, K. M. Innite-dimensional linear systems, optimal
control and Algebraic Riccati equations. Doctoral dissertation;
Helsinki University of Technology, Finland, 2002.
Staffans, O. J. Passive and Conservative Continuous-Time
Impedance and Scattering Systems. Part 1: Well-Posed Systems. Math. Contr. Signals Sys., 15, 2002, pp 291315.
Luo, Z. H.; Guo, B. Z.; Morgul, O. Stability and Stabilization
of Innite-Demiensional Systems with Applications; SpringerVerlag, Berlin, 1999.
Curtain, R. F.; Logemann, H. L.; Staffans, O. J., Stability Results of Popov-Type for Well-Posed Innite-Dimensional Systems with Application to Integral Control. Proc. London Math.
Soc., 86, 2003, pp 779816.
Opmeer, M. R. Distribution Semigroups and Control Systems.
J. Evolution Equations, 2006, pp 145159.
Walker, J. A., Dynamical Systems and Evolution Equations:
Theory and Applications, Plenum Press: New York, 1980.
Chen, G., Lasiecka, I., Zhou, J. Eds. Control of Nonlinear
Distributed Parameter Systems; Marcel Dekker: New York,
2001.
Li, X.; Yong, J., Optimal Control Theory for InniteDimensional Systems. Birkhauser:
Boston, 1994.
Fattorini, H. O. Innite Dimensional Optimization and Control Theory, Vol.62 of the Encyclopedia of Mathematics and
Applications; Cambridge University Press, Boston, 1999.
Ahmed, N. U.; Teo, K. L. Optimal Control of Distributed Parameter Systems. North Holland: Oxford, 1981.
Delfour, M. C. Shape Optimization and Free Boundaries, Series C: Mathematical and Physical Sciences; Kluwer Acadmic
Publishers: Dordrecht, 1992.
Smith, R. C. and Demetriou, M. A. Eds. Research Drections in
Distributed Parameter Systems; SIAM: Philadelphia, 2003.
RUTH CURTAIN
University of Groningen,
Groningen, The Netherlands
58
SUPERCONDUCTORS, METALLURGY OF BETA TUNGSTEN
SUPERCONDUCTORS, METALLURGY OF BETA

TUNGSTEN
When discussing technical superconductors one distinguishes
between high-temperature superconductors and low-temperature superconductors. The characteristic temperature is the
critical temperature, Tc. Above this temperature superconductivity ceases. When thinking about high temperatures one
typically addresses temperatures around the point of liquification of nitrogen (77 K); when considering low temperatures, values in the region of the boiling point of liquid helium
(4.2 K) are of interest.
The low-temperature superconductors can be further subdivided. For example, there are solid-solution superconductors and -tungsten superconductors. In this article, we
consider the latter. Another name for -tungsten superconductors is A-15 superconductors, a classification according to
the crystal structure of the material.
Additional main parameters to characterize superconductors are the upper critical magnetic field, Bc2, and the critical
density, Jc. Again, superconductivity disappears for values of
the magnetic field B or the current density J above their critical values.
Solid-solution superconductors, for example, NbTi, are limited to be used up to a magnetic field of about 9 T. Some
superconductors of the -tungsten group have the potential to
reach magnetic fields well above 20 T. To take advantage of
this potential, many different production methods of wires
and tapes have been established. As all A-15 compounds are
very brittle, the processing to final dimension takes place in a
more ductile state and mainly in combination with substrate
materials for electrical and mechanical stabilization. At final
dimensions, or even after winding to coils, the superconducting layers are formed during a defined heat treatment.
Many procedures have been developed in order to increase
Bc2 and Jc by adding elements for alloying the basic metals.
Because of poor ductility and differences in thermal expanJ. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
(010)-Direction
A ion
B ion
(100)-Direction
(001)-Direction
Figure 1. Part of an A15 lattice emphasizing the bcc-sublattice of

the A ions and the nonintersecting linear chains of the B-ions.
sion coefficients of the materials in such a composite, stress

and strain needs to be carefully controlled during handling
and in the final application, for example, high-field magnets
of different sizes.
This name, used for the class of materials addressed by the
title, is based on the -form of the element tungsten (W), first
reported by Hartmann et al. in 1931 (1). This modification is
taken as a representant of the A15 structure. As one was not
entirely sure whether or not this structure occurs in pure W,
but only in connection with oxygen, other prototypes have
been sought. The testing of the elements has not been fruitful,
and so the lattice of a binary intermetallic compound of chromium (Cr) and silicon (Si), that is, Cr3Si, became a commonly
used indicator. In the following text, the term A15 will be
utilized. In the unit cell of a binary compound A3B, which
shows the A15 structure, the ions are ordered in the following
manner (see Fig. 1): The B ions are arranged in such a way
as to build up a body-centered cubic (bcc) sublattice. The A
ions form nonintersecting chains in the (100)-, (010)-, and
(001)-directions on the faces of the unit cell.
In this group of superconductors, however, the A ions commonly are niobium (Nb) or vanadium (V), and the B ions are,
for example, recruited from tin (Sn), aluminum (Al), germanium (Ge), or gallium (Ga), without trying to be exhaustive.
Thus, typical representants are materials such as Nb3Sn,
Nb3Al, Nb3Ge, or V3Ga, to name only a few. Other members,
together with their transition temperature Tc, can be found in
Table 1 (24).
The explanation of why a crystal structure is applied in
order to characterize a class of superconductors, in spite of
the fact that it is not entirely possible to assign superconductivity to a certain combination of qualities, becomes evident
noting the following: Until the so-called high Tc superconduc-
59
tors were discovered, the record for the maximum Tc has always been held by members of the A15 family.
All metallic materials being superconducting in the range
from 14 K to 23 K have the crystal structure of the A15s. The
high superconducting transition temperature Tc for V3Si and
Nb3Sn has been discovered by Hardy and Hulm (5) in 1953
and Matthias et al. (6) in 1954. A15 materials are expected
not to be superconducting above 25 K (7), due to the increasing instability of their structure related to the electron
phonon interaction.
Nevertheless, there are compounds showing the A15 structure which do not have a remarkably high Tc. For example,
the intermetallic compound containing Nb and osmium (Os),
that is, Nb3Os, has rather a low one: Tc 1 K (3), whereas
the compound of V with cobalt (Co), that is, V3Co, does not
show superconductivity at all, down to a temperature of T
0.015 K (8). That fact made the search for an explanation for
the qualities of those materials even more difficult.
In general, it is possible to describe the physical phenomena relevant for the A15s, for example, the high superconducting transition temperature, which is also related to the
BCS formula (9), by one-dimensionality, partially localized
states near to or at the Fermi level, elastic softness, and
strong electronphonon coupling.
In the A15s, a nonatomic diffusion, characterized by an
atomic motion during the transformation at low temperatures, not exceeding the distance of an unit cells size, takes
place. Such martensitic transformations are quite common in
solids. The most famous example is an alloy of iron (Fe) and
carbon (C), the FeC transformation to the Fe3C, the bcc-iron
or martensite. The martensitic transformation in A15, above
Tm (martensitic phase transition temperature) is from cubic
A15 to a more tetragonal structure at a few tenths of a kelvin
below this temperature. This transformation has been observed first at V3Si and Nb3Sn by Batterman and Barrett (10)
by x-ray diffraction.
Further investigations have been made by susceptibility
and NMR measurement of Knight Shift and nuclear spin lattice relaxation rate (11). Results are that, in the superconducting state, the A15s have the behavior of type-II-superconductors. It is remarkable that, the higher the transition
temperature, the higher the temperature dependency of the
susceptibility becomes.
The development of technically usable A15s was correlated
to the research work done for conductors having a high critical temperature Tc, and being adequate for the production of
high-field magnets.
Early progress for commercial use has been achieved with
powder metallurgical systems, by filling Nb and Sn powder
into a Nb tube, compacting the powder, and drawing the entire piece to the final wire diameter. Other methods targeting
the production of tapes, which was possible by passing tapes
of the substrate material, for example, V or Nb, through a
bath of molten Ga or Sn, respectively. Additional experiments
Table 1. Transition Temperatures [in K] of Some A15 Superconductors (24)

A15
Cr3Si
Mo3Ir
Nb3Al
Nb3Au
Nb3Ga
Nb3Ge
Nb3Sn
Ta3Sn
Ti3Ir
V3Al
V3Ga
V3Ge
V3In
V3Si
Tc
1.2
8.5
19.1
11.3
20.7
23.2
18.2
5.8
4.3
9.6
14.8
14
17.1
60
used the same principal approach, by producing pancake-like

tapes, made by rolling together already alloyed tapes. Chemical vapor deposition (CVD) and cathodic sputtering are being
used, as well.
The advantage of A15 conductors at high Tc and high Bc2,
the upper critical field (To be precise, the upper critical field
would be Hc2, which is related to the upper critical flux density Bc2 in the following way: Bc2 0Hc2. But the common
combination used in the superconducting society is the one
above.) is faced by the disadvantage of the brittle intermetallic A15 compounds. That applies for conductors having reduced mechanical tension in the diffusion layer. In the case
of tapes this is given due to the smallness of the distance of
the neutral phase to the diffusion layer, resulting in bending
stresses low enough to wind magnets after the A15 forming
heat treatments.
Tapes, having naturally a large width-to-thickness ratio,
are unsuitable for rapidly excited magnets. The A15 layer
causes instability because of flux jumps originating from the
currents perpendicular to the superconducting layer. Fulfilling the demand of intrinsic stability is one of the important
criteria for modern superconducting magnets, and can only
be achieved by dividing the superconductive parts into very
small portions.
First ideas for the production of fine filaments in a substrate have been laid down in the British patent GB No.
1203292 of 1966 (12). Nevertheless, the advantages of the
properties of tapes (Nb3Sn; V3Ga) for the design and construction of pancake coil magnets has been used on several highfield magnets. A magnet of 17.5 T (13.5 T with Nb3Sn plus 4.0
T with V3Ga), being constructed by Intermagnetics General
Corporation (IGC), has been installed at the National Research Institute for Metals (NRIM) in Japan in 1975 (13).
In the case of fine filaments, materials in a ductile state
are drawn down to final dimensions. The A15 compound is
then formed by a heat treatment, using the solid-state diffusion. This method of solid-state diffusion has first been applied by Tachikawa for a V3Ga wire (14). The A15 compound
is made via a heat treatment at approximately 700C, which
enables the Ga, solved in the Cu, to diffund into the V filaments, forming V3Ga.
As the problem of brittleness still exists, it has now become
common to wind the magnets coils first, having the treatment
afterwards. This so-called wind-and-react method is now used
for almost all magnets assembled of A15 superconductors.
The structures of V3Ga and Nb3Sn are very similar, but material costs and production procedures are in favor of Nb3Sn. It
is possible to perform the solid-state diffusion by several
methods. Examples are diffusion of Sn from a bronze with a
high tin content into Nb filaments, or by an external diffusion
of Sn to a Cu-Nb system with intermediate steps of forming
a bronze by the Sn diffusion procedure. The bronze process is
applicable, for example, for Nb-Sn, V-Ga, V-Si, V-Ge, Nb-Au,
and Nb-Al.
In order to get high critical currents, it is obvious that the
reaction of Nb and Sn or V and Ga to Nb3Sn or V3Ga must be
as complete as possible. Different methods have been developed to reach this target, for example, Nb-Sn internal Sn conductors, powder-in-tube, jelly roll, or in situ technologies.
Such methods are necessary, as otherwise the amount of
Nb3Sn in the cross section is limited to the Sn content of the
bronze. Compared with conductors produced via the bronze
route, these materials are superior, in terms of the critical

current density Jc. Forced by design and fabrication technology, those conductors tend to have filament coupling and high
hysteresis losses, caused by relatively large effective filament
diameters. These values of the effective filament diameters
are significantly better for the bronze route conductors, also
showing higher figures for Tc and Bc2.
Nb3Al is one of the about 47 known A15 superconductors,
having the third highest Tc of 19.1 K (see also Table 1), ranking behind only impractical materials like Nb3Ge (23 K) and
Nb3Ga (20.3 K). While having those promising figures, the
producibility of Nb3Al turned out to be a difficult task. Different production methods of powder metallurgy have been utilized, as well as the tube and the jelly roll processes.
A15 superconductors are strain sensitive and, while
Nb3Sn may drop about 50% in Jc at a strain of 0.5%, the Jc
of Nb3Al is reduced by 20% only at a strain up to 0.8% (15).
That gave reason for renewed interest in that material, especially for large-scale applications like the International Thermonuclear Experimental Reactor (ITER). For the very big
coils of such a machine, the conductors are cabled and welded
into a stainless-steel conduit, resulting in additional axial
strain.
A significant amount of research work has been done toward powder metallurgical solutions, but no commercially usable process has been developed for Nb3Al. The jelly roll and
Nb tube processes seem to be more promising to yield high
Jc values at reasonable length. Wires of A15 type superconductors used for the wind-and-react technique need to be insulated with an insulation material like glass, quartz, or ceramic, withstanding the heat-treatment temperature. In the
case of Nb3Sn, this temperature is at approximately 700C,
leading to a braided insulation type of E- or S-glass. The
higher temperatures, which are necessary to form the Nb3Al
phase, are another obstacle for a broader use of this material.
While Nb3Sn became the main choice of the A15 materials
being used in a wide section of high-field magnets, a development leading to higher critical currents and better mechanical values is necessary.
The main factor that is limiting high-magnetic field is the
Lorentz force, which increases proportionally to the square of
the generated magnetic field. The stresses induced into the
conductor and the strains require reinforcement, reducing the
overall current density and thus leading to magnets being
bigger and, therefore, less efficient. The bronze substrate of
A15 superconductors is highly resistive and the stabilization
is poor. Cu of very good conductivity has to be added [compare
Fig. 2(a) with Fig. 2(b)]. Different methods, different with respect to the method of processing the conductors, like internal
Sn, jelly roll, and bronze process, are used. In order to prevent
the diffusion of B ions, for example, Sn into the Cu, diffusion
barriers are necessary. In case of the bronze process conductors, diffusion barriers of Ta or Nb are used.
To influence the superconducting parameters, the critical
current density Jc and the upper critical field Bc2, by getting
better heat-treatment conditions and higher pinning forces,
third elements, creating ternary and quarternary alloys, for
example, (Nb, Ta)3Sn, (Nb, Ta, Ti)3Sn, or Ge, Hf, or Al are
added. As not only Jc but also the overall current, usually
denoted by Ic,overall, is important, the size of conductors has to
be increased. This is possible by producing larger monolithic
superconductors, mainly of rectangular shape. Another alter-
61
with already existing A15 formation is only possible with

complicated, not very practical methods, for example, extremely high hydrostatic pressure.
METHODS OF PRODUCTION
Surface Diffusion Process
In this process, a tape of V or Nb is dipped into a Ga or Sn
bath. The heat treatment then forms V3Ga or Nb3Sn. It is also
possible to hold the bath at the reaction temperature, forming
the A15 phase during immersion. To use the better flexibility
of tapes and getting the relatively thin layers of the brittle
A15 only, other methods, such as sputtering the V or Nb to
the core material by cathodic deposition or chemical vapor deposition, or by condensing under high vacuum conditions have
been applied. For special MRI magnets, tapes of (Nb, Zr)3Sn
are introduced, operating at 9 K (16).
Composite Diffusion Process
Members of the composite are cold-worked together to final
shape. The final heat treatment forms, by solid-state diffusion, the superconducting phase or a microstructure, which
also has normal conducting phases. Tapes or even wires made
of the components of the superconducting phase, probably in
combination with other elements for electrical or mechanical
stabilization, or just to improve the workability, are coldworked together, tapes of Nb and Al or Cu, Sn, and Nb being
stacked and rolled. It is also possible to use tapes of the highmelting component with layers of the lower-melting component received by CVD for stacking and rolling. Clad chip extrusion (17), by producing the components as described, chipping and filling them into a Nb-lined Cu can for extrusion and
drawing, is another development of this production.
Powder Metallurgy Process
Figure 2. Different cross sections of Nb3Sn bronze conductors; (a)

with TaCu core for stabilization; and (b) unstabilized (courtesy of Vacuumschmelze).
native is the cabling of a number of round wires into flat or

round cables. Special attention to possible reasons for the reduction of Jc by filament size, the principle conductor design,
or the cabling process, has to be given.
To fulfill basic requirements of technically reliable superconductors like high transition temperature Tc, high critical
field Bc2, and high critical current density Jc, many different
approaches have been used. This was mainly done by developing production and material treatment methods adequate
to overcome the sometimes difficult material parameters of
the A15 components and in order to optimize the parameters
of the final conductor configurations. As all A15 materials are
brittle by nature, which cannot be overcome by any means, it
is necessary to use methods of deformation to get tapes or
wires being cold-worked prior to forming the A15 layers. Once
the layers are formed, the handling of the conductors needs
utmost care, in order not to reduce or completely to destroy
the superconducting properties. Deformation of conductors
Matrix material like Cu is mixed with other components, for

example, Nb and Sn, or Nb and Al, after the compaction to
final shape of wires or tapes. The formation of the superconducting phase takes place on the outside of the powder layers.
Producing a superconducting wire by PM methods has been
approved by Kunzler et al. (18) and was the first successful
process for A15 wires. The Nb tube has been filled with (Nb,
Sn) powder with a ratio of 3 to 1 or a (Nb3Sn, Sn) powder.
Heat-treatment temperatures of 970C to 1400C are used.
More recently, powder of the intermetallic compound NbSn2,
mixed with Sn powder, is filled into a thin-walled tube of Cu
or Cu-based alloy, and reduced in size for compaction. The
powder is then placed into a Nb tube within a Cu tube. After
that, it is possible to use this combination and stack it again
into a Cu can with other matrix parts. Such a billet can be
cold-worked to final size without further heat treatment. During the heat treatment, at typically 650C to 700C during
15 h to 45 h, a Nb3Sn layer of about 2.5 m thickness is
formed on the inner side of the Nb tube, resulting in filament
diameters of 10 m to 20 m, depending on the number of
cores. The reaction process is, in fact, a two-step process of
NbSn2 Sn Nb6Sn5 from the core, and then forming with
the Nb tube and, by depleting, of Sn the Nb3Sn. Similar methods can be used for Nb3Al, especially by using sintered rods
of Nb powder, metal-impregnated to get it self-supporting,
62

Table 2. Increase of Hardness of Niobium and Tin Bronze During Deformation (20)
Sample
7.7 at. % Sn
8.5 at. % Sn
Nb, original
Nb Core
Degree of
Deformation
(%)
Heat-Treatment
Time (h)
Temperature
(C)
Hardness
(N/mm2)
0
21
27
21
21
21
0
21
21
21
21
0
25
57
68
75
97
99.7
1
5
8
1
5
8
1
1
1
500
500
500
400
400
400
550
550
550
1260
2230
2330
1340
1160
1140
1350
2560
1640
1370
1330
1020
1330
1480
1460
1530
1510
2030
due to immersion into Al-Ge or Al-Si baths at temperatures

of 600C or 580C, respectively (19). Reaction treatment at
1700C is followed by final annealing, at 750C.
Bronze Process
This technology is very typical for Nb3Sn and V3Ga superconductors. In a substrate of Cu-Sn or Cu-Ga, a number of filaments of Nb or V are inserted. The solubility of Sn and Ga is
limited to approximately 8 at.% or 20 at.%, respectively. This
solubility restrains the amount of the A15 compound that can
be formed by solid-state diffusion. The high-strength component, the bronze, is getting the low-strength component, NbGa, to a laminated flow, but multiple intermediate heat treatments due to work hardening are necessary; see, for example,
Table 2 (20). Relatively thin filaments, with a maximum diameter of about 5 m, are necessary to receive good overall
current density. The Sn reservoir and the diffusion path is
the bronze between the filaments, but the decreasing content
of Sn during the reaction treatment reduces the speed of diffusion. Heat treatments at about 700C, in many different
variations, form the intermetallic A15 compound at the interlayer of the bronze to Nb or V. This also creates Kirkendall
voids (21), which disturb the diffusion paths; see Fig. 8. To
get high filament numbers for conductors of bigger size and
due to the limitations in filament diameter, second and third
stacking of billets may be necessary. For electrical stabilization, Cu can be embedded in the matrix or in the center of
the conductor. An outer shell of Cu is also possible, especially
in the case where a larger amount of Cu is needed; see Fig.
3. Diffusion barriers of Ta, Nb, or V are necessary, to protect
the Cu during the diffusion process by preventing B ions entering from the bronze into the Cu, which would significantly
reduce the conductivity.
Internal Sn Process
A Cu jacket, having a Sn rod in the center and Nb cores distributed around it in the remaining cross section, is the start-
ing element of this process. Those subelements may be surrounded by Nb or Ta barriers, to protect the stabilizing Cu
from diffusion of the Sn. This Cu is the outer tube into which
the subelements are inserted. The Cu-Nb-Sn composite is
cold-worked to final dimension, without intermediate annealing. The bronze matrix is formed when the conductor already
is in its final shape. Therefore, the heat treatment needs two
cycles. The first typically starts at 200C for 100 h, continues
at 375C for 24 h, and finishes at 580C for 50 h. It is needed
for the homogenization of the Cu-Sn bronze, containing different phases according to the phase diagram in Fig. 4 (23). A
consecutive second heat treatment of one or more steps is necessary to form the Nb3Sn layer. The typical temperature is
around 725C. For a better distribution of the Sn, methods of
placing it rather close to the filaments instead, as a core in
the center of the subelements, is used, resulting in reduced
heat-treatment time and a homogeneous A15 layer distribution over the cross section of the conductor (24).
In Situ Process
Casting ingots of two-phase Cu-Nb or Cu-V with dendrites
dispersed into the matrix of Cu, cold-working, and then coating with Sn or Ga, are the basic steps to start this process
[Fig. 5(a)]. To reduce the likelihood of wire-breakage, a bronze
matrix may be used. After dispersing the Sn or Ga throughout
the matrix, Nb3Sn or V3Ga are formed during the heat treatment at the interfaces between the matrix and the filaments.
Highly homogeneous Cu-V ingots are difficult to produce by
conventional casting methods, due to a large miscibility gap
in the liquid region. A continuous arc casting plus Ga coating
give the opportunity to optimize the material toward better
mechanical values and better current-carrying capacity.
Large ingots up to 150 mm in diameter [Fig. 5(b)] have been
produced by smelting the Cu-Nb into a CaO mold for solidification. Wires produced thereof have shown, after heat treatment at 800C for 25 h, good mechanical values in untwisted
condition. No degradation is found up to 1.2% strain, but in
Preconductor billet
Final preconductor
Final conductor billet
Extrusion
drawing
annealing
Bundling
Bundling
Extrusion
drawing
annealing
Nb/CuSn core
63
CuSn tube
Cu tube + Ta barrier
Final conductor
Figure 3. Fabrication of bronze route conductors (courtesy of Vacuumschmelze).
10
1300
20
30
Cu-Sn-Concentration
70
50
60
40
20
900
80 %Gew
25
30
90
35
40
798
15.5 (25.5)
13.1
755
(22.0)
700
676
640
640
600
586
582
590
14.9 (24.6)
520
500
16.5 (27.0)
800
1100
1083
1000
900
Temperature
800
798
7.7
(13.5)
19.1 (30.6)
700
400
640
43.1 (58.6)
600
500
586
9.1
520
9.1
(15.8)
300
10
45.5 (59.0)
350
20
25
30
86.7 (92.4)
415
6.2 (11.0)
45.5 (60.9)
227
232
189 44.8 (60.3)
200
0.7 (1.3)
100
15
400
300
350 20.5 (32.55)
0
Cu
10
20
30
40
50
60
Cu-Sn-Concentration
186
70
98.7 (99.3)
80
90 %At. 100
Sn
Figure 4. Detailed phase diagram of CuSn (23).
64
Nb
Cu
;;
Casting
Drawing
Sn planting
(a)
(b)
Figure 5. (a) Schematic diagram of in situ process of Cu-Nb composite (courtesy of Fujikura).
(b) Transverse section showing Nb dendrite solidified in Cu mold (courtesy of Fujikura).
the case of twisted wires, the degradation is significant (25a).

Multistage cables made from fine wires with a diameter of 0.2
mm with small in situ filaments of 0.5 m have been developed for ac applications. Such cables are showing even more
improved values in strain resistance (25b).
External Diffusion Process
This process is also typically used for V3Ga and Nb3Sn composites. In a Cu billet with drilled holes, rods of V or Nb are
embedded and cold work is applied to get the final dimensions. Prior to heat treatment, the surface is coated with Ga
or Sn having first the diffusion into the matrix, and then
forms at the interface between matrix and the cores the
V3Ga or Nb3Sn layers, respectively. The amount of A15 phase
produced by this method is limited by the quantity of the B
material which can be stored in the coating of the Cu-A wire.
This fact dictates the maximum diameter of the wire. The
combination of this method with the bronze process, in order
to increase the tin content, is also possible.
Jelly Roll Process
A foil of Nb with slit meshes and a foil of Cu-Sn bronze are
spirally rolled into a cylinder, cold-worked, and heat-treated.
The Nb3Sn is received at the interface of the Cu-Sn and the
Nb. Despite the successful use of the jelly roll process for the
Nb3Sn, this method was originally developed for Nb3Al (26).
Foils of Nb and Al have been wrapped around one Cu cylinder
and inserted into another Cu cylinder. Wires with a diameter

of 0.2 mm receive a heat treatment of several hours at about
850C, resulting in a mixture of Nb3Al and nonreacted Nb.
Placing some, for example, nine, of the unreacted jelly rolls
into a Cu rod with drilled bores, is a logical way to increase
the current by getting a larger cross section. The modified
jelly roll technology is performed (27) by wrapping layers of
Cu sheets and expanded metal sheets of Nb around a Sn rod.
Conductor designs with 54 of such jelly rolls assembled with
a barrier of Nb or V are placed in a Cu can for cold-working
to final dimensions, providing a high Jc and large Cu crosssection, likewise. In principal, the modified jelly roll process
proceeds like the internal Sn process.
Tube Process
To form the basic composite, a Cu tube with a Sn or Cu-Sn
alloy core is inserted into a Nb tube. This combination is introduced into another Cu tube and then cold-drawn. A number of those single-core elements are stacked for assembling
into a Cu can and cold-worked to final dimension. For the
production of basic composites of larger conductors, cold working is performed down only to intermediate dimensions, followed by a second assembly. During heat treatment, the Sn
diffusion into the matrix takes place first. Then the Nb3Sn
layer is formed at the interface of the Cu to the inner side of
the Nb tubes. This technology has also been used for V3Ga
wires and tapes, rolled from wires (28).
21.5
T
[C]
2000
Wire speed: 1 m/s

206010C
A2
1940C
251
Heating
current
Supply reel
Winding-up
reel
Analyzing
recorder
Electrode
pulley
Heating voltage
10
cm
22.50.5
65
1500
A2+A15
A15+
A15
In vacuum
2105 Torr
Ga bath
Coolant
Electrode
Power source
dc constant
voltage
Figure 8. Schematic diagram of rapid-heating/rapid-quenching apparatus (30).
Nb Al
1000
17
20
25
at % Al
Figure 6. Detailed phase diagram Nb-Al (29).
Specialties of the Production of Nb3Al

The A15 phase in Nb3Al can be attained from the melting
bath using high temperatures, and at reduced temperatures
for the reaction between Nb and Nb2Al. The -phase of
Nb2Al needs temperatures above 1600C. By reacting with Nb
at about 1800C, A15 is formed. The desired stoichiometry of
the A15 phase seems to be stable at high temperatures of
1770C only (1940C in the corner of the phase diagram of
Fig. 6 (29). In general, the high reaction temperature for the
conventionally processed materials leads to grain growth. Be-
cause many production techniques failed to get stoichiometric Nb3Al, numbers of different approaches have been tried,
like laser alloying or rapid quenching by melt spinning. Powder metallurgical processes are unfavorable, because of the
high oxygen (O) content in the powder, which does not allow
the high deformation rate required. To get reliable conductor
lengths, especially for magnets with magnetic fields above 21
T, a distinct rapid-quench process has been established (30).
By the jelly roll technology, the wire consisting of several elements made from Nb and Al sheets wrapped around an Nb
core is produced, using multistacking and extrusion procedures, resulting in a cross-section pattern, as can be seen in
Fig. 7. The wire produced this way is subject to ohmic heating
at 1900C to 2000C, and rapidly but continuously quenched
in a bath of molten Ga (melting point: 30C); see Fig. 8. Such
a treatment leads to a supersaturated Nb-Al-bcc phase. This
metastable phase is transformed by a heat treatment of 700C
to 900C to microcrystalline A15 of nearly stoichiometric composition. The resistivity ratio RRR in the Nb matrix has a
value of about 17, giving reduced concern about bridging of
filaments in this kind of conductor. Due to the processing, the
wire consists of Nb-Al only, with access of Nb. For the use at
a higher current J, Cu has to be clad, in order to stabilize
the wires.
BRONZE CONDUCTORS
Figure 7. Cross section of Nb/Al composite wire (30).
To fabricate high-field magnets, flexible tapes, having the advantage of a small distance from the brittle A15 compound to
the neutral phase in bending direction, have been used successfully. The large area-to-thickness ratio of the A15 layer
leads to instability (flux jumps), especially if magnets have a
rapid ramping rate. The solid-state diffusion process, as used
for bronze conductors of Nb3Sn and V3Ga (31), has solved this
problem, by dividing the core material into plenty of fine filaments. The formation of A15 layers is principally limited by
the amount of Sn and Ga in the bronze. The solubility of Sn
in Cu is 8.5 at.% and for Ga in Cu is 20 at.%. Bronze with
about 7.5 at.% Sn or about 18 at.% Ga has been used. The
66

Table 3. Maximum Area Ratio, Remaining Tin Concentration, and Transition
Temperature of Solid-State-Diffused Nb3Sn Samples (20)
Tin Concentration
in Alloy (at. %)
7.7
8.5
b
Maximum
Area Ratio
ANb3Sn /ACu(Sn)
Remaining Tin
Concentration
(at. % (calculated)]
Transition
Temperature
Tc (K)
700
730
750
750
0.15
0.18
0.26
0.30
5.1
4.6
3.3
3.4
17.5 a
diffusion process forms the A15 layer until the equilibrium

for a given temperature is established. At a temperature of
620C to 700C, the diffusion for Ga ends at a remaining concentration at 14 at.% to 15 at.% Ga in the matrix. Sn diffusion
from the bronze proceeds at approximately 700C to 850C,
leaving a Sn concentration in the bronze of approximately 3
at.% to 4 at.%; see Table 3.
The heat-treatment time and temperature has to be controlled in such a way as to receive an optimum layer thickness, but without increasing too much the grain size. Especially for a long heat treatment of 200 h, the matrix volume
has to be increased, in order to provide enough Sn or Ga.
Small distances between the filaments seem to be desirable,
due to reduced bending strain, but the space between the filaments acts as a diffusion path for the B ions from the conductor periphery, too. Those diffusion paths are reduced in
their effective width by the Kirkendall voids (see Fig. 9) (32)
caused by the diffusion mechanism during heat treatment. At
a given temperature and a constant concentration gap, the
quantity of B ions diffusing through a cross section in a given
time is proportional to the area of this cross section (Ficks
first law). From this follows that the cross section of the cores
of Nb or V should be divided in as many portions as feasible,
to increase the interlayer between the bronze and the core
material. This leads to an increase of the total amount of A15
material, even with reduced heat treatment time. Optimization studies of diffusion treatment versus layer thickness
have shown that filament diameters should be in the range of
3 m to 5 m. For conductors with a diameter of 1.5 mm, and
taking into account the cross section needed for stabilization
and diffusion barrier, approximately 15,000 filaments are necessary. Workability of the component is an essential request
Bronze
Nb
Nb3Sn
18.1 b
18.04 b
Heat-treatment time: 24 h.
Heat-treatment time: 66 h.
Direction of
diffusion
Kirkendall
void
to arrive with technically and commercially usable conductors. While the basic components, Nb or V, electron beam- or
arc-melted, are high-purity materials of excellent ductility,
they are sensitive to imbrittlement by interstitials of oxygen
(O), nitrogen (N), or C. This is especially true for the V, but
the more problematic part is the bronze; see Table 2. Normally Cu-Sn bronze contains about 10 wt.% Sn, and as desoxidizer phosphorus (P) is used. In the Nb-Sn system, P prevents the diffusion procedure. The amount of Sn should be as
close as possible to the solubility limit of 8.5 at.%. For many
years, the technically attainable Sn content was limited to
about 8 at.%. Newer processes made homogeneous bronze at
8.5 at.% Sn available (33). The positive influence of the Sn
content on Jc is shown in Fig. 10. During cold work, the hardness of the bronze is increasing rapidly, as shown in Table 2,
and a significant number of intermediate heat treatments
have to be applied. From the workability point of view, this is
certainly a disadvantage of the bronze process.
STABILIZATION AND BARRIERS
To have maximum Jc, the core material has to be converted
into A15, theoretically to 100%. Regarding mechanical and
electrical stability, a small core of unreacted material, even
in fine filaments, has a positive influence on the overall per-
1500
Matrix:
Cu-15 % Sn-0.3 % Ti
Non Cu Jc (A/mm2)
Diffusion
Temperature
(C)
650C 100 h
1000
670C 100 h
Matrix:
Cu-14 % Sn-0.3 % Ti
500
T = 4.2K
Ec = 10 V/m
0
Figure 9. Tin diffusion in the system tin-bronze with Nb filaments

showing the narrowing of the diffusion paths by the Kirkendall voids;
after (32).
10
12
13
14
11
Applied magnetic field (T)
15
16
Figure 10. Diagram of non-Cu Jc versus applied magnetic field H

showing the enhancement of Jc by the Sn concentration (30).

Magnetization per total sample volume
Magnetization (kA/m)
400
200
0
200
400
3
0
1
Applied field (T)
(a)
200
Peak due to Nb
Magnetization (kA/m)
150
100
50
0
50
100
150
200
3
Ta barrier
Nb backed Ta barrier
2
0
1
Applied field (T)
(b)
Figure 11. Magnetization curve for material; (a) containing a 100%

Nb barrier: The relatively large flux jumps in the low field region are
very apparent; and (b) containing an interrupted Nb barrier: No flux
jumps in the low field region are observed (24).
formance of the conductor. For reasons of electrical and thermal stabilization, for example, during the occurrence of a
quench, which may result from wire movement in connection
with energy dissipation, it is necessary to have a highly conductive material in the cross section. The bronze itself has a
rather low conductivity (specific resistance 70 nm at a
temperature of 4 K), which is important to reduce the alternating current (ac) losses. Experiments have shown that, in
case of bronze conductors, a reversed transport reaction can
lead to a very small (0.3 at.%) Sn content, compared with the
values of Table 3. In this low-Sn bronze, the value for the
specific resistivity is reduced by a factor of 10. This method is
not very practical and has, in view of ac losses, disadvantages.
It is therefore necessary to increase the electrical stability by
designing conductors with Cu included in the cross section.
This can be done by a few percent distributed throughout the
matrix, up to 20% in the wire center and up to a maximum
part of the cross section as an outer shell. The composite
needs a barrier to protect the Cu from the diffusion of B ions
into the stabilizing part, which would reduce the conductivity
of the Cu. Barrier materials fulfilling this task are V, Nb, Ta,
or alloys and combinations thereof (34). The use of a Nb barrier seems to be the natural choice, as it fits the material
parameters of the complete conductor. The Nb3Sn layer
formed at the interface of bronze and barrier acts just like a
large filament, bringing additional high ac losses. The effect
of the barrier materials on the hysteresis losses is shown in
Fig. 11(a) and Fig. 11(b). The use of Ta barriers avoids mag-
67
netic disturbances, even in cases where high-temperature

heat treatments are being used. Furthermore, Ta barriers are
effective as reinforcement due to their high Youngs modulus
and remarkably high strength at 4 K. Access of O must be
prevented during any heat treatment in the course of production and the diffusion procedure, as Ta interacts intensively
with O. Conductors with peripheral stabilization should have
25% (area) of Cu, in order not to get a Cu layer 10 m, as
O might penetrate into the Ta (32). Barriers of Ta penetrated
by O are likely to burst during the final diffusion treatment.
Further solutions might be the use of a Ta core, increasing
the yield strength significantly by the larger cross section of
Ta. Enlarging the Ta portion of the cross-section area to
10% results in yield strength Rp0.2 of 250 MPa. See Fig.
12(a) and Fig. 12(b) (35) for the characteristics and the cross
section of a high-yield-strength conductor.
Al may act as a stabilizing material due to its attractive
properties: low weight, high specific residual resistivity ratio
RRR (1000), high thermal conductivity, and low magnetoresistance. For practical, use, restrictions arise from the poor
mechanical values, not compatible with the other components
of bronze conductors. The melting point of Al is lower than
the reaction temperature. Thus, Al is only useable for reacted
conductors, giving the need for the react-and-wind technology,
which does not implement too high a strain in the conductor
during winding, but which is feasible for large magnets only.
Monolithic and cable conductors can be coextruded together
with high-purity Al and reinforcing elements of special steel
or cobalt-(Co)-based alloys (36). In case a high overall current
Ic is needed, electrical stabilization is possible by cabling of
unstabilized or stabilized wires together with Cu wires. It is
essential that those Cu wires are also protected by diffusion
barriers, which must have an outer Cu or Cu-Sn layer to
avoid O penetration into the Ta. An example of a cable with
additional stabilizing CuTaCu wires is shown in Fig. 13. Prereacted cables, soft-soldered to Cu clad tapes of Al, have been
used for special laboratory magnets. To build very large coils,
as necessary for Tokamak fusion technology like ITER, conductors with critical currents of about 100 kA at 12 T and
excellent mechanical values are needed. The present way to
fulfill such demands is to produce a round cable of many superconducting wires (ITER: 1000 wires) in a multicabling
process, embedded into a stainless-steel jacket. By this
method, good cooling conditions are received as well.
A direct relationship exists between Jc and ac losses: With
increasing Jc the ac losses are increasing as well. The most
important reason for such losses in internal-Sn and jelly roll
conductors is filament bridging during the reaction treatment
and barrier material like Nb. The increase in volume caused
by the diffusion of Sn is about 30% to 40% in the Nb3Sn layer,
while the outer wire dimension is practically unaffected. The
filament bridging results in huge effective filament sizes d*.
In powder-in-tube conductors, the filament size depends on
the grain size of the NbSn2 powder. As the forming of the
Nb3Sn layer in the tube process takes place at the interface
between the Cu and Nb tubes, large filaments are received.
Barriers of Nb behave like large filaments, too. Bronze-processed conductors have the lowest ac losses and smallest d*,
as bridging in the bronze matrix is negligible. Bridging is also
not a problem for Nb3Al conductors. The filament diameter of
conductors produced by the jelly roll technique is in the range
of 50 m. The specification of the ITER conductors is, in re-
68

600
Stress (MPa)
HYS
T = 4.2K
500
400
HJC
300
100
80
200
60
Figure 13. Superconducting cable consisting of eight bronze conductors with a TaCu core and eight additional stabilizing wires of CuTaCu (courtesy of Vacuumschmelze).
40
100
20
0
0.2
0.4
0.6
Strain (%)
0.8
0.05
1
0.1
1.2
(a)
gard to this fact, divided into two partsone of high Jc and

high ac losses (HP I), and the other with a low Jc and low ac
losses (HP II); see Table 4. A reduction of ac losses in internal-Sn conductors is possible, by reducing the Sn content, but
matching the properties of a bronze process conductor, like
the one shown in Fig. 14(a) will be hardly possible, see Fig.
14(b).
Degradation by Cabling
Figure 12. Stress/strain characteristics at a temperature of 4.2 K of

a high yield strength conductor compared with a high-Jc conductor
(35) and cross section of the high yield strength conductor with a Ta
core (courtesy of Kobe Steel).
The selection of the jacket material has to take into account

that almost no additional strain should be induced into the
conductor by the properties of the conduit. The critical current of A15 superconductors is depending on the stress strain
conditions. It is also necessary to consider the behavior of the
conduit material during the heat-treatment cycles (react-andwind) applied for reaction. While cabling seems to be a logical
approach to increase the current-carrying capacity, degradation during this process may happen, especially for Rutherford type, flat cables. The sensitivity to the deformation which
takes place at the edges of flat cables is in dependence to the
conductor design and making. Bronze conductors, with and
without stabilization, exhibit a degradation of only 5% and,
even as specially enhanced designs, not more than about 10%.
Other configurations like internal-Sn have shown a significantly higher degradation of the critical current Jc (37). Generally, cables of A15 are workable with bronze conductors or
with powder-in-tube conductors. In the case where cables
made from Nb3Al wires of the continuous-quench method are
applicable, internal-Sn conductors have potential for improvement (24). Jelly roll conductors of Nb3Sn exhibit high sensitivity to strain-induced damage and are thus not suitable (38).
For cables consisting of many single superconductor wires,
sintering of the wires during heat treatment has to be
avoided, and coupling losses must be reduced by the resistance between them. Cr plating of about 2 m thickness appears to be suitable to fulfill those topics. Problems with the
RRR of the stabilizing Cu may arise, due to the heat treatment. Thus measures have to be taken against this effect.
The Cu content of modified jelly roll and internal-Sn conductors is, in the basic design, larger than that of the bronze
conductors. Due to the high Sn content in these composites,
the Jc, in particular, the non-Cu Jc, taking into account the
cross-section area without stabilizing portions, is superior as
well. The Cu content can be varied up to above 60% (area).
However, a minimum of about 30% is needed, in order to have
69
Table 4. ITER Strands: Test Results and Specifications (38)

Technique
Internal
Tin
Internal
Tin
Internal
Tin
0.806
Ta Nb
123
9.9
220/175
340/96
650/180
0.802
Ta/Nb
3
9.3
185/120
340/72
650/200
partially
1.49
10.2
27.7
3.2
147
9.14
570
91
0.801
Ta
5
18.0
200/6
350/18
450/28
580/180
650/240
fully
1.59
9.31
24.9
2.6
130
9.04
780
136
fully
1.38
9.49
25.3
1.7
80
8.98
710
595
partially
1.61
9.31
24.6
2.3
213
9.01
680
599
3.1
1.3
5.9
6.3
Bronze
Bronze
Diameter (excluding Cr-layer) (mm)

Barrier
Thickness barrier (m)
Twist pitch (mm)
Heat-treatment temperature and time (C/h)
0.803
Ta
1015
8.8
570/220
650/175
0.802
Ta
68
18.4
650/240
Filament reacted
Cu/non-Cu ratio
Critical temperature T* at 13 T (K)
Upper critical field Bc2* (T)
Cr thickness (m)
RRR (resistivity at 273 K/resistivity at 20 K)
Overall strand density (g/cm3)
Non-Cu Jc at 12 T, 4.2 K, 0.1 V/cm (A/mm2)
Non-Cu hysteresis losses, 3 T cycle at 4.2 K (mJ/cm3)
n value at 12 T, 4.2 K, 0.1 V/cm
Coupling loss time constant (ms)
partially
1.49
10.2
28.3
2.1
150
9.33
550
94
0.62
an outer Cu shell, which is necessary for mechanical stability

during processing.
Stress and Strain
The mismatch of thermal expansion coefficients of the conductor components is creating compressive strain in the A15
layer during cooling. The difference from the heat-treatment
temperature to 4.2 K is approximately 1000 K. The influence
of strain on the Ic, values was presented first by Buehler and
Levingstein in 1965 (39). The compressive strain influences
the Tc and Bc2 values as well. The relative difference in linear
expansion, l/l, of separated bronze and Nb3Sn for T of 1000
K, is about 1.05% (32). Depending on the volume-ratio of
Nb3Sn to the bronze and for a conductor that has good bonding between its elements, the contraction is between 0.00%
and 1.05%. This is because the filaments are under tensile
stress and the bronze are under compressive stress. Measurements on Nb3Sn with removed bronze matrix show a Tc close
to the maximum 18.2 K. The values for prestressed conductors are reduced by about 1 K. Conductor designs which have
besides the matrix and the twisted filaments, stabilizing Cu,
diffusion barrier, or reinforcing components, for example, CuNb, are a rather complicated system. For example, regard the
following values: A conductor with 22% (area) Cu and 5% Ta
has a relative thermal contraction from room temperature
down to 4 K of 0.29%, whereas conductors with 33% Cu and
10% Ta have 0.26%.)
The critical values decrease by the compression but increase again under axial tension. Maximum Ic is gained at the
strain m, where the tensile force in the filament is reduced to
a minimum. The Youngs modulus in the filament area is in
the range of 130 GPa and for the bronze between 50 GPa to
80 GPa (40). The value for the bronze depends on the depletion of the Sn. By the diffusion process the Sn content is reduced and Kirkendall voids are created. These effects, the
high temperature for annealing and the length of the heat
treatment influencing the grain structure, are responsible for
HP I
(spec.)
0.81
HP II
(spec.)
0.81
10
10
2
100
2
100
700
600
20
500
200
20
a low yield strength in the bronze. Plastic deformation of the

bronze caused by tensile stress, due to mismatch of the thermal expansion coefficients (bronze: 16 106 K1; Nb/Nb3Sn:
7 106 K1), results in reduced differences in the thermal
contraction. Bare Nb3Sn shows breaks at strain of about 0.2%.
Prior to breaking, slip-steps with an angle of 45 can be observed. The precompression built up in the conductor has, as
a consequence, the drop of Ic, Bc2, and Tc. The same precompression acts as a mechanical reserve during elongation of the
conductor, though strain of 0.7% to 0.8% is possible for the
complete composite without irreversible Ic degradation (32).
This decreasing under compression and increasing under tension is reversible, as long as a critical strain in the filament
was not exceeded. If strain was too high, Ic would not recover
completely after the load was released, because of cracks in
the filaments; see Fig. 15.
The increased strain tolerance is an important reason for
the fact that bronze conductors can be used as technical superconducting wires. Consequently, a thicker layer of Nb3Sn
leaves a more Sn-depleted bronze with reduced Youngs modulus. Possible plastic deformation and the reduced Youngs
modulus are diminishing the compressive strain of the filaments and, therefore, the degradation of the critical values is
smaller. The mechanical reserve against axial stress is
smaller, too. The filaments themselves are less strain sensitive if not fully reacted. Besides the thermal contraction during the cooling from about 1000 K to 4 K axial stress is applied to the conductor by several manners. The force used for
winding a magnet is giving tension, while the bending force
gives rise to tensile and compressive strain above or below
the neutral wire axis. In the finished magnet, the Lorentz
force F J B leads to hoop stresses, related to the radius
of the winding. Therefore, J B r, where r is the
radius, leads to very large forces, which makes special reinforcement measures necessary in the magnet or the winding
package.
At the stress-compensated state, m, the upper magnetic
flux density B*c2, which is strain dependent but always below
70
Figure 15. Scanning electron microscope micrograph showing

Nb3Sn layers after strain of the conductor of 3% (courtesy Vacuumschmelze).
50
40
4.2 K
30
description of Ic /Icm concerning the dependency on both magnetic field and strain, is given by the strain scaling law (42):
oMs (mT)
20
10
Ic
=
Icm
0
10
20
30
mJ
non Cu Wh = 97
cm3
NST 4675 Ta A
40
50
4
0
B (T)
(b)
Bc2 ( )
Bc2m
n p
1 B/Bc2 ( )
1 B/Bc2m
In this formula n, p, and q are scaling parameters, which can

be found together with the values of B*c2m for Nb3Sn and
V3Ga in Table 5.
The mechanical behavior of V3Ga depends mainly on the
volume portion of V, while an increase of the Ga concentration reduces the tolerance with respect to mechanical loads.
Conductors fabricated by the in situ technology show considerably higher mechanical values than filamentary bronze pro-
Figure 14. (a) Cross section of an ITER conductor for the central
solenoid (HP II specification, i.e., low loss) with a Cu to non-Cu ratio
of 1.5 and 4675 (Nb, Ta) filaments. (courtesy of Vacuumschmelze); (b)
Magnetization curve of the conductor shown in Fig. 14(a), indicating
the low hysteresis losses of 97 mJ/cm3 (measurements: Vacuum
schmelze).
1.0
B*C2()/B*C2m
Bc2, can be calculated from the measured Ic by using Kramers

law (41).
At m, the axial strain of the conductor at which Ic,max is
reached is in the range of 0.3% to 0.7%, leading to intrinsic
strain 0 m. The maximum critical current is not correlated to the magnetic field. Nevertheless, the difference of Ic
at and Ic at m strongly depends on the applied magnetic
field. At 12 T, between the compressive states and m, the
difference in Ic is about 20% and at 16 T there is a factor of
about 2. The detailed effects of intrinsic strain on the critical
magnetic field Bc2 and the critical current density Jc can be
seen in Fig. 16 and Fig. 17. The influence of the other conductor components besides the filaments and the bronze, like stabilizers or diffusion barriers is, of course, not negligible. A
0.9
B*C2()/B*C2m = 1 a0u
0.8
0.7
Material
a(0<O)
a(0>O)
Nb3Sn
V3Ga
Nb-Ti
1.7
1.7
1.7
900
450
1250
650
23
0.8
0.4
0
0.4
Intrinsic strain, 0, percent
0.8
Figure 16. Upper critical field B*c2 as a function of intrinsic strain 0

(42).
71
1.0
Jc/Jcm
0.6
Nb3Sn
14
12
Scaling parameters
n=1
B*c2m = 21 T
p = 0.5
q = 2.0
u = 1.7
0.4
a=
0.2
0.4
6
8
10
12
14
Temperature = 4.2 K
16
0.2
Rapid-heating/
Rapid-quenching processed
Nb3Al
1.0
B=4T
16
900 (0<0)
1250 (0>0)
0
0.2
Intrinsic strain, percent
0.4
0.6
Figure 17. Relative critical current density Jc /Jcm as a function of

intrinsic strain 0 for different magnetic fields (42).
Normalized critical current, Ic/Icm
0.8
4T
6 8
10
0.9
0.8
0.7
Bronze
processed
(Nb, Ti)3Sn
0.6
0.5
12 T
4.2 K
0.0
cess conductors (43). There is not a clear Ic,max, but degradation is also smaller or completely recovered, respectively,
after the load has been released.
Additions of third elements influence the Ic, Bc2, and Tc values. Effects of strain can be seen as a function of B or Bc2.
Due to the increase of Bc2 by addition of Ti, Hf, or Ta to the
matrix or to the core material, the effect of strain on Bc2 is
reduced. Further influence is given by the growth rate of the
layer and, therefore, the remaining unreacted part of the core.
Wires of Nb3Al are less strain-sensitive; even with an intrinsic strain of 0.5% Ic is reduced only by approximately 10%, as
shown in Fig. 18.
The martensitic phase transition temperature Tm increases
with the compressive strain, showing an influence of the cubic-to-tetragonal distortion of the lattice and the degradation
of Tc and Jc (44). At the strain m with Jc having its maximum,
the Nb3Sn phase becomes cubic again. The effect of transverse
compression on Ic is similar to that of axial strain. An increase
of transverse strain t is leading to a small Ic enhancement.
At higher t, the Ic encounters a strong reduction. The sensitivity of Ic on transverse strain is higher than for axial strain
and the irreversible behavior for t starts at a level which is
about 25% of the comparable axial strain (45). Transverse
compression may occur in large magnet assemblies like Tokamaks, with each magnet having close and strong neighboring
magnets. Because of the high currents necessary in such magnets, conductors are likely to be cabled and enveloped in a
stainless-steel conduit. A mixture of compressive radial
stresses and transverse pressures is obtained. Especially
braiding procedures of the wires are leading to many crossover points with high stress concentrations. Tetragonal distortion is also caused by compression of the unit cell by radial
Table 5. Scaling Parameters for the Use with the Strain

Scaling Law (42)
Material
Nb3Sn
V3Ga
B*c2m
1
1.3
0.5
0.4
2
1.0
21
21
Ag-sheath
processed
Bi-2212
0.2
0.4
0.6
Intrinsic strain (%)
0.8
1.0
Figure 18. Critical current Ic versus intrinsic strain 0 for bronzeprocessed (Nb, Ti)3Sn wire, and a new Nb3Al wire processed by rapidheating/rapid-quenching. The strain sensitivity of the Nb3Al compared with the (Nb, Ti)3Sn (30).
stresses (45). Axial stresses are applied more frequently to

conductors, but in view of the higher sensitivity to t, the
transverse compression has to be carefully taken into account
for magnet engineering as well.
HEAT-TREATMENT PRINCIPLES AND CONDITIONS
The enhancement of the critical properties of practical superconductors depends on improvements in the composite. It is
further necessary to use production methods and diffusion
treatments that are optimized toward the required features
of magnets with high magnetic fields (20 T) and high field
homogeneity. Multiple, sometimes interacting measures are
necessary to increase and to stabilize the values of Jc and
Bc2. By far, not all effects in the many different A15 members
are understood. Most of the information, gained by research
work, is available for bronze conductors, especially for those
of the Nb3Sn system, including internal-Sn and jelly roll conductors.
The speed of the reaction of the material in the course of
the heat treatment, during which the A15 phase is formed, is
correlated to the quotients of the atomic radii. The characteristic diffusion speed responsible for the ordering of the A ions
to form the characteristic chains, depends on the radii of the
B atoms. The smaller these radii are, the faster the diffusion
process may occur (46). However, stoichiometric systems call
for limits in the heat treatment. To have high concentrations
of B ions, the solubility and workability of the components
have to be shifted toward their limits. The thickness of the
A15 layers formed per unit of time can be calculated by Ficks
diffusion law, applied at the interface of two diffusion layers.
The theoretical prediction for the amount of A15 phase that
72
high. For jelly roll and internal-Sn composites this can be

achieved to a good degree. The so-called bronze route is limited by solubility and workability reasons to about 15 wt.%
Sn. It is further necessary that the bronze be uniform, especially that the variation of the Sn concentration over the cross
section is small. Filament sizes and filament spacings have to
be watched, to leave sufficiently wide diffusion paths and to
avoid bridging of filaments, in order not to end up with large
effective filament diameters d*.
TERNARY ELEMENTS
Figure 19. Layers (thickness 1 m) of (Nb, Ta)3Sn around the

unreacted (Nb, Ta) core of a bronze conductor (courtesy of Vacuumschmelze).
is formed, depending of the time t, is a proportionality of t0.5.

Due to influences such as grain growth or decreasing B ion
concentration during the transition process, the kinetic parameters are reduced in relation to the theoretical prediction.
Measurements of the speed of layer formation in Nb3Sn have
shown proportionalities to powers of the time t in the range
from 0.30 to 0.35 (32). In case where ternary or quarternary
alloys are utilized, the mechanism of the diffusion process is
similar, but the formation speed is influenced.
The pinning of fluxoids penetrating into type-II-superconductors needs structural imperfections. Such imperfections
may be grain boundaries, lattice disturbances, grain morphology, impurities, or combinations thereof. As the coherence
length in Nb3Sn is only about 3.5 nm, it is difficult to get a
complete picture of the interactions which are necessary, in
order to trap a fluxoid in the superconducting layers. In general, the increase of Ic depends on the heat-treatment temperature and time. Additional time and/or a higher temperature
lead to a larger layer thickness. In Fig. 19, filaments with an
A15 layer and an unreacted core of Nb in a bronze process
conductor have been prepared so as to visualize the layer
thickness of about 1 m. Nevertheless, Jc in the layer may be
decreased by grain growth, because in the intermediate field
range, the maximum pinning force is related to the average
grain size and the grain size distribution in the A15 layer.
Thus, the critical current Ic is inversely proportional to the
grain size and, therefore, proportional to the number of grain
boundaries per unit volume which act as pinning center.
Grain growth reduces the specific grain boundary area and
diminishes the amount of fast diffusion paths. Diffusion at
higher temperatures leads to faster layer formation and finer
grain, but by depleting the bronze of B ions, Kirkendall voids
are formed. This leads to a reduction in the diffusion rate. If
those voids are located at the bronze-to-layer interface, Ic degradation may occur due to their influence on the strain behavior. Therefore, the gradient of the concentration of Sn over
the cross section has to be taken into account for all heattreatment models. It is also important to have stoichiometric
conditions, that is, a Sn concentration that is sufficiently
The increase of Jc in the A15s is dominated at the intermediate field range by flux pinning at the grain boundaries. The
pinning force density Fp is equal to the product of Jc and the
corresponding magnetic flux density B(Fp Jc B). According to Kramers law, the pinning force shows saturation
in the high field region B*c2. Increasing Bc2 leads to an increasing Jc within the A15 layer. The value of Bc2 is dominated by
the normal state resistivity 0 and the critical temperature
Tc. As it is not easy to increase Tc remarkably, the main
means for varying Bc2 is given by the normal state resistivity
0, measured just above Tc or by the resistivity ratio. Such an
increase in 0 results in an increase of the GinzburgLandauparameter / , where is the penetration depth (47).
Because of the proportionality of the upper critical field Bc2 to
the GinzburgLandau parameter , the former is raised, too.
Raising 0 is possible by impurities, irregularities in the
chemical composition, causing microstructural defects. Bc2 is
not depending on the grain size, but its upper limit is determined by the susceptibility according to Paulis paramagnetic
effect (43). For that reason, the flux-pinning force and the
grain boundaries are not relevant if the magnetic field B is
close to Bc2.
Besides the more principal aspects, there are different
other reasons which are influencing the performance of practical A15 superconductors. Disturbances in the microstructure are originating from the production process, chemical
nonhomogeneities, or variations of filament diameters over
the length (sausageing). Nonuniformity of A15 layers due to
nonuniform distribution and supply of B ions is strongly influenced by the conductor design. Those macroscopic effects
are also observed for designs leading to irregular working and
deformation conditions due to the combination of materials
with quite different ductility, like Nb, Cu, and Sn. Microcracks occur in the layer itself, caused by thermal or handling
defects. Bronze matrix conductors need, because of work
hardening, intermediate heat treatments to preserve or to restore the ductility. Prereaction to a substantial degree may be
encountered. It leads to heterogeneous deforming conditions,
reduction of the Sn supply for the final diffusion treatment,
and mechanical defects in the conductors. Therefore, intermediate heat-treatment temperatures must be chosen carefully
and should not exceed 500C. Additionally, time has to be restricted.
To achieve better properties of A15 conductors, doping
with defined impurities like Zn, Mg, Fe, and Ni, and also
alloying with higher contents of Ti, Ta, or Ga has been performed. The stoichiometry in a ternary or a quartenary compound is a rather demanding field. The variety of metallurgical treatments like alloying, in combination with numbers of
NbTi
NbTi
Nb3Sn
Nb3Sn
(NbTa)3Sn
(NbTa)3Sn
(NbTaTi)3Sn
(NbTaTi)3Sn
jcnon-Cu (A/cm2)
106
105
4.2 K
2K
4.2 K
2K
4.2 K
2K
4.2 K
2K
104
0.1 V/cm
103
10 12 14 16 18 20 22 24 26
B (T)
73
The heat-treatment time and temperature is in interaction

with additional elements, or combinations thereof, responsible for the formation of the A15 layer. For wind-and-react
technology, it is indispensable to limit the temperature, due
to the insulation materials available. Practical glass-braid insulations for temperatures of up to 700C and 800C, respectively, are at hand. The temperaturetime combination is further determined by the application of conductors. Additions
of Ta, for example, Nb 7.5 wt.% Ta, reduces the formation
rate of the A15 phase (48), but increases Jc due to a longer
heat treatment. In the case of intermediate magnetic fields,
filament diameters should be smaller, leading, even with Ta
doping, to a relatively short heat-treatment time, to realize,
in this range of magnetic field, the required fine grain. Especially for conductors which are to be used at higher fields, the
Figure 20. Non-Cu critical current density Jc versus magnetic field

for Nb3Sn (undoped, doped with Ta, or Ta and Ti, respectively) at
temperatures of 4.2 K and 2 K (measurement: Vacuumschmelze).
different heat treatments, result in remarkable effects. Some

of the additives, for example, in the bronze matrix, are influencing the diffusion process like speed and grain refinement,
which enhances the performance of conductors in the intermediate field region. Others are more effective by increasing
Jc at high fields due to an increase of Bc2. There is not a complete correlation given between composites and increase of Jc
for the entire field range. This is not surprising at all, since
additives may be deposited at the grain boundaries. Therefore, crossover of Jc versus B for doped or undoped conductors
or for conductors doped with different additives is observed,
as shown in Fig. 20.
The different atomic radii of the additives in relation to
the basic alloy is also of influence. Ti with a smaller atomic
radius is incorporated more completely into the Nb lattice
than zirconium (Zr) or hafnium (Hf) with larger atomic radii.
The embedding of the Ti, originating from the matrix Cu-Sn
into Nb3Sn, is at a larger degree than for the Ti alloyed with
the Nb core. Elements like Ti, Zr, and Hf for alloying Nb result in A15 layers with fine grains. Mg in bronze, like Cu-Sn
or Cu-Ga, is increasing the formation rate of the A15 phase,
which causes grain refinement. The low Tc of V3Ge can be
raised from 6 K to about 10 K by adding 8 at.% Al (43). Ti in
Nb cores (about 2 at.%) is increasing the layer thickness. The
improvement of Jc in Nb3Sn by alloying Hf to the core and
additional Ga to the matrix is remarkable and leads also to
an increase of the irreversible strain d (43), but especially
alloying with Ga is difficult and impractical. V3Ga, which has
already superior high field values, improves further above 20
T, when adding Ga to the V core and Mg to the Cu-Ga matrix.
As Ti speeds up the diffusion rate of Sn in Nb, a doping with
0.3 wt.% Ti, in combination with high-Sn-bronze (15 wt.%)
shows improved values of non-Cu Jc (33), see Fig. 10. By introducing Ge into the matrix of a Nb3Sn composite, the thickness of the A15 layer is reduced significantly. This is most
probably related to the formation of an additional phase with
the Ge (Nb6Ge5) at the interface between core and matrix (22).
The grain size, however, is smaller, and so Jc in the layer
is enhanced.
Figure 21. Microscopic photographs of reacted layers in fractured

cross sections of samples heat treated at (a) 700C for 100 h; and (b)
750C for 150 h (courtesy of Kobe Steel).
74

10000
Critical current density, Jc (A/mm2)
Rapid-heating/
Rapid-quenching processed
Nb3Al
Figure 22. Externally stabilized (Nb, Ta)3Sn conductor with a Ta

diffusion barrier and outer Cu stabilization for high-resolution NMR
magnets (courtesy of Vacuumschmelze).
filament diameters are increased, as well as heat-treatment

time and temperature. Often, more than one different cycle is
executed, to improve the formation of the A15 layer, as not
too much attention should be paid to grain growth, in view of
the high field application. It is furthermore understood that,
for each grain size, pinning interaction may be different, and
flux linelattice spacing is reason for different numbers of
flux lines in each grain (49). The variability of the grain sizes
for different heat treatments, but also within one sample, becomes evident in looking at the microscopic photographs in
Fig. 21(a) and Fig. 21(b).
The necessity to fulfill the requirements of high critical
currents Ic at high magnetic fields Bc2 leads automatically to
large conductor cross-sections with a high number of filaments. Therefore, more than 100,000 filaments and cross-section areas of more than 6 mm2 are unavoidable. Figure 22 is
an example of such an externally stabilized Nb3Sn conductor.
For several years, the demand for high magnetic fields and
magnets with larger bores has been the driving force in the
development of A15 superconductors. While the first successful magnets were built from tape conductors (13), round or
rectangular wires became the more favorable solutions.
For magnetic field strengths of above 9 T in the center of
the magnet, materials with high Bc2, like conductors with A15
structure, are necessary. The study of polymers or macromolecules demands NMR systems of up to 1 GHz proton-resonance frequency (Larmor frequency); this corresponds to a
magnetic field of 23.5 T. Thus all NMR systems working at
frequencies of more than 400 MHz (corresponding to a magnetic field of 9.4 T) need other conductor material than NbTi.
At a temperature of 2 K and a field of up to 21 T, Nb3Sn still
shows reasonable Ic values. Other composites, like the rapidquenched Nb3Al, may perform even better according to Fig.
23. Besides the NMR applications, which are dominated by
bronze route conductors, diffusion technology is substantial
for conductors that are subject to a high magnetic field. For
big machines like ITER, more than 1,000 tons of A15 conductors will be necessary.
Laboratory scale solenoids with magnetic fields of up to 20
T or high field split coils take advantages of the improved
1000
Ag-sheath
processed
Bi-2212
100
Bronze processed
(Nb, Ti)3Sn
Nb-Ti
10
4.2 K
1
10
15
20
Magnetic field, B (T)
25
Figure 23. Non-Cu critical current density Jc versus magnetic field

B for a bronze-processed (Nb, Ti)3Sn and a Nb3Al rapid-heating/rapidquenching processed wire at a temperature of 4.2 K (30).
current densities Jc of, for example, internal-Sn, modified jelly

roll, or tube-processed wires. Special dipole and quadrupole
magnets for accelerators have to reach magnetic fields of
more than 11 T, which is only achievable with A15 conductors
like the internal Sn type shown in Fig. 24. Hybrid magnet
systems, made of a combination of superconductive and resistive magnets, are capable of reaching fields of up to 45 T
or 50 T. Not only the high critical fields Bc2, but also the superior values of the critical temperature Tc2, are further advantages of these materials, which can even be used, with refrigerator cooling, for magnetic fields of 5 T to 10 T. Tapes of
Nb3Sn with a Nb-1% Zr-core are used to build split-coils acting as open MRI systems, operating at 9 K (16).
Figure 24. Cross section of an internal tin conductor for use in highfield dipole magnets (courtesy IGC).
SUPERCONDUCTORS, METALLURGY OF DUCTILE ALLOYS
The results showing the improvements of the properties of

A15 superconductors have been obtained, to a considerable
degree, on an empirical basis. To force this technology forward, even more empirical work will be necessary. Fundamental tasks, like improvement of the composite, layer homogeneity, and grain morphology, will be inevitable for
controlling the microstructural influences in future designs.
Introducing artificial pinning centers (APC) and their possible
variations and controlled grain refinement will be another
route to be followed.
BIBLIOGRAPHY
1. H. Hartmann et al., Z. Anorg. Chem. 198: 116, 1931.
2. M. Weger and I. B. Goldberg, Some lattice and electronic properties of the -tungstens, Solid State Phys., 28.
3. W. Buckel, Supraleitung, Weinheim, Germany: VCH, 1984.
4. J. Muller, A15 type superconductors, Rep. Proc. Phys., 43: 1980.
5. G. F. Hardy and J. D. Hulm, Phys. Rev., 87: 884, 1953.
6. B. T. Matthias et al., Phys. Rev. 95: 1435, 1954.
7. W. Klose, Sommerschule fur Supraleitung, Pegnitz, Germany:
DPG, 1970, p. 14.
8. O. Henkel et al., Supraleitende Werkstoffe, Leipzig, Germany:
VEB Verlag fur Grundstoffindustrie, 1982.
9. J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Theory of superconductivity, Phys. Rev., 108: 1175, 1975.
10. B. W. Batterman and C. S. Barrett, Phys. Rev., 13: 390, 1964.
11. H. J. Williams and R. C. Sherwood, Bull. Amer. Phys. Soc. 5:
430, 1960.
12. British patent, GB.No.1203292, 1966.
13. W. D. Markiewicz et al., A 17.5 T superconducting concentric
Nb3Sn and V3Ga magnet system, IEEE Trans. Magn. MAG-13:
35, 1977.
14. K. Tachikawa and Y. Tanaka, Japan. J. Appl. Phys. 5: 834, 1966.
75
27. W. K. McDonald, Expanded metal containing wires and filaments,

US patent 4414428, 1983.
28. R. G. Sharma, Multifilamentary V3Ga wires and tapes with composite covers, Cryogenics, 25: 381, 1985.
29. J. L. Jorda and R. Fluckiger, Departement de Physique de la Matiere Condensee, Geneve, J. Less-Common Metals, 75: 227 (1980).
30. K. Inoue et al., New Nb3Al Multifilamentary Conductor and Its
Application to High-Field Superconducting Magnets, TsukubaShi, Ibarak: 305, Japan: RHMF, 1997.
31. K. Tachikawa, Proc. ICEC, 3, 1970.
32. H. Hillmann, Fabrication technology of superconducting material, Superconductor Material Science, New York: Plenum, 1981.
33. T. Miyazaki et al., Improvement of critical current density in the
bronze processed Nb3Sn superconductor, Paper CPB-7, CEC/
ICMC, 1997.
34. H. Krauth et al., Int. Workshop on High Magn. Fields, 1996.
35. Y. Miyazaki et al., Development of bronze processed Nb3Sn superconductors for 1 GHz NMR magnets, Paper CPB-6, CEC/ICMC,
1997.
36. M. Thoner et al., Aluminum stabilized Nb3Sn superconductors,
Adv. Cryog. Eng., 34: 507, 1987.
37. E. Wang et al., Evaluation of Nb3Sn superconductors for use in a
23.5 T NMR magnet, IEEE Trans. Magn., 30: 23442347, 1994.
38. H. G. Knoopers et al., Third round of the ITER strand benchmark
test, EUCAS conference applied superconductivity, no. 158, IOP
conference series, 1997.
39. E. Buehler and H. J. Levingstein, J. Appl. Phys., 36: 3856, 1965.
40. G. Rupp, The importance of being prestressed, in M. Suenaga
and A. F. Clark, eds., Filamentary A15 Superconductors, New
York: Plenum, 1980.
41. E. J. Kramer, J. Appl. Phys., 44: 1360, 1973.
42. J. W. Ekin, Stress/strain effects on critical current, Cryogenics,
35: S25S28, 1995.
43. K. Tachikawa, Recent developments in filamentary compound superconductors, Adv. Cryog. Eng., 28: 1982.
15. E. W. Collings, Processing of Nb3Al superconductors, Rep., Columbus, OH: Ohio State University, 1997.
44. R. W. Hoard et al., The effect of strain on the martensitic phase

transition in superconducting Nb3Sn, IEEE Trans. Magn., MAG17, 1981.
16. C. G. King et al., Flux jump stability in Nb3Sn tape, IEEE Trans.
Appl. Supercond., 7: 15241528, 1997.
45. W. Specking et al., Effect of transverse compression on Ic of

Nb3Sn multifilamentary wire, Adv. Cryog. Eng., 34: 569, 1988.
17. E. W. Collings, Recent advances in multifilamentary Nb3Al

strand processing, Rep., Columbus, OH: Ohio State University,
1997.
46. T. Luhmann and A. R. Sweedler, Phys. Lett., 58A: 355, 1976.

47. V. I. Ginzburg and L. L. Landau, Zurn. Ekspr. Teor. Fyz., 20:
1064, 1950.
18. J. F. Kunzler et al., Phys. Rev. Lett., 6: 89, 1961.
48. W. Specking, F. Weiss, and R. Flukiger, Effect of filament diameter and spacing on Jc of Nb3Sn wires in the intermediate filed
range and at high fields, IEEE Trans. Magn., 23: 11881191,
1987.
19. M. R. Pickus et al., Filamentary A15 Superconductors, New York:

Plenum, 1980, p. 331.
20. BMFT, Entwicklung von Hochfeldsupraleitern, BMFT-FB T,
1976.
21. D. Larballestier et al., Rutherford Lab. Report 74-135, IEEE
Trans. Magn. MAG-11: 247, 1970.
22. T. Pyon and E. Gregory, Some effects of matrix additions to internal tin processed multifilamentary Nb3Sn superconductors, IEEE
Trans. Appl. Supercond., 5: 17601763, 1995.
23. M. Hansen, Constitution of Binary Alloys, New York: McGrawHill, 1958, p. 634.
24. E. Gregory et al., Development of Nb3Sn wires made by the internal tin process, CEC/ICMC, AB-6, 1997.
25a. Y. Ikeno et al., Development of Nb3Sn superconducting wire using an in-situ processed large ingot, in Advances in Cryogenic
Engineering, vol. 36a, New York: Plenum, 1990.
25b. H. Fuji et al., Development of react and wind coils using in-situ
Nb3Sn wires for ac applications, MT15, 1997.
26. S. Ceresara et al., IEEE Trans. Magn. MAG-15: 639, 1979.
49. D. Rodrigues, Jr. et al., Flux pinning mechanisms in superconducting A15 materials and the optimization of their transport
properties, to be published in Adv. Cryog. Eng., 44: 1998.
REINHARD DIETRICH
Vacuumschmelze GmbH
SUPERCONDUCTORS, METALLURGY OF
DUCTILE ALLOYS
The most widely used superconducting materials are based
on NbTi alloys with Ti contents ranging from 46 wt. %
to 50 wt. % Ti. These alloys of Nb and Ti have both high
strength and ductility and can be processed to achieve high
critical current densities that make them ideal candidates
for magnets and applications. NbTi-based superconductors are commercially produced in long uniform lengths
and cost signicantly less to produce than other superconductors. The main drawbacks of this material are a low
critical temperature, typically requiring cooling by liquid
helium, and a low upper critical eld which limits the applied eld at which they can be used to below 12 T.
Although several other ductile superconductors have
been investigated (most importantly niobiumzirconium),
the niobiumtitanium alloy system has been the only ductile superconductor in use since the mid-1960s. The vast
majority of all superconducting magnets have been constructed from a small NbTi alloy range of 46 wt. % Ti
to 50 wt. % Ti (62 at. % Ti to 66 at. % Ti), with most of
the superconductor used being 47 wt. % Ti (63 at. % Ti).
It is the strength, ductility, critical current density, and
relatively low cost of this material that makes it stand
out as a commercial superconductor. The critical temperature, Tc , and upper critical eld, Hc2 , of these NbTi alloys
are unexceptional (Tc 9.3 K, Hc2 (4.2 K) 11 T, Hc2 (2
K) 14 T), but their very high critical current densities
(3000 A/mm2 at 5 T, 4.2 K) make them the ideal choice
for applications in magnetic elds up to 9 T at 4.2 K (liquid helium coolant) and 12 T at 1.9 K (superuid liquid
helium coolant). The primary applications of NbTi-based
magnets are magnetic resonance imaging (MRI), nuclear
magnetic resonance (NMR), particle accelerators, magnetic
connement for fusion reactors, ore separators, electrical
power conditioners, and superconducting magnetic energy
storage (SMES). Replacing some Nb with 15 wt. % Ta to
25 wt. % Ta further extends the upper critical eld when
cooled with superuid liquid helium but has yet to achieve
signicantly higher critical current densities. This article
will concentrate on Nb47 wt. % Ti and similar alloys because of their complete dominance as ductile superconductors. Collings (1, 2) has compiled extensive reviews of the
properties and applications of ductile Ti-based superconductors and has covered the history of their development
up to 1983 in an earlier volume (3).
Almost all NbTi strands are manufactured as a composite of NbTi laments in a Cu (or occasionally CuNi
or Al) matrix. The number of laments can vary from 50
to 5000, and their diameters can vary from 5 m to 50
m. The Cu matrix typically represents 50% to 70% of the
strand volume and provides the composite with thermal
and electrical stability. Subdividing the laments provides
adiabatic ux-jump stability and allows greater cold work
to be applied to the strand, which is benecial in achieving high critical currents. Commercial scale composites are
manufactured on a scale of 200 kg billets, and the resulting wire lengths can exceed 10 km at nal size without
breakage.
Figure 1. A transmission electron microscope image of the microstructure of a Nb47 wt. % Ti superconductor in transverse
cross section reveals a densely folded array of second phase pins
which are 1 nm to 4 nm in thickness. For comparison a schematic
illustration of the uxoid diameter (10 nm) and spacing (22 nm)
at 5 T and 4.2 K is superimposed on the top left-hand corner.
In type II superconductors, such as NbTi, high critical current densities in magnetic elds are only possible
if uxoid motion is inhibited. In NbTi, strong uxoid pinning is made possible by the creation of a nely dispersed
nanometer scale nonsuperconducting that closely matches
the uxoid spacing. The method of creating that pinning
microstructure distinguishes the two basic methods of producing NbTi strand. Figure 1 shows such a microstructure
achieved by heat-treating the strand during processing to
produce Ti-rich -Ti which fold into a densely packed array
of sheets when the NbTi wire is drawn to nal size. The Ti sheets are typically 1 nm to 4 nm thick with a separation
of 5 nm to 20 nm. The folding of the microstructure during
wire drawing is a result of the limited deformation orientations available in the body-centered cubic (BCC), crystal
structure of the NbTi grains. Almost all commercial NbTi
strands are fabricated using precipitation heat treatment
and are termed conventionally processed in order to distinguish them from the second and newer method of NbTi
production called Articial Pinning Center (APC). In the
APC process the pinning microstructure is engineered by
mechanically assembling rods or sheets of the component
materials at a size large. Extrusion and wire drawing reduces the assembled array to the nal nanometer dimensions. Like the conventionally processed strand, the deformation of the engineered microstructure is dominated
by the folding of the NbTi matrix grains, and the resulting APC microstructures are very similar in appearance to
those produced by conventional processing. The APC approach allows a large degree of freedom in the selection of
matrix and pinning materials as well as their ratios and
physical distribution. Consequently, APC strands outperform conventionally processed strands in elds up to 5 T
and promise greater performance at higher elds. The additional cost associated with assembling the pinning array
and reducing it to nanometer scale has limited its commercial application.
THE NBTI ALLOY SYSTEM
Stable Phases
The atomic volume difference between Ti and Nb is only
about 2%, resulting in a -isomorphous system where the
Superconductors, Metallurgy ofDuctile Alloys
Figure 2. The variation in Hc2 at 4.2 K (4), Tc (5) and resistivity

(6) with composition for single-phase NbTi. Hc2 is dened as the
linear extrapolation of the high-eld pinning force (Fp ) to zero.
phase has a BCC structure with a lattice parameter of

approximately 0.3285 nm. Figure 2 shows the variation
in Hc2 at 4.2 K (4), Tc (5) and resistivity, n (6) with composition for single-phase -NbTi. The upper critical eld
peaks sharply in the range of 40 wt. % Ti to 50 wt. % Ti
with a maximum value of 11.5 T at 44 wt. % Ti. The critical temperature drops continuously over this range with
increasing Ti content. The only other stable phase in this
system is the Ti-rich phase which has a hexagonal closepacked (HCP) structure and a composition of 1 at. % Nb to
2 at. % Nb. The low Nb content of the -Ti phase suggests
that -Ti precipitates should have a low Tc (approaching
the 0.39 K Tc of pure Ti) and should be nonsuperconducting
under practical operating conditions. The benecial role of
precipitating the normal-phase -Ti precipitates was rst
shown by Pfeiffer and Hillman in 1968 (7). The alpha phase
is only stable below 882 C (at atmospheric pressure); and
for the alloy composition range of interest, -Ti is only
stable below 570 to 600 C. In Figure 3 the widely used
high-temperature phase boundaries of Hansen et al. (8)
are combined with the calculated low temperature boundaries of Kaufman and Bernstein (9) modied by Moffat and
Kattner (10) to provide a composite equilibrium phase diagram that generally reects production experience. The
use of the calculated low-temperature phase boundaries is
a result of the difculty of achieving equilibrium at the low
temperatures (compared to the melting point) at which the
to + transformation occurs in NbTi. The interdiffusion rate in -NbTi decreases exponentially with decreasing Ti content, and it was calculated in 10 that at 500 C it
would take 10 years to reach the same condition in a 30 at.
% Ti alloy that it would take in 3 s in an 80 at. % Ti alloy. The
slow diffusion rates mean that in the composition range for
superconducting application the single phase can be retained even with relatively slow quenching from above /
+ boundary (600 to 650 C). The shape and position of the
/ + boundary is important because it determines the
maximum volume of -Ti precipitate that can be formed for
a given heat treatment temperature and alloy composition.
The calculated boundary indicates that both increasing the
Ti content and decreasing the heat treatment temperature
should increase the maximum volume of precipitate that
can be produced.
Figure 3. A hybrid equilibrium phase diagram for NbTi combining the experimentally determined high-temperature phase
boundaries of Hansen et al. (8) with the calculated lowtemperature phase boundaries of Kaufman and Bernstein (9) modied by Moffat and Kattner (10). Also shown is the martensite
transformation curve (Ms ) of Moffat and Larbalestier (11).
Metastable Phases
There are three metastable phases of importance: two
martensite ( and ) and an phase. The martensite
is HCP with lattice parameters identical to -Ti, and it
forms in alloys up 7 at. % niobium. The orthorhombic
is transitional between the HCP and the BCC phase,
and it forms at higher Nb concentrations. The martensite
transformation boundary of Moffat and Larbalestier (11)
is shown in Fig. 3 and shows that the most commonly used
NbTi alloys are outside the range of the martensite transformation. The phase has a hexagonal crystal structure
(c/a = 0.613). It can be formed athermally in the alloy range
86 at. % Ti to 70 at. % Ti by quenching from the -phase region, or it can be formed by aging in the temperature range
of 100 to 500 C. The phase is typically observed as small
ellipsoids roughly 5 nm to 10 nm in their longest dimension. In cold-worked and heat-treated NbTi strands, they
can grow to 50 nm in diameter. All the metastable phases
can be transformed to single-phase -NbTi or two-phase
+ microstructures by heating long enough in the or
+ phase elds, respectively.
Cold-Worked Microstructures
In order to achieve high critical current densities a ne and
homogeneous dispersion of ux pinning material must be
introduced that is of sufcient volume for signicant pinning but does not deleteriously affect the other Hc2 or Tc .
The process by which the rst high critical current density
microstructures were achieved was arrived at empirically
before the resulting microstructures were characterized
(12). The processing involved a high cold-work strain followed by three or more heat treatments in the + phase
range, each separated by additional cold work with the nal heat treatment being followed by another large coldwork strain. An understanding of the microstructural development was key, however, to the further optimization of
NbTi and the reproducible production of high critical current strand. Initial observation of the microstructure was
hindered by the difculty in preparing transverse cross sections of micron-sized laments suitable for examination by
transmission electron microscopy (TEM). Once techniques
had been developed to prepare the TEM specimens, it became clear that folded sheets of -Ti precipitates were the
dominant microstructural features of the nal strand (see
Ref. 13). Systematic analysis of the production process (as
in Ref. 14) revealed that the high prestrain heat treatments produced -Ti precipitates only at the intersections
of grain boundaries. The location of precipitation at the
grain boundary triple points meant that the precipitation
was homogeneously distributed if alloy composition and
grain size were uniform. The grain boundary triple-point
-Ti was also sufciently ductile that it could be drawn
down to the nanometer scale with breaking up or causing the strand itself to become difcult to draw. This contrasted with the other commonly observed -Ti precipitate
morphology, Widmanstatten
-Ti, which formed in densely

packed rafts in the interior of grains and resulted in a great
increase in the lament hardness. The next section reviews
each step of the process in more detail.
THE CONVENTIONAL PROCESS
Alloy Fabrication
A high-purity ne-grained NbTi alloy with chemical homogeneity over both a large and small microstructural
scale is an essential starting point to the production of
NbTi strands. The large liquidsolid phase separation
shown in the phase diagram (Fig. 3), along with the high
melting point of the Nb, makes it particularly difcult and
expensive to produce a high-quality NbTi alloy suitable
for superconductor application. The main driving force for
high homogeneity is the key role that precipitate quantity
and morphology play in determining critical current density, both of which are highly sensitive to composition. The
development of a high-homogeneity NbTi alloy was a crucial step in the advance toward high critical current NbTi
(see Ref. 15). The desired properties of the initial alloy billet
are as follows:
1. The correct overall alloy composition to optimize
Hc2 , Tc and precipitation for pinning. The acceptable
range is Nb46-wt. % Ti to Nb48 wt. % Ti.
2. Uniform composition over the entire billet to ensure
optimum physical and mechanical properties over
the entire lament.
3. Chemical homogeneity on a microstructural level in
order to ensure uniform precipitation of the correct
morphology (typically 1.5 wt. % Ti).
4. Low and controlled levels of impurity elements in order to ensure predictable superconducting and mechanical properties.
5. Elimination of hard particles (typically Nb-rich) because any particle that does not co-reduce with the
alloy can result in lament drawing instability and
ultimately strand breakage. The exterior of the nal
NbTi rod must also be free of hard particles and
must be smooth enough that it does not easily pick
up particles during subsequent handling.
6. A ne (typically ASTM grain size 6 or smaller) and
uniform grain size as it controls the distribution of
precipitate nucleation sites. A ne grain size also improves diffusion barrier uniformity. Where high critical current is less important, a larger grain size has
been used increase ductility.
7. Low hardness (typically a Vickers hardness number
of 170 or less) to ease co-deformation with softer stabilizer material.
The NbTi alloy is prepared from high-purity Nb and Ti
by consumable electrode vacuum-arc melting (where the
electrodes are composites of Nb and Ti) and by electronbeam or plasma-arc melting. It is usually necessary to
remelt the ingot two or three times in order to achieve the
necessary chemical homogeneity. Primarily produced for
the aviation industry, the high-purity source Ti is reduced
from TiCl4 by Mg (the Kroll process). High-purity Nb is rened from lower-purity Nb by two or three electron-beam
remelts. The lower-purity Nb source itself is extracted from
niobitetantalite (Nb2 O5 and Ta2 O5 ) or pyrochlore (0.25%
to 3% Nb2 O5 ) ores by way of an intermediate ferroniobium
alloy which is used on a relatively large scale for steel production. Table 1 lists the typical allowable ranges for impurities, typied by specications for the superconducting
supercollider. The small level of allowable Ta has a historical origin, and it is unlikely that additions of less than 1.5
wt. % Ta will have a signicant impact on superconducting
or mechanical properties. Increasing the level of Fe from
200 L/L (the specication of the superconducting supercollider as found in Ref. 16) to 500 L/L (L/L is equivalent
to the more commonly used ppm) actually has a benecial
effect as shown in Ref. 17.
The chemical inhomogeneities that may be observed in
the alloy at this stage in production can be divided into two
types based on size: macroinhomogeneities (those visible to
the eye) and microinhomogeneities (those requiring identication using microscopes). The most common macroinhomogeneities are Ti-rich freckles and hard Nb-rich particles. The Ti-rich freckles are so called from their appearance in ingot cross sections and are a result of Lorentz
and buoyancy-driven ow of Ti-rich material between dendrites (see Ref. 18). Control of radial heat transfer and uid
ow in the melt pool eliminates the occurrence of freckles.
Because of their relatively small size (typically 1 mm to
2 mm in diameter), compositional deviation (Ti-rich by 8
wt. % Ti to 10 wt. % Ti), and ductility, freckles are not in
themselves particularly deleterious to strand production.
The importance of the presence or rather the absence of Tirich freckles is as an indicator of good melt control. Ti-rich
freckles are readily identied from ash radiographs of ingot cross sections. If ash radiography indicates that an
ingot cross section is freckle-free, it is likely that smallerscale microinhomogeneities, which are more difcult and
expensive to quantify, have been kept to a minimum. A
more serious macroinhomogeneity is the presence of hard
Nb-rich particles which result in strand breakage failures
(see Ref. 19). Nb-rich regions are a result of the high freezing point of Nb and can be eliminated by good process control and remelting. The Nb-rich particles were the cause of
many early strand failures but are rarely seen in modern
production.
Microchemical inhomogeneity in the cast ingots of
NbTi is inevitable because of the coring produced by the
large liquidsolid phase separation. The microhomogeneity level can be qualitatively revealed by metallography
using a composition-sensitive etch as shown in Fig. 4. In
this example of a high-homogeneity-grade alloy, the microchemical variation is 1 wt. % Ti and has a wavelength
of 100 m to 200 m. Commercial NbTi alloys have microchemical variations of 1 wt. % Ti to 4 wt. % Ti,
with higher-homogeneity alloys costing more. Where high
critical current density is less important, reduced microhomogeneity can be acceptable in order to reduce cost but
not to an extent that will reduce strand yield by causing
drawability problems during subsequent processing.
The diameter of the initial cast NbTi ingot ranges from
200 mm to 600 mm, and this is typically reduced to 150 mm
by hot forging before being fully annealed in the singlephase region (approximately 2 h at 870 C). Extended anneals can be used to reduce microchemical inhomogeneity
but will increase the grain size and consequently reduce
the density of precipitate nucleation sites.
Figure 4. Microchemical inhomogeneity in an NbTi alloy can be

revealed using a composition-sensitive etch, as in this example of
an high-homogeneity Fe-doped Nb46 wt. % Ti alloy produced by
Teledyne Wah Chang.
Composite Assembly
Stabilizer. All superconducting strands are fabricated as
a combination of one or more continuous laments of superconductor within a high electrical and thermal conductivity matrix. The matrix material provides electrical and
thermal stability and protection from burnout if the superconducting device reverts to the normal state (quenches).
High-purity Cu is most commonly used because it has good
electrical and thermal conductivity (a resistivity of 0.42
nm and a thermal conductivity of 260 W/mK at 6 T, 4.2
K), a high heat capacity, and good strength at both low temperature and during processing and it combines and processes well as a composite with NbTi. A CuNi stabilizer
is used for alternating-current (ac) applications where a
high-resistivity matrix is required to reduce eddy current
loss in the matrix and coupling between the submicron laments (e.g., Refs. 20 and 21. Where high transverse resistivity is not required, Mn additions to Cu can be used to
suppress the proximity coupling of laments (22). Combinations of high-purity Cu and CuNi or CuMn can be used
within the same composite because the alloy stabilizer is
only required between laments. High-purity aluminum
has a greater in-eld thermal conductivity and electrical conductivity as well as a lower heat capacity, a lower
density, and greater radiation resistance than Cu. Despite
these considerable advantages, Al is rarely used as a sta-
bilizer because it is extremely difcult to coprocess with

NbTi. Furthermore, conventional processing requires the
use of precipitation heat treatments that further soften the
Al with respect to the NbTi. The use of APC processing
avoids the need for precipitation heat treatment and has
been used to fabricate Al-stabilized strand with 15 vol. %
of Al (23). Additional stabilizer can be added after or late
in the fabrication process by external application, by soldering the strand into a channel machined in the external
stabilizer, coextruding the strand inside the external stabilizer, or other cladding techniques (24). These methods
of applying external stabilizer allow for a greater variety
of conductor cross-section designs and stabilizer compositions.
Diffusion Barriers. Good fabrication practice results in
an excellent metallurgical bond between the composite
components by the stage of the process at which precipitation heat treatments are applied. At -Ti, precipitation
temperatures Cu and Ti at an NbTi/Cu interface will react
to form hard intermetallic TiCu (most commonly TiCu4 )
compounds (25, 26). The hard intermetallics do not codeform with the laments and will agglomerate as nal
wire drawing proceeds. The agglomeration of hard particles results in lament nonuniform lament cross sections
and ultimately can lead to strand failure. The practice of
applying Nb diffusion barriers between the NbTi and the
stabilizer to improve lament uniformity and strand predates the acute intermetallic problems caused by aggressive multiple precipitation heat treatments (27). Nb diffusion barriers became commonplace after it was clear that
they would be required to achieve the high critical current densities required by the superconducting supercollider project (28). Nb is effective at inhibiting the formation
of the TiCu intermetallics and is mechanically compatible
with the NbTi alloy rod.
The Nb barrier is normally applied as a sheet wrapped
around the NbTi rod before inserting into the stabilizer
material. Because the Nb sheet does not contribute to the
critical current density, the stability reduces the overall superconductor cross section and is kept to a minimum. The
Nb sheet is not impervious to Cu or Ti diffusion and must
be kept to a sufcient thickness so that Cu levels in the
NbTi do not reach high enough levels during heat treatment that TiCu4 is formed. As the composite is deformed
to smaller and smaller cross section, so is the thickness
of the diffusion barrier, thus the most critical processing
stage for the Nb diffusion barrier is the nal heat precipitation heat treatment. At this stage, not only is the barrier at its thinnest during heat treatment, but earlier heat
treatments have produced Ti-rich -Ti precipitates, some of
which will be adjacent to the barrier. Although Cu and Nb
have a very low mutual solubility, the high density of Nb
and NbTi grain boundaries in the cold-worked composites provides sufcient pathways for Cu diffusion. Faase et
al. (29) have calculated that for an aggressive nal heat
treatment of 80 h at 420 C a minimum barrier thickness
of 0.6 m is required to prevent reaction layers in the lament (there is no signicant reaction in the Cu for Custabilized composites). For smaller laments, this can represent a signicant proportion of the nonstabilizer area.
Figure 5. High-resolution back-scattered electron scanning electron microscope image of the NbTi adjacent to an Nb diffusion
barrier (white) after nal precipitation heat treatment. Nonuniformity of the barrier is shown along with a zone of reduced precipitate (black) next to the barrier.
For instance, for a 6 m nal lament diameter design, a

0.6 m barrier at nal heat treatment size represents 4%
of lost superconductor area. For a 2.5 m nal lament diameter the same thickness diffusion barrier represents a
lost superconductor area of over 9%. The calculated predictions of required Nb thickness agree well the development
of 6 m lament strands for the superconducting supercollider where it was shown that 4 area % Nb barriers were
required as opposed to 2 area % barriers used in early prototypes (30). Faase et al. also observed that the -Ti precipitates adjacent to the Nb barrier were much smaller than
elsewhere in the NbTi. In addition to the formation of
brittle intermetallics, Moreland et al. (31) found that Cu
concentrations of <3% in the NbTi signicantly degraded
the local energy gap. Reducing the temperature of the nal
precipitation heat treatment signicantly reduces the required minimum barrier thickness but must be balanced
against a reduced rate of precipitation. Figure 5 shows a
transverse cross section of NbTi lament adjacent to a
Nb barrier after it has received the third of three precipitation heat treatments. The nonuniformity of the barrier
thickness is clear in Fig. 5 as is a zone of reduced precipitate size adjacent to the barrier.
The grains of the BCC Nb barrier and the NbTi superconductor deform under plane strain conditions such that
their 100 directions align with the wire-drawing axis,
and they thin in their 001 direction. Maintaining grain
boundary continuity requires that both the NbTi and Nb
grains must intercurl around their neighbors during cold
work (see Ref. 32). At the NbTi/Nb interface the Nb grains
intercurl with the NbTi grains resulting in an uneven
barrier thickness. As the grain size increases, the scale of
NbTi penetration into the barrier increases. Hessner et al.
(33) showed that there was a linear relationship between
increasing NbTi grain size and increasing variation in Nb
barrier thickness, providing an additional incentive to use
ne-grained NbTi alloy.
Strand Geometry and Filament Spacing. A wide variety
of strand geometries are manufactured from NbTi, ranging from large single-lament conductors to multilamentary strands with more than 40,000 laments. The low-
resistivity stabilizer tends to be much softer than the

NbTi superconductor, so attention has to be paid to producing a composite that is mechanically stable over the
large extrusion and drawing reductions. Gregory et al.
(34) established that the optimum mechanical stability for
NbTi/Cu multilamentary composites was for a lament
spacing-to-diameter (s/d) ratio of 0.15 (ratios of 0.15 to
0.20 are now typical). This ratio xes the local stabilizerto-superconductor ratio; and if additional stabilizer is required, it can be designed outside the lament pack. Ghosh
et al. (35) established that for direct-current (dc) magnet
application a minimum Cu thickness of 0.4 m to 0.5 m
was required to reduce the magnetization associated with
proximity lament coupling. For s/d ratios of 0.15 to 0.20,
the minimum Cu thickness requirement limits the minimum lament diameter to 2.7 m to 3.3 m when using a
pure Cu matrix. If smaller laments are required, Ni or Mn
can be added to the Cu between the laments as explained
earlier.
Composite Assembly. In its simplest form, an NbTi rod
with or without a barrier wrap is coextruded inside a highpurity Cu ne-grained can. The warm extrusion produces
a well-bonded monolament. For a multilamentary application the monolament composite can be drawn to an
intermediate size so that it can be restacked in another Cu
can (typically 250 mm to 360 mm in diameter and shaped
to t the extrusion chamber) to form a second extrusion
composite. The monolaments for restacking are usually
drawn through a nal hexagonal die to facilitate uniform
strand packing. In addition to the NbTi/Cu subelements
the stack will also contain shaped Cu spacers and Cu hexes
to ll voids and provide additional stabilizer if necessary.
Some composite designs utilize a central Cu core (usually
assembled from hexagonal Cu rods) which is thought to
reduce center burst during extrusion (superconducting supercollider specications required a Cu core). The composite must be densely packed to avoid upset during extrusion. Even after packing with Cu shims, some void space
is inevitable in any multilamentary stack. Normally the
amount of void space in a well-packed composite is not
sufcient to cause problems during extrusion; but in some
cases particularly, when there are large numbers of small
laments, the extrusion billet may have to be isostatically
compacted prior to extrusion. Single stacks of more than
22,000 laments have been successfully fabricated. Alternatively, composites with 200 or fewer laments can be
fabricated by inserting bare NbTi rods into holes in a
stabilizer billet prepared by gun-drilling (see Ref. 36 for
more examples). A good metallurgical bond is required between the composite components if uniform lament diameters and high yield are to be achieved, and this requires that contamination of the surfaces be kept to a minimum. The mobility of oxygen in Cu at processing temperatures also means that surface oxides must be removed
from the high-purity Cu in order to maintain low resistivity
and good mechanical properties. When the monolament
is fabricated at large size (100 mm to 200 mm diameter),
it benets from a low surface-to-volume ratio for the components. The restacked monolaments suffer from a relatively large surface-to-volume ratio that increases as the
diameter of the stacked laments decreases. The importance of reducing chemical and particulate contamination
cannot be overemphasized. Long lengths of strand (typically 10 km or more) are desired for many applications, and
small particulate pickup that signicantly reduces yield by
strand breakage can occur in even tightly controlled largescale production (37).
After the billet is assembled, the laments are sealed
under vacuum by electron-beam (or sometimes tungsten
inert gas) welding a Cu lid onto the billet.
Extrusion
After preheating to 500 to 650 C the extrusion billets are
conventionally extruded at ratios of 10:1 to 20:1 and then
water-quenched. Lower temperatures are favored because
they maintain a greater degree of cold work in the NbTi
but they require greater forces. Low extrusion speeds avoid
excessive heating of the NbTi. Hydrostatic extrusion may
also be used when available and has the advantages of
lower-temperature operation, higher yields available from
higher length-to-diameter ratios, and an ability to coextrude a wider variety of material combinations. After successful extrusion, the composite is a well-bonded composite
of a suitable size (50 mm to 90 mm in diameter) for cold
work to proceed.
Cold Work
Superconducting NbTi strands are fabricated in a coldworked condition that is far greater than any other metallurgical process, with the possible exception of piano wire.
The object of the cold work is to produce the nal nanometer scale dispersion of pinning center. Understanding the
role and importance of this highly strained state in the
production of NbTi strands is the key to the conventional
process. The amount of cold work in the NbTi normally
can be represented by the true strain, T , which is given
by Eq. (1):
where d0 and A0 are the starting diameter and crosssectional area of the alloy rod at the nal recrystallization
anneal, respectively, and d and A are the diameter and area
of the NbTi after cold work. Figure 6 illustrates the conventional process in terms of cold-work strain for a composite fabricated from a cold-worked rod and given three heat
treatments. The total available cold-work strain available
for processing the NbTi is termed the strain space and
is limited by the initial annealed NbTi rod size and the
nal lament diameter. Warm extrusions increase the cold
work in the NbTi at a reduced rate compared with similar reductions performed by rod or wire drawing and consequently reduce the effective strain space. Increasing the
temperature of extrusion improves the bonding between
the composite components but reduces the available coldwork strain.
The strain space can be further subdivided into three
regions, each requiring a minimum cold-work strain to be
effective:
Figure 6. Schematic illustration of the conventional process for

NbTi superconductors in terms of cold-work strain.
1. Prestrain, p , the cold-work strain before the initial

precipitation heat treatment.
2. The inter-heat treatment strain, HT , the strain required between precipitation heat treatments.
3. The nal strain, f , the strain required to reduce the
precipitate size to nal optimum pinning size.
Figure 7. The morphology and location of precipitation in coldworked NbTi alloys are determined by the amount of cold work
in the microstructure. As the Ti content is increased, the cold work
required to produce optimum -Ti (only at grain boundary intersections) increases. In this compilation of data (3944), optimum
precipitation is denoted by the symbol , and mixed precipitate
morphology is denoted by the letter m for heat treatment at 420 C.
The number 3 identies where mixed-mode precipitation has also
been observed for 375 C heat treatments.
The cold work performs seven primary functions:

1. Encouraging the formation of the preferred precipitate phase and morphology.
2. Improving microchemical homogeneity by mechanical mixing both prior to heat treatment and after
heat treatment when local Ti depletion has occurred.
3. Increasing the density of precipitate nucleation sites.
4. Increasing the grain boundary density, thereby increasing diffusion rates (grain boundary diffusion being considerably faster than bulk interdiffusion).
5. Reducing the average diffusion distance to the precipitate nucleation site.
6. Increasing the volume of precipitate by multiple
strain/heat treatment cycles.
7. Reducing the precipitate dimensions from the precipitation scale of 100 nm to 300 nm diameter to the
pinning scale of 1 nm to 5 nm.
The cold work is applied to the extruded composite by
standard rod- and wire-drawing techniques. The rod or
wire is pulled through a shaped die that uniformly reduces
the cross-sectional area by 15% to 25%. As the strand becomes smaller in diameter, multiple die machines are used
to apply more than area reduction in a single pass. The
amount of cold work in a single-phase NbTi alloy can
be monitored using hardness testing: As the strain in the
NbTi increases, so does the hardness (38).

Buckett and Larbalestier (39) established that a minimum cold-work strain of approximately 5 was required
before precipitation heat treatment to ensure the optimum precipitation mode, that of -Ti precipitates located
at grain boundary intersections. At lower strains, phase
and/or intragranular Widmanstatten
were formed, producing inhomogeneous microstructures that reduced workability. Lee et al. (40) systematically studied the relationship
between prestrain, precipitate morphology, and alloy and
showed a relationship between NbTi composition and the
required cold-work prestrain required to avoid strains
phase and/or intragranular Widmanstatten
-Ti (40). Figure 7 illustrates the strong increase in required prestrain

with increasing Ti content. This relationship is very important because it not only explains the importance of a homogeneous alloy composition but also explains the difculties
associated with processing high Ti alloys. An increase in
p of roughly 0.77 is required for each wt. % Ti increase in
composition. Figure 8 shows the large amount of phase
and or intragranular Widmanstatten
-Ti that can be produced by heat treating a high-Ti alloy (in this case, Nb54
wt. % Ti) at too low a prestrain. By a true strain of 5 to 7
a high-angle grain structure of regular, small diameter (50
nm to 100 nm) grains has been produced by the large coldwork strain. Increasing the prestrain further homogenizes
and renes the microstructure.
Figure 8. Transmission electron microscope image in bright eld

(a, b), and dark eld using an omega reection (c) of an Nb54
wt. % Ti alloy after precipitation heat treatment. The heat treatment was applied at a cold-work prestrain of 5, which is insufcient to avoid the deleterious phase and or intragranular Widmanstatten
-Ti forms of precipitation. In (a) the microstructure

is shown in longitudinal cross section with the drawing axis running down the page. Images (b) and (c) are of the same transverse
cross-sectional area with the dark-eld image bringing out contrast from the phase.
Figure 9. The peak critical current density increases linearly

with volume percent of -Ti precipitate over the range of 3% to
25 vol. % -Ti for Nb47 wt. % Ti (41, 43). Similar results have
been shown for 50 wt. % Ti (42). Least squares t dependencies
for the Nb47 wt. % Ti data are shown for applied elds of 5 T and
8 T.
Precipitation Heat Treatment

Lee et al. (41) rst established a linear relationship between the optimized critical current density and the volume of precipitate in a laboratory-scale monolamentary
composite fabricated from Nb47 wt. % Ti alloy as shown
in Fig. 9. The relationship extended from 0% of the strand
volume being precipitated (non-heat-treated) to 25 vol. %.
Chernyj et al. (42) extended this relationship to an Nb50
wt. % Ti alloy and a maximum volume percent of -Ti of
28%. A wider study of strands produced by different manufacturers for the superconducting supercollider conrmed
the linear relationship for industrial-scale strands (43).
The importance of maximizing the amount of precipitate
in the strand is unambiguous. Precipitate is produced in
the NbTi by heat treatments at 375 to 42 C for a duration of typically 40 h to 80 h. Increasing the temperature
increases the precipitation rate but increases the precipitate diameter and the low-eld Jc to high-eld Jc ratio (43).
The amount of precipitation is also dependent on the alloy
composition; the quantity of -Ti produced by the rst precipitation heat treatment increases strongly with Ti content (44) as shown in Fig. 10. This relationship shows how
too low a Ti content in the NbTi alloy can result in insufcient precipitation for high critical current density and
how a large local variation in Ti content can lead to an inhomogeneous distribution of precipitates and, subsequently,
ux-pinning sites.
After approximately 10 vol. % precipitate has been produced in the rst heat treatment, it becomes very difcult to produce signicantly more without excessively long
heat treatment times. By applying additional cold-work
strain to the microstructure, more precipitate is produced
Figure 10. The precipitation rate in NbTi increases strongly

with Ti content in NbTi alloys. Additional heat treatments further increase the amount of precipitate. As additional heat treatment and strain cycles are applied, and more precipitate is produced, the residual Ti content of the -NbTi matrix drops until
insufcient Ti is left to drive further precipitation. In this graph,
data from one and two heat treatments (40) are combined with
data from three heat treatments of 80 h at 420 C. The average
residual matrix composition is calculated assuming an -Ti composition of Nb3.75 at. % Ti.
(as shown in Fig. 10 for the second heat treatment). An optimum balance between increased precipitate volume and
minimum strain space is at a strain of approximately 1.2
(12). Three or more heat-treatment and strain cycles are
normally required to produce the 20 vol. % or more precipitate in the microstructure required for high critical current densities (Jc > 3000 A/mm2 at 5 T and 4.2 K). As the
-Ti is precipitated, the composition of the -NbTi is depleted in Ti until it reaches between 36 wt. % Ti and 37
wt. % Ti, at which point there is insufcient Ti to drive further precipitation. More aggressive heat treatment is more
likely to compromise the Nb diffusion barrier and coarsen
the precipitate size.
After the nal heat treatment the microstructure
viewed transverse to the drawing axis consists of a uniform
distribution of roughly equiaxed -Ti precipitates, 80 nm to
200 nm in diameter, in a matrix of equiaxed NbTi grains
of similar dimensions. Viewed in longitudinal cross section
the -Ti and -NbTi grains are somewhat elongated along
the drawing axis with an aspect ratio of 4 to 15 depending
on the processing history. Further cold-work strain is required to reduce the dimensions of the precipitates so that
they can pin ux efciently. During the -Ti precipitation
heat treatments, the -NbTi matrix has been depleted in
Ti to a level of 37 wt. % Ti to 38 wt. % Ti, and the Hc2 and Tc
of the composite at this point in processing are the same as
the values of single-phase material of these lower Ti levels
(5).
Final Wire Drawing
The plain strain-imposed intercurling of the NbTi grains
that is so deleterious to barrier uniformity also results in
the distortion of the -Ti precipitates into densely folded
sheets during nal wire drawing. The folding process
rapidly decreases the precipitate thickness and spacing
and with a dependence of d1.6 (where d is the strand diameter) and increases the precipitate length per area with
a dependence of d1.6 as measured by Meingast et al. (5). As
the microstructure is rened toward optimum size the bulk
pinning force increases and the peak in the bulk pinning
force moves to higher eld as shown in Fig. 11 (data from
Ref. 45). The Hc2 and the Tc gradually return to the values
of the original single phase starting alloy as the precipitate
are rened toward and below the superconducting coherence length, (5). The critical current density increases as
the microstructure is rened until it reaches a peak, after which there is a steady decline. The peak in Jc for a
monolament or a multilamentary strand with uniform
laments occurs at a nal strain of approximately 5. If the
laments are nonuniform in cross section (sausaged), the
peak occurs earlier and at a lower critical current density.
A strand that has a premature (and lowered) peak in Jc
during nal drawing is described as extrinsically limited
because it has not attained the intrinsic critical current of
the microstructure. The most common source of extrinsic
limitation is sausaging of the laments due to intermetallic
formation or lack of bonding between the components of the
composite. The degree to which a composite has been extrinsically limited can be observed by examining the sharpness of the resistive transition when measuring the critical
current, Ic . Volker (46) showed that the shape of the transition curve near its onset can be approximated by
where V is the voltage across, I is the current in the superconductor, and n is the resistive transition index. For a
nonextrinsically limited superconductor the value of n at
5 T, 4.2 K, can be 70 or higher. By quantifying the variation in lament cross-sectional area by image analysis, the
amount of lament sausaging can be measured directly
(47). A high critical current density superconductor with
a high n-value is shown in Fig. 12. The strand is one of a
number of high-performance wires developed for the superconducting supercollider (48). The lament sausaging
in this strand has been reduced to a very low level (a coefcient of variation for the lament cross-sectional areas
of approximately 2%). With tight quality control, uniform
properties and piece lengths exceeding 10 km should be
expected.
The specic pinning force for the -Ti precipitates falls
from 360 N/m2 for an average sheet thickness of 2.6 nm,
to 200 N/m2 for a 1 nm average sheet thickness (14, 49)
but this is more than compensated for by the increase in
precipitate density caused by the continued folding of the
-Ti sheets.
Final Processing
Twisting. Just before a multilamentary strand has
reached nal size, it is usually twisted about its drawing
axis. The twisting is required to reduce ux-jump instability caused by varying external elds, reduce instabilities
caused by self-eld, and reduce eddy-current losses. The
tightness of the required twist increases with the expected
rate of change of eld. The required twist pitch for a superconducting supercollider strand, a relatively steady-state
magnet, was approximately 80 rotations along the drawing axis per meter, while for ac application with a similarly
sized strand the number of twists per meter might be 300.
The twisting occurs just before the strand is reduced to nal size so that it can be locked in by a nal drawing pass
or by nal shaping.
Final Shaping. The nal shape of the strand cross section
need not be round in cross section: It can also be shaped
into square or rectangular cross section by the use of independently adjusted rollers operating along the strand
surface.
Cabling. Individual strands can be cabled or braided together to form a conductor with a higher current-carrying
capacity. The most common design for NbTi magnets is
the Rutherford cable, which consists of two parallel at
layers of strands. Using this approach, high-aspect-ratio
cables can be produced with as many as 46 strands (50).
As was the case for the individual laments, the strands
are transposed around the cable, forming a densely packed
square or rectangular cross-section spiral. The design consideration for Superconducting Supercollider Laboratory
cable are discussed in Ref. 51. The compaction of the strand
around the squared cable edges severely distorts the strand
10
Figure 11. For conventionally processed NbTi the bulk pinning

force increases in magnitude with drawing strain after the last
heat treatment. The increase occurs at all elds as the precipitate
size and spacing are reduced to less than a coherence length in
thickness (45). The renement of the microstructure with increasing strain for the same strand is shown schematically in transverse cross-sections with the -Ti precipitates in black.
cross section, but the excellent mechanical properties of

NbTi/Cu composites combined with good strand design
and advances in cabling technology have reduced cabling
degradation to minimal levels (52).
Figure 12. Partial cross-section of a strand designed for the

Large Hadron Collider at CERN by IGC Advanced Superconductors (now Luvata Waterbury, Inc.). 250,000 km of Nb-Ti strand
were required in order to produce magnets for the 27 kilometer LHC ring, including 1232 dipoles and 858 quadrupoles. Each
dipole was 15 m in length and weighed 35 tonnes. The LHC uses
1.9 K operation to push the Nb-Ti based magnets beyond 8 T. Inset is the full strand cross-section showing the individual lament
stacking units. Each LHC strand has 6425 or 8800 laments of 6
or 7 m diameter respectively.
attain a critical current density 1000 A/mm2 at a eld of

11.5 T (2.05 K) using an Nb37 wt. % Ti22 wt. % Ta alloy.
Ta has an even higher melting point than Nb, making the
fabrication of chemically homogeneous ternary alloys particularly difcult. The behavior of NbTiTa alloys under
the conventional process is similar to that of binary alloys,
but the precipitates do not appear to pin as efciently (57).
THE APC PROCESS
NbTiTa
The addition of Ta to NbTi alloys suppresses the paramagnetic limitation of Hc2 by the large orbital moment of the
alloys (53). Although Ta is only of benet below 4.2 K (54),
it has a relatively long history of study because it should
extend the useful eld range of ductile superconductors by
1 T or more (55). So far, however, improved Hc2 has not
translated effectively into improvements in Jc , except very
near to Hc2 (above 11 T). Lazarev et al. (56) were able to
The quantity, composition, and distribution of pinning center as well as the composition of the matrix are limited,
in the conventional process, by the thermodynamics of the
NbTi phase diagram. Additional precipitate can be produced by increasing Ti content of the alloy (as shown in
Figure 10), but that is more than offset by the decrease in
Hc2 (Fig. 2). The result is a critical current limit in conventionally processed NbTi superconductors of approximately 3800 A/mm2 , at 4.2 K and 5 T. An alternative ap-
proach is to fabricate the microstructure by mechanically

assembling the desired components of the microstructure
at large size and reducing the microstructure to the appropriate size by extrusion and cold drawing (58, 59). The
engineered microstructural rods can be restacked into a
composite just as for a conventional NbTi superconductor, but no precipitation heat treatments are required. An
intermediate approach developed by Supercon, Inc. (60)
uses a low-temperature diffusion heat treatment to modify a densely packed microstructure fabricated from layer
of pure Nb and Ti. The diffusion-modied APC has been
successfully used in solenoid, model dipole (61), and MRI
magnets (62). Round-wire APC superconductors and multilayers have developed zero-eld Jc up to 10% of the theoretical upper limit provided by the depairing current density
Jd . (Jd Hc /) (e.g., Refs. 63 and 64), where is the penetration depth. APC superconductors fabricated with Nb
pins perform particularly well at low elds (up to about 5
T to 7 T), and Jc values approaching 7500 A/mm2 at 3 T
(65, 66) have been achieved (25% of Nb pinning center in
an Nb47 wt. % Ti matrix). Nb has been a preferred pinning material because of its mechanical compatibility with
the NbTi matrix. Even using Nb, however, poor workability and increased costs associated with assembly and yield
have so far limited the commercial application of APC composites. The components of an engineered microstructure
must initially be large enough to be stacked by hand (or
possibly machine); consequently the engineered pins must
undergo a far greater deformation to reach optimum size
than for -Ti precipitates which start at 100 nm to 200 nm
in diameter. The larger deformation and multiple extrusions and the restacks required by the APC process result
in a microstructure that can be much less uniform than for
the conventional process (67). For this reason, processes
that can use smaller cross-sectional starting dimension,
such as stacked or wrapped sheet, can result in superior
properties such as the Jc of 4250 A/mm2 at 5 T and 4.2 K,
achieved by Matsumoto et al. (68) with stacked sheets of
Nb50 wt. % Ti and 28 vol. % of Nb sheets. Because of the
large amount of cold work in the engineered microstructure, it is extremely sensitive to heating during extrusion;
the highest round-wire Jc (5 T, 4.2 K) of 4600 A/mm2 was
achieved by Heussner et al. (69). For the Nb pins, similar
volumes of pinning material are required as for conventionally processed materials; but by using ferromagnetic
pins (Fe or Ni) the required pin volume to achieve high
critical current density has been reduced to only 2 vol. %
(70). Such developments suggest that there are still exciting advances that can be made in the development of
ductile NbTi-based superconductors.
2 K OPERATION
The relatively low Tc of Nb-Ti allows means that considerable gains in Hc2 and and Jc can be gained from lowering the temperature of operation using superuid liquid He, typically at 1.8 to 2 K. This has been recently
exploited by the Large Hadron Collider project at CERN,
which operates at 1.9 K so that the magnetic eld can be
pushed beyond 8 T (71). Boutboul et al. (72) have shown
11
that the critical current density of production LHC strands,

from ve different sources, when measured at 1.9 K and 9
T (22752376 A/mm2 ) is approximately that of the same
strands at 4.2 K and only 6 T.
BIBLIOGRAPHY
1. E. W. Collings, Applied Superconductivity, Metallurgy, and
Physics of Titanium Alloys, vol.1: Fundamentals, New York:
Plenum, 1986.
2. E. W. Collings, Applied Superconductivity, Metallurgy, and
Physics of Titanium Alloys, vol.2: Applications, New York:
Plenum, 1986.
3. E. W. Collings, A Sourcebook of Titanium Alloy Superconductivity, New York: Plenum, 1983.
4. H. Muller, The upper critical eld of niobiumtitanium, PhD
thesis, Univ. WisconsinMadison, 1990.
5. C. Meingast, P. J. Lee, D. C. Larbalestier, Quantitative description of a high Jc NbTi superconductor during its nal optimization strain. I. Microstructure,Tc , Hc2 and resistivity, J.
Appl. Phys., 66: 59625970, 1989.
6. D. L. Moffat, Phase transformations in the titaniumniobium
binary alloy system, PhD thesis, Univ. WisconsinMadison,
1985.
7. I. Pfeiffer, H. Hillmann, Der Einuss er Struktur auf die
Supraleitungseigenschaften von NbTi50 und NbTi65, Acta
Metall., 16: 14291439, 1968.
8. M. Hansen et al., Systems titaniummolybdenum and
titaniumcolumbium, J. Metals, 3: 881888, 1951.
9. L. Kaufman, H. Bernstein, Computer Calculations of Phase
Diagrams, New York: Academic Press, 1970.
10. D. L. Moffat, U. R. Kattner, The stable and metastable TiNb
phase diagram, Metall. Trans., 19A: 23892397, 1988.
11. D. L. Moffat, D. C. Larbalestier, The competition between
martensite and omega in quenched TiNb alloys, Metall.
Trans., 19A: 16771686, 1988.
12. Li Chengren, D. C. Larbalestier, Development of high critical
current densities in niobium 46.5 wt % titanium, Cryogenics,
27: 171177, 1987.
13. D. C. Larbalestier, A. W. West, New perspectives on ux pinning in niobiumtitanium composite superconductors, Acta
Metall., 32: 18711881, 1984.
14. P. J. Lee, D. C. Larbalestier, Development of nanometer scale
structures in composites of NbTi and their effect on the
superconducting critical current density, Acta Metall., 35:
25262536, 1987.
15. D. C. Larbalestier et al., High critical current densities in
industrial scale composites made from high homogeneity
Nb46.5T, IEEE Trans. Magn., 21: 269272, 1985.
16. Specication for Niobium Titanium Alloy for the Superconducting Supercollider, MSD Document Control, Dallas, Texas:
Superconducting Supercollider Laboratory, 1992, SSC-MagM-4000A.
17. D. B. Smathers et al., Improved niobium 47 weight % titanium
composition by iron addition, Mater. Trans., Jpn. Inst. Metals,
37 (3): 519526, 1996.
18. R. Schlatter, Electrical and magnetic interactions in vacuumarc remelting and their effect on the metallurgical quality
of speciality steels, J. Vac. Sci. Technol., 11 (6): 10471054,
1974.
12
19. H. Hillmann, Fabricaton technology of superconducting material, in S. Foner and B. B. Schwartz (eds.), Superconducting
Materials Science, New York: Plenum, 1981, p. 275.
20. I. Hlasnik et al., Properties of superconducting NbTi superne
lament composites with diameter 0.1 m, Cryogenics, 25
(10): 558564, 1985.
21. J. R. Cave et al., Reduction of AC loss in ultra-ne multilamentary NbTi wires, IEEE Trans. Magn., 25 (2): 19451948,
1989.
22. T. S. Kreilick et al., Reduction of coupling in ne lamentary
Cu/NbTi composites by the addition of manganese to the matrix, Adv. Cryog. Eng., 34: 895900, 1988.
23. M. K. Rudziak, T. Wong, J. Wong, Development of APC NbTi superconductors with internal high purity aluminum stabilizer,
IEEE Trans. Appl. Supercond., 7: 11971200, 1997.
24. H. C. Kanithi, D. Phillips, B. A. Zeitlin, Further development
of aluminum clad superconductors, IEEE Trans. Magn., 27:
18031806, 1991.
25. D. C. Larbalestier et al., High critical current densities in ne
lament NbTi superconductors, IEEE Trans. Nucl. Sci., 32:
37433745, 1985.
26. M. Garber et al., Effect of Cu4Ti compound formation on the
characteristics of NbTi accelerator magnet wire, IEEE Trans.
Nucl. Sci., 32: 36813683, 1985.
27. E. Gregory, in R. W. Meyerhoff (ed.), Manufacture of Superconducting Materials, Metals Park, OH: Amer. Soc. Metals, 1977,
pp. 116.
28. R. M. Scanlan, J. Royet, R. Hannaford, Evaluation of various
fabrication techniques for fabrication of ne lament NbTi
superconductors, IEEE Trans. Magn., 23: 17191723, 1987.
29. K. J. Faase et al., Diffusional reaction rates through the Nb
wrap in SSC and other advanced multilamentary Nb46.5
wt. %Ti composites. Adv. Cryog. Eng., 38: 723730, 1992.
30. Y. High et al., Quantitative analysis of sausaging in Nb barrier
clad laments of Nb46.5 wt % Ti as a function of lament
diameter and heat treatment, Adv. Cryog. Eng. (Mater.), 38:
647652, 1992.
31. J. Moreland, J. W. Ekin, L. F. Goodrich, Electron tunneling into
superconducting laments: Depth proling the energy gap of
NbTi laments from magnet wires, Adv. Cryog. Eng. (Mater.),
32: 11011108, 1986.
32. W. F. Hosford, Jr., Microstructural changes during deformation
of [011] ber textures metals, Trans. Metall. Soc. AIME, 230:
1215, 1964.
33. R. Heussner, P. Lee, D. Larbalestier, Non-uniform deformation
of niobium diffusion barriers in niobiumtitanium wire, IEEE
34. E. Gregory et al., Importance of spacing in the development of
high current densities in multilamentary superconductors,
Cryogenics, 27: 178, 1987.
35. A. K. Ghosh et al., Anomalous low eld magnetization in ne
lament NbTi conductors, IEEE Trans. Magn., 23: 17241772,
1987.
36. T. S. Kreilick, Niobiumtitanium superconductors, in Metals
Handbook, 10th ed., vol. 2: Properties and Selection: Nonferrous Alloys and Special-Purpose Materials, ASM Int., 1990,
pp. 10431057.
37. P. Valaris et al., A statistical evaluation of recent SSC conductors produced at IGC/ASI, IEEE Trans. Magn., 27: 17521754,
1991.
38. J. Parrell, P. Lee, D. Larbalestier, Cold work loss during heat
treatment and extrusion of Nb46.5 wt % Ti composites as
measured by microhardness, IEEE Trans. Appl. Supercond.,

3: 734737, 1993.
39. M I. Buckett, D. C. Larbalestier, Precipitation at low strains in
Nb%46.5 wt % Ti, IEEE Trans. Magn., 23: 16381641, 1987.
40. P. J. Lee, J. C. McKinell, D. C. Larbalestier, Microstructure
control in high TiNbTi alloys, IEEE Trans. Magn., MAG-725:
19181924, 1989.
41. P. J. Lee, J. C. McKinnell, D. C. Larbalestier, Restricted novel
heat treatments for obtaining high Jc in Nb46.5 wt % Ti, Adv.
Cryog. Eng. (Mater.), 36: 287294, 1990.
42. O. V. Chernyj et al., NbTi superconductors of a high currentcarrying capacity, Supercond. Sci. Technol., 4: 318323,
1991.
43. P. J. Lee, D. C. Larbalestier, An examination of the properties
of SSC Phase II R&D strands, IEEE Trans. Appl. Supercond.,
3: 833841, 1993.
44. P. J. Lee, J. C. McKinell, D. C. Larbalestier, Progress in the understanding and manipulation in high Jc NbTi alloy composites, in Y. Murakami (ed.), Proc. of New Developments in Appl.
Supercond., Singapore: World Scientic, 1989, pp. 357362.
45. C. Meingast, D. C. Larbalestier, Quantitative description of a
very high critical current density NbTi superconductor during its nal optimization strain. II. Flux pinning mechanisms,
J. Appl. Phys., 66: 59715983, 1989.
46. F. Voelker, Resistance in small, twisted, multicore superconducting wires, Particle-Accelerators, 1 (3): 205207, 1970.
47. Y. High et al., Quantitative analysis of sausaging in Nb barrier
clad laments of Nb46.5 wt % Ti as a function of lament
diameter and heat treatment, Adv. Cryog. Eng. (Mater.), 38:
647652, 1992.
48. P. Valaris et al., SSCL phase 1 inner cable development program at IGC Advanced Superconductors, Inc., IEEE Trans.
Appl. Supercond., 3: 705708, 1993.
49. P. J. Lee, D. C. Larbalestier, Determination of the ux pinning
force of -Ti ribbons in Nb46.5 wt. % Ti produced by heat
treatments of varying temperature, duration and frequency,
J. Mater. Sci., 23: 39513957, 1988.
50. R. Scanlan et al., Design and fabrication of a high aspect ratio cable for a high gradient quadrupole magnet, IEEE Trans.
Appl. Supercond., 7: 936938, 1997.
51. D. Christopherson et al., SSC 40 mm cable results and 50 mm
design discussions, IEEE Trans. Magn., 27: 18811883, 1991.
52. M. Garber et al., Superconducting wire and cable for RHIC,
IEEE Trans. Magn., 30: 17221725, 1994.
53. M. Suenaga, K. M. Ralls, Some superconducting properties of
TiNbTa ternary alloys, J. Appl. Phys., 40: 44574463, 1971.
54. D. G. Hawksworth, D. C. Larbalestier, Enhanced values of Hc2
in NbTi ternary and quaternary alloys, Adv. Cryog. Eng., 26:
479486, 1980.
55. H. R. Segal et al., NbTi based conductors for use in 12 Tesla
toroidal eld coils, Proc. 8th Symp. Eng. Problems Fusion Res.,
New York: IEEE Press, 1980, pp. 255259.
56. B. G. Lazarev et al., The study of the microstructure and Jc in
Nb37Ti22Ta superconductor produced with different duration of treatments, in H. W. Weber (ed.), Proc. 7th Int. Workshop Critical Currents Supercond., Alpbach, Austria, Singapore: World Scientic, 1994, pp. 601604.
57. P. J. Lee et al., Microstructure property relationships in
NbTiTa, IEEE Trans. Appl. Supercond., 3: 13541357,
1993.
58. G. L. Dorofejev, E. Yu Klimenko, S. V. Frolov, Current carrying capacity of superconductors with articial pinning centers,
in C. Marinucci and P. Weymuth (eds.), Proc. 9th Int. Conf.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
Magnet Technol., Villigen, Switzerland: Swiss Institute Nuclear Res., 1985, pp. 564566.
B. A. Zeitlin, M. S. Walker, L. R. Motowidlo,Superconductors
having controlled laminar pinning centers, and method of
manufacturing same,US Patent 4,803,310, 1989.
M. K. Rudziak et al., Development of APC NbTi composite
conductors at Supercon, Inc., IEEE Trans. Appl. Supercond.,
5: 17091712, 1995.
R. M. Scanlan et al., IEEE Trans. Magn., 30: 16271632,
1994.
C. Renaud et al., First commercial application of NbTi superconductor employing articial pinning centers, IEEE Trans.
Appl. Supercond., 5: 11891192, 1995.
G. Stejic et al., Effect of geometry on the critical currents of
thin lms, Phys. Rev., B49: 12741288, 1994.
E. Kadyrov, A. Gurevich, D. C. Larbalestier, High critical current densities in Nb47%Ti multilayers with planar copper
ux pinning nanostructure, Appl. Phys. Lett., 68: 15671569,
1996.
L. R. Motowidlo et al., Multilament NbTi with articial pinning centers: The effect of alloy and pin material on the superconducting properties, Appl. Phys. Lett., 61 (8): 991993,
1992.
H. C. Kanithi et al., Further developments in NbTi superconductors with Articial Pinning Centers, Adv. Cryog. Eng.
(Mater.), 38: 675683, 1992.
P. J. Lee, D. C. Larbalestier, P. D. Jablonski, Quanticaton of
pinning center thickness in conventionally processed and powder processed articial pinning center microstructures, IEEE
K. Matsumoto et al., Enhanced Jc properties in superconducting NbTi composites by introducing Nb articial pins with a
layered structure, Appl. Phys. Lett., 64 (1): 115117, 1994.
R. W. Heussner et al., Increased critical current density in
NbTi wires having Nb articial pinning centers, Appl. Phys.
Lett., 70 (7): 901903, 1997.
N. D. Rizzo et al., Ferromagnetic articial pinning centers in
superconducting Nb0.36 /Ti0.64 wires, Appl. Phys. Lett., 69 (15):
22852287, 1996.
L. Rossi, Superconducting magnets for the LHC main lattice.
IEEE Trans. Appl. Supercond 14.2: 153158, 2004.
T. Boutboul, S. Le Naour, D. Leroy, L. Oberli, V. Previtali, Critical current density in SC Nb-Ti strands in the 100 mT to 11
T applied eld range, IEEE Trans. Appl. Supercond., 16.2:
11841187, 2006.
PETER J. LEE
University of
WisconsinMadison,
Madison, WI
13
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering

c 1999 John Wiley & Sons, Inc.
Copyright
SUPERCONDUCTORS: PROCESSING OF HIGH-TC BULK,

THIN FILM, AND WIRES
Superconductors are a class of materials possessing two unique properties: the complete loss of electrical resistivity below a transition temperature called the critical temperature (T c ) and the expulsion of magnetic flux
from the bulk of a sample (diamagnetism) in the superconducting state. The latter property is also known as
the MeissnerOchsenfeld effect or more commonly, the Meissner effect (1). At temperatures above T c , these
materials possess electrical resistivity like ordinary conductors, although their normal state properties are
unusual in many aspects. The abrupt change from normal conductivity to superconductivity occurs at a thermodynamic phase transition determined not only by the temperature but also by the magnetic field at the
surface of the material and by the current carried by the material. Several metals and metallic alloys exhibit
superconductivity at temperatures below 22 K, and will be henceforth called low temperature superconductors
(LTS). In 1950, superconductivity was explained as a quantum mechanical phenomenon by the London phenomenological theory (2). Later, the two-fluid phenomenological model explained the electronic structure of a
superconductor as a mixture of superconducting and normal electrons, with the proportion of superconducting
electrons ranging from zero at the onset of superconductivity to 100% at 0 K (3). In 1955, Bardeen, Cooper, and
Schrieffers (BCS) theory explained that superconductivity was the result of the formation of electron pairs
of opposite spins (known as Cooper pairs), primarily owing to electronphonon coupling (4). The BCS theory
proved to be the most complete theory for explaining the superconducting state and the normal state of LTS
materials. A major development in superconductors was the discovery by Josephson in 1963 that Cooper pairs
show macroscopic phase coherence, and that such pairs can tunnel through a thin insulating layer sandwiched
between two superconducting layers [the superconductorInsulatorsuper-conductor (SIS) junction known as
the Josephson junction] (5). This effect, called the Josephson effect (5), caused a flurry of activity in the fields of
high-speed computer logic and memory circuits in the 1960s and 1970s, since it can be used to make high-speed
low-power switching devices. Problems were encountered in the mass fabrication of Josephson junctions for
complex systems such as digital computers. Although the applications of LTS for electrical applications and
electronics were demonstrated, the cost of cooling was too high for the commercial development of LTS.
The era of high-temperature superconductors (HTS) began in 1986 when two IBM Zurich researchers, K.
A. Muller and J. G. Bednorz, reported the occurrence of superconductivity in a lanthanum barium copper oxide
(LaBaCuO) at 30 K(6). Soon after, M. K. Wu, P. W. Chu, and their collaborators at the University of Alabama
and University of Houston (7), respectively, announced the discovery of 90 K superconductivity. Since these two
historic discoveries, there has been substantial progress in HTS technology. Several new families of cuprates
including BiSrCaCuO (8), TlCaBaCuO (9), and HgCaBaCuO (10) have been found to be superconducting above
90 K. These discoveries make feasible electrical and electronics applications at temperatures above the boiling
point of liquid nitrogen (77 K). Cuprate superconductors with a T c value higher than 30 K have been classified as
high-temperature superconductors. The obvious advantage of using liquid nitrogen rather than liquid helium
for cooling is its higher heat of vaporization, which not only simplifies the design of cryostats but also the
cost of cooling. Furthermore, liquid nitrogen (at $0.25/liter) is more than an order of magnitude cheaper than
liquid helium (at $5/liter). Progress made in cryocoolers has made feasible HTS applications in electrical wires,
1
SUPERCONDUCTORS: PROCESSING OF HIGH-TC BULK, THIN FILM, AND WIRES
Fig. 1. A generalized road map for the HTSs technology. YBCO 123 compound has been the most studied material among
the HTSs. Most applications have been demonstrated with the YBCO superconductor. Most applications in HTS wires have
been demonstrated using BSCCO.
magnets, and electronics. The excitement and challenges posed by these HTS materials have touched multiple
disciplines, such as physics, chemistry, material science, and electrical engineering. Tremendous progress
has been made in the application of HTS materials in such areas as Superconducting Quantum Interference
Devices (SQUIDs), passive microwave devices, and long-length wires, as illustrated in the road map for the HTS
technology, shown in Fig. 1. Better-quality materials emerging from refined processing methods have made
it possible to separate the intrinsic properties of HTS from its extrinsic properties. The interrelationships of
processing with structural, physical, electrical, and magnetic properties continues to be an area of intensive
scientific research. In this article, we provide an overview of important high-temperature superconducting
materials, their properties, and promising procedures for synthesizing bulk, thin film and wire forms of HTS
conductors for engineering applications.
Properties of High-Temperature Superconductors

Magnetic Properties. The critical temperature (T c ) is defined as the temperature below which a
superconductor possesses no dc electrical resistivity or the temperature below which a superconductor exhibits
diamagnetic behavior. The useful range of operation of a superconductor is typically below 0.6 T c , as other
Fig. 2. Temperaturecurrentmagnetic field (TJH) three-dimensional surface defining the operating limits for a superconductor.
important superconducting properties are enhanced below this temperature. Critical current density (J c )
is one of the most important superconducting properties for engineering applications. It is an estimate of
maximum current density (current per cross-sectional area of the conductor) a superconductor can support
before becoming a normal conductor. The critical field (H c ) for a superconductor is the maximum magnetic
field below which a superconductor exhibits diamagnetic behavior and above which the superconductivity is
quenched. The T c , H c , and J c parameters define a point in three-dimensional space, and for superconductors
these points span a volume as shown in Fig. 2.T c , H c , and J c values are relatively low in type I superconductors.
Type II superconductors are generally suitable for most electrical and electronic applications because of higher
T c , H c and J c values (11). In type I superconductors, the flow of shielding current in the superconductor is
restricted to a thin layer from the surface, called the penetration depth (), when the magnetic field is below
H c . The penetration depth is very small near 0 K, and increases dramatically as the temperature approaches
T c . The penetration depth at 0 K ranges from 100 to 1500 for type I materials (11). Above the critical field,
the magnetic field completely penetrates a type I superconductor, quenching superconductivity, as shown in
Fig. 3. Figure 3 also illustrates the T-dependence of resistivity in a superconductor at T > T c .
In type II superconductors,H c1 represents the lower critical field above which magnetic flux penetrates a
superconductor to form a mixed state in which superconducting and normal electrons coexist. When H > H c2 ,
the upper critical field, superconductivity is largely confined to the surface of the material. In the mixed state,
magnetic flux penetrates through small tubular regions on the order of the coherence length () (a length scale
that characterizes superconducting electron pair coupling), called vortices (or flux tubes), with each vortex
containing one quantum of flux, 0 (12). Abrikosov, in his study of type II superconductors determined that 0
= h/2e, where h is Plancks constant and e is the electronic charge (12). The vortices form a periodic lattice called
the Abrikosov vortex lattice. The resistivity of a superconductor may be vanishing in the mixed state, provided
the vortices are pinned or trapped. As the applied magnetic field (H a ) approaches H c2 , the number of vortices
increases until there can no longer be any more addition of vortices, at which point the material becomes a
normal conductor. Figure 4 shows the magnetic properties of type II superconductors. In the mixed state, each
vortex resides in a normal region, which is separated by superconducting regions. The vortices experience three
different types of forces: one is the Lorentz force due to the flow of external current, experienced in the direction
of and proportional to the vector product of the current and the vortex field. Second is the force of repulsion
from other vortices, and third is the pinning force from metallurgical defects. The Lorentz force causes motion
of vortices (also called the flux flow). The vortex motion produces an opposing electric field to the flow of current,
Fig. 3. A type I superconductors magnetic flux density versus applied magnetic field. A type I superconductor will not
have any flux enclosed in the bulk below the critical temperature. Above the critical field, the applied field completely
permeates in the bulk of the material.
essentially contributing to ohmic losses. However, vortices can become trapped or pinned (called flux pinning)
at metallurgical defects, in secondary phases or at impurity sites. Owing to the repulsive forces between each
vortex, pinning a few vortices may lead to a frozen vortex lattice, or a lossless state, true only for direct currents
and low-frequency alternating currents. The frozen vortex lattice occurs only below a critical field called the
irreversibility field (H irr ). Some of the trapped vortices could remain in the superconductor, contributing to a
hysteretic behavior when an alternating current is applied, which produces further dissipation of energy. The
trapped flux is analogous to remanent flux in ferromagnetic materials. While pinning centers tend to prevent
the movement of vortices, there is a tendency for vortices to jump over the pinning defects. This phenomenon is
called flux creep. At H a > H irr , vortices will move, causing additional energy dissipation. Resistive transitions
in a superconducting sample in applied magnetic fields are shown in Fig. 5. Transitions at higher fields clearly
show additional ohmic losses (13). Although the upper critical field is generally higher in type II materials, the
limiting field is the irreversibility field, which is an order of magnitude lower than H c2 . Owing to the complex
nature of the cuprate ceramic superconductors, and operation at higher temperatures, the ac losses in HTS
materials are generally higher than in LTS materials (14,15).
Structural Properties. The presence of one or more copper oxide (CuO2) planes in the unit cell is a
common feature of all HTS materials, also referred as cuprate superconductors. The most popular cuprate
materials are Y1 Ba2 Cu3 O (henceforth referred to as YBCO), Bi2 Sr2 Can 1 Cun O2n+4 (where n = 2, 3) BSCCO
(henceforth referred to as Bi2212, Bi2223 for n = 2, and n = 3, respectively), Tl2 Ba2 Cam 1 Cum O2m+4 (henceforth
referred to as Tl2201, Tl2212, and Tl2223), and HgBa2 Cam 1 Cum O2m+2 (where m = 1, 2, 3). Table 1. lists
the well developed cuprate superconductors, their superconducting properties, and important applications
demonstrated to date. In YBCO, there are two square planar CuO2 planes stacked in the c-direction, separated
by an intercalating layer of barium and copper atoms and a variable number of oxygen atoms. The conventional
wisdom is that the CuO2 planes are the conduction channels of superconductivity, whereas the intercalating
layers provide carriers or act as charge reservoirs necessary for superconductivity, although this view is
not shared universally (16). The charge density, the number of superconducting charge carriers per unit
volume, is determined by the overall chemistry of the system and by the charge transfer between the CuO2
planes and the CuO chains. The charge density in a HTS material (1019 /cm3 ) is two orders of magnitude
lower than conventional LTS (1021 /cm3 ). Remarkably, the oxygen content in the system changes the oxidation
Fig. 4. The magnetic behavior of a type II superconductor, showing the Meissner state behavior below H c1 , mixed state
behavior between H c1 and H c2 , and the surface superconductivity between H c2 and H c3 .
Fig. 5. Temperature dependence of resistivity for an epitaxial c-axis-oriented YBCO thin film at different magnetic fields,
with the field parallel to the c-axis. Courtesy of 13.
states of the copper chain atoms, which, in turn, affects their ability for charge transfer, charge density,
and superconducting properties. Depending on the oxygen content, the YBCO 123 material could have a
nonsuperconducting tetragonal (a = b = c) phase or a 90 K superconducting orthorhombic (a = b = c) phase.
When fully oxygenated, YBCO possesses an orthorhombic unit cell with typical dimensions of a = 3.85 , b =
3.88 , and c = 12.0 , and a T c = 90 K. Figure 6 shows the crystal structure of YBCO 123 and BSCCO 2223
superconductors, showing the conduction layers and the binding layers in each case.
BSCCO superconductor contains a weakly bonded double BiO layer that separates the CuO2 planes. The
Bi2223 structure is part of a family of several other HTS compounds in which Bi is replaced by Tl or Hg
(with different oxygen coordination) and partially by lead. In some cases, the double layer of these metal-oxide
layers can be reduced to a single layer-yielding another family of superconductors such as 1212 or 1223 [e.g.,
Fig. 6. The crystal structure of YBCO 123 and BSCCO 2223 compounds, showing the conducting layers, and the binding
layers.
(TlPb)1 Ba2 Ca2 Cu3 O10 henceforth is referred as Tl1223]. The single-layer compounds offer strong flux pinning
and low intrinsic defect structure compared to the double-layer compounds. The presence of a weakly bonded
BiO layer is crucial for their superconducting properties in the Bi2212 and 2223 systems. The mechanical
properties of Bi2212 and 2223 are micaceous, (mica-or clay-like), and they have highly anisotropic growth
rates along the ab- plane and c-direction. The latter is important to process long-length wires and to enhance
the electromagnetic connectivity of the grains, thus making high transport current densities over long lengths
possible (17). This is mainly due to the weak interlayer bonding of the BiO layer with the CuO2 planes. On
the other hand, the weak interlayer bonding of the BiO layer also leads to intermixing of Bi2212 and Bi2223
phases. In spite of this weakness, better grain connectivity and micaceous crystalline morphology are attractive
for developing long-length HTS wires. Tl2212 and 2223 compounds have a structure similar to the Bi2212 and
2223 compounds with (TlO) double layers replacing the BiO layers.
Other Important Properties. All the popular HTS materials possess a fundamental limitation, crystalline anisotropy, (i.e., they possess different structural and electrical properties in different directions). Superconducting properties such as critical current density (J c ) and critical magnetic field (H c ) along the ab-planes
(xy) are superior to those along the c-axis (z). A major challenge for researchers has been to develop textured
samples to take advantage of high J c in the ab-planes. The HTS materials also exhibit higher penetration
depths compared with LTS materials. The penetration depth is an important parameter for high-frequency
applications of superconductors. It is defined as the depth through which a magnetic field penetrates into a
superconducting sample and decays to 1/e of the field at the surface. The magnetic field decays in the form of
H = H 0 e x/ , where H 0 is the field at the surface,x is the depth through the sample, and is the penetration
depth of the superconductor, analogous to the skin depth in conventional electrical conductors. Penetration
depth (T) increases with temperature. Penetration depth is a frequency independent parameter in contrast
to the frequency-dependent skin-depth of normal conductors. This means that little or no dispersion will be
introduced in superconducting components, and that it will be negligible up to frequencies as high as tens of
gigahertz, in contrast to dispersion present in normal metals. Furthermore, lower losses in superconductors
lead to a reduction in physical size, and this feature represents another advantage for HTS thin-film based
circuits. Compact delay lines, filters, and resonators are possible with a high-quality factor (Q) due to lowconductor losses (see Superconducting filters and passive components). The challenge in processing involves
developing HTS materials with smooth-surface morphology to minimize high-frequency ac conductor losses.
The extremely short coherence length in HTS materials (<30 along the ab-plane) increases the difficulty of
making Josephson junctions (see (Tunneling and josephson junctions).
Processing of High-Temperature Superconductors

Although T c and H c values of a superconductor are generally intrinsic properties of a specific material,J c and
values, on the other hand, are a function of sample microstructure and can vary by several orders of magnitude
owing to various processing techniques. A high J c in large magnetic fields is desirable for electrical applications
of HTS. In general, J c greater than 105 A/cm2 at 77 K is required for wires, tapes, and magnets for high
magnetic field applications (17). Although high J c s in HTS are achieved in bulk, thin films, and wires, several
processing-related problems need to be addressed. These problems relate to grain boundaries, metallurgical
defects, flux pinning mechanisms, and flux creep. Grain boundaries are interfacial regions between adjacent
grains and could contain impurity phases, normally conducting or insulating, extending beyond the coherence
length of these superconductors. These grain boundaries create superconductornormal conductor or insulator
superconductor Josephson junctions, the critical currents of which are much lower than in a homogenous
superconductor, and hence they are called as weak links. The grain boundaries are particularly a problem in
polycrystalline bulk superconductors because they affect the bulk critical current density. Grain boundaries
also contribute to surface resistance of a superconductor at high frequencies because they increase the residual
ac losses. Processing of HTS materials to improve J c requires ways to minimize the grain-boundary effects
and enhance the flux pinning sites in the superconductors. Addition of a small percentage of silver into
HTS materials improves the critical currents of grain boundaries. Also, texturing of a bulk superconductor
to align grains along ab-planes reduces the detrimental effects of grain boundaries. It has been determined
that the presence of low-angle grain boundaries [i.e., adjacent grains interfacing at a low angle (10 )], is not
deleterious to J c (18). However, the elimination of high-angle grain boundaries in HTS materials has been a
difficult challenge to overcome. Recent developments in YBCO-coated conductors show promise in reducing
the presence of high-angle grain boundaries, discussed later. To date, YBCO is the most successful candidate
for enhancements in flux pinning (19). Enhancement of flux pinning and suppression of flux creep have been
achieved by the introduction of controlled amounts of defects, such as co-existing normal secondary phases by
alloying, or by neutron or proton irradiation to define normal regions along the particle trajectory that give
rise to higher J c s and higher H c s (20).
Tremendous progress has been made in the synthesis of bulk single-phase HTS conductors, silversheathed wires, tapes, and thin-film-coated conductors in the past decade as shown in Table 1.. Bulk conductors
with large volume fraction of superconductivity are being routinely synthesized in YBCO-, Tl-, and Bi-based
HTS materials. Improved powder synthesis, understanding of reaction chemistry, and phase diagrams have
led to increased phase purity and improved flux pinning (resulting in higher current density at self field).
Solid-state reaction and melt texturing are the two widely used processing methods for bulk HTS.
Processing of Bulk Superconductors

There are three important process considerations for the choice of bulk superconductors for applications such
as magnets, current leads, and superconductor magnetic energy storage systems (1) growth of large-area,
single-phase bulk conductors, (2) ease of introducing strong flux pinning centers in the bulk, to enhance the
critical currents, and (3) long-term stability, (i.e., ability to withstand thermal cycling to modest temperatures)
and aging under ambient conditions. Among the various bulk superconductors, YBCO appears to be the most
plausible for electrical applications since the details of the crystal structure have been reliably established,
and it is easy to synthesize single-phase specimens. Moreover, in YBCO, introduction of flux pinning centers to
enhance J c has been demonstrated, although this is not the case in Bi, Tl, and Hg superconductors. Further,
melt-processing growth of YBCO superconductors provides grain-alignment resulting in J c as high as 105
A/cm2 at self field and 77 K (21). In the case of YBCO, the melt-powder-melt growth (MPMG) processing has
significantly improved performance of bulk materials. The thallates are also attractive as T c as high as 125
K, J c in bulk greater than 105 A/cm2 at 77 K and zero field (22) have been achieved. An advantage with the
thallates is that multiple phases, especially Tl2223 and Tl2212 phases, can coexist, without degrading the
superconducting properties in bulk conductors, thin films, and wires. The magnetic field dependence of J c is
comparable to the best YBCO superconductors. A disadvantage of thallates is the toxicity of Tl, which needs very
careful handling and processing. The volatility of Tl gives rise to its losses during high-temperature synthesis.
Most of the thallates contain structural defects, which affect the superconducting properties negatively. The
thermodynamics of Tl-based compounds is quite different from other HTS materials due to the volatility of Tl
(23).
At the present time, the synthesis of mercury-based HTS has unique problems mainly due to the volatility
of mercury compounds. Since HgO decomposes around 500 C, all processing must be done in a sealed environment in the production of bulk superconductors. Also, the Bi compounds are not attractive for bulk conductors
primarily due to the polycrystalline nature of the grains and weak flux pinning above 30 K (19). Melt-texturing
Bi compounds is not as simple as the YBCO. Bismuth compounds are generally thought to be attractive for
wire applications owing to their micaceous morphology. In the following section, we provide processing details
on YBCO, Tl2212, and Tl1223 bulk superconductors.
Processing YBCO 123 Bulk Superconductors.

Conventional Solid-State Reaction Method. The conventional means of producing bulk HTS materials
is the solid-state reaction method, or calcination. Carbonates, or oxides of Ba, Ca, Cu, and Y as starting materials, are mixed and ground to the desired size and then heat-treated in an alumina crucible at temperatures
as high as 950 C for 24 h. After cooling to room temperature, the reacted sample is finely ground and mixed
again and heat-treated for an additional 24 h; this step is repeated to promote homogeneity. The samples are
then pelletized in desired shapes using a hydraulic press at a typical pressure of 1500 psi. The pellets are
placed in a crucible made of gold or platinum or alumina, introduced into a furnace preheated to 900 C, and
sintered typically in air for 24 h. Oxygen is introduced into the furnace at around 500 C, during slow cooling
of the samples to room temperature. This gives rise to the growth of the orthorhombic superconducting phase.
Although the technique is fast, there are undesirable phases present in samples prepared this way, typically
BaCuO2 , CuO, and possibly some contamination from the heat-treatment process. In spite of these limitations,
it is possible to get T c of 90 K and current densities of 104 A/cm2 at zero field at 77 K. The polycrystalline
nature of grains in such samples results in poor field dependence of J c , with J c decreasing by three orders of
magnitude in a magnetic field of 0.1 T.
Texturing of YBaCuO 123 Superconductors. Texturing of YBCO can be achieved using mechanical, magnetic, and melt-growth techniques. The mechanical and magnetic means produce only a marginal
improvement in superconducting properties, mainly owing to the presence of large-angle grain boundaries,
whereas melt growth techniques reduce the large-angle grain boundaries and improve the critical currents
significantly. The melt-texturing techniques are the most suitable for enhancement of J c . Among the melttexturing methods, melt texture growth (MTG) (21), quench and melt growth (QMG) (24), and melt powder
melt growth (21) are techniques developed for nondirectional solidification. In these techniques, large YBCO
crystals grow stacked along the c-axis, with improved connectivity between grains. Directional solidification is
also possible by providing the right thermal gradient during the melt processing, either by the movement of
the furnace or the sample. Directional solidification methods align the grains along the ab-planes, the direction
of current flow, thus improving the current-carrying capability. It is also possible to align the grains along a
preferred orientation using seeded melt growth (25) and liquid phase removal methods (26). In the seeded melt
growth technique, a directionally oriented single crystal Sm123 or Nd 123 is used as a seed material to grow
YBCO, oriented along the same preferred direction.
The melt texture growth of YBCO 123 involves melting the YBCO above its peritectic temperature,
approximately 1010 C, in air. When heated above this temperature, the 123 compound undergoes incongruent
melting to form a solid and a liquid phase according to the following reactions (27)
On subsequent slow cooling, large oriented domains of 123 phase with 211 (green phase) inclusions are formed.
In such samples, grains grow preferentially along ab-planes, are stacked along the c-direction, and are coupled
by low-angle grain boundaries. This microstructure is in contrast to small randomly oriented polycrystalline
grains obtained by solid-state reaction. Quenching studies performed on MTG samples show the presence of
faceted growth interfaces where the 211 volume fraction and particle size decrease abruptly upon being included
into the growing 123 grains. The 211 phase appears to provide Y into the growing phase of 123. The MPMG
process is the most commonly used melt growth process designed for smaller sizes of the 211 precipitates, and
homogenous distribution of the precipitates. This technique involves reaction of Y2 O3 with a liquid phase to
form the 211 phase. The nonuniform distribution is handled by taking the powder and crushing and remixing
the melt quenched samples. This powder is pressed into a pellet and subjected to the melt growth. It was
found that the samples should not be held above the peritectic temperature for too long because the 211 phase
will grow into coarse grains. After a short time above the peritectic temperature, it is cooled to just below the
peritectic temperature and then slow-cooled in flowing oxygen to room temperature. The temperature profiles
for MTG and MPMG processing techniques are shown in Fig. 7. The critical temperatures are identified for
10
Fig. 7. Temperature profiles for (a) the melt texturing growth and (b) the melt process melt growth. Courtesy of 27.
Fig. 8. Schematic representation of microstructure evolution in the MPMG growth of YBCO: (a) after quenchingyttria
and solidified liquid; (b) during heating to the 211 + L regionyttria particles redesolving; (c) in the 211 + L region; and
(d) final microstructure showing the 123 and 211 phases. Courtesy of 27.
both processes. The evolution of the microstructure during the MPMG process is shown in Fig. 8 (27). The
figure illustrates the formation of 211 inclusions into the liquid phase.
The seeded melt growth is the most promising for large area domains of YBCO (27). In this technique,
a single crystal seed of Sm123 or Nd123 is embedded in a presintered Y123 bar, and the whole assembly is
introduced into a modified Bridgeman furnace for 15 min to 30 min and then withdrawn at a very slow rate
of 1 mm/h. As the growth front of Y123 passes by the seed, seeded growth occurs at the interface. Because of
the favorable alignment along ab-planes, large current densities, as high as 46,000 A/cm2 at 77 K and zero
field have been demonstrated. In the melt growth techniques, the hold time and cooling rate have a strong
influence on the microstructure of the samples, determining the grain size and nature of the grain boundaries.
The hold time determines the amount of liquid phase formed for the peritectic recombination. The cooling rate
determines the time available for the recombination of the liquid (25).
11
Processing of TBCCO 2212 and 1223 Superconductors. In general, processing of thallates with
high superconducting volume fraction (V sc ) has been difficult, primarily because of the volatility of Tl. Processing of Tl based HTS can be done either using an open system, or a closed system. In the open system, a sample
is heat treated in an oxygen atmosphere or air at temperatures between 850 C and 900 C, for 10 min to 15 min.
In a closed system, samples are sealed in a gold tube and heat-treated for longer times of 24 h to 48 h (28). The
open system is susceptible to loss of Tl and, hence, requires additional compensation for Tl. The closed system
has the disadvantage of sealing the samples in a container or a foil, resulting in reaction between the Tl-based
compound and the container. In our work, we were able to synthesize samples with T c s greater than 100 K for
the Tl2212 phase and 115 K for the Tl1223 phase, with superconducting volume fractions greater than 50%
in each case, by tuning the process parameters (29). The processing parameters that were optimized included
the starting compositions of the compounds, sintering temperatures, and sintering times. The starting compositions are critical mainly because of the loss of Tl during heat treatments. In the 1223 superconductor the
Hitachi composition of Tl0.5 Pb0.5 Sr1.6 Ba0.4 Ca2 Cu3 Oy has been a popular starting composition for high-quality
1223 superconductor (30). In our study, we examined the role of different Ba concentrations and excess Tl and
Pb doping concentrations to obtain a single phase 1223 superconductor in an open system (29).
Tl2212 samples were synthesized starting from BaCuO2 , CaO (99.999%) and Tl2 O3 (99.999%) compounds.
BaCuO2 was prepared by mixing BaO2 (99%) and CuO (99.99%) in equal molar portions and heat treating at
900 C for 24 h in air. The latter step was repeated after regrinding the pellet to promote homogeneity. BaCuO2 ,
CaO, and Tl2 O3 were mixed in a molar ratio of 2:1:1.07, pelletized, and reacted in an enclosed Pt crucible. Our
results indicate that samples sintered at 850 C for 10 min followed by a second heat treatment at 860 C for 10
min displayed optimum superconducting properties, which included a sharp T c of 97 K, with a superconducting
volume fraction as high as 54%. The porosity of the samples appears to be the limiting factor for the V sc .
The first step performed in the synthesis of (Tl,Pb)(Sr, Ba) CaCuO 1223 superconductor, was the preparation of a SrCa CuO precursor, starting from 99.999% pure CaO and CuO and 99% pure SrO. Appropriate molar
portions of SrO, CaO, and CuO were mixed, heated to 900 C for 24 h, cooled, and ground into a fine powder;
then the same process was repeated to obtain a homogeneous precursor compound. For the Ba-doped samples,
appropriate amounts of 99.99% pure BaO2 were alloyed with SrO, in the precursor. The second step was to add
appropriate amounts of 99.999% pure PbO and Tl2 O3 to the precursor. Pellets of approximately 500 mg were
made and heat-treated in an enclosed Pt crucible, at a temperature of 870 C for 10 min, followed by a slow
cooling.
Characterization of Bulk Superconductors

Structural Characterization. The primary characterization of any form of a superconductor consists
in establishing the structure by X-ray diffraction (XRD). Xrays from a source are collimated to strike the
sample at an incident angle and Bragg scattered from the sample. For X rays, atomic planes of the material
serve as a diffraction grating, causing the monochromatic X-ray beam to diffract and form a pattern. From an
XRD pattern, one can establish the various phases populated and, in principle, the crystal structure including
lattice parameters of each phase. In practice, since the c-axis lattice parameter depends on the oxygen content,
XRD patterns can be used effectively to correlate structural results with superconducting parameters. If there
is more than one phase present in a sample, the XRD patterns can be used to obtain quantitatively the
concentration of various phases present in a sample. Since bulk materials synthesized by solid-state reaction
are polycrystalline, the XRD pattern from such a sample has peaks corresponding to scattering from different
crystallographic planes such as (100), (110), and (001). Powder XRD is an effective means to establish the major
phases present in a sample.
12
Magnetic Properties. Vibrating sample magnetometer (VSM) is a widely used technique for the study
of all types of magnetic materials as it permits a measurement of the dc magnetization of a sample. In HTS
materials, VSM permits establishing the T dependence of the diamagnetic susceptibility in both a zero-fieldcooled (ZFC) and a field-cooled (FC) configuration. In the VSM technique, a sample is placed in a uniform
magnetic field and vibrated at a low frequency so as to induce a voltage signal in a pickup coil proportional
to the magnetic moment of the sample. The magnetic moment is proportional to the product of the sample
susceptibility and the applied field (30). Measurements are typically performed at several temperatures to
obtain the magnetic and superconducting properties. A plot of the magnetization versus applied magnetic field
can yield the magnetization J c using the Beans critical state model. For a cylindrical sample, the magnetization
J c is approximately 30 M/d where M represents the width of the hysteresis loop at a particular applied
field H, and d the effective thickness of the sample normal to the applied field (31). This model typically results
in an overestimate of the actual J c . A more direct method to determine the transport J c is by using the fourpoint probe method, in which a 1 V/cm electric field criterion is used to establish the transport J c . Other
magnetization measurement techniques include ac and SQUID magnetometers. In the ac magnetometer, a
sinusoidally oscillating field is used. The primary coil is surrounded by a set of oppositely wound detector coils.
Introducing a magnetic sample into the center of one or the other secondary detector coil leads to a voltage
imbalance in the secondary circuit (32). The resulting flux change leads to a measurable induced emf, which
is proportional to the sample susceptibility. Magnetic moment sensitivities as low as 10 8 emu are possible
using a commercial ac susceptometer. Enhanced sensitivity can be obtained using a SQUID magnetometer,
as SQUIDs are used in the detector circuitry. There are a large number of tools available for researchers to
correlate the physical properties with the processing variables. Table 2. lists characterization tools used by
HTS researchers, the outcomes from each tool, and corresponding observable.
Structural, Electrical, and Magnetic Properties of Bulk Superconductors. YBCO (123) Superconductors
The T c of YBCO is near 90 K, provided the oxygen content in YBCO is optimal. YBCO, in particular,
has remarkable of T c sensitivity to oxygen stoichiometry as displayed in Fig. 9 (33). Specifically, one finds
that at = 7.0 in the Y1 Ba2 Cu3 O superconductor,T c = 90 K. For < 6.2, T c = 0 K. The plot of Fig. 9 also
reveals that in the range 6.6 < < 6.8, T c = 60 K, and displays a plateau, a feature that is identified with
the existence of a superstructure in oxygen bonding. Figure 9 also reveals that it is not only the value of
, but also the distribution of oxygen in the structure that controls the value of T c . For YBCO, it is well
established from neutron Bragg diffraction measurements that oxygen in the CuO2 planes remains intact,
whereas oxygen in the chains (CuO3 ) diffuses at modest temperatures T > 300 C and effuses out of the
structure at T > 500 C. An orthorhombic to tetragonal phase transformation occurs near an oxygen content
of = 6.5. Oxygen can be reversibly desorbed and absorbed from YBCO either by heating in a vacuum or an
oxygen ambient, respectively. Plasma oxidation has been demonstrated as an attractive means to oxygenate
HTS at rather modest temperatures for microelectronic applications. At a local level, it appears that the T c ()
variation can be traced to the length of the Cu(1)O(4) bond that displays a threshold behavior as shown in
Fig. 10 (33). Superconductivity in YBCO commences when the Cu(1)O(4) bond length acquires a value of 1.82
or larger. Bond lengths less than 1.82 , apparently couple the pyramidal oxygen O(4) to the Cu(1) chain
cations, inhibiting charge transfer from the chains to the CuO2 planes. Oxygenation of YBCO to a value of
6.5 gives a Cu(1)O(4) bond length close to the threshold value.
The melt-textured samples have higher J c compared with conventional solid-state-sintered samples.
Figure 11 shows the enhanced J c typical of melt-processed samples (14,15). For comparison,J c of an epitaxial
YBCO thin film is shown. The figure shows improvement in J c for melt-textured samples compared with the
conventionally sintered samples. Melt-processed samples have at least an order of magnitude higher in-field
J c and do not drop as rapidly at higher fields as the sintered samples.
2212 superconductors. Synthesis of the Tl2212 superconductors was performed by reacting appropriate amounts of the Ba2 Ca1 Cu2 O5 precursor with Tl2 O3 . The results of three samples each processed differently
13
14
Fig. 9. Critical temperature versus oxygen content for YBCO samples. The dots and crosses were measured on samples
with oxygen removed by high-temperature quenching and low-temperature gettering, respectively. Courtesy of 33.
are presented here. Sample 1.1 was processed at 850 C for 10 min. Sample 1.3 was processed at 850 C for 10
min followed by an additional heat treatment at 860 C for 10 min. Sample 1.6 was prepared by heat treatment
at 850 C for 10 min followed by heat treatment at 860 C for 45 min. From XRD measurements, all samples
showed the characteristic Bragg reflection peaks of the Tl2 Ba2 CaCu2 O8 phase. No other impurity phases were
detected in the XRD scans. Figure 12 shows the XRD reflections for the three samples in which characteristic
peaks of the Tl2212 phase are clearly visible. Figure 13 shows the magnetization hysteresis loops of the three
samples recorded at 15 K. The magnetization scale for all samples was normalized to their mass. As seen
in Fig. 13, the diamagnetism displayed by sample 1.3 is the largest among the samples investigated. This
result has been reproduced in several samples. Figure 14 displays the temperature dependence of magnetic
susceptibility. In this figure, the scans are labeled according to the second heat-treatment temperature and
time. One finds that the second-step heat treatment at 860 C for 10 min not only increases the saturation
diamagnetization at T = 15 K but also significantly narrows the width of the superconducting transition, without affecting the T c value. Our conclusion is that the second-step heat treatment produces more strain-free
material, or chemical ordering of the phase. The underlying structural growth does not alter the bulk T c , but do
significantly enhances the diamagnetic response (Fig. 14), owing due to growth of a more strain-free material.
The superconducting volume fractions for samples 1.1, 1.3, and 1.6 were 37%, 54%, and 27%, respectively.
Clearly, the process conditions used for sample 1.3 appear to optimize growth of the superconducting phase.
Further increase in the second-step heat treatment time, resulted in an increase of T c to above 110 K. It has
been shown by Sugise et al. (34), that prolonged heating of Tl2212 samples in air can grow bulk 2223 samples.
Depletion of Tl2 O3 from Tl2212 samples essentially leads to the transformation of the two-layered cuprate
Tl2212 into the three-layered cuprate Tl2223 structure. Since the XRD scans provide no evidence of presence
of the 2223 phase, it must be present as stacking faults, or planar defects, as documented by Raveau et al. (35)
using high-resolution electron microscopy. Also, in Fig. 14, note that the sample with second-step 20 min heat
treatment in a sealed gold tube displays much smaller diamagnetism than the samples heat-treated in a Pt
15
Fig. 10. Critical temperature versus Cu(1)O(4) (copper in the chain bridging oxygen) bond length in the 123 phase. Also
shown are data for dopants introduced into the YBCO phase, with similar structural changes. As shown in the inset, the
CuO2 CuO2 plane distance increases, the O(4) moves toward the CuO chains, and the Ba moves toward the CuO2 planes.
Courtesy of 33 and references therein.
crucible. This is evidence of a reaction between the thallates and the gold tube at 860 C in a closed system and
suggests that gold should be avoided as a crucible for processing in a closed system of thallate superconductors.
Tl1223 Superconductors. For the synthesis of 1223 compound, the starting composition of
Tl0.5 Pb0.5 Sr2 Ca2 Cu3 O9 (1223) stoichiometry upon thermal processing yielded samples with 1223 and 1212
phases that had a T c of 107.5 K and a volume fraction of 55%. Replacement of some Sr by Ba in the starting
composition Tl0.5 Pb0.5 Sr2 x Bax Ca2 Cu3 O9 , on the other hand, yielded samples that had the 1223 phase. Samples with a Ba content x < 0.4 clearly were found to have both 1212 and 1223 phases, whereas samples with x
> 0.4 showed principally the 1223 superconductor. At x = 0.75, samples containing only the 1223 phase were
obtained upon heat treatment at 870 C for 11 min, using starting materials in which 50% additional Tl2 O3 and
PbO were compensated. Low-field ac susceptibility measurement of the sample shown in Fig. 15 reveals an
onset T c of 117 K with a superconducting volume fraction of about 50%. Note that the in-phase susceptibility
does not indicate the presence of minority phases. Samples with a Ba content of x = 0.75 showed the highest
T c of 117 K. Our results indicate that a minimum Ba content of x = 0.4 is necessary to produce a majority
1223 phase, which appears to be consistent with the Hitachi composition. Our study also demonstrates that
the highest T c , with the largest superconducting volume fraction are obtained at a Ba content of x = 0.75, in
Tl and Pb over compensated samples.
Metal-Doping Effects in YBCO Bulk Superconductors. Doping transition metals of group III (Fe,
Co, Ni) in YBCO exemplifies the trials and tribulations of working with high T c superconductors. As material
quality has improved, so have the challenges to understand the nature of the pairing state. One of the areas
that has been intensively studied is the doping of Fe in YBCO. More than 1000 papers have been published
16
Fig. 11. Comparison of field dependence of critical current density at 77 K, for conventional solid-state-sintered YBCO,
melt-textured YBCO, and the epitaxial YBCO thin films. The figure shows improvement in J c as a function of processing.
Courtesy of 14 and 15.
in this area, with the main conclusion that the broken line curve in Fig. 16 displays the T c (x) variation
separating the superconducting phase (x < xc ) from the antiferromagnetic insulators (x > xc ) near x = xc =
0.13 (36). This is the typical result for samples prepared by standard solid-state reaction at ambient pressure
of oxygen (henceforth referred to as APO). Recently, it has been demonstrated that the underlying phase
diagram is actually an artifact of sample processing. When these samples undergo an additional sinter at
915 C at elevated pressures of oxygen (220 atm of pure oxygen),T c (x) increases qualitatively, and all the
samples studied up to x = 1/3 are rendered superconducting with a T c greater than 50 K. In particular, samples
that were considered to be antiferromagnetic insulators (x > 0.13) have been rendered superconducting, thus
altering the presently accepted superconducting-antiferromagnetic phase diagram. Powder X-ray diffraction
measurements have shown that the average structure of the samples synthesized at high pressures of oxygen
(henceforth referred to as HPO) are also tetragonal as the usual samples synthesized at APO, except for a small
excess of oxygen. The Fe local environments as revealed directly in Mossbauer spectroscopy measurements in
the two types of samples (APO and HPO) at the same value of x are, however, qualitatively different, suggesting
that aspects of dopant-centered local structures control in a remarkable way the superconducting behavior in
these cuprates. There are plausible reasons to suggest that the majority of Fe resides in the CuO3 chains, but
that in APO samples at least Fe3+ does not replace Cu2+ but instead acquires a tetrahedral co-ordination as
exemplified by Ga3+ cations in Y1 Sr2 Ga1 Cu2 O7 superconductors (37). In HPO samples,Fe3+ is oxidized to Fe4+
with localization of an additional oxygen atom in its near-neighbor coordination, thus qualitatively changing
the local site symmetry. The Co doping effects in YBCO bear a similarity to those of Fe doping, although the
effects are less dramatic. The HPO sintering of APO samples increases the T c quantitatively in general at
doping concentrations x < xc = 0.20. At x > xc , antiferromagnetism sets in, and even HPO synthesis is unable
to transform such samples to superconductors. The Ni doping effects in YBCO are apparently unaffected by
17
Fig. 12. Powder X-ray diffraction spectra obtained three Tl2212-sintered samples processed at different conditions as
outlined in the text.
HPO synthesis as revealed by the results shown in Fig. 16. The T c (x) variation of Ni-doped YBCO samples
sintered at APO is similar to the variation in HPO samples. In contrast to Fe and Co doping, Ni doping in
YBCO does not alter the lattice symmetry, which remains orthorhombic, probably because the dopant enters
the structure substitutionally as Ni2+ replacing Cu2+ . The Fe and Co dopants enter the structure in higher
charge states Fe3+ and Fe4+ , Co2+ and Co3+ largely in planes and bring in additional oxygen to alter not only the
18
Fig. 13. Diamagnetic response of the samples 1.1, 1.3, and 1.6 obtained using VSM at 15 K. Sample 1.3 shows the largest
diamagnetic susceptibility among the three samples, indicating that the heat treatment procedure used for sample 1.3 is
close to the optimum.
lattice symmetry, which changes from orthorhombic to tetragonal, but also to create new local structures. The
nature of these local structures continues to be a subject of current interest in large part because there appears
to be a close correlation between the existence of specific structures and appearance of bulk superconductivity.
These new results also demonstrate that Fedoped YBCO samples synthesized at HPO are chemically more
19
Fig. 14. The temperature dependence of magnetic susceptibility obtained from VSM measurements. Sample 1.3 shows
the sharpest transition among the samples studied.
Fig. 15. The ac susceptibility measurement of a 1223 sample with x = 0.75 obtained using an ac susceptometer.
stable that those synthesized at APO. The higher chemical stability of HPO-synthesized YBCO samples may
be attractive for various applications of bulk, thin films, and wires.
Processing of Hts Thin Films
In-Situ Versus Ex-Situ Processing. Thin films of HTS materials (see THIN FILMS) have been deposited
on a wide variety of substrates using two different procedures: in-situ processing and ex-situ processing. Both
processes require high-vacuum deposition systems. In-situ processing is a technique in which the samples
are deposited and processed inside a vacuum chamber and may not require any postprocessing outside the
20
Fig. 16. T c plotted as a function of metal doping concentration x for M = Fe (top panel),M = Co (middle panel) and M =
Ni (bottom panel) in YBCO samples synthesized a high P (filled symbols) and at ambient P (open symbols). Lines drawn
through data points are guides to the eye. The T c values determined either magnetically or resistively are representative
of a vast literature. Magnetization measured in a 20 G field, in the zero field cooling mode, are displayed in insets.
chamber (i.e., the samples they will be superconductors when they are removed from the vacuum chamber).
The ex-situ processing method, on the other hand, requires postprocessing of samples outside the chamber.
A deposition process will yield an amorphous material if the substrates are not heated; therefore, additional
heat-treatment becomes necessary to obtain superconducting samples. The In-situ procedure is preferred
because it leads to thin films with uniformity, with good control over the growth process and stoichiometry.
21
An in-situ deposition process usually uses a complex deposition system configured with a substrate heater
assembly and process-monitoring capabilities. in-situ processing allows growth of multilayers with precise
thickness control, essential for applications such as coated thin film conductors on metallic substrates with
buffer layers. A drawback of the method is that it is not suitable for deposition on large area substrates. An exsitu procedure is a simpler process because it is easier to optimize, and one can obtain reasonably good quality
HTS thin films with postprocessing outside the vacuum chamber. The system requirements are less expensive
compared to in-situ processing and are possibly applicable for thick or coated films of superconductors.
Physical and Chemical Deposition Methods. Deposition of HTS thin films can be done using either
a physical or a chemical deposition method. The physical deposition methods include techniques such as
sputtering, electron beam evaporation, and pulsed laser ablation or deposition. Chemical deposition methods
include techniques such as metal-organic chemical vapor deposition (MOCVD) and metal-organic decomposition
(MOD) (see THIN FILMS). Primarily, physical deposition techniques include some form of high-energy ion or
electron bombardment of the target to release the material for deposition onto substrates. A chemical deposition
method involves chemical vapors (transported using a gas such as hydrogen or nitrogen) reacting at the surface
of heated substrates to form a thin film of desired composition. Both physical and chemical methods can be
tailored for in-situ or ex-situ processing. Among the deposition methods, sputtering and pulsed laser deposition
(PLD) are the most popular methods used, especially for the in-situ processes (see THIN FILMS).
Pulsed Laser Deposition. Pulsed laser deposition has been successfully used to deposit HTS thin
films for in-situ processing; PLD is designed to replicate target stoichiometry in the deposited thin films (38).
A typical PLD system consists of an ultra high vacuum chamber containing fused quartz windows for in-situ
spectroscopic investigations. Pulsed laser deposition is a vapor phase deposition process in which a focused
pulsed laser beam of energy density (also called fluence) exceeding 1 J/cm2 strikes a sintered composite target
material at an angle of 45 , vaporizes the target, and grows a thin film on a heated substrate placed nearby.
A CO2 laser, a Nd:YAG laser, a XeCl, or a KrF excimer each have been successfully used for PLD growth of
HTS. Multiple targets can be used with each target ablated for a required amount of time. In multitarget
systems, a target is rotated and synchronized with an optical encoder for pulsing the laser. This helps reduce
the focused laser beams heating effects on the target. A water-cooled target holder is normally used for further
heat removal. in-situ deposition of YBCO thin films is typically performed at an oxygen partial pressure in
the 70 mTorr to 200 mTorr range, on substrates heated in the range of 600 to 800 C depending on the choice
of substrates. The presence of oxygen is required for compositional control of the oxide superconductors. After
deposition, an oxygen anneal of the deposited film is performed for several hours at a reduced temperature
(400 to 500 C), with slow cooling, in a higher oxygen partial pressure of 500 mTorr to 700 mTorr to obtain
superconductivity (38,39).
Typical growth rates using a PLD system are in the range of a few angstroms per second. The growth
parameters controlling the quality of thin films include the choice of the substrate, substrate temperature,
background pressure of the vacuum system, laser energy density, frequency of the laser pulse sequencer, and
oxygen partial pressure in the system. The HTS thin films prepared by PLD have excellent surface morphology,
although one of the problems encountered has been the presence of particulates on the surface of such thin
films. Understanding the chemistry of the laser interaction with emitted radicals of the target material is
essential to control the growth quality of thin films over large areas and continues to be a subject of current
research.
Sputtering of Thin Films. Sputtering is perhaps one of the more popular deposition methods used
in the semiconductor industry for metals and dielectric materials. Broadly, there are two methods used: dc
sputtering and RF sputtering. Both methods use ion plasma, typically argon, because of its inertness. Highenergy ions bombard a target material to release target atoms, which are deposited on a substrate. Substrates
can be heated to a high temperature for in-situ processing of materials. A modified method in sputtering, called
magnetron sputtering, uses a magnetic field at the target to confine the plasma, which leads to higher deposition
rates at low working gas pressures. Sputtering systems can employ either a single target or multiple targets. A
22
multitarget sputtering system gives precise control in multilayer depositions, because one can deposit a chosen
target for a desired time, close a shutter and move on to the next target, essential for deposition of multilayers.
Single-target systems are inexpensive and easier to use. Due to differing sputtering rates of various atoms,
control of their relation content in a thin-film is difficult. In general, sputtering is susceptible to effects such as
negative ion bombardment, which are primarily due to oxygen ions bombarding and resputtering the deposited
thin film. One way to minimize negative ion bombardment is to use an off-axis geometry in which a substrate
is mounted transverse to the target assembly. Process optimization is difficult to achieve for deposition of
complex oxides such as the HTS materials. Some of the process parameters that need to be optimized include
the target stoichiometry, partial pressure of the inert gas, RF or dc power density, substrate-target distance,
and the substrate temperature. High-quality YBCO, TBCCO, and TPSCCO superconducting thin films and
multilayers have been deposited using off-axis magnetron sputtering (40,41).
Metal-Organic Chemical Vapor Deposition. Metal-organic chemical vapor deposition is a proven
technique for deposition of high-quality compound semiconductors. In principle, the advantages of MOCVD
are the ease of large-area depositions, with high deposition rates and without a need for sophisticated high
vacuum systems. However, the control of precise composition of the precursors is very difficult, and only small
area films have been made to date. The MOCVD uses a deposition technique in which the required atomic
constituents are transported in the form of vapors of metal-organic precursors by an inert transport gas such
as argon or nitrogen, premixed with an oxidizing gas. Metal-organic precursors are placed in stainless-steel
containers and heated to temperatures between 200 to 500 C to release vapors of the metal-organics. These
vapors react with the heated substrate inside a reaction chamber. Mass flow controllers are used to control flow
rates for each precursor. Mass flow rate, oxygen partial pressure and temperature of the substrate are some
of the important process parameters. Typical metal-organic precursors for YBCO include metal -diketonates
such as Y(C11 H19 O2 )3 , Ba(C11 H19 O2 )3 , and Cu(C11 H19 O2 )3 (42). A modification of this deposition technique is
also currently used and is called the metal-organic decomposition (43). In this technique, metal carboxylates
or acetates dissolved in an organic solvent are spin-coated on a substrate. The coated material is heat-treated
at high temperatures to obtain the desired HTS phase. The technique is simpler to use because it does not
require high vacuum equipment and is easily scalable for large area samples. The MOD does show promise for
long-length wires. However, it is not pursued by many researchers due to the difficulty of producing films with
only c-axis-oriented grains.
In our study, we have deposited in-situ films of YBCO on MgO substrates using the off-axis RF magnetron
sputtering (18) process. Typical sputtering conditions for YBCO thin films are substrate temperature of 700 C,
sputtering gas pressure of 500 mTorr, oxygen partial pressure of 10 mTorr, RF power of 60 W, and a target
dc bias of 30 V. We have also processed Tl2212 thin films by on axis RF magnetron sputtering, and ex-situ
processing. Tl2212 films were fabricated from both Tl2212- and Tl2223-sintered superconducting compounds
at an RF power of 220 W, chamber pressure of 5 mTorr. Such thin films had to be postprocessed in an excess
Tl2 O partial pressure, provided in the form of sintered 2223 pellets in a platinum crucible used for sintering. In
both cases, heat treatment times and temperatures were optimized to obtain smooth morphology, high phase
purity, and superior electrical and microwave properties (44).
Sputtering of TlCaBaCuO thin films from a sintered powder target was performed using pure argon
gas. The reason for not using oxygen and reactive sputtering is the volatility of Tl from the target. Tl readily
combines with oxygen and forms Tl2 O3 which can be easily pumped out from the chamber. The target gets
depleted of Tl much faster when oxygen is used as part of the reactive gases. The sputter-deposited thin films
were postprocessed in two steps: first, sintering in air at 850 C for 12 min to 15 min in an excess Tl2 O partial
pressure and, second, annealing in an oxygen flow of 500 sccm at 750 C for 15 min to 30 min. An excess Tl
partial pressure was maintained during the annealing process. Sintering was performed in a small box furnace
in the free surface configuration (45). Thin films were placed on a sintered pellet of Tl2223 with the film side
facing the free surface in a small covered platinum crucible. A second pellet was placed above the sample in a
platinum wire mesh support. The pellets provided the excess Tl partial pressure in the crucible to minimize the
23
loss of Tl from the thin film. Sintering of TlCaBaCuO thin film at 850 C in air for 12 min to 15 min established
the superconducting phase and morphology. At 850 C, Tl2 O3 decomposes into a liquid Tl2 O phase, which rapidly
sinters with other constituents. After sintering, the crucible was removed from the furnace and cooled rapidly.
Oxygen annealing of the air-sintered TlCaBaCuO thin film was carried out in the same configuration as used
for sintering, with an oxygen flow of 500 sccm to 1000 sccm. During oxygen annealing, thin-film grains grow
into large platelets. Oxygen annealing thus improves the electrical properties of the thin film by increasing its
oxygen content.
Another promising method for the growth of high-quality Tl-based HTS thin films is called the Tl2 O3 vapor
process or the thallination process (46). In this process, thin films of the precursor materials of BaCaCuO or
SrCaCuO are deposited using any of the thin-film deposition methods, and then the samples are heat-treated
in a Tl2 O3 vapor pressure. Tl2 O3 vapor diffuses into the precursor and reacts to form the superconducting
phases in the films. This technique is safer because the vacuum deposition chamber is not contaminated with
the Tl2 O3 vapors.
Characterization of Processed Superconducting Thin Films. The XRD provides one of the direct
characterizations of grown thin films because it gives information on the phase purity, their lattice constants,
and their crystallographic orientation with respect to the film plane. Electrical transport measurements are
performed for accurate measurement of transport current density at zero field as well as at finite magnetic
fields. The transport current measurements involve patterning a four-probe test structure on the HTS films.
The four-probe test devices are designed with line widths of 10 m to 1 mm depending on the nature of the
measurements.
The geometry of a typical test device is shown in Fig. 17. The test structure consists of two outer pads
through which a constant current source is applied and a voltage is measured across the voltage sensing lines,
which are 1 mm apart. The width of the voltage sensing should be less than the width of the line connecting
pads 1 and 2 in order to approximate a point contact as closely as possible. The test devices are patterned on
superconducting thin films using standard positive photoresist photolithography and a wet chemical etching
technique using a 1:100 phosphoric acid:DI water solution. A pulsed current is applied between the outer pads,
and the voltage across the outer sense lines is monitored. The electric field criterion of 1 V/cm is typically used
(i.e., the current through the sample at which measurement of 1 V potential difference over a 1 cm spacing
develops gives an estimate of the superconductors critical current). Also, the same test structure can be used
to determine the T c from temperature (T) dependence of resistivity (47). In such a measurement, a constant dc
current of 10 A is applied through the outer terminals, and the voltage across the sense lines is related to the
resistance of the sample. Knowing the exact thickness of the superconducting thin film and the cross-section,
the T dependence of resistivity plot can be obtained.
Structural, Electrical, and Magnetic Properties of Thin Films. High-quality YBCO and Tl2212
HTS thin films have been grown on various substrates such as YSZ, LaAlO3 (LAO), SrTiO3 (STO), CeO2 , and
MgO. Figure 18 shows an XRD scan of an annealed Tl2212 thin film on LAO substrate. The figure shows
characteristic reflections of Tl2212, Tl2223 phases as well as those of the LAO substrate. From the prominent
(0 0 l) reflections observed in the XRD spectrum, the highly c-axis-oriented growth is evident. The Tl2212
phase is the dominant one in the film, as determined from the intensities of the XRD peaks. The c-axis lattice
constant calculated from the XRD scan is found to be 29.2 . The a-axis lattice constant of LaAlO3 is 3.7801 .
The a-axis lattice constant of Tl2212 phase is 3.8503 . Thus, the lattice mismatch is less than 2%.
Figure 19 shows the temperature dependence of resistivity for one of the Tl2212 thin films deposited
on LAO, using a 50 m wide four-probe device with zero applied magnetic field. The zero-resistance T c is
approximately 100 K for the thin film. Figure 20 shows the typical zero field J c variation with temperature
obtained using a four-probe test device.J c Values at zero magnetic field as high as 5 105 A/cm2 at 77 K
and approximately 1 106 A/cm2 at 60 K were obtained. The surface morphology of the films was essentially
featureless and smooth, which is typical of high-quality films. In general, the J c of Tl2212 and Tl1223 thin
films are lower than the J c of epitaxial YBCO thin films. The presence of grain boundary weak links and
24
Fig. 17. Geometry of four-probe test devices used for electrical transport measurements. A current source is connected
between pads 1 and 2, and corresponding voltage measurements are taken across either pads 3 and 4 or pads 5 and 6.
weak flux pinning in thallate thin films may be the main reasons for the lower J c compared to epitaxial insitu-grown YBaCuO thin films. However, among the polycrystalline HTS, Tl2212 and Tl1223 thin films have
shown superior electrical properties; hence, such films are very attractive for electronic applications (48). insitu processing of Tl2212 and TlPb1223 superconducting thin films has not been highly successful owing to
the complex chemistry of the thallic oxides. Tl2212 and Tl1223 thin films with very low surface resistance (at
least two orders of magnitude below the oxygen-free copper conductor) at frequencies as high as 12 GHz and a
current density greater than 106 A/cm2 are routinely obtained (48).
Processing of Superconducting Wires and Tapes

HTS Wires and Tapes, First Generation. The Bi2223 compound is perhaps the more attractive
compound for processing conductors because its T c (110 K) is higher than that of the Bi2212 compound (90
K). Surprisingly, the Bi2212 compound has better magnetic field dependence of J c at 4 K, primarily due to
better electromagnetic connectivity. At 77 K, the Bi2223 phase clearly has higher J c s compared to the Bi2212
phase. The best BSCCO HTS conductors have J c s KA/cm2 as high as 20 to 70 kA/cm2 at zero field with J c also
depending on the length of the conductor (49). Oxide powder-in-tube (OPIT) has been a widely used technique
for processing long lengths of BSCCO wires.
OPIT Method for BSCCO 2212 and 2223 Superconducting Wires. Lead (Pb)-doped Bi2223 and
Bi2212 are prime candidates for OPIT synthesis of long-length HTS wires. Of these two phases, Bi2212 is
easier to synthesize. The reaction kinetics and thermodynamics for synthesis of the pure Bi2223 phase are
quite complex and generally result in mixed phase samples, with the presence of Bi2223, Bi2212, and other
secondary phases. Partial substitution of lead for bismuth is generally found to stabilize growth of the Bi2223
25
Fig. 18. XRD reflections observed on a Tl2212 thin film on a LAO substrate. The reflections correspond to the Tl2212
phase, the Tl2223 phase, and the substrate.
Fig. 19. Temperature dependence of electrical resistivity for a 50 m wide Tl2212 thin film four-probe test device.
phase with a typical starting composition of Bi:Pb:Ca:Sr:Cu given by 1.8:0.4:2:2:3 for such growth (50). Figure
21 pictorially shows the four steps involved for the OPIT method of preparing HTS wires and tapes. In step
1, the precursor powder is prepared from a solid-state reaction (or calcination) of mixtures of Bi-, Cu-, and
Pb- oxides and Sr- and Ca- carbonates in an alumina crucible at temperatures between 800 and 860 C for
48 to 60 h in air. Another approach for synthesis of the precursor material is to mechanically alloy the
chemicals for higher packing density. In step 2, the precursor powder is filled inside a silver billet and then
drawn into fine filaments less than 1 mm in diameter. Several tens to a hundred of these filaments are stacked
in a silver tube, drawn, and then rolled into tapes 4 mm to 6 mm wide, and typically 0.2 mm thick in step 3.
Finally, in step 4, OPIT tapes are processed in the presence of a partial melt, which aids in densification and
26
Fig. 20. Typical temperature dependence of zero field current density for a postprocessed Tl2212 thin film. Measurements
were performed on the four-probe test devices.
grain growth. To create a partial melt, the sample is heated above 900 C for a short time, followed by a long
anneal of approximately 100 h at 840 to 850 C. At present, such tapes can be reliably manufactured in several
kilometer scale lengths. The main process parameters involved in the synthesis of wires thus include precursor
material synthesis, mechanical processing, and heat treatment conditions. The precursor composition, particle
size, and density are important factors in the quality of precursor material synthesis (51). Billet dimensions,
filament restacking geometry, and mechanical properties of the tape are the important mechanical aspects for
wires and tapes of HTS. Heating rate, sintering temperature, sintering time, and cooling rate are important
parameters for heat treatment conditions that control electrical and mechanical properties of drawn wires.
Heat treatment is performed after cutting wires to required lengths. Each filament is embedded in a silver
matrix, which adds mechanical strength and flexibility. A threshold value for silver content is required to
enhance electromagnetic grain connectivity, essential for long length conductors. At the same time, one needs
to maximize the superconducting volume fraction so that the HTS rather than the silver present in the
conductor provides the percolative path for charge carriers. Fill factor is one of the critical parameters in HTS
wires, defined as the ratio of the volume of the superconductor core to the entire volume of the conductor.
Typical fill factor in most HTS conductors to date is below 60% (51). Magneto-optic studies indicate that most
of the current transport occurs at the interfacial region between silver and the bulk BSCCO rather than in the
bulk BSCCO (52). The interfacial region with well-aligned grain structure appears to be the key to improve J c s
in these wires. The Bi2212 phase is attractive for low-temperature applications at 4 K because it is economical
to produce long length multifilamentary conductors (52).
HTS Wires and Tapes: Second Generation, YBCO-Coated Conductors. Primarily because of
the higher J c s demonstrated in biaxially textured thin films;. YBCO has enormous advantage over BSCCO for
wire applications. One approach is to grow YBCO on a biaxially textured chemically compatible buffer layer.
A processing technique proposed for biaxial texturing of chemically compatible buffer layers on untextured
substrates such as alloys of nickel is called the ion-beam-assisted deposition (IBAD) (53). Another approach
proposed is the biaxial texturing of the substrates. It is well known that metals when deformed into wires
27
Fig. 21. Schematic diagram showing the general OPIT process used to produce BSCCO tapes, developed by American
Superconductor Corporation. Courtesy of 15.
develop alignment along preferred directions. Recent work at Oak Ridge National Laboratory has shown that
biaxially textured [i.e., both in-plane and out-of plane alignments (ab)] nickel can be produced over long lengths
(54). Such substrates are called rolling assisted biaxially textured substrates (RABiTS) (54). An interesting
example of biaxial texturing of metallic sheets already used in electrical engineering is in silicon-doped steel
stampings used in the electric power industry, in which texturing results in the easy direction of magnetization
parallel to the length of sheets, reducing hysteresis losses.
The IBAD process is used primarily for deposition of buffer layers such as yttrium-stabilized zirconia
(YSZ), or cerium oxide (CeO2 ) on untextured nickel alloy tapes (53). The IBAD process is a thin-film deposition
process developed by IBM for preferential growth of biaxially textured buffer layer material on untextured
substrates. In this process, a rare gas ion beam is used to bombard the buffer layer along a preferred crystallographic direction as it is being deposited. The angle of incidence of the ion beam with respect to the plane
of the substrate is the critical parameter for the preferential orientation of the buffer layer grown. Typical
buffer layer of choice for IBAD has been YSZ, grown to a thickness of 200 nm to 500 nm. A thin layer of CeO2
is interposed between YSZ and YBCO for better lattice matching. Subsequent growth of postprocessed YBCO
thin or thick films on the biaxially oriented buffer layers have yielded high J c s > 105 A/cm2 at 75 K and in
magnetic fields up to 5 T. Figure 22 shows the J c versus magnetic field characteristics obtained for a short
sample of a 1.6 m thick YBCO/IBAD-YSZ/Ni alloy tape, at 75 K (53). The figure also shows the magnetic field
dependence of J c for a proton-irradiated Bi-2223 OPIT wire for comparison. The figure clearly shows the vastly
improved performance of the IBAD tape compared to the best OPIT wire. An advantage of the IBAD-buffered
HTS tapes compared to OPIT wires is that the tapes can be wound with the HTS in compression, attractive for
electromagnetic applications such as electric motors, and generators. Current efforts are underway to scale-up
the process for long lengths necessary for commercialization.
The Rolling assisted biaxially textured substrates (RABiTS) process was developed by Oak Ridge National
Laboratory for epitaxial growth of HTS thin films over long lengths. Figure 23 schematically shows the process
for the development of RABiTS (54). Nickel is the primary candidate for the RABiTS because of its higher
oxidation resistance compared with copper. Also, thermomechanical texturing of Ni gives a cube-textured substrate. A typical YBCO thin-film-coated conductor fabrication process involves thermomechanical processing
of Ni substrate (125 m), followed by PLD of buffer layers of CeO2 (0.2 m), YSZ (0.2 m), and finally 1 m
thick YBCO film PLD deposited at 780 C in an oxygen pressure of 185 mTorr (54). After deposition, the film
is cooled at 10 C per min, in an oxygen pressure of 700 mTorr at 400 C. All oxide buffer layers are grown at
temperatures of 780 C. RABiTS of lengths up to 100 feet (only the substrates with no HTS) have been processed
28
Fig. 22. J c versus applied magnetic field for a 1.6 m IBAD tape as a function of magnitude and orientation with respect
to the tape plane, compared with a proton-irradiated OPIT BSCCO 2223 wire. Courtesy of 14.
Fig. 23. Schematic representation of the RABiTS process developed at Oak Ridge National Laboratory. Starting with a
randomly oriented Ni bar/plate, cold rolling is used to produce a distinct copper-type rolling texture. This is followed by
recrystallization to a cubic structure. Epitaxial metal and/or oxide buffer layer(s) are then deposited on the textured Ni.
Courtesy of 54.
already.J c values of a short YBCO thin-film-coated conductor are shown in Fig. 24. For comparison, thin films
of YBCO on LAO and thin films of Tl2212 and 1223 on LAO are also shown. Because PLD and sputtering
techniques are not scalable, MOCVD may be the primary technique for long-length HTS-coated conductors on
RABiTS. The MOCVD-grown YBCO on RABiTS has yielded zero field J c as high as 0.64 MA/cm2 at 77 K.
29
Fig. 24. The critical current density versus magnetic field for YBCO thin film on a RABiTS substrate of configuration
ML3 corresponding to YSZ(110) (0.5 m)Ni(100) (125 m). Data for both Hc and Hab are shown. Also shown are data
for YBCO thin film on STO, Tl2212, and Tl1223 thin films on LAO substrates. For Hc, the high-field properties of YBCO
thin films on RABiTS are better than those on STO. Courtesy of 54.
Applications of High-Temperature Superconductors

Bulk superconductors can be used in applications such as high Q cavities for microwave applications see Superconducting microwave technology, frictionless bearings, magnetic levitation (see Superconducting levitation),
and fault current limiters (see Superconducting fault current limiters). Melt-textured as well as single crystal
bulk YBCO conductors have been used as a microwave cavity with very high Q values greater than 10,000 at
77 K and at frequencies below 20 GHz. Progress in bulk single-crystal growth could be attributed to newer
crucible materials such as BaZrO3 (55). The use of BaZrO3 crucible precludes impurities such as Al, Zr in melts
or crystals for the growth of 123 rare-earth-based HTS.
Thin film HTSs are most attractive for microwave applications because of their low surface resistance at
frequencies upto 35 GHz (see Superconducting microwave technology and Superconducting filters and passive
components). A large number of microwave components have been demonstrated using HTS technology, (e.g.,
high Q filters resonators, phase shifters, delay lines). Epitaxial growth of HTS thin films on low-loss microwave
substrates such as MgO and sapphire (with a CeO2 buffer layer) has been key for high Q microwave components.
A promising new application in this area involves integration of HTS thin films and ferroelectric thin films
for electrically tunable components (56). Tunable filters, resonators, and phase shifters have been successfully
demonstrated. We have recently reported a two-pole tunable bandpass filter using a thin-film YBCO/STO/LAO
multilayer microstrip configuration. The structure consists of a LAO substrate (254 m thick) on which an
epitaxial STO thin film (0.3 m thick) was deposited by PLD, followed by a YBCO thin film (0.35 m thick)
deposited by PLD. The two-pole filters were designed for a center frequency of 19 GHz and 4% bandwidth.
Tunability is achieved through the nonlinear dc electric field dependence of relative dielectric constant of
STO thin films. Figure 25 shows the swept frequency response of a tunable filter at 77 K, for bipolar bias
voltages between 0 and 500 V (56). Note that the insertion and return losses are reduced with applied bias.
Although high voltages are required, the power consumption is low owing to negligible currents through the
high dielectric constant ferroelectric thin film.
Another area of electronics in which HTS thin films have a niche is in the use of HTS SQUIDs (see
SQUIDS) as magnetic field sensors. A SQUID operates on the principle that the critical current through two
Josephson junctions connected in a loop is a function of the magnetic flux threading the loop. Since each SQUID
30
Fig. 25. Electric field tunability of a tunable bandpass filter using a thin film YBCO/STO/LAO multilayered microstrip
structure. The filter uses the electric field dependence of the nonlinear dielectric constant of the STO ferroelectric thin film.
stores multiples of the basic flux quanta, SQUIDs are the most sensitive sensors of magnetic fields at levels of
10 14 T at 77 K, thus making possible measurements of magnetic fields produced by current flow in a human
brain, about 10 13 T (see SQUIDS). The key to the HTS SQUID technology is the reproducible fabrication of
grain boundary Josephson junctions. This technology has currently matured to the level of commercial usage
in magnetic imaging for medical diagnostics such as a magnetoencephalography (MEG), mapping the brains
magnetic activity. Another area of application for HTS thin films is in interconnect technology for semiconductor
multichip modules. With the promise of high-quality YBCO thin films on various buffer layers such as CeO2 ,
and YSZ, it is feasible to conceive of applications of HTS thin films on silicon substrates.
Applications such as nuclear magnetic resonance (NMR) require uniform magnetic fields over large
volumes. Nuclear magnetic reasonance requires a field level of 10 T to 20 T. The NMR technique uses RF
and magnetic fields to study transitions between nuclear spin states. An RF coil surrounds a sample, placed
inside a magnetic field. The magnetic field is swept through a small range with a separate coil. A secondary
RF coil (may be a superconducting coil) picks up the resonance signal. Use of superconducting coils in magnet
technology has made possible high magnetic fields with minimal energy dissipation. Use of superconducting
lines for power transmission is attractive for energy efficiency. The current densities needed for the utility
power lines are considerably lower than most other power applications. It is estimated that current densities
below 104 A/cm2 at 77 K are sufficient for utility power lines. The Bi-based superconductors are the most
attractive for these applications. The design of underground cables cooled to liquid nitrogen temperatures is a
challenging aspect because these superconductors are detrimentally affected by moderate mechanical stresses.
Superconducting magnetic energy storage (SMES) is an energy storage device wherein energy is stored in the
magnetic field associated with circulating currents in superconductors. An SMES system consists of an ac/dc
31
power converter to charge a large torroidal magnet. When needed, the energy stored in the magnetic field can
be tapped and converted by ac power. Again HTS wires are needed for the large magnets used in such a system.
Other applications attractive for HTS wires and tapes include power devices such as motors, generators and
transformers. With the current technology, several prototypes of HTS wires and tapes have been demonstrated.
American Superconductor Corporation had tested a 125 hp ac synchronous motor built using a BSCCO 2223
composite tape. A 5 kJ HTS SMES magnet, a high-current pancake coil (100 A), and current development of
500 and 1000 hp motors indicate that HTSs have potential for large-scale power applications (49). Table 3.
shows some of the HTS companies which provide superconducting materials and fabrication services.
BIBLIOGRAPHY
1.
2.
3.
4.
5.
6.
7.
W. Meissner R. Oschenfeld Ein Neuer Effekt bei Eintritt der Supraleitfahigkeit, Naturwiss., 21: 787, 1933.
F. London H. London The electromagnetic equations of the superconductor, Proc. R. Soc. A, A149: 7188, 1935.
F. London Superfluids, Macroscopic Theory of Superconductivity, vol. 1 New York: Wiley, 1950, pp. 3-4.
J. Bardeen L. N. Cooper J. Schreiffer Theory of superconductivity, Phys. Rev., 108 (5): 11751204, 1957.
B. D. Josephson Possible new effects in superconductive tunneling, Phys. Lett., 1 (7): 251253, 1962.
J. G. Bednorz K. A. Muller Possible HTS superconductivity in the BaLaCuO system, Z. Phys. B, 64: 189193, 1986.
M. K. Wu et. al. Superconductivity at 93 K in a new mixed phase Y-Ba-Cu-O compound system at ambient pressure,
Phys. Rev. Lett., 58 (9): 908910, 1987.
8. H. Maeda et al. A new HTS oxide superconductor without a rare earth element, Jpn. J. Appl. Phys., 27 (2): L209210,
1988.
9. Z. Z. Sheng A. M. Hermann Bulk superconductivity at 120 K in the T1-Ca-Ba-Cu-O system, Nature, 332 (6160):
138139, 1988.
32
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.

S. N. Putilin et al. Superconductivity at 94K in HgBaCuO, Nature, London, 362: 226, 1993.
J. D. Doss Engineers Guide to High Temperature Superconductivity, New York: Wiley, 1989.
A. A. Abrikosiv On the magnetic properties of superconductors of the second group, Soviet Phys. JETP, 5: 1174, 1957.
R. P. Huebener et al. Thermoelectric and thermomagnetic effects in high temperature superconductors, Proc. Materials
Res. Soc. Symp., 1992; pp. 1322.
P. M. Grant Superconductivity and electric power: Promises, Promises . . . Past, present, and future, IEEE Trans. Appl.
Supercond., 7: 112135, 1997.
T. P. Sheehan Introduction to High Temperature Superconductivity, New York: Plenum, 1994.
H. A. Blackstead J. D. Dow Implications of Abrikosov-Gorkov exchange scattering for theories of high temperature
superconductivity, Phys. Lett. A, 206: 107110, 1996.
D. C. Larbalestier Road to conductors of high temperature superconductors: 10 years do make a difference!, IEEE
C. Blue P. Boolchand In-situ preparation of superconducting YBCO 123 thin films by on-axis rf magnetron sputtering
from a stoichiometric target, Appl. Phys. Lett., 58: 20362038, 1991.
S. Tanaka Reviews, prospects and concluding remarks. Materials needs for applications, Physica C, 282287: xxxi
xxxix, 1997.
M. A. Kirk H. W. Weber Electron microscopy investigations of irradiation defects in the high T c superconductor YBCO,
in Studies of High Temperature Superconductors, vol. 10 Commack, NY: Nova, 1992, p. 243.
S. Jin et al. Melt textured growth of polycrystalline YBCO with high transport J c at 77K, Phys. Rev. B, 37: 78507853,
1988.
S. S. P. Parkin et al. Bulk superconductivity at 125 K in Tl2 Ca2 Ba2 Cu3 Ox Phys. Rev. Lett., 60 (24): 25392542, 1988.
M. Greenblatt et al. Chemistry and superconductivity of thallium based cuprates, Studies of the HTS Superconductors,
vol. 4 Commack, NY: Nova, 1992.
M. Murakami et al. Melt processing of bulk YBCO superconductors with highc , J. of Eng. Materials Tech., 114: 189,
1992.
M. Murakami et al. Melt processing of bulk high Tc superconductors and their applications, IEEE Trans. Magn., 27:
1479, 1991.
V. R. Todt et al. Processing of large YBCO domains for levitation applications by a NdBaCuO seeded melt growth
technique, J. Electron. Materials, 23: 1127, 1994.
K. Salama S. Sathyamurthy Melt texturing of YBCO for high current applications, Appl. Superconductivity, 4 (1011):
547561, 1996.
Z. Sheng Tl-based high T c superconductors, Mater. Sci. Forum, 130132: 134, 1993.
G. Subramanyam et al. Processing and physical properties of single phase TlBaCaCuO (2212) and (Tl-Pb) (Sr-Ba)CaCuO
(1223) superconductors, Appl. Superconductivity, 4: 591598, 1996.
T. Kamo et al. Appl. Phys. Lett., 59: 31863188, 1991.
C. P. Bean Magnetization of high field superconductors, Rev. Mod. Phys., 36: 3139, 1964.
W. M. Reiff Magnetic susceptibility measurements: An important facet of modern solid-state characterization, Amer.
Lab. 26: February 1994.
L. H. Greene B. G. Bagley Oxygen stoichiometric effects and related atomic substituitions in the high T c cuprates, in
Physical Properties of High Temperature Superconductors, vol. II, Singapore: World Scientific, 1990, pp. 509569.
R. Sugise, et al. Jpn. J. Appl. Phys., 27: L2310, 1988.
B. Raveau et al. Layered thallium cuprates: Non-stoichiometry and superconductivity, Solid State Ionic. 39: 4962,
1990.
F. Shi et al. Metal (M) dopant centered local structures, high-pressure synthesis and bulk superconductivity in
YBa2 (Cu x M)3 O7 : MFe, Co, Ni, J. Condens. Matter, 9: L307L313, 1997.
A. Rykov V. Caignaart B. Raveau Quadrupole interactions and vibrational anisotropy of tetragonal Fe (111) in the 123
derivative LnSrCuGa1 x Fex O7 (Ln = Y, Ho), J. Solid State Chem., 109: 295, 1994.
J. Qiao C. Y. Yang High Tc superconductors on buffered silicon: Materials properties and device applications, Materials
Sci. Eng. R14: 157202, 1995.
D. C. Payne J. C. Bravman (eds.) Laser ablation for material synthesis, Materials Research Society (MRS) Fall Meeting,
MRS Symp. Proc., Vol. 191, Boston, 1990.
33
40. D. S. Ginley Thallium films for microelectronic applications, in A. M. Hermann, and Y. Yakhmi (ed.), Thallium based
superconducting Compounds, Singapore: World Scientific, 1994.
41. W. Holstein et al. Superconducting epitaxial Tl2212 films on sapphire with cerium oxide buffer layers, Appl. Phys. Lett.,
61 (8): 982984, 1992.
42. J. Zhao P. Norris Metalorganic chemical vapor deposition of YBCO 123 thin films, Mater. Sci. Forum, 130132: 233254,
1993.
43. G. Braunstein et al. Process of formation and epitaxial alignment of SrTiO 3 thin films prepared by metal-organic
decomposition, J. Appl. Phys., 73: 96170, 1993.
44. G. Subramanyam et al. Studies on sputtered TlCaBaCuO high T c superconducting thin films for microelectronics
applications, Materials Sci. Forum, 130132: 613632, 1993.
45. D. S. Ginley et al. Morphology control and high critical currents in superconducting thin films in the Tl-Ca-Ba-Cu-O
system, Physica C, 160: 4248, 1989.
46. B. Johs et al. Preparation of high Tc Tl-Ba-Ca-Cu-O thin films by pulsed laser evaporation and Tl2 O3 vapor processing,
Appl. Phys. Lett., 54: 18101812, 1989.
47. W. Pelster Accurate automated measurement of superconductor material resistance, Res. & Develop., 30: 8084, 1988.
48. DuPont Superconductivity, Technical Note on Tl based superconducting thin films, Wilmington, DE, 1998.
49. A. P. Malozemoff Q. Li. S. Fleshler Progress in BSCCO-2223 tape technology, Physica C, 282287: 424427, 1997.
50. U. Balachandran et al. Processing and fabrication of high T c superconductors for electric power applications, Appl.
Superconductivity, 5 (1-6): 187191, 1997.
51. V. Selvamanickam et al. High temperature superconductors for electric power and high energy physics, J. Metals, 50
(10): 2730, 1998.
52. S. Brody The last days of BSCCO?, Superconductor and Cryoelectronics, featured article on BSCCO wires, 11: 2630,
Spring 1998.
53. Y. Ijjima et al. In-plane aligned YBCO thin films deposited on polycrystalline metallic substrates, Appl. Phys. Lett., 60:
769771, 1992.
54. A. Goyal et al. Epitaxial superconductors on rolling assisted biaxially textured substrates (RABiTS): A route towards
high critical current density wire, Appl. Superconductivity, 4 (1112): 403427, 1996.
55. A. Erb E. Walker R. Flukiger The use of BaZrO3 crucibles in crystal growth of the high T c superconductors: Progress
in crystal growth as well as in sample quality, Physica C, 258: 920, 1996.
56. F. A. Miranda et al. Tunable microwave components for Ku and K-band satellite communications, Integrated Ferroelectrics, 22: 269278, 1998.
GURU SUBRAMANYAM
University of Dayton
PUNIT BOOLCHAND
University of Cincinnati
SUPERCONDUCTING MAGNETS,
QUENCH PROTECTION
spread of quench in potted, pool-boiling, and cable-in-conduit

superconductors (CICC). Because of the difficulties in predicting disturbances, initial quench zones, and quench propagation, a conservative design criterion is to assume that local
hot spots are adiabatic and that all energy must be dumped
externally. For internal quenching, the maximum time to
heat a long quench zone with a cowound or surface heater has
to be known.
The problem of magnet quench protection is a subset of
the generic problem of magnet protection, (1) during normal
operation, (2) off-normal conditions such as quench, and (3)
faults, whether in the coil, bus, or power supply. The two
most fatal flaws are mechanical rupture and electrical arcing.
They are often preceded by excessive displacement and/or
partial electrical discharges and leakage currents. Flaws that
are sometime repairable can end a magnets useful life when
cracks cause leakage of helium under pressure, leakage current causes enough heat to quench the magnet, or displacements create unacceptable field errors. All of the structural
design may be considered as part of the magnet protection
design. This subject is too vast to be treated here, but should
be discussed for specific applications in the articles on Superconducting Magnetic Energy Storage, motors/generators, fusion, and Magnetic Resonance Imaging magnets. The design
for electrical integrity will be discussed here, since it isnt discussed elsewhere and because the internal and external voltages during a quench dump are usually significantly higher
than those during normal operation.
Finally, we briefly review the actual history of failures to
protect magnets. Case histories provide a cautionary tale: this
article can only go so far in helping to protect magnets, since
most failures are caused by mental lapses that are frequently
not design errors.
Superconducting magnets are subject to thermal instability,

leading to the loss of superconductivity, known as quench, in
which the critical values of field, temperature, and current
density are exceeded and fail to recover. This phenomenon
generally begins in a localized region of the coil, then spreads
to the rest of the magnet or magnet system with a quench
propagation velocity. All magnets use some form of composite superconducting wire, in which superconducting filaments
carry current in parallel with normal conductor, known as the
stabilizer. The stabilizer has the dual purpose of preventing
quench in the face of disturbances and of protecting the magnet from excessive temperatures and pressures when unwanted quenches occur. Following a quench, current rapidly
transfers from the superconducting material to the stabilizer,
since composites are always designed so that the resistance
of the stabilizer is much less than that of the superconducting
material in its normal state.
In the case of a quench, it is almost never possible to allow
continued operation at constant current without unacceptable
temperature rises in the magnets. The current must be reduced to zero and the stored energy in the magnet eliminated,
either by absorbing the energy within the magnet or by
dumping the energy externally. If the energy is dumped externally, it may be absorbed in either warm, generally room
temperature, dump circuits, or in cold, generally liquid helium temperature, dump circuits. Some magnets have sufficient enthalpy in their stabilizer and magnetically coupled
passive structures to absorb their own stored energy without
excessive temperature or pressure rises. In fact, most magnets have sufficient energy, if the ratio of peak local heating
to average global heating can be held to acceptably low levels.
The peak/average ratio can be controlled by design either by
activating internal resistive or inductive heaters or by rapidly
dumping or heating all helium coolant in order to guarantee
that large portions of the coil will heat up together. The other
method of protecting magnets is to dump the magnet energy
into an external resistor. This requires an absolutely reliable
method of interrupting current flow from the power supply
and diverting it into the dump resistor. Both methods require
reliable and rapid detection of a quench.
The fundamental limit on protecting magnets against
quench is the detection of that quench. This can be particularly difficult in the case of coils in an electromagnetically
noisy environment, pulsed coils, and multicoil systems. It is
also difficult when the coils are very conservatively stabilized,
as is often the case for very large coil systems and buswork.
Quench detection systems can be active or passive. Active systems usually involve some sort of balanced voltage bridge.
The signal/noise ratios of voltage bridges can be improved by
using cowound sensors and active cancellation. Passive systems use transformer-fed heaters to trigger superconducting
switches or voltage thresholds to trigger cold diodes.
In order to size a magnet for protection, it is usually necessary to know something about the physics of quench propagation. Different physical theories are needed to predict the
COIL PROTECTION CIRCUITS

When a superconducting magnet quenches, all of its magnetic
energy is converted into heat. If a magnet has enough total
mass to absorb the heat and is small enough to guarantee
that a quench will propagate into a large fraction of the magnet, then no protection circuits are needed, except to disconnect the power supply, when the current isnt freely circulating. However, when the magnet is too large and stores too
much energy to guarantee completely passive protection,
some active measure must be taken. The main distinctions
between the most commonly used coil protection circuits are
whether the dump resistors are internal or external to the
cryostat and whether the quench detection is triggered by active logic or passive breakdown of a switch. External dump
resistors usually correspond to the design philosophy of saving the magnet by depositing almost all of the energy into
a large, inexpensive structure at room temperature. Internal
dump resistors are usually designed to lower the peak local
to average heating of the magnet to a manageably low level.
External Dump
Neumeyer has recently reviewed the external quench protection circuits for superconducting magnets (1). He schematizes
1
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
SUPERCONDUCTING MAGNETS, QUENCH PROTECTION
Mi
OS
CS
Lstray
To power supply
Li
DS
Ri
SR
OS
Rbus
Figure 1. Simple dump circuit schematic [Neumeyer et al. (1)].
PS
CS
Lcoil
the basic external dump circuit as shown in Fig. 1. The magnet is represented by an inductance L, while the mutual inductance M, and the coupled inductance and resistance, Li
and Ri, represent the sum of all coupled magnets and passive
conducting structures in the magnet system. The external
dump circuit consists of a power supply, a closing switch CS,
to shunt out the power supply during quench dump, an opening switch OS to interrupt magnet current, and a dump resistor R (). The basic principle is that the dump resistor is
much, much larger than the resistance of the magnet normal
zone, so that almost all of the energy is deposited, at room
temperature, in a resistor sized to safely absorb all of the
magnet energy.
Leads to the dump resistor should generally be coaxial, in
order to minimize the voltage overshoot, due to Lleads dI/dt.
Several equivalent simple dump circuits may be used, as
shown in Fig. 2. Under normal operation, they all have the
same effect on the magnet. The tradeoffs are in cost versus
reliability and in the specifics of the power supply and magnet
grounding system. A magnet ground/interrupter switch configuration should be selected that allows the switch and magnet to float on a single short to the magnet case ground without drawing large fault currents.
Iload
Figure 3. Typical Artificial Zero Counterpulse Circuit [Neumeyer et

al. (1)].
Neumeyer represents a typical counterpulse interrupter

with the circuit schematic shown in Fig. 3.
The operating coil current flows through the power supply
(PS) and the normally closed switch. When a quench is detected, the counterpulse capacitor C is discharged by closing
the switch DS, producing a current zero in OS. The inductor
SR is a saturable reactor, which desaturates near current
zero, decreasing the dI/dt. This helps to extinguish arcs or to
restore voltage-holding capability in a solid-state switch. The
key parameter, set by selecting the counterpulse capacitor, is
the time during which current must be reversed and held
near zero. This is on the order of 10 s for vacuum bottle
interrupters, 5 s to 50 s for thyristors, and 50 s to 200 s
for air-blast interrupters. The faster the interrupter clears,
the less expensive the counterpulse circuit. For solid-state
DCCB
Dump
DCCB
Magnet
2a R series, DCCB NC
Power
supply
Switch
Switch
Magnet
Dump
Figure 2. Equivalent superconducting

magnet quench dump circuits.
2c R parallel, DCCB NC,

Switch NO
Magnet
2b R parallel, DCCB NC
Switch
DCCB
Power
supply
Dump
Power
supply
Dump
Power
supply
2d R parallel, DCCB NC,

Left switch NC, right switches NO
Magnet
switches with antiparallel diodes, the energy stored in the

commutating capacitor must be
1 V (0)I(0)
1
CV 2 = toff
( 2)
2 c
2
sin
(1)
where toff is the specified time during which reverse voltage is

maintained across the solid-state switch (s), is the phase
angle in radians when the switch current is zero, and V(0)
I(0) R.
The alternative to counterpulsed circuits are dc switches
that can sustain a high enough voltage to force current zero
without a resonant pulse. Here, the opening switch develops
a high enough arc or resistive voltage to drive the device current close enough to zero to extinguish itself and shunt the
current to a dump resistor. The most commonly used forced
current zero devices are air-blast breakers and explosive
switches. The air-blast breakers use a blast of compressed air
and rapidly parting contacts to create a very long, contorted
arc with high-voltage and a high tendency to quench. The
blast also cools and further constricts the arc. Explosive
switches use a redundant number of explosive charges and
small arcs with a moderately high voltage in each arc. Both
types of switch can also be counterpulsed to increase the probability of current interruption. However, even in a forced-current zero circuit with no capacitor, an inductor is still needed
as a snubber, in order to limit the dV/dt rise across the superconducting magnet, as discussed next.
Interrupters. In order to protect a superconducting magnet
with an external dump resistor, an absolutely reliable current
interruption switch is needed. Several current interruption
switches have been used in magnet design for quench dump
or other high voltage pulses. Whatever technology is selected,
the interrupters will usually have two opening switches in
series in order to provide adequate reliability. The dump
strategy will then be either to open both series switches at
once or to detect a failure to open in a nondestructive opening
switch (e.g., a solid-state switch), then to open a more reliable
but destructive switch (e.g., a series explosive switch). An interrupter may also include a switch to shunt conductor current during normal operations in order to reduce the steadystate ampacity requirement of the main interrupter. In this
case, the high-current, inexpensive, normally closed mechanical switch transfers current into the quench dump interrupter
after quench detection. The interrupter then carries the magnet current only as long as is needed to open and transfer its
current to the external dump resistor.
Interrupting switch technologies include:
1. Thyristor breakers with counterpulse circuits

2. Mechanical breaker (air, air blast, vacuum, vacuum and
magnetic field)
3. Explosively actuated breaker (fuse and fuseless)
4. Water cooled fuses (activated by water flow interruption)
5. Gate turn off (GTO) thyristor breaker
6. Insulated gate bipolar transistor (IGBT) switches

7. Superconducting switches
Turn-on switches that are used for the counterpulse circuit include:
1. Ignitrons
2. Thyristors
3. Vacuum switches
In the past, mechanical interrupters were favored for large
magnets, because of the high power handling capabilities of a
single device (e.g., up to 73 kA 24 kV in the JET air-blast
interrupter (2). However, with mechanical interrupters, the
inevitable electrode erosion by current interruption arcs tends
to limit the number of reliable operations to 104 operations
with periodic maintenance every 103 interruptions. The probability of failure (to interrupt current) in a mechanical interrupter has also been typically 103 to 104 at best, although
this is clearly dependent on the specific design. For example,
Yokota reported vacuum bottle tests in which there were five
interruption failures in 10,000 interruptions with single bottles and no failures in 10,000 with two seriesed bottles (3).
Clearly, any degree of reliability can be achieved with mechanical switches with adequate redundancy and maintenance. However, high-reliability requirements are usually
met by the use of solid-state devices. Because they have no
moving parts and are erosion-free, their lifetimes can easily
exceed 106 operations, limited only by thermal fatigue. While
individual solid-state devices used to be limited to the range
of 1 kV 1 kA, it is now possible to purchase thyristors with
ratings of 6 kA 6 kV. In the case of thyristor solid-state
switches, the reliability of interruption will probably be set
by the counterpulse circuit with the failure mechanism being
either capacitor burnout or failure to close of the counterpulse
circuit switch. The counterpulse circuit and its reliability limitations can be eliminated by the use of GTOs. They have always had less power-switching capability than normal thyristors, but are currently available with ratings of 3.3 kV (4
kA, turn-off, 1.2 kA, ss). A new technology, IGBTs, is beginning to be used in high-power applications requiring fast
switching, with device ratings up to 3 kV (1.2 kA, turn off
400 A, ss). IGBTs can be switched an order of magnitude
faster than conventional thyristors, making them useful in
switching converters that reduce the amount of filtering or
voltage ripple on the magnets.
Explosive fuses are now capable of operating with very
high reliability. They are inexpensive and incorporate redundancy in a single unit by including several in series explosive
charges and arcs. They are frequently counterpulsed for further redundancy. Explosive fuses have the particular problem
that they wont interrupt currents below a certain level. They
are therefore most appropriately used in quasi-steady-state
operations, in which low current quench is highly unlikely, or
in those magnets that can guarantee passive internal absorption of the quench energy. They also favor applications where
quench is highly unlikely, because they can only be used once.
A clever hybrid was proposed by Kuchinski to minimize
the total cost of mechanical and solid-state interrupters (4),
S1
S2
T1
S3
+
C1
T2
+
C2
Cold zone
VS
Figure 4. An arc-free current interrupter with pulse-rated solidstate components.
as shown in Fig. 4. This circuit allows all of the solid-state

components to be pulse rated, while eliminating arcing in all
of the mechanical components. S1 and S2, low-voltage mechanical switches, carry the magnet operating current. To interrupt, the thyristor T1 fires, suppressing any arcs, while S1
opens. T2 fires to initiate the counterpulse through T1 that
turns it off. With D and T2 carrying magnet current with a
low voltage drop, S2 opens. Then the vacuum switch is ignited
to provide the reverse counterpulse through T2 to switch it
off. All switches are interrupted and magnet current flows
through the dump resistor.
Dump Resistors. The most common dump resistor is a meander of steel bars. Alternative dump resistor concepts include:
Dump Resistor
Steel bars in air
Steel bars or ribbon in water
(5)
Liquid rheostats (6)
Voltage clamps (Zeners,

MOVs, ZnO)
Advantages
Simplicity, cost, maintainability
Simplicity, energy density
Energy density, elimination
of solid structure; Disadvantage: negative temperature coefficient
Faster dump for fixed peak
voltage; Disadvantage:
High cost/Joule
The voltage across a linear resistor declines with the current. Since electrical integrity is limited by voltage, it would
be more efficient to use a resistor that discharged at constant
voltage. With a perfect voltage source, either the peak dump
voltage could be lowered by one-third or the dump time could
be improved one and a half times. This can be approximated
by highly nonlinear resistors, such as Zener diodes, Metal Oxide Varistors (MOVs) or Zinc Oxide (ZnO) arresters (7). At
very high energy levels, these are prohibitively expensive. An
inexpensive alternative with a useful degree of nonlinearity
is stainless steel, which has a resistance temperature coefficient of 0.001/K. If the temperature of a stainless steel resistor is allowed to rise 500 K by the end of a dump, its resistance will have increased by 50%. With nickeliron alloys, the
resistance can be quadrupled by the end of a pulse (1).
External Quench of Multiple Magnets. Both magnets and
switches are limited in voltage and current. When a magnet
Ri
Figure 5. Series interleaf protection circuits [Dudarev et al. (8)].
system becomes too large, the dump circuit must subdivide

into several parallel, series, or independent circuits. A particularly elegant circuit topology is that of the series interleaf,
used in the tokamak systems T-15 and the Tokamak Fusion
Test Reactor (TFTR), a normal magnet system. The series interleaf connection is shown in Fig. 5 (8).
This circuit has two advantages over independent or parallel protection circuits. With the interleaf, the voltage drops
alternatively up and down, because of the alternation of superconducting inductors and external resistors, as shown in
Fig. 6. This prevents high voltage from building up through
the coil system, as it would do if there was only a single dump
resistor. If the large coil system had simply been broken into
the same number of independent or parallel dump circuits,
there would be a possibility of unbalanced currents and
forces, during a quench dump. With a floating power supply,
the interleaf circuit also prevents unbalanced forces during a
single ground fault.
Power
Cold
switch
supply
Ri
Li
Cold
Zone
Um
Figure 6. Series interleaf circuit dump waveform [Dudarev et al.

(8)].
Alternative options for dumping energy in a mutually coupled multicoil system are summarized in the table.
Dump Strategy
Advantages
Disadvantages
Dump all coils in series
Simplicity, no unbalHigh terminal voltage

anced currents or
forces even with single
ground fault
Dump all coil in series
No unbalanced currents As many VCLs and
interleaf
or forces even with sindump circuits as intergle ground fault; low
leafs
voltage
Dump all coils in paral- Low voltage; indepenPossibility of unbalanced
lel; dump all coils indedent control
currents and forces;
pendently
large number of VCLs;
least reliable for given
component reliability
Dump faulted coil exter- Smallest refrigeration
Overcurrents in persisnally; other coils rerequirement for recool
tent magnets when
main in persistent
coupled to dump coil
mode (ORNL EPR)
Internal Dump
Magnets are dumped internally when it is desirable to eliminate helium loss through the vapor-cooled leads and when
terminal voltages have to be severely limited, as with commercial products, such as MRI magnets, where users have to
be in the same room as the magnet. The least aggressive
action that is typically taken with a pool-boiling magnet is
simply to disconnect the superconducting coil from the powersupply and allow its current to die down slowly as it freewheels through a superconducting switch or normal joint.
This would be an appropriate response to a low liquid-level
reading, in which an active dump might do more harm than
good.
Cold dump resistors may still be either internal or external
to the magnet. That is, a large fraction of the energy can be
deposited into resistors or diodes that are outside of the magnet, but inside the cryostat. However, since there is now no
benefit in refrigeration or cooldown requirements in removing
energy from the magnet, it is almost always desirable to return as much of the heat as possible to the magnet, thereby
creating longer quench zones and more uniform heat deposition with the magnet. This can be done by making the resistive element a heater, closely coupled to the outer layer of
the magnet or cowound with the magnet superconductor.
Cold Dump Circuits. In a cold dump circuit, either the interrupter or the resistor or both will be inside the cryostat at
the magnet temperature. Cold switches include superconducting switches and fuses, cold diodes, and cold transistors. Cold
resistors include cowound, insulated normal metal, surface
heaters, and power dissipation in the switch itself. One benefit of having all elements of the dump circuit cold, when the
switch is superconducting, is that the vapor-cooled leads can
be detached, allowing current to circulate losslessly in the
magnet. Cold heaters have two additional benefits, especially
when they are cowound through the entire magnet. They can
prevent hot spots by reducing the peak local/average heating,
and they can also cancel the dI/dt voltage of the magnet with
the resistive voltage, greatly reducing internal voltages in the
coil. Both the thermal peaking factor and the internal voltage
can be reduced by orders of magnitude from the external

dump option in a well-designed internal dump circuit.
The disadvantage of a cold resistor is that all of the magnet energy is absorbed at cryogenic temperature, greatly increasing the time and cost for recool. Therefore, strings of accelerator magnets, which involve a very large number of
magnets and training (see SUPERCONDUCTORS, STABILITY IN
FORCED FLOW) quenches, use cold switches, but dump externally in order to achieve a large number of rapid cooldowns.
CICC magnets and absolutely stable pool-boiling magnets are
not supposed to quench. If they do, the engineering postmortem would generally take more time than cooldown, and recool is not a major consideration. The disadvantage of a cold
switch is that the power-handling capability of an individual
switch is much smaller than that of a warm switch. Above
approximately 200 V 15 kA, cold-switching becomes impractical.
Cold switches can be triggered actively or passively. Cold
diodes begin to conduct when a forward voltage higher than
a threshold is applied. Cold transistors begin to conduct when
a forward voltage higher than the turn-on voltage is applied
to their base. The transistor gain allows them to be over an
order of magnitude more sensitive than the diodes. Superconducting switches can be driven normal by external heaters or
induction coils, coupled closely with the switch. These triggers
may either be driven from a small, external power supply, or
be driven passively by the magnetic induction of the quench
itself.
A fully actively driven resistor doesnt actually require any
switch. Usually, the entire magnet can be driven normal, by
applying heat that is less than 12% of the magnet stored
energy. A small external circuit can drive the current in a
cowound resistor or a heating pad on the outer layer in order
to ensure a large quench zone. A totally nonresistive method
for creating a long quench in a large coil has also been proposed, in which helium would be rapidly drained from the
cryostat (9). Advantages and disadvantages of each method
are listed in the table.
Cold Dump Circuit
Metal strips
Superconducting
switches
Superconducting fuses
Internal resistive
heaters
Fast helium drain to induce global quench
Advantages
Low $/kg; uniform heating easy
Ideal for low energy coils
with persistent currents
Can develop higher voltage, faster dump than
superconducting
switches
Guarantee uniform heating of coil, without imposing high quench
propagation velocities. High degree of redundancy can be designed in. No bulky
dump structures outside coil
No bulky dump structures outside coil
Disadvantages
Extremely massive,
bulky
Much too expensive for
large coils; both massive and high $/kg
Much too expensive for
large coils
Extraction of leads from

CICC must be leaktight, resistive drop
must be low enough to
prevent arcing within
sensor
Slower, less uniform

heating of coil than resistive heater. Not applicable to CICC
It is our opinion that cowound, resistive heaters have the fewest theoretical limits in almost all cases. They will always

QQ
;;
Current
leads
LHe
reservoir
NC
Bypass
switch
Storage coil
Figure 7. A representative dump circuit using cold diodes.
achieve the least peaking in energy deposition and internal

voltage. They can also be used in all applications as advanced
quench detection sensors, as discussed below; and they can
sometimes be used as structural backing elements.
Cold Switches. A typical cold switch is shown in Fig. 7.
Here, by splitting a coil into two halves it is possible to apply
a resistive dump voltage to both sides without opening the
normally closed switch across the coil as a whole.
In this circuit, the magnets are charged through the power
supply, which is then shunted through the normally closed
switch, allowing current to circulate through the magnets. If
a quench begins in either half of the center-tapped coil, the
voltage across the switch is zero, but the inductive voltage on
the unquenched half is equal and opposite to the inductive
and resistive voltage across the quenched half. Above the cold
diode threshold of 10 V to 15 V, two of the four diodes begin
conducting. The resistor is sized so that its voltage drop is
much larger than the 1 V to 1.5 V forward drop across the
diode, but small enough to satisfy Underwriter Laboratory
safety limits, typically 100 V. The resistor itself might be a
heater pad on the outer layer of the winding, inducing a
longer quench zone and more uniform heat distribution in
the coil.
In the circuit shown in Fig. 8, cold transistors are used,
instead of cold diodes, in order to block high voltages during
both charging and discharging of the magnet system. Kaerner
(10) found that the only active devices blocking bipolar voltages at helium temperatures are NPT (non-punchthrough)
IGBTs (Insulated Gate Bipolar Transistors). In 1995, a single
device would carry 300 A at 5.9 K and block 1200 V. In this
system, after quench is detected, the weak coil is shunted
by the IGBT, the coil is rapidly dumped by internal heaters,
then the rest of the coils are ramped down. With this scheme,
only 1/n coils of the stored energy has to be dumped at helium
temperatures, but there is only one set of vapor-cooled leads.
Probably the most commonly used cold switch is a normally closed superconducting switch. During coil charging, it
is held normal by an actively driven heater. The heater is
then turned off and frequently the charging leads are then
disconnected to reduce losses. These can be used in conjunction with a cold transformer in order to lower the voltage
threshold for passive dumping. Anashkin et al. designed a
passive circuit that is capable of responding to very low rates
of current decay (11). The circuit in Fig. 9 demonstrated a
superconducting switch normal transition for a magnet field
decay of 2 104 T/s.
Gate connector
bushings
Figure 8. Arrangement of cold bypass switches in toroidal SMES
system [Kaerner et al. (10)].
2
4
Rs
3
5
6
1
Figure 9. Highly sensitive passive protection circuit: (1) Superconducting magnet, (2) Shunt resistor, (3) Superconducting switch, (4)
Main switch heater, (5) Auxiliary switch heater, (6) Secondary winding, (7) Detachable current leads, (8) Current supply [Anashkin et
al. (11)].
The current decay due to a normal zone induces current in

the secondary winding and auxiliary heater which drives the
superconducting switch normal. Most of the magnet current
is now forced through the shunt resistor. The shunt resistor
can be placed either inside or outside the cryostat. An external resistor would be favored for applications with a large
number of magnets or expected training quenches, such as
accelerator rings, in which cooldown time and refrigeration
requirement dominate. Internal placement is favored for very
high performance magnets, where the resistor can also be
used as a magnet heater to force uniform quench.
QUENCH DETECTION
Superconducting magnets have traditionally used relatively
simple methods for detecting a quench. Voltage taps on the
surface of a winding are the most common, and changes in
temperature, pressure, and flow have also been used, sometimes as supplements to voltage taps. These methods become
inadequate when a magnet has to operate in a strongly
pulsed field. The problem of induced noise voltage is exacerbated by large size and large transients in flow or temperature. In the future of large commercial systems, requiring
high reliability, quench detection is likely to be the weak link
in the magnet protection system. While series redundancy
can provide arbitrarily high reliability in external protection
circuits, it will require advanced concepts to guarantee high
signal-noise ratio, rapid detection, along with leak-free and
discharge-free quench detection under all operating conditions.
Voltage Sensors
Conventional Voltage-Taps and Bridges. The voltage across a
coil or a section of a coil is measured by tapping into the external surface of the conductor through the insulation. This is
commonly done at sections of the winding pack that are physically close to each other, in order to avoid unnecessary inductive pickup, but which may still have significant voltages between them, such as the voltage across two layers or two
pancakes. This method, being the crudest and most likely to
be overwhelmed by inductive noise, is used with very slowcharging dc magnets. The signal-noise ratio of voltage taps is
most frequently enhanced by the use of a bridge circuit to
cancel out the inductive signal. Two nearly equal inductive
signals, such as the inductance of two adjacent double pancakes or an external inductance with the same or proportional
dI/dt as the coil are put in two arms of the bridge. An external
resistance is then balanced against the coil section whose resistance is being measured, and the ramped superconductor
signal is zeroed out. Any voltage should then be equal to the
resistive drop across the magnet during a quench. A typical
bridge circuit design by Purcell (12) is shown in Fig. 10.
Pick-Up Coils. If a coil is connected to a low transient impedance external circuit, it is possible to detect a quench by
inductive pickups to an overall magnet change in current. An
external dump circuit can then be triggered to accelerate the
coil dump. A typical use of a pickup coil quench detector by
Sutter (13) is shown in Fig. 11.
A variation on this technique is to place higher-order
pickup coils around a superconductor as a multipole antenna
50K
V2
L2
10K
Helipot
V1
L1
50K
Figure 10. Quench detection bridge circuit [Purcell et al. (12)].
(14). During a quench, circulating supercurrents will be suppressed more rapidly than overall current, causing a rapid
collapse of diamagnetism in the conductor that can be detected by the antenna. Both techniques have the advantage
of not interrupting the coil insulation.
Detection Circuits. The basic detection circuit by St. Lorant
(15), shown in Fig. 12, would trigger a quench dump, based
on a resistive voltage level that exceeds a preset threshold,
typically 10 mV to 200 mV. In a bridge circuit, resistive voltages of either polarity must be expected. Low-pass filters prevent false positive signals due to ambient noise. The comparator is the quench detector itself. The rectifier allows either
positive or negative voltages to be used as unipolar digital
triggers. A signal inhibit may be used to prevent quench triggers during coil ramping, and the Schmidt trigger creates a
trigger pulse of fixed amplitude and duration.
Advanced Detection Techniques
In a large magnet, the terminal voltage during pulsing may
be as high as 5 kV to 25 kV. The POLO coil set a worlds
record for CICC at 23 kV (16), while recent designs of large
CICC coil systems, such as ITER (10 kV), TPX (7.5 kV), and
NAVY SMES (10 kV) have specified voltages in this range,
while the important pool-boiling Anchorage SMES system is
being designed to withstand 4.2 kV. By contrast, a large number of quench simulations have shown that in order to hold
hot-spot temperatures to 150 K, a quench must be detected
at a threshold voltage of 0.2 V to 1.0 V (17,18). If, as in the
TPX design (19), a desired value of signal/noise of 10 : 1 is
specified, the quench detection sensors must be capable of reducing noise levels to 20 mV to 100 mV. This implies that
the voltage rejection capability of the quench detection system
should be on the order of 100,000500,000 : 1. Although this
may seem optimistic, recent experiments at MIT, the Lawrence Livermore Laboratory, and the Ecole Polytechnique
Federale de Lausanne have demonstrated the feasibility of
such high levels of voltage noise rejection (17,20).
New techniques that promise the greatest cancellation of
inductive noise and the highest signal/noise ratios include the
use of internal sensors, digital differencing and signal processing, and fiberoptic temperature sensors.
Advanced Voltage Sensors. Several noise rejection techniques can be used simultaneously in order to obtain ultra-
D
Superconducting
magnet
L = 1.48mH, R = 5.92m
+
70m
Transrex 500-5
power supply
100V, 5000A
102mF
102mF
Water
cooled
dump
Pick-up
coil
25
50W
D
Magnet
shunt
Power supply
shunt
Ramp
generator
circuit
Dewar
Quench
detection
circuit
Figure 11. Use of pickup coil quench detector in the Fermi National Accelerator Laboratorys
energy doubler magnet [D. F. Sutter et al. (13)].
high system noise rejection. Individual concepts that can be

used include:
1. Cowinding an insulated voltage sensor with the CICC
cable
2. Extracting voltage sensors at joints, but terminating
them within the winding; thus subdividing the terminal
voltages and localizing quench information, without exacerbating electrical integrity or leak tightness
3. Placing the cowound sensor in the part of the cable best
calculated to reject transverse, longitudinal, and selffield voltages. Placement of voltage taps in the center of
the final stage subcable was favored by TPX, as shown
in Fig. 13 (17)
4. Terminating the voltage sensor internally, further localizing and subdividing the winding into sections, and
Vcoil
or
Filter
Variable
attenuator
Comparator
Rectifier
permitting another level of voltage differencing, such as

simple differencing and central difference averaging.
The two halves of the sensor form one solid wire that
can be cabled and wound with the rest of the conductor.
A method for forming an internal termination in a continuous sensor is shown in Fig. 14
5. Further signal processing, such as using integrated
volt-seconds, rather than simple voltage thresholds as
detection criteria
;;;
;;;
;;;
These techniques are now described in more detail.

Cowinding an Insulated Voltage Sensor with the CICC Cable. A
cowound sensor has been used previously on the US-DualPurpose Coil (US-DPC) coil, where an insulated wire was
wound along an edge of the conduit on the outside (21). A
cowound, insulated aluminum strip in the B&W SMES coil is
Center of
final stage
Fiberoptics
Triplet
Schmidt
trigger
Signal
inhibit
Signal out
+
Reset
Valley
Figure 12. Basic quench detection circuit block diagram [S. J. St.
Lorant et al. (15)].
Central tap
Figure 13. Placement of sensors within cable.
Copper
braid
Braze joint
S-glass
braid
Copper wire
;;;;

QQQQ
;;;;;
;;;;;

QQQQ
Stainless steel
capillary tube
Figure 14. Two-sided voltage sensor, showing internal termination

between two halves.
also being used as structural reinforcement. If the sensor is

cabled on the inside of the conduit, as shown in Fig. 13, the
degree of noise cancellation will improve by at least another
order of magnitude. The ITER QUELL (Quench on Long
Lengths) experiment demonstrated the cabling and extraction
of voltage sensors on the surface of a cable, but inside the
conduit, for a 100 m length of conductor. In the QUELL experiment, the rejection of transverse voltage was more than
400 times better than that of conventional voltage taps (21).
Two TPX experiments, one with a copper cable, the other with
a NbTi cable, demonstrated cabling of sensors in different positions of a full-scale cable (17). Depending on the sensor position, noise rejection ranged from 60060,000 : 1.
Extracting Voltage Sensors at Joints. Joint extraction from the
ends of a continuous winding, while difficult, is much less intrusive than extraction every two layers or pancakes through
the winding pack. The joints must already be accessible for
servicing and capable of accommodating helium stubs and
lines and instrumentation feedthroughs. The biggest problem
with extracting voltage sensors at the joints is ensuring that
they will not leak helium and that they wont be subject to
electrical breakdown. The approach to preventing leaks is
two-fold: (1) injecting a moderately sized seal area with a
mineral-filled sealant, such as Stycast, with a good thermal
match to the seal metal, and (2) designing the seal to be long,
in compression, and mechanically redundant in terms of interrupting individual microcracks, and (3) made up of a separate, weldable piece that can be thermal shock tested or cycled before installation. A simple version of this joint was
demonstrated at MIT that, after two cooldowns to nitrogen,
was capable of holding 70 bars on one side and vacuum on
the other (17).
Design against electrical breakdown is ensured by designing the CICC itself, so that there can never be 160 V across
the voltage sensor. In this case, the sensor insulation will be
below the Paschen minimum for helium, irrespective of uncertainties in helium pressure due to leaks or in stray transverse
or longitudinal magnetic fields. Almost any large magnet will
satisfy this design approach automatically. For example, a
normal zone as long as 1 km with a stabilizer current density
of 200 A/m2 at an average field as high as 13 T, would have a
resistive voltage of only 133 V between joints. Depending on
the time needed for internal current dump, the current density can be lowered slightly, if needed, in order to compensate
for resistivity rising with temperature. This should seldom
be necessary.
Placing the Cowound Sensor in the Part of the Cable Best Calculated to Reject Transverse, Longitudinal, and Self-Field Voltages. The surface of the cable is not the best position for emulating the trajectory and therefore the flux linkage of a typical
strand in a cable. The best job would be done by a sensor that

was cabled into a first triplet, as though it were a strand. The
second best position, identified so far, is to place the sensor in
the center of the final stage of the cable, which is easier to
cable and not vulnerable to conduit broaching or welding.
Both positions are far superior to the center of the cable or a
natural cabling valley on its surface. Equations for approximating the induced noise voltage for different types of field
and sensor placement were derived by Martovetsky (22) and
are summarized in Table 1.
Terminating the Voltage Sensor Internally. Internal termination of sensors localizes the signal by subdividing the winding
into sections and permits another level of voltage differencing, such as simple differencing and central difference averaging. In TPX, the voltage sensors were terminated internally
at 1/6, 1/3, 1/2, 2/3, 5/6, and completion of the distance
through the winding pack, as shown in Fig. 15. Even before
using differencing techniques, this should further reduce the
noise voltage by another factor of six, in a way similar to putting voltage taps on each one of six double pancakes. Taking
a simple difference between, say, the signal at 1/6 and 1/3 is
frequently ineffective, since there may be a systematic gradient through the winding pack due to eddy currents being
turned around by a break on one side of the winding pack,
but not the other (23). However, central difference averaging,
in which one-half the first and third sensor signal are subtracted from the middle sensor, as first proposed by Yeh and
Shen (24), can be effective in canceling out gradients. For example, if voltages are measured across double pancakes, the
threshold voltage across the first six pancakes would then be
0.5V12 V34 0.5V56. In simulations of TPX, the degree of
further cancellation using CDA was 520 (17). Internal termination has the additional benefit of localizing quench initiation and of following the sequence of quench propagation
through a winding pack.
Using a sensor insulation that is compatible with inexpensive seam-welding of the steel capillary tube permits the redundant use of multiple voltage sensors in a cable. Fabrication of sensors with S-glass braid, as shown in Fig. 16, was
demonstrated in the QUELL experiment and the ITER CS
Model Coil. Some combination of Formvar, teflon, and/or kapton would be used for NbTi. Teflon would probably be most
desirable, because of its ability to eliminate seams at modest
temperature and pressure, thus eliminating tracking. XMPI
Kapton, at higher temperature and pressure, should have
even higher performance.
Further Signal Processing, Such as Using Integrated Volt-Seconds, Rather Than Simple Voltage Thresholds as Detection Criteria. This is particularly effective in screening out short disturbances, such as stick-slip, flux jumps, and plasma
disruptions. In TPX simulations, an additional factor of
1020 in signal/noise was achieved by using a volt-second
window, instead of a voltage threshold (17). Further improvements in signal/noise ratio through signal processing have
been proposed through the use of carrier signals and synchronous detectors.
In summary, although there is clearly a broad range of design specific signals and signal/noise ratios, the following
rules of thumb might be used for preconceptual design of a
voltage noise reduction system with the goals of achieving a
signal/noise level improvement of 105 (i.e., 10 kV down to
100 mV): Split the winding pack into sections (510) Place
10
Table 1. Noise Voltages for Different Sensor Positions

Transverse Field
Self Field
Grad-B
Ir 2
L
V
4R 2
Wire in center
1
V
2
z
B
r 2last stage dl
0 r
L
Parallel Field
B r l
2
subcable, n
n1
dl
p,n
Geometry
VB
Wire in valley
Nturns d 2cable
2
Grad-B
1
V
2
IL IL
4
12
z
B
r 2last stage dl
0 r
L
B r l
2
subcable,n
n1
p,n
r 2cable
dl
lp,cable
Geometry
V f (Inhom)
Wire in triplet
1
2
B r
L
z cable
dl
Grad-B
1
V
2
V f (Inhom)
z
B
r 2last stage dl
0 r
L
IL
12
V f (Inhom)
B r l
2
subcable,n
n1
dl
p,n
Geometry
V f (Inhom)
Wire in center
of final stage
1
2
B r
L
z last stage
dl
Grad-B
1
V
2
z
B
r 2last stage dl
0 r
L
IL IL
12 16
B r l
3
2
subcable,n
n1
dl
p,n
Geometry
V f (Inhom)
1
2
B r
L
z last stage
dl
internal sensor in cable final stage (1001000) central difference average (510) filter, integrate, signal process (5
10) 104 106. Two TPX noise injection experiments, one with
copper, the other with NbTi cable, demonstrated transverse
field noise rejections of up to 60,000 (17). The ITER QUELL
coil had both conventional and cowound voltage taps on the
outside of the cable in its natural valleys. In this experiment,
the voltage rejection of the cowound sensor was 80 times better than that of the noninductive winding with a voltage rejection of 6 : 1 or 500 : 1 total. Another noise-rejection method
used in ITER of differencing multiple in-hand windings was
calculated to achieve noise rejection ratios of 300400. A better way to normalize the results to design for all noise sources
is to show that the V/(km-T/s) are 1 for transverse field and
10 for parallel and self-field, corresponding to 1 V for 1
T/s over a kilometer. Placement of the sensor in the center of
the final stage or in a triplet have, so far, met this goal in

all experiments.
Advanced Fiberoptic Temperature Sensors. Conventional
temperature sensors have a number of disadvantages: they
cannot be inserted inside a cable-in-conduit, dont measure
cable temperature directly, and have a significant time lag.
Like conventional voltage taps, they cant give coverage to a
winding without frequent penetration of the insulation system. Furthermore, thermocouples are insensitive at helium
temperatures, while carbon glass resistors and resistance
temperature detector (RTD) sensors are insensitive at quench
temperatures. The use of fiber optic temperature sensors, protected by a steel capillary, as shown in Fig. 17, has several
advantages over conventional sensors: (1) They can be inserted directly into the helium flow channel with a very short
;;;;;
;;
;;;;;
Stainless steel
capillary tube
Superconducting cable
Voltage
taps
Figure 15. Inexpensive voltage sensor concept.
Cu braid
Sglass
braid
Cu wire
Figure 16. Six internal voltage sensors, terminated at equal distances through cable.
;;;;;
;;;;;;;
;;;;;
Sglass
braid
Stainless steel
capillary tube
Figure 17. Fiber-optic sensor.
Stainless steel
capillary tube
Fiber optic
2
[nL + nL]
(2)
where is the phase shift and n is the index of refraction.

The major problem with this technique is the rejection of
path length changes due to mechanical strain in the glass.
One strain-rejection technique is to decouple the glass from
the conduit containing it, so that they dont share strain. At
M.I.T., a fiber was inserted in a 1.0 mm stainless steel capillary tube and several turns were wound on an 80 mm steel
tube. The assembly was heated to 700C, then cooled to 4 K
without damage, thus demonstrating strain decoupling and
the absence of a capstan multiplying effect on fiber tension
(26). The QUELL experiment demonstrated that the relatively loose fit of a copper-clad fiber can reduce the strainsharing by a factor of 10100. However, mechanical strain
was still a dominant effect, since the changes in index of refraction are 1%. It has been demonstrated by Smith that the
mechanical strain and temperature-dependent signals can be
almost totally decoupled by using two independent signals
with different strain and temperature dependences (27). This
can be accomplished by the use of polarization maintaining
(PM) optical fiber, two color operation, or two mode operation.
Each polarity, color, or mode has a different, calibrated ratio
of strain and temperature dependence, so that the gain on
one phase shift can be adjusted to tune out the strain signal.
This method has the additional benefit that it can also be
used to measure the integrated conductor strain. Overall system cost-performance analysis indicates that two mode operation will probably be the most cost-effective long-term solution. Typical sensitivities are shown in Table 2.
Table 2. Sensitivity of Dual-Mode and Dual-Polarity Fiber to

Temperature and Strain
Method
2 Polarities
2 Modes
2 Colors
Temperature Sensitivity
(radians/m-K)
Strain Sensitivity
(radians/m-)
1.2
2.18
0.1
5 103
55 103
0.5 103
11
Sglass
braid
Gold sputtered
mirror
thermal time constant, (2) they are insensitive to pulsed magnetic fields (25), helium flow, and pressure, (3) they are very
small (50 m), so as many fibers as desired can be placed
in a single steel can, (4) the length of a fiber between joints is
practically unlimited, 50 km being a routine commercial
length, and (5) they have great scientific potential in the use
of signal processing to provide a complete profile of temperature and field versus length and time.
Fiberoptic temperature sensing works on the principle of
measuring optical path length, taking advantage of the temperature dependence of the glass index of refraction. The
change in phase from either effect is
=
;;;;

QQ

QQ
Fiber optic
Figure 18. Internal termination of fiber optic temperature sensor.
Glass-fiber must be clad in order to prevent the entrance

of water during handling. Commercial acrylic fibers are adequate for NbTi, but will not survive Nb3Sn, NbAl, or BSSCO
heat treatments at 600C. Although the surrounding steel
can contain any spattering of the cladding, this solution is, at
best, messy. Coatings of copper, gold, or graphite can protect
the glass fiber and survive heat treatments. Unjacketed
acrylic coatings should be used outside of the joint region, so
that high-voltage cryostat feedthroughs wont be necessary.
The same design principles of internal termination applied
to voltage sensors within a winding pack apply to fiber optic
temperature sensors. If they are internally terminated by
sputtering mirrors on polished cuts, as shown in Fig. 18, the
integrated noise temperatures are reduced by the number of
subdivisions, then reduced further by differencing techniques.
Subdivision also allows greater localization of quench
events. In the special case of irradiated magnets, it also helps
keep the signal attenuation manageable (60 dB). The
method for subdividing a fiber is to cut the fiber at the desired
length, polish and silver the end. The fiber end is then a mirror and the laser light signal returns to the splitter and detector to form the information-carrying half of a Michelson interferometer.
After converting phase shifts to voltages, the same signal
processing concepts used for voltage sensors can also be used
for fiber optics. Integrating the number of fringe shifts over
a time window, such as one second, can achieve an order of
magnitude improvement in signal-noise ratio. The signalnoise ratio of the fiber optic quench detector is naturally enhanced by the increase in sensitivity with rising temperature.
Simulations of quench and disturbances in TPX showed minimum signal/noise ratios of 600 : 1 for quench detection within
a second (26). This was both within design criteria and superior to internal voltage sensors. The temperature sensitivity
of the fibers has been measured with and without cladding
and can be characterized by an initial quadratically increasing fringe count according to the equation:
a dT = 1.75T 2 17.5T + 43.25; 5K < T < 9K
(3)
The increasing sensitivity saturates at about 15 fringes/mK, and then only increases to about 25 fringes/m-K at room
temperature, as shown in Fig. 19.
If the fiber optic sensor is considered to be a length-temperature rise integral measurement, the sensitivity is 200
times higher at 30 K than at 5 K. Therefore, a global disturbance that raises the helium temperature throughout a 1 km
winding from 5 K to 5.2 K will give a signal that is smaller
than a quench that raises 1 m of conductor to 30 K.
12
16
350
14
300
250
10
200
8
150
Fringes /m
Fringes / Km
12
100
4
2
50
11 13 15 17 19 21 23 25 27 29
Figure 19. Temperature sensitivity of glass fibers.
For an unclad fiber, the sensitivity is almost exactly onetenth that of a clad fiber, the thermal strain of the plastic
cladding acting as an amplifier of the temperature signal. The
curves in Fig. 19 can be used for design of temperature sensors with NbTi; but fringes/m-K should be multiplied by 0.1
for design with Nb3Sn.
Fiber optic sensors would use the same sort of prefabricated and pretested seal as the voltage sensors. However, a
few inches of clearance would be needed between a joint and
the initial position of the seal piece, in order to use a handheld field splicer. The optical fiber can then be coiled into the
pocket of the seal piece.
Quench Detection Conclusions
1. Fiber optic temperature sensors and internal voltage
sensors have been shown by simulation and experiment
to improve signal/noise ratios in quench detection systems by several orders of magnitude.
2. A leak-free method for extracting sensors has been demonstrated, and a redundant and replaceable sealing system has been designed.
3. A method for coil/sensor electrical design has been defined that is robust against arbitrary helium pressures
and magnetic fields. Enormous safety margins are feasible with NbTi and fused teflon or kapton insulation.
4. Advanced quench detection sensors can also be used as
scientific instruments, measuring the internal properties of CICC conductors.
MAGNET PROTECTION CRITERIA
Adiabatic Protection Criterion
A popular and conservative protection criterion is to assume
that there is no heat transfer from the local hot spot where
quench is initiated and that all Joule heating is absorbed by
the stabilizer. In this case, the relation between the peak allowable hot spot temperature and the J2t integral of the conductor stabilizer during a coil dump is a unique property of
the stabilizer material, usually copper. The maximum allowable current density is then determined by the peak allowable
hot spot temperature and the peak allowable terminal voltage
for a coil dump. Typical values of peak temperature allow-
ables are 80 K to 200 K. If a coil is completely supported in

compression by external structure, as an accelerator coil in a
large iron yoke, it is possible to design up to 450 K or the
melting point of solder (28). A typical allowable terminal voltage for a pool-boiling magnet is 1 kV to 3 kV, and 3 kV to 20
kV for a CICC magnet. However, as discussed in the section
on electrical protection, the fundamental limits on voltage for
both topologies are strong functions of specific design.
Irrespective of the coil temperature allowables, the selection of the peak terminal voltage fixes the minimum L/R
dump time constant (s) at
D =
2Wm
VmaxIcond
(4)
where Wm is the stored energy in the magnet (J), Vmax is the

peak allowable terminal voltage on dump (V), and Icond is the
conductor current (A). For a system dumping its energy into
a linear external resistor with negligible resistive voltage
within the coil system, the magnet current after switching is
simply

R t
(5)
I(t) = Io p exp D
Lm
where Rd /Lm is the dump time constant (s).
Since a quench will not be detected immediately, there is
also a delay time before any action is taken. This has sometimes been specified as 12 s as a design goal, but it is a
function of the signal/noise ratio of the quench detection system, as explained in that section. The allowable current density in the copper then is fixed by the combination of delay
plus dump time as

2
tdelay + D = Z(T f )
j0cu
(6)
2
where tdelay is the maximum possible time delay (s), Tf is the
final temperature at the end of dump (K), and G(Tf) or, alternatively in some texts Z(Tf), is a property, unique to each material or combination of materials, defined as

Z(T f ) =
Tf
Tb
C(T )
dT
(T )
(7)
This integral can also be thought of as the J2t integral of the

current in the conductor during a dump, and is thus usually
represented with the units (A2 /m4-s). An analytical approximation for G(Tf) for copper from a bath temperature of 4.2 K
is
Z(T f ) =
1363
1016
4173
Tf +
T f + 47.89
(8)
Another way of stating the design constraint, derived by

Iwasa and Sinclair (29) defines Z(Tf) as the integral from the
detection temperature to the final temperature and models
the presence of materials other than substrate by a correction
factor a, which is the enthalpy change ratio of substrate/all
other materials, which can be determined by table lookup or
20
8 108
Copper
Iron
Nickel
Niobium
Titanium
Tin
Aluminum
Teflon
Helium: 130g/m^3
Helium: 150g/m^3
6 108
5 108
1
2
15
Z (1016 A2 s/m4)
7 108
Specific enthalpy (J/m3)
13
4 108
3 108
3
4
10
5
5
2 108
1 108
0
50
100
150
200
Temperature (K)
250
300
Figure 20. Specific enthalpies of magnet materials (J/m3) versus

temperature (K).
approximated as a volumetric ratio. The maximum allowable

current density in the substrate is then

1 + a VD I0 Z(TD , TF )
j0 =
(9)
a
EM
where VD is the dump voltage (V), I0 is the operating current
(A), and Em is the stored energy (J).
Iwasas a-factor can be quickly estimated with a curve of
volumetric specific enthalpies and the volumetric ratios of the
different constituent materials in a design. The specific enthalpies of copper, aluminum, iron, nickel, niobium, titanium,
tin, teflon, helium at 130 kg/m3 and helium at 150 kg/m3 are
plotted from 4.5 K to 300 K in Fig. 20.
Copper and nickel are the two best materials above a maximum temperature of 100 K to 120 K. However, there is no
more than a factor of two difference between the best and the
worst metals. Copper is about 60% better than aluminum; but
since aluminum is three times lighter than copper, it is twice
as good as copper, if energy/mass is a more important consideration than energy/volume. Copper is clearly also the best
material from an adiabatic J2t criterion. Therefore, if the design is limited by the hot spot temperature in a given volume,
copper is the material of choice.
The curves of Z(Tf) versus T (K) in Fig. 21 appear in Iwasas casebook (30).
While high-purity silver has the best Z(TF), far-less expensive oxygen-free coppers are nearly as good. Copper is two
three times as good as high purity aluminum. The Z(TF) versus TF(K) curves in Fig. 22 were calculated specifically for a
tradeoff between copper and stronger, less high purity aluminums. The curves show that copper is 20 to 25 times better
than aluminum alloys. The implication is that the cross-sectional area of copper needed for protection in a design where
the available enthalpy is dominated by the stabilizer would
be four to five times less than that of aluminum. In a design
100
200
300
T(K)
Figure 21. Z(Tf) functions: (1) Silver (99.99%); (2) Copper (RRR 200);
(3) Copper (RRR 100); (4) Copper (RRR 50); (5) Aluminum (99.99%)
(30).
in which other components, such as superconductor, helium,

or other structures were important, the superiority of copper
would be reduced. Magnetoresistivity would also reduce the
quantitative superiority of copper. In magnetic energy storage
designs, the main factor that improves the relative position of
aluminum is that an aluminum such as 2219 can also be used
as a structural material. In a design that requires a structural cross-section that is several times larger than the cross-
Copper(RRR-200)
Copper(RRR-100)
Copper(RRR-50)
Al Alloy(RRR-5.5)
A12219
1.6E + 17
4E + 16
1.4E + 17
1.2E + 17
3E + 16
1.0E + 17
8E + 16
2E + 16
6E + 16
4E + 16
Aluminum Z(A2s/mm4)
Copper Z(A2s/mm4)
0 100
1E + 16
2E + 16
0
50
100
150
T (K)
200
250
0
300
Figure 22. Z(Tf) functions: (1) Copper (RRR 200), (2) Copper
(RRR 100), (3) Copper (RRR 50), (4) Aluminum alloy (RRR
5.5), (5) Aluminum alloy 2219.
14

Table 3. Z(Tf ) Functions for Copper and Aluminum
Temperature
4
10
20
50
75
100
145
190
240
273
300
Copper,
RRR 200
Copper,
RRR 100
0
3.173
4.776
3.847
6.757
8.633
1.097
1.285
1.410
1.505
1.561
0
1.597
2.481
2.798
5.457
7.274
9.567
1.142
1.268
1.360
1.416
1014
1015
1016
1016
1016
1017
1017
1017
1017
1017
1014
1015
1016
1016
1016
1016
1017
1017
1017
1017
section needed for protection, the volumetric advantage of

copper would disappear.
The values of Z(Tf) from Fig. 22 are listed in Table 3.
QUENCH PROPAGATION
Adiabatic (Potted) Magnets
Adiabatically cooled magnets are selected for applications
that are relatively small, and quasi-steady state, since they
are incapable of absorbing large amounts of local energy.
They are particularly suited to applications where there cant
be any cryogenic fluid within the magnet and where small,
compact winding packs are required. Iwasa has argued convincingly (31) that most high-temperature superconductor
(HTS) magnets are also likely to be adiabatically cooled, because the local energy absorption is improved by orders of
magnitude at higher temperatures.
Adiabatically cooled magnets may be protected either by
internal or external energy dumps. If the simplifying assumption is made that the thermal conductivity and heat capacity
of all materials is temperature independent, the longitudinal
quench propagation velocity is expressed by the balance of
constant local heating density and thermal diffusion through
the winding pack as (30)

vpropagation = J

CnCs
n kn

Tcs + Tc
Top
2
(10)
where vpropagation is the longitudinal quench propagation velocity in the winding direction (m/s), J is the current density in
the composite wire (A/m2), n is the electrical resistivity of the
composite wire (W-m), kn is the thermal conductivity of the
normal wire (W/m-K), Cn and Cs are the heat capacities of
the wire in its normal and superconducting states respectively (J/kg-K), and Top, Tc, and Tcs are the operating and superconductor transition temperatures, respectively (K) (32).
Again, because the heat capacity of all materials rises much
more quickly with temperature than resistivity does, the
quench propagation velocity of HTS quenches should be much
lower than that of LTS quenches and the coils will be harder
to protect.
Copper,
RRR 50
0
7.953
1.256
2.005
4.283
5.990
8.206
9.793
1.116
1.192
1.247
1013
1015
1016
1016
1016
1016
1016
1017
1017
1017
Al Alloy
(RRR 5.5)
Al2219-T85
0
2.280
2.584
9.466
2.898
5.538
1.050
1.504
1.941
2.195
2.386
5.688
6.422
2.517
8.521
1.827
4.055
6.451
9.016
1.062
1.186
1012
1013
1014
1015
1015
1016
1016
1016
1016
1016
1011
1012
1014
1014
1015
1015
1015
1015
1016
1016
Pool-Boiling Magnets
Pressure Rise. The quench pressure rise during the quench
of a pool-boiling magnet involves a design tradeoff. The maximum pressure can most easily be controlled by the setting
of external pressure relief valves or rupture disks (33). The
additional pressure in the magnet due to pressure drops in
the vent lines and the disks or valves can be solved by the
time-dependent model of Krause and Christensen (34), assuming frictional, adiabatic (Fanno) flow in the vent lines. An
alternative, recommended by Powell (35) is repressurization
with warm helium gas. The motive for maintaining a high
pressure, discussed in the section on electrical integrity, is
that it will improve the dielectric strength of warm helium.
The disadvantage is that it will increase the required thickness of the liquid helium cryostat, the pulsed eddy currents
in the thicker cryostat, and the conduction losses through the
cold mass supports.
Quench Propagation. Under normal conditions of cooling by
a pool of liquid helium, the propagation velocity should obey
the proportionality (36):
v p = a( j0 jr )
(11)
where jr is the current density at which the conductor would

recover (A/m2), and j0 is the operating current density
(A/m2). If the quench condition causes local dry out, then the
propagation velocity is simply (37)
v p = a j0
(12)
CICC Magnets
Quench propagation in CICC conductors is usually treated as
a one-dimensional problem, quench propagating from an interior normal zone toward the inlet and outlet of a hydraulic
channel. Bottura has written a general three-dimensional numerical solution for quench propagation (38), which, to the
best of our knowledge, is also the only commercially available
general quench propagation solution for any coil topology.
However, since three-dimensional effects have only a secondorder effect on the key design parameters of temperature,
pressure, and expulsion velocity, one-dimensional solutions
are still used.
Whole Coil Normal. In the extreme case of the whole coil

going normal at once, quench propagation isnt an issue. This
may help to place conservative upper bounds on peak pressure and expulsion velocity for design purposes. Note that simultaneous quench of a whole coil is not a worst case for hot
spot temperature. If the coil is designed for internal energy
absorption, it is a best case, because the peak/average energy
absorption in the coil would be 1.0. If it is designed for external energy absorption, it may be a worst case for the refrigerator by absorbing energy at cryogenic temperature, instead of
externally at room temperature; but in terms of the hot spot
defined by the local J2t, it is not a worst case, because the
nonlinear dump time can only be accelerated by adding a
larger internal coil resistance to the external resistance.
Dresner (39) derived a simple expression for the pressure rise,
if an entire hydraulic channel goes normal at once. This case
would correspond to practical designs in which a resistive
heater was used to ensure uniform internal energy absorption
or to an external, uniform energy source being applied to a
layer with nearly uniform temperature and field:

Pmax = 0.65
Q (Lcoil/2) f
Dh
2
behavior is more complicated than that described by Dresner

and the Shajii/Freidberg theory described next should be
used.
Dresners time-dependent equation for the pressure rise is

3/2

D
p p0 = Cc 4 fZ
4 fct
D
CZ3/2 = 0.95t 2.01

Z=
p=

vexpulsion 0.952
Qc0
C p
2/3
Dh t
f
2/3
Q2 Dt 4
2 c2 f
1/3
(18)
3( 1)
Qt, p p0
3 + 4
(19)
For helium, 5/3, and if C 0.83 for helium, as proposed

by Dresner, these equations reduce to

Z = 0.25 3
(14)
The same assumptions also predict (40) a helium expulsion

velocity of
3( 1)
2C(3 + 4 )
Similarly, Dresner solves for central pressure as function of

time as
Pmax
(17)
Dresners equation implies that the quench zone Z is
(13)
0.36
0.36
Q2 (Lcoil/2)3 f
p0
1
= 0.65
Dh
pmax
(16)
where C is a function of quench zone acceleration O(1), p0 is

the initial pressure (Pa), is the helium density (kg/m3), c is
the velocity of sound in helium (m/s), f is the friction factor,
Z is the length of the hot helium piston (m), D is the hydraulic
diameter (m), and t is the quench time (s). In order to calculate the equivalent piston size, Wachi (41) makes the substitution:
0.36
where Q is the volumetric heating of the helium (W/m3), l is

the half-length of the channel (m), and f is friction factor, and
Dh is the hydraulic diameter (m). A cable-in-conduit has approximately three times the friction factor of a smooth tube
with the same Reynolds number. For a quench pressure
wave, Dresner adopts an approximate value of f 0.013.
When the pressure rise is not much greater than the initial
pressure P0 (Pa), the more exact formulation is
15
Q2 Dt 4
2 c2 f
(20)
and:
p = 0.207Qt, p p0
(21)
1/3
(15)
where Q is the volumetric heating of the helium (W/m3), is

the constant pressure thermal expansion coefficient co is
the isentropic sound speed (m/s), is the helium density
(kg/m3), Cp is the helium specific heat at constant pressure (J/
kg-K), Dh is the hydraulic diameter (m), f is friction factor,
and t is the time since quench initiation (s). This solution is
valid only for the beginning of a quench, since it assumes constant helium properties and neglects inertia and frictional
heating.
Shajii and Freidberg (42) rewrite the time dependences of

Dresners quench propagation equations for temperature,
pressure, and quench propagation velocity as
0.10
TDresner(t) =
R0 LIQZ
4dh
f02 c20
1/3
(0 J 2 )5/3t 7/3
(22)
where o is the resistivity of copper (-m).

pDresner = 0.210 J 2t
(23)
and
Time-Dependent Normal Zone. Quench propagation scaling
for long coils and uniform helium properties were first derived
by Dresner (40). Dresners scaling laws remain valid for describing the early stages of quench in a long hydraulic channel. However, the engineering limits of CICC coils are usually
defined by the hot spot temperature, peak pressure, and peak
helium expulsion flow toward the end of a quench. By this
time, the material properties have changed significantly and
heat absorption is dominated by the cable-and-conduit metal,
rather than the helium. In this regime, the scaling of quench
5
VqDresner X q = 0.42
4
4dh
f02 c20
1/3
(0 J 2 )2/3t 1/3
(24)
Shajii Quench Theory. The Dresner equations were shown

by Shajii and Freidberg to apply only in the operating space
of short times, low conductor temperatures (25 K), and long
initial quench zones (43). Since this operating regime almost
never includes the regimes of greatest interest for design (hot-
16
spot temperature, peak pressure, peak expulsion velocity), it

was necessary to develop solutions for other, more relevant
regimes. Shajii and Freidberg derived analytic expressions for
temperature, pressure, and quench zone propagation velocity
for five other regimes, identified as the (1) short coil, low p,
(2) short coil, high p, (3) long coil, low p, (4) long coil, high
p, and (5) thermal hydraulic quenchback (THQB) regimes.
In a short coil, the coil length is much shorter than the diffusion length of the quench zone, so that quench propagation is
affected by end conditions. In a long coil, the mass of the coil
channel is constant and propagation is unaffected by the
ends. In the low p regime, the pressure rise due to the
quench is p0, while it is p0 in the high p regime. An actual quench may have a trajectory in quench regime space
that traverses two or more of these regimes. We use the nomenclature long coil and short coil here, because the usage
has become accepted. However, it should be clarified that we
are always discussing the length of a hydraulic channel,
which is typically an order of magnitude shorter than a coil
length. In all four of the constant mass quench zone solutions,
Shajii derived scalings for the temperature and pressure,
valid in the high-temperature (20 K) regimes that are of
interest as coil allowables. Assuming that the heat transfer
coefficient h (W/m2-K) is large:

Aw wCw 2
T (t) =
(T )J 1 et/ w
2hPw
(25)
where T is the average of the cable and conduit temperatures

(K), and w is a characteristic time constant for heat exchange
between the cable and conduit (s):

1
1
1
+
(26)
= hPwall
AcablecableCcable
AwallwallCwall
w
where h is the wall heat transfer coefficient (W/m2-K), Pwall is
the conduit wetted perimeter (m), Acable and Awall are the cable
and wall cross-section areas (m2), cable and wall are the mass
densities of the cable and wall (kg/m3), and Ccable and Cwall are
the specific heats of the cable and wall (J/kg-K), respectively.
Short Coil Solutions. The criterion for a short coil solution
is that the length of the coil is longer than the length of the
quench zone, but less than the thermal diffusion length of the
quench:

L<
24dh c20tm
fVq
(27)
where c0 is the initial sound speed (m/s), dh is the hydraulic

diameter (m), Vq is the quench front velocity (m/s), and tm is
the time needed to reach the maximum allowable temperature tm Tmax /0 J2.
Short Coil, High Pressure. A short coil quench will be in the
high-pressure rise regime (p p0) when
R0 0 J LIQZ
p

>1
p0
2p0Vq
ground helium pressure (Pa), dh is the hydraulic diameter

(m), f is the friction factor, assumed to be a constant in the
range of 0.060.08, and 0 is a diffusion constant of the conduit material (typically 57 106 m4-K/A2s), defined as

Acu c (T )
(29)
0
= min
T Ac cCc (T ) + Aw wCw (T )
where c is the thermal resistivity (W/m-K).
For a short coil in the high-pressure regime, the asymptotic quench velocity, once quench has been well initiated, is

2dR
Vq = 3
L J2
(30)
f Lcoil IQZ 0
where d is the hydraulic diameter (m) and f is the friction
factor. The position of the forward quench front Xq (m), using
the convention that all quenches are symmetrically centered
about x 0, is

Xq =
LIQZ
2
(28)
where R is the universal gas constant (8314.3 J/kg-mole-K,

He 4.003 kg/mole), 0 is the background helium mass density (kg/m3), J is the current density in the stabilizer (A/m2),
LIQZ is the initial length of the quench zone (m), p0 is the back-
+ (Vqt)
3/2
(31)
Short Coil, Low Pressure. A short coil quench will be in the

low-pressure rise regime (p p0) when

f0 Lcoil
p

Vq2 < 1
(32)
p0
4dp0
For a short coil in the low-pressure regime, the quench velocity is
Vq =
R0 0 J 2 LIQZ
2p0
(33)
The short coil limit also provides a simple analytical expression for the density outside the normal zone as a function of
length and time:

2
0 f V q
Lcoil
x
(34)
(x, t) = 0 +
2dh c20
2
where 0 is the initial density (kg/m3). The density decreases
linearly with x, while the velocity outside the normal zone is
a constant versus both space and time.
Long Coil Solutions
Long Coil, High Pressure. The long coil solution is defined
by the two criteria that the coil length is much greater than
the length between diffusion edges (L/2)2 XD2 (tm), and that
the length of the quench zone is much less than the length
between diffusion edges [Xq2 XD2 (tm)]. The criterion for the
long coil solution can then be stated as
L2coil
2/3
3/2
24dh c20tm
2
4Vq2 tm
f Vq
(35)
The long coil quench is in the high-pressure rise regime

(p p0) when
R0 0 J 2 LIQZ
p
>1
p0
2p0Vq (tm )
(36)
The position of the forward quench front in the high-pressure

regime is

Xq = (LIQZ /2)
5/3
+ (Vq t)

5/3 3/5
(37)
where Vq is the asymptotic quench propagation velocity (m/s):
2dh
Vq = 0.613
f
1/5
RLIQZ 0 J 2
2/5
1
t 1/5
c0
v(x, t) =
0.8Vqt
t + 21 (x LIQZ /2)2

cot
K K + 2
32t 2
302
2
(x LIQZ )3/2
t 3/4 + 3/2
2
(40)
where 2(t) (3/4) 1 and 0 is

1/3

0 =
2dh 0 c20
(41)
Equation (11) implies that the expulsion velocity is given by

24dh c20
v(x = Lcoil /2, t)
t
f L2coil
(42)
This is the same expulsion velocity as that predicted by

Dresner. However, with the exception of the expulsion velocity, Dresners solutions for temperature, pressure, and quench
velocity in Eqs. (1618) have functional dependencies that are
significantly different from Shajiis, even in the long-coil case.
Long Coil, Low p Regime. In the small p regime, the pressure rise remains small in comparison with the initial pressure (p p0), corresponding to a weak quench in which the
helium coolant removes most of the heat generated by a
quench. In the small p regime, the length of the quench zone
is
Xq
R0 LIQZ (T + T Tt=0+ )
2p0
LIQZ
2
(43)
where T (Tcable Twall)/2 and T (Tcable Twall)/2. t 0

means the time immediately after the initial quench zone has
been established, assuming sudden energy deposition.
The quench velocity is then
R0 0 J LIQZ

p(t) = 902 c20t 2
L coil /2
Xq
dx
(x LIQZ /2)2 + XD2
(45)
2
where the leading edge of the diffusion front XD (m) is

XD2 =
302 t
0 Vq
(46)
For the combined short coil plus small p case, the solution
for p reduces to Eq. (23). For the combined long coil plus
small p case, the solution for p reduces to

p(t) = 1.36
f
2dh
1/2
0 c0 t 1/2Vq3/2
(47)
(39)
where 12 0 0.8Vq /302. The density profile for the density

in the outer region (kg/m3) is given by
902
(x, t) = 0 +
2K 3/2
The low p itself is given by
(38)
(The multiplier of 0.613 differs from the two published values

of 0.766 by a factor of 5/4, removing an ambiguity in interpreting the Shajii equations self-consistently.) The helium velocity (m/s) in the region outside the quench zone is
17
Universal Scaling Regimes. Shajii recast the preceding criteria into a universal form that predicts the boundaries between the four quench regimes. Two dimensionless variables
x and y are defined and all quench regimes are shown as filling four corners of xy space. First a dimensionless variable
4/3
5/3
and a dimensional variable L0J4/3
0 (A /m ) are defined:

2

c0 0
0 RT max
(48)
= 1.7
p0
p0
where 0, c0, and p0 are the density (kg/m3), sound speed (m/
s), and pressure (Pa) of the background helium. R is the universal gas constant (8314.3 J/kg-mole-K), and Tmax(x,t) is the
maximum temperature of the quench zone (K).

p50 d
2.6
4/3
3
L0 J0 =
(49)
2
R
f 0 c20 05 Tmax
and L0J4/3
0 tend to be relatively constant over a wide range
of time and space for a given design. In order to distinguish
better between the widely different quench regimes, these
variables are reordered by being normalized to the strongly
varying J (A/m2) and Lq (m):
x=
Lq
L
(50)
y=
Lq J 4/3
L0 J04/3
(51)
This now allows the criteria for entry into each of the four
quench regimes to be written in the following simple universal form:
Quench Regime
Long coil, high pressure
Short coil, high pressure
Long coil, low pressure
Short coil, low pressure
Pressure
Condition
y
y
y
y
1
1.2x1/3
0.8
1.2x1/3
Length
Condition
y
y
y
y
x5/6
1.1x5/6
x2/3
x2/3
The four universal scaling regimes are shown in Fig. 23.
Vq
2p0
(44)
Thermal-Hydraulic Quenchback. The misnomer thermal-hydraulic quenchback (THQB), which is certainly thermohy-
18
y = Lq J4/3 / L0 J04/3
y = Lq J 4/3 / L0 J04/3
;;;;;;;
;;;;;;;
;;;;;;;
102
102
I
10 1
II
10 0
III
10 1
10 2
IV
10 1
100
x = Lq/L
10 1
10 1
10 0
III
10 1 2
10
10 2
Figure 23. Boundaries in xy space defining the four quench regimes: (I) long coil-high pressure rise, (II) short coil-high pressure
rise, (III) long coil-low pressure rise, and (IV) short coil-low pressure
rise [Shajii, 1995 (44)].
II
IV
10 1
100
x = Lq/L
10 1
10 2
Figure 24. Scaling diagram showing dimensionless quench regime

boundaries for M 4.
The temperature ahead of the quench front is then

draulic, but has nothing to do with quenching backwards, has
become sufficiently popular as a term that we wont try to
rename it Joule-Thomson quench propagation. It refers to a
condition in which compression heating of helium in front of
a quench zone leads to rapid propagation of quench in which
the thermal/quench wave travels much faster than the mass
flow of helium expulsion in front of the quench zone. A key
dimensionless parameter in predicting whether there can be
a rapid onset of THQB, again introduced by Shajii (44), is
the safety margin between current sharing and background
temperature M:

M=
Ch (T0 )
C (T0 )

0 c20
p0

Tcs T0
T0

(52)
Typically M 2 to 5 for practical coils. Another parameter

is needed to account for the finite ratio of frictional to compressive heating in the THQB regime:
=
C (T0 )0 T0
p0
(53)
is typically of order unity.

THQB cannot exist long in the low pressure rise regime.
In the long-coil, high pressure rise regime, the temperature
just ahead of the quench front can be written as:

1 C (T0 ) RT 0 0 J 2 Lq
T f (t) T0 +
2 Ch (T0 )
c20Vq (t)
(54)
This expression includes the effect of compression heating,

but not of frictional heating.
The condition for THQB to occur before T Tmax is
y>M
5/3
(55)
For the short coil-low pressure rise regime, the condition for
THQB to occur before T Tmax is
y > 1/3 M 1/3
(56)
T f (t) = T0 +
f Vq3
2dCh (T0 )

t
(57)
In the short coil-high pressure rise regime, frictional heating

must again be included, and the condition for THQB to occur
before T Tmax is
y>

1/2
x1/3
1 + 1 + 4(x M)
2
(58)
where and other numerical coefficients have been set to

unity.
The intersection of the THQB regime with the four universal regimes for a typical case of M 4 is shown in Fig. 24.
Unique THQB shaded regions can be calculated for every
value of M.
ELECTRICAL INTEGRITY
A superconductor has no voltage in the quiescent state and
may have a very low voltage, during startup and shutdown,
depending on the application. In the majority of applications,
the superconducting magnet will experience its highest design voltages during a quench dump. It is also during a
quench dump that the helium temperature will be highest,
helium density in electrical isolators may be decreased by
pressure relief valves, and liquid will be boiled into vapor. All
of these effects can simultaneously increase electrical fields,
while decreasing the ability of helium or insulation voids to
withstand the electrical fields. Arcs have developed during operation of real superconducting magnets and are a known
cause of failure and life limitation.
Breakdown in Helium
In a pool-boiling magnet, helium is the primary insulation.
Design against helium breakdown must include the windingpack, joints, supports, vapor-cooled leads, feedthroughs, and
room-temperature isolators. In CICC magnets, the problem is
restricted to specially designed helium isolators, providing
the electrical isolation between the magnets and its grounded
helium headers. However, CICC magnets have the special
problem of protecting feedthroughs and leads against low-
19
Table 4. Paschen Minimum Gaps and Gaps at 20 kV, According to Olivier Equation
Pressure
(atm)
Density
(kg/m3)
Gap at
Paschen
Minimum
(mm)
Gap at
20 kV
(mm)
127, liquid
15.2, vapor
117
0.973
103
0.178
7.9 105
6.6 104
8.5 105
0.011
10
0.056
0.133
1.17
0.144
20.0
20,000
112
Temperature
(K)
1.0
4.229
3.0
3.0
104
1.0
5.0
150
5.0
273
density helium breakdown, in the event of a helium leak. The

primary motivation for dry superconducting magnets is to
avoid this problem altogether, while accepting a low energy
margin against disturbances.
The dielectric strength of liquid helium is comparable to
that of air at standard temperature and pressure. Unfortunately, since breakdown accompanies heating due to a normal
event and rapidly heats local helium, the actual breakdown
strength of liquid helium in a magnet is hard to interpret. It
is conservative and probably correct to always consider gaseous helium to be the insulator in a pool boiling magnet.
The breakdown strength of gaseous helium at ambient
temperature is only a small fraction of that of air because the
electrons can gather kinetic energy from electrical field drift
up to the ionization level in the noble gas helium. Paschens
law should hold for gaseous helium at any temperature. At
all density-gap products that are well above the minimum of
the Paschen curve, the dielectric strength of helium is at least
a factor of 10 worse than that of nitrogen or air.
In the high-pressure regime of the Paschen curve, Olivier
showed that a direct exponential correlation can predict the
voltage breakdown for a broad range of gases, including helium, in a uniform field over a broad range of gap lengths and
pressures (45). At any temperature, the breakdown voltage
between spherical electrodes is
Vbreakdown = K d
Vbreakdown = 645, 375 0.878d 0.901
(60)
The density-gap product that produces the minimum of 160

V in helium is 105 kg/m3-m. The gaps at the Paschen minimum and the gaps predicted for breakdown at 20 kV by the
Olivier equation are shown in Table 4.
A Paschen curve through several sets of experimental data
is shown in Fig. 25.
At cryogenic temperatures the breakdown strength of helium in uniform field allows for a straightforward design in
pool boiling magnets, where it can be guaranteed that breakdown will occur in the high density-distance regime. Figures
26 and 27 (46) show that with minimum breakdown fields on
the order of 10 MV/m, each millimeter of gap should be adequate to protect against 10 kV.
However, breakdown in helium is highly sensitive to electric field nonuniformities and to field polarity. Figure 28 (47)
depicts low temperature dc breakdown characteristics for he-
20 C barmm
Low temp. (g/cm3)mm
Townsend
Fallou et al.
McCallum and Klatzow
Gerhold
Levi
Meats
Brand
Comsa and Cohen
Kusko
Bortnik
Schhuber
105
Volts
(59)
where K is a constant, is the mass density of helium (moles/

liter), d is the gap (mm), and and are the exponents. According to Olivier, for helium 0.878, 0.901, and K
4310. Thus, for helium at STP, where one mole 22.4 L and
the density of helium is 4 g/22.4 L 0.178 kg/m3, the general
equation for voltage breakdown with the gap d in m, the
breakdown voltage V in volts, and the density in kg/m3
would be
10 2
10 3
Pressure-spacing product @ 20 C
10 2
10 3
10 2
10 6
10 1
10 5
10
10 4
10 3
Density-spacing product
10 2 bar-mm 10 3
10 1
10 2
(g/cm3)-mm
Figure 25. Paschen curve for helium.
+
p = 0.1 MN m 2
5.2
45 MVm1
40
35
30
10 k
15 MVm1
20
25
20
1.0
20
0.4
U (KV)
=5
4.0
0.6
4.5
P (MN m2)
0.8
al
Critic rm
e
h
t
o
is
10
+
10
20K
0.2
Phase line
0.1
20
40
5K
60
80 100
(kg m3)
120
20K
140
Figure 26. Depicts the breakdown strength of helium as a function

of pressure and density, under uniform field conditions [Gerhold,
1979 (48)].
0
0
10
d (mm)
Figure 28. Helium nonuniform field breakdown voltages in low temperature helium gas at atmospheric pressure [Gerhold, 1979 (48)].
lium gas under nonuniform (point to plane) field conditions at

atmospheric pressure (48).
The strong polarity dependence is evident in this figure,
suggesting that the design of the electrode configuration to
maintain a field as close to uniform as possible is particularly
important, when feasible. The breakdown level is also a function of electrode surface material and conditions (49), conditioning, and electrode polarity (50).
A design approach proposed to avoid striking an arc across
a helium gap is to use the dielectric strength of helium at the
design temperature and atmospheric pressure for a sharp
point at a 3 mm gap between the point and the plain (51). At
4 K, the breakdown strength of helium is 60 V/mil (2.36 kV/
mm), at 10 K, the breakdown strength is 24 V/mil (944 V/
mm), and at 100 K, the breakdown strength is 2.4 V/mil (94
V/mm).
Electric Field Concentrations
Solid insulations are seldom designed with electrical fields
higher than 1020% of their intrinsic dielectric strength. This
Conditioned sparking
voltage Vs (kV)
50
40
d
d
d
d
d
30
20
10
0
5
6
nd (cm2)
=
=
=
=
=
0.05cm
0.10cm
0.15cm
0.25cm
0.30cm
8 9 1020
Figure 27. Depicts low temperature (10 K) dc breakdown characteristics for helium gas under uniform field conditions [Meek, 1979
(47)].
is not merely an expression of extreme conservatism toward

catastrophic punchthrough and arcing, but also reflects the
reality of electrical field concentrations in practical designs.
While many solid insulations have dielectric strengths 100
kV/mm, it has been shown that the presence of small voids
in a solid will cause the inception of particle discharges in low
temperature helium at electric fields of 10 kV/mm to 20 kV/
mm (52). There is some evidence that the minimum breakdown voltage in a helium void may saturate at low density
and that there is no left-hand side of the Paschen curve in
voids. Measurements by Hiley and Dhariwal (53) appear to
saturate at 4.5 kV/mm in polyethylene at zero density for a
cavity depth of 0.2 mm (900 V). In epoxy resin, they appear to
saturate at 1.2 kV/mm for a cavity depth of 0.2 mm (240 V).
Partial discharges in solid insulation voids are caused by
high electric fields in the voids. These will not destroy the
insulation, if they are infrequent, but can cause erosion of
organic insulation in a pulsed application.
The electric field in a void tends to be higher than that of
the solid insulation itself because the dielectric constant of
solid insulations is always higher than that of free space. Further electric field multipliers are caused by shape factors in
the void, where breakdown can be further enhanced by nonuniform electric field within the void itself. Analytic solutions
are available for the electric field concentrations in planar,
spherical, and cylindrical voids. For other shapes, numerical
techniques are available to calculate the electric field concentrations for both 2-D and 3-D shapes. Engineering rules of
thumb also exist for the most likely void shapes. Zahn (54)
developed the following derivation of the electric field concentrations in spheres and cylinders.
Voids in Solid Insulation. The simplest model assumes that
a void region exists within a uniform dielectric that has an
essentially uniform electric field in the vicinity of the void.
21
Table 5. Dielectric Constant

Material
Kapton polyimide film 120CI-1
Kapton polyimide film 135RCI
G10-CR
G11-CR1
Ea
Ea
Ea
(a)
(b)
(c)
We consider three simple cases shown in Fig. 29 of uniform

electric field incident on (1) a long, thin void, either planar or
cylindrical, with the long axis in the direction of the electric
field; (2) a long thin void, either planar or cylindrical, with
the long axis perpendicular to the electric field; and (3) a
spherical void. Cases (1) and (2) form easily at an interface
between dissimilar materials, while case (3) can arise due to
gassing. If the void region is air at ambient temperature and
pressure, the electrical breakdown strength is Eb 3 kV/mm.
Fields above Eb will result in spark discharges. The maximum
external electric field outside the void can then be calculated
that would keep the voidage electric field below Eb.
Long Thin Void Along Electric Field. Because the tangential
component of an electric field is continuous across an interface, the electric field
(61)
Long Thin Void Perpendicular to Electric Field

Planar Void. For highly insulating dielectrics with dielectric relaxation times ( / ) much greater than the time
scales of a quench, the normal component of the electric displacement field, D E must be continuous across an interface. Thus

Ea
0
(62)
The higher the dielectric permittivity of the insulation surrounding the void, the lower the applied electric field must be
to keep Ev Eb.
Cylindrical Void. In the absence of any volume charge in a
uniform permittivity dielectric, Laplaces equation can be
solved in cylindrical coordinates for the electrical field inside
and outside the cylindrical void:

Ea ix
+ 0

R2 ( )
cos ir
E = Ea 1 + 2 0
r ( 0 + )

R2 ( )
sin i
1 2 0
r ( 0 + )
E =2

Ea < Eb
+ 0
(64)
Spherical Void. The general form of solution to Laplaces

equation in spherical coordinates for a uniform z directed
electric field is

B
(65)
= Ar + 2 cos
r
The total electric field inside and outside the cylindrical void
is
3
Ea iz
2 + 0

R3 ( 0 )
cos ir
E = Ea 1 + 2 3
r (2 + 0 )

R3 ( 0 )
sin i
1 3
r (2 + 0 )
E=
0<r<R
(66)
r>R
3
Ea < Eb
2 + 0
(63)
(67)
Dielectric constants of some widely-used solid insulations are

listed in Table 5.
The electric field multipliers for other commonly found
void shapes are listed in Table 6.
The multipliers for laps and voids in corners are typical
values, based on numerical analysis.
A conservative way to look at the design rules for solid
insulation would be to design so that partial discharges in the
solid insulation were impossible, irrespective of the partial
Table 6. Electric Field Multipliers for 3

Insulation System
Shape
0<r<R
r>R
The internal electric field is purely z directed, while the external electric field has the applied uniform electric field plus a
point dipole field. To avoid breakdown in the spherical void:
This result is valid for planar and cylindrical voids.
E v = E a E v =
3.5
3.8
4.95.0
5.15.2
The internal electric field is purely x directed, while the external electric field has the applied uniform electric field plus a
line-dipole field. To avoid breakdown in the cylindrical void
Figure 29. Three void orientations versus electric field in dielectric

medium.
Ev = Ea
Dielectric Constant (r /0)
Infinite plane or cylinder, parallel to field

Infinite plane, transverse to field
Infinite cylinder, transverse to field
Sphere
Cusp/crescent (debonding at a rounded corner)
Triangle (e.g., epoxy/kapton lapping)
Multiplier
1
3
1.5
9/7 1.286
1.61.7
1.61.8
22
pressure in the insulation voids. As an example, in a superconducting magnet, the worst gas and the most likely gas to
have a partial pressure is helium, whose Paschen minimum
is 160 V. Adopting 2 kV/mm as a typical allowable for glassepoxy and a geometry/dielectric constant mismatch multiplier
of 2, this design rule would then put the burden on the mechanical design to assure that the largest possible void in the
solid insulation would be
dmax,allowable
VPaschen min
=
Emax,allowableRmult
160V
= 0.040 mm
kV
x2
2
mm
(68)
For example, if a conduit with a slip-plane wrap bowed in 10

m, this would be acceptable, since breakdown of helium
would still be impossible.
where Cg is the shunt capacitance to ground and Cs is the

series capacitance. For fast rise times and large values of ,
half of the terminal voltage can appear between the first two
turns of the coil. Measurements on the POLO coil (55) and
analyses at Karfruke (KfK) and M.I.T. showed that large voltage enhancements could be avoided, if the voltage rise time
were long enough. This is a tradeoff, since switch losses or
counterpulse circuit size is proportional to the voltage rise
Stray Coil Capacitance

The turn-turn and layer-layer voltages across a winding pack
are not exactly equal, even in the absence of any resistance.
The turns in a winding pack dont all link exactly the same
amount of flux, particularly in multicoil systems. The peak-
Rb
average ratios in coil systems we have investigated were in

the range of 1.11.3 : 1. A more serious, but avoidable, problem comes from the possibility of overvoltages due to distributed capacity charging of the winding pack. There is a distributed capacitance between every turn of the winding pack and
to ground, as illustrated by the circuit model, shown in Fig.
30. Mutual capacitance to nonadjacent turns is neglected.
The peak local electrical field is higher than what would
be predicted considering only the coil resistance because of
stray coil capacitance in the insulation. The voltage enhancement factor is a function of :

Cg
=
(69)
Cs
Ctg
Ctg
Ctg
Ctg
Ctg
Ctg
Ctg
Ctg
Ctg
Ctg
N
Ctg
Ctg
Pancake
Cpp
Cpp
Ctt
Cpp
Ctt
Cpp
Ctt
Cpp
Ctt
Ctt
3
Ctg
Cpp
Cpp
Ctt
Cpp
Ctt
Cpp
Ctt
Ctt
Ctg
Cpp
Ctt
2
Ctg
Cpp
Cpp
Cpp
Ctt
Ctt
Cpp
Ctt
Ctg
Cpp
Ctt
Ctt
Node
1
Ctg
2
Ctg
Ctg
1
Figure 30. Distributed capacitance/inductance network model of a winding

pack and ground insulation.
3
Ctg
2
n+1
4
Ctg
3
Ctg
Ctg
Tum
time. Optimization of the cost/performance trade can be done

for a specific design, using commercial circuit codes. However,
since the ringing time is primarily a function of conductor size
only, analysis indicates that it is safe over a broad range to
limit the voltage rise time to 100 s, if the conductor is 5
cm square.
Tracking Along an Insulating Surface
Tracking is the leakage current due to the formation of a conducting path across the surface of an insulation. In most
cases, the conduction results from degradation of an organic
insulation. The conducting film is usually moisture from the
atmosphere absorbed by some form of contamination, such as
dust. Tracking does not depend upon Paschen breakdown and
can occur at well below 100 V in air, while a gaseous discharge in air cannot exist below 380 V. Degradation of the
insulation is caused by heat from tracking, which either carbonizes or volatilizes the insulation. This implies that insulations in helium or vacuum at liquid helium temperatures
should be more immune to tracking than magnets in air, because of the absence of chemical reactions. Carbonization results in a permanent extension of the electrodes and usually
takes the form of a dendritic growth; but erosion of the insulation also occurs.
Configurations for which tracking is particularly important are those where a high voltage conductor emerges from
an insulated lead. Generally a conducting electrostatic shield
or ground plane will surround the lead insulation, but will be
terminated at some point, leaving a tracking path along the
surface of the lead insulation from the end of the ground
plane to the conductor itself. The electric field is enhanced at
the end of the electrostatic shield in this configuration, and
can lead to local ionization which effectively extends the
ground electrode, due to space charge, and reduces the useful
length of the tracking path, such that the discharge advances
along the surface toward the lead, with eventual flashover.
Figure 31 illustrates tracking at a high voltage lead termination.
Common practice is to utilize a stress cone (a specially
shaped electrode) or a stress grading material or coating on
the surface of the lead insulation just after the termination
of the conducting electrostatic shield. The effect of the stress
grading is to reduce the maximum field gradient based on the
nonlinear voltage versus current characteristic of the stress
grading material. Silicon carbide is a commonly used material
for stress grading, and is available in tape and paint binders.
Equipotentials
Conductor
Area of field
enhancement
Insulation
Ground plane
Tracking path
Figure 31. Tracking at high voltage lead termination.
23
Heat shrinkable stress grading tubing is also available, and

is normally used in the termination of high voltage cables.
This material could offer a solution to the problem of temporary stress grading during test, when the termination of high
voltage leads is not in its final configuration.
Yet another technique is to cover the entire termination in
an insulating jacket. The effect is to contain the areas of highest stress in a solid dielectric, and to increase the tracking
path length that must be taken by a discharge which would
directly connect the high voltage electrode with the ground
electrode.
To the extent that the high voltage conductors can be fully
insulated and enclosed by an electrostatic shield, the tracking
problem in the near vacuum environment can be avoided.
However, the high voltage conductors such as the main high
current leads and any instrumentation (e.g., voltage tap) conductors must eventually emerge from the insulation.
The helium supply lines which must bridge the gap between the high voltage conductors and ground present another potential tracking path. Considering a configuration
consisting of a metallic pipe emerging from the main conductor which delivers the coolant to an insulating tube, which in
turn connects to a grounded metallic pipe associated with the
manifold. Tracking on the outer surface of the insulating tube
could be a problem, particularly in the cryostat vacuum environment which can be compromised by helium leaks. This
problem can be solved by encapsulating the region of transition from high voltage, through the insulator, to ground, in a
solid dielectric contained within an electrostatic shield. In
this case the path which would otherwise be available for
tracking, and the entire region subject to electric stress, can
be fully contained in solid insulation, as illustrated in Fig. 32.
Experimental Basis for Tracking Allowables. Migliori,
Schermer, and Henke (56) measured dielectric tracking fields
in liquid helium for four gaps in a G-10 circuit board. The
minimum tracking strength that they measured was 13.7 kV/
mm at 0.48 mm. They reported that the cleanliness of the
sample, that is, whether it was abraded with copper or wetwound with metal inclusions, made little difference in the
tracking strength. They also reported that the breakdown
field declined as d0.25 and that the tracking strength was that
of helium vapor, rather than that of liquid helium. This is a
somewhat misleading conclusion, because the tracking
strengths that they measured were similar to those measured
earlier by Haarmon and Williamson (57) in liquid helium, and
orders of magnitude superior to Haarmon and Williamsons
measurements in gaseous helium. The earlier experiment reported nearly identical tracking breakdown fields for paper
phenolic, polyethylene, nylon, and teflon. Their measurements of the tracking strength for series of five 60 kV, 500
ms, half sine-wave pulses with descending gap are shown in
Table 7.
For design purposes, the relevant numbers are those for
gaseous helium at 4 K and 293 K. Fortunately, the temperature sensitivity is rather small, varying only a factor of two
at most over a temperature range of 75 : 1. A factor of four
safety margin would imply that tracking should be limited
to 100 V/mm (2.5 V/mil). Designs have been identified with
tracking fields in the range of 60 V/mm to 200 V/mm.
24
Insulating
tube
HV piping
Insulating
tube
Grounded
piping
Grounded
piping
HV piping
Coolant
Coolant
Insulation
Insulation
Ground plane
Ground plane
(b)
(a)
Figure 32. Avoidance of tracking with full-insulated coolant connections. (a) Tracking at coolant
connection. (b) Fully insulated coolant connection.
Electrical Design Allowables

For design purposes, typical electrical design limits in a cryogenic environment are summarized in Table 8.
Breakdown Due to Helium Leaks
High voltages can be maintained over short distances in a
good vacuum. Typically, in the cryostat surrounding CICC
and potted magnets, or the vacuum tank surrounding poolboiling magnets, a vacuum of 106 to 105 torr can be maintained. In vacuum systems, where residual gas density is low,
the breakdown voltage is to the left of the Paschen minimum.
Therefore, electrical integrity within the cryostat can be compromised by helium leaks into the vacuum space. Helium
leaks can be caused by diffusion through welds in a conduit,
helium feed stubs, and feedthrough seals. If there is a gas
leak, the gas density increases and the breakdown voltage is
lowered toward the minimum, which could lead to a spark
discharge. This is the usual situation in cryostats and junction boxes, where leaks from helium coolant lines or extraction of sensor wires are difficult to suppress completely. Irmisch has confirmed low voltage breakdown in helium gas at
cryogenic temperatures and low pressures (58), simulating
the effect of a helium leak into a vacuum space with a 10 mm
gap. At 6 K, the Paschen minimum was 240 V, instead of the
classic minimum of 160 V.
Leaks are suppressed by the avoidance of nonwelded seals,
design layouts that permit field repair of leaks, and vigilant
leak detection at room temperature and at cryogenic tempera-
tures. Some rules of thumb for leak rate testing are that the
leak rate at 1 atmosphere and room temperature is approximately 1000 times better than the leak rate at 10 atmospheres and 4 K. A state-of-the-art measurement, using mass
spectrometry, might test helium stubs down to leak rates of
1010 torr-L/s at room temperature, corresponding to 107
torr-L/s at 4 K.
It is generally acknowledged that quality assurance alone
wont guarantee an absence of electrical discharges in vacuum, because of the difficulty of guaranteeing zero leaks and
because of the deleterious effects on breakdown of modest
transverse magnetic field. Three design ideas have been proposed that can decrease the probability of discharges by several orders of magnitude: (1) grading of the insulating ground
planes, (2) the use of a guard vacuum, and (3) the use of insulating beads.
1. The use of graded ground planes around all insulating
surfaces has the greatest generality as a design concept.
If all metallic surfaces that are facing each other with
high potential differences are insulated, then there
cant be an arc between those surfaces acting as electrodes; although there could still be partial discharges
that would gradually degrade the insulation. However,
if each insulating surface or wrap has a ground plane,
the resistance of the ground plane can be graded, so
that the insulating or ground plane surfaces facing each
other never have a potential difference greater than 160
V. In this case, no combination of pulsed fields, capaci-
Table 7. Tracking Strength for 60 kV Pulse Waveforms (kV/cm)

Material
Mylar
Paper phenolic
Polyethylene
Nylon
Teflon
Permalin
Pressure
(atm)
LHe
4K
GHe
4K
GHe
293 K
LN2
77 K
GN2
77 K
GN2
293 K
0.9
1.6
0.9
1.6
0.9
1.6
0.9
1.6
0.9
1.6
0.9
1.6
79
79
95
118
118
158
79
118
95
158
68
79
3.6
2.7
2.8
2.3
3.3
2.6
2.7
2.0
2.0
2.8
3.1
2.2
2.8
158
189
158
189
158
236
158
236
79
158
118
158
39
79
32
39
30
59
47
59
43
53
30
68
4.7
6.8
4.3
7.9
5.9
7.9
4.7
7.9
4.7
7.9
4.7
7.9
4.3
4.3
3.8
3.9
3.9
3.9

Table 8. Typical Electrical Insulation Sizing Rules
Type
Flashover along insulating spacer in helium
Tracking along insulating wrap
Breakdown through epoxy
Breakdown through kapton wrap
Units
Design
Criterion
V/mil
V/mm
V/mil
V/mm
V/mil
kV/mm
V/mil
kV/mm
57
100300
1.55
60200
50150
1.56
125500
520
tive charges, or helium pressures can cause a breakdown.

2. The use of a guard vacuum around a winding pack can
be very effective, where space constraints arent significant, as in the Mirror Fusion Test Facility-B (MFTF-B)
coils (59). The secondary vacuum space greatly decreases the probability of a leak into the cryostat by requiring two series independent leaks in the coil case
and the guard vacuum shell. It can further lessen the
possibility of operational problems by the use of independent on-line leak detection and/or differential vacuum pumping. In MFTF-B, it also had a fringe benefit
of being used as an auxiliary helium channel to reduce
thermal stresses in the winding during cooldown.
3. Fast and Hart (60) have shown that filling an evacuated
box for power feedthroughs box with glass beads increased the Paschen minimum voltage and increased
the pressure at the Paschen minimum, as shown in Fig.
33. Testing in helium as a function of pressure from
1 Pa to 0.13 MPa (105 to 1 atm) at room temperature,
77 K, and 4.2 K in helium, they found that the minimum breakdown voltage across a 4.8 mm gap was increased by more than a factor of 10 at all three tempera-
Breakdown potential (kV)
10
tures. Insulating beads could be used to raise the

Paschen minimum by an order of magnitude, either in
evacuated spaces, such as junction boxes, or in high
pressure chambers, such as helium isolators with
filters.
CASE HISTORIES
The majority of actual failures to protect superconducting
magnets have had little relation to the design principles described in this chapter. Errors in operation have accounted
for more failures than errors in fabrication, and errors in design have been less important than either. Several studies
and workshops have addressed the problems of real-life magnet protection (61).
Operational errors causing magnet component failures
have included:
Bypassing protective circuits or not replacing them when
they fail
Wiring magnet or instrumentation leads backwards
Interpreting anomalous or changed sensor readings as
indicators of interesting experimental phenomena, instead of burnout
Poorly written, long and boring operating instruction
manuals, with incorrect or ambiguous instructions; the
ignoring of instructions
Absence of routine inspection and replacement of components with finite lifetimes
Energization of coils in an incorrect sequence (the classic
error is energizing an outer solenoid and crushing an inner solenoid with no strength in radial compression
through induced diamagnetic currents)
Increasing current when critical properties proved to be
better than expected, leading to burnout of vapor-cooled
lead
Unattended dewar operation leading to ice blocks, causing overpressure in the dewar
Removal of a rubber stopper in a dewar, leading to inadequate vapor flow to the leads; failure to monitor lead gas
flow or voltage drop
Manufacturing QA problems that led to coil or prototype
failure included:
1.0
0.1
25
10
100
Pressure (mm Hg)
(a)
1000
Breakdown voltage as a function

of residual GHe pressure:
,77K without beads;
,300K without beads;
,77K with beads;
,300K with beads;
(b)
Figure 33. Paschen curve for helium. (a) Measured in uniform gap
and (b) measured in gap with and without glass bead particles [Fast
and Hart (60)].
Helium leaks at fittings

Voids at the conductor/monolith interface leading to mechanical rupture of the conductor
Small metallic slivers being sprung from a conduit by the
tube mill
Metallic inclusions in an open, pool-boiling magnet causing turn-turn shorts and arcing
Insulation tearing during winding, especially of kapton
Joint installation without solder
Weld contamination due to contamination by injected
polyurethane spacers
Turn-turn short due to mechanical abrasion of insulation
during transportation and installation
Vacuum leaks through mechanical (i.e., nonwelded) seals
26
Exposure of Incoloy conduit to oxidants during heat

treatment, leading to stress aggravated grain boundary
oxidation (SAGBO) cracking
Reliability problems caused by inadequate redundancy in the
design of the protection circuits have included:
Dump circuit interrupters failing to clear
Power supplies failing to shut down on receiving a correct trigger signal from the protection circuit
False positive dump signals leading to arcing
Programming error in a programmable logic controller
leading to deliberately applying overcurrent to the coil.
Same problem with overvoltage
Arcing in a dump resistor causing a dump that was too
slow
Structural design errors have included:
An overly complex load path causing structural failure of
a brittle, aluminum primary structure
Inadequate mechanical protection of sensor insulation
Inadequate mechanical support of leads, causing lead
shorting or arcing
Peeling of solder lap joints from ends, neglecting strain
incompatibility
not only of the magnets, but also the leads, bus, and instrumentation feedthroughs. The goals of these improvements
would be to achieve orders of magnitude decreases in the
probability of failure to detect a quench rapidly, to successfully interrupt current in a magnet, and to simulate and adequately predict the ability of quench propagation to absorb
magnet energy without damage.
BIBLIOGRAPHY
1. C. Neumeyer et al., Quench protection circuits for superconducting magnets, Proc. IEEE Symp. Fusion Energy, U. Illinois, p.
1275, Sept 30Oct 5, 1995.
2. P. Dokopoulos and K. Kriechbaum, DC circuit breaker for 73 kA,
24 kV. Electrotech. Zeitschrift A. 97 (8): translated by H. Vogel,
Los Alamos Report LASL-TR-77-27, 1976.
3. S. Yokota, Poloidal field power supply using vacuum circuit
breaker, IAEA Second Large Tokamak Meeting, Princeton, New
Jersey, Nov. 1976.
4. V. Kuchinski et al., High-power fast switches, IPEC-Yokohama,
899, 1995.
5. Argonne National Laboratory Superconducting Magnet Group,
Final design of a superconducting MHD magnet for the CFFF,
Argonne National Laboratory, ANL-MHD-79-6, pp. 250251,
March, 1979.
6. C. Neumeyer, Liquid rheostat dump resistors, Princeton Plasma
Physics Laboratory 40-940217-PPPL/CNeumeyer-01, Feb 17,
1994.
SUMMARY
7. T. Ishigohka and Y. Kushiro, Quench protection of superconducting magnet using ZnO arrester, Cryogenics, 31: 562, 1991.
Superconducting magnet protection during a quench is an important but special case of the larger problem of magnet protection against structural, electrical, and thermal failures.
The usual method of guaranteeing protection is to build conservative magnets with more structure, stabilizer, and superconductor than they need. The techniques of providing adequate conservatism in the design of magnet protection
circuits, quench detection, protection criteria, quench propagation modeling, and electrical integrity have already been
described. However, experience with magnet failures suggests
that new design approaches may be more successful in the
future.
In most cases, redundancy should be favored, where possible, as a method of guaranteeing reliability, over just adding
more material, because of its ability to multiply low probabilities of failure. Redundant techniques include the use of in
series, independent interrupters; paralleled independent
quench detectors and paralleled independent cowound heaters. It also includes the use of leak-tight welds and guard
vacuums, insulation systems with no voids below the Paschen
minimum and long discharge or tracking life, and simultaneous use of signal-noise improvement techniques, such as selecting the best spot in the cable for a cowound sensor, differencing, and filtering the optimized signals.
Given the certainty that data processing will continue to
improve more rapidly than magnet technology, future magnets are likely to concentrate on using redundancy and intelligent signal processing in order to improve performance and
reliability simultaneously. Smart interlocks should check for
wiring errors, while simulators check and calibrate sensors.
Reviewed projects must adopt more stringent review standards, insisting on a complete design of the protection system,
8. A. Dudarev, V. E. Keilin, and Y. Kuroedov, Quench protection of

very large superconducting magnets, IEEE Trans. Appl. Supercond., 5: 226229, 1995.
9. S. M. Schoenung et al., Liquid helium dump concept for a large
scale superconducting magnetic energy storage plant, Adv. Cryo
Eng, 31: 1986.
10. J. F. Kaerner et al., A protection system for small high power
SMES with power semiconductors working at cryogenic temperatures, IEEE Trans. on Appl. Supercond., 5: June 1995.
11. O. P. Anashkin et al., Superconducting magnet protection system, Adv in Cryogenic Eng, 37: Part A, 339, Plenum Press, 1992.
12. J. Purcell et al., The superconducting magnet system for the 12foot Bubble Chamber. Argonne National Laboratory Report ANL/
HEP6813, 1968.
13. D. F. Sutter et al., Electrical protection of superconducting magnet systems, Fermi National Laboratory Technical Memorandum
TM-559, March 1975.
14. T. Ogitsu et al., Quench observation using quench antennas on
RHIC IR quadrupole magnets. IEEE Trans. Mag., 32: 3098,
July 1996.
15. S. J. St. Lorant communication reported in J. Allinger, G. Danby,
S. Y. Hsieh, J. Kean, J. Powell and A. Prodell, Fusion magnet
safety studies program superconducting magnet protection system and failure: Interim Report. Brookhaven National Laboratory, BNL 20787, Nov 1975.
16. M. Darweschsad, The POLO coil, a prototype tokamak poloidal
field coil, design features and test results. MT-14, Tampere, Finland; June 1995.
17. J. H. Schultz, Feasibility of the TPX voltage sensor quench detection system. M.I.T. Plasma Science and Fusion Center Report
PSFC/RR-97-3, March 5, 1997.
18. N. Mitchell et al., ITER No: N11 R1 03 04-09-07 W1, Magnet
design description: Appendix A: conductor design, April 14, 1995.

19. I. Zatz et al, TPX No: 94-950117-PPPL-IZatz-01, TPX structural
and cryogenic design criteria, Rev. 1, Jan 17, 1995.
20. J. H. Schultz, ITER No ITER/US/96/EV-MAG/J.H.SCHULTZ/
5.31/-1 J. H. Schultz, P. W. Wang, and S. Smith, Interpretation
of voltage sensor data in the QUELL experiment, May 31, 1996.
21. M. M. Steeves et al., The US Demonstration Poloidal Coil. IEEE
Trans. Mag., 27: 23692372, 1991.
22. Nicolai N. Martovetsky, and Michael R. Chaplin, Normal-zone
detection in tokamak superconducting magnets with co-wound
voltage sensors. IEEE Magnet Technol. Conf., Tampere, Finland,
July, 1995.
23. P. W. Wang, TPX Memo: 13-930719-MIT-PWANG-01, TF voltage
tap noise due to plasma disruption. July 19, 1993.
24. H. T. Yeh, J. S. Goddard, and S. Shen, Inductive voltage compensation in superconducting magnet systems, 8th Symp. Eng. Probs.
Fus. Res, 1802, 1979.
25. S. Smith and S. Ezekiel, TPX No:1314-941130-MIT/SSmith-01,
Improved fiberoptic quench detection at the Francis Bitter National Magnet Laboratory. Rev. 1, Nov 30, 1994.
26. J. H. Schultz and S. Smith, TPX Memo 1314-950815-MITJSchultz-01, Feasibility of the fiber optic temperature-sensor
quench detection system. Aug 15, 1995.
27. S. P. Smith, TPX No:1314-950424-MIT-SSmith-01, Report on fiberoptic strain sensitivity reduction experiments at the Francis
Bitter National Magnet Laboratory on 3/16/1995, April 24, 1995.
28. K. Koepke, P. Martin, and M. Kuchnir, Doubler system quench
detection threshold. IEEE Trans. Mag., MAG-19: 696, May 1983.
29. Y. Iwasa and M. Sinclair, Protection of large superconducting
magnets: maximum permissible undetected quench voltage.
Cryogenics, Dec. 1980.
30. Y. Iwasa, Case Studies in Superconducting Magnets: Design and
Operational Issues. New York: Plenum Press, 1994.
31. Y. Iwasa, Design and operational issues for 77 K superconducting
magnets, IEEE Trans. Mag., MG-24: 1211, 1988; Y. Iwasa et al.,
Stability and quenching in high-temperature superconductors,
32. C. H. Joshi and Y. Iwasa, Prediction of current decay and terminal voltages in adiabatic superconducting magnets, Cryogenics,
29: 157, 1989.
33. L. R. Turner, Safety of superconducting fusion magnets: twelve
problem areas, Argonne National Laboratory Report ANL/FPP/
TM-121, May 1979.
34. R. P. Krause and E. H. Christensen, Calculation of quench pressures in pool boiling magnets, IEEE Trans. Mag., 1979.
35. J. Powell, ed., Aspects of safety and reliability for fusion magnet
systems. Brookhaven National Laboratory, BNL 50542, 72-75,
Jan 1976.
27
41. Y. Wachi et al, Investigation of the pressure rise during the

quench of a force cooled superconducting coil, IEEE Trans. Mag.,
25: 15001503, 1989.
42. A. Shajii and J. P. Freidberg, Quench in superconducting magnets. II. Analytic solution, J. Appl. Phys., 76 (5): 1 Sept 1994; L.
Dresner, IEEE Trans. Mag., 25: 1710, 1989; L. Dresner, Proc 11th
Symp Fus Eng, IEEE, 1218, 2: 1986.
43. A. Shajii and J. P. Freidberg, J. Appl. Phys., 76 (5): 1 Sept 1994.
44. A. Shajii, Universal scaling laws for quench and thermal hydraulic quenchback in CICC coils, IEEE Trans. Appl. Supercond., 5:
477482, 1995.
45. G. Olivier, Y. Gervais, and D. Mukhedar, A new approach to compute uniform field breakdown of gases, IEEE Trans. Power Appar.
Syst., PAS-97 (3): May/June 1978.
46. Dielectric breakdown of cryogenic gases and liquids. J. Gerhold,
Cryogenics, Fig. 22, 1979.
47. Meek and Craggs, Electrical Breakdown of Gases. 662, Fig. 6.14,
New York: Wiley, 1978.
48. J. Gerhold, Dielectric breakdown of cryogenic gases and liquids,
Fig. 10, Cryogenics, 1979.
49. D. May and H. Krauth, Influence of the electrode surface condition on the breakdown of liquid helium. IEEE Trans. Mag., MAG17 (5): Sept 1981.
50. Ch. Olivier, Performance of electrodes at the first voltage breakdown in liquid helium, IEEE Trans. Mag., MAG-17 (5): Sept.
1981.
51. Fallou, Dielectric breakdown of gaseous He at very low temperatures. Cryogenics, April 1970.
52. R. J. Meats, Butt gap discharges in laminated insulation impregnated with very cold helium gas. Third Int. Conf. Gas Discharges,
IEEE, London, 1974.
53. J. Hiley and R. S. Dhariwal, Dielectric breakdown in high density
helium and in helium impregnated solid dielectrics. Cryogenics,
25: 334, June 1985.
54. M. Zahn, Preliminary engineering design criteria to prevent
spark discharges in TPX, Nov 27, 1994, TPX Memorandum No.
1314-941127-MIT-MZahn-01.
55. A. M. Miri, Transient voltage oscillations in a large superconductive coil. 9th Int. Symp. High Voltage Eng., Graz, Austria, Aug
28Sept 1, 1994.
56. A. Migliori, R. I. Schermer, and M. D. Henke, Dielectric tracking
in liquid helium. Cryogenics, 442, 1978.
57. R. A. Haarman and K. D. Williamson, Jr., Electrical breakdown
and tracking characteristics of pulsed high voltages in cryogenic
helium and nitrogen. Adv Cryo Eng. 21 (102): Plenum Press,
1976.
36. J. R. Miller, J. W. Lue, and L. Dresner, IEEE Trans. Mag., MAG13 (24): 1977.
58. M. Irmisch et al, Breakdown characteristic of helium gas at cryogenic temperatures and low pressures with respect to a local helium leak. KFK Report 31-03-05 P02B, June 1992.
37. Z. J. J. Stekly, Adv. Cryo. Eng., 8 (ed.) Timmerhaus K.D. New

York, Plenum Press, 1963, 585.
59. C. Henning, Reliability of large superconducting magnets

through design. IEEE Trans. Mag., MAG-17 (1): 618, Jan. 1981.
38. L. Bottura and O. C. Zienkiewicz, Quench analysis of large superconducting magnets, Cryogenics, 32 (7): 659, 1992; L. Bottura, A
numerical model for the analysis of the ITER CICCs, J. Comp.
Phys, 125 (26): 1996; L. Bottura, Numerical aspects in the simulation of thermohydraulic transients in CICCs. J. Fusion Eng.,
14 (1): 13, 1995.
60. R. W. Fast and H. L. Hart, Use of glass beads to increase the

breakdown voltage in subatmosphere, cold helium gas. Advances
in Cryogenic Engineering, 35: Plenum Press, New York, 1990,
809812.
39. L. Dresner, Protection considerations for force-cooled superconductors, Proc. 11th Symp. Fusion Eng., II: 12181222, Austin TX;
Nov. 1822, 1985.
40. L. Dresner, Thermal expulsion of helium from a quenching cablein-conduit conductor, Proc. 9th Symp. Eng. Prob. Fusion Res., 618,
Chicago, 1981; reformulated in L. Bottura, Quench propagation
and protection in cable-in-conduit superconductors. Encyclopedia
of Applied Superconductivity, Comett 4090 Cb, 1996.
61. P. G. Marston et al, Magnet Failure Workshop, Journal de Physique, Colloque C1, Tome 45, Jan 1984, C1-637; F. Arendt and
P. Komarek, Potential failures and hazards in superconducting
magnet systems for fusion reactors. Nuc Tech/Fusion, 1: Oct,
1981; R. J. Thome, J. Bart Czirr, and J. H. Schultz, Survey of
selected magnet failures and accidents.
JOEL H. SCHULTZ
Massachusetts Institute of
Technology
SUPERCONDUCTORS, CRYOGENIC STABILIZATION

Kamerlingh Onnes (1) discovered superconductivity in
1911 when he was measuring the electrical conductivity of
mercury as a function of temperature down to the temperature of liquid helium. He found that at 4.15 K the electrical
resistivity (, m) became too small to measure. Although
it is not possible to prove experimentally that a quantity is
exactly zero, experiments to date have been able to show
that the resistivity of a metal in the superconducting state
is less than 1022 cm (compared to the resistivity of pure
copper at low temperature, 109 cm) (2). Before long, a
number of other elements were also found to exhibit this
same phenomenon at similar low temperatures. A list of a
few of these materials, called superconductors, with their
transition (or critical) temperatures, is given in Table 1 (2
4).
Even at the time of its discovery, Kamerlingh Onnes
realized that the phenomenon of superconductivity could
have important technological uses. However, it was soon
discovered that these early superconductors (the so-called
type I superconductors) remained in the superconducting
state only if they were not carrying a substantial electric
current (I, A) and if they were not in the presence of any
substantial magnetic eld (H,A/m or T: see Table 1, footnote
b). As the external magnetic eld is intensied, or as the
electrical current within the superconductor is increased,
its normal-mode electrical resistivity is restored at a critical value of either current or eld. Hence, the superconductors were inherently unstable under certain conditions. As
Silsbee suggested in 1916, these are not two unrelated phenomena, because the magnetic eld generated by passing
a current through the wire destroys the superconductivity at the same value as does an externally applied eld
(5). The temperature at which superconductivity appears
in zero applied magnetic eld and carrying no current is
called the critical temperature (Tc , K). However, the practical transition temperature (i.e., the maximum temperature at which a superconductor exhibits superconductivity)
is a function of both external magnetic eld and the current within the wire, as shown in Fig. 1. It can be seen
that practical operation of a superconductor must be made
within the parameter space under the surface F = f(I, H,
T) shown, and thus below certain limits on temperature,
current, and magnetic eld.
Both the critical currents and the critical elds of the
early superconductors were small. Consequently, although
the discovery of superconductivity was made early in twentieth century, its practical use for producing strong magnetic elds was not realized until much later. An efcient
design of superconducting magnets for large physics experiments and energy devices has had to wait until near the
end of the twentieth century, (6).
By the 1950s, experimentation with intermetallic compounds and alloys had led to the discovery of materials that
greatly relieved the above limitations. These materials remain in the superconducting state to somewhat higher
temperatures (see Table 1) in higher magnetic elds, and
also are able to transport larger electrical currents. They
Figure 1. Critical linearized F(I, H, T) surface of a superconductor. It should be noted that in a typical magnet-grade superconductor the critical current I, magnetic eld H, and temperature T are
not linearly related. However, the classical theory of cryostability
assumes linear relationships.
were given the name type II superconductors. Prior to

the 21st century it was believed that the upper temperature limit of type II superconductivity was that of Nb3 Ge
at 23.2 K. In 2001, the discovery of superconductivity in
MgB2 raised that upper temperature limit of the previously called low temperature superconductors to approximately 40 K.
A fundamental difference in the behavior of these two
types of superconductors is in the way in which a magnetic
eld enters the material, as illustrated in Fig. 2(a) and
2(b). In type I materials, magnetic eld is excluded from
the superconductor [Fig. 2(a)]. However, when any one of
the limits of critical temperature (Tc ), magnetic eld (Hc ),
and current (Ic ) is exceeded, there is an abrupt and total
entry of any external eld into the material as the material loses its superconductivity. In type II superconductors,
above a threshold eld Hc1 , the eld begins to enter in discrete units of eld called uxoids, and superconductivity is
totally destroyed only when the eld Hc2 , has completely
entered the superconductor [see Fig. 2(b) and 2(c)].
The important characteristic of type II superconductors (alloys such as NbTi or intermetallic compounds like
Nb3 Sn) is their capacity to sustain high transport currents
(7), which makes them suitable for use in high-current devices. However, for large enough magnetic eld (and/or current) these superconductors eventually pass into the normal state as well. Consequently, these superconductors can
also become unstable. For H < Hc1 , a type II superconductor is in the superconducting state; for Hc1 < H < Hc2 , the
superconductor is in a mixed state (the magnetic eld penetrates into the regions existing in the normal state, but
bulk superconductivity is not extinguished); and for H >
Hc2 , the superconductor is in the normal, resistive state
[Fig. 2(c)]. Type I superconductors have a critical eld, Hc ,
below which there is no eld within the material and it
is superconducting [Fig. 2(a)]. In contrast, type II superconductors have two critical elds: the lower critical eld
Hc1 , at which the magnetic eld begins to move into the
superconductor, and the upper critical eld Hc2 , at which
the penetration is complete and superconductivity is de-
Superconductors, Cryogenic Stabilization
Figure 2. Superconductor types. (a) Magnetization (magnetic

moment per unit volume, M; 4M is also used) versus magnetic
eld for a type I superconductor. The magnetization decreases
abruptly from the Meissner value to zero. (b) Magnetization versus magnetic eld for a type II superconductor. The magnetization decreases monotonically from the Meissner value, hence
providing three distinct regions: (1) the Meissner state (magnetic
ux completely excluded), (2) the vortex state, and (3) the normal
state. In the vortex state magnetic ux penetrates in the form of
vortices but the material is still superconducting. The dotted line
represents the comparison with the case for a type I superconductor. (c) Magnetic eld versus temperature fora low-temperature
superconductor. (d) Magnetic eld versus temperature for a hightemperature superconductor.
stroyed; see Fig. 2(b) and (c).

An important physical property of the intermetallic
compound type II superconductors is that they are brittle
and their fabrication into useful shapes may require that
the superconducting compound be produced by heat treatment after the wire has been formed in its nal shape (3).
In the case of Nb3 Sn, the wire is drawn to its nal dimension before heat treatment causes the reaction between the
Nb and the Sn to form the compound.
In the early 1980s new compounds were discovered that
were able to retain superconductivity up to very much
higher temperatures. The rst of these compounds was
La2x Bax CuO4 with a critical temperature of 35 K (8).
Since that time, other compounds with higher and higher
operating temperatures have been developed. Now, practical superconductors that retain superconducting properties to temperatures near 135 K are available (4, 9). These
high-temperature superconductors (HTSCs) are different
from the conventional superconductors in that they are
complicated oxide compounds. In addition, they are granular and exhibit their superconductivity only along certain
planes within the crystal. Thus, in an application requiring
a signicant length, it is necessary to fabricate the nished
superconductor with the grains aligned along the superconducting plane with no more than about 7 misalignment
(9). A greater angle between the grains results in weak
links between them, which drastically limits their ability
to transmit current. Methods of addressing the alignment
problem have been developed.
One type of HTSC consists of the oxides of barium, strontium, calcium, and copper, commonly referred
to as BSCCO, one example being Ba2 Sr2 Ca2 Cu3 O8 , or
2223BSCCO. For the BSCCO compounds the oxidepowder-in-tube (OPIT) process (4) has succeeded in reducing the weak link problem to a manageable level by produc-
ing continuous conductors in lengths up to one kilometer

capable of carrying acceptably high current densities. However, in a plot of critical magnetic eld versus temperature,
only a part of the area under the curve showing the phase
space for superconductivity is available for the transport of
current. Such a plot is shown in Fig. 2(d), which includes
an additional line called the line of irreversibility, which
varies considerably from one compound to another. This
line shows a practical limit of current-carrying capability
in that above it the ux is no longer pinned and the currentcarrying capability vanishes, even though operation may
still be within the superconducting envelope. The line of irreversibility is quite low for the highly anisotropic BSCCO:
operation in magnetic elds above 1 T is only possible at
temperatures below 40 K. Another HTSC is YBa2 Cu3 O7 ,
or YBCO. The more isotropic YBCO compound has a much
higher line of irreversibility, making possible operation at
useful current densities in magnetic elds up to 5 T at 77
K (liquid nitrogen temperature). The OPIT process does
not work to reduce weak links in YCBO, and useful YBCO
conductors have only been produced by deposition onto specially textured substrates (10).
Like the intermetallic compounds of the type II superconductors, the HTSCs are brittle and difcult to fabricate.
A major part of the effort to produce useful HTSCs has been
toward their fabrication into useful shapes (9).
EXPERIENCE FROM EARLY MAGNET APPLICATIONS

The discovery of the type II superconductors allowed the
superconducting state to be maintained in the presence of
higher currents and at more elevated magnetic elds (compared to the previous situation with type I superconductors). Applications of superconductivity have been made in
instrumentation devices, and a considerable amount of investigation into superconducting electric power transmission lines has been done (11, 12), but no such power lines
have been built to date. The greatest application of superconductivity on a large scale has been in the building of
magnets with very high eld capabilities (13).
The achievement of the higher operating parameters of
the type II superconductors was not in itself sufcient for
the construction of high-eld magnets with maximum performance. All the superconducting magnets designed until the early sixties of the twentieth century, without exception, suffered from so-called degradation, i.e., from the
loss of the superconducting state before reaching the full
design eld. This phenomenon has been attributed to the
fact that internal and/or external disturbances trigger an
irreversible instability causing the appearance of normal
zones followed by a so-called quench (the destruction of
the magnetic eld, with the stored energy being converted
to heat). Degradation can be attributed to a number of
sources, which may be mechanical, magnetic, or thermal
in nature. Any source of heat in the system can cause local
temperature increases in the superconductor to the point
where the appearance of a normal resistance region is imminent. In early attempts to build large magnets it was
discovered that because of effects like these, the performance of superconducting wires fell far short of the shortsample results, a phenomenon called coil degradation (14).
A magnet would quench after reaching only a fraction of
its design eld. However, after each such quench, the next
try resulted in a higher operational eld. After repeated attempts, the eld attained can be considerably higher than
that attained at rst, but not as high as the short-sample
results. This phenomenon is called training.
At rst it was thought that degradation was the result
of weak spots in the long lengths of superconducting wire.
However, this would not explain training, and production
techniques have been shown to produce remarkably uniform properties in long lengths of superconducting wires
(many kilometers, even in small magnets). It was concluded that degradation must be caused by one or more
of a number of possible disturbances, such as a source of
heat or the penetration or rearrangement of the magnetic
eld.
Such disturbances can lead to severe consequences, in
which both functional and mechanical integrity of the device can be compromised: It can become unstable. An engineering application cannot tolerate such an instability.
It has become obvious that a new technological solution is
needed to overcome these difculties.
CAUSES OF INSTABILITIES (DISTURBANCES)

There are several mechanisms that can cause the generation of heat within a superconductor carrying current in
a magnetic eld. Because of the very low heat capacity of
most materials at the low operating temperature, the generation of heat can cause a sufcient rise in temperature
to create a local zone in the superconductor that is above
the critical temperature and therefore has normal resistance. The normal zone then becomes an additional source
of heat because of Joule heating (I2 R, R being the electrical

resistance, ). The consequent increase in temperature can
exacerbate one or more of the processes creating the heat.
This positive feedback (without a stabilizing inuence) can
lead to a runaway situation. Regardless of the cause of instability, an uncontrolled growth of the normal zone will
promote further heat generation, and in the worst case
cause a quench, and all of the energy in the entire assembly,
which may be hundreds of megajoules, can be released as
heat, generating a pressure increase in the cooling medium
and overheating the entire assembly, possibly even catastrophically. The prevention of such an occurrence is a task
of stabilizing the superconductor assembly.
Disturbances to superconducting systems can be abrupt
and cause local heating, so that a maximum local effect
results. Other disturbances are prolonged in nature and
tend to heat larger portions of the system, so that their
effect is distributed more widely. Causes of disturbances
have been classied as magnetic, mechanical, and thermal,
as indicated in Fig. 3.
The forces generated by the magnetic eld tend to
stretch the magnet structure radially while causing it to
contract axially (similar to the reaction to a pressure increase of a cylindrical pressure vessel with ends closed by
sliding pistons). This straining of the magnet system can
cause sliding friction, or even the sudden rupture of components such as epoxy used for potting. Any sudden motion
like this will generate heat as the motion is stopped. Once
the external magnetic eld reaches Hc1 , magnetic eld penetration begins and the motion of the uxoids also generates heat. Because the magnet system is operated well below ambient temperature, there is a constant input of heat,
which may heat some portions of the system unequally.
Any one of these disturbances can result in a normal zone
within the superconductor.
Flux Jumps
The most serious disturbance is caused by ux jumps. In
the mixed state, magnetic ux is present within the superconductor. In the presence of a current passing through the
superconductor the uxoids are acted upon by the Lorentz
force, which is perpendicular to the surface of the wire
and is proportional to the product of the current density J
(A/m2 ) and the magnetic induction B (T)in vector form, J
B. The uxoids are pinned to the crystal latice by imperfections or impurities, so that motion does not occur unless
a disturbance creates a location where the Lorentz force
exceeds the pinning force.
Flux-jump instability is a characteristic of the superconductor itself, not the magnet as a whole. According to
Wada et al. (15), uxoids can be dened as quantized magnetic ux lines distributed within a type II superconductor.
Degradation caused by a ux jump is related to a sudden
motion of the uxoid within the superconductor (16). The
collective, discontinuous motion of uxoids in such a superconductor can be caused by mechanical, thermal, magnetic,
or electrical disturbances. Breakaway of the moving uxoid
vortices is inherently associated with Joule heating. The
increase of the local temperature may be sufciently large
to cause an avalanche of a uxoid movement, thus creat-
Figure 3. Instability sources. The instability disturbances are

either magnetic (notably a ux jump), mechanical (mostly responses to thermal stresses), and thermal (such as heat inleaks).
ing additional Joule heating and a subsequent increase in

temperature. This redistribution of the magnetic eld in
the superconductor is called a ux jump.
Mechanical Disturbances
Mechanical inuences may be responsible for a disturbance in a superconductor that can lead to instability (17).
For example, during the cooldown period before the startup
of operation, as well as during the subsequent period of
charging, stresses of various origins are always present in
superconducting windings. These may be the thermoelastic stresses caused by the difference in thermal contraction
of composite components such as superconductors, normal
conducting material (e.g., copper), and epoxy impregnate.
Similar stresses can arise from thermal gradients within
the system, and stresses may arise during the manufacture
of the composite conductor (drawing) and magnetic coils
(winding and banding). Stresses are also caused by magnetic forces during charging. Small conductor movements
activated during eld cycling may cause local heating (18).
These stresses may lead to yielding or even tensile failure of a component in the magnet system such as the
potting resin, or to shear failure of the bond between an
epoxy impregnant and the conductor. If thermal or magnetic stresses become great enough to cause some of the
magnet material to yield, the yielding can take place in a
series of discontinuous jumps (serrated yielding), which,
because of their rapidity, cause rapid heating. Heat is produced as the motion is arrested. Because of the low temperature, many materials are subject to cold embrittlement.
Brittle fracture can be more serious than serrated yielding
because the energy is all deposited locally.
The friction between the metallic and insulated surfaces
can cause local heating. If this stress results in a stickslip
process, the effect will be similar to the brittle fracture
mentioned above.
Any of these mechanisms can constitute a disturbance
energy that may be strong enough to initiate a quench.
Studies of thermomagnetomechanical instabilities are numerous (1922); however, the problem is open for further
studies.
Distributed Heat Sources

Some of the above disturbances can result in slower, more
distributed heat sources. However, there are additional
sources of heat that are distributed widely through the
magnet system. Although these heat inputs may not directly cause local heating within the magnet system, the
heat transfer to the coolant must be sufcient to absorb the
corresponding energy input.
Heat Leaks. Because essentially all superconductors are

operated at temperatures in the cryogenic range, an ever
present source of heat is the heat that enters the system
through the thermal insulation, piping, and the support
system. No insulation is perfect, so that this heat source
is inevitable. Where, and how much, heat enters is determined by the design of the cryogenic system. The heat leak
through the thermal insulation is usually distributed uniformly. However, the heat admitted through current leads,
instrumentation, and piping may be more localized.
Hysteretic Losses. In some cases superconductors must

carry an alternating or pulsed current. In these cases, a
continuous energy loss occurs and manifests itself as a heat
source because of the Lorentz force on the unpinned ux
lines that move in and out of the superconductor as a result
of the varying current. These losses are called hysteretic
losses (3). They are inuenced by the roughness of the surface of the superconductor. The smoother the surface, the
lower the losses. Even more important is the twist pitch
of the superconducting laments and the resistivity of the
matrix material (discussed below).
METHODS OF STABILIZATION
Because there are inevitable inputs of heat to the system
as well as disturbances that can affect the temperature of
the superconductor, it is necessary to provide methods to
maintain the system at its operating temperature to prevent the catastrophic loss of the superconductivity. In the
case of a superconducting magnet, the key problem of stability is that of sustaining the successful operation of the
magnet system without loss of the magnetic eld and without damage to the system. The problem is not only a possible appearance of normal zones within the superconductor,
but the need to sustain conditions for the reestablishment
of the superconducting state after the appearance of instability. Furthermore, the continuous operation of the device
during the transition period must be assured.
Flux-jump stabilization suppresses an initiation of a cycle of disturbances that may cause the transition into the
normal state. Thus, it is directed toward the prevention of
an instability. The role of cryostabilization, in contrast, is to
restore the superconducting mode of operation, once a disturbance has already initiated the existence of a normal
zone.
There are several ways to approach the problem of stabilizing a superconducting system. First, the design of the superconductor wire can be used to minimize, or possibly even
prevent, the damaging effect of a ux jump. The energy
released during the passage of the ux lines through the
conductor is proportional to the distance traveled across
the wire. Thus the conductor is usually fabricated to consist of many very ne laments (10 m to 100 m in diameter). These laments are distributed within a matrix
of a metal, which, while not a superconductor, does provide a highly conductive path for the current in case the
superconductor can no longer handle the entire current.
Typically the matrix metal is made of copper. However, for
HTSCs silver is also used. One of the ac loss mechanisms
arises from a coupling of the currents in adjacent superconductor laments, and this loss can be minimized by twisting the superconductor composite so that the laments are
transposed along the length of the wire (3). An additional
measure that can be directed toward a suppression of the
ac losses is adding a more resistive metal, such as CuNi,
around the outer surface of the superconductor, thus decoupling the superconductor laments and reducing the eddy
currents owing in the normal metal matrix (3).
Second, heat generated within the magnet system must
be removed to maintain an acceptable operating temperature. This cooling can be provided by pool boiling in a
suitable cryogenic uid, or by the forced ow of the cryogen, either as a liquid, or as a uid at supercritical pressure to avoid the problems of two-phase ow. In some cases
where very low-temperature operation is required, use has
been made of superuid helium to obtain its excellent heat
transfer capabilities. In this case an operating temperature
in the vicinity of 2 K must be used.
The stabilization of the HTSCs utilizes the same methods as those for type II superconductors. In one respect,
the task is eased by the fact that the specic heats of the
magnet system become considerably higher as the temperature is raised. Thus the temperature increase from a given
amount of heat is less than it would be at lower temperature. Another factor in favor of stabilization comes into
play when the system is operated at the higher end of its
temperature range. In this case, the current cannot be as
high, which results in lower Lorentz forces, and thus the
likelihood of ux jumps is diminished.
Cryogenic Stabilization
According to Reid et al. (23), cryogenic stabilization is
achieved if, after the release of a certain amount of Joule
heat and a local rise in temperature (caused by either internal or external perturbations), efcient cooling is provided
to remove that thermal energy more rapidly than it is generated. This goal assumes balancing of the following energy ows: (1) the energy brought to an intrinsically stable
superconductor in the form of a thermomechanical disturbance or thermal energy from any source, including that
generated by the current that is redirected into the stabilizing, resistive matrix, and (2) the energy removed from a
superconductor element by convective cooling and by conduction. These processes can be highly transient in nature;
indeed, their duration is usually very short (on the order
of 103 to 102 s). Consequently, the use of a steady-state
energy balance should lead to a very conservative stability
criterion.
Cryostabilization can be full or limited. Full cryostabilization means stable operation after the entire conductor
has been driven normal by a large disturbance (23). Design
based on full cryostabilization is as a rule the most conservative and involves large conductors. Limited cryostabilization refers to recovery from a disturbance of limited
size. By using the stability criteria, one can decide whether
the superconductor is going to be stable (in the sense of either full or limited stabilization) or unstable. For example,
full cryostability can be dened either using the so-called
Stekly criterion or the MaddokJamesNorris model (see
below).
All the early-developed stability criteria discussed in
the following sections have been based on quasi-steadystate balances. The theory behind these static criteria will
be (somewhat arbitrarily) called the classical theory of
cryostabilization. The formulation of a transient problem
and some of the issues involved will be given subsequently.
The Stekly Criterion. Stekly and his collaborators (24
26) were the rst to formulate a method to prevent a catastrophic quench in a magnet. They discovered that reliable
stabilization can be achieved by the simultaneous application of two measures: (1) an alternate path for the current
through an adjacent material (such as copper) with high
electrical conductivity (although nonsuperconducting) on
the appearance of normal zones in the superconductor, and
(2) very efcient cooling, such as in liquid helium. The
so-called current sharing between a superconductor that
abruptly loses its superconducting capability (becoming
highly resistive) and the normal conductor (the matrix in
which the superconductor is embedded, having a smaller
resistivity than the adjacent superconductor in its normal
state) secures a continuation of the magnet operation. By
the intense cooling, the conductor can then be cooled back
below the critical temperature for a given magnetic eld

and electrical current. This, in turn, leads to a reestablishment of the superconducting state in the coil. Therefore,
heat transfer to a cooling stream in addition to conduction
through the conductor and the assembly will remove the
generated heat and provide the conditions for the superconductor to reassume the total current ow, along with
a disappearance of the current ow through the normal
conductor, but without losing the magnetic eld. Consequently, the coil becomes stable, and no quench will result.
This approach constitutes cryogenic stabilization.
The Concept of Stability and the Energy Balance. The key
modeling tool needed to dene exactly the concept of stability is the energy balance of a conductor and/or magnetic
device (a coil), which may be described in general as follows:
This balance assumes a composite conductor material surrounded by a coolant and subject to transients as well as
internal and external instabilities. The goal of a design is to
keep the operating point of the conductor within the limits imposed by the critical surface (T, I, H)c but with an
additional requirement formulated as follows: If the disturbance upsets this operation, the restoration of the superconducting state is still possible.
Let us rst formulate the classical theory of cryostability, the concept originally introduced by Stekly and collaborators (25, 26). We will not present this approach either
in its entirety or in a chronological perspective. Rather, we
will discuss the main points, which depend upon an energy
balance that is astonishingly simple compared to the complexity of the superconducting instability phenomena. This
simplicity commends this criterion as the most conservative of several that have been developed over the years.
First, the physical background for the analysis should
be emphasized. Three distinct physical situations can be
distinguished in a conductor operating at a given magnetic
eld H, carrying constant current I, and operating at various temperatures T. These situations are indicated in Fig.
4(a) as a superconducting mode, a current-sharing mode,
and a normal mode.
In a (completely) superconducting mode, the conductors

operating current is less than critical, for a less than critical magnetic eld, at a less than critical temperature. In
a current-sharing mode, current ows partly through the
matrix because of the appearance in the superconductor of
normal zones characterized by a resistance that is much
higher than the resistance of the matrix. Finally, when the
temperature increases above the critical temperature, the
superconductor operates in the normal mode and the current is carried exclusively by the matrix.
In an adjacent diagram [Fig. 4(b)], the corresponding
distribution of thermal energy generated by Joule heating
is presented. In the superconducting-mode Joule heating
is zero. In a current-sharing mode thermal energy generation per unit of area of the conductor surface is equal to GA
= IIm (/AP)m = I(I Ic )(/AP)m (W/m2 ), where m (m) is
the electrical resistivity of the resistive part (matrix), and
Im = I Ic (A) is the current through the matrix. The symbols I and Ic represent the operating and critical currents
as dened in Fig. 4(a). The quantities Am (m2 ) and Pm (m)
are the cross sectional area of the matrix and the conductor perimeter, respectively. (Note that the units for electrical resistivity, area, and length, often used in practice, are
cm, cm2 , and cm, respectively. Then the generated thermal energy per unit of conductor area is in W/cm2 .) In a
normal mode, say for I = Ic,b , thermal energy generation
has a constant value of I2 c,b (/AP)m .
To start the analysis we introduce a series of farreaching assumptions but still preserve the main features
of the process:
The conductor is a composite, that is, it consists of

a superconducting core and a highly conducting (but

nonsuperconducting) matrix.
The heat transfer phenomena involved, including the
eventual release of thermal energy caused by a disturbance, are quasisteady.
The temperature of the conductor is uniform across its
cross section, and there is no heat conduction through
the conductor.
The electrical resistance between the core and matrix
of the conductor is negligible.
The electrical resistivity of a normal zone in the superconductor is several orders of magnitude larger than
the resistivity of the matrix (thus, if the conductor operates in a normal, resistive mode, the electrical current tends to ow through the matrix).
The electrical and thermal properties of the conductor materials and kinetic properties of the processes
involved are independent of temperature.
The general energy balance equation in this case should include only thermal energy generation and heat convection,
that is,
The concept of cryostability assumes that the rate of thermal energy generation caused by instability is, in a limit,
equal to the rate of heat removal by either pool boiling or
forced convection of liquid helium (usually supercritical,
and in some cases superuid). This assumes that current
may ow partly through the superconductor and partly
through the matrix (current sharing) or it may be completely rerouted to the matrix if the superconductor becomes resistive (normal mode of operation). If the energy
generated by Joule heating (both in the superconductor
and in the matrix, in the current sharing mode, or entirely
Figure 4. (a) ITH characteristic of a conductor: a linearized

relationship f3 (I, T, H xed); see also Fig. 1. Increase in the
local temperature at given operating current I forces a superconductor to change from superconducting mode (state
1 ) to a
current-sharing mode (state
3 ) and ultimately into the normal
4 ). In a current-sharing mode a part of the curmode (state
rent ows through the resistive nonsuperconducting matrix and
a part through the partially resistive superconductor. In the normal mode current ows only through the matrix, being excluded
from the resistive superconducting part by its much higher electrical resistivity. (b) Thermal energy generation by Joule heating
versus temperature for a given current. In the case of no current
sharing, Joule heating starts abruptly at, say Tc,0 (dotted line). In
the case of current sharing, thermal energy generation increases
linearly with T in the current-sharing zone. In the superconducting mode there is no Joule heating (GA = 0). In the normal zone
the Joule heating has a constant value (for a given current).
through the matrix, in the normal mode) is more than compensated by the heat removed from the conductor, its temperature will return to below the critical temperature, and
the conductor will stay stable. So the stability criterion can
be expressed as follows:
where is the so-called Stekly stability parameter. The

above-dened cryostability criterion can be written in an
explicit form taking into account the energy generation under the condition that all the current is owing through the
matrix at the onset of normal mode of conductor operation.
In this case the density of heat transfer rate generated
within the conductor and normalized to the unit of heat
transfer area between the conductor and coolant is given
by
where VL in (Vm1 or Vcm1 ) is the voltage drop per unit

length of the composite conductor. The right-hand side in
the last equality of Eq. (4) expresses the thermal energy
generation in terms of the current density (J = I/A), the
conductor diameter D, and the so-called lling factor =
Asuperconductor /Atotal . The heat transfer rate density QA (W/m2
or W/cm2 ) on the matrix surface is
Therefore, the Stekly parameter is given as
When a composite conductor operates under current sharing, the total current I is the sum of the currents owing
through the superconductor, Ic , and the matrix, Im . In general,
Under the current-sharing conditions, the voltage drop per

unit length of the composite conductor, VL , is given as follows (27):
Here a linearity of the critical IT curve has been assumed.

Equation (8) can be rearranged as follows:
or
where v = VL /[Ic,b (m /Am )], i = I/Ic,b , and = (TI Tb )/(Tc,0

Tb ) are the reduced values of the voltage drop, current,
and temperature, respectively. The modes of conductor operation can be presented formally using a current-sharing
factor (27) f dened as follows:
and consequently,
From Eqs. (5) and (6), after rearrangement and introduction of the reduced voltage drop, current, and temperature,
as well as the Stekly parameter, one can obtain
By eliminating the reduced temperature from Eqs. (10) and

(13) one can obtain the reduced voltage as a function of the
reduced current in the matrix. This situation describes the
variation of voltage with current. An interesting insight
can be gained from the vi diagram using the relations
given by Eqs. (10) and (13) and recognizing that the v = i
line in that diagram depicts the normal state [Fig. 5(a)].
The voltagecurrent relationship features are described in
the legend of Fig. 5(a).
If the current is assumed to be xed, an insight into
a conductor modes of operation can be gained by using
a diagram of the reduced thermal energy generation g =
GA /[h(Tc,0 Tb )], and/or the reduced heat transfer rate
Ref. q = QA /[h(Tc,0 Tb )], versus reduced temperature [Fig.
5(b)], according to Gauster as reported in Ref. 27.
Finally, it should be pointed out that several denitions
of the stability factor exist in the literature. In addition to
the Stekly parameter , a stabilization parameter = 1/
can be used (28):
where also a parameter p has been introduced, equal

to the fraction of matrix perimeter actually exposed to
Figure 5. Reduced characteristics. (a) Reduced voltage versus

reduced current. If the conductor is fully superconducting (i.e., f
= Im /I = 0), then 0 < i < 1 (line 01). Current-sharing cases [f
= ( + i 1)/i, for linear Ic versus T] correspond to the range of
reduced current 1 < i < 1/1/2 . Full cryostability corresponds to the
01C curve ( < 1) and sufcient cooling. The 01A curve ( > 1)
corresponds to insufcient cooling. For < 1 the reduced voltage v
is a single-valued function of i. With an increase in i from 0 to 1 the
superconductor quenches at i = 1. With decreasing current while
in normal mode (along the v = i line) the conductor switches back
to the superconducting mode at i = 1/1/2 . (b) Reduced thermal
energy generation g and reduced cooling rate q versus reduced
temperature for a xed reduced current (for i > 1/1/2 , > 1).
Stable conditions for g < q.
coolant. If the stabilization parameter as dened by Eq.

(14) is to be used, note that > 1 implies a stable system
and < 1 an unstable one. According to Ref. 17 the stabilization parameter can be dened either at the critical
temperature and the operating current, or at the critical
temperature and zero current and magnetic eld. In addition, a separate parameter can be dened for the currentsharing case.
Prediction of the stability using the Stekly cryostability parameter has opened a new era in building large superconducting magnets. From Eq. (6) is obvious that, for
a given superconductor and matrix material (usually copper) as well as for the dened cooling conditions, to achieve
greater stability, one should increase the ratio of matrix to

superconductor in the cross section of the composite conductor. This criterion has led to very conservative design
( can be smaller than 0.1, i.e., larger than 10), leading
to very expensive solutions.
Let us consider an application of the Stekly criterion
in an actual conductor design (for more details consult
Ref. 27). For a typical critical current density of the order of magnitude of 105 A/cm2 and a standard cooling heat
transfer rate of 2 101 W/cm2 (a composite conductor
with a low-temperature superconductor cooled by liquid
helium), a change of the critical current between 102 A and
104 A leads to conservative conductor designs as follows.
The ranges of the conductor diameters and the copper-tosuperconductor ratios are between 1 mm and 21 mm and
between 7 and 35, respectively. The design constraint assumes = 1. These rather large matrix-to-superconductor
ratios can be reduced by further reducing the operating
current (the standard operation is at i = 0.75 0.2). A more
elaborate theory, based on the same approach, takes into
account thermal resistance between the matrix and the
superconductor and their actual sizes. Still, the stability
criterion stays conservative, 1. It should be noted that
the transition from nucleate to lm boiling of the coolant
(liquid helium) may cause instability when > 1 if the
current sharing is not complete.
MaddockJamesNorris Equal-Area Theorem. The Stekly
cryostability criterion has been developed taking into account removal of the generated thermal energy from a conductor only by convection, that is, the conduction mechanism has been neglected. This is a consequence of the
assumption of the existence of a uniform temperature
throughout the conductor. This assumption, however, can
easily be violated, because the matrix material (usually
copper) is thermally conductive. Hence, a question may be
posed: How must the steady-state stability criterion be formulated if the conduction effect is to be taken into account?
An elegant approach to the solution of this problem was
given by Maddock et al. (29). We will summarize a onedimensional formulation as introduced by Maddock (29),
although a two-dimensional (or for that matter a threedimensional) formulation can be readily devised.
Let us assume that a composite conductor carrying constant current I and with one-dimensional geometry (i.e., a
long cylindrical rod with D/L < 1, D being its outer diameter and L its length) is submerged in a liquid helium pool.
Let a disturbance, caused by an instability, initiate the appearance of a normal zone in the superconducting part of
the conductor, thus leading to current sharing. Thermal
energy generation caused by Joule heating in the normal
current-carrying parts of the conductor will cause an increase of the conductor temperature. Eventually, when all
the current starts owing through the matrix, any further
increase in temperature will be accompanied by a constant
thermal energy generation G (the current intensity is constant). Due to the conductor surface temperature increase,
the boiling heat transfer mode of liquid helium may change
from nucleate (at an initially small temperature differences) to lm boiling (at more pronounced temperature differences). The boiling characteristic (i.e., the curve of heat
Figure 6. An equal-area theorem: boiling characteristic curve

(QV ) and thermal energy generation curve (GV ). The area between
the QV and GV curves on the segment between the points S1 and
U represents an excess of heat transfer rate removed by boiling at
the cold end. The area between the same curves in the zone US2
represents an excess of thermal energy generated at the hot end.
Equality of the two energy rates (equality of the corresponding areas) corresponds to a balanced condition QV = GV over the entire
conductor.
transfer rate density versus temperature difference) will

determine the rate of heat removal. In Fig. 6, both thermal
energy generation and heat transfer rate are presented as
functions of temperature difference between the conductor
surface and the liquid helium pool (note that Tb remains
unchanged; consequently the axis in Fig. 6 may be considered as the conductor surface temperature with Tb as a
xed parameter).
The area enclosed between the thermal energy generation curve and the boiling (heat transfer) curve over the
segment S1 U is proportional to the excess thermal energy
removed from the conductor. Similarly, the area enclosed
over the segment US2 between the thermal energy generation curve and the corresponding lm boiling segment of
the boiling characteristic curve is proportional to the excess
of thermal energy generated within the conductor over the
heat transfer rate density removed by boiling. Because of
the temperature difference between the hot and cold ends,
heat conduction will take place along the conductor, leading to a removal of the excess thermal energy from the hot
end by the excess of boiling heat transfer from the cold
end. A stable equilibrium of this heat transfer process will
be reached if the two excess thermal energy rates (represented by the shaded areas between the curves in Fig.
6) are exactly the same. This straightforward phenomenological description leads directly to the required stability
criterion: Equality of thermal energy rates, represented by
equal areas in Fig. 6, corresponds to a limiting case of a
cryogenic stability. If the cold-end thermal energy rate is
larger than the hot-end thermal energy rate, the conductor
temperature will ultimately return to its initial temperature, equal to the liquid bath temperature, thus leading to
the disappearance of the normal zone and a restoration of
the superconducting mode of the conductor. The limiting
condition can be interpreted as an equality of the areas in
Fig. 6 with, say, conductor temperature as abscissa. This
conclusion constitutes the so-called equal-area theorem introduced by Maddock et al. (29).
Formal mathematical proof of the above described theorem is straightforward. The general energy balance equation (with the uniform temperature assumption relaxed)
10
reads as follows:
In an analytical form for a one-dimensional heat transfer

problem this balance reads as follows:
where A (m2 ) is the area of the conductor cross section, and

k in (W/mK) is the thermal conductivity of the matrix. The
bracketed term on the left-hand side and the two terms on
the right hand side are all functions of temperature; thus,
in a compact form,
where F, QV , and GV are the conductive heat ux, heat

transfer rate per unit volume of the conductor, and thermal energy generation per unit volume of the conductor,
all exclusive functions of temperature. After the formal integration of Eq. (17) one obtains
where we assume that the integration boundaries (T1 and

T2 ) are well apart so as to reach the zones in the conductor
where the temperature gradients are equal to zero. Making the reasonable assumption that thermal conductivity
k(T) depends linearly on temperature in the given temperature range [k(T) = cT], a transformation of the dependent
variable in Eq. (18), t = cT2 /2, will lead to
A geometric interpretation of the Eq. (19) is straightforward. If one replaces the abscissa in Fig. 6 using the same
transformation that led to Eq. (19), then Eq. (19) represents the equality of the areas presented in Fig. 6. If this
condition is satised (i.e., the energy balance is preserved),
any excess of thermal energy generated within the hot zone
of a superconductor will be compensated by the excess of
the heat transfer rate removed from the conductor in the
cold zone thanks to the conduction between the two parts
of the conductor.
Wipf Minimum Propagation Zone. The concept of the
minimum propagation zone (MPZ) was introduced by Martinelli and Wipf (30). It has since been developed into a
comprehensive stabilization theory (17). The original approach was based on a simple energy balance of a superconducting material idealized to be innite in space, but
with a localized thermal energy generation source within
a preexisting normal zone of nite size. This balance, in a
generalized form, is as follows:
Hence, the approach assumed a balance between thermal

energy generated by Joule heating (within the already established normal zone) and the rate of heat transfer from
the normal zone by conduction through the superconductor.
Therefore, the physical size of the normal zone that evolves
into a quench has been assumed to depend on a tradeoff
between Joule heating in the normal zone and the heat
carried out of the normal zone by conduction. As a consequence, assuming the validity of the energy balance condition described above, the normal zone will neither grow nor
collapse. The normal zone dened in such a way is called
the MPZ (17).
The simplest realistic situation is the one that leads to a
conservative stabilization limit, which requires the consideration of non-current-sharing conditions, and additional
convective cooling. Let us assume that a linear superconductor, with an already developed normal zone of length
2X, is cooled by (1) a coolant at temperature Tb , and (2)
conduction in the axial directions (Fig. 7). In such a situation, the energy balance has an additional term [see Eq.
(16)]. The conductor has a circular cross section (radius r),
with a constant electrical resistivity (in the normal zone)
and uniform thermal conductivity k (in both normal and
superconducting parts). Under these conditions, the differential energy balance per unit length of the conductor is as
follows (17):
with the following boundary conditions:

1. Superconducting zone [note that in this zone the second term on the right-hand side in Eq. (21) vanishes]:
2. Normal zone (with extra heating qtr at x = 0)
Note also that at the boundaries between the localized normal zone and both superconducting zones, the temperature gradients must be equal. Hence, Eqs. (21)(23) (and
also an additional equation representing the equality of the
temperature gradients at the junctions between the zones)
11
Figure 7. A conductor with a formed normal zone ( = 0) of total

length 2X. Outside the normal zone, the conductor is superconducting ( = 0). The conductor is exposed to convective cooling,
and to thermal energy generation within the normal zone. The
heat conduction through the conductor removes heat from the
normal zone.
dene the mathematical model of the temperature distribution in the conductor. The closed-form solution can be
readily obtained (17). Of particular interest is an analytical relationship between the heat ux caused by heating,
qtr [dened by Eq. (23)] and the length of the normal zone.
This relationship can be obtained from the rst of the two
equations in Eq. (23) by differentiating the temperature
distribution inside the normal zone, and by subsequently
applying the result at the normal zone boundary. The nal
result is as follows:
where is a stability parameter dened in the same manner as the one introduced by Eq. (14). The relationship
given by Eq. (24) is presented graphically in Fig. 8. The
abscissa in Fig. 8 represents the dimensionless length of
the normal zone, l = 2X/(rk/2h)1/2 . Note that the length of
the MPZ is equal to twice the normal-zone half length X
for the critical magnitude dened by Eq. (24), that is, for
qtr = 0:
In Fig. 8, three distinct regions can be identied. The

rst region corresponds to the pairs of values of the dimensionless heating and the dimensionless length of the
normal zone, l, such that < 0.5. The second region corresponds to the stability-parameter range 0.5 < < 1, and
the third region to > 1. In the rst region, say for =
0.49, the superconductor is in a least stable situation (15).
In this region a MPZ exists for any 0 < < 0.5. At = 0.5
the MPZ is theoretically innitely large; see Eq. (25). The
second region (0.5 < < 1) is characterized by increasing
stability. It should be noted that any = const curve in
this region has a minimum. Hence, the left-hand branches
of these lines correspond to an unstable equilibrium (between the l = 0 value and l at min , for any ). The locus
of these minima denes the bifurcation line between the
unstable and stable equilibrium. The maximum size of recoverable transition (MSRT) line is dened by X/(rk/2h)1/2
= ln(2 1). Finally, the third region ( > 1) corresponds
to fully stable (i.e., cryostable) conditions. In that region
any created normal zone is in stable equilibrium. The size
of the normal zone increases monotonically with increase
of the heat release per unit conductor volume. Note that

1 corresponds to the condition imposed by the conservative
Stekly criterion.
The Wipf approach has been rened even further for periodic temperature distributions and for a current-sharing
situation (1731).
It is instructive to provide a simplied representation
of the main features of all three stabilization criteria discussed so far (Stekly, Maddock, and Wipf). A diagram of
reduced cooling rate versus reduced temperature may be
used (13): see Fig. 9. There the reduced cooling rate is assumed to be a linear function of reduced temperature. The
non-current-sharing case will be considered. The following
situations may be distinguished. For the reduced heat generation, the one described by curve a, the superconductor is
in a fully stable mode (a cryostable case). The limit of this
type of behavior is at the condition described by the Stekly
criterion (the reduced heat generation denoted by curve b).
If current further increases (represented by an increase in
reduced current i in Fig. 9), the superconductor reaches
the region of MRZ stability, as described by Wipf (curve
c). Further increase in current may eventually provide the
condition that corresponds to the Maddock equal-area stability criterion (curve d). Note that for curve d, the areas
between the curves representing the reduced cooling rate
and reduced heat generation are equal. If current increases
even more, the MPZ stability zone will be reached (curve
e).
ADVANCED STABILITY AND THERMAL MANAGEMENT

PROBLEMS FOR SUPERCONDUCTORS
The conventional theory of cryogenic stability of both lowtemperature and high-temperature superconductors has
not been able to address a number of issues related to the
design of modern superconducting devices. Without trying
to provide a comprehensive review (for the details consult
the bibliography contained in Ref. 27), let us discuss briey
two characteristic topics.
As commented on in the introductory part of this article, instability phenomena are inherently transient. Consequently, the quench and recovery are also inherently transient. The existence of a normal zone, as introduced rst
by Wipf and others, should be considered in the light of
the propagation velocity. This velocity is positive during
quench and negative during recovery. Under the equilib-
12
Figure 8. Dimensionless extra heating versus dimensionless

length of the normal zone. No current sharing case. The region
for 1 corresponds to a fully stable conductor (cryostable region). The region for 1 > > 0.5 respresents the so-called minimum recovery zone (MRZ) of stability. The length of the MRZ is
the length of a superconducting zone in equilibrium with neighboring normal zones of innite length. A longer superconducting zone will spread into the normal zones; a shorter one will
shrink. The region for 0.5 corresponds to the so-called minimum propagation zone (MPZ) of stability. The MSRT line is the
line of maximum size of recoverable transition. The locus of minima in the MRZ region provides a boundary between stable and
unstable equilibrium (17).
rium conditions it is equal to zero. A good review of the

early attempts to determine this velocity is provided in
Ref. 27. A modern treatment of the related problems (such
as the existence of the so-called traveling normal zones
(TNZs), the propagation velocities in the uncooled superconductors, the inuence of nonlinearities introduced by
temperature-dependent material properties, and the issues related to thermal and hydrodynamic management of
internally cooled superconductors) is reviewed in Ref. 31.

It should be added that the stability conditions dened by
the conventional theory based on the presence of a continuous disturbance does not reect the proper conditions for
stability against transient heat pulses. So transient stabilization must be considered (27). An important part of
these studies is an appreciation of transient heat transfer
phenomena. For example (31), for very short transient phenomena (of the order of magnitude of tenths of a millisecond), the heat transfer coefcient may be in the range of
Figure 9. Reduced cooling rate, q = QA /[h(Tc,0 Tb )], and reduced thermal generation, g = GA /[h(Tc,0 Tb )], versus reduced
temperature, = (TI Tb )/(Tc,0 Tb ). Curves 01Aa, 01Bb,
01Cc, 01Dd, and 01Ee, (each denoted by a dotted line)
correspond to various values of the reduced thermal generation g,
each with different but xed current. The reduced cooling curve
is the same for all these cases. In a fully cryostable region (curve
01Aa) the corresponding current causes a Joule heating lower
than the cooling rate for any reached temperature. The stability
conditions for the curve 01Bb satisfy the Steckly criterion; see
the curve for = 1.0 in Fig. 8. The stability conditions for the curve
01Dd satisfy the Maddock criterion (the shaded areas both below and above the cooling curve are equal to each other); see the
curve for = 0.5 in Fig. 8. This representation is valid in the absence of current sharing (compare with Fig. 8).
0.5 W/cm2 K to 1.5 W/cm2 K (the propagation velocity was

between 5 m/s and 20 m/s). These transient heat transfer
coefcients are larger by an order of magnitude than the
lm boiling heat transfer coefcient.
The introduction of HTSC materials brought an additional appreciation of the need to include in the analysis the
temperature-dependent thermophysical properties and kinetic heat transfer characteristics over a much broader
temperature range. A direct consequence of this fact in
any thermal management problem for a superconductor
(not necessarily related to cryostability) is a requirement
to model the problem with its increasingly nonlinear character. Under these circumstances, the originally introduced
assumptions must be revisited (32).
THERMOPHYSICAL AND HEAT DATA ON CRYOGENS
Although the HTSCs can exhibit some superconducting
properties even at temperatures approaching ambient,
superconductivity applications under practical conditions
still depend upon cooling to cryogenic temperatures. This
fact does not diminish the value of the HTSCs, because
of the greater ease of operation and savings in refrigera-
13
tion power that are effected by operating at temperatures

nearer to that of liquid nitrogen than to that of liquid helium. The theoretical power input required to produce one
unit of refrigeration at cryogenic temperatures is given as
(Ta T)/T, where Ta is ambient temperature, or the temperature at which the heat must be rejected, and T is the
refrigeration temperature, the temperature at which the
heat is to be removed. For an ideal cryogenic refrigerator to
remove heat at the temperature of liquid helium (4 K) and
reject it at ambient temperature (300 K) would requires
74 W of input power per watt of refrigeration. At liquid
nitrogen temperature (77 K) this ratio is reduced to a little less than 3 W power input per watt of refrigeration. As
Strobridge (33) has shown, cryogenic refrigerators to date
do not approach this theoretical limit very closely. The degree to which this limit is approached is not dependent
upon the temperature level of refrigeration, but depends
strongly upon the capacity of the unit. The larger the refrigerator, the better is the degree of approach, with the
best (largest) units reaching only 35% to 40% of the limit
(33).
Table 2 lists some of the thermophysical properties of
cryogens (34) that might be used in the cooling of a superconducting system.
If cooling is to be done by pool boiling of a liquid, the
range of temperature that is available is of interest. For a
given cryogen, the boiling temperature can be xed at any
temperature between the the triple point and the critical
point by maintaining the corresponding system pressure.
From Table 2 it can be seen that over the range from below
4.15 K to over 150 K, there are only two gaps; from 5.25 K
to 13.8 K, and from 44.4 K to 54.4 K.
Also, the application of supercritical pressure can allow
forced-ow cooling without encountering two-phase ow
with its consequent complications of pressure and ow oscillations and excess pressure drop in the ow channels.
However, only in the case of liquid helium can this be accomplished at a pressure as low as 0.23 MPa.
The main heat transfer phenomena involved with cooling of superconducting magnets are as follows: (1) forced
convection of a single-phase coolant (say, gaseous or supercritical helium), (2) phase-change heat transfer (nucleate and lm boiling) in both pool boiling and channel ow
conditions, and (3) heat transfer in superuid helium. The
most important mode of heat transfer for cryogenic stabilization is boiling (both nucleate and lm boiling). In Table
3 typical data for (1) the critical heat ux (i.e., the maximum heat ux for nucleate boiling at the given temperature difference) and (2) the Leidenfrost-point heat ux
(i.e., the minimum heat ux for lm boiling) are given. It
should be noted that the available heat transfer data scatter widely and depend greatly on heat transfer surface orientation, geometry, and surface conditions. Transient heat
ux data for nucleate boiling of cryogens are greater by an
order of magnitude than steady-state values.
SAFETY
In working with any cryogenic uid, safe operation requires that there be a satisfactory understanding of the
14
hazards that can arise and also a strict compliance with

safe operating principles for these uids. These hazards
can stem from the temperature of the uid, or from its reaction with other materials that come in contact with it.
The low temperatures can embrittle some structural materials, or can cause freezing of human tissue if personnel
should contact the cold uid or the exterior of a cold, uninsulated pipe. The low temperatures also cause a signicant
thermal contraction which, if not sufciently compensated,
can give rise to high stresses that can cause their own hazard.
Any of the cryogens that have boiling points below that
of liquid oxygen can condense the atmospheric air, resulting in a condensate that is enriched in oxygen (as much
as 50% oxygen) which is even more hazardous than liquid air. If this condensed air is allowed to fall on sensitive
equipment, some materials can be sufciently embrittled
to crack, other equipment may no longer function properly,
and, if the enriched liquid air should fall on a combustible
material (such as asphalt), an explosion can result. In an
improperly purged system, condensation of air, or its constituents such as water vapor or carbon dioxide, can result
in the obstruction of pressure relief passages or the connections to instrumentation needed for the safe operation
of the system.
The use of liquid hydrogen or liquid oxygen is less probable in the application of superconductivity. However, such
an application is not impossible. Use of these uids entails
the additional hazard of combustion, or even explosion, if
strict safety measures are not observed.
The most likely safety hazard to be encountered in applications of superconductivity is the excessive buildup of
pressure within a cryogenic system. If a cryogenic uid is
totally conned, as in a pipe between two closed valves,
the pipe must try to maintain the density of the liquid as
the contained uid becomes a gas that approaches ambient
temperature. Because of the increase in the compressibility
factor of the gas at high pressure, the pressure obtained is
higher than what would be computed by applying the ideal
gas law to the appropriate density ratios. Consequently,
helium trapped in this fashion could reach a pressure of
103 MPa (15,000 psi), and nitrogen could reach a pressure
of 296 MPa (43,000 psi) if the container did not rupture
beforehand (35). In the case of a quench of a superconducting magnet there is a very rapid release of energy that
will quickly enter the cryogenic coolant. If sufcient venting capacity is not provided, a rapid, hazardous buildup
of pressure in the coolant passages or the container can
result.
ADDITIONAL READING
Several references mentioned in the bibliography deserve
additional attention as useful sources for further reading.
The book of Wilson (14) is a classical text and must be read
as an upper-level introduction to the problems of design of
superconducting magnets. A two-volume book by Collings
(27) provides extensive insight into many highly technical aspects of metallurgy and physics of low-temperature
superconductors, as well as a very comprehensive bibliog-
raphy. The book by Dresner (31) is an advanced text that

provides an insight into the modern treatment of the stability of superconducting devices. For technical calculations
related to design of cryogenic devices, including cryostability aspects, the book by Iwasa (28) will be very useful. A
number of useful sources can be found that deal with cryogenic and heat transfer aspects of the design of cryogenic
devices. In addition to those mentioned in the bibliography, the following one may be consulted: S. W. van Sciver,
Helium Cryogenics, New York: Plenum, 1986.
BIBLIOGRAPHY
1. H. K. Onnes Commun. Phys. Lab. Univ. Leiden, Nos. 120b,
122b, 124c, 1911; see also K. Gavroglu and Y. Goudaroulis
(eds.), Heike Kamerlingh Onnes: Through Measurement to
Knowledge, The Selected Papers, Norwell, MA: Kluwer, 1990.
2. K. D. Timmerhaus T. M. Flynn Cryogenic Process Engineering,
New York: Plenum, 1989.
3. R. V. Carlson Applications of superconductivity, in K. D.
Williamson, Jr., and F. J. Edeskuty (eds.), Liquid Cryogens,
vol. II, Boca Raton, FL: CRC Press, 1983, chap. 4.
4. D. C. Larbalestier M. P. Maley Conductors from superconductors: Conventional low-temperature and new hightemperature superconducting conductors, Mater. Res. Soc.
Bull., 18: 5056, 1993.
5. K. Mendelssohn Superconductivity and superconducting devices, in C. A. Bailey (ed.), Advanced Cryogenics, London:
Plenum, 1971, chap. 10.
6. D. S. Beard et al. The IEA large coil task, Fusion Eng. Des., 7:
1232, 1988.
7. V. A. Altov et al. Stabilization of Superconducting Magnetic
Systems, New York: Plenum, 1977.
8. F. J. Edeskuty Cryogenics, in E. A. Avallone and T. Baumeister III (eds.), Marks Standard Handbook for Mechanical Engineers, 10th ed., New York: McGraw-Hill, 1996, chap. 19.2.
9. M. P. Maley personal communication, 1998.
10. X. D. Wu et al. Properties of YBa2 Cu3 O7 thick lms on
exible buffered metallic substrates, Appl. Phys. Lett., 67:
23972399, 1995.
11. P. Chowdhuri F. J. Edeskuty Bulk power transmission by
superconducting dc cable, Electr. Power Syst. Res., 1: 4149,
1977/78.
12. E. B. Forsyth The high voltage design of superconducting
power transmission systems, IEEE Electr. Insulation Mag., 6:
716, 1990.
13. R. J. Thome Magnet program overview for the international thermonuclear test reactor, IEEE Trans. Magn., 30:
15951601, 1994.
14. M. N. Wilson Superconducting Magnets, Oxford: Claredon
Press, 1983.
15. H. Wada et al. VII-1: Appendixterminology for superconducting materials, Cryogenics, VAMAS Suppl., 35: S113S126,
1995.
16. A. A. Abrikosov On the magnetic properties of superconductors
of the second group, Sov. Phys.JETP, 5: 11741182, 1957,
transl. of Zh. Exp. Teor. Fiz., 32: 14421452, 1957.
17. S. L. Wipf Stability and degradation of supeconducting
current-carrying devices, Tech. Rep. LA-7275, Los Alamos Scientic Laboratory, Dec. 1978.

18. M. N. Wilson Stabilization of superconductors for use in magnets, IEEE Trans. Magn. MAG-13: 440446, 1977.
19. O. Tsukamoto T. Takao S. Honjo Stability analysis of superconducting magnet: An approach to quantication of energy disturbance caused conductor motion, Cryogenics, 29: 616620,
1989.
20. R. G. Mints A. L. Rakhmanov Instabilities in Superconductors
(in Russian), Moscow: Nauka, 1984.
21. Y. Iwasa et al. Experimental and theoretical investigation
of mechanical disturbances in epoxy-impregnated superconducting coils, part 1: General introduction, Cryogenics, 25:
304306; part 2: Shear stress-induced epoxy fracture as
the principal source of premature quenches and training
theoretical analysis, 307316; part 3: Fracture induced premature quenches, 317322; part 4: Pre-quench cracks and frictional motion, 323336, 1985.
22. V. I. Dotsenko et al. Experimental study of thermomechanical instabilities in superconducting composites, part 1: Magnetic ux jumps induced by plastic deformation, Cryogenics,
29: 1015; part 2: Dynamics of normal zones, 1621, 1989.
23. D. T. Reid et al. Denitions of terms for practical superconductors. 3. Fabrication, stabilization and transient losses. Cryogenics, 19: 327332, 1979.
24. A. R. Kantorowitz Z. J. J. Stekly A new principle for the construction of stabilized superconducting coils, Appl. Phys. Lett.,
6 (3): 5657, 1965.
25. Z. J. J. Stekly J. L. Zar Stable superconducting coils, IEEE
Trans. Nucl. Sci., 12: 367372, 1965.
26. Z. J. J. Stekly R. Thome B. Strauss Principles of stability in cooled superconducting magnets, J. Appl. Phys., 40:
22382245, 1969.
27. E. W. Collings Applied Superconductivity, Metallurgy, and
Physics of Titanium Alloys, New York: Plenum,1986.
28. Y. Iwasa Case Studies in Superconducting Magnets. Design
and Operational Issues, New York: Plenum, 1994.
29. B. J. Maddock G. B. James W. T. Norris Superconductive composites: Heat transfer and steady state stabilization, Cryogenics, 9: 261273, 1969.
30. A. P. Martinelli S. L. Wipf Investigation of cryogenic stability and reliability of operation of Nb3 Sn coils in helium gas
environment, in H. M. Long and W. F. Gauster (eds.), Proc.
1972 Appl. Superconductivity Conf., IEEE Publ. 72CH06825-TABSC, 1972, pp. 331340.
31. L. Dresner Stability of Superconductors, New York:
Plenum,1995.
32. D. P. Sekulic F. J. Edeskuty Z. Uzelac Heat transfer through a
high temperature superconducting current lead at cryogenic
temperatures, Int. J. Heat Mass Transfer, 40 (16): 39173926,
1997.
33. T. R. Strobridge Cryogenic Refrigeratorsan Updated Survey,
NBS Technical Note 655, Boulder, CO: US Department of Commerce, National Bureau of Standards, 1974.
34. R. L. Mills F. J. Edeskuty Cryogens and their properties, in K.
D. Williamson, Jr., and F. J. Edeskuty (eds.), Liquid Cryogens,
vol. II, Boca Raton, FL: CRC Press, 1983, chap. 1.
35. F. J. Edeskuty W. F. Stewart Safety in the Handling of Cryogenic Fluids, New York: Plenum, 1996.
D. P. SEKULIC
F. J. EDESKUTY
University of Kentucky,
Lexington, KY
Edeskuty Engineering, Los

Alamos, NM
15
16

Copyright
SUPERCONDUCTORS, FORCED FLOW CONDUCTOR MANUFACTURING

Electric windings that use low temperature superconductors need to be actively cooled. An easy and effective
way to maintain the winding at operating temperature is to fill the space around the coil with liquid helium
(pool-cooled magnets), allowing the liquid to penetrate between the turns. Alternatively, the superconductor
may include a cooling path where the coolant is forced to circulate under pressure. This article is a review of
the manufacturing methods for forced flow conductors, starting from the early prototypes (30 years ago) to the
very large cable-in-conduit conductors of the fusion magnets. The design aspects of forced flow conductors are
not deeply discussed here, although they have sometimes driven the effort toward advanced manufacturing
technologies. Essential references on the design aspects of forced flow conductors can be found in the articles
Superconductors, STABILITY IN FORCED FLOW CONDUCTORS; HYSTERESIS AND COUPLING LOSSES IN SUPERCONDUCTORS; AND SUPERCONDUCTING MAGNETS, QUENCH PROTECTION.
The manufacturing methods for forced flow superconductors sound like a brand new subject. Actually, it
is old enough to have a history. The initial proposal dates back to 1965, when 1 suggested using the circulation
of supercritical helium for reliable cooldown and heat transfer in large superconducting magnets. The first
milestone is the spark chamber detector built in 1970 at CERN (2), with 50 MJ stored energy and 24 tonnes of
NbTi hollow conductor.
Besides the clear advantage in thermal behavior, a key argument in favor of forced flow conductors is
electromechanical integrity. In contrast with pool-cooled windings, where the individual turns are insulated
only by spacers to allow the coolant to wet the metallic surface, the turns of a winding made of forced flow
conductor can be fully insulated and potted. Such a rigid structure offers superior mechanical performance
under electromagnetic load and withstands very high operating voltages (e.g., in pulsed or quench mode),
not limited by the dielectric strength of the coolant. Clearly, the choice of forced flow conductors becomes
mandatory for magnets with very large stored energy, subjected to high operating loads and requiring highvoltage discharge in case of quench.
Comparing the latest forced flow conductors with the early ones, manufactured thirty years ago, the
difference is impressive. On one side, the conductor design has made substantial progress, using the results of
extensive R&D work (mostly on ac loss and stability) and sophisticated thermohydraulic codes. On the other,
the manufacturing methods have evolved under the pressure of the design goals, and advanced technologies
have become reliable and affordable.
The present review is focused on the manufacturing aspects. In general, only practical industrial manufacturing experience is quoted; short prototype conductors are reported only if an original method was used.
The review is organized in three main sections. The first describes the conductor generation where the helium
flows in smooth pipes and the superconductor is attached outside the pipes, without direct contact with the
coolant. The second generation is a transition group of conductors, where a flat cable is encased in a welded
steel sheath. The third generation includes the rope-in-pipe, or cable-in-conduit, conductors, where the coolant
is forced to flow through a bundle of strands with large void fraction, encased in a structural conduit.
Fig. 1. Three extruded hollow superconductors: (a) Al matrix with NbZr filaments (1967, courtesy of CERN), (b) Cu matrix
with NbTi filaments (IMI, 1968), (c) round in square Cu matrix with thick NbTi filaments (MFO, later ABB, about 1968).
Superconducting cables for power transmission lines are not in the scope of this review. Their manufacturing methods, dominated by the high-voltage insulation requirements, are similar to those for conventional
high-voltage conductors for underground lines.
Conductors with Smooth Pipes

In the late sixties, the use of a helium-tight conduit produced by extended welded seams was considered too
risky for large-scale applications (particle detectors, fusion magnets), where a single leak in the winding would
have either severely delayed or compromised the entire project. It is understandable that copper pipes were
initially selected for forced flow conductors, because of their high thermal conductivity, small pressure drop,
and reliable leak-tightness.
Four conductor subgroups are identified according to the assembly procedure for attaching the superconductor to the copper pipes.
Superconducting Pipes. The first experiments with forced flow conductors look today like small
masterpieces of technology. The idea was not to attach a superconducting wire to a pipe, but to create a hole in
a superconducting composite.
A cross section of the first ever reported hollow superconductor is shown in Fig. 1(a). The extruded
composite has a pure aluminum matrix with six NbZr filaments (0.25 mm thick) in the midplane, to allow
bending radii down to 10 mm. The copper-plated NbZr filaments were supplied by Supercon (United States),
and the hollow extrusion was done in 1967 at the Atelier e lectromechanique de Gascogne (France). Over 300
m of conductor have been produced and successfully wound into a solenoid at CERN (3).
The hollow conductor shown in Fig. 1(b) was commercially produced by Imperial Metal Industries (England) at the end of the sixties. The rectangular, extruded pipe, 6.5 5 mm, is made of copper, with 14 NbTi
filaments (0.45 mm diam) embedded in it. The production lengths ranged up to 200 m. At CERN, the conductor
sections were electron beam welded (4) and wound into a solenoid of 32 double pancakes, for a total of 2500 m
of conductor. At Saclay, France, a smaller, racetrack coil (5) was wound with 300 m of the same conductor.
At M.F.O. (Switzerland), later BBC and ABB, a larger, square superconducting pipe was developed (6) by
hollow extrusion of a copper billet with 16 NbTi filament ( 0.55 mm diameter); see Fig. 1(c). Because of the
large cross section, the conductor unit length was limited to 100 m.
Fig. 2. A hollow superconductor (7), manufacture in 1974 by a tube mill method (courtesy of K. Agatsuma, ETL, Japan).
The manufacture of a hollow superconductor at the Electro-Technical Laboratory, Tsukuba, Japan (ETL)
was reported in 1974 (7). The method is shown schematically in Fig. 2(a). Twelve square rods of superconducting
composite (6 6 mm), each with 19 NbTi filaments in a copper matrix, were inserted in a rectangular copper
tube and rolled with intermediate annealing to a flat strip, 1.6 mm thick. The strip was then welded in a
conventional tube mill and drawn down to form a rectangular hollow conductor, 4 5 mm; see Fig. 2(b). Due to
the rolling process, the 116 filaments, with equivalent diameter 0.25 mm, are flattened and nonhomogeneously
spaced.
The major drawbacks of such superconducting pipes are the large size (leading to poor critical current
density) of the superconducting filaments due to the limited reduction ratio, and their nontransposition with
respect to the transverse field, due to the lack of twisting. As a result, large flux jumps were observed in
operation.
Conductors Assembled by Electroplating. An original method was developed at the Kurchatov
Institute (Moscow) in the late sixties to bond together the superconducting strands into large composites by
continuous electroplating in a CuSO4 electrolyte (8). The same method was later used to assemble the force
flow conductors for the toroidal coils of the Tokamak-7 (T-7) and Tokamak-15 (T-15).
The T-7 conductor (9) is made from a linear array of nine copper pipes, 2 mm inner diameter, with 16
multifilamentary NbTi strands sitting in the grooves between the pipes; see Fig. 3(a). Another 32 thin, single
core NbTi wires are placed in the grooves between the multifilamentary strands and the copper pipes to obtain
an even surface. The pipestrands assembly is bonded by electroplating a 0.6 mm copper layer up to the final
size of 28 4.5 mm. The strands are not transposed, and the conductor has suffered from severe flux jumps,
triggering quenches during ramp up and down. The large number of parallel cooling channels, supplied by two-
Fig. 3. Cross section of the conductors assembled by copper electroplating at Kurchatov Institute: (a) the NbTi-based T-7;
(b) the react-and-wind Nb3 Sn conductor for the T-15. (Courtesy of Kurchatov and Bochvar Institutes, Moscow.)
phase helium, results in a highly nonhomogenous distribution of the mass flow rate. Over 10 km of conductor
have been manufactured in units longer than 200 m.
In the conductor for the T-15 (10), a flat cable of 11 nonstabilized Nb3 Sn strands (1.5 mm diam) around a
bronze strip was bonded after heat treatment to two copper pipes (3 mm inner diameter) by an electroplated
Cu layer, 1.2 mm thick; see Fig. 3(b). Some copper wires were included between the cable and the pipes to
obtain a more regular envelope. To minimize the bending strain during winding, the heat-treated cable is close
to the neutral bending axis of the conductor. In the winding of the T-15 coils, a wet insulation method had to be
applied to smooth the uneven conductor surface. Over 100 km of conductor have been manufactured, in units
of 200 m.
After the T-7 and T-15 manufacturing experience, the technique of assembly by electroplating was not
applied any more. The low process speed, the large electric power requirement, and the poor dimensional
tolerance overcame the advantage of a good, low-resistance bonding of the components.
Strands Soldered on a Central Copper Pipe.

The OMEGA Conductor. The conductor for the OMEGA detector at CERN is a historical milestone
in the development of forced flow superconductors (11). The development started in 1968, with a prototype
consisting of a round-in-square copper pipe with eight multifilamentary strips located in longitudinal slots.
The 40 40 mm assembly was fed into an outer copper pipe and compacted by drawing to the final size of 18
Fig. 4. OMEGA conductor: (a) prototype (left) and (b) final soldered cable. Courtesy of ABB.
18 mm [Fig. 4(a)]. This first approach was abandoned because of the high cost, the limited manufacturing
length (40 m), and the lack of transposition.
The final cross section is shown in Fig. 4(b). Manufactured in 1970 at BBC (Switzerland), the 18 18 mm
conductor is a square copper pipe with residual resistance ratio (RRR) >250, surrounded by two layers of wires
(1.5 mm diam) cabled in opposite directions. In the first layer, 30 out of 36 wires are superconducting, with
four NbTi thick filaments (250 m) in a low RRR copper matrix. The strands are initially not twisted, but, as
they are cabled on the copper pipe, no back rotation of the cage strander is allowed, so that the filaments in the
strand have the same twist pitch as the cable (200 mm). The second layer, with protective function, contains
40 copper wires and is cabled in the opposite direction, with pitch 180 mm.
The cabled conductor was driven into a SnAg5 bath and pulled through a square die, with a die angle of
45 , cooled at constant temperature to solidify the solder; see Fig. 5. Twelve sections, about 1 km each, were
produced for a total of 24 tonnes (11.4 km). The soldering process for each section took about 35 h, at a speed
of 0.45 m/min. The online quality assurance included eddy currents and ultrasonics to check the quality of the
solder impregnation.
The OMEGAs Brothers. The smallest brother of the OMEGA is a rectangular conductor developed
in the early seventies at ETL, Japan (7), consisting of a copper pipe with one layer of 52 wires (0.32 mm
diam) cabled around and impregnated with SnAg solder. Half of the wires are multifilamentary NbTi strands,
alternated to copper wires. In contrast with the OMEGA, the wires are cabled and soldered as a round conductor
and later drawn down to the final rectangular size of 5 4 mm; seeFig. 6(a)
In the mid seventies, a conductor of identical size to the OMEGA (18 18 mm) was manufactured at
CERN for a large superconducting dipole (12). The copper pipe, with 10 mm bore, is round, and only one layer
of NbTi multifilamentary strands and copper wires is cabled on it. The cable is encased between two copper
profiles and bonded by SnAg soldering; see Fig. 6(b). The unit length is up to 270 m, and the overall production
is more than 5 km.
In the late seventies, a hollow conductor [see Fig. 6(c).] was manufactured at VAC, Germany (13). A loose
layer of 20 NbTi multifilamentary strands and 10 copper wires were cabled and soldered in the same process
Fig. 5. Soldering of the OMEGA conductor (courtesy of ABB).
Fig. 6. Soldered, hollow conductors, manufactured similarly to the OMEGA conductor. See text for details.
around a square copper pipe. The solder is SnPb eutectic alloy, and the final size is 8.2 8.2 mm. A first
conductor batch had to be replaced, as it did not withstand the tight bending radius (over 4% bending strain).
About 500 m of conductor was used for the test facility SAFFO at ENEA (formerly CNEN), Italy.
A similar, bigger conductor [see Fig. 6(d)] was manufactured in 1981 at Europa Metalli, Italy (14). The
33 NbTi strands (1.3 mm diam) were first cabled on a round copper pipe and soldered by SnPb alloy. The
conductor was then shaped approximately to a rounded square, 14.6 14.6 mm. About 9 km of conductor have
been manufactured and wound into the outer module of the SULTAN test facility (Villigen, Switzerland).
Fig. 7. Nb3 Sn hollow conductor, with four braids soldered in the slots of the Cu pipe (16), according to the react, wind,
and solder process (courtesy of K. Agatsuma, ETL, Japan).
Another little brother of the OMEGA [see Fig. 6(e)] has been manufactured at BBC in 1981, for a Sector
Cyclotron, SuSe (15), in Munich, Germany. The rectangular copper pipe has a tiny hole, 1 2 mm. Two kinds
of NbTi strands, with 0.7 mm diameter and different Cu: non-Cu ratios, were cabled around the copper pipe
and soldered with SnAg alloy. The overall conductor size is 4.5 5.5 mm. About 4.5 km of conductor has
been wound into the winding of the sector cyclotron, with a minimum bending radius of 68 mm (3.3% bending
strain).
The Nb3 Sn Hollow Conductor at ETL, Japan. Apparently, the method of cabling and soldering the
superconducting strands on a copper pipe does not apply to Nb3 Sn conductors: soldering must be carried out
after heat treatment, but a heat-treated strand cannot be cabled because of its brittleness. Agatsuma et al.
(16) reported in 1978 an ingenious method to prepare a Nb3 Sn conductor soldered on a copper pipe, according
to the steps in Fig. 7. called the react, wind, and solder method.
Initially, two round, hollow braids (12 and 24 strands) are made from thin, Sn-plated Nbbronze composites (0.1 mm diam). The hollow braids are flattened to a 0.6 mm thick ribbon, wound on a large-diameter
holder, and heat treated to form the Nb3 Sn composite. In a single process, two small and two large braids are
driven, together with a SnAg solder foil, into longitudinal slots on a rectangular copper pipe. The conductor,
wrapped with a polyimide film and a reinforcing steel tape, is wound into a coil with 75 mm minimum bending
radius. After winding, the coil is heated in vacuum to melt the solder and bond the braids to the copper pipe;
see Fig. 7.
If the conductor were a solid monolith during the winding process, the bending strain at the Nb3 Sn
filaments would be in excess of 3%, which would be fatal for the conductor (to avoid irreversible damage in
Nb3 Sn, the bending strain should be below 0.5%). However, as the braids have not yet been bonded to the
copper pipe, the strands can settle in the slots and the actual bending strain is much smaller. A length of 160 m
conductor was used for the test coil, which was successfully tested, demonstrating the viability of the process.
The strands in the braids are transposed, but not the four braids placed in the longitudinal slots. To
obtain a high-current conductor, a very large number of thin strands must be braided. The conductor layout,
with thin braids placed on the perimeter of the pipe, does not allow a large operating current density. Since the
successful demonstration, no further application of the process has been reported.
The Swiss LCT Conductor. The most complex conductor belonging to the OMEGA family is the Swiss
LCT conductor, manufactured in 19811983 at BBC for the Swiss contribution to the Large Coil Task at Oak
Ridge. Initially, two forced flow conductor layouts were developed, one being a two-stage cable in conduit and
the other a multistage soldered cable around a copper pipe (17). The reasons to prefer the second option were
mostly the cost and the slowness of the first. Today, this choice may sound surprising and the same reasons
would rather lead us to prefer the cable-in-conduit option. However, in 1981 BBC had developed a broad
knowhow on soldered conductors, and almost no industrial experience was available on welding of long lengths
of steel conduit.
The cable layout is summarized in Fig. 8. The first cable stage consists of 10 multifilamentary NbTi
strands (0.46 mm diam) cabled around a copper wire and solder-impregnated. To limit the ac losses, the first
cable stage is tightly wrapped with a 0.05 mm thick CuNi strip and soldered again to bond the wrap to the
cable. To maintain full transposition, the second cable stage is made of six subcables around a thick copper core
(2.25 mm diam). Eight second cable stages are eventually cabled with the tight pitch of 150 mm around a round
copper pipe (7.5 4.5 mm), wrapped with CuNi foil. The cable is compacted and shaped by rolls to a square
of 18.5 18.5 mm. Despite the large load required to compact the rigid, solder-impregnated components down
to a local void fraction of 30%, neither damage nor performance degradation was observed in the compacted
cable.
For the final solder filling of the multistage cable, a special CdZnAg alloy was selected (18), with superior
mechanical properties to those of PbSn and SnAg alloys. As the melting point of the solder is about 350 C, the
process speed had to be maintained in the range of 2 m/min to avoid degradation of the NbTi critical current
density. Due to the high speed, the die needed to be specially laid out. To prevent solder oxidization and for
environmental reasons (Cd vapor), the process was carried out under nitrogen atmosphere. The overall length
produced for the Swiss LCT conductor was 5.5 km, in units of 250 m.
The Swiss LCT conductor marked the end of a successful series of hollow conductors cabled and soldered
around a copper pipe. As in the OMEGA and its brothers, a single-stage cable was used, so the achievable
current density was low, because only one layer of strands could be placed on the outer perimeter of the
pipe. The option to use a multistage cable around the pipe was viable, but turned out to have drawbacks: the
solder cross section climbed above 25%, and complex, high-resistivity barriers had to be included to avoid large
coupling-current losses.
Flat Cables Soldered to Copper Pipes. In Nb3 Sn react-and-wind conductors, the serconducting strands
must stay as close as possible to the neutral bending axis to avoid substantial strain-induced degradation during
winding. This requirement, together with the wish to maintain the advantages of the copper pipes, drove in
the eighties the development of a number of flat cables assembled by soldering to copper pipes. The ECN (The
Netherlands) and the SIN (later CRPP, Switzerland) carried out the conductor development and manufacture
as part of the construction of the SULTAN test facility.
The smallest of this conductor subgroup is a NbTi-based flat cable soldered to the broad side of a rectangular, half-hard copper pipe to build a square conductor, 8.4 8.4 mm; see Fig. 9(a). The cable is made by
16 multifilamentary strands with 75 mm pitch and is not at the neutral bending axis, but this is not crucial
for the ductile NbTi conductor: the bending strain in the winding is smaller than 1%. The soldering process
Fig. 8. The Swiss LCT conductor. Dimensions in millimeters. (Courtesy of ABB).
Fig. 9. Flat cables soldered to copper pipes: (a) NbTi ECN 8 T, 8.4 8.4 mm; (b) Nb3 Sn SIN 12 T, 21.5 8.4; (c) Nb3 Sn
ECN 12 T, 26.2 7.0 mm; (d) Nb3 Sn SIN 9 T, 26.3 9.6 mm. (Courtesy of the SULTAN group.)
was carried out at MCA, Massachusetts, in 1982 (19). A wave soldering device was not sufficient to fill the
interstices satisfactorily and had to be modified by adding a restraining die, where the solder becomes solid, as
in the OMEGA process. The solder cross section is less than 5%. Ten sections have been produced, each 500 m
long.
The Nb3 Sn react-and-wind conductor in Fig. 9(b) was manufactured at SIN (later CRPP) in 1985 (20).
The flat cable is made of 2016 multifilamentary strands (0.125 mm diam) according to the external Sn method.
The first cable stage (not fully transposed) is composed of 7 strands (1 + 6). The second cable stage (also
nontransposed) is composed of two layers of strands (6 + 12) cabled around a copper core. The last stage is a
970 m long flat cable (3.3 13.8 mm) of 16 subcables around a bronze strip. The flat cable is heat-treated on
an Inconel drum (7 layers) at a radius of 600 mm. This allows straightening the cable after heat treatment
without irreversible damage of the Nb3 Sn filaments and bending it eventually to the minimum winding radius,
300 mm, with the tolerable bending strain of 0.28% in Nb3 Sn.
The other components to be attached to the flat cable are a copper pipe, a copper profile as stabilizer,
and two stainless steel strips as a structural reinforcement. These four parts have been assembled with three
brazing foils and brazed at 630 C for a total length of 930 m in a continuous process, at the rate of 0.3 m/min.
10
The process could be stopped and restarted to replace the heating elements. The 200 m long sections of the
copper pipe were preliminarily brazed to a full length, using a higher-melting alloy.
The brazed steelcopper composite and the heat-treated Nb3 Sn cable were assembled in a combined
continuous millingsoldering process. A slot is milled in the copper strip, and the cable is carefully placed in
the slot. Bonding of the cable to the copper is achieved by a wave soldering device: the PbSn alloy fills up to
80% of the void area in the cable. The process speed is 0.5 m/min. Stop and restart was necessary every 70 m to
replace the cutting tool. As the thickness of the heat-treated cable varied at the layer transition, the thickness
of the slot had to be periodically adjusted. The 21.5 8.4 mm conductor was produced in a single length of
870 m and wound into a solenoid. The coil performance confirmed that, besides the expected bending strain,
no significant degradation occurred during the manufacturing process. The dc-operated coil did not suffer from
the lack of transposition in the cable.
For the Nb3 Sn conductor manufactured in 1986 at ECN (The Netherlands), see Fig. 9(c), some features
have been simplified (21). The flat cable, 18.5 1.85 mm, is made of 36 multifilamentary strands (1 mm
diam, powder-in-tube method), without central strip. The seven cable sections, about 140 m long, are wound
as loose pancakes on steel supports and heat-treated. The other conductor components (two steel strips, two
copper pipes, and two copper strips) are all assembled to the heat-treated cable in one run by a wave soldering
device. The steel strips were preliminarily electroplated with solder to ease the bonding. The process speed
was 1 m/min. Six out of seven conductor sections (total about 800 m) were wound as double pancakes with
interpancake joints.
To complete the SULTAN test facility in the split-coil configuration, three additional conductor lengths
were manufactured at SIN in 1988 and 1990; see Fig. 9(d). (22). The flat cable, 13.7 3.54 mm, is made of 98
multifilamentary strands (0.65 mm diam, bronze method), cabled in two stages, (1 + 6) 14, around a central
bronze strip. After heat treatment, the components are assembled in a single run (See Fig. 10), as in the ECN
conductor, by a wave soldering device, at the rate of 0.3 m/min. Two conductor lengths, 27.0 10.3 mm, each
800 m long, were wound into the two 9 T coils of the SULTAN facility. A third conductor section, 26.1 7.6
mm, 800 m long, with thinner steel strips, was used for the innermost, 12 T coils.
Sheathed Flat Cables

The first attempts to place the active superconducting elements inside a hydraulic conduit date back to the
late seventies. Besides the group of bundle conductors (see below), a small number of conductors have been
developed by encasing a more or less rigid flat cable into a welded conduit.
The EU-LCT Conductor. Vacuumschmelze (Germany) manufactured the EU-LCT conductor in 1980
1981 (23) after three years of development (24). The design optimization effort (ac loss, stability, and structural
aspects) led to a sophisticated conductor layout, requiring four continuous manufacturing processes to assemble
the components at very tight tolerance. The single-stage, NbTi flat cable, (stainless-steel-sheathed, 40 10
mm) is shown in Fig. 11.
Twenty-three NbTi rectangular composites, 2.35 3.1 mm, are cabled and soldered on a flat core, which
has a central insulating foil to avoid large interstrand coupling-current losses. The core is prepared starting
from a NiCr 0.5 mm thick strip, clad on one side with a 50 m thick CuNi layer to provide a base for soldering.
A 4 m thick Sn coating is electroplated on the CuNi side of the strip. The strip is folded on a 50 m thick
Kapton foil to form the core of the cable. To improve the strength, the folded strip is stretched, adding 6% cold
work. The tolerance on the final size of the strip is (33.8 0.55) (1.0 0.03) mm. Two foils of SnPbIn solder,
50 m thick, are then attached to the strip by fusing at both edges. The rate of the process is 5 m/min.
A cage strander with full back twist is used to wind the 23 composites edgewise on the flat strip, without
adding torsion to the bending. A Roebel head at the cabling point bends the composites accurately at the edge
of the flat core, always maintaining the short side of the composite parallel to the flat core; see Fig. 11. Spacing
11
Fig. 10. The millingsoldering line for the final assembly of the Nb3 Sn react-and-wind conductor for SULTAN (courtesy
of G. Pasztor, CRPP).
wires cabled together with the superconducting composites keep the 0.8 mm space between the strands. In the
same process, the strands are bonded to the core by melting the solder foils. This is achieved by pressing hot
metal blocks on the cable: the strand temperature is 200 to 210 C for about 10 s. The load is maintained till full
solidification occurs. Afterwards, the NiCr spacing wires, coated with a refractory varnish, are removed. The
transposition pitch is long (400 mm); however, the interstrand coupling-current loss is small, as the strands
are electrically in contact only through the thin solder layer. The speed of the cablingsoldering process (See
Fig. 12) is 1 m/min.
The 0.8 mm thick sheath is a folded strip of nitrogen-alloyed austenitic steel (316LN). To create additional
cooling channels and to limit the heat load to the superconductor during the welding, six steel spacers, 0.6 mm
thick, are placed between the cable and the sheath; see Fig. 11. The six spacers are preliminarily attached to
the 98 0.8 mm strip in a continuous process, by spot electron beam welding, with minimum heat load to
the thin strip. To avoid excessive outgasing in the vacuum environment of the electron beam, the steel for the
spacers is 316Ti (not nitrogen-alloyed).
12
Fig. 11. The EU-LCT conductor (courtesy of Vacuumschmelze, Germany).
The cable sheathing is the fourth and last continuous process for the conductor assembly; see Fig. 12.
Six shaping rollers fold the ribbed strip into a U where the cable is fitted, and another six rolls fold the strip
around the cable, leaving a welding gap <0.1 mm. A tungsten inert gas (TIG) pulsating torch carries out the
closing weld. The strand temperature underneath the weld increases to 260 C during 5 s; the temperature
rise does not affect the NbTi critical current density. Due to the electrode replacement, the process needed to
be interrupted every 50 m. The welding parameters at the restart points are adjusted to guarantee a reliably
tight weld transition.
The final tolerance on the conductor size is 0.1 mm. The corner radius is 2 mm. The conductor has been
produced in units of about 500 m, for a total length of 7 km.
Nb3 Sn Sheathed Flat Cables. The flat cable geometry is attractive for the Nb3 Sn react-and-wind
method because the active, strain-sensitive superconductor can be placed, after the heat treatment, on the
neutral bending axis, allowing the conductor to be wound into a coil with minimum degradation. Several such
developments were started in Europe and Japan in the eighties, but only few led to significant manufacturing
applications.
The three prototype conductors, developed in the mid eighties for the initial phase of the Next European
Torus (NET) project, were eventually manufactured only in 10 m sections (25,26,27). The rectangular
conductors include a heat-treated Nb3 Sn flat cable, 4.0 to 6.6 mm thick. An insulating core in the center of the
cable to reduce the coupling loss was shown not to be necessary (27), so the cable thickness could be reduced.
The stabilizer, assembled either as a mixed matrix monolith or as a transposed cable, is soldered to both sides
of the flat cable. The conduit is assembled around the soldered conductor by welding four steel strips. The
cooling channels are placed underneath the full-penetration weld seams: a strip folding process is not allowed,
due to the required conduit thickness (2.5 to 3.2 mm). Laser beam welding was needed to reduce the heat load
to the soft soldered cable and to maintain the tight assembly tolerance.
A Russian prototype react-and-wind conductor (28) has a layout similar to the NET prototypes, with four
stainless steel strips welded to form the outer sheath. However, the helium channels are eight parallel copper
pipes soldered to the central Nb3 Sn flat cable, and the weld seams do not need to be helium-tight.
The ETL Wind-and-React Flat Conductor. An unusual example of a Nb3 Sn flat cable-in-conduit, windand-react conductor was developed in the early eighties at ETL, Japan (29). Two coreless flat cables, each
made of 15 stabilized Nb3 Sn strands 1.4 mm in diameter, are separated by a CuNi strip to reduce the coupling
loss (lack of transposition). Two thick copper strips are rolled to form ribbed, U-shaped profiles with large
13
Fig. 12. The cablingsoldering (upper) and the sheathing (lower) processes for the EU-LCT conductor (courtesy of Vacuumschmelze, Germany).
cooling channels for the helium flow (the coolant is also allowed to flow in the interstices between cables and
stabilizers). The cable sandwich is assembled with the copper stabilizer and wrapped with a punched CuNi
tape: the components are not bonded together. The 0.75 mm thick outer copper sheath is welded in a tube mill.
The wrapped cablestabilizer is pulled through the welded copper pipe and drawn down to the final dimension
14
Fig. 13. The 13 23 mm Nb3 Sn wind-react conductor developed at ETL: coppersheathed, conductors, left and middle;
stainless-steel-sheathed conductors, right (courtesy of K.Agatsuma, ETL. Japan).
Fig. 14. React-and-wind Nb3 Sn conductors developed at JAERI for toroidal field coils: (a) STEP-1, and (b) TMC-FF
(courtesy of JAERI, Naka).
of 13 23 mm; see Fig. 13.. After the die, the conductor passes through a high-frequency annealing furnace to
soften the sheath. The production unit length ( 40 m) is limited by the size of the drawing bench.
Eight conductor sections (total about 300 m) were insulated by glass tape, wound into a coil with 300 mm
inner diameter (over 4% bending strain) and eventually heat-treated to form the Nb3 Sn.
The same manufacturing process was also applied for a stainless-steel-sheathed conductor (30). The
selected steel was 316L, the wall thickness 0.6 mm. Wrinkling of the sheath was observed at the smallest
winding radius. The properties of the longitudinal TIG welding seam were investigated before and after heat
treatment (800 C, 50 h). A pancake coil was wound, heat-treated, and tested with eight other pancakes based
on Cu-sheathed conductors. The busbars for the coil are manufactured with a thicker (1.5 mm) steel sheath;
see Fig. 13.
The TMC-FF Conductor at JAERI. The development of Nb3 Sn flat cable, react-and-wind conductors
for toroidal field coils started at JAERI in the early eighties with a short prototype, called STEP-1 (31); see Fig.
14(a). A Nb layer was included at the outer surface of the strand as a thermal barrier, to delay the diffusion
inside the strand of heat from friction at the strand surface. Eventually, the strand was drawn through a starshaped die cutting four grooves in the Nb outer shell to allow heat exchange between coolant and stabilizer. The
sheath was planned to be a thin CuNi strip with an outer steel reinforcement. However, the short prototype
was jacketed in a steel U profile with a welded lid.
The actual toroidal model pancake was wound in 1991 with 90 m of monolithic flat cable manufactured
by Hitachi Cable 32; see Fig. 14(b). The 4.8 mm thick Nb3 Sn cable consists of 23 nonstabilized strands soldered
after heat treatment to a number of U-shaped copper profiles, to build a monolithic, ribbed assembly. To
15
Fig. 15. React-and-wind poloidal Nb3 Sn conductors developed at JAERI: (a) the DPC-EX, 10.2 40.8 mm; (b) the DPC-TJ
with the steel shell machined reinforcement welded longitudinally around the thin conduit. (Courtesy of JAERI, Naka.)
increase the loop resistance and lessen coupling-current losses, a few epoxy spacers are interleaved between
the stabilizer elements. It was planned to build the conduit by longitudinal welding of two hot-rolled, 3 mm thick
steel profiles to obtain sharp outer corners and improve the conductors rigidity. However, this development
was skipped, and the conduit for the 90 m model conductor was a 2 mm thick steel sheath, TIG-welded in a
tube mill process.
The DPC-EX Conductor at JAERI. The Nb3 Sn conductor for the Demonstration Poloidal Coil (DPC-EX
coil) was manufactured in 198889 at Mitsubishi Electric (33); see Fig. 15. Over 500 m of the 10.2 40.8
mm conductor have been produced in two sections and wound after heat treatment into two double pancakes.
From the design point of view, the DPC-EX represents the transition between the generation of the flat hollow
conductors and the emerging group of bundle conductors (rope-in-pipe or cable-in-conduit).
The 153 Cr plated strands (0.81 mm diam) are cabled in three stages, 3 3 17. The last cable stage has
a steel strip core to cut the interstrand coupling loss. The helium void fraction in the cable space is in excess of
40%. At both sides of the cable, two stainless steel squared pipes provide the main cooling channels. The helium
in the cable space is quasistagnant: every 0.3 m, the steel pipes have mixing holes (3 mm diam) to improve the
heat exchange with the cable and to avoid a large pressure rise in case of quench. The outer sheath and the
two pipes are made of a stainless steel (JK-1) specially developed (34) to maintain high strength and ductility
at low temperature after the Nb3 Sn heat treatment at 700 C, 30 h. The conduit has constant thickness, 1.5
mm, and is assembled from two strips, folded and welded by two simultaneous TIG seams at the edge side, in
a modified tube mill (the steel pipes underneath the seams allow achieving full penetration without damaging
the strands).
Cable-In-Conduit Conductors
The development of cable-in-conduit conductors (CICCs) started slowly in the late seventies, with only few
significant applications till the late eighties. However, as soon as the manufacturing methods had been mastered
by the industry, the CICCs became very popular, and today they dominate the restricted market for forced flow
superconductors. The main design motivation for introducing the cable-in-conduit conductor is its large wetted
perimeter, improving the heat exchange and stability for pulsed field applications. The initial development was
aimed at Nb3 Sn react-and-wind technology. In the nineties, the use of CICCs was extended to the wind-andreact method, as well as to the NbTi technology.
16
Investigations of the stability and ac loss performance of the CICCs stimulated a number of experiments
with short-length, small-size conductors manufactured on purpose. This was in contrast with the former forced
flow conductors, where the experiments were carried out only on full-size conductors, manufactured to wind a
real magnet. The short conductors for small-scale experiments are ignored in this review, as they are mostly
irrelevant for the issue of manufacturing methods.
Instead of a one-by-one description of each conductor, the manufacturing aspects are grouped into four
categories, with cross-references that cannot always reflect the historical sequence.
Strand Coating. A coating of the strand surface may be applied to control the transverse resistance
of the cable as well as the friction and bonding at the strand contact points. The choice of coatings is broad,
ranging from the bare strand surface (no coating) to insulating varnish. The design criteria for ac loss and
stability (interstrand current sharing) drive the selection of the strand coating; see also the article HYSTERESIS
AND COUPLING LOSSES IN SUPERCONDUCTORS. The coating options for Nb3 Sn strands are restricted by the
requirement of withstanding the heat treatment.
The properties of metallic coatings for NbTi strands have been thoroughly investigated for accelerator
cables (not forced flow conductors). The coating methods include galvanic deposition of Ni, Cr, and Zn, as
well as dipping in molten SnAg alloy (35,36). For NbTi-based CICCs, copper oxide has been used as a strand
surface treatment in a SMES coil (37), and Cr plating for a prototype SMES conductor (38). The bare copper
surface has been selected for the dc-operated poloidal coils of the Large Helical Device (39), as well as for a
100 kA conductor of a flux pump (40). Organic-varnish-insulated strands have been used for the CICC of the
DPC-U coils (41). In the NbTi CICC for the 45 T hybrid magnet (42), the drawing lubricant is intentionally not
removed from the strands to enhance the contact resistance. The bare copper surface, as well as the oxidized
copper surface, proved to be not satisfactory for pulsed operation (because of high interstrand coupling loss).
Conductors with fully insulated strands showed severe instabilities due to their inability to redistribute the
current among the strands.
In Nb3 Sn-based CICCs, an antibonding coating is highly recommended to avoid intermetallic diffusion
at the strand crossovers during heat treatment. If the strands bond together without possibility of sliding
during the winding process (react-and-wind), the performance degradation due to the bending strain and the
interstrand coupling loss may become unacceptably high. An example of noncoated Nb3 Sn in a CICC, with very
high coupling-current losses, is provided by the DPC-TJ coil (43).
A 5 m copper sulfide layer was originally specified, but eventually not applied, for the WestinghouseLCT
conductor (44). The antibonding properties of the copper sulfide have also been assessed in the JA development
for Nb3 Sn CICC (45). In the US-DPC conductor,some lubricant was accidentally used during the cabling,
providing an oil film on the Crcoating (46) Based on this experience and on special ac loss experiments (47),
the bare Nb3 Sn strands of the CICC for the 45 T hybrid magnet were intentionally coated with Mobil 1 oil;
during the heat treatment the organic film partly vaporizes, leaving a blackened strand surface. The organic
vapors may have a detrimental effect on the other conductor components during the heat treatment, and the
loose carbonized particles may affect the flow dynamics in operation.
The Cr Electroplating. Today, the large majority of Nb3 Sn strands for CICCs are Cr-coated in a continuous electroplating process. The plating of wire with Cr is not a standard process, and a number of new
plating lines have been set up, especially for Nb3 Sn strands. Because of the environmental issues associated
with Cr electrolyte, the plating lines are mostly assembled at Cr plating companies rather than at companies
for continuous galvanic coating of wires.
Compared to other galvanic processes (e.g., Cu, Sn, and Ni), Cr electroplating is much slower (and more
expensive too): at a current density of 50 A/dm2 , the growth rate of the Cr layer is in the range of 1 m/min.
To obtain reasonable process speed and prevent the Cr plating from becoming bottleneck in strand production,
several strands can be pulled in parallel through the same galvanic bath, or the same strand can make multiple
passes through the bath. The process speed is obviously inversely proportional to the required thickness of the
Cr layer. As the antibonding effect of the Cr plating is due to the surface oxide, a thin coating (in the range of
17
Fig. 16. Scheme of a continuous Cr plating line (courtesy of ABB).
2 m) is preferable, for cost reasons, to a thicker one. Typical process rates for a 2 m coating are in the range
of 1 m/min for a single pass and can be substantially increased by multiple passes.
Prior to Cr plating, the strand needs to be cleaned, usually by galvanic etching. This step may be either
integrated in the Cr plating line (48) (see e.g., Fig. 16) or done separately, together with a surface passivation
(49).
The antibonding properties of the Cr plating are well established. However, the contact resistance at
the strand crossovers (i.e., the interstrand coupling loss) and the worsening of the copper RRR (due to grain
boundary diffusion during the heat treatment) may vary over a broad range, depending on the proprietary
electroplating process of the different vendors (50). A discussion of the role of the electrolyte composition is
given in Ref. 51.
Strand Bundling. The way of bundling the strands in a CICC may have a crucial effect on the conductors performance. For an even current distribution under time-varying current operation, the path, and
thus the self-inductance, of each strand in the bundle must be identical. This requirement may also be formulated as full transposition with respect to the perpendicular applied field. Layered cables of superconducting
strands, for example 1 + 6 or 1 + 6 + 12 configurations, should be avoided. However, it is allowed to place a
nonsuperconducting core at the center of a cable, as for example in the Swiss LCT cable (17): this option is
satisfactory for transposition issues, but further investigations are required to assess the role of the central
core (segregated copper) as a stabilizer for transient events. Among the coreless stranded bundles, the triplet
is definitely the most stable geometry. With increasing number of strands (e.g., five or six) there is a tendency
for one strand to slip to the center, making an imperfect, non-fully-transposed, 1 + 4 or 1 + 5 cable.
To bundle together a reasonable number of superconducting wires, multiple stranding (multistage cabling)
is necessary. To preserve the transposition, the pitches of each cable stage must not be integer multiples. To
avoid strand damage (45) and to reduce the interstrand coupling loss (52), the pitch direction must be the
same in all the cable stages. The typical pitch lengths are in the range of 10 to 20 times the diameter of the
cable stages. A sequence of tight pitches turns into a stiff cable with a large average angle of the strands
to the conductor axis, leading to ineffective use of the cross section. On the other hand, a sequence of loose
pitches may lead to strand slippage during the bundle compaction, with local transposition errors. The pitch
of the last cable stage is crucial for coupling loss. In very large bundles, the subcables may be wrapped with a
high-resistivity metal strip to cut the largest coupling-current loops (53). The wrap direction must be opposite
to the pitch direction to avoid loosening and crinkling of the strip when cabling the next stage with back twist.
The wrap may be spaced (no overlapping) to allow some coolant mass exchange. The use of the subcable wraps
to cut the interstrand coupling currents must be carefully watched in the design, as it may lead to a severe
penalty in interstrand current sharing and current redistribution.
The selection of the cable stage sequence, especially the number of elements in the last cable stage, is
made according to the final cable shapea rectangle, an annulus with central hole, a square bundle, or the
18
like. Table 1 (54,55,56,57,58,59) lists a number of fully transposed cable configurations for large bundles (more
than 200 strands).
The natural void fraction of a stranded cable prior to compaction is larger than 50%. However, the typical
void fraction required in the cable space is in the range of 30% to 45% (higher void fraction may allow strand
movement in operation; lower void fraction may affect the integrity of the strands and limit the stability or
the hydraulic performance). Compaction of the strand bundle may be achieved either with a die at the cabling
point or with shaped rolls (either freely revolving or engine-driven). For large compaction loads, a die is not
recommended, as abrasion may occur at the strand surface and large pulling loads may damage the cable.
In large multistage cables, progressive compaction is advisable at each cable stage for an even distribution of
the voids and to avoid very high compressive loads at the outermost strands. The springback of the bundle
after passing the compacting tool must be allowed for in setting the size of the rolls or die. The cross-section
reduction of the bundle is preferably done in one step. A thin, overlapped metal wrapping is sometimes applied
after compaction to protect the bundle in the subsequent handling and manufacturing processes. A final
calibrationshaping pass may be necessary if tight dimensional tolerance is required.
The hydraulic diameter of a strand bundle with 40% void fraction is in the same range as the strand
diameter. To reduce the pressure drop, whenever large mass flow rate is required in operation, parallel helium
channels, with larger hydraulic diameter, may be attached to the strand bundle. In the DPC-EX, the extra
channels are two perforated steel pipes assembled at both sides of the flat cable (33). In the US-DPC, the fourcorner area between the central, wrapped bundle and the outer square conduit provides high-speed helium
channels (46).
A practical way to add a pressure-relieving helium channel is to cable the strand bundle as an annulus
around a central hole. This also provides a sound, fully transposed layout for the last cable stage. The central
hole can be obtained by preshaping the subbundles with rollers (in the last but one cable stage) as wedges
and fitting them together in the last cable stage (60). In this way, the subbundles form a vault and the central
hole does not need a mechanical support. However, matching the preshaped subbundles correctly requires an
accurate, continuous adjustment of the back twist of the individual components. The cabling rate is very low,
and only a limited compaction load can be applied after cabling. In the ITER conductors (53) [Fig. 17(d, e)], as
well as in the prototype 200 kA SMES (61) [Fig. 17(b)], the QUELL conductor (62) [Fig. 17(c)], and the HT-7U
(63) [Fig. 17(f)], the last cable stage is formed on a steel spiral that allows helium exchange between the central
hole and the bundle. No preshaping is necessary in the subbundles of the ITER conductor, which are compacted
against the spiral core, yielding a homogeneous filling of the annulus.
19
Fig. 17. Examples of CICC with pressure-relieving helium channels: (a) the Nb3 Sn US-DPC, 22.3 22.3 mm with central
heater (courtesy of MIT); (b) the NbTi 200 kA SMES, 55.1 mm diam (courtesy of Bechtel); (c) the Nb3 Sn QUELL CICC, 19.4
mm diam (courtesy of CRPP); (d) the Nb3 Sn ITER-CSMC, 51 51 mm (courtesy of ITER); (e) the Nb3 Al insert conductor,
45.7 mm diam (courtesy of JAERI); (f) the NbTi HT-7U conductor (from Ref. 63).
The subbundles of the Polo conductor (64) are cabled around a solid steel pipe (See Fig. 18), which forms
a separate, central conduit for forced flow of two-phase helium at 0.12 MPa. The inner conduit is also wrapped
with adhesive layers of Kapton and can be used in the quench detection system to balance the inductive voltage
component. The annular space contains stagnant supercritical helium pressurized at 0.4 MPa. The 13 subbundles of six NbTi strands around a CuNi core are spaced by an insulating wrap covering 70% of the surface.
Neither instabilities nor ramp rate limitations have been observed in the perfectly transposed conductor, but
the dc performance was degraded by 30% (65).
Braided Bundles. Instead of multiple stranding, an alternative way to bundle together the superconducting strands for a CICC is braiding. The hollow braids in common use in electrical engineering (e.g., for
20
Fig. 18. The Polo conductor, 22.5 22.5 mm. with double, separate cooling channels and insulated subbundle (courtesy
of VAC).
high-frequency shields) consist of two interlocked multiple spirals with opposite pitch direction. Such hollow
braids are transposed for perpendicular field, but are not suitable for a CICC, because the symmetry of the
hollow geometry is destroyed during compaction and, in case of parallel time-varying field component, large
voltage may be induced between the two opposite by running spirals. The flat braids proposed for superconducting accelerator dipoles are fully transposed for perpendicular and longitudinal applied field, but do not fit,
because of the large aspect ratio, into a practical CICC.
In lattice braids, similar to some type of alpine ropes, all the elements are interlocked in a bulk bundle.
For selected layouts, the strand bundle is fully transposed for both transverse and longitudinal applied field
(notice that stranded cables are never transposed for longitudinal applied field). Lattice braids were proposed
as an option for the WH-LCT conductor (66). The full transposition, high flexibility, and mechanical stability
of the bundle, as well as the large wetted perimeter, made very attractive the idea of a lattice braid as the first
cable stage for a CICC bundle. If all the strands are braided at once into a large bundle, the aspect ratio of the
braided bundle can be designed starting from the layout of the braiding platform (See Fig. 19) without the need
of final shaping as with stranded bundles. The only drawback for practical application of the lattice braids is
that such braiding machines are not industrially available and must be specially built. The complexity (and
the cost) of the braiding machine sharply increases with the number of elements to the braided at once. For
21
Fig. 19. The 29-strand lattice braid: scheme of the braiding platform with the movements of the carriers (left) and
operating braiding machine (right). (Courtesy of ABB.)
this reason, the braid option was abandoned in the WH-LCT, after a few meters of 34-strand bundle had been
manually braided.
A powered braiding machine, able to handle 29 strands at once, was built in 1989; see Fig. 19 (48). The
29-strand lattice braid was used as a first cable stage for a rectangular Nb3 Sn prototype CICC (67), with a total
of 609 strands (29 3 7). Compared to a similar conductor with purely stranded cable configuration ( 3
3 73 3 3 7) and identical overall void fraction, it is evident from Fig. 20 that the braid-based conductor
has a more homogeneous distribution of voids. Mechanical tests (68) also indicate a higher transverse modulus
for the braided bundle because of the interlocked structure (no line contacts among strands).
Conduit Materials. The selection of the conduit material is dictated by the mechanical requirement
in operation, as well as by manufacturing issues (weldability, industrial availability of strips or pipes, and
compatibility with the heat treatment, if any).
Copper is seldom used as a conduit material because of its low strength. The ETL wind-and-react Nb3 Sn
flat cable (see above) has a 0.75 mm thick, Cu welded sheath (29), manufactured in straight, 40 m long sections
by longitudinal welding. A 200 m long section of NbTi CICC with 55 strands was manufactured in a continuous
tube mill process, using a 0.6 mm thick copper strip (69). Although the coils wound with the copper-sheathed
conductors were small (inner radius 0.15 m and 0.26 m respectively), in both cases the sheath cross section
was not designed to withstand alone the hoop load in operation.
Low-carbon-content, austenitic steel is the favorite material for CICC sheaths, because of its ductility,
strength, broad availability in almost any shape, and low cost. Whenever higher strength is required, nitrogenalloyed austenitic steels (e.g., AISI 316LN and AISI 304LN) may be used. Some precautions must be taken using
the nitrogen-alloyed steels with Nb3 Sn conductors, due to the possible reduction of the fracture toughness after
22
Fig. 20. Cross sections of two prototype CICCs for NET: braided bundle, 29 3 7 (left) and stranded bundle, 3 3 3
3 7 (right). (Courtesy of ABB.)
Fig. 21. Use of Al alloys for CICCs: (a) an MIT test conductor (courtesy of B. Montgomery) with Al primary conduit and
co-wound profiles; (b) the W7-X conductor, with coextruded All6060 alloy (courtesy of EM-LMI); (c) the GEM conductor
with steel primary conduit and coextruded outer Al cladding. (Courtesy of N. Martovetsky, LLNL.)
heat treatment. A large effort has been carried out, mostly in Japan, to develop and select suitable stainless
steels to prevent degradation of the mechanical properties after heat treatment (70,71,72,73).
Among the relevent Nb3 Sn-based CICCs with austenitic steel conduits, we recall the 45 T-A and B hybrid
(74), the DPC-EX (33), the DPC-TJ (54), the ENEA conductor (75), the ITER-TFMC (76), the HFTF (77), the
FC 150 m coil (78), and the 12 T TFMC pancake (79).
Most of the NbTi-based CICCs, have an austenitic steel conduit, for example, the poloidal LHD coils (55),
and the DPC-U (58), the EU-LCT (24), the Italian MHD (80), the Polo conductor (64), the 45 T-C hybrid (42),
the JF-30 (81), the 200 kA SMES (61), and the HT-7U (63).
Aluminum-Based Conduit. An Al-sheathed CICC [nontransposed cable of 3 (+ 6 + 121 + 6 + 12)
NbTi strands] was manufactured in short lengths in 19771978 (82). The 0.9 mm thick Al round pipe, 12 mm
diam, was co-wound with two matching Al extruded and anodized profiles; see Fig. 21 The test conductor was
bent to small radius and used for stability experiments.
For the W7-X coil system, consisting of nonplanar windings with small bending radius, a NbTi based
CICC was proposed (83) with conduit made from a coextruded Al alloy; as extruded, the CICC is soft (y
100 MPa) and can be easily bent to small radius. After aging at 130 to 180 C, which can be combined with
23
the epoxy impregnation process, the Al 6060 alloy hardens (y 280 MPa at 4 K), providing the required
mechanical stiffness to the winding (56). The square conduit, 14.8 14.8 mm (See Fig. 21), is coextruded
on the 10 mm diam cable, without any protective wrapping, in unit lengths of 600 m. To avoid degradation
of the NbTi conductor performance, the temperature and speed of the coextrusion process must be carefully
controlled: a decrease by 5.5% of the strand Ic performance was observed for the W7-X conductor (84). The
coextrusion of superconducting cables and pure Al is well developed for particle detector magnets. For W7-X,
the process was first used to produce an internally cooled conductor.
A coextrusion method was also used for the GEM detector conductor (85). However, here the cable (See
Table 1) is preliminarily sheathed with a 1.5 mm thick 304L steel conduit and the thick, pure Al clading has
no helium containment function. The bonding of Al to steel was proved to be effective for stability. With an
overall size of 49.8 68.5 mm (84% consisting of the pure Al clading), the GEM conductor is the largest forced
flow conductor manufactured in useful length; see Fig. 21
Incoloy 908 Conduit. The use of a superalloy as a conduit material for CICC was first considered for the
Nb3 Sn WH-LCT conductor, due to the disappointing experience after heat treatment with the nitronic 40 (a
steel with 0.33% nitrogen content). The R&D work (86) led eventually to the choice of JBK-75, a 30% Ni alloy
specially developed from the A-286 superalloy, as a sheath material.
The main drawback of the steel as conduit material for Nb3 Sn CICCs is its large coefficient of thermal
expansion: from the heat treatment down to the operating temperature, the differential contraction of steel
and Nb3 Sn causes a degradation of the superconducting properties, especially severe at high field. In the scope
of the US-DPC coil R&D program, a new superalloy of the Incoloy group was developed as sheath material
for Nb3 Sn CICCs, with coefficient of thermal expansion matching that of Nb3 Sn (87). Several test programs
have been carried out to compare the mechanical properties of Incoloy 908 and 316 LN steel (see, e.g., Ref.
88). However, the most impressive advantage in the use of Incoloy 908 as a conduit material is the gain in
the high-field performance of the Nb3 Sn (89). On the other hand, a big concern in the use of Incoloy 908 is
the possible occurrence of stress-accelerated grain boundary oxidation (SAGBO) during the heat treatment,
leading eventually to catastrophic failure of the material. Relaxation of the surface stress in the conduit by
shot peening, as well as tight control of the oxygen content (below 1 ppm) in the heat treatment atmosphere,
are the key measures to avoid SAGBO (90,91).
The US-DPC conductor was sheathed with a 2.35 mm thick strip of Incoloy 908, TIG-welded in a tube
mill (46). The first large-scale application of Incoloy 908 as a conduit material for a Nb3 Sn CICC is ITER-CS
model coil conductor (53). Several tens of tonnes of round-in-square tubing, as 6 m to 11 m long sections, have
been manufactured by combined hot, hollow extrusion and cold drawing (92).
Pure titanium has been proposed as an alternative low-coefficient-of-thermal-expansion material for
sheathing of Nb3 Sn-based CICCs. Its low-temperature mechanical properties after aging are reported in Ref.
93. The only important CICC with Ti conduit is the 100 m long QUELL conductor, with 3 3 4 6 Nb3 Sn
strands and 1.2 mm thick pure Ti round pipe (62).
Jacketing Methods. The selection of the jacketing method depends to a large extent on the dimension
and cross section of the jacket. A key issue affecting the jacketing is the wall thickness. Conduits with constant
wall thickness can be manufactured and assembled by a variety of methods, some of which are the cheapest.
On the other hand, a conduit with variable wall thickness at the corners allows better packing of the conductor
in the winding and offers superior mechanical performance.
The Tube Mill. The large majority of CICCs are jacketed by the tube mill method. The generic scheme
of the process is shown in Fig. 22, from Ref. 54. The metal strip is formed by rolls around the cable in a large
number of steps; see Fig. 23(a) for the WH-LCT conductor (94) Fig. 23(b) for the DPC-U (58), and Fig. 23(c)
for the 45 T hybrid (59). The longitudinal TIG weld is usually done while the conduit is oversize, to limit
the thermal load on the strand bundle underneath the welding seam and to avoid contact between the weld
bead and the bundle. However, it is also possible to fold the strip exactly to the final size and weld it without
damaging the superconductor, as for example in the DPC-EX (33) and the EU-LCT (24).
24
Fig. 22. Scheme of the tube mill method for CICC, courtesy of JAERI.
After forming and welding the round pipe (a highly standardized industrial process), the conduit is
compacted in a number of steps to suppress the gap between pipe and bundle, and to give the square or
rectangular shape required for winding; see Fig. 23 Typical conduit elongation during the compaction process,
by rolling and/or drawing, is up to 10%. It is crucial to control the cable elongation during the compaction:
at large cross section reduction and low void fraction, plastic axial deformation of the bundle may start, with
fatal consequences for the stranded cable, which does not withstand elongation larger than 2% to 3% without
permanent damage. On the other hand, good mechanical engagement between cable and conduit during the
final shaping steps is highly desirable, to restrain dangerous strand movements under the operating load.
Starting from a metal strip, it is not possible to achieve sharp corners in the final, squared conductor.
The typical range for the outer corner radius is 3 mm to 6 mm, depending on the wall thickness. The final cold
work of the conduit (starting from the fully annealed strip) depends on the cross-section reduction, the conduit
thickness, and the corner radius.
The thickness of the conduit (typically 1.5 mm to 2.5 mm) ranges from 0.8 mm (EU-LCT) up to 3.5 mm
(LHD-OV). The tube mill method is used to form conduit from a variety of metals, including copper, titanium,
Incoloy 908, and austenitic steels of different composition. A list of selected CICCs sheathed by a tube mill is
given in Table 2 (95). Some cross sections of CICCs jacketed in tube mill are shown in Fig. 24.
Quality control of the weld (59) is obtained through tight monitoring of the process parameters. The
presence of the strand bundle underneath the weld seam precludes a check by X rays or eddy currents after
compacting and shaping. Stop and restart of the process (e.g., to replace the TIG electrode and the gas bottles)
is common practice for the tube mill method.
The Pullthrough Method. In the pullthrough method (see scheme in Fig. 25) the sheath is first manufactured as a straight, oversize conduit of the same length of the final conductor section. In a second step, the
bundle, wrapped with a protective metal foil, is pulled through the oversize conduit. Eventually, the conduit
cable assembly is compacted by rolling or drawing and coiled on a take-up reel.
The pullthrough method can be applied to any conduit geometry; however, it is especially attractive for
thick-walled conduit and conduit with nonconstant wall thickness (which cannot be welded in a tube mill).
Compared to the tube mill, an advantage of the pullthrough method is that the welds of the conduit can be
fully inspected (and repaired) before assembling the superconductor.
25
Fig. 23. The forming steps for the metal strip in a tube mill: (a) the WH-LCT (courtesy of P. Sanger); (b) the 45 T hybrid
(courtesy of J. Miller); (c) the DPC-U (courtesy of JAERI).
The straight, oversize conduit can be assembled by different methods, including longitudinal welding of
strips or profiles. However, all the pullthrough applications to date have used the butt welding of extruded
tubing sections (typical unit length 6 m to 12 m). In one case (96), the 1 mm thick Incoloy 908 pipe was first
manufactured in a tube mill, then cut in 10 m sections, and eventually reassembled by butt welding at the
jacketing site. The pullthrough method was first applied for short conductor length and prototype conductors
26
Fig. 24. Examples of CICCs jacketed in a tube mill: top left, the WH-LCT (94); top middle, the LHD-IV (courtesy of
NIFS); bottom left, the DPC-U (courtesy of JAERI); bottom middle, the JF-30 (courtesy of JAERI); right, the 45 T hybrid
conductors (courtesy of NHMFL).
(e.g., Refs. 30,60). Table 3 reports the pullthrough jacketing applications with useful (>50 m) conductor length.
Except for the ITER-CS conductor, the conduits quoted in Table 3 have constant wall thickness.
The butt welding of the thin-wall conduits (2 mm) is done with a TIG single-pass orbital welder without
filler. A second pass with filler was necessary for the 3 mm thick conduit of GEM (85). For the circle-in-square
27
Fig. 25. Scheme of a jacketing process according to the pullthrough method.
Incoloy tubing of ITER-CS, a number of orbital passes with filler were carried out after the first root pass.
Eventually, the corner area was filled manually. Special attention has to be paid to the weld protrusions (drops)
inside the conduit: these can be monitored either by an endoscope or by inserting a calibrated gauge. The
quality assurance on the butt welds may include X rays, ultrasound, dye penetrant, and leak tightness.
The applicable pulling load, F p , is limited by the yield strength of the cable (i.e., the threshold d where
strand damage occurs) and by the strength of the conduit, which must react the pulling load at the insertion
point (this is an issue only for thin-walled jackets). The pulling load per unit length has been found (96) to be
a function of the clearance between the cable and the oversize conduit. The coefficient of friction, , defined
as the pulling load divided by the cable weight can also be monitored. There is broad agreement (53,75,76,96)
28
that the insertion clearance must be larger than 1.5 mm for friction coefficients < 1. However, according
to the actual tolerance on the size of cable and conduit, a larger clearance may be necessary. The longer the
conductor section to be jacketed, the more crucial becomes the issue of clearance. For short length, the risk of
higher friction may be accepted. For long section jacketing, high initial friction may lead to an avalanche effect
and the cable may eventually get stuck in the conduit (76).
After pullthrough, the conductor is compacted by rolling or drawing to suppress the insertion clearance.
For conduits with nonconstant wall thickness, large cross-section reduction should be avoided to keep tight tolerance in the final conductor geometry. On the other hand, the compaction must achieve a positive engagement
of cable and conduit to restrain strand movement in operation, which may result in dangerous instabilities.
The jacketing of long sections of CICC by the pullthrough method has been successfully demonstrated at
VNIIKP, Moscow, where a 1 km long jacketing line has been set up and operated (96). The maximum conductor
length Lmax that can in principle be jacketed by the pullthrough method can be assessed by imposing the
constraint that the pulling load at the head of the cable does not exceed the tensile cable strength where strand
damage occurs, d :
where is the density of the cable, typically in the range of 7 103 kg/m3 to 8 103 kg/m3 . The range of
d depends on the cable pattern and strand properties. For the friction coefficient , a safety margin must be
considered. Assuming, conservatively, d = 50 MPa and 1, the maximum allowed length would be about
700 m.
Laser Beam Welding. CO2 laser beam welding was proposed in 1986 (97) as an effective method to form
thick conduit by longitudinal welding. The laser beam allows deep penetration and continuous joining with
negligible deformation of thick-walled conduits in a single pass, without filler. Due to the small heat load on
the underlying cable, laser welding can be applied to the final conductor size without the final compaction and
shaping steps, which are mandatory in the pullthrough method and in most tube mills to suppress the gaps
between cable and welded conduit. Compared to the electron beam, also used for narrow welding of thick walls,
the laser beam is most suitable for continuous processes, as it does not need a vacuum environment.
Conduit segments with variable wall thickness can be manufactured by hot rolling and cold drawing in
sections up to about 20 m, butt-welded and coiled to the required conductor length. Two longitudinal welds are
required to join the jacket segments, with the cable fed in between, as in the tube mill. This can be obtained
either by two synchronized laser beams or by a single, optically split beam. For a process rate in the range of 2
m/min, the required beam power is about 1 kW per millimeter of wall thickness. The penetration depth of the
seam can be controlled in practice to 10% (mostly due to the plasma oscillations in the weld). To guarantee
full penetration of the weld and avoid contact of the strand bundle with the laser beam, protection underneath
the weld is mandatory: this can be either a wrap on the cable or a longitudinal strip or gap under the seam.
Longitudinal laser beam welding has been applied only to two conductors: Polo (64) and the NET-ABB
prototype conductors (48). The jacket of Polo is built from four quadrant profiles (see Fig. 18), preassembled by
laser welding into two U profiles and eventually welded by two synchronized beams on the wrapped cable. The
weld seam is 1.5 mm thick. Four conductor sections, each 150 m long, have been produced. The actual welding
time for one section was less than 1 h, with several stops and restarts (97). Eddy currents were used to check
the quality of the welds. Leaking spots were repaired by TIG welding.
In the NET-ABB prototype conductor (see Fig. 20) the weld seam is 3.9 mm thick. An air gap and a
longitudinal steel strip underneath the weld protect the strand bundle from the contact with the beam. Quality
assurance by ultrasound can be applied online in the process.
Despite the satisfactory employment and attractive features of laser beam welding (high rate, no deformation, minimum heat load for large penetration depth), no further application of it has been reported for
29
CICC jacketing in the last decade. The main reasons are the investment cost for the laser device, which can be
justified only for large series production, and the limited experience of industrial conductor suppliers with the
laser welding technology (a similar situation to that for lattice braiding; see above).
The Coextrusion Method. Because of the need for temperature compatibility, only aluminum alloys (see
above) can be used to jacket a CICC by the coextrusion method. For the W7-X (56), the square, coextruded
conduit fulfils at once the functions of helium containment, structural support, and stabilization; see Fig. 21.
In the case of the GEM conductor (85), the pure Al cladding is coextruded on the steel jacketed conductor and
has only the function of limiting the hot spot temperature in case of quench.
Double Conduit. To avoid the problem of assembling a thick-walled jacket, a double conduit has been
used in a few cases. The additional conduit has no helium containment function, but its cross section adds to
the main conduit from the structural point of view.
The secondary conduit may be inside the main conduit, as in the US-DPC [see Fig. 17(a)], where a 1 mm
thick Incoloy 908 strip is formed in a tube mill and welded (not helium-tight) on the round cable (46). The inner
conduit stabilizes the shape of the cable, defines the void fraction, and protects the cable during the welding of
the thicker, outer conduit. During the development of the Polo conductor (64) (Fig. 18), an inner conduit with
punched holes was initially foreseen as a protection for the cable during the outer conduit welding (the inner
conduit was later replaced by a double wrap).
An outer secondary conduit can be used to give a structural reinforcement and provide a better shape for
winding. This possibility is of special interest when the primary conduit is made in a tube mill, with constant
wall thickness and large corner radii. In the MIT test conductor (82), both primary and secondary conduits are
aluminum (see Fig. 21): the two extruded U profiles are fitted to the round NbTi conductor without welding. In
the DPC-TJ (99) (see Fig. 15), the Nb3 Sn conductor, jacketed in a tube mill, pancake-wound, and heat-treated,
is eventually encased and spot welded into two steel U profiles (preformed armor). The thin-walled, sharpcornered U profiles are manufactured by 3-D milling a spiral into a thick steel plate, in order to fit the shape
of the heat-treated pancake.
Choice of Manufacturing Methods

The conductor layout, and hence the manufacturing methods, are to large extent dictated by design considerations. However, the cost and the reliability of the manufacturing process should also be taken into account in
the selection of the conductor layout.
Cable-in-conduit conductors are today more popular than soldered conductors, although many applications
actually do not require the special features of the CICC. From the manufacturing point of view, a soldered
conductor (e.g., a layer of NbTi strans cabled and soldered on a square copper pipe) is preferable and cheaper
for low operating current and overall cross sections smaller than 60 mm 2 . On the other hand, a CICC is more
effective for high current and cross sections larger than 100 mm2 .
For cable-in-conduit conductors, a strand coating is mandatory whenever low ac loss is required, but
may be omitted for dc-operated conductors. The electroplated Cr coating is the only practical choice for Nb3 Sn
strands. More options are available for NbTi strands, including hot dipping in low-melting alloys and galvanic
coating with Ni or Cr.
The requirement of transposition is applied currently to all the strand bundles for CICCs. However,
nontransposed cables have been successfully operated in dc mode. An attractive, cost-saving option is to cable
a layer of strands on a central copper core as the first cable stage, reducing the copper ratio in the strands
and hence the overall mass of superconducting strand, and saving one cabling stage compared to the multipletriplets pattern. Lattice braids have proved to be an elegant and effective method for strand bundling, but
require the use of dedicated machinery.
30
The most popular conduit material for CICC is low-carbon stainless steel (nitrogen-alloyed, whenever
required), readily available in any shape, easy to weld, and cheap. For Nb3 Sn conductors, the Incoloy 908, due
to the matching coefficient of thermal expansion, offers a superior performance at high field and competing
mechanical properties. The higher cost, the precautions to be taken at welding and heat treating, and the
single-source supplier to some extent balance the advantages of the Incoloy 908 over stainless steel.
The tube mill is a highly developed industrial process for pipe production. For CICCs with thin, constantthickness conduits, the tube mill is the cheapest, easiest jacketing method, with virtually no limit on the
conductor length. For thick-walled conduit, as well as for nonconstant thickness, the tube mill and TIG welding can be replaced by laser beam longitudinal welding of preassembled profiles. The pullthrough method,
definitely preferable for prototypes and short-length jacketing because of the small investment and the low
risk, has become unexpectedly popular also for long length and series production, where its competitiveness is
questionable.
Three decades of forced flow superconductors show clearly a trend toward easy, low-technology manufacturing methods despite the successful demonstrations of sophisticated layouts and advanced technology (e.g.,
EU-LCT and Polo). Initially, the research institutes were the leaders in conductor development. Today, many
companies can supply forced flow superconductors without the support of design and R&D activities from the
lab community.
BIBLIOGRAPHY
1. H. H. Kolm, A closed loop cooling system for superconducting bubble chamber magnets, Proc. Int. Symp. Magn. Technol.,
Stanford, CA, 1965, p. 611.
2. M. Morpurgo, The design of the superconducting magnet for the OMEGA project, Part. Accel., 1: 255, 1970.
3. M. Morpurgo, Construction of a superconducting test coil cooled by helium forced circulation, CERN Report 68-17,
1968.
4. M. Morpurgo, A superconducting solenoid cooled by forced circulation of supercritical helium, CERN Report 69-25,
1969.
5. C. Lesmond, J. C. Lottin, S. Shimamoto, Experiment with hollow conductor superconducting magnet, Proc. Magn.
Technol. Conf., Hamburg, 1970, Vol. 3, p. 925.
6. G. Meyer R. Maix, Superconductors and superconducting magnets, Brown Boveri Rev., 57(7/8): 355, 1970.
7. Y. Naganuma, et al. The manufacture of hollow superconductors, Proc. ICEC, Kyoto, 1974, Vol. 5, p. 508.
8. V. N. Agureev, et al. Electroplated stabilized multifilament superconductor, IEEE Trans. Magn., 11: 303, 1975.
9. D. P. Ivanov et al Test results of tokamak-7 superconducting magnet system (SMS) sections, IEEE Trans. Magn., 15:
550, 1979.
10. E. Yu. Klimenko et al, Superconducting conductor for T-15 toroidal magnet, Sov. At. Energy (Engl. Transl.), 63: 756,
1987.
11. N. Schaetti, Superconductors for the magnet coils of the Omega spark chamber at CERN, Brown Boveri Rev., 59 (2/3):
73, 1972.
12. M. Morpurgo, A large superconducting dipole cooled by forced circulation of two phase helium, Cryogenics, 19: 411,
1979.
13. H. Hillmann, Fabrication technology of superconducting material, in S. Foner and B. B. Schwartz, (eds.), Superconductor
Material Science, Ser. B68, New York: 1981. NATO Adv. Study Inst., 1981.
14. G. Pasotti, et al. SULTAN: An 8 T, 1 m bore test facility, the outer solenoid, IEEE Trans. Magn., 17: 2007, 1981.
15. U. Trinks, et al. A protoype coil for the superconducting separated sector cyclotron SuSe, J. Phys., Cl. Suppl. 1, 45: 217,
1984.
16. K. Agatsuma, et al. Braided multifilamentary Nb3 Sn hollow superconductor and its magnet, IEEE Trans. Magn., 15:
787, 1979.
17. H. Benz, et al. The conductor for the Swiss LCT coil, IEEE Trans. Magn., 17: 2213, 1981.
18. H. Benz, et al. Design and manufacture of the conductor for the Swiss LCT coil, IEEE Trans. Magn., 19: 711, 1983.
31
19. W. M. P. Franken, et al. Manufacture of the hollow supercritical He cooled conductor for the ECN/SULTAN project,
IEEE Trans. Magn., 19: 368, 1983.
20. B. Jakob G. Pasztor, Fabrication of a high current Nb3 Sn forced flow conductor for the 12 tesla SULTAN test facility,
IEEE Trans. Magn., 23: 914, 1987.
21. J. A. Roeterdink, et al. Design and construction of the ECN 12 T niobium tin magnet insert for the SULTAN facility,
IEEE Trans. Magn., 24: 1429, 1988.
22. B. Jakob, G. Pasztor, R.G. Schindler, Fabrication of high current Nb3 Sn forced flow conductors and coils for the SULTAN
III test facility, Fusion Techology 1992, Amsterdam: Elsevier, 1993, p. 872.
23. H. Krauth, et al. Manufacturing and testing of a forced flow cooled superconductor for tokamak magnets, IEEE Trans.
Magn., 17: 918, 1981.
24. H. Krauth, et al. Development and testing of a forced flow cooled superconductor for LCT, Proc. 8th Symp. Eng. Probl.
Fusion Res., 1979, p. 1451.
25. B. Jakob, et al. Design and fabrication of a 17 kA preprototype Nb3 Sn conductor for the Tf coils of the NET fusion
project, IEEE Trans. Magn., 24: 1437, 1988.
26. J. M. Plaum, et al. Development toroidal field conductor for NET, IEEE Trans. Magn., 24: 1433, 1988.
27. R. Flukiger,
et al. The Nb3 Sn react and wind conductor for NET toroidal field coils and its boundaries, Fusion Technology
1990, Amsterdam: Elsevier, 1991, p. 1584.
28. V. A. Glukhikh, Programme of the conductor development for the ITER toroidal field coils, Proc. Magn. Technol. Conf.,
Tsukuba, 1989, Vol. 11, p. 886.
29. K. Agatsuma, et al. Fabrication and test of a forced cooled Nb3 Sn superconducting coil, IEEE Trans. Magn., 21: 1040,
1983.
30. K. Agatsuma, et al. Stainless steel sheathed forced internally cooled Nb3 Sn superconductor and its coil test, IEEE
Trans. Magn., 23: 1535, 1987.
31. T. Ando, et al. +2 T test module coil (TMC-II) in the cluster test program, Proc. 10th Symp. Fusion Eng., Philadelphia,
1983, p. 1346.
32. M. Sugimoto, et al. Development of hollow cooling monolithic conductor for ITER TF coil, IEEE Trans. Magn., 28: 218,
1992.
33. T. Ando, et al. Fabrication and test of the Nb3 Sn demo poloidal coil (DPC-EX), Fusion Technology 1990, Amsterdam:
Elsevier, 1991, p. 243.
34. H. Nakajima, et al. Tensile properties of new cryogenic steels as conduit material for forced flow superconductors at 4
K, Adv. Cryog. Eng. Mater., 34: 173, 1988.
35. M. D. Sumption, et al. Contact resistance and cable loss measurements of coated strands and cables wound from them,
36. J. M. Depond, et al. Examination of contacts between strands by electrical measurement and topographical analysis,
37. T. Satow, et al. Present status of 480 MJ/40 MW SMES development project, Proc. Int. Conf. Electr. Eng., Matsue,
Japan, 1997.
38. K. Kwasnitza, A. Sultan, S. Al-Wakeel, AC losses of a 10 kA NbTi cable-in-conduit superconductor for SMES application,
Cryogenics, 36: 27, 1996.
39. K. Takahata, et al. Stability tests of the NbTi cable-in-conduit superconductor with bare strands for demonstration of
the large helical device poloidal field coils, IEEE Trans. Magn., 30: 1705, 1994.
40. G. Pasztor, et al. Design fabrication and testing of a 100 kA superconducting transformer for the SULTAN test facility,
Proc. Magn. Technol. Conf., Beijing, 1997, Science Press, 1998, Vol. 15, p. 839.
41. Y. Takahashi, et al. Experimental results of stability and current sharing of NbTi cable-in-conduit conductors for the
poloidal field coils, IEEE Trans. Appl. Supercond., 3: 610, 1993.
42. J. E. C. Williams, et al. The development of a NbTi cable-in-conduit coil for a 45 T hybrid magnet, IEEE Trans. Magn.,
32: 1633, 1994.
43. K. Okuno, et al. Ac loss performance of 1 m bore, large-current Nb3 Sn superconducting coils in JAERI demo poloidal
coil project, IEEE Trans. Appl. Supercond., 3: 602, 1993.
44. C. J. Heyne, et al. Westinghouse design of a force flow Nb3 Sn test coil for the large coil program, Proc. 8th Symp. Eng.
Probl. Fusion Res., San Francisco, 1979, p. 1148.
45. N. Aoki, et al. Development of forced-cooled Nb3 Sn bundle conductor, IEEE Trans. Magn., 19: 733, 1983.
32
46. M. M. Steeves, et al. The US demonstration poloidal coil, IEEE Trans. Magn., 27: 2369, 1991.
47. B. J. P. Baudouy, et al. AC loss measurements of the 45 T hybrid/CIC conductor, IEEE Trans. Appl. Supercond., 5: 689,
1995.
48. P. Bruzzone, Fabrication of a short length of wind and react conductor, ASEA Brown Boveri Rep. HIM 20420, Zurich,
1990.
49. Y. Ipatov, et al. Galvanic chrome plating of copper wire for the ITER program. Proc. ICEC/ICMC 96, Kitakyushu,
Japan, Elsevier, 1996, 1969.
50. P. Bruzzone, A. Nijhuis, H. H. J. ten Kate, Effect of Cr plating on the coupling current loss in cable-in-conduit conductors,
Proc. ICEC/ICMC 96, Kitakyushu, Japan, Elsevier, 1996, p. 1243.
51. Y. Ipatov, P. Dolgosheev, V. Sytnikov, Prospective barrier coatings for superconducting cables, Supercond. Sci. Technol.,
10: 507, 1997.
52. K. Kwasnitza I. Horvath, Experimental evidence for an interaction effect in the coupling losses of cabled superconductors, Cryogenics, 23: 9, 1983.
53. P. Bruzzone, et al. Conductor fabrication for the ITER model coils, IEEE Trans. Magn., 32: 2300, 1996.
54. N. Aoki, et al. Fabrication of superconductor for the DPC-TJ coil, Cryogenics, 33: 581, 1993.
55. K. Nakamoto, et al. Design and fabrication of forced-flow superconducting poloidal coils for the Large Helical Device,
Fusion Technology 1994, Amsterdam: Elsevier, 1995, p. 909.
56. R. Heller, Superconductor for the coils of the modular stellarator Wendelstein 7-X, IEEE, Trans. Magn., 30: 2383, 1994.
57. J. V. Minervini, et al. Conductor design for the GEM detector magnet, Proc. IISSC, San Francisco, 1993, Vol. 5, p. 595.
58. T. Kumano, et al. Development of superconductors for the DEMO poloidal coils (DPC-U1, U2), Proc. Magn. Technol.
Conf., Tsukuba, 1989, Vol. II, p. 841.
59. J. R. Miller, et al. Experience on sheathing 10 km of cable-in-conduit conductor for the NHMFL hybrid, Adv. Cryog.
Eng. Mater., 41: 489, 1996.
60. D. Bessette, et al. Fabrication and test results of the 40 kA CEA conductors for NET/ITER, Fusion Technology 1992,
Amsterdam: Elsevier, 1993, p. 788.
61. D. L. Walker, et al. Design of a 200 kA conductor for superconducting magnetic energy storage (SMES), Adv. Cryog.
Eng., 35: 573, 1990.
62. A. Anghel, et al. The ITER quench experiment on long length at the SULTAN facility, Fusion Technology 1994 Amsterdam: Eberier, 1995, p. 881.
63. B. Z. Li, et al. Conductor fabrication for the HT-7U model coil, Fusion Technology 1998, Amsterdam: Elsevier, 1998, p.
775.
64. S. Forster, U. Jeske, A. Nyilas, Fabrication of a 15 kA NbTi cable for the 150 T/s high ramp rate polo model coil, Fusion
Technology 1988, Amsterdam: Elsevier, 1989, p. 1557.
65. R. Heller, et al. Stability of a poloidal field coil under rapidly changing magnetic field, IEEE Trans. Magn., 32: 2336,
1996.
66. M. A. Janocko, Lattice braided superconductors, IEEE Trans. Magn., 15: 797, 1979.
67. P. Bruzzone, Fully transposed braids for the prototype cable-in-conduit conductors of NET, IEEE Trans. Magn., 28:
190, 1992.
68. P. Bruzzone, N. Mitchell, J. Kubler,

Mechanical behavior under transversal load of the 40 kA Nb3 Sn cable-in-conduit
conductor for the NET inner poloidal coil, Proc. Magn. Technol. Conf., Tsukuba, 1989, Vol. 11, p. 926.
69. C. R. Walters, et al. Quench transients in internally cooled conductors, IEEE Trans. Magn., 19: 680, 1983.
70. M. Shimada S. Tone, Effect of niobium on cryogenic mechanical properties of aged stainless steel, Adv. Cryog. Eng., 34:
131, 1988.
71. J. Kubler,
H. J. Schinder, W. J. Muster, Influence of aging on the fracture toughness of cryogenic austenitic materials,
evaluated by a simple method, Adv. Cryog. Eng. Mater., 38: 191, 1992.
72. R. P. Reed, R. P. Walsh, C. N. McCowan, Effect of Nb3 Sn heat treatment on the strength and toughness of 316 LN
alloys with different carbon content, Adv. Cryog. Eng. Mater., 38: 45, 1992.
73. H. Nakajima, et al. Development of high strength austenitic stainless steel for conduit of Nb3 Al conductors, Adv. Cryog.
Eng., 42: 323, 1996.
74. R. P. Walsh, L. T. Summers, J. R. Miller, The 4 K tensile and fracture toughness properties of a modified 316LN conduit
alloy, Proc. ICEC/ICMC 96, Kitakyushu, Japan, Elsevier, 1996, p. 1891.
33
75. R. Bruzzese, et al. The cable-in-conduit Nb3 Sn conductor for the EURATOM-ENEA 12 T wind-and-react magnet, IEEE
76. A. della Corte, et al. Successful completion of the conductor manufacture for the ITER-TF model coil, Fusion Technology
77. M. M. Olmstead M. O. Hoenig, Constructing, fabricating and forming of internally cooled cabled superconductors, IEEE
Trans. Magn., 17: 922, 1981.
78. M. Nishi, et al. A 12 T forced flow type superconducting magnet, IEEE Trans. Magn., 23: 1531, 1987.
79. M. Yamaguchi, et al. Development of a 12 T forced cooling toroidal field coil, Proc. ICEC, 10: 169, 1984.
80. F. Negrini, et al. First results of the Italian national program on superconducting MHD magnets technology, IEEE
Trans. Magn., 30: 2086, 1994.
81. Y. Takahashi, et al. Development of a 30 kA cable-in-conduit conductor for pulsed poloidal coils, IEEE Trans. Magn.,
19: 386, 1983.
82. M. O. Hoenig, A. G. Montgomery, S. J. Waldman, Experimental evaluation of a 1 m scale D-shaped test coil fabricated
from a 23 m length of internally cooled, cabled superconductor, Adv. Cryog. Eng., 25: 251, 1979.
Wendelstein VII-X, ABB Rep. HIM 20442, 1989.

83. Machbarkeitsstudie uber
einen Supraleiter fur
84. H. Kronhardt, O. Dormicchi, J. Sapper, Design and manufacture of a Wendelstein 7-X demonstration coil, Fusion
Technology 1998, Amsterdam: Elsevier, 1998, p. 735.
85. N. N. Martovetsky, et al. GEM detector conductor manufacturing experience, IEEE Trans. Appl. Supercond., 5: 761,
1995.
86. R. E. Gold, et al. Evaluation of conductor sheath alloys for a forced flow Nb3 Sn superconducting magnet coil for the
large coil program, Adv. Cryog. Eng. Mater., 28: 759, 1982.
87. I. S. Hwang, et al. Mechanical properties of Incoloy 908an update. Adv. Cryog. Eng. Mater., 38: 1, 1992.
88. A. Bussiba, R. L. Tobler, J. R. Berger, Superconductor conduits: Fatigue crack growth rate and near threshold behavior
of three alloys. Adv. Cryog. Eng. Mater., 38: 167, 1992.
89. W. Specking, J. L. Duchateau, P. Decool, First results of strain effects on critical current of Incoloy jacketed Nb3 Sn
CICCs, Proc. Magn. Technol. Conf., Beijing, 1997, Science Press, 1998, Vol. 15, p. 1210.
90. J. S. Smith, J. H. Weber, H. W. Sizek, Control of stress-accelerated oxygen assisted cracking of Incoloy alloy 908 sheath
for Nb3 Sn cable-in-conduit, Adv. Cryog. Eng., 42: 407, 1996.
91. N. Mitchell, et al. Avoidance of SAGBO in IncoloyR 908 used as a jacket material for Nb3 Sn conductors, Proc. Magn.
Technol. Conf., Beijing, 1997, Science Press, 1998, Vol. 15, p. 1163.
92. J. H. Weber J. M. Poole, Production and properties of IncoloyR alloy 908 tubing for sheathing of Nb3 Sn superconducting
cables, Adv. Cryog. Eng., 42: 383, 1996.
93. H. Nakajima, et al. 4 K mechanical properties of pure titanium for the jacket of Nb3 Sn superconductors, Proc.
ICEC/ICMC 96, Kitakyushu, Japan, Elsevier, 1996, p. 1895.
94. P. Sanger, et al. The trials and tribulations of fabricating the pipe for the rope-in-pipe Nb3 Sn superconductor, Adv.
Cryog. Eng., 28: 751, 1982.
95. N. Koizumi, et al. Design of the Nb3 Al insert to be tested in ITER central solenoid model coil, IEEE Trans. Magn., 32:
2236, 1996.
96. V. Sytnikov, et al. RF jacketing line for manufacturing ITER cable-in-conduit conductor, Proc. ICEC/ICMC 96, Kitakyushu, Japan, Elsevier, 1996, p. 799.
97. S. Forster, et al. Development of components for poloidal field coils within the KFK Polo project, Fusion Technology
98. R. Garre, et al. Development and manufacture of superconducting cable-in-conduit conductors, Cryogenics, 34 (Suppl.):
619, 1994.
99. M. Nishi, et al. Development of high current density, large superconducting coil for fusion machines: The DPC-TJ
program, Cryogenics, 33: 573, 1993.
PIERLUIGI BRUZZONE
Centre de Recherches en Physique des Plasmas

Copyright
HIGH-Tc SUPERCONDUCTORS, PHYSICAL STRUCTURES, AND ROLE OF

CONSTITUENTS
Since the discovery of the phenomenon of superconductivity in 1911 by H. Kamerlingh Onnes (1), superconductors have been a topic of great interest from fundamental as well as application points of view. With the
discovery of high-temperature superconductors (HTSCs) in 1986 (2), the field of superconductivity achieved
even greater interest. Since then HTSCs have moved into the realm of practical applications because they are
mostly superconducting above the boiling point of liquid nitrogen (77 K), which is cheaper than and has a much
greater cooling capacity than liquid helium. Therefore, for any application in which liquid nitrogen can replace
liquid helium, the refrigeration cost for operating superconductive devices will be about 1000 times less.
The newly discovered HTSCs are copper-oxidebased ceramics that remain superconducting near 100 K.
This temperature regime does not seem high, but it is high enough for liquid nitrogen to cool most of HTSCs
into the superconducting range. For example, the compound yttrium barium copper oxide (YBCO) has been
found to be superconducting up to 92 K (3). Additional important high-temperature superconductors include
bismuth strontium calcium copper oxide (BSCCO) (4), thallium barium calcium copper oxide (TBCCO) (5), and
mercury barium calcium copper oxide (HBCCO) (6). HBCCO has the highest critical temperature, T c = 135
K (7,8). Table 1 presents the chemical formula and T c values of each of these compounds. The HTSCs can be
subdivided into three classes: three-dimensional compounds based on BaBiO3 , layered copper-oxide compounds
with hole conductivity (based on La2 CuO4 , YBa2 Cu3 O7 , and compounds of Bi and Tl), and compounds based on
Nd2 CuO4 with electron conductivity. With increasing number of CuO2 planes in copper-oxide superconductors,
a certain increase in T c is observed. Its values range from T c = 36 K in La2 x Srx CuO4 with a single CuO2
plane to T c = 125 K in compounds of Tl (2223) with three CuO2 planes.
Common Features of High-Tc Superconductors

Coherence Length and London Penetration Depth. In addition to the crystallographic aspects
of the HTSC oxides, there are more fundamental parameters contributing to the anisotropic nature of these
materials. Central to the theory of superconductivity is the Cooper electron pair discovered by Cooper in 1956
(9), when he showed that it is possible for electrons to have an attractive interaction leading to a lowering
of energy of the system. One result of the pairing is that electric current can flow without resistance, giving
rise to superconductivity. The critical temperature is related to the thermal energy required to destroy the
pairing between the paired electrons. Earlier in 1950, Ginzberg and Landau showed that the spatial variation
of the particle or superelectrons carrying the resistanceless current can be described by the coherence length
. The parameter gives the length scale for the density variation of Cooper pairs, and simply put, the greater
the length, the larger the hurdle the pairs can cross without breaking up. The breaking of pairs means the
destruction of superconductivity, and so care must be taken to keep the material free of features that break
these pairs. The other important length in superconductivity is the London penetration depth L , which sets
the scale for the decay of an external magnetic field as a function of depth in the superconducting material.
1
HIGH-TC SUPERCONDUCTORS, PHYSICAL STRUCTURES, AND ROLE OF CONSTITUENTS
There is a strong parallel between the skin depth of metals in response to high-frequency fields and the
London depth in response to magnetic fields. In 1935, the London brothers first introduced the concept of
penetration depth while following a phenomenological approach to superconductivity via Maxwells equations
for electromagnetism (10).
Ginzberg and Landau (GL) showed that the ratio of = /L determined the nature of the response
of the superconductor to a magnetic field. For 1/2, type-I superconductivity exists, and there exists a
critical field H c when superconductivity and diamagnetism are abruptly destroyed, while for 1/ , type-II
superconductivity exists, and there exists a lower critical field H c1 above which materials are not perfectly
diamagnetic and an upper critical field H c2 beyond which superconductivity and diamagnetism are completely
destroyed. All the HTSC oxide superconductors are type II and are potentially useful because they can sustain
high magnetic fields and remain superconducting.
Anisotropy. In addition to the crystallographic anisotropy, there are a number of important properties
that also show anisotropy such as electrical conductivity, thermal conductivity, magnetic field penetration,
and current density. The large variation in electrical and thermal conductivity in directions parallel to and
perpendicular to the c axis is a direct consequence of the conductivity of the CuO2 layers. Ratios of 100 and 17 for
the electrical and thermal conductivities parallel to the ab plane and perpendicular to it have been measured
for YBCO (11). These values are even higher in Bi- and Tl-based systems, indicating strong anisotropy parallel
to and perpendicular to the CuO2 layers. Anisotropy in response to magnetic fields and current density is a
result of the large variation in the coherence length with direction. For YBCO, is a few nanometers along the
ab plane and an order of magnitude less perpendicular to it. This produces a large variation in the magnetic
field and current density it can support perpendicular to and parallel to the ab plane. Similar features are
observed in the other HTSC systems.
Structures of High-T c Superconductors

The knowledge of structure is a prerequisite for understanding the properties of high-T c superconductors. The
structures of high-T c superconductors can be easily analyzed by considering the structure of a perovskite. The
original mineral called perovskite was CaTiO3 and was described in the early nineteenth century by Gustav
Rose. In addition to the importance of perovskite structures in superconductivity, dielectric, piezoelectric,
pyroelectric, and ferroelectric properties of compounds with perovskite structures have found widespread
applications in memory devices, capacitors, tunable filters, microphones, loudspeakers, etc.
The chemical formula for a perovskite structure is ABX 3 , where A and B are metal cations and X is a
nonmetal anion. X is either oxygen (O) or a member of the halogen family, such as fluorine, bromine, or chlorine.
The perovskite cell consists of a relatively large metallic atom (A) at the corners (or center), smaller metallic
atoms (B) at the center (or corners), and the nonmetallic atom (X) at face centers (or midpoint between the
Fig. 1. A schematic representation of a perovskite structure.
corners). Figure 1 illustrates the position of the various atoms on a cube. The chemical formula of ABX 3 is
easily deduced from this figure if the eight unit cells sharing the corner A atom and the two unit cells sharing
the X atoms are considered. The atom positions can be designated as A: (0,0,0) corner sites; X: (0, 12 , 12 ), (1, 21 , 12 );
( 21 , 21 ,0), ( 12 , 12 ,1); ( 21 ,0, 12 ), ( 12 ,1, 21 ) edge centers;B: ( 21 , 21 , 12 ) body center.
Another representation of the perovskite structure that is extremely useful to the understanding of HTSC
oxide structures is the layered or planar approach in which the structure is formed by stacking of layers in the
z direction. In this approach, the molecular grouping comprising each layer is also shown mainly in order to
stress the importance of the layered approach to an understanding of the properties of the HTSC oxides. The
layers themselves usually do not have any significance and may not even exist as independent structures. In
HTSC oxide terminology, the long axis is denoted by c (or z) as compared to the smaller a or b axis.
An ideal perovskite is cubic, like CaTiO3 . Its strong ionic bonding makes CaTiO3 electrically insulating
and a hard mineral. It also shows three-dimensional isotropy in many properties due to the cubic symmetry. The
crystallographic nature of anisotropy or isotropy does not necessarily determine the nature of fundamental
properties of materials such as its electrical conductivity. For instance, the element gallium exists as an
anisotropic hexagonal structure, but the momentum distribution of its electrons termed momentum space is
isotropic, and similar features could occur for superconductors. In general, there are a number of variations
from the ideal form of CaTiO3 . A typical example is the diversity in properties demonstrated by BaTiO3 . The
crystal structure of BaTiO3 undergoes structural phase transitions as a function of the ambient temperature,
changing its symmetry and response to external electric fields. Below 5 C it has an orthorhombic structure,
between 5 C and 210 C it is tetragonal, while above 210 C it is cubic. In the tetragonal state, the structural
distortion results in a permanent electric dipole moment, making it ferroelectric, while in the cubic phase
the symmetry prevents the existence of a permanent dipole moment. In noncubic perovskite minerals the
phenomenon of twinning is commonly observed. Twinning is most easily described by considering a structure
obtained by interchanging the a and b axis in a tetragonal phase, resulting in regions that are mirror images
of each other.
In addition to these common structural features, the role of the A and B cations is also crucial in the
properties of perovskites. The replacement of the smaller Ca2+ with an ionic radius of 0.99 with the larger
Ba2+ (1.34 ) plays a central role in the tetragonal distortion leading to the ferroelectric property of BaTiO3 . The
Fig. 2. Comparison of structures of La- and Nd-based superconductors. (a) Structure of La2 x Bax CuO4 with sixfoldcoordinated Cu. This is a p-type superconductor. (b) Structure of Nd2 CuO4 with planar fourfold-coordinated Cu. When
some Nd3+ ions are replaced by Ce4+ , this material becomes an n-type superconductor.
doping or substitution of A or B cations with other ions having radii within 15%, but with different valence,
results in a charge imbalance that is compensated by the creation of holes or equivalently by vacancies in
the X sites. This charge imbalance results in nonstoichiometric perovskite structures. All the HTSC oxide
superconductors contain Cu2+ instead of Ti4+ , resulting in a CuO2 layer that plays a central role in the nature
of electric current transport in these materials. Because of this feature all the HTSC oxides exhibit an almost
uniform size of a and b axes (usually between 3.81 and 3.83 ). In addition to these aspects, structural
transformations, twinning, anisotropy, etc., are all key features in controlling the superconducting properties
and need to be studied in detail to obtain a clearer picture of the behavior of these complex and potentially
useful structures.
Structure of La(Ba,Sr,Ca)CuO and Nd(Ce)CuO Compounds. Superconductivity in the highT c oxides was first discovered in the La2 CuO4 system when Bednorz and Muller 2 observed a sharp drop in the
electric resistance in Ba-doped La2 x Bax CuO4 . Subsequent observation of the Meissner effect confirmed that
the compound was superconducting. The T c was maximized at 30 K for a value of x = 0.15. The structure of the
superconducting phase was identified as a derivative of the layered perovskite K2 NiF4 and is shown in Fig. 2.
The CuO2 layers consist of octahedrally coordinated Cu, with the CuO bond length of 1.9 in the ab plane
and 2.4 in the c direction. Also, the Cu atoms occur at (000) and ( 21 21 21 x4 ,) lattice sites in the unit cell, similar
to a body-centered structure. The substituted Ba2+ ions occupy a fraction of the La3+ sites, and the difference in
valence brings about an increase in the number of holes. Superconductors with hole conduction are called p type,
a terminology similar to that followed in semiconductors. While substitution with atoms of different valence
is a well-known technique to introduce excess holes or electrons (once again in parallel with semiconductors),
oxidation or reduction to change the oxygen content also changes carrier concentration. Interstitial oxygen
between the LaO sheets or missing oxygen from the lattice sites, that is, vacancies, can also lead to excess
charge carriers. For example, the electrochemical oxidation of La2 CuO4 results in p-type superconductivity
(12). Substitution of the smaller (1.12 )Sr2+ ions instead of Ba2+ (1.34 ) results in a higher T c (39 K) for x
= 0.15. The increase in T c by substitution of a smaller atom and also the fact that increasing pressure tends
to increase T c [i.e., in the La1.8 Ba0.2 CuO4 phase, T c increases from 31 K to 36 K for a pressure increase from
atmosphere to 1.2 GPa (13)] is a significant property for a fundamental understanding of high-temperature
superconductivity and it also points the way to the synthesis of oxide superconductors with higher T c .
Fig. 3. Structural phase changes in La-based copper oxides as function of Sr concentration. (Adapted from Ref. 15.)
The structure of compounds in the Nd2 CuO4 system is identical to the compounds in the La-based
copper oxides. The only difference arises in the positions of the oxygen atoms, giving rise to an O2 layer
instead of the LaO layer present in La-based structures. While this does not change the CuO2 layers, it
results in a square planar fourfold copper coordination as seen in Fig. 2, which compares the structures of
the La- and Nd-based superconductors. The most important feature of Nd-based materials derives from the
fact that superconductivity in this system is by electroconduction or n-type rather than hole conduction as
in La-based superconductors. The difference in conductivity between the two systems, which in the undoped
state have almost identical structures and trivalent lanthanum or neodymium, is owing to the effect of doping
the substituting La with divalent Sr and the Nd with tetravalent Ce. In one case it results in a decrease
in the number of electrons (La3+ + e Sr2+ ), leading to excess holes, while in the other it increases the
number of electrons (Nd3+ + h+ Ce4+ ), leading to n-type conductivity. The electroconducting compound
Nd1.85 Ce0.15 CuO4 with T c = 24 K is the most studied material in this system (14).
Another common feature observed among the HTSC oxide families is easily observed in La2 x (B =
Ba, Sr)x CuO4 . Here, the lattice parameters and crystallographic symmetry depend on the value of x. For
x = 0.2 and B = Sr, the structure is tetragonal, while it is orthorhombic for x < 0.2. Figure 3 shows the
structural phase diagram for the La2 x Srx CuO4 system (15). Similar features can be found for B = Ba. One
aspect of the structures in La-based compounds is that superconductivity is observed both in the orthorhombic
and tetragonal phase, a feature not seen in the YBaCuO system, where only the orthorhombic phase
superconducts.
YBaCuO Compounds. The increase in T c observed by increasing pressure and by substitution of
smaller atoms led Chu and his co-workers (13) to replace La3+ (ionic radius of 1.14 ) by the smaller Y3+ (0.94 ),
leading to the first superconductor with a transition temperature (T c ) higher than the boiling point of liquid
N2 (77 K). The compound synthesized was YBa2 Cu3 O7 with a T c of 92 K. YBa2 Cu3 O7 is also commonly
referred to in the literature as Y123 or generally as YBCO and is the most widely studied of all the HTSC
oxides. Its superconducting properties, such as T c , the maximum current that it can carry (measured by the
current flowing per unit cross-section area of the superconducting material, i.e., the critical current density,
J c ), and the maximum strength of magnetic field that it could expel via the Meissner effect (H c ), are all highly
sensitive to the crystal structure and crystal quality of the YBCO material.
Fig. 4. Structure of Y1 Ba2 Cu3 O7 . The oxygen content () varies from 1 (rightmost figure) to 0 (leftmost figure) changing
the structure from a tetragonal insulating phase to the superconducting orthorhombic YBCO phase. The fivefold coordination of the Cu in the copper oxide plane is also shown. The chain oxygen content (in the ab plane) varies with . The
distortion of the CuO2 planes is also seen.
In terms of the perovskite structure, YBa2 Cu3 O7 is modeled by stacking three perovskite-type cells of
type BaCuO3 , YCuO2 , and BaCuO2 as shown in Fig. 4. A BaO plane separates two layers containing Cu and O.
While one contains atoms in the ratio of 1:2, that is, CuO2 , the other contains a varying quantity of O, ranging
from CuO to Cu. The two layers are commonly referred to as the CuO2 plane or CuO chain, respectively. The
quantity of oxygen in the chains is the key to properties of YBa2 Cu3 O7 and is represented by the value of
, known as the oxygenation parameter. For = 1, the chains do not contain O (Fig. 4), while for = 0, the
chains are fully oxygenated, giving a CuO layer. The value also determines the crystal structure of YBCO,
and Fig. 4 (left to right) shows the transition from an orthorhombic structure (a = 3.82 , b = 3.89 , c = 11.68
) for = 0 to a tetragonal structure for = 1 (a = b = 3.86 , c = 11.82 ). Significantly, in the tetragonal
phase obtained for 0.6, the material is insulating and does not have a superconducting transition, while in
the orthorhombic state obtained for 0.6, it is metallic and superconducting. T c is seen to maximize at 92
K as approaches zero. can be measured either using the titration method or from the lattice parameter of
YBCO obtained from X-ray diffraction (XRD) measurements using an empirical formula, = (c 11.68)/0.1501,
suggested by Tranquada et al. (16). Here, c is the value of the c parameter obtained from XRD measurements.
Figure 5 shows the experimentally observed correlation between the c-axis lattice parameter and T c for YBCO
in both thin-film form (17) and bulk form (18). The higher values of the c-axis lattice parameter for YBCO thin
film than those for bulk YBCO with the same T c are attributed to the expansion of the unit cell under the
influence of the substrate material.
In many cases, it is easier to interpret the properties of the HTSC oxides by understanding the stacking
sequence of the individual CuO layers making up the unit cell. But, unlike the case of true layered compounds
such as graphite and mica, where the weak bonds between layers makes intercalation with foreign atoms easy,
this is not the case with the HTSC oxides, and so the layers must not be interpreted as such. In the case of the
Bi-based HTSC oxides, though, the binding between BiO sheets is weak and intercalation of iodine is possible.
Figures 6(a) and 6(b) compare the layering sequences in YBCO and La2 CuO4 . The main difference seems to
Fig. 5. Experimental correlation between the c-axis lattice parameter and T c for YBCO thin film form (14) and YBCO
bulk form. (Adapted from Ref. 18.)
Fig. 6. Layered approach to the structure of copper oxide superconductors. (a) The Y1 Ba2 Cu3 O7 structure with the
individual planes and plane spacings in the c direction marked. (b) Similar approach for the La-based superconductors.
(Adapted from C. P. Poole, H. A. Farach and R. J. Creswick, in Su, San Diego, CA: Academic 1995, pp. 182191.)
Fig. 7. Twinning in Y1 Ba2 Cu3 O7 . The figure shows a transmission electron microscope (TEM) image of a YBCO film
grown on MgO by pulsed laser deposition. The striations correspond to the twin boundaries.
be in the form of the position of the individual Cu atoms. For the case of YBCO, all the Cu atoms are of one
type; that is, edge type or all centered (depending upon the representation), while for the La-based structure,
they occupy (000) and ( 12 12 12 ) positions. The Y layer acts as a spacer for the two CuO2 planes in YBCO. Also
unlike the case of LaSrCuO, the atoms on the CuO2 layer in YBCO do not all lie on the same plane, that is,
their z values differ slightly and so they are actually CuO2 sheets as compared with CuO2 planes. This slight
distortion is visible in Figs. 4 and 6(a). Heating YBCO through 350 C transforms the orthorhombic structure
to the tetragonal phase owing to the loss of oxygen. This displacing transition is the reason for twinning and is
usually seen when the strain accompanying the phase change cannot be completely released, as in the case of
crystalline thin films deposited on substrates. Thus, YBCO thin films undergo twinning when grown at high
temperatures and subsequently cooled through 350 C in an oxygen ambient. Figure 7 shows a typical YBCO
film showing twinning observed under an electron microscope.
In addition to YBCO, the YBaCuO family also contains other phases of which YBa2 Cu4 O8 (Y124) has
a T c of 80 K and Y2 Ba4 Cu7 O15 (Y247) has a T c of 55 K. These two phases are not observed under normal
processing conditions and require extremely high oxygen pressures or the addition of alkali-metal compounds
in air at normal pressures (19). The structure of Y124 is similar to that of YBCO with the addition of an extra
CuO chain layer displaced with respect to the first by a lattice parameter of b/2 (where the b axis contains the
oxygen of the chains). Y124 is orthorhombic with a = 3.84 , b = 3.87 , c = 27.2 . The structure of Y247
Fig. 8. Crystal structure of the Tl-(Bi-) based superconductors. Changing the number of Cu atoms introduces copper oxide
layers ( n). The structure of the Bi-based compounds is similar, with the Tl positions occupied by Bi. (Adapted from Ref. 21.)
can be represented by the intergrowth of alternating layers of YBCO and Y124, resulting in an orthorhombic
structure of a = 3.85 , b = 3.87 , c = 50.3 . As compared to YBCO, Y124 has a fixed oxygen stoichiometry
and so does not undergo the displacive transformation resulting in the twinning morphology.
Bi-, Tl-, and Hg-Based Compounds. In early 1988, two new superconducting systems based on
bismuth and the toxic metal thallium with transition temperatures considerably higher than YBCO were discovered (20,21). These two systems have a and b lattice parameters similar to the Y- and La-based compounds,
but with a much larger c parameter. The general formula for compounds occurring in this system is given by
A2 B2 Can 1 Cun O2n+4 , where A = Bi or Tl, B = Sr or Ba, and n = 1 to 4. The compounds in these two systems have
identical structures and differ only in atom positions, so it is convenient to deal with their structures together.
The highest T c recorded so far in these systems are 123 K for n = 3 in the Tl system, giving Tl2 Sr2 CaCu3 O10 ,
also referred to as Tl2223, and 110 K for the structurally similar Bi2223. The toxicity of the Tl oxides has
prevented extensive characterization of these compounds as compared to the Bi structures. The crystal structures (Fig. 8) are similar to those of YBCO, but the CuO chain layers are instead replaced by a double layer
of TlO. The absence of chains makes the Bi- and Tl-based oxides less susceptible to the stoichiometry of oxygen, though nonstoichiometry is commonly present in these systems due frequently to Ca(Sr) or Cu depletion.
The multiplicity of phases is a major problem in the synthesis of homogeneous material in these systems.
The Cu coordination is fivefold square pyramidal with O as for YBCO (Fig. 4). But for n = 3, the middle
CuO layer contains Cu, which is planar fourfold coordinated. The layer in these systems consists of CuO2
CaCuO2 layers sandwiched between alkaline-earth-metal oxide B(Sr or Ba)O layers and by a double layer
of A(Tl or Bi)O. An interesting feature shown by these compounds is the variation of T c with the number of
CuO2 planes, n. The T c is found to increase with increasing n, maximizing at n = 3. While this confirms the
widely accepted fact that the CuO2 layers play a vital role in superconductivity, for example, the structures
with n = 0 do not superconduct; increasing the number of layers indefinitely does not necessarily increase T c .
10
In addition to these stoichiometric compounds, long sintering times can result in the evaporation of part
of an AO layer, resulting in a monolayer of TlO or BiO, giving nonstoichiometric oxides. T c s in the Tl-based
compounds vary from 10 K to 121 K depending on the number of copper oxide planes. The family of systems
given by the formula A2 B2 Can 1 Cun O2n+4 or equivalently by A2 B2 Can Cun+1 O2n+6 (here n does not correspond
to the number of CuO2 layers) can be extended to other cations such as mercury (Hg), which forms a series of
compounds (with A2 replaced by Hg and B by Ba) with T c s ranging from 95 K to 133 K, again maximizing for
three CuO2 layers.
Role of Microstructures in High-T c Superconductors

After the initial excitement and directions towards large-scale technological applications of the HTSC oxides
following their discovery in 1986, scientists have realized the difficulty in putting these ceramic oxides to use
because they are hard, brittle, and difficult to draw into sheets or wires. The major emphasis in materials
science and technology is now on developing the processing and fabrication of thin films for devices. Their novel
properties, such as the ability to shield closed volumes from electromagnetic interferences via the Meissner
effect, the low microwave surface resistance for use in filters for cellular base stations, the Josephson effect in
superconducting quantum interference devices (SQUIDs), and the sharp transition in the use of bolometers to
detect infrared radiation are in stages of development and use. One exciting application is the use in microelectronics as interconnectsthe reduction in Joule heating due to resistanceless current flow will increase the
density of devices per chip and the efficiency of packaging. The improvement in RC time constants, increasing
the speed of devices, also makes it attractive to consider the integration between superconductors and semiconductors to operate at liquid-nitrogen temperatures. In fact, there are a number of transistors that improve
performance when cooled to 77 K.
However, there are a number of problems to be solvedsome, such as the magnitude of electric current
that HTSC oxides carry and the erratic behavior of weak links in Josephson junctions across grain boundaries,
are functions of the Cooper pair coherence length and material quality, while others, such as the degradation
on exposure to moisture of YBCO and the formation of ohmic contacts to these oxides, can be solved by proper
combination of materials and processing conditions. Thus the material quality or microstructure of thin films
seems to be the most important aspect of thin-film study of HTSC superconductors. This leads us to the large
influence of grain boundaries on the current-carrying ability of HTSC oxides. While the discussion here pertains
directly to YBCO, its extension to other HTSC oxides seems highly likely owing to the central role played by
coherence length. As mentioned earlier, the coherence length in oxide superconductors is on the order of a few
nanometers, meaning that any imperfection in the material with an influence scaling across dimensions of the
same magnitude will break Cooper pairs, mainly because they are formed owing to certain properties exhibited
by the perfect crystal structure.
Figure 9 shows an electron micrograph of a grain boundary occurring in a YBCO film with a misorientation
between the grains being quite large (28 ). It is now well accepted that when the Cooper pairs strike the
boundary at an angle, some pass through unbroken, while others break and become nonsuperconducting,
resulting in a reduction in the current that emerges on the other side of the boundary. In HTSC terminology,
the processes are referred to as pair tunneling and pair breaking, respectively. The grain boundary is said
to constitute a weak link as a result of this process. Actually, there are two subtle processes involved across
boundaries in the HTSC oxides: the weak-link behavior is a result of pair tunneling across an insulating
region separating the two superconducting regions or semiconductorinsulatorsemiconductor tunneling, as
it is referred to, while the other is the proximity effect, and occurs across a metallic or normal boundary, that
is, a superconductornormal conductorsuperconductor structure. These two processes have characteristic
currentvoltage behavior and are easily discernible in classical superconductors where the coherence lengths
are large, almost 10 nm to 100 nm. In HTSC oxides, these two processes cannot be differentiated because they
occur on length scales of a few nanometers, the same as the coherence length.
11
Fig. 9. Large-angle (35 ) grain boundary in Y1 Ba2 Cu3 O7 . This figure shows a high-resolution TEM image of a typical
high-angle pure (001) tilt grain boundary in YBCO.
The significant role of grain boundaries has been emphasized by some lucid experimental demonstrations
like those by researchers at IBM (22,23), where they showed that not only does the critical current density (J c )
of the material decrease due to the presence of the boundary, but the J c also depends on the misorientation
angle of the boundary. Figure 10 is a plot of the experimental result of J c across grain boundaries in YBCO
films grown by pulsed laser deposition (PLD) (24,25). The plot clearly shows that increasing the misorientation
decreases the J c . This has important implications for devices. High-current-carrying films need boundaries with
extremely small misorientations or better still no boundaries at all. To explain the reason for this behavior
a number of researchers (23,26,27) have invoked the dislocation structure of low-angle grain boundaries first
proposed by Read and Schockley in 1950 (28). Figure 11 demonstrates this idea in a low-angle boundary in a
12
Fig. 10. Experimentally observed variation of the critical current density ( J c ) with boundary misorientation (). (Adapted
Ref. 23.)
Fig. 11. Low-angle (6 ) grain boundary in Y1 Ba2 Cu3 O7 showing the dislocation array. This figure shows a high-resolution
TEM image of a low-angle pure (001) tilt boundary in a YBCO film grown by pulsed laser deposition. The boundary
misorientation is generated by the equally spaced [100] edge dislocations.
13
Fig. 12. The variation of dislocation spacing with boundary misorientation (). The geometrically calculated variation in
spacing parallels the experimental observation of J c with misorientation shown in Fig. 10.
Fig. 13. Dislocation core. This high-resolution TEM image shows the highly distorted structure of the YBCO lattice
around a [100] edge dislocation.
YBCO film grown by PLD. The almost regular separation of the dislocations marked in the figure generates the
boundary misorientation, and in fact a plot of the dislocation separation versus misorientation (Fig. 12) shows
a trend similar to that followed by the J c in Fig. 10. This similarity was an irresistible attraction to model
the current flow as a strong function of the grain-boundary features arising from the dislocation arrays. This
includes the finite region along the boundary where the lattice was highly distorted and so superconductivity
can be destroyed (29). This is usually observed near or around dislocations like that shown in Fig. 13.
The long-range strain fields are due to the dislocations (30), that distort the crystal structure and affect superconductivity by changing the order parameter (31). The order parameter is the complex quantummechanical wave function used to describe the density of Cooper pairs as a function of space, similar to a wave
function describing the electron density in a metal, for example, the ability of Cooper pairs to tunnel across
14
small distortions that are present along the boundary surface due to strains (32). This feature is a function of
the boundary width, which describes the region along the boundary in which current flows via tunneling rather
than normal current. This parameter is experimentally difficult to observe and various approximations such as
the strength of the Burgers vector or via energy minimization (33) are used to calculate it. Another important
influence of grain boundaries is the effect of nonstoichiometry commonly observed along the boundary (34),
leading to excess charge and subsequently scattering of the Cooper pairs.
Jagannadnam and Narayan have proposed an elegant model to represent the J c across low-angle boundaries that incorporates all the features mentioned before. While the previous calculations can be done fairly simply, the actual nature of the grain boundaries in HTSC materials needs to be understood fully. It has been shown
by various authors that there is a large variation in grain-boundary microstructure even in films prepared under identical conditions (35,36). Recently, it has been seen that even boundaries with a similar misorientation angle along identical crystallographic directions can have different atomic scale structures (37,38).
Compared with strong variation in the low-angle region of 0 to 10 observed in Fig. 10, the J c in the highangle region of 10 or more is uniformly low. An early explanation of the IBM group for this saturation effect was
the overlapping of dislocations as the boundary misorientation increased. While the notion of the dislocation
structure boundaries loses significance at higher angles, it is difficult to even model the current in the high-angle
regime for the lack of a uniform description of the boundaries occurring there. Various representations such as
the near-coincidence site lattice model (near CSL) or more recently the grain-boundary structure representation in terms of distinct structural units (38,39) are being used to understand the properties of the high-angle
boundaries. It has also been difficult to interpret exceptions to the low J c in high-angle cases, especially for the
90 twist boundary, formed by rotating one unit cell of YBCO 90 with respect to an adjoining cell with the rotation axis being perpendicular to the c axis or [001] direction (40,41). Here the absence of the weak-link behavior
is a strong suggestion for the fact that high J c s are associated with boundaries possessing low energy, also
usually accompanied by high symmetry, which is true for the 90 twist boundary. In this light the predominance
of asymmetry in grain boundaries in YBCO grown by PLD is an important feature and must be understood
more clearly. All these factors lead one to realize that the exact nature of the effect of grain boundaries
is far from being fully comprehended and detailed research must be continued for a better understanding.
Role of Substituents in High-T c Superconductors

The simple copper oxide pervoskites are insulators. By substituting for certain atoms in the unit cell, these
materials are made to behave as metals and hence possibly as superconductors. The key characteristic of
any superconductor is an energy gap exactly at the Fermi level. Owing to this characteristic, Cooper pairs of
electrons do not scatter off the lattice site and break. In the absence of scattering, the propagation of electrons
encounters no resistance and the phenomenon of superconductivity is observed. The transition temperature
depends strongly on the electronic density of states at the Fermi level. That parameter in turn, is strongly
affected by doping of the initial ceramic with other atoms of a different valence to provide extra electrons (or
too few electrons, called holes), which are then available for participation in the superconducting mechanism.
The practice of doping has a long history in the field of semiconductors, and chemists have developed
great skills in modifying the properties of materials to allow practical electronic devices to be made. It is not
surprising to see similar efforts applied to the HTSCs. A tremendous array of substitutions have been tried
to improve the mechanical, magnetic, or current-carrying properties of these materials. The complexities of
multicomponent phase diagrams prevent some attempted compounds from forming at all. Of those that can be
made, either in chemical equilibrium or via a narrow path of kinetics, success is not guaranteed because it is
hard to change the electronic properties precisely at the Fermi level. Moreover, the Fermi level is not one single
number, but rather there is a Fermi surface in energy space, which is badly distorted from a simple spherical
shape by the anisotropy present in the HTSC compounds. Consequently, there is considerable empiricism and
guesswork present in any particular choice of chemical substitution in the HTSCs.
15
Fig. 14. (a) T c versus oxygen content in YBCO. (b) Bond valence sum around the copper atom in CuO(2) layer. The purpose
of showing the two figures together is to illustrate that T c is proportional to the oxidation state of the copper atoms in the
conduction layer.
Role of Oxygen and Oxygen Vacancies. The role of oxygen in influencing the electronic properties
of HTSCs is very crucial. Normally, an oxygen atom takes on two electrons from another atom. Therefore, if it
is absent, two more electrons are free to go elsewhere in the crystal. This is how vacancies affect the charge
balance in a crystal. The first thing that oxygen vacancies do is change the number of free carriers available in
the crystal lattice, which in turn adjusts the Fermi level slightly. The density of states at the Fermi level ( N 0 )
is a key parameter of superconductors: the transition temperature T c depends on N 0 as exp(1/N 0 V), where V
is the Cooper pairing potential. Small changes in N 0 caused by oxygen vacancies can translate into substantial
changes in T c . It does not take much to change N 0 appreciably. There are about 3.5 1013 vacancies/cm2 in a
double-layer compound such as BSCCO, which corresponds to having about 1% of the oxygen atoms missing.
Explaining the role of oxygen vacancies depends upon the charge-transfer model, by which the electrons
normally in the CuO2 planes are transferred to sites elsewhere in the unit cell (42). In YBCO, the oxygen
deficiency shows up especially in CuO chains, making their formula CuO1 , while the CuO2 layers remain
chemically complete. A key experimental quantity in this model is the oxidation state of copper atoms in the
CuO2 planes. Any deviation from 2.0 indicates that charge transfer is occurring. Because of the mild geometric
distortion in the unit cell, the copperoxygen bonds are stretched slightly in this plane. By measuring the bond
lengths around that copper atom, a bond valance sum can be calculated, and this gives the oxidation state (43).
As the number of oxygen vacancies varies, both this sum and the superconducting transition temperature T c
vary in the pattern shown in Fig. 14. This presents a very convincing argument for the charge-transfer model.
16
In TBCCO, there is competition between several different means of causing charge transfer. Those means
may be missing metal atoms, extra metal atoms, missing oxygens, extra oxygens, etc. BSCCO exhibits all these
conditions, plus the size mismatch between the normal dimensions of the CuO(2) planes and the other planes in
the stack is so large that atoms occasionally wind up in the wrong layer. The presence of many slightly different
but similar structures in one crystal makes it difficult to interpret experimental data. Nevertheless, despite
all the complexity, the average oxidation state for copper is 2.21, which tends to confirm the charge-transfer
model.
It should also be noted that an excess of oxygen atoms also acts as a dopant, because it increases
the number of sites where electrons can reside, which amounts to creating hole carriers in the crystal. The
compound La2 CuO4 is normally an insulator, but when extra oxygen is forced in, changing it to La2 CuO4+d , it
becomes a superconductor. Thus, it appears that a small number of oxygen vacancies offer a way to make small
changes in the carrier concentration. Oxygen vacancies are equivalent to the substitution of metal atoms in
terms of their effect on the availability of charge carriers.
Role of Substituents in La(Ba,Sr,Ca)CuO and NdCeCuO Compounds. Substituting for
La3+ by divalent ions M = Ba2+ , Sr2+ , Ca2+ leads to the appearance of metallic conductivity and for 0.05
< x < 0.3 superconductivity appears in La2 x Mx CuO4 y (LMCO) compounds. T c (x) is plotted in Fig. 15(a),
where the black dots correspond to the appearance of the Meissner signal, and the white dots to its saturated
value (44). For larger concentrations of Sr, x > 0.15, the oxygen vacancies can appear, y > 0; their formation,
however, can be suppressed by performing the annealing under high oxygen pressure. Metallic conductivity
and superconductivity may also be obtained by increasing the content of oxygen above the stoichiometric
value La2 CuO4+d , d > 0. In this case, oxygen occupies interstitial positions in layers of LaO. In general, for
La2 x Mx CuO4 y one observes the typical dependance T c (n), which is shown in Fig. 15(b), with x substituted
by n = x2y, the number of free charge carriers (holes) per cell in the CuO2 planes.
Under the substitution of La by the trivalent rare-earth ions RE = Nd, Sm, or Gd, a smooth decrease of
T c with decreasing ionic radius in the series of these ions (45) occurs. At the same time, the value of magnetic
moment of RE ion has no effect on T c , which indicates that the Cooper pairs in the CuO2 plane are weakly
coupled to the magnetic moments of ions in the LaO layers. An analogous situation is observed for electronic
superconductors Nd2 x Cex CuO4 under the substitution of Nd by the rare-earth ions Pr, Sm, and Eu. The
decrease of T c with decreasing radii of RE ions and primitive cell volume may be related to the decrease of
CuO bond length.
However, this simple picture contradicts the dependence of T c on the external pressure. In the LaSr
compounds T c increases with an increase of pressure, attaining a certain maximum, and even decreases at
high pressures. Meanwhile, in NdCe compounds T c is pressure independent. This difference in the T c (p)
dependence can be related to a special role of apex oxygen: In the tetragonal phase of LaSr compounds there
are two apex oxygens in the complete CuO6 octahedron, and in the tetragonal phase of NdCe compounds the
apex oxygen is absent.
An isovalence substitution of copper by the ions of 3d metals has a much stronger effect on T c . At a
concentration of x = 5% to 7% for Ni and Fe and x = 2% to 3% for Zn ions, the superconductivity in LMCO
disappears (45,46,47). While the disruption of superconductivity due to scattering of the Cooper pairs on
magnetic impurities is well known, the suppression of T c due to Zn impurities in CuO planes is characteristic
of copper-oxide superconductors. Figure 16 shows the T c (x) dependence for the Zn impurity in La2 x Srx CuO4 y
LSCO and YBCO (47). It is possible that Zn ions, which have a filled 3d shell (3d10 ), destroy a rather complicated
correlation conduction band in CuO2 planes that is related to CuO charge transfer as shown in Fig. 17 (48).
This leads to suppression of superconductivity. In this context, one can consider Zn2+ ions as effective magnetic
scatterers, because they substitute Cu2+ ions, which have a local magnetic moment.
Role of Substituents in YBCO Compounds. A large number of studies have been carried out to
investigate the role of constituents in YBCO compounds. Immediately after superconductivity with T c = 90
K in the YBCO system was observed, a large class of RBCO compounds with a similar T c was prepared. This
17
Fig. 15. (a) T c as a function of Sr concentration in ( La1 x Srx )2 CuO4 [5.13] and the general dependence T c (n) on the
concentration of holes n in the CuO2 planes for some copper-oxide superconductors. (After Tiokura, 1992.)
was done by substituting Y by lanthanides Ln = La, Nd, Sm, Eu, Gd, Dy, Ho, Er, Tm, Yb, or Lu. For these no
separation of T c in the compounds with rare-earth ions having a large magnetic moment has been observed.
This indicates their weak coupling with the in-plane holes. The exceptions are Ce and Er ions, whose formal
valence is +4, which violates the isovalence of the substitution of Y ions and can change the concentration
of carriers in CuO2 planes. In order to investigate this concept, the system (Y1 x y Cay ) PrxBa2 Cu3 O7 has
18
Fig. 16. T c as a function of impurity x in a La1.8 Sr0.2 Cu1 x Znx O4 y compound. For comparison the variation of T c in
YBa2 Cu1 x Znx O7 is also shown as a function of x.
Fig. 17. Schematic of electronic structure of copper oxide superconductors accounting for (a) the correlation splitting of a
pd band, and some models for its change upon p or n doping: (b) rigid band, (c) impurity band, (d) filled gap, and (e) local
density approximation. [48]
been examined (49). The T c (y) dependence for different x and T c (x) at y = 0 is shown in Figs. 18(a) and 18(b),
respectively. The substitution of Pr4+ for Y decreases the number of in-plane holes, whereas substitution of Ca2+
has the opposite effect. This allows one to investigate separately the dependence of T c on the concentration
of holes n (yx) and on the magnetic scattering on the localized moments of Pr. Indeed, it follows from
Fig. 18(a) that T c (y) has a typical hole dependence with a maximum at an optimal concentration of holes n for
various concentrations x of Pr ions. The location of these maxima shifts to larger values of y with an increase
of Pr content, which proves directly that the concentration of carriers n (yx) decreases with an increase of
x. At the same time, a decrease of the maximum value of T c (x,y) for the optimal n is observed, which indicates
19
Fig. 18. The dependence of T c in Y1 x y Cay Prx Ba2 Cu3 O6.95 (a) on the concentration of Ca at a fixed content of Pr and (b)
on the concentration of Pr for y = 0. The dotted line is the T c dependence on the concentration of holes without magnetic
scattering. [After Neumeier et al. 49]
a decrease of T c due to magnetic scattering leading to the destruction of Cooper pairs. Figure 18(b) shows
the effect of a change of the number of carriers on Tc(x,y = 0). One can see that the maximum value of T c in
YBCO at 7 = 6.95 is attained under a small decrease of the number of in-plane holes due to the tetravalent
impurities. In the course of investigations of the effect of impurity substitution on T c , one must control the
content of oxygen as it influences the number of carriers in CuO2 planes and determines the value of T c .
An investigation of the effect of substitution of Ba in YBCO by the rare-earth ions has been carried out
for the Ln(Ba2 x Lnx )Cu3 O7 compounds, where Ln = La, Nd, Sm, Eu, Gd. All the lanthanides produce an
equal decrease in T c with respect to the concentration of impurities x. This indicates a weak sensitivity of
the superconducting transition to the appearance of magnetic moments on the Ba sites. A phase transition
is also observed from orthorhombic to tetragonal (at x = 0.2 to 0.3), which, however, did not have significant
effect on the superconducting properties. In view of an often uncontrolled increase of oxygen content 7 > 7
under the substitution of Ln for Ba and a complicated rearrangement of charge in the layers BaO(4), Cu(1)O,
Cu(2)O(2,3), it is not possible to unambiguously conclude regarding the dependence of T c on the concentration
of holes in CuO2 planes.
In the YBCO compounds a more complicated effect of the substitution for copper ions has been observed.
The complicated effect arises due to two reasons. First, in YBCO compounds there are two nonequivalent
copper positions, whose substitution has a different effect on their electronic structure and superconductivity.
Second, some impurities such as Fe and Co affect the oxygen content and the short-range order in Cu(1)O(1)
layer, which may change the number of carriers in the Cu(2)O(2,3) planes.
It is interesting to study the effect of substituting Cu by ions of 3d elements M 2+ = Ti(3d2 ), Cr(3d4 ),
Mn(3d5 ), Fe(3d6 ), Co(3d7 ), Ni(3d8 ), and Zn(3d10 ). It has been observed that for x = 0.1 in YBa2 (Cu1 x Mx )3 O7
the strongest suppression of T c occurs for the ions of Fe and Co, having the maximum magnetic moment, and
also for the Zn ions (50). However, some investigations have demonstrated that in case of Fe and Co an extra
20
Fig. 19. The dependence of lattice parameters and T c on the concentration of impurities in YBa2 (Cu1 x Mx )O7 for (a)
Zn, and (b) Ga.
annealing in an oxygen atmosphere restores the value of T c for x less than or equal to 0.35 almost completely,
but does not relax the decrease of T c by the impurities of Zn and Ni (51). Such different effects of annealing on
T c in impurity compounds of YBCO is explained by the fact that at low concentrations the ions of Co and Fe
preferentially occupy the position of Cu(1), and the ions of Zn and Ni the position of Cu(2). Therefore, annealing
in oxygen restores the oxygen content and its coordination in the Cu(1)O(1) layer for the Co and Fe impurities,
but has no effect on the plane Cu(2)O(2,3) for the Zn and Ni impurities.
A more detailed analysis of the effect of diamagnetic impurities under a substitution for copper is also
available in literature (52). In that study Zn2+ (3d10 ) and Ga3+ (3d10 ) impurities were used. Their ionic radii for
a filled3d shell, R(Zn2+ ) = 0.75 , R(Ga3+ ) = 0.62 are close to R(Cu2+ ) = 0.73 , which allows one to obtain
single-phase samples in a wide range of impurity concentrations. Neutron diffraction measurements have
indicated that Zn primarily occupies the position of Cu(2), and Ga the chain positions of Cu(1). The dependence
of T c and lattice constants are shown in Fig. 19(a) for Zn and Fig. 19(b) for Ga. Zn impurities in the layers
Cu(2)O(2,3) have only a small effect on the parameters of the lattice, preserving the orthorhombic phase, but
lead to a rapid suppression of T c . The oxygen content remains close to optimal (6.8 and 7.0 for samples with
Zn and Ga, respectively), and therefore the transition to the tetragonal phase in the case of Ga is not related
21
Fig. 20. T c versus the concentration of impurities M = Ni2+ and Zn2+ and M = Al3+ , Fe3+ and Co3+ in YBa2 Cu3 z Mz O7 .
[After Taraskon et al. 53
to the deficiency of oxygen. The resistance (x) in the normal phase increases much faster for Ga impurities
than that for Zn, while they have the converse T c (x) dependence. These experiments clearly demonstrate that
the main role in the appearance of superconductivity in copper-oxide compounds is played by the CuO2 planes,
where the specific properties of copper in 3d9 states are necessary to attain high T c .
According to studies of structural, magnetic and superconducting properties of YBa2 Cu3 z Mz O7 compounds, paramagnetic and diamagnetic impurities affect these properties differently (49). Here M = Ni,Fe,Co
represent paramagnetic impurities and M = Zn,Al represent diamagnetic impurities. It has been observed
that the divalent impurities Zn2+ and Ni2+ do not change the content in the sample and, when occupying
positions in CuO2 planes, preserve the orthorhombic phase in the domain of single-phase states (for z 0.3
or the concentration x = z/3 < 0.1). Al3+ impurities as well as Co and Fe being in the trivalent state occupy
the chain Cu(1) positions. They lead to an increase of oxygen content, filling the vacancies near the impurity
site and leading to a transition to the tetragonal phase for z < 0.1. The dependence of T c (z) for these impurity
substitutions is shown in Fig. 20. It is clear from this figure that substituting in-plane Cu(2) by Zn impurities
leads to a rapid decrease of T c while substituting the in-chain Cu(1) by trivalent impurities Al, Fe, Co has
a much weaker effect on T c . Apparently, both substitution of impurities for in-chain Cu(1) and changing the
oxygen content have the same effect on the charge transfer from the chains to the planes, causing the transition
from insulating to metallic phase which becomes superconducting. Note that in a series of experiments a more
complicated dependence of T c on the concentration of impurities that substitute for copper in YBCO compounds
has also been observed (54,55).
There are certain elements in the periodic table that have been shown to be unreactive with YBCO
materials (56,57). When such elements are added to YBCO compound, they do not substitute any of the
elements in the lattice and tend to reside in intergranular regions (58). However, they have been found to affect
significantly the superconducting properties of the YBCO material by modifying the grain-boundary properties.
Silver is one such element that has most widely been used as a dopant to modify the grain-boundary properties
of YBCO system (59). Shown in Fig. 21 is the variation of T c and J c of YBCO films doped with different amount of
silver. All the films were deposited in situ on (100)LaAlO3 substrates using a pulsed laser deposition techniques.
Different levels of doping in YBCO films were obtained by using targets containing different concentration of
silver metal. Figure 21 shows that there is a one-to-one correlation between the resistivity and critical current
22
Fig. 21. Variation of room-temperature resistivity and critical current density ( J c ) at 77 K of films obtained from YBCO
targets having different dopant level of Ag. [After Kumar et al. (58)
densities. This correlation is due to weak-link coupling in which the critical current density of weak links
should increase with lowered resistivity (60). Figure 21 also shows that films made using targets having 5% of
silver gives highest J c and the lowest room-temperature resistivity. This can be explained on the basis of the
fact that a lower percentage of silver in the target may not be able to provide sufficient silver atoms required to
enhance grain growth and liberate oxygen at the periphery of the grains of films, whereas a higher percentage
of silver in the target would result in higher doping of YBCO films. Higher levels of silver in the targets and
films may give rise to the formation of some impurity phases. According to Kao et al. (61), it is also possible
that beyond the optimum Ag content, Ag atoms may enter the CuO basal panes in the orthorhombic structure
and cause deterioration in T c and J c of the sample.
Since the grains in the superconducting films are coupled to each other through the materials in the
grain boundaries, the presence of silver in the grain boundaries affects the properties of YBCO materials very
significantly. The materials in the boundaries are the rejects of the grains, and they may be either metallic
or insulating in nature. Depending upon the nature of the material in the grain boundary, the couplings are
known as SIS (60) or SNS (62,63) where S, I, and N stand for superconductor, insulator, and normal
metal, respectively. The temperature dependence of the critical current density as a function of temperature
gives an idea of the nature of the weak links present in superconducting samples with different amount of
grain-boundary dopants (60). If the superconducting grains are coupled to each other by an insulating phase
in the grain-boundary region, the current density of such junctions is determined by the expression suggested
by Ambegoakar and Baratoff (60),
where Rn is junction resistance and (T) is energy gap. When T is close to T c ,
23
Fig. 22. Critical current density versus temperature plots for undoped and Ag-doped YBCO films on LaAlO3 substrates
grown under identical conditions. [After Kumar et al. (1994).]
If the suppression of the order parameter is taken into account, calculation through GinzbergLandau
theory gives the following J c (T) relation for an SIS network (64):
However, if the superconducting grains are coupled to each other by normal-metal type material in the grainboundary region, the junctions are known as SNS junctions and the current density of such junctions is
given by the expression suggested by de Gennes (62) and Clarke (63),
where d is the thickness of the grain boundary layer and n is the normal-metal coherence length. Ignoring the
weak temperature dependance of n compared to the ( 1 T/T c )2 term, we obtain
or
The value of the slope of the J c versus T c T plot can be used as a figure of merit of the film, since the
larger the slope of the plot the better the quality of the film. Having identified the nature of the weak link
from the J c versus T c T plot, one can derive other useful information from this plot regarding the width of
the grain boundary in different films. From Eq. (6) it is obvious that the slope of the J c versus T c T plot is
proportional to the inverse exponent of the width of the grain boundary d. This concept was used by Kumar et
al. (65) to understand the reason for the difference in critical current densities of undoped and Ag-doped films.
They fabricated a series of YBCO films with different concentration of Ag. Plots of the temperature dependance
of the critical current density of these films are shown in Fig. 22. Determining the values of J c and T c T
24
Fig. 23. J c versus T c T plots of undoped and Ag-doped YBCO films on LaAlO3 substrates grown under identical
conditions. [After Kumar et al. 65]
from Fig. 22, the J c versus T c T plots of undoped and Ag-doped YBCO films were obtained and are shown in
Fig. 23. It is clear from this figure that both undoped and Ag-doped YBCO films follow the SNS model, but
their grain-boundary widths are different. The variation in grain-boundary widths in undoped and Ag-doped
films is as follows: d0% :d2% :d5% :d10% :d20% :: 1.0:0.6:0.5:0.65:2.0. Owing to differences in grain-boundary width,
the grain-boundary resistance and hence the strength of SNS coupling in these films are also different. In
the present example, the coupling is maximum in the case of 5 wt % Ag-doped film and is minimum in the case
of 20 wt % Ag-doped film, and therefore the former has the maximum current density and the latter has the
least.
Role of Substituents in Bi-, Tl-, and Hg-Based Compounds. The copper superconductors containing layers of bismuth oxide, thallium oxide, or mercury oxide exhibit superconductivity at higher transition
temperatures than YBCO. Because all three types of compounds have very similar structures, it is convenient
to discuss the role of dopants in these compounds together. Among the bismuth-oxide-based superconductors,
the structures of greatest interest are Bi1 Sr2 Ca1 Cu2 O10 (BSCCO-1212) and Bi2 Sr2 Ca2 Cu3 O10 (BSCCO-2223).
The most familiar substitution in bismuth-oxide-based superconductors is to replace bismuth with lead. This
has relatively minor effects on the superconducting properties, but it strongly affects the kinetics and phase
equilibrium of the mixture. The result is that (Pb,Bi)2 Sr2 Ca2 Cu3 O10 is much easier to make than BSCCO-2223
without lead. In a similar analogy with the rare-earth substitutions in YBCO, the partial replacement of Bi
with Pb makes little difference electronically, and thus changes the T c only very modestly.
In both the thallium- and bismuth-based superconductors, the strategy of doping is driven by the conjectures of each investigators about what will produce desired properties. The number of substitutions possible
on those unit cells, together with the complexity of their phase diagrams, requires verification of hypotheses
by empirical means. The familiar series of thallium superconductors are made of Tl, Ba, Ca, Cu, and O. The
transition temperatures of various phases are generally above 100 K. A less well-known series is based on
Tl-Sr-Ca-Cu-O (TSCCO) with Sr playing the equivalent role of Ba. TSCCO is nonsuperconducting. Sheng et
al. (66) have shown that the substitution of the trivalent Y for the divalent Ca brings an extra electron onto
that layer in the unit cell, leading to the conversion of material to the superconducting phase with a T c of 80
K. The goal of most substitutions is to change the number of carriers (holes, usually) by doping the various
layers of the unit cell. As in YBCO compounds, the carrier concentration can be manipulated to optimize T c by
doping with elements of different valance. Using this concept a T c of 110 K has been achieved in TSCCO by
25
Fig. 24. Lattice parameters a and c versus T, for HgBa2 CuOx , samples with different oxygen contents. [After Wagner et
al. (70).]
combining the leadbismuth substitution for thallium with a substitution of yttrium for calcium (67,68). The
chemical formula is typically (Tl1 x Mc )Sr2 (Ca1 h Yh )Cu2 O7 , with various choices of x and h between 0 and
1, and around 0.1. The T c of this series has been found to lie between 40 K and 100 K.
The first mercury-based HTSC is HgBa2 CuO4+ (69). As in the case of La2 CuO4+ , the primary doping
mechanism is an interstitial oxygen. The maximum T c = 95 K when is approximately equal to 0.06, but drops
to 59 K when = 0.01. The change in oxygen content changes the unit-cell a and c dimensions. The associated
change in T c correlates with these dimensions as shown in Fig. 24 (70). Experiments have shown that about
8% of copper occasionally substitutes on the mercury site, but a systematic pattern of changes in T c is yet to
be established. The new sister compounds with two or three CuO2 layers that have T c = 133 K also contain a
slight excess of oxygen (6). It is expected that the effect of oxygen interstitial and vacancies may be similar to
that found in previous HTSCs.
Conclusion
This article deals with the structure and role of constituents in high-temperature superconductors. After introducing certain basic terminology, we have presented schematic drawings of the unit cells of several different
high-temperature superconductors. The high-temperature superconductors have in common the presence of
copper-oxide layers, with superconductivity taking place between these layers. The unit cells are not perfectly
symmetric, which has important consequences for superconductivity. For the role of substituents, the range of
possible substitution in high-T c superconductors is enormous. This article presented some examples with an
effort to explain why researchers preferentially try particular kinds of doping. The motivation is to manipulate
the interacting electrons so as to learn more about the metallic state of the ceramic oxides. The goal of doping
research is not to raise T c through empiricism but to establish an experimental foundation on which a theory
can be built to explain the mechanism of high-T c superconductors.
BIBLIOGRAPHY
1. H. K. Onnes, Report on the research made in the Leiden Cryogenic Laboratory between the 2nd and 3rd Int. Congr. on
Refrigeration, Suppl. No. 34b, 1913.
2. J. G. Bednorz K. A. Muller, Z. Phys. B, 64: 189, 1986.
26
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.

X. D. Wu, et al.Appl. Phys. Lett., 51: 861, 1987.
H. Maeda, et al.Jpn. J. Appl. Phys., 27: L2, 1998.
Z. Z. Sheng A. M. Hermann, Nature, 332: 138, 1988.
A. Schilling, et al. Nature, 363: 56, 1993.
M. Nunez-Reguerio, Science, 262: 97, 1993.
C. W. Chu, Nature, 365: 323, 1993.
L. N. Cooper, Phys. Rev., 104: 1189, 1956.
F. London H. London, Proc. R. Soc. London, Ser, A., 149: 7, 1935.
J. C. Phillips, in Physics of High-T, Superconductors, New York: Academic Press, 1989.
A. Wattiaux, et al. C. R. Seances Acad. Sci., 310: 1047, 1990.
C. W. Chu, et al. Phys. Rev. Lett., 58: 405, 1987.
J. Fonteuberta, L. Fabrega, A. V. Narlikar (ed.), Studies in High Temperature Superconductors, vol. 16, New York: Nova
Science, 1996.
D. Jerome, W. Kang, S. S. P. Parkin, J. Appl. Phys., 63: 4005, 1988.
J. M. Tranquada, et al. Phys. Rev. B, 38: 8893, 1988.
D. Kumar, Electrical, structural and chemical properties of laser ablated YBCO thin film, Ph.D. thesis, Indian Institute
of Technology, Bombay, 1993.
S. J. Rothman, et al. Proc. DIMETA-88, Int. Conf. Diffus. Met. Alloys, Hungary, 1988.
R. J. Cava, Nature, 338: 328, 1989.
T. Siegrist, et al. Nature, 334: 231, 1988.
C. C. Torardi, et al. Science, 240: 631, 1988.
P. Chaudhari, et al. Phys. Rev. Lett., 60: 1653, 1988.
D. Dimos, P. Chaudhari, J. Mannhart, Phys. Rev. B, 41: 4038, 1990.
J. Narayan, et al. Appl. Phys. Lett., 51: 1845, 1987.
D. Dijkkamp, et al. Appl. Phys. Lett., 51: 619, 1987.
D. Dimos, et al. Phys. Rev. Lett., 61: 219, 1988.
K. Jagannadham J. Narayan, Philos. Mag., 59: 917, 1989.
W. T. Read W. Shockley, Phys. Rev., 78: 275, 1950.
K. Jagannadham J. Narayan, H. S. Kwok (ed.), Superconductivity and Applications, New York: AIP, pp. 3749.
M. F. Chisholm S. J. Pennycook, Nature, 351: 47, 1991.
V. L. Ginzberg L. D. Landau, Zh. Eksp. Teor, Fiz., 20: 1064, 1950.
G. Deutsher, Physica C, 153-155: 15, 1988.
K. Jagannadham J. Narayan, Mater. Sci. Eng. B, 8: 5, 1991.
N. D. Browning, et al. Physica C, 212: 185, 1993.
S. McKernan, M. G. Norton, C. B. Carter, J. Mater. Res., 7: 1052, 1992.
D. M. Kroeger, et al. J. Appl. Phys., 64: 331, 1988.
S. Oktyabrsky, et al. Proc. Microse. Microanal., 4, 1998.
R. Kalyanaraman, et al. Mater. Res. Soc. Proc., 526: in press.
M. M. McGibbon, Philos. Mag. A, 73: 625, 1996.
S. E. Babcock, et al. Nature, 347: 167, 1990.
C. B. Eom, et al. Science, 251: 780, 1991.
J. D. Jorgensen, et al. Phys. Today, 44: 34, 1991.
J. D. Jorgensen, et al. Phys. Rev. B, 36: 3608, 1987.
H. Takagi, et al. Phys. Rev. B, 40: 2254, 1989.
L. H. Greene B. G. Bagley, D. M. Ginsberg (ed.), Physical Properties of High-Temperature Superconductors, vol. 2,
Singapore: World Scientific, 1990, p. 509.
J. T. Markert, Y. Dalichaouch, M. B. Maple, D. M. Ginsberg (ed.), Physical Properties of High-Temperature Superconductors, vol. 1, Singapore: World Scientific, 1989, p. 509.
A. V. Narlikar, C. V. N. Rao, S. K. Agarwal, A. V. Narlikar (ed.), Studies of High Temperature Superconductors, vol. 1,
New York: Nova Science, 1989, p. 341.
J. Fink, et al. Proc. Int. Semin High Tc . Supercond., Dubna, JINR E17-90-472, 1990, p. 8.
J. J. Neumeier et al., Phys. Rev. Lett., 63: 2516, 1989.

50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
27
G. Xiao, et al. Phys. Rev. B, 35: 8782.

Y. Oda, et al. J. Phys. Soc. Jpn., 57: 4079, 1988.
G. Xiao, et al. Phys. Rev. Lett. 60: 1446, 1988.
J.-M. Taraskon, et al. Phys. Rev. B, 37: 7458, 1988.
A. M. Balagarov, J. Piechota, A. Pajaczkowska, Solid State Commun., 78: 407, 1991.
S. Katsuyama, Y. Ueda, K. Kosuge, Mater. Res. Bull., 24: 603, 1989.
Williams Chaudhary, 1988.
Kubaschewki Alcock, 1979.
D. Kumar, et al. Appl. Lett., 62: 3522, 1993.
J. Joo, et al. Appl. Supercond., 2: 401, 1994.
V. Ambegaoakar A. Baratoff, Phys. Rev. Lett., 10: 486, 1963.
Y. H. Kao, et al. Appl. Phys. Lett., 67: 353, 1990.
P. G. de Gennes, Rev. Mod. Phys., 36: 225, 1964.
J. Clarke, Proc. R. Soc. London, Ser. A, 308: 447, 1969.
M. Tinkham, in Introduction to Superconductivity, New York: McGraw-Hill, 1975.
D. Kumar, P. R. Apte, R. Pint, J. Appl. Phys., 77: 5802, 1995.
J. P. Sheng, et al. Appl. Phys. Lett., 54: 280, 1989.
R. S. Liu, et al. Appl. Phys. Lett., 57: 2492, 1988.
R. S. Liu, et al. Physica C, 159: 385, 1989.
S. N. Putalin, et al. Nature, 362: 226, 1993.
J. L. Wagner, et al. Physica C, 210: 447, 1993.
READING LIST
IEEE Trans. Electron. Dev., 34, 1987, special issue.
Y. Jeon, G. Liang, J. Chen, M. Croft, M. W. Ruckman, D. Di Marizo, M. S. Hegde, (1990), Phys. Rev. B, 41, 4066.
W. Jin, M. H. Dagani, R. K. Kalia, P. Vashishta, Phys. Rev. B, 45, 5535.
D. Kumar, (1994) in Electrical, Structural and Chemical Properties of Laser Ablated YBCO Thin Films, Ph. D. Thesis
(Indian Institute of Technology, Bombay, India).
G. Xiao, Cieplak, D. Musser, M. Z., A. Gavrin, F. H. Streitz, A. Bakhshai, C. L. Chien, J. J. Rhyne, J. A. Gotaas, (1988),
Nature, 332, 238.
D KUMAR
RAJIV K. SINGH
R. KALYANARAMAN
S. OKTYABRSKY
K. JAGANNADNAM
University of Florida
J. NARAYAN
North Carolina State University
HIGH-ENERGY PHYSICS PARTICLE DETECTOR MAGNETS
739
ple who had developed the cryogenic systems for hydrogen

bubble chambers.
The hype and hopes of type two superconductivity soon
faded with the realization that the niobium tin then available
was not a usable superconductor for large magnets. Early on
the experimental work shifted to NbTi and NbZr alloys.
Before 1964, NbZr was the alloy of choice because it was
more stable than NbTi. The alloy superconductors that performed well in short samples would reach only 30% or 50% of
their critical current in a magnet (2). Wires plated with copper appeared to perform somewhat better than bare wire (3).
Degradation due to flux jumps was the topic of the day. Improvements in magnet performance were not spectacular because there was no general understanding of what was happening within the superconductor. By 1964, the construction
of large superconducting particle detector magnets appeared
to be nearly hopeless.
Early Superconducting Detector Magnets
The paper on cryogenic stability of superconductors by Stekly
and Zar (4) caused excitement in the particle physics community. The paper stated that if the superconductor was put in
a low resistivity matrix, it didnt matter whether the superconductor flux jumped as long as the matrix remained at a
temperature below the superconductor critical temperature.
The discovery of cryogenic stability led to the first large detector magnets being built for particle physics. The first of these
magnets was the 12 foot bubble chamber magnet at Argonne
in 1969 (5,6). The 12 foot bubble chamber was followed by a
7 foot bubble chamber magnet at Brookhaven in 1970 (7), a
15 foot bubble chamber at the Fermilab 1973 (8), a 3.8 meter
bubble chamber at CERN 1973 (8,9), the LASS magnet at
SLAC 1974 (10), and a number of smaller devices. In 1972, M.
Morpugo tested a hollow conductor, forced-cooled, cryostable
solenoid for the OMEGA experiment at CERN (11,12). Cryostability solved the scale problem for large superconducting
magnets, but magnets built in this way operated at low current densities and were far from being thin from the standpoint of particle transmission through the magnet.
Low Mass Thin Detector Magnets
HIGH-ENERGY PHYSICS PARTICLE

DETECTOR MAGNETS
THE DEVELOPMENT OF DETECTOR SOLENOIDS
The discovery of type two superconductivity in 1961 (1) was
celebrated by the particle physics community. Suddenly it appeared to be possible to create a large volume of magnetic
field at an induction not heretofor considered to be economical
using conventional magnets. In 1960, one of the largest operating particle detectors using a magnetic field was probably
the 72 inch bubble chamber at Berkeley. Within days of the
announcement of the discovery of niobium tin as a high field
superconductor, particle physicists at Berkeley and other locations set up research groups to make superconducting magnets for particle detectors. There was excitement and constant communication between groups in both the United
States and Europe. The cryogenic expertise for the development of the new superconducting magnets came from the peo-
Truly thin detector solenoids required a superconductor that

could operate at higher current densities without flux jumping. Work by Bean et al. (13), Hancox (14), Chester (15), and
Smith et al. (16,17) paved the way to understanding the intrinsic stability of superconductors, which led to the development of modern, twisted multifilamentary conductors with a
low matrix metal to superconductor ratio. Increasing the current density in the magnet winding was one way of making
the magnet more transparent to particles.
In addition, thin superconducting solenoids had to be
cooled in a different way. Helium bath cryostats contain too
much material for them to be transparent to particles. In order to reduce the mass of the cryostat, it was found that thin
solenoids had to be cooled indirectly by conduction to tubes
that contain helium. Experimental work in the 1970s suggested that two-phase helium cooling would result in a lower
operating temperature than supercritical helium cooling (18).
The first experiment calling for a thin solenoid was at the
ISR at CERN (19). In 1975 a thin solenoid was proposed for
the MINIMAG experiment proposed by the Lawrence Berke-
740
Table 1. Parameters for Various Thin Superconducting Detector Solenoids
Magnet
CLEO-1
PEP-4
CELLO
CDF
TOPAZ
VENUS
ALEPH
AMY
GSI
ZEUS
DELPHI
H-1
CLEO-2
g-2
KEK Balloon
SDC Test Coil
BaBar
CMS
Central
Induction
(T)
Warm Bore
Diameter
(m)
Cryostat
Length
(m)
Matrix
Material
Conductor
Location
Radiation
Thickness
(Rad Len)
Stored
Energy
(MJ)
Matrix
J
(A mm2)
Type of
Cooling
1.5
1.5
1.3
1.5
1.2
0.75
1.5
3.0
0.6
1.8
1.2
1.2
1.5
1.45
1.2
1.5
1.5
4.0
2.0
2.04
1.5
2.85
2.72
3.4
4.96
2.39
2.4
1.72
5.2
5.2
2.9
14.1a
0.852
1.7
2.76
6.0
3.7
3.84
4.02
5.4
5.4
5.6
7.0
2.11
3.3
2.9
7.4
6.0
3.8
0.18b
2.0
2.4
3.85
12.5
Cu
Cu
Al
Al
Al
Al
Al
Al
Al
Al
Al
Al
Al
Al
Al
Al
Al
Al
Outside
Outside
Outside
Inside
Inside
Inside
Inside
Outside
Inside
Inside
Inside
Inside
Inside
Inside
Inside
Inside
Inside
Inside
0.7
0.83
0.6
0.84
0.70
0.52
1.6
2
1.0
2
1.7
1.8
2.2
2
0.21
1.2
1.4
2
10.0
10.9
5.0
30
20
12.0
136
40
3.4
16
108
130
25
5.5
0.815
45
23
2500
350
645
64
56
?
30.8
50
?
46.3
46
41.3
81.8
241
63.4
37 & 67
15.4
Forced
Forced
Forced
Forced
Forced
Forced
Natural
Pool
Natural
Forced
Forced
Forced
Natural
Forced
Pool at End
Forced
Natural
Natural
Beam orbit diameter, outer solenoid coil diameter 15.1 m, inner solenoid coil diameter 13.4 m.
Total gap between the iron poles (the iron return path is C-shaped, the total gap in the iron is about 0.23 m.
a
b
ley Laboratory in 1975 (20). This experiment required a one

meter diameter solenoid that was 0.35 radiation lengths
thick, including the cryostat. Two 1 m diameter test coils
were built and tested in 1975 and 1976 (21,22). The conductor
in the test coils was operated at matrix plus superconductor
current densities as high as 1250 A mm2. The MINIMAG
experiment was not built, but a larger detector for an experiment at PEP colliding beam ring at the SLAC was embarked
upon. This detector required a clear bore diameter of 2 m with
a gap of 3.3 m between the iron poles. A uniform 1.5 T induction (better than 1 part in 1000 within a 2 m diameter, 2 m
long volume) was required. The coil and cryostat had to be
less than 0.7 radiation lengths thick so that calorimeters and
muon detectors could be located outside of the magnet. Work
began on a 2 m diameter test coil in late 1976. This coil was
tested in 1977 and 1978 (23). The thin coil experimental work
at Berkeley led directly to the CLEO-1 detector at Cornell
University (19) and the PEP-4 detector (24,25) at the PEP
colliding beam facility at the SLAC.
A group at CEN Saclay outside Paris decided to build their
detector magnet using a conductor that had a low copper to
superconductor ratio soldered to very pure aluminum high residual resistance ratio (RRR) matrix. The advantages of the
aluminum matrix were as follows: The minimum propagation
zone was lengthened so that the energy needed to induce a
quench in the magnet was increased by over three orders of
magnitude, and the quench propagation velocity along the
wire was faster than for a comparable copper matrix conductor. The 2 m diameter CELLO detector magnet was first
tested in 1979 (26). A conductor made with the copper matrix
superconductor coextruded in pure aluminum (RRR 1000)
was developed in a number of locations at about the same
time (27). This type of conductor was used on the CDF detector magnet at Fermilab (28), on the VENUS detector (29), the
TOPAZ (30) detector, and the AMY detector (31) at KEK in
Japan, on the ALEPH (32) and DELPHI (33) detectors at
CERN, on the H-1 (34) and ZEUS (35) detectors at DESY in

Germany, on the GSI solenoid (36) at Darmstadt, on the
CLEO-2 detector (37) at Cornell University, and on the CLOE
detector at Frascati. A thin solenoid SDC experiment test coil
was tested before the SSC was canceled (38). The Japanese
flew a 1 m diameter balloon solenoid (39,40) that used a low
matrix to superconductor ratio RRR 1000 aluminum matrix
conductor to achieve a very low radiation thickness (about
0.25 radiation lengths) for a cosmic ray experiment. Detector
solenoids for the BaBar (41) experiment at the B factory at
SLAC, the ATLAS (42,43) toroidal magnet detector at the
LHC, and the CMS (44) solenoidal detector at the LHC are
currently under development or construction. All of these
magnets will use a pure aluminum matrix superconductor
that will be wound on the inside of a hard aluminum support structure.
The use of thin solenoid magnet construction techniques
has proven to be less costly even when thinness was not required. As a result, the thin detector solenoid construction
techniques were used to build two 13.4 m diameter and one
15.1 m diameter solenoid for the g-2 experiment at the Brookhaven National Laboratory (45). These solenoids were successfully tested to full field in the summer of 1996 (46). Table
1 summarizes the design parameters for a number of the thin
superconducting detector magnets.
THE DEFINING PARAMETERS FOR THIN SOLENOIDS

In the literature, thinness is defined in terms of interaction
lengths, absorption lengths, and radiation lengths. In highenergy physics detectors, there is no one universal definition
of thinness. Thus, discussion of interaction lengths must identify the particle, and absorption lengths must identify the
particle and its energy. The most common definition of thinness uses radiation lengths as a defining parameter. One ra-
diation length occurs when 63.2% (1 1/e) of the neutral particles have formed charged particle pairs. This definition is
appropriate in many experiments because the calorimeters
and muon detectors are the only detectors that are located
outside the magnet.
The physical thickness of a material that is one radiation
length thick is a function of the material atomic number Z
and the material specific density . In order for a superconducting magnet to be thin, it must be made from low-density,
low-Z materials. The radiation thickness of a detector magnet
is the sum of the radiation thicknesses of the windings, the
coil support structure, the cryostat, and the intermediate temperature shields.
The radiation thickness X0 of a magnet component can be
estimated using the following expression:
X0 =
t
Lr cos()
(1)
where X0 is the radiation thickness of the magnet component

(given in radiation lengths), t is the physical thickness of the
741
material in the magnet component, Lr is the thickness for one

radiation length of the material in the magnet component,
and is the particle angle with respect to a line perpendicular to the component. In most cases, radiation thickness is
defined when 0.
The value of Lr used in Eq. (1) can be obtained from Table
2 (47,48) or it can be estimated using the following expression:
Lr = 158
Z0.73
(2)
where Z is the atomic number for the heaviest element in the

compound that makes up the component and is the mass
specific density for the material in the component. For pure
elements, Eq. (2) yields a good estimate of Lr, except for ordinary hydrogen, which has no neutrons in its nucleus. For
components made from compounds, the use of the Z for the
heaviest element in the compound will tend to overestimate
radiation thickness whereas using an average value of Z will
Table 2. The Radiation Thickness of Various Materials
Material
Pure elements
Hydrogen
Deuterium
Helium
Lithium
Beryllium
Boron
Carbon
Nitrogen
Oxygen
Neon
Magnesium
Aluminum
Argon
Titanium
Chromium
Iron
Nickel
Copper
Niobium
Tin
Tungsten
Lead
Uranium
Compounds, Alloys and Other Materials
Water
Polyethylene
Epoxy resin
Glass fiber epoxy
Carbon fiber epoxy
Boron aluminum (45% B)
Mylar
Sodium iodide
Lithium floride
304 Stainless steel
Nb47% Ti
a
Liquid state.
Graphite or carbon fiber.
One Radiation Length Lrad
Mass Density
(kg m3)
1
1
2
3
4
5
6
7
8
10
12
13
18
22
24
26
28
29
41
50
74
82
92
70.8a
163a
125a
534
1,848
2,370b
1,550b
808a
1,142a
1,207a
1,740
2,700
1,400a
4,540
7,200
7,870
8,902
8,960
8,570
7,310
19,300
11,350
18,950
630.5
1,261.0
943.2
827.6
651.9
553.9
427.0
379.9
344.6
289.4
254.6
240.1
195.5
168.7
146.7
138.4
131.9
128.6
101
88.6
67.6
63.7
60.0
8,900
7,640
7,550
1,550
353
234
275
470
302
240
146
88.9
140.0
37.2
20.4
17.6
14.8
14.3
11.8
12.1
3.5
5.6
3.2
1,000a
950
1,450
1,750
1,600
2,550
1,390
3,670
2,640
7,900
6,520
360.8
447.8
406
330
418
381
399.5
94.9
392.5
137.9
132.8
360.8
470
280
189
261
149
287
25.9
149
17.4
20.4
kg m2
mm
742
often underestimate the radiation thickness. For components

made from alloys or composites, the method of mixtures can
be applied to achieve a good estimate of Lr.
liding beam storage ring. The number of ampere turns needed

to generate a uniform magnetic induction within the detector
solenoid can be estimated by using the following expression
(49):
THIN DETECTOR SOLENOID DESIGN CRITERIA

NI =
The strategy for minimizing the radiation thickness of a superconducting detector magnet requires the following steps:
(1) Massive parts such as current bus bars, gas-cooled electrical leads, cold mass support structures, vacuum services, and
cryogenic services should be located at the ends of the magnet
away from the region that is supposed to have a minimum
radiation thickness. (2) The superconductor should have a
minimum amount of copper and niobium titanium. The stabilizer matrix material for the superconductor should be made
of a low-resistivity, low-Z material such as ultrapure aluminum. (3) The support structure on the outside of the coil,
which will carry the hoop forces in the solenoid, should be
made from a strong, ductile, low-Z, low-density material with
a high thermal conductivity. (4) The magnet should be cooled
indirectly with helium in tubes that are attached to the coil
support structure. (5) Intermediate temperature shields for
the cryostat should be made of a low-Z, low-density, high
thermal conductivity material such as aluminum. (6) The inner cryostat vacuum vessel should be made from a strong lowZ, low-density material. (7) The outer cylinder of the cryostat
vacuum vessel should be made from a material with a low-Z,
a low density, and an elastic modulus that is reasonably high.
The typical physics detector solenoid is usually between
two unsaturated iron poles that have an average relative permeability that is greater than 20. The magnetic flux generated by the solenoid winding is returned by an iron yoke that
carries the magnetic flux from one pole to the other. The relative permeability of the iron in the return yoke is usually
above 50. Figure 1 shows a typical thin detector solenoid
within an experiment located around the collision point of col-
B0 Lg
0
(3)
where NI is the total number of ampere turns in the detector

solenoid coil needed to generate a magnetic induction B0 in a
solenoid that has unsaturated iron poles that are a distance
Lg apart. 0 is the permeability of air (0 4 107 H
m1).
Equation (3) underestimates the ampere turns needed to
generate the magnetic induction in the solenoid bore anywhere from 3% to 30% depending on the design of the magnetic circuit and the central induction within the solenoid.
The equation underestimates the required ampere turns because the relative permeability of the iron in the poles and
the return yoke is not infinite and the iron in the pole pieces
is often segmented, with detectors between the segments. Often the extra ampere turns are put at the ends of the solenoid
so that the desired field uniformity within the solenoid can be
achieved. Computer codes such as POISSON (50) and OPERA2D (51) can be used to determine the number of ampere
turns needed to generate the desired central induction and
the desired field uniformity within the detector volume.
The amount of superconductor needed to generate the
magnetic field is quite small. When NbTi with a critical current density of 2500 A mm2 at 4.2 K and 5.0 T is used, about
0.3 mm of NbTi is needed for every tesla of central magnetic
induction produced (52). The copper to superconductor ratio
for the conductor can be as low as 0.8. The amount of stabilizer (usually annealed 0.99999 pure RRR 1000 aluminum)
in the conductor is dictated by the type of quench protection
chosen.
;;;
;;;
;;;;;;
;;;
;;;;;;
;;;
;;;
;;;
;;;
;;
;;;
Solenoid thin region
Solenoid thick region
Iron return yoke

with muon
detectors
Barrel calorimeter
Super conducting solenoid
Iron pole
End cap
calorimeter
Axis of rotation
;;
Vertex detector
Figure 1. A quarter section view of a typical colliding beam

physics detector with the solenoid cryostat shown. (The thin
and thick sections of the superconducting magnet are shown.)
Central detector
Meters
Beam collision point
Line of
symmetry
The physical thickness of the superconducting coil is determined by the thickness of stabilizing matrix material in the
conductor. The average conductor current density Jm is determined by the safe quench condition for the coil. For safe magnet quenching through a dump resistor, the magnet E0Jm2
limit can be estimated using the following expression (53,54):
2
E0 Jm
= V I0 F (Tm )
r
r+1
(4)
where E0 is the magnet stored energy when it is operated at

its design current I0, V is the discharge voltage for the magnet
during the quench (for large magnets V is limited to about
500 V), I0 is the magnet design current (I0 is typically greater
than 3000 A), r is the matrix to superconductor ratio, and
F*(Tm) is the integral of Jm2 dt needed to raise the stabilizer
adiabatic hot spot temperature from 4 K to a maximum hot
spot temperature Tm. (For RRR 1000 aluminum, F*(Tm)
6 1016) A2 m4 when Tm 300 K.)
The magnet stored energy E0 can be estimated if one
knows the solenoid coil diameter Dc, the central induction
B0, and the gap between the iron poles Lg. An approximate
expression for the magnet stored energy is as follows:
E0 =
D2c B20 Lg
80
(5)
If the E0Jm2 limit for the magnet is increased, then the magnet design current I0 or the magnet discharge voltage V must
be increased as well. Quench back from the coil support structure can be helpful in improving the quench protection for
the magnet. Magnets that employ quench back (22,55) as the
primary means for quench protection can be operated at a
much higher E0Jm2 limit, but the typical solenoid that is protected with a dump resistor across the leads has the E0Jm2
limit given by Eq. (4).
From Eqs. (4) and (5), one can determine the thickness of
the superconducting coil tc using the following expression:
tc =
D2c B40 Lg
830V I0 F (Tm )
0.5
r
r+1
(6)
In order for the coil thickness to be thinner than the value

given by Eq. (6), quench back must turn the whole coil normal
in a time that is significantly faster than the L over R time
constant of the coil and dump resistor circuit.
In virtually all of the large detector solenoids, the superconducting coil is wound inside the support cylinder (56).
When the coil is inside the support cylinder, the joint between
the coil and the support structure is in compression as the
magnet is charged. An additional advantage is that the coil
package cools down from the outside. Thus, the support cylinder shrinks over the coil. A few of the smaller detector magnets were wound with the coil on the outside of a bobbin or
support cylinder. In all these cases, the conductor was designed to carry all of the magnetic hoop forces and the helium
cooling tubes were attached to the outside of the coil. The
superconducting solenoid coil can be wound in one or two layers. A two-layer coil has the advantage of having both leads
from the coil come out at the same end of the coil package.
There are a number of accepted ways of winding coils so that
743
they have more current per unit length at the solenoid ends
than in the center. One approach is to make the matrix current density higher at the ends by making the conductor thinner along the coil axis.
The thickness of the support shell outside the superconducting coil is governed by the magnetic pressure on the coil
windings (49). Total strain of the coil should be limited to
prevent plastic deformation of the conductor matrix. If the
conductor has a pure aluminum matrix, the strain limit for
the coil should be set to about 0.1% (57,58). A conservative
view assumes that virtually all of the magnetic forces are carried by the support shell, and the calculated shell thickness
is given by the following expression:
ts = 250
B20 Ds
0 E s
(7)
where ts is the design thickness for the support shell, Ds is

the inside diameter of the support shell, and Es is the modulus of elasticity of the material in the support shell. If the
superconductor is included in the overall strain calculation,
the thickness of the support shell can be reduced.
The coil cryostat is primarily the vacuum vessel that provides the insulating vacuum for the magnet. The two primary
cryostat elements are the outer cryostat vacuum vessels and
the warm bore tube. The multilayer insulation and shields
make up only a minor part of the cryostats radiation thickness. A design thickness of a solid outer cryostat wall can
be calculated using the following expression, which has been
derived from the equation for elastic buckling of a cylinder
under external pressures (59,60):
P0 L0 D1.5
0
t0 = 1.08
E0
0.4
(8)
where t0 is the thickness of the outer cryostat wall, P0 is pressure on the outer wall of the cryostat (usually P0 1 atm
1.013 105 Pa), L0 is the length of the thin unsupported section of the outer cryostat wall, D0 is the diameter of the outer
cryostat wall, and E0 is the elastic modulus of the material in
the outer wall of the cryostat.
The minimum thickness of the inner wall of the cryostat
can be derived if one knows the design ultimate stress for
the material in the inner wall (58,61). The margin of safety
normally applied to a pressure vessel wall, such as the cryostat inner wall, is usually four (62). An expression for the
minimum inner cryostat wall thickness is given as follows:
ti = 2
Pi Di
u
(9)
where ti is the minimum wall thickness for the inner cryostat

wall, Pi is the design internal pressure on the inner cryostat
wall, Di is the diameter of the inner cryostat wall, and u is
the ultimate stress for the material used in the inner cryostat
wall. Sometimes, the cryostat inner wall thickness is greater
than the thickness given by Eq. (9) so that one can mount
particle detectors and other equipment on this wall.
The material thicknesses calculated using Eq. (6)(9) can
be used to estimate the radiation thickness of the detector
solenoid. Table 3 compares four cases where the coil diameter,
744
Table 3. A Comparison of Four Thin Solenoids

Component
Central induction B0 (T)
Solenoid coil diameter (m)
Gap between the iron poles (m)
Length of the solenoid thin section (m)
Cryostat inside diameter (m)
Cryostat outside diameter (m)
Cryostat overall length (m)
Magnet ampere turns (MA)
Magnet stored energy (MJ)
Magnet design current (A)
Magnet selfinductance (H)
Number of conductor layers
Number of coil turns
Quench discharge voltage (V)
Matrix current density (A mm2)
NbTi plus copper thickness (mm)
Total coil thickness (mm)
Coil support structure thickness (mm)
Inner cryostat thickness (mm)
Outer cryostat thickness (mm)
Magnet radiation thickness (Rad Len)
Magnet cold mass (metric tons)
Magnet overall mass (metric tons)
Case 1
Case 2
Case 3
Case 4
1.5
2.0
3.3
3.3
1.84
2.24
3.85
3.94
9.28
5000
0.74
2
788
500
127.1
0.90
9.39
12.97
1.24
13.11
0.495
2.12
4.00
1.5
4.0
6.6
6.6
3.80
4.28
7.30
7.88
74.25
5000
5.94
2
1576
500
44.9
0.90
26.58
25.95
2.57
25.23
0.974
14.4
23.5
0.75
4.0
6.6
6.6
3.80
4.22
7.30
3.94
18.56
5000
1.48
2
788
500
89.9
0.45
13.28
6.49
2.57
25.23
0.589
6.16
15.3
1.5
4.0
3.3
3.3
3.80
4.26
3.85
3.94
37.13
5000
2.97
2
788
500
63.6
0.90
18.75
25.95
2.57
19.12
0.829
7.00
11.8
the gap between the iron poles, and the central induction are
varied. In all four cases, the cryostat walls and coil support
structure are made from solid aluminum. The superconductor
is NbTi with a thick aluminum stabilizer. The assumed insulation system inside and outside the cold mass consists of
60 layers of aluminized mylar and netting with a single 1 mm
thick aluminum shield on other side of the coil. Figure 2
shows a cross section of a coil and cryostat for CASE 2 given
in Table 3. In order to make a significant reduction in the
radiation thicknesses shown in Table 3, quench back must be
the primary mode of quench protection and the outer cryostat
vacuum vessel must be made from a cellular (honeycomb)
composite structure that is physically thicker than a solid aluminum vessel (63,64).
MAGNET POWER SUPPLY AND COIL QUENCH PROTECTION
The power supply parameters are set by the coil charge time
tch and the design operating current I0 for the solenoid. The
charge time for a detector solenoid is rarely an issue. Charge
times as long as one hour are acceptable. The charge voltage
V L1 di1 /dt, where L1 is the self-inductance of the magnet
circuit; and di1 /dt, is the magnet current charge rate. (For a
typical magnet, di1 /dt, I0 /tch.) To determine the power supply voltage, one must add the IR voltage drop across the gascooled electrical leads and the cables connecting the power
supply to the magnet. In addition, a voltage drop of 0.9 V
should be allocated to the power supply back wheeling diodes
and a current shunt.
The Quench Protection Dump Resistor
Most large detector magnets are protected by a dump resistor
across the gas-cooled electrical leads. When a quench is detected, the power supply is disconnected and the dump resistor is put across the leads. The design of a magnet dump re-
sistor circuit is determined by the following relationship (54):
F (Tm ) =
j(t)2 dt =
r
(r + 1)
Tm
T0
C(T )
dT
(T )
(10)
where j(t) is the current density in the magnet superconductor cross section as a function of time t, C(T) is the superconductor volume specific heat as a function of temperature T,
(T) is the superconductor matrix material electrical resistivity as a function of temperature, and r is the ratio of matrix
material to superconductor in the magnet conductor. T0 is the
starting temperature of the magnet (about 4 K), and Tm is the
maximum allowable hot spot temperature for the magnet conductor (usually 300350 K). For a conductor with a very pure
aluminum matrix with an RRR of 1000, the value of F*(Tm) is
around 6.0 1016 A2 m4 s when Tm is 300 K.
When the magnet is discharged through a dump resistor,
the current decay is exponential with a decay time constant
1 (1 L1 /Rex, where Rex is the resistance of the external
dump resistor). The value of F*(Tm) at the magnet coil hot
spot is given as follows:
F (Tm ) = j02

(r + 1) 1
+ ts0
r
2
(11)
where ts0 is the time needed to detect the quench and switch
the resistor across the magnet coil (in most cases ts0 is less
than one second) and j0 is the starting current density in the
coil superconductor plus matrix material (I0 divided by the
conductor cross-sectional area). If a constant resistance dump
resistor is used, the value of the resistance Re that results in
a hot spot temperature less than or equal to Tm can be expressed as follows:
Re
j02
2F (Tm )
(r + 1)
L1
r
(12)
The design value of Re should be larger than the value calculated by Eq. (12). For a constant resistance dump resistor, the
maximum discharge voltage across the leads V ReI0 will
occur when the dump resistor is just put across the magnet.
Figure 3 shows a circuit diagram of the coil, its power supply, and the magnet dump circuit. A quench detection system
is also shown. The values for inductances and R given in Fig.
3 would apply to CASE 2 in Table 3. The quench detection
system shown in Fig. 3 compares the voltage across the superconducting coil with the dB/dt voltage due to changes in flux
in the coil. If a voltage is measured across the coil and there
is no corresponding dB/dt voltage, there is a normal region in
the coil. The normal region detected by the quench detector
will open the switch, putting the dump resistor across the coil.
Other methods can also be used to detect short normal sections within a magnet (65).
745
of quench protection (66,67). If the resistance of the external

resistor induces quench back in a time less than the time tQ,
then the hot spot temperature is less than Tm, the maximum
allowable hot spot temperature. The maximum allowable
quench-back time tQ is the quench-back time required for fail
safe quenching tF minus the time required for a heat pulse to
cross the insulation between the quench back circuit (usually
the coil support structure) and magnet coil tH. (For a layer of
ground plane insulation that is two millimeters thick, tH varies from 0.3 to 0.5 seconds depending on a number of factors.)
For large detector solenoids, tH is usually small compared 1.
The value of tF can be determined using the following expression:
r
tF =
r+1
F (Tm ) F (Ts )
j02

(13)
The Role of Quench Back

It has been observed in most of the thin detector solenoid
magnets that when the dump resistor is put across the electrical leads, the entire magnet becomes normal through the
process of quench back (22). Quench back ensures that the
coil current will decay faster than is predicted by the L over
Rex time constant. As a result, the magnet hot spot temperature is reduced.
There is a maximum time tQ before which quench back
must occur in order to have quench back be a fail-safe method
;;
;;
;;
where r and j0 are previously defined. F*(Tm) and F*(Ts) are

defined by the right-hand term in Eq. (10) for the maximum
hot spot temperature Tm and the maximum temperature the
coil would go to if the entire coil quenched instantaneously
Ts. The value of Ts depends on how the stored energy of the
magnet is split between the hard aluminum support shell and
the coil. For detector solenoids it is usually safe to split the
magnet stored energy between the support shell and the coil
according to their masses.
Stainless steel cup
;;;;
;
;;
;;;;;;;
;;;;;;;;
;
;;
;;;;;;;
;;
;;;;;;;
;;
;;;;;;;;;;;
Thick A1 outer support flange
Support strut
A1 coil support shell
Thin outer shell L = 6600 mm
Or = 2180 mm
Stainless steel
Stiffening Rings
Thin A1 outer wall
Thin outer shell Or = 2140 mm

Copper 80 K shield
He cooling tube
Support block
Seal plate
IR = 1900 mm
Cryostat L = 7300 mm
0
100
Multilayer insulation
Thin A1 inner Wall
A1 matrix S/C
80 K honeycomb structure
S/C radius = 2000 mm
Thin shell L = 5945 mm
Coil L = 6600 mm
Thick A1 inner vacuum wall
200
mm
Figure 2. A cross section through the end of a 1.5 T thin solenoid with a 4.0 m coil diameter. A
self-centering support strut is shown along with the stiff end ring for the superconducting coil
package. (See Case 2 in Table 3.)
746
Mutual inductance = 3.78 mH

Selfinductance = 5.94 H
Number of turns = 1576
dB/dt
Coil
Current
shunt
Current
shunt
R = 10
R = 10
Quench
detector
Dump resistor
R = 0.1
Solenoid coil
Gascooled lead
Gascooled lead
Dump switch
Once tQ has been determined, it is possible to calculate the

resistance of an external resistor needed to cause the coil to
quench back from the support tube in a time that is less than
tQ. The minimum resistance needed for quench back Rm can
be calculated using the following relationship for a solenoid
coil that is well coupled inductively to its quench-back circuit
(66):
Rm =
L1 N2 A2
2 N1 I0
H2
2 tQ
Power
Supply
5000 A
12V
To voltage
control
Solenoid support structure

and coupled shorted secondary
selfinductance = 2.46 H
Circuit resistance ~ 0.5 m
4.4 K Time constant = 2.4 s including the gascooled leads
Figure 3. A schematic circuit diagram for

the coil, power supply and quench protection
circuit for a large detector solenoid with a
support cylinder. (See Case 2 in Table 3.)
To current
control
0.5
(14)
where L1 is the magnet coil self-inductance, N2 is the number

of turns in the quench-back circuit (N2 1 when the support
shell is the quench back circuit), N1 is the number of turns in
the magnet coil, A2 is the cross-sectional area of the quench
back circuit, I0 is the coil current, H2 is the enthalpy change
per unit volume needed to raise the quench-back circuit temperature from 4 K to 10 K (for aluminum, H2 13,200 J
m3), 2 is the resistivity of the quench-back circuit material,
and 2 is the L over R time constant for the quench-back circuit (the support tube).
If the minimum quench-back resistance Rm is less than the
resistance of the quench protection resistor Re, quench back
will always occur during a magnet dump. Therefore, the hot
spot temperature of the coil is lower than Tm. When quench
back is present, one can use a varistor (a resistor where the
voltage across the resistor is nearly independent of current)
as a dump resistor to speed up the quench process without
increasing the coil voltages during the quench (68).
DESIGN CRITERIA FOR THE ENDS
OF A DETECTOR SOLENOID
The previous sections have dealt primarily with the center
section of a detector solenoid. Much of the engineering for a
detector solenoid is in the ends of the magnet, where thinness

is not an issue. For example: (1) The support system for the
solenoid cold mass is attached to the ends of the magnet. (2)
The outside ends of the cryostat vessel are often where physical connections are made between the magnet and the rest of
the detector. (3) The current leads and voltage taps into the
coil will come out of the coil package at its ends. Gas-cooled
electrical leads that connect the coil to the room-temperature
outside world may also be located inside the magnet insulating vacuum vessel in the end region. (4) Cryogenic cooling is
usually fed into the solenoid coil from the ends. Cooling
should also include the intermediate temperature fluid (either
liquid nitrogen at 80 K or helium gas at 50 to 80 K) used to
cool the shields. (5) cryostat vacuum pumping ports will be
located at the ends of the solenoid. (6) Room-temperature feed
throughs for voltage taps, quench detection coils, temperature
sensors, and pressure transducers will enter the magnet at
the ends.
Cold Mass Support System
The cold mass supports to room temperature must carry gravity forces, seismic forces, magnetic forces, and shipping forces.
Most detector solenoids are designed to be at a neutral magnetic force point when the coil is at its operating temperature,
so the cold mass support system must have a spring constant
that is higher than the magnetic force constant.
Solenoids that are surrounded by iron are usually, but not
always, in stable equilibrium in the radial direction and in
unstable equilibrium in the axial direction. In the tortional
direction (about the solenoid axis), there are almost no magnetic forces in a well-built solenoid, although asymmetric
holes in the iron can introduce some of these forces. Stable
equilibrium indicates that the magnetic forces will act in a
direction that reduces a placement error; unstable equilibrium indicates that the magnetic forces will act in a direction
that increases the placement error. In the direction of stable

equilibrium, the spring constant of the support system is not
a critical issue except when determining how the magnet responds to vibration. In the direction of unstable equilibrium
(usually the axial direction) the spring constant of the support
system must be larger than the force constant for the magnet
at its maximum design field. In general, the magnet force constant is linear with the location error and it increases with
the magnet current squared. The magnetic force constant is
a function of the design of the coil, the iron return yoke, and
the pole pieces. Magnetic force constants will change as the
iron in the magnetic circuit saturates.
Two types of cold mass support systems are commonly
used in detector solenoids. The first is the self-centering support system where the position of the center of the solenoid
coil does not change during the magnet cooldown or as the
magnet is powered. The second support system carries axial
forces with pushpull rods at one end of the magnet while
the radial forces are carried by gravity support rods at both
ends of the magnet. Both types of support systems must be
designed to handle magnet shrinkage during the cooldown. A
coil that is 6.6 m long and 4 m in diameter will shrink almost
28 mm in the axial direction and the radius will decrease
about 8.4 mm. The external cryostat support system should
be in line with the cold mass support system in order to avoid
bending within the cryostat. The spring constant for the combined internal and external support systems must be greater
than the magnetic force constant.
The self-centering support system has several advantages:
(1) the position of the magnet center is the same both warm
and cold. The PEP-4 solenoid magnetic center changed less
than 0.3 mm during the magnet cooldown. (2) The radial and
axial supports can be combined using either tension or compression rods. Two of the rods can also carry the tortional
forces about the solenoid axis (tortional forces). The angle of
the support rods can be set so that rod stress is not changed
during the coil cooldown. (3) Since the axial spring constants
must be high, the spring constant will be high in all directions. The self-centering support system will have a relatively
high first mode vibration frequency. (4) The self-centering
support system is robust in all directions, so earthquake and
transportation forces should not be a problem.
The two disadvantages of the self-centering support system are as follows: (1) As the magnet coil cools down, it will
move with respect to the ends of the cryostat vacuum vessel
at both ends of the magnet. This movement must be considered when designing electrical leads, cryogen feed-throughs,
and other attachments to the coil. (2) Flexure of the coil package (at the ends of the support cylinder) will affect the spring
constant of the support system. The stiffness of the ends of
the coil package and the number of radialaxial supports are
the determining factors for the spring constant of this type of
support system. Finite element stress and strain calculations
can be used to determine the spring constant of the cold mass
support system. A description of the design of a self centering
support system can be found in Ref. 69. A location of a typical
self-centering support compression strut for a detector solenoid is shown in Fig. 2. The strut rotates in its sockets as the
solenoid cold mass contracts. The distance between the ball
sockets does not change as the solenoid cools down from room
temperature to 4 K. The angle of the strut with respect to the
747
solenoid axis changes as the coil end of the strut moves toward the center of the solenoid.
The Solenoid Support Structure, the Cryogenic Heat Sink
The support cylinder outside the superconducting winding
serves the following functions: (1) The outer cylinder carries
the magnetic pressure forces that are generated by the coil.
(2) The outer cylinder transfers magnetic, gravitational, and
seismic forces from the coil structure to the cold mass support
system. (3) The outer cylinder carries the helium cooling
tubes and acts as the heat sink for the coil and all attachments to it. This means that the outer support cylinder must
be made from material that conducts heat well in both the
radial and axial directions.
The end ring of the support cylinder should be as stiff as
possible in bending. End-ring stiffness can be increased by
making the ring thicker, thus increasing its moment of inertia, or one can fabricate a laminated end ring with a high
elastic modulus material such as 304 stainless steel (elastic
modulus of 200 GPa as compared to 69 GPa for aluminum)
on the outside and the inside of the ring with aluminum in
the center. The need for stiff end rings on the support cylinder
is reduced as the number of cold mass supports per end is
increased for a given coil diameter (69).
Coil Electrical Connections and Leads to the Outside World
Connections to the superconducting coil that come through
the end ring should be mounted on copper bus bars that are
electrically insulated from the end rings. These bus bars
should be cooled in liquid helium in order to avoid heat from
outside the coil being deposited directly into the superconducting windings. Heat leaks down pulsed current leads,
which are usually not gas cooled, can be particularly troublesome. The cooling circuit used to cool bus bars at the ends of
the coil should be part of the magnet helium cooling system.
Since much of the cooling circuit is electrically grounded, inline electrical insulators will be required in the cooling lines
that cool the electrical bus bars connected to the superconducting coil.
Most detector solenoids have gas-cooled electrical leads
that are fed from a liquid helium pot located somewhere near
the solenoid. The current buses between the lead pot and the
coil are often cooled by conduction, a practice that has led to
a number of failures. All current buses should be heliumcooled. The lead pot commonly used in detector magnets can
be eliminated by using gas-cooled electrical leads that are
attached to the ends of the coil structure. The helium used to
cool these leads comes directly from the liquid helium cooling
circuit. Gas-cooled leads attached to the end of the magnet
are located within the cryostat vacuum, so these leads must
be completely vacuum tight and they must withstand any increase in pressure that might occur in the cooling circuit during a quench (70). The bundled nested tube leads that were
used on the PEP-4 experiment (71) and the g-2 solenoids (72)
can be operated at any orientation within the cryostat vacuum vessel. Properly designed gas-cooled leads are stable and
they are capable of operating for more than 30 min without
gas flow.
CRYOGENIC COOLING OF A THIN DETECTOR SOLENOID
Most of the detector solenoids shown in Table 1 are cooled by
helium in tubes attached to the superconducting coil or the
748
support cylinder outside the coil. This technique has the following advantages over the bath cooling used for early cryostable detector magnets (73): (1) Tubular cooling eliminates
the cryostat helium vessels. As a result, the solenoids are
thinner and less massive. (2) The volume of helium in a tubular cooling system is small. Once this helium is evaporated
during a quench, it is expelled from the tube. Large quantities
of helium gas are not produced during a magnet quench. The
helium expelled during a magnet quench can be returned to
the refrigerator where it is recovered. (3) Tubes can withstand
high pressures during a quench. Relief valves for the system
can be moved from the magnet cryostat to the helium supply
system, which can be outside the detector. (4) Magnet cooldown can be done directly using the helium refrigerator. (5)
Recovery from a quench can be simplified using a well-designed tubular cooling system.
Detector magnets are cooled with two-phase helium rather
than supercritical helium for the following reasons: (1) The
operating temperature for the superconducting solenoid is
lower (18). As two-phase helium flows down the cooling circuit, it gets colder as its pressure goes down. The temperature
of a single-phase cooling circuit increases as one goes along
the cooling circuit. (2) The mass flow through the cooling circuit is minimized. As a result, the pressure drop along the
flow circuit is lower. (3) There is no need for auxiliary helium
pumping in a two-phase flow circuit. Helium flow can be provided directly by the J-T circuit of the refrigerator. (4) A properly designed two-phase helium flow system can be operated
at heat loads greater than the capacity of the refrigerator for
a period of time. Thus fluctuations in the heat load can be
tolerated by two-phase flow circuits that are designed for the

average heat load. The most often stated disadvantage of twophase cooling is the potential existence of flow and pressure
fluctuations in the cooling tube. This has not been a problem
in detector magnets when the two-phase helium cooling circuit is properly designed. Experiments with extensively
looped cooling tubes that are hundreds of meters long have
shown that proper design of the flow circuit can nearly eliminate the flow oscillation problem (73,74).
The types of two-phase helium flow circuits are commonly
used in detector solenoids are the forced two-phase flow system and the natural convection two-phase flow system.
Forced two-phase flow is appropriate when the flow circuits
are long and when the control dewar is below the top of the
magnet. Natural convection two-phase flow is appropriate
when there is a large vertical head between the helium dewar
and the load and when there are many parallel flow circuits
so that the mass flow in any one circuit can be kept low. Either type of two-phase helium flow circuit can be made to
work in most detector solenoids.
The key to stable operation of forced two-phase helium
cooling circuits is the control dewar and heat exchanger
(25,73). Flow for the magnet cooling circuit comes from the
J-T circuit of the helium refrigerator. Cooling flow can also
come from a positive displacement helium pump (75), but allowances must be made for the pump work heating generated
by such a pump. Two-phase flow from the refrigerator J-T
circuit flows through a heat exchanger that is cooled in a bath
of liquid helium at the suction pressure of the cold end of the
refrigerator. The temperature of the helium bath is the lowest
Helium
compressor
T = 300K
Mixing circuit
for cooldown
F
LN2
Refrigerator
cold box
Bypass circuit
for cooldown
Gas handling is not shown

as part of the compressor
flow system
F
Lead gas
Flow control
Helium-gascooled leads
JT valve
LN 2
Insulators
2 F Helium flow circuit

Liquid He
Figure 4. A schematic representation of

a forced two-phase helium cooling system
for a large detector solenoid while the solenoid is cold. (Note: Open valves are unshaded; closed valves are shaded. Dark
lines carry flow; shaded lines carry no
flow.)
T = 4.4 K
Heat
exchanger
500 Liter control
dewar
T = 4.6 K
Thin S/C solenoid
cryostat
Not included are the backup relief valves or
the vacuum system relief valves. Relief valves
are not needed on the magnet piping
749
JT valve
LN2
T = 4.4 K
Refrigerator
cold box
Cooldown
shut-off valve
Mixing circuit
for cooldown
500 Liter
phase
separator
Liquid He
T = 4.4 K
F
Lead gas
F flow control
Helium-gascooled leads
Insulators
Bypass circuit
for cooldown
T = 300 K
LN2
Natural convection
2 F helium flow circuit
Helium compressor
Gas handling is not shown as
part of the compressor flow system.
Refrigeration in and out

T = 4.4 K
Thin S/C solenoid
cryostat
stor
ium
l
e
H
age
tank
LHe
Not included are the backup relief
valves or the vacuum system relief
valves. Relief valves are not needed
on the magnet piping.
man
gas
ium
Hel
u
Nat
Liqu
ral
id
con
ifold
tio
vec
um
heli
n2
man
e
has
coo
ling
ifold
Figure 5. A schematic representation of a natural convection two-phase helium cooling system

for a large detector solenoid while the solenoid is cold. (The insert at the lower right shows the
physical arrangement of the natural convection cooling system.)
temperature in the two-phase flow circuit. Within the heat

exchanger, helium in the gas phase is condensed to liquid so
that the helium leaving the heat exchanger either is on the
saturated liquid line or is slightly subcooled. As a result, the
average density of the helium in the flow circuit is maximized,
which will cause the flow pressure drop through the flow cir-
cuit to be reduced a factor of two or three as compared to the

same flow circuit without a heat exchanger in the helium
bath. The use of the heat exchanger in the control dewar
allows the operation of the cooling circuit with a heat flow
into the magnet that exceeds the capacity of the refrigerator
by as much as 50%. Under this condition, the magnet can be
750
kept cold as long as the heat exchanger in the control dewar

is kept covered with liquid helium. The control dewar enhances flow circuit stability. Figure 4 shows a schematic representation of a forced two-phase helium cooling circuit in its
simplest form. The valves in Fig. 4 are shown as they would
be when the magnet is operating at 4 K.
The advantages of forced two-phase cooling are as follows:
(1) The entire detector solenoid can be cooled using a single
helium flow circuit. As a result, the cooldown of the magnet
is straight forward because all of the sensible heat of the helium can be employed during the cooldown process. (2) The
operating temperature for a two-phase cooled magnet is lower
than it would be for any supercritical helium-cooled magnet.
(3) The control dewar with its heat exchange can be located
flexibly with respect to the magnet coil. Transfer lines to and
from the coil can be long, if desired. (4) Gas-cooled electrical
leads and shields can be cooled directly from the two-phase
cooling circuit. The connections for leads and shields can be
made inside the magnet cryostat vacuum vessel. (5) Since the
liquid helium inventory in contact with the magnet is limited
to the helium in the cooling tube, the amount of helium gas
produced during a quench is small. (6) The liquid helium in
the control dewar can be used to speed up the recovery of the
superconducting magnet after a quench. The primary disadvantage of forced two-phase cooling is that when the refrigerator stops, the magnet cooling stops. The magnet will quench
within minutes after the refrigerator stops running. For some
users this is a serious consideration.
The natural convection two-phase cooling system overcomes the primary disadvantage of a forced two-phase cooling
system in that the magnet will remain cold and operating
even when the refrigerator is not in operation. The cooling for
a natural convection cooling system comes from the helium
that is stored in a tank that is above the top of the coil. The
greater the head between the storage tank and the top of the
magnet, the better the natural convection two-phase flow system operates. In order for the natural convention flow system
to operate effectively, the following conditions must be present: (1) The pipe from the bottom of the helium storage dewar
to the manifold at the bottom of the magnet should be short
and well-insulated. There should be no boiling in helium
transferred to the lower manifold on the coil package. (2) In
order to reduce the flow circuit pressure drop, there should be
many short up-flow circuits in parallel going up and around
the coil to the manifold at the top of the magnet. Boiling
should occur in these tubes. This increases the helium flow
through the cooling system. (3) The pipe from the manifold at
the top of the coil package should dump two-phase helium
into the top of the storage tank, where phase separation occurs. This pipe should be insulated from the helium that is in
the storage tank. The difference in helium density in the pipe
connecting the tank and the lower manifold and the twophase helium in the cooling tubes circling the magnet provides the driving force for the flow circuit. Figure 5 shows a
schematic representation of a natural convection two-phase
helium cooling circuit in its simplest form. The upper part of
Fig. 5 shows a schematic of how the refrigerator, its compressors, and the magnet would be hooked up. The lower part of
Fig. 5 shows the physical arrangement of the helium storage
tank (phase separator) and the detector solenoid. The storage
tank shown in Fig. 5 is 500 liters, but that tank could be
much larger if needed. The valves in Fig. 5 are shown as they

would be when the magnet is operating at 4 K.
Natural convection flow has some disadvantages, which
are as follows: (1) Cooldown of the magnet is not as straightforward as with the forced two-phase cooling system. This difficulty can be overcome by having a separate forced flow circuit for the magnet cooldown. (2) In a natural convection
cooled magnet system, all of the helium that is in the storage
tank may be boiled during a magnet quench. This can be overcome by installing an automatic shut-off valve, which is triggered by the quench detector, in the pipe between the storage
tank and the liquid helium manifold on the bottom of the
magnet. (3) The thin section of the solenoid has a larger radiation thickness top and bottom in the regions where the liquid helium and two-phase helium manifolds are located. (4)
The helium storage tank for natural convection cooling must
be located directly above the magnet coil, and the transfer
lines between the storage tank and the coil should be as short
as possible. The physics experiment must accommodate these
transfer lines.
BIBLIOGRAPHY
1. J. E. Kunzler et al., Phys. Rev. Lett., 6: 89, 1961.
2. R. C. Wolgast et al., Superconducting critical currents in wire
samples and some experimental coils, in Advances in Cryogenic
Engineering, Vol. 8, New York: Plenum Press, 1962, p. 38.
3. C. Laverick, The performance characteristics of small superconducting coils, in Advances in Cryogenic Engineering, Vol. 10, New
York: Plenum Press, 1964, p. 105.
4. Z. J. J. Stekly and Z. L. Zar, Stable superconducting coils, IEEE
Trans. Nucl. Sci., NS-12: 367, 1965.
5. J. R. Purcell, The 1.8 tesla, 4.8 m bubble chamber magnet, Proc.
1968 Summer Study Superconducting Devices Accelerators, BNL
50155 (C-55), p. 765, 1968.
6. R. E. Jones et al., Construction of the 12 foot bubble chamber
superconducting magnet cryostat, in Advances in Cryogenic Engineering, Vol. 15, New York: Plenum Press, 1969, p. 141.
7. D. P. Brown, R. W. Burgess, and G. T. Mulholland, The superconducting magnet for the Brookhaven National Laboratory 7 foot
bubble chamber, in Proc. 1968 Summer Study Superconducting
Devices Accelerators, BNL 50155 (C-55), p. 794, 1968.
8. E. G. Pewitt, Superconductivity in high energy physics, in Advances in Cryogenic Engineering, Vol. 16, New York: Plenum
Press, 1970, p. 19 (a review of large magnets).
9. E. V. Haebel and F. Wittgenstein, Proc. Int. Conf. Magnet Technol., Hamburg, Germany, 1970.
10. T. H. Fields, Superconductivity applications in high energy physics, IEEE Trans. Magn., MAG-11: 113, 1975.
11. M. Morpurgo, Construction of a superconducting test coil cooled
by helium forced circulation, Proc. 1968 Summer Study Superconducting Devices Accelerators, BNL 50155 (C-55), 953, 1968.
12. M. Morpurgo, Proc. Int. Conf. Magnet Technol., Hamburg, Germany, 1970.
13. C. P. Bean et al., Research Investigation of the Factors that Affect the Superconducting Properties of Material, AFML-TR-65431, Air Force Materials Laboratory, Wright Patterson Air Force
Base, Ohio, 1965.
14. R. Hancox, Phys. Lett., 16: 208, 1965.
15. P. F. Chester, Rep. Prof. Phys., 30 (2): 361, 1967.
16. P. F. Smith and J. D. Lewin, Pulsed superconducting synchrotrons, Nucl. Instr. Methods, 52: 248, 1967.
751
17. M. N. Wilson et al., Experimental and theoretical studies of filamentary superconducting composites, Part 1: Basic ideas and
theory, J. Phys. Appl. Phys., 3: 1517, 1970.
38. A. Yamamoto et al., Development of a prototype thin superconducting solenoid magnet for the SDC detector, IEEE Trans. Appl.
Supercond., 5: 849, 1995.
18. M. A. Green, Cooling intrinsically stable superconducting magnets with super-critical helium, IEEE Trans. Nucl. Sci., NS-18:
664, 1971.
39. T. Mito et al., Prototype thin superconducting solenoid for particle astrophysics in space, IEEE Trans. Magn., MAG-25: 1663,
1989.
19. H. Desportes, Superconducting magnets for accelerators, beam

lines and detectors, IEEE Trans. Magn., MAG-17: 1560, 1981.
40. Y. Makida et al., Ballooning of a thin superconducting solenoid

for particle astrophysics, IEEE Trans. Appl. Supercond., 5: 658,
1995.
20. M. A. Green, Large superconducting solenoid for the MINIMAG

experiment, in Advances in Cryogenic Engineering, Vol. 21, New
York: Plenum Press, 1975, p. 24.
21. P. H. Eberhard et al., Tests on large diameter superconducting
solenoids designed for colliding beam accelerators, IEEE Trans.
Magn., MAG-13: 78, 1977.
22. M. A. Green, The Development of Large High Current Density
Superconducting Solenoid Magnets for Use in High Energy Physics Experiments, LBL-5350, doctoral dissertation, UC Berkeley,
May 1977.
23. M. A. Green, Large superconducting detector magnets with ultra
thin coils for use in high energy accelerators and storage rings,
Proc. 6th Int. Conf. Magnet Technol., Bratislava, Czechoslovakia,
p. 429, 1977.
24. P. H. Eberhard et al., A magnet system for the time projection
chamber at PEP, IEEE Trans. Magn., MAG-15: 128, 1979.
25. M. A. Green, W. A. Burns, and J. D. Taylor, Forced two-phase
helium cooling of large superconducting magnets, in Advances in
Cryogenic Engineering, Vol. 25, New York: Plenum Press, 1979,
p. 420.
26. J. Benichou et al., Long term experience on the superconducting
magnet system for the CELLO detector, IEEE Trans. Magn.,
MAG-17: 1567, 1981.
27. J. M. Royet, J. D. Scudiere, and R. E. Schwall, Aluminum stabilized multifilamentary Nb-Ti conductor, IEEE Trans. Magn.,
MAG-19: 761, 1983.
28. E. Leung et al., Design of an indirectly cooled 3-m diameter superconducting solenoid with external support cylinder for the
Fermilab collider detector facility, IEEE Trans. Magn., MAG-19:
1368, 1983.
41. P. Fabbricatore et al., The superconducting magnet for the BaBar

detector of the PEP-II B factory at SLAC, IEEE Trans. Magn.,
MAG-32: 2210, 1996.
42. J. M. Baze, Progress in the design of a superconducting toroidal
magnet for the ATLAS detector on LHC, IEEE Trans. Magn.,
MAG-32: 2047, 1996.
43. D. E. Baynham et al., Design of the superconducting end cap toroids for the ATLAS experiment at LHC, IEEE Trans. Magn.,
MAG-32: 2055, 1996.
44. F. Kircher, H. Deportes, and B. Gallet et al., Conductor developments for the ATLAS and CMS magnets, IEEE Trans. Magn.,
MAG-32: 2870, 1996.
45. G. Bunce et al., The large superconducting solenoids for the g-2
muon storage ring, IEEE Trans. Appl. Supercond., 5: 853, 1995.
46. G. Bunce et al., Test results of the g-2 superconducting solenoid
magnet system, IEEE Trans. Appl. Supercond., 7: 626, 1997.
47. Review of Particle Properties, compiled by the Particle Data
Group, Rev. Mod. Phys., 48 (2): Part II, 50, 1978.
48. Y. S. Tsai, Pair Production and Bremsstrahlung of Charged Leptons, SLAC-PUB-1365, Table III.6, 1974.
49. W. R. Smythe, Static and Dynamic Electricity, 2nd ed., New York:
McGraw-Hill, 1950.
50. POISSON/SUPERFISH Reference Manual, LANL Publication
LA-UR-87-126, 1987.
51. Part of the TOSCA series of programs produced by Vector Fields
Ltd., Oxford, United Kingdom.
52. M. A. Green, Calculating the Jc, B, T surface for niobium titanium using the reduced state model, IEEE Trans. Magn., MAG25: 2119, 1989.
29. A. Wake et al., A large superconducting thin solenoid magnet

TRISTAN experiment (VENUS) at KEK, IEEE Trans. Magn.,
MAG-21: 494, 1985.
53. B. J. Maddock and G. B. James, Protection and stabilization of

large superconducting coils, Proc. IEE, 115 (4): 543, 1968.
30. H. Hirabayashi, Detector magnets in high energy physics, IEEE

Trans. Magn., MAG-24: 1256, 1988.
54. P. H. Eberhard et al., Quenches in large superconducting magnets, In Proc. 6th Int. Conf. Magnet Technol., Bratislava, Czechoslovakia, 654, 1977.
31. K. Tsuchiya et al., Testing of a 3 tesla superconducting magnet

for the AMY detector at Tristan, in Advances in Cryogenic Engineering, Vol. 33, New York: Plenum Press, 1987, p. 33.
55. M. A. Green, Quench back in thin superconducting solenoid magnets, Cryogenics, 24: 3, 1984.
32. J. M. Baze et al., Design, construction and test of the large superconducting solenoid ALEPH, IEEE Trans. Magn., MAG-24:
1260, 1988.
56. A. Yamamoto et al., A thin superconducting solenoid with the

internal winding method for collider beam experiments, in Proc.
8th Int. Conf. Magnet Technol., September 1983.
33. R. Q. Apsey et al., Design of a 5.5 meter diameter superconducting solenoid for the DELPHI particle physics experiment at LEP,
IEEE Trans. Magn., MAG-21: 490, 1985.
57. A. Yamamoto et al., Design study of a thin superconducting solenoid for the SDC detector, IEEE Trans. Appl. Supercond., 3: 95,
1993.
34. P. T. M. Clee, Rutherford Appleton Laboratory, Didcot, United

Kingdom, private communication.
58. R. J. Roark and W. C. Young, Formulas for Stress and Strain, 5th
ed., New York: McGraw-Hill, 1975.
35. A. Bonito Oliva et al., Zeus thin solenoid: Test results analysis,
59. S. Timoshenko, Theory of Plates and Shells, New York: McGrawHill, 1940.
36. D. Andrews, Oxford Technology Ltd., Oxford, United Kingdom,

private communication.
60. H. E. Sauders and D. F. Wittenberg, Strength of thin cylindrical

shells under external pressure, ASME Trans., 53: 207, 1931.
37. C. M. Monroe et al., The CLEO II Magnetdesign, manufacture

and tests, Proc. Twelfth Int. Cryogenic Eng. Conf., Southampton,
United Kingdom, Guildford, United Kingdom: Butterworth, 773,
1988.
61. E. P. Popov, Mechanics of Materials, Englewood Cliffs, NJ: Prentice-Hall, 1959.

62. ASME Boiler and Pressure Vessel Code, Section 8, Division 1,
ANSI/ASME BPV-VIII-1.
752
HIGH-FIELD EFFECTS
63. R. Fast et al., Isogrid vacuum shell for large superconducting solenoids, in Advances in Cryogenic Engineering, Vol. 39, New York:
Plenum Press, 1993, p. 1991.
64. H. Yamaoka et al., Development of a brazed-aluminum-honeycomb vacuum vessel for a thin superconducting solenoid magnet,
in Advances in Cryogenic Engineering, Vol. 39, New York: Plenum
Press, 1993, p. 1983.
65. M. N. Wilson, Superconducting Magnets, Oxford, United Kingdom: Oxford Clarendon Press, 1983, p. 219.
66. M. A. Green, The role of quench back in the quench protection of
a superconducting solenoid, Cryogenics, 24: 659, 1984.
67. M. A. Green, PEP-4, TPC Superconducting Magnet, A Comparison of Measured Quench Back Time with Theoretical Calculations of Quench Back Time for Four Thin Superconducting Magnets, Lawrence Berkeley Laboratory Report LBID-771, August
1983, unpublished.
68. J. D. Taylor et al., Quench protection for a 2 MJ magnet, IEEE
Trans. Magn., MAG-15: 855, 1979.
69. M. A. Green, PEP-4, Large Thin Superconducting Solenoid Magnet, Cryogenic Support System Revisited, Lawrence Berkeley
Laboratory Engineering Note M5855, March 1982, unpublished.
70. M. A. Green, Calculation of the pressure rise in the cooling tube
of a two phase cooling system during a quench of an indirectly
cooled superconducting magnet, IEEE Trans. Magn., MAG-30:
2427, 1994.
71. R. G. Smits et al., Gas-cooled electrical leads for use on forced
cooled superconducting magnets, Advances Cryogenic Eng., Vol.
27, New York: Plenum Press, 1981, p. 169.
72. M. A. Green et al., A design method for multi-tube gas cooled
electrical leads for the g-2 superconducting magnets, in Advances
in Cryogenic Engineering, Vol. 41, New York: Plenum Press, 1996,
p. 573.
73. M. A. Green et al., The TPC Magnet Cryogenic System, Lawrence
Berkeley Laboratory Report LBL-10552, May 1980, unpublished.
74. J. D. Taylor and M. A. Green, Garden Hose Test, Lawrence
Berkeley Laboratory, Group A Physics Note 857, November
1978, unpublished.
75. W. A. Burns et al., The construction and testing of a double-acting bellows liquid helium pump, Proc. 8th Int. Cryogenics Eng.
Conf., Genova, Italy, Guildford, United Kingdom, IPC Science
and Technology Press, 383, 1980.
MICHAEL A. GREEN
Lawrence Berkeley National
Laboratory

Copyright
HTS FILM GROWTH

Thin films of superconductors are primarily of interest for electronics applications. High-performance superconducting thin films are essential, for example, for high-frequency passive devices and Josephson junction
circuits. Thin films are also important for fundamental studies of superconducting materials, where specially
designed experiments can take advantage of the thin layer geometry and the capability of layering various
combinations of materials, as well as lithographically defining fine features in them. Additionally, there is currently growing interest in using thin film deposition technology for deposition of thick films on polycrystalline
substrates for power applications.
In the decade since the discovery of high-temperature superconductivity (HTS) in cuprate compounds,
thin film materials have played an important role in the evolution of this field. The discovery of HTS has led
to a rapid development of many different thin film deposition techniques. HTS thin films are now routinely
made in hundreds of laboratories around the world, and thin film research is actively pursued in areas such
as new superconducting materials and epitaxial oxide heterostructures. More recently, advances in deposition
processes have brought about the synthesis of larger-area HTS films, making fabrication less expensive and
allowing for high-throughput manufacturing of single-superconducting-layer films.
In this article, we review the physical vapor deposition (PVD) techniques most frequently used for HTS
thin film synthesis. Chemical vapor deposition techniques are reviewed in another article of this encyclopedia.
Here, we review only the methods and issues that are relevant for film deposition. This does not cover the
details of the film growth processes, nor the physical properties of HTS films and their applications. The reader
is referred to other references for a more complete discussion of HTS thin films and their applications (1,2).
Materials
The HTS material of choice for applications is still YBa2 Cu3 O7 (YBCO), the first cuprate discovered to superconduct above 77 K. There are several compelling reasons why this material is still the most suitable for many
of the applications. Some of the fundamental ones are its smaller conduction anisotropy, higher superconducting critical currents in a magnetic field, and greater chemical stability than in other HTS compounds. However,
the most important reason is the ease of fabricating high-quality, single-phase YBCO thin films.
In this review, YBCO will be considered the canonical high-T c superconductor, and the discussion will
be limited to this example, partly for reasons of conciseness and partly because of the prominence of this
compound in the HTS community. Additionally, unless otherwise mentioned, one is generally considering
growth of c-axis-oriented films, i.e. where the CuO2 layers are parallel to the substrate.
Table 1 reviews the HTS compounds and methods used to date to prepare these thin films. For a review of
materials, we refer the reader to Refs. 2 and 3. There has been a substantial effort related to BiSrCaCuO
thin films (2212 and 2223 phases) by a variety of techniques, particularly in Japan. Some industrial work in
the US has focused also on TlBaCaCuO (mostly 2212), primarily for passive electronics applications. One
of the obvious reasons for looking at these other compounds is their higher critical temperatures. Nevertheless,
1
HTS FILM GROWTH
the advantages of YBCO have been hard to surpass, particularly at temperatures below 65 K. Substantial
thin-film work has also been devoted to LaSrCuO, NdCeCuO, HgBaCaCuO, and infinite-layer
compounds, mostly driven by academic interest in basic properties of these HTS materials.
Thermodynamic Issues
Film growth is inherently a nonequilibrium process. Nevertheless, thermodynamic stability is important as a
driving force for the reactions taking place during growth. HTS compounds possess large unit cells which can
have additional complications, such as a wealth of metal-atom defects and oxygen nonstoichiometries. These
complex structures require temperatures for formation close to their melting points, typically (0.80.9)T m ,
much higher than is usual for growth of epitaxial metals and semiconductors. The intricacies of film growth
and the thermodynamics of phases produced are certainly not sufficiently well understood at present. Below,
we summarize some of the key thermodynamic issues as they are currently accepted.
Oxygen Stability. HTS compounds are oxides and, as such, require the presence of oxygen during
synthesis. Although it is technically possible to deliver oxygen through the substrate, in practice, oxygen is
delivered as a gas impinging on the growth surface together with the cation species. Figure 1 shows the
pressuretemperature thermodynamic stability diagram for the bulk YBCO compound. One can see from the
diagram that a certain minimum pressure must be maintained for the stability of YBCO. Below that oxygen
pressure, depicted by the line labeled d1 , the YBCO compound is not stable and will decompose. For the
typical temperatures during film formation, this translates into a minimum oxygen pressure of about 100
mPa (1 mtorr). At higher oxygen pressures, there is another decomposition line for YBCO, labeled d2 , usually
not reached during in situ growth of thin films. During cooldown of films to room temperature, this second
decomposition is usually not observed because it is strongly limited by diffusion kinetics. However, the stability
diagram also implies that the YBCO compound is metastable at room temperature and pressure. In fact, most
HTS compounds are believed not to be thermodynamically stable at lower temperatures. References 4 and 5
have a more detailed discussion of the YBCO stability lines. Hammond and Bormann (6) argued that growth
of thin films is optimal close to the high-temperature thermodynamic decomposition line of the compound.
HTS FILM GROWTH
Fig. 1. Thermodynamic stability diagram for YBCO in oxygen pressure and temperature. The shaded elliptical areas
indicate regimes where in situ film growth for YBCO has been reported successful. The rectangles show ex situ film growth
regimes.
In fact, the empirical data confirm this, as can be seen also in Fig. 1, where successful film growth regions
are denoted. For processes that possess a more reactive form of oxygen, such as atomic oxygen or ozone, the
equivalent decomposition line is shifted compared with that in Fig. 1. Such activated oxygen species have been
proven helpful for growth of YBCO when the total pressure is lower than about 1 Pa (10 mtorr).
Compositional Phase Diagram. The HTS compounds contain typically three to five different metal
species in addition to oxygen. Some of these materials are line compounds (and this is presumed to be the case
for YBCO) and, as such, do not accept a solid solution of atoms in their chemical formula. This means that one
will always be synthesizing a number of phases in addition to the desired HTS material. For YBCO, the Gibbs
phase rule implies two other compounds as impurity phases. Figure 2 shows the present understanding of the
ternary phase diagram for YBaCu oxides at the low oxygen pressures relevant for in situ thin film growth
(7,8). The corners of the triangles define the phases which will be present for any composition in that triangle.
There are some caveats to the description given above. First of all, as already stated, film growth is not
an equilibrium process, and therefore metastable phases could be formed during synthesis. Secondly, the HTS
material that is formed during film growth may not be, and probably is not, exactly like the bulk material. This
has now been well documented in the literature for YBCO films. In fact, it is also fairly well established that
YBCO film growth does not match exactly the thermodynamic phase diagram shown in Fig. 2. In particular,
the Y2 BaCuO5 phase has not been observed in in situ grown films. On the other hand, CuO and Y2 O3 second
HTS FILM GROWTH
Fig. 2. Thermodynamic ternary phase diagram for YBaCu oxides at temperatures below 850 C and oxygen pressures
below one atmosphere.
phases are commonly observed in YBCO films prepared close to 1:2:3 stoichiometry and are not predicted by
the phase diagram.
Epitaxy
Due to the significant anisotropy of the HTS materials, the easy direction for electrical current flow is in the
ab plane, along the CuO2 planes. For devices which require current to flow parallel to the substrate, the
growth direction of the HTS films has to be in the c-axis direction. In order to achieve c-axis-oriented growth,
good epitaxial growth of all grains is necessary. Furthermore, a high-angle, in-plane grain boundary of two
c-axis-oriented grains has been found to behave as a superconducting weak link. This implies that one needs to
align all grains in the plane of the substrate as well as out of the plane. The requirements of in-plane epitaxy
are very well illustrated in the case of yttria-stabilized zirconia (YSZ) substrates where the YBCO films grow in
the c direction, but the c axis grains can have two major different orientation in the ab plane, 45 apart (15). In
the case of applications which require the current flow to be perpendicular to the substrate, a-axis growth has
been implemented. Many multilayer Josephson devices have been tried in this way. a-axis growth is typically
initiated with a lower substrate temperature, which in turn reduces the mobility of ions necessary for the
c-axis growth. Due to the lower growth temperature, films are formed which are structurally less perfect. To
overcome this problem, many workers have used a technique of starting growth at a lower temperature for aaxis nucleation, making a template, and then raising the temperature for the remainder of film deposition (16).
Substrates. The most frequently used substrate for fundamental research is SrTiO3 because of its
good thermal and lattice constant match with most HTS materials. However, high dielectric loss, small wafer
size, and high cost of SrTiO3 render it unfavorable for technological applications. LaAlO3 has emerged as an
alternative to SrTiO3 as a low-loss perovskite substrate, although its critical dielectric properties are not under
complete control. MgO substrates, on the other hand, have an even lower loss for microwave applications.
However, the reproducibility of the MgO surface quality has been a problem in the earlier days, but more
recently, the surface quality has improved. For applications where very low loss is required, MgO is the best
HTS substrate at present. MgO and LaAlO3 are commercially available in wafer sizes larger than 5 cm (2 in.).
HTS FILM GROWTH
More conventional substrates such as Si and sapphire have been used and found to be incompatible with
thick HTS thin films, mainly due to microcracking in the HTS films resulting from the thermal expansion
mismatch between the substrates and the HTS materials. The maximum thickness of YBCO one can grow
without significant microcracking is about 50 nm on Si and about 600 nm on sapphire. Table 2 shows a list of
substrates most commonly used today for growth of HTS films. References 17 and 18 have a more extensive
discussion of HTS compatible substrates and buffer layers.
In most cases, excellent epitaxial alignment can be achieved by depositing oxide buffer layers, such as
CeO2 , before HTS growth. The CeO2 buffer layer has been found to suppress the unwanted a-axis nucleation for
the c-axis growth on substrates such as SrTiO3 and LaAlO3 . On substrates such as YSZ and r-plane sapphire,
the CeO2 buffer layer promotes in-plane alignment of each c-axis grain, thereby eliminating potential highangle grain boundaries. In fact, by using an appropriate buffer layer and a seed layer, one can create a process
to produce 45 -angle grain boundaries in a controlled fashion. On Si substrates, the YSZ buffer layer was found
to grow epitaxially and to be very effective in preventing chemical reaction between HTS materials and Si.
Ion-Beam-Assisted Deposition. Recently, a new technique, ion-beam-assisted deposition (IBAD), has
been implemented to grow HTS layers on polycrystalline substrates. By bombarding the growing film surface
with an ion beam incident at a specific angle to the substrate, one is able to achieve partial in-plane crystalline
alignment of the deposited material where the substrate has none. Iijima et al. (19) first utilized this technique
to grow in-plane textured zirconia buffer layers on substrates made of polycrystalline Ni-based alloys. YBCO
was then deposited by PLD. Due to the absence of high-angle grain boundaries in YBCO, the critical currents
attained are much higher than without the IBAD process. Other groups, including the one at Los Alamos
National Laboratory, has followed on this work and achieved critical currents over 106 A/cm2 at 77 K (20).
More recently a group at Stanford University has demonstrated very good IBAD results with 10 nm thick
MgO films (21). While more development is needed to establish whether the IBAD process is a viable one for
large-scale applications, it certainly opens up a whole new range of substrate materials for HTS film growth.
Film Growth Methods

Most of the methods currently employed for fabrication of HTS thin films involve formation of the cuprate
crystal structure during film deposition. These are usually called in situ processes, and they will comprise most
cases covered in this review. Chemical vapor deposition processes, which are also in situ growth, are covered in
another chapter of this volume. Historically, the processes that were developed first, required a postdeposition
anneal, or simply postanneal, in order to crystallize the material. Such methods are now less frequently
utilized, since they are not well suited for fabrication of multilayer structures and are thus technologically
HTS FILM GROWTH
more limiting. However, they are still used for compounds such as TlBaCaCuO or HgBaCaCuO, where the in
situ processes are often not practical due to the required high vapor pressures of Hg and Tl. Finally, there are
other processes such as solgel, plasma spray deposition, and liquid phase epitaxy that are generally used for
fabrication of thick films (thicker than 1 m); these also will not be covered in this brief review.
Postanneal Growth Methods. In the post-anneal, or ex situ, growth process, the metallic elements are
deposited on a substrate in the correct composition as an amorphous or multilayer film, usually in compound
form with oxygen and possibly fluorine. Subsequent annealing in air or oxygen at a high temperature, typically
850 C for YBCO, forms an epitaxial film of the desired HTS phase (9). A very commonly used postanneal
method for growth of YBCO films is achieved by using BaF2 in the precursor film. Water is then required in the
annealing step in order to eliminate the fluorine and start the HTS growth process. Since the critical growth
step takes place separately from deposition, the actual technique used to deposit the layers is of secondary
importance. Sputtering and evaporation are typically used.
Ex situ methods have initially focused on synthesis under atmospheric oxygen pressure and correspondingly high temperatures. This region of phase space is depicted in Fig. 1 with the rectangle in the upper left
corner. Later work has shown that lower-temperature ex situ growth can also be achieved if the oxygen pressure is reduced at the same time (see also the lower rectangle in Fig. 1). Such films grown under lower oxygen
pressures were observed to have properties closer to the in situ grown films (10,11).
In situ Growth Methods. Dominant methods in use today for physical vapor deposition of HTS are
sputtering and laser ablation (also referred to as pulsed laser deposition, or PLD). Both of these methods
are most commonly done from a single target and as such, became rather popular mostly because they are
relatively simple to implement and fairly reproducible in the films they produce. In addition, targets for PLD
are relatively inexpensive and easy to fabricate, and hence, the technique is well suited for investigation of
many different materials. Less widespread today, but currently growing in popularity, are coevaporation and
molecular beam epitaxy. After the initial slow start in contending with a sufficiently oxidizing environment in
high vacuum, evaporation methods have now emerged as an established way to grow high-quality HTS films.
The following is a list of in situ physical vapor deposition techniques used for deposition of HTS compounds,
which will be covered in this review:
(1) Sputtering
On-axis magnetron
Off-axis magnetron
Inverted cylindrical magnetron
Ion-beam sputtering
(2) Laser ablation
On-axis
Off-axis
LaserMBE
(3) Evaporation
Flash evaporation
Reactive coevaporation
ReactiveMBE
For more discussion on various PVD techniques, we also refer the reader to Refs. 12,13, and 14.
HTS FILM GROWTH
Fig. 3. Schematic for the (a) on-axis and (b) off-axis sputter deposition.
Sputtering
Sputtering is a very commonly used technique for metal deposition in semiconductor and magnetic storage
industries. As such, it was applied early on to HTS materials. The first event of significance was the use of a
single composite target, which became commonplace for HTS materials because of the inherent difficulty of
reproducing the metal-atom stoichiometry. The major complication compared to more conventional sputtering
of metals is the energetic negative-ion (oxygen) bombardment of the substrate, due to the ionic nature of the
oxide target. This causes resputtering and degradation of the sample. This has been recognized even before
the advent of HTS and has been fully investigated by Rossnagel and Cuomo (22). Two types of solutions had
emerged: (1) using a higher gas pressure in order to thermalize the energetic species (23), and (2) off-axis
sputtering, where one eliminates the high-energy particles coming directly from the target (24,25). Figure 3
shows the schematic of the two processes. Other variations include on-axis unbalanced-magnetron sputtering
(26) and inverted cylindrical magnetron (ICM) sputtering (27).
The energies of ejected atoms from the sputtered targets are in the range of tens of electron volts, and
they get scattered by the background Ar and O2 gas. The angular distribution of each cation species is different,
and the sweet spot for the proper cation composition is relatively small. In order to cover a 5-cm (2-in.) wafer
uniformly from a single 5-cm target in an off-axis geometry, one has to resort to some sort of scanning method,
such as rotation of the wafer. The deposition rate is very low, not only because the deposition rate for oxides
is much lower than for metals, but also because of the off-axis geometry. It takes several hours to deposit
a few hundred nanometers of film. Larger targets and a number of sputter guns depositing simultaneously
have been used at several laboratories to increase the deposition rate. When all the deposition conditions
are optimum, sputtering has demonstrated a capability to produce YBCO films of excellent crystallinity and
surface condition. However, the deposition conditions have to be changed as the targets erode, because the
changing target surface geometry results in a different plasma distribution. In addition, the substrate heating
method for off-axis sputtering is not as simple as in the case of the off-axis laser ablation or evaporation which
will be described later.
HTS FILM GROWTH
Fig. 4. Schematic for the (a) on-axis and (b) off-axis laser ablation processes.
Laser Ablation
Laser ablation, or pulsed-laser deposition (PLD), is a relatively new technique that gained much popularity
because it is ideally suited for deposition at a high oxygen pressure. The relative ease of this technique in
depositing multicomponent oxides (and nitrides) has made it especially effective in exploring new materials
for HTS electronics, such as epitaxial dielectrics or barrier layers. A short-wavelength (170 to 260 nm) excimer
laser is focused onto a rotating target of the material to be deposited. Under the energy of the laser beam
(0.1 to 2 J per pulse), the matter emitted from the target forms a plume that carries it to the substrate at
supersonic velocities. In general, a higher gas pressure is required during laser ablation, due to the very high
energy of the vaporized material from the surface of the target. The laser plume glows brightly from the
target, and deposition is usually done near the end of the plume, about 5 cm away; see Fig. 4. Again, as in
the case of sputtering, the various species scatter differently, and therefore the sweet spot of the deposition
process is relatively small, usually one to two centimeters, depending on the geometry. The deposition rate per
laser pulse ranges from a fraction of an angstrom to a few angstroms. The technique was found to be fairly
reproducible from the early days and has been used extensively for research and development purposes during
the past ten years. Even though other techniques are potentially more manufacture-friendly, laser ablation is
still very actively used for prototyping devices made up of several complex materials, such as superconductors,
ferroelectric oxides, and magnetic oxides.
A particular problem associated with PLD is the deposition of micron-size droplets, so-called boulders, on
the grown film. These particles originate at the target and are emitted from the action of the laser pulse. A
variety of procedures have been utilized to reduce this problem, so that the boulder density can be very low in
the best films. Such procedures include target preparation (such as frequent polishing), defocusing the laser
spot, mechanically chopping the plume, and spatially filtering the beam.
Although PLD can produce high deposition rates (up to tens of nanometers per second), the area on which
one deposits is small. A straightforward way to increase the deposition area is to scan wafers over the plume
either by moving the substrate vertically and horizontally or by rotating the substrate. This method has been
tried in several laboratories and is still pursued. Another scheme utilizes a rotating cylindrical target with a
linear laser profile to obtain larger deposition areas.
Another major difficulty of these techniques is uniform heating of large wafers. Most of the heating
methods for small-area laser ablation have been to mount a substrate on a heated metal surface with silver
HTS FILM GROWTH
paste, which is difficult to extend to larger sizes. In order to overcome this difficulty of heating a large wafer,
an off-axis laser ablation technique has been developed; see Fig. 4(b). A large wafer (5 to 8 cm) can be mounted
parallel to the direction of the plume inside a relatively simple blackbodylike heater. The deposition takes place
when the atoms collide with the background pressure and are scattered to the surface of the wafer. By rotating
the wafer and selecting an appropriate pressure for the geometry, a fairly uniform deposition was achieved
over 5 cm wafers. This technique allows for simultaneous deposition on both surfaces of the wafer, which is an
important benefit for microwave applications requiring a ground plane. A drawback of this technique is its low
deposition rate because of the off-axis geometry. Typical conditions for deposition of a 5 cm wafer resulted in a
deposition rate about a factor of 10 lower than for the on-axis deposition.
By using laser fluence just enough to evaporate a few atomic layers of the target in a low-oxygen environment (less than 10 2 Pa 10 4 torr) and at the same time using a sequence of metal or metal oxide targets,
one can obtain a process similar to molecular beam epitaxy (MBE) by evaporation, here called laser MBE. Some
in situ diagnostic tools can then be used to characterize the growth of the materials. This technique has mainly
been used to grow artificially layered superconducting materials, such as the infinite-layered superconductor
(28,29).
Evaporation and Molecular Beam Epitaxy

Deposition of HTS thin films by evaporation follows a tradition of such deposition of metal films. The added
complexity here is the required partial pressure of oxygen during growth. One can distinguish several variations
in the evaporation approaches: flash evaporation, reactive coevaporation of metals, and sequential deposition by
MBE. Except for the first process, these evaporation techniques utilize individual metallic sources. However,
in attempting to control individual sources, two difficulties arise. One is the need for very fine control of
individual sources, including development of species-specific sensors. The second is the need to work at a low
enough pressure to minimize beam scattering and at the same time achieve the highly oxidizing thermodynamic
conditions required for the growth of these compounds. For an in situ process, one is confined to work at a
pressure of molecular oxygen above 100 mPa (1 mtorr) or to use a more reactive form of oxygen supply than
molecular oxygen.
Flash Evaporation. The simplest approach to evaporation of HTS materials is to evaporate the compound in small batches, i.e. in flashes of evaporant material. Since the material does not melt congruently, it
is not possible to establish a continuously constant rate of metal fluxes, but for short enough intervals one can
average out the compositional variations. Usually, the evaporant material is a powder of YBCO located in a
feed mechanism that drops small quantities onto an evaporation source (30). In most cases, the films require
a postannealing treatment to oxidize the film sufficiently, since the fast deposition does not allow for sufficient
incorporation of oxygen. This method is at present less significant.
Reactive Coevaporation. Historically, the first attempts to make films by evaporation utilized a high
molecular oxygen pressure. The problems associated with a high background oxygen pressure are rate control
of the individual species and the degradation of the sources. In order to circumvent this difficulty, several
approaches were taken. One was to accommodate a high differential pressure between the sources and the
sample by introduction of nozzles in close proximity to the sample. Another method is to utilize a more reactive
species for oxygen incorporation, such as atomic oxygen or ozone.
The evaporation technique allows one to tune the composition of the film by adjusting the relative rates
of the sources. Several groups have worked over the past years on developing process control for coevaporation
and studying YBCO film properties as a function of metal atom composition. Figure 5 shows SEM micrographs
of films with various metal compositions. The general finding is that films that are grown slightly Y- and
Cu-rich have better performance than films with exactly 1:2:3 stoichiometry. This is true for a majority of in
situ techniques and not just evaporation. The reasons for this are still not completely clearin particular,
10
HTS FILM GROWTH
Fig. 5. Morphology of YBCO films as a function of their composition during reactive coevaporation.
whether it is a materials issue or a process control issue. It is known that superconducting properties of films
degrade significantly as one goes into the Ba-rich composition. It is possible that process fluctuations around
the desired metal-atom stoichiometry are responsible for degradation of films close to 1:2:3 composition.
Probably the most technologically significant advance in HTS thin film evaporation has been the large
area heater developed for coevaporation by the group of H. Kinder at the Technical University of Munich
(Germany) and depicted in Fig. 6. They used a blackbody-type rotating disk heater, similar to the one used by
others in PLD deposition, but they added an oxidation pocket (31). This heater has a narrow slit that allows for
a differential pressure between the oxidation pocket and the rest of the chamber of about 1000:1. The deposition
and the oxidation processes are therefore separated, and a low pressure in the chamber improves the stability
of thermal evaporation sources. Uniform YBCO films were grown on wafers up to 20 cm in diameter using this
technique. More importantly, this method is significantly more cost-effective and has higher throughput than
any of the other PVD processes.
Molecular Beam Epitaxy. In general, one can distinguish between reactive molecular beam epitaxy
(MBE) approaches to HTS film deposition and mere reactive evaporation by the lower background pressure
of the former technique. Another difference is that MBE deposition is usually done sequentially rather than
simultaneously. Typically, MBE systems have multiple sources with individual shutters, as well as some in
situ monitoring tools; see Fig. 7. Reference 32 has a thorough discussion of relevant issues in reactive MBE of
HTS films. The work by a number of groups in this field has focused on careful atomic layering to produce very
smooth films, as well as customized growth of new materials and heterostructures containing these phases.
The reactive MBE technique also lends itself to careful control of chemical doping of these materials, an issue
that is very important for the superconducting properties of the cuprates.
In situ Monitoring and Diagnostics. Evaporation and, especially, MBE, with their inherent low
background pressure, lend themselves well to vacuum techniques for in situ deposition monitoring and film
HTS FILM GROWTH
11
Fig. 6. Schematic of the heater for large-area HTS deposition by evaporation.
Fig. 7. Schematic for the reactive MBE process used in HTS film growth.
diagnostic tools such as reflection high-energy electron diffraction (RHEED). RHEED has been shown to be
particularly helpful in controlling deposition of BiSrCaCuO films, where the growth occurs in blocks of
subunit cells, but where many similar phases are possible. RHEED has been less helpful to date in the growth
of YBCO, which occurs in blocks of unit cells and where intergrowths are more difficult to tailor due to the
higher stability of the primary phase. Terashima and coworkers (33) have shown that growth of YBCO can
exhibit oscillations in the RHEED pattern commensurate with unit cell deposition, suggesting that growth
proceeds in a unit-cell-by-unit-cell fashion, rather than in smaller building blocks. More recently, RHEED has
also been extended to high-pressure processes such as PLD (34).
Another area of technological development has been in the use of optical absorption techniques for measurement and control of atomic fluxes. Both hollow-cathode lamp systems (35) and tunable diode lasers (36)
have been used to monitor fluxes in situ in close proximity to the substrate. Such monitoring of individual
fluxes is a prerequisite for careful control of HTS growth.
12
HTS FILM GROWTH
Fig. 8. An STM image of a laser ablated YBCO film, showing spiral growth structures. Image courtesy of Prof. Darrell
Schlom.
Concluding Remarks
In spite of a wealth of research, growth of cuprate films has remained a complicated matter. This is due to the
materials rather complex multicomponent crystal structures. They are prone to a variety of defects and growth
morphologies. Much work still remains to be done until films are better understood and more reproducible.
As already described, HTS films are now routinely made with out-of-plane as well as in-plane alignment
on single-crystalline substrates. Aligned films are still not routine on polycrystalline substrates, but there has
been progress in this area as well. This has been manifested in high critical currents in films, particularly
for YBCO. Typical critical currents in good quality YBCO films are over 2 106 A/cm2 at 77 K and above
107 A/cm2 at low temperatures. There has been much work in characterizing defect structures, but a good
understanding of their influence on physical properties is still lacking. One growth structure that seems to be
nearly ubiquitous in YBCO films is the spiral, which forms during growth and at the core of which is a screw
dislocation (36). Figure 8 shows a scanning tunneling microscopy (STM) image of a surface of a YBCO film
with a spiral. Steps in this image have a height of one unit cell of YBCO. Such spirals have been identified as
contributing to pinning of vortices, important for high critical currents.
Maximally oxygen-doped YBCO bulk materials, made at higher temperatures than thin films, have a
superconducting transition temperature of 92 K and a resistivity of 35 cm at 100 K. Although having nominally the same crystal structure as the bulk material, YBCO films often have different electronic properties.
For example, lower T c (85 to 90 K) and expanded c-axis lattice constants (1.170 to 1.172 nm) are frequently
found in thin films. Additionally, flux-pinning centers appear to be highly dispersed in in situ grown films of
YBCO, although the structural origin of these centers has not yet been identified. It is important to have a
HTS FILM GROWTH
13
close feedback loop between film deposition and structural and electrical characterization in order to further
improve HTS films.
Technology for deposition and monitoring of HTS thin films has developed significantly over the past
decade. At the same time, the materials understanding of film nucleation, growth, and oxide interfaces has
started to unfold. Together, these developments in the future will bring new capabilities for HTS film growth.
BIBLIOGRAPHY
1. J. M. Rowell, A decade of progress towards a high temperature superconducting electronics technology, Solid State
Commun., 102: 269282, 1997.
2. T. R. Lemberger, Films of high-temperature oxide superconductors, in D. M. Ginsberg (ed.), Physical Properties of High
Temperature Superconductors III, Singapore: World Scientific, 1992.
3. C. P. Poole, H. A. Farach, R. J. Creswick, Superconductivity, San Diego: Academic Press, 1995.
4. T. B. Lindemer et al., Decomposition of YBa2 Cu3 O7 x and YBa2 Cu4 O8 for pO2 0.1 MPa, Physica C, 178: 93, 1991.
5. R. K. Williams et al., Oxidation induced decomposition of YBa2 Cu3 O7 x , J. Appl. Phys., 70: 906913, 1991.
6. R. H. Hammond, R. Bormann, Correlation between the in-situ growth conditions of YBCO thin films and the thermodynamic stability criteria, Physica C, 1624: 703704, 1989.
7. R. Beyers, B. T. Ahn, Thermodynamic considerations in superconducting oxides, Rev. Mater. Sci., 21: 335, 1991.
8. P. Karen, O. Braaten, A. Kjekshus, Chemical phase diagrams for the YBa2 Cu3 O7 family, Acta Chem. Scand., 46:
805840, 1992.
9. M. Naito et al., Thin film synthesis of the high T c oxide superconductor YBa2 Cu3 O7 by electron beam codeposition, J.
Mater. Res., 2: 713725, 1987.
10. R. Feenstra et al., Effect of oxygen pressure on the synthesis of YBaCuO7 x thin films by post deposition annealing, J.
Appl. Phys., 69: 65696585, 1991.
11. J. M. Phillips et al., Comparison of Ba2 YCu3 O7 thin films grown on various perovskite substrates by coevaporation,
J. Mater. Res., 7: 26502657, 1992.
12. R. G. Humphreys et al., Physical vapor deposition techniques for the growth of YBa2 Cu3 O7 thin films, Supercond. Sci.
Technol., 3: 3852, 1990.
13. J. A. Alarco et al., High quality YBCO thin filmslaser deposition, co-evaporation, and device fabrication, Phys. Scripta,
44: 95101, 1991.
14. C. H. Stoessel et al., Thin-film processing of high-T c superconductors, J. Supercond., 6 (1): 117, 1993.
15. S. M. Garrison et al., Observation of two in-plane epitaxial states in YBa2 Cu3 O7 films on yttria-stabilized ZrO2 , Appl.
Phys. Lett., 58: 21682170, 1991.
16. H.-U. Habermeier et al., Preparation and properties of YBCO thin films with the c-axis aligned in the film plane, in L.
Correra (ed.), High Temperature Superconductor Thin Films, Amsterdam: North-Holland, 1992, pp. 343352.
17. A. Perrin, M. Sergent, in A. Narlikar (ed.), Studies of High Temperature Superconductors, vol. 7, Commack, NY: Nova
Science Publishers, 1991.
18. J. Qiao, C. Y. Yang, High-T c superconductors on buffered silicon: Materials properties and device applications, Mater.
Sci. Eng., R14: 157202, 1995.
19. Y. Iijima et al., In-plane aligned YBa2 Cu3 O7 x thin films deposited on polycrystalline metallic substrates, Appl. Phys.
Lett., 60: 769771, 1992.
20. X. D. Wu et al., High current YBa2 Cu3 O7 d thick films on flexible nickel substrates with textured buffer layers, Appl.
Phys. Lett., 65: 19611963, 1994.
21. C. P. Wang et al., Deposition of in-plane textured MgO on amorphous Si3 N4 substrates by ion-beam-assisted deposition
and comparisons with ion-beam-assisted deposited yttria-stabilized-zirconia, Appl. Phys. Lett., 71: 29552957, 1997.
22. S. M. Rossnagel, J. J. Cuomo, Negative ion effects during magnetron and ion beam sputtering of YBa2 Cu3 Ox , in Thin
Film Processing and Characterization of High-Temperature Superconductors, American Institute of Physics Conf. Proc.
165, New York: AIP, 1988, pp. 106113.
23. U. Poppe et al., Direct production of crystalline superconducting thin films of YBa2 Cu3 O7 by high-pressure oxygen
sputtering, Solid State Commun., 66: 661665, 1988.
14
HTS FILM GROWTH
24. C. B. Eom et al., Synthesis and properties of YBa2 Cu3 O7 thin films grown in situ by 90 degrees off-axis single magnetron
sputtering, Physica C, 171: 354382, 1990.
25. J. R. Gavaler et al., Critical parameters in the single-target sputtering of YBa2 Cu3 O7 , J. Appl. Phys., 70: 43834391,
1991.
26. N. Savvides, A. Katsaros, In situ growth of epitaxial YBCO thin films by on-axis unbalanced direct current magnetron
sputtering, Appl. Phys. Lett., 62: 528530, 1993.
27. X. X. Xi et al., Preparation of YBa2 Cu3 O7 thin films by inverted cylindrical magnetron sputtering, J. Less-Common
Metals, 151: 349355, 1989.
28. T. Kawai et al., Superconducting artificial lattices grown by laser MBE, in L. Correra (ed.), High Temperature Superconductor Thin Films, Amsterdam: North-Holland, 1992.
29. H. Koinuma, M. Yoshimoto, Controlled formation of oxide materials by laser molecular beam epitaxy, Appl. Surface
Sci., 75: 308319, 1994.
30. Y. Yasuda et al., YBaCuO superconducting thin films prepared by plasma-assisted flash evaporation, Appl. Phys.
Lett., 55: 307309, 1989.
31. P. Berberich et al., Homogeneous high quality YBa2 Cu3 O7 on 3 and 4 substrates, Physica C, 219: 497, 1994.
32. D. G. Schlom, J. S. Harris, Jr., MBE growth of high T c superconductors, in R. F. C. Farrow (ed.), Molecular Beam
Epitaxy: Applications to Key Materials, Park Ridge: Noyes, 1995.
33. T. Terashima et al., Reflection high-energy electron diffraction oscillations during epitaxial growth of high-temperature
superconducting oxides, Phys. Rev. Lett., 65: 26842687, 1990.
34. G. J. H. M. Rijnders et al., In situ monitoring during pulsed laser deposition of complex oxides using reflection high
energy electron diffraction under high oxygen pressure, Appl. Phys. Lett., 70: 18881890, 1997.
35. J. N. Eckstein, I. Bozovic, G. F. Virshup, Atomic layer-by-layer engineering of high-T c materials and heterostructure
devices, Mater. Res. Soc. Bull., 19 (9): 4450, 1994.
36. W. Wang et al., Direct atomic flux measurement of electron-beam evaporated yttrium with a diode-laser-based atomic
absorption monitor at 668 nm, Appl. Phys. Lett., 71: 3133, 1997.
37. C. Gerber et al., Screw dislocations in high-T c films, Nature 350: 279280, 1991.
KOOKRIN CHAR
VLADIMIR MATIJASEVIC
Conductus, Inc.

Copyright
HTS JOSEPHSON JUNCTION DEVELOPMENT

Josephson junctions are the fundamental building blocks for a variety of superconducting electronics applications, including high-speed, low-power digital logic, and sensitive magnetic field and high-frequency electromagnetic detectors. A Josephson junction consists of a weak connection between two superconductors
which exhibits the Josephson effects (described below). While low-temperature superconductor (LTS) Josephson junction technology is well-developed, high-temperature superconductor (HTS) Josephson junctions are
still relatively immature. Nonetheless, extensive HTS junction fabrication efforts are in progress due to the
possibility of applying Josephson effects at temperatures compatible with reliable, low-cost refrigerators. In
this article we discuss the more common approaches to HTS junction fabrication and optimization, with a focus
on potential digital circuit applications.
Introduction to Josephson Junctions

According to the BardeenCooperSchreiffer (BCS) theory of superconductivity, the electrons in a superconductor are weakly bound into Cooper pairs and can be described mathematically by a complex wave function with
spatially dependent amplitude and phase. The fact that all of these electrons occupy a macroscopic quantum
state leads to several profound properties of superconductivity, such as zero resistance, quantization of the
magnetic flux threading a hole in a superconductor, and the Josephson effects.
If two pieces of superconductor, each characterized by their own wave function, are brought very close
together, but separated by a thin insulating layer, then the two wave functions can overlap. It was predicted
by Brian Josephson (1) that this overlap would lead to novel phenomena associated with the dissipationless
tunneling of Cooper pairs through the insulating barrier. The first of these, the direct current (dc) Josephson
effect, is that a dc current can flow through this Josephson junction without the appearance of a voltage
drop across the junction, and furthermore that the magnitude of this supercurrent is a function of the phase
difference between the two electrodes. The maximum value of the supercurrent is called the critical current,
Ic . Josephsons second prediction was that if a dc voltage were applied across the junction, then the phase
difference would evolve at a rate proportional to the voltage, leading to a time-oscillating Cooper pair current.
This is the alternating current (ac) Josephson effect.
Most of the above discussion can also be applied in cases where two superconducting electrodes are
separated not by an insulator, but by other nonsuperconducting or weakly superconducting regions. These
include normal metals, semiconductors, or even extremely narrow superconducting constrictions. This broader
class of Josephson devices is generally known as weak links, and essentially all HTS Josephson devices
being developed fall under this heading. In particular, the use of a normal metal, or of a superconductor above
its transition temperature, is the focus of much of the development work, and thus of the discussion in this
article.
1
HTS versus LTS Materials

Fabrication of HTS Josephson junctions is complicated by the materials properties of the high-temperature superconductors. Low-temperature superconductor junctions are typically produced using polycrystalline metallic
Nb as the superconductor. In contrast, the new HTS materials are multicomponent ceramic oxides with four or
more elements, such as yttriumbariumcopperoxide (YBa2 Cu3 O7 x or YBCO). The oxide superconductors
have perovskite-based crystal structures with two to three copperoxygen planes in a layered structure within
a unit cell. As a consequence, these materials are anisotropic, with higher critical current densities and longer
superconducting coherence lengths parallel to the CuO2 planes. The superconductive coherence length is the
characteristic length over which the Cooper pair density decays, as for example in approaching an interface
between a superconductor and a normal metal. It also corresponds to the average physical separation of the
two electrons which make up a Cooper pair. For a clean material (mean free path much longer than coherence
length) it is given by the approximate expression 0.18vF /kB T c , where vF is the Fermi velocity and T c is the
superconducting transition temperature (1). In contrast to most LTS materials, the coherence lengths in the
oxide superconductors are on the scale of atomic dimensions (30 parallel to the copper oxide planes, and
4 perpendicular to the planes). Since damaged layers thicker than the coherence length result in degraded
superconducting properties, surface cleaning and film growth at device interfaces are especially critical in the
HTS materials.
Also unlike in most low temperature superconductors, grain boundaries in the HTS oxide superconductors
lead to reduced critical current densities, and in fact can behave like Josephson junctions (see the section
entitled HTS Josephson Junction Types). Because of the anisotropy and grain boundary problems associated
with the high-temperature superconductors, epitaxial (single crystal) HTS films grown at high temperatures
(650 C to 800 C) are required for many applications. High-temperature growth and the need to maintain high
epitaxial quality in each layer of multilayer structures make HTS junction and circuit fabrication considerably
more complex than the corresponding LTS processes.
Selection of YBCO for HTS Junctions

Nearly all Josephson junction development in HTS materials has been based on YBCO or nearly identical
compounds with another rare-earth element in place of yttrium. The superconducting critical temperature, T c ,
for this family of compounds is approximately 90 K. The main advantage of using YBCO compared with other
HTS material families is the low volatility of all cation elements which permits films to be grown in a single
step from a sourcefor example, a laser ablation target or sputtering targetfabricated with a stoichiometric
ratio of yttrium, barium, and copper.
In contrast, films of the HTS families with the highest critical temperatures, TlBaCaCuO (TBCCO)
and HgBaCaCuO (HgBCCO), are typically prepared in a two-step process in which a Tl- or Hg-deficient
as-deposited film is annealed to the point of partial melting in a container sealed to obtain a high pressure
of Tl or Hg vapor. Although grain boundary junctions can be formed by using this technique with bicrystal
substrates or substrates with patterned steps, it is not extendible to multilayers. In the case of the BiSr
CaCuO (BSCCO) or BaKBiO (BKBO) families, films can be produced in a single step, but the volatility
of K and Bi has limited film reproducibility and has discouraged junction development efforts. The TBCCO,
HgBCCO, and BSCCO materials are also more highly anisotropic than YBCO, which introduces additional
complications in 3-D structures.
Fig. 1. Currentvoltage characteristic of a Josephson weak link in the zero-capacitance limit, without thermal noise,
showing definitions of the critical current, Ic , and normal resistance, Rn . The dashed line is the high-current asymptote of
the IV curve.
Basic HTS Josephson Junction Characterization

The currentvoltage characteristic for an ideal Josephson junction described by the Resistively-ShuntedJunction (RSJ) model is shown in Fig. 1. The RSJ model adequately describes the behavior of many HTS
junctions, as explained in more detail in the section entitled Josephson Effects. The primary junction parameters of interest are the critical current, Ic , the normal state resistance, Rn , and the product of these two factors,
Ic Rn . The critical current is the maximum current that can flow through the junction without a voltage drop,
while the normal state resistance is given by the inverse of the slope of the IV characteristics at a few times
Ic , as shown in Fig. 1. Rn is due to dissipation in the device, for example, by current flow through the normal
metal in a superconductor/normal metal/superconductor weak link. Ic and Rn are often expressed in terms
of area-normalized quantities: the critical current density, J c = Ic /A, and the resistancearea product, Rn A.
Characterization for an HTS Josephson junction typically includes (a) measurement of the currentvoltage
(IV) characteristics as a function of temperature and (b) measurement of Ic modulation in a magnetic field.
In some cases, the IV characteristics are also studied under microwave irradiation as a measure of the ac
Josephson effect. In a well-behaved HTS junction the IV shape will show the concave-up curvature consistent
with the resistively shunted junction model. The temperature dependence of Ic and Rn gives information on
the nature of coupling across the Josephson junction (e.g., proximity-effect coupling or tunneling), while the
dependence of Ic on magnetic field provides a gauge of the uniformity of Cooper pair transport across the weak
link. More comprehensive Josephson junction studies can also include determination of the J c dependence on
tunnel barrier or interlayer thickness, as well as measurement of J c uniformity across multiple junctions on a
chip (important for circuit applications).
Survey of Junction Congurations

Josephson effects in HTS materials were first observed at naturally occurring grain boundaries in films and
in weak links between two HTS samples created by either breaking a sample in vacuum and coupling across
the vacuum gap or bringing two pieces in contact and coupling through a surface layer that was degraded by
reaction in air. Most of the configurations used since that time to fabricate junctions with controllable and
reproducible properties are shown in Fig. 2. All configurations require epitaxial films to prevent naturally
Fig. 2. The wide range of fabrication methods for HTS Josephson junctions can be loosely grouped into four classes: grain
boundary junctions, damage-induced junctions, deposited-interlayer junctions, and microbridges. The deposited-interlayer
approaches are thought to be the most promising for applications requiring more than a few junctions.
occurring and randomly placed grain-boundary junctions from interfering with the engineered junctions. The
fill pattern for HTS film layers in Fig. 2 indicates the orientation of CuO planes.
The junction configurations shown in Fig. 2 are grouped in the four categories of grain-boundary weak
links, damage junctions, deposited interlayer junctions, and microbridges. The chronology of development was
that single-HTS-film configurations that did not require special substrate treatments were explored first
for example, the junctions in Figs. 2(c), 2(f), and 2(j). These were followed by more complex processes where
substrates were fabricated to promote the formation of a junction during HTS film growth as in Figs. 2(a), 2(b),
and 2(g). Since the configurations in Figs. 2(d), 2(h), and 2(i) are direct analogues of LTS junctions, they were
identified as candidate structures soon after discovery of HTS, but development did not begin until multilayer
growth and patterning techniques were developed.
Application Requirements
The key electrical parameters of an individual junction, which determine its suitability for a given application,
are principally its Ic and RN and secondarily such parameters as capacitance C and inductance L. For applications which require many junctions, such as digital circuits, the uniformity of these parameters is of critical
importance, while for sensor applications, such as superconducting quantum interference devices (SQUIDs),
low intrinsic noise is a key characteristic.
For digital circuits a number of constraints dictate the range of parameters required. First, the product of Ic and a typical gate inductance Lg should be not much more than a flux quantum, 2.07 10 15
Wb (2.07 pH-mA). Since it is difficult to fabricate gates with inductance less than about 4 pH in HTS (see
the section entitled Circuit Integration), Ic can be no more than about 0.5 mA. At the same time, Ic should be
large enough that the Josephson energy, Ic /2e, is much larger (say by a factor of 100 to 1000) than the thermal
energy, kB T, otherwise there will be too many thermally induced, erroneous switching events. For operation
at 40 K this dictates that Ic be greater than 0.04 mA to 0.4 mA. At the same time, the product of Ic and Rn
establishes the maximum reliable operating frequency of the circuit, 2eIc Rn /h. For Josephson devices to be
competitive, this frequency must be at least tens of gigahertz. This means that Ic Rn must be greater than
approximately 0.3 mV, and Rn must be of order 1 to 10 . While different arguments are applied for sensor
applications of SQUIDs, such as magnetometry, the resulting requirements for Ic and Rn are quite similar.
Josephson Effects
The dc and ac Josephson effects introduced in the section entitled Introduction to Josephson Junctions can
be stated mathematically as follows:
where 1 and 2 are the phases of the wavefunction in the two electrodes. In the presence of a magnetic field,
with vector potential A, Eq. (1a) is generalized as follows:
where 1 and 2 refer to two specific points on opposite electrodes, and the integral is taken along a straight
line between those two points (1). While the phase difference across the junction cannot be directly measured,
Josephsons predictions have several measurable consequences. For example, as a result of the periodic depen-
Fig. 3. Currentvoltage characteristics of a YBCO/CoYBCO/YBCO edge SNS junction, showing Shapiro steps in response
to 13 GHz radiation. Such steps are a manifestation of the ac Josephson effect.
dence of supercurrent on phase difference, and the field-dependence of the phase difference given by Eq. (1d),
the critical current of a spatially extended junction displays a dependence on the magnetic field described by
the Fraunhofer pattern characteristic of single-slit diffraction in optics: Ic |sin(/0 )/|, where is the magnetic flux through the junction and 0 = h/2e is the superconducting flux quantum. The ac Josephson effect is
responsible for Shapiro steps, which are constant voltage steps in the IV characteristics of the junction in the
presence of microwave radiation. Such steps occur when the average voltage across the junction has values such
that the oscillation frequency of the supercurrent is a multiple of the frequency of the applied radiation. This
effect is the basis for the definition of the standard volt, which uses a precisely known frequency to generate a
fixed voltage across a series array of many junctions. Figure 3 shows an example of Shapiro steps induced in a
YBCO/Cobalt-YBCO/YBCO edge superconductor/normal metal/superconductor (SNS) junction (see the section
entitled Edge-Geometry Weak Links) by 13 GHz radiation.
Real Josephson junctions can be modeled by the so-called resistively and capacitively shunted junction
(RCSJ, or, for small capacitance, just RSJ) model, which takes the ideal Josephson element described by Eq.
(2) and shunts it with a resistor and a capacitor. The resistor models the path by which quasiparticles cross
the junction. For HTS junctions we can use the RSJ model with a linear resistor and zero capacitance for
comparison with the data. In this case the RSJ model, without thermal noise, predicts a hyperbolic shape for
the IV characteristic:
where V is the time average of the instantaneous voltage, V(t), which oscillates at the Josephson frequency
with amplitude Ic Rn . The dc behavior is illustrated in Fig. 1. In the presence of thermal noise the sharp voltage
onset at Ic is smeared out.
Hts Josephson Junction Types

As described in the section entitled Survey of Junction Configurations, there are a wide variety of HTS
Josephson junctions. In fact, it has proven relatively easy to fabricate HTS devices which exhibit Josephson
effects, with the technology ranging in sophistication from naturally occurring grain boundary weak links to
all-epitaxial structures incorporating superconductors, insulators, and deposited interlayers. However, it has
proven difficult to meet the requirements of some of the more demanding applications. For example, single
flux quantum (SFQ) digital circuits require junctions with high resistance, high Ic Rn products, and 1 Ic
spreads less than 10%. No junction technology to date has consistently met these constraints. At present the
most widely used HTS junctions are grain boundary weak links and SNS edge junctions with doped YBCO
interlayers. The grain boundary devices have excellent characteristics but relatively poor Ic spreads, while the
SNS edge junctions are more difficult to produce but appear to be the most viable candidate for fabrication of
complex circuits.
Grain Boundary Junctions.

Naturally Occurring Grain Boundary Junctions. Within a few months of the discovery of YBCO, Josephson effects were measured in bulk ceramic samples. An example of these measurements can be found in Robbes
et al. (2), where a polycrystalline ceramic pellet was bonded to a glass slide, polished to a thickness of 0.25 mm,
and constrictions were engraved to pattern a dc SQUID. Shapiro steps and modulation of the critical current
by a magnetic field were observed.
Although randomly oriented grain boundary junctions in polycrystalline material are too poorly controlled
to be practical for electronics, they are the most important obstacle in fabrication of HTS conductors for
carrying large currents for large-scale, high-power applications. Many of the possible mechanisms for weak
link formation at grain boundaries, such as impurities or second phase formation, have been eliminated by
careful synthesis. However, the grain boundary continues to be the subject of research on topics such as the
relationship between local strain fields and oxygen deficiency (3).
Bicrystal Grain Boundary Junctions. Dimos et al. (4) were the first to fabricate grain boundaries in
HTS materials with controlled angles of misalignment. They cut and polished bulk SrTiO3 single crystals with
symmetric [100] tilt boundaries. The two crystals were sintered together with parallel [001] axes. The bicrystal
was then cut and polished to obtain [001] surfaces for growth of epitaxial c-axis-oriented YBCO films. At least
on a macroscopic scale, the misorientation angle in the basal plane of YBCO matched the misorientation angle
in the SrTiO3 bicrystal.
The bicrystal experiments elegantly quantified the need for biaxial alignment in YBCO high-current
conductors by demonstrating the rapid decrease in J c with increasing grain misalignment. They also provided
a route for synthesis of grain boundary junctions, so-called bicrystal junctions, with controlled misalignment
between grains and controlled placement of junctions along a single line. Strontium titanate bicrystals with
24 and 37 tilt rotationsangles where particularly well-defined, clean interfaces can be formedare commercially available.
Bicrystal junctions are still in use for low-junction-count applications, usually where an integrated HTS
groundplane is not required. Specifically, most commercially available dc SQUIDs are fabricated in this way.
Bicrystal junctions are the simplest and least expensive way for a university laboratory to pattern a few
junctions for research. However, in addition to the obvious disadvantage imposed by placement of junctions
along a single line, junction uniformity has not matched the level obtained by other junction configurations.
Several different explanations for the lack of uniformity are based on the observation that the junction interface
meanders along the substrate bicrystal boundary on a microscopic scale as YBCO grains overgrow the boundary
from each side.
Biepitaxial Grain Boundary Junctions. Biepitaxial junctions are a variant of bicrystal junctions in
which the bicrystal template for junction formation is provided not by a bulk bicrystal substrate but by the
tendency of particular buffer layers to grow with different epitaxial orientations depending on the presence
or absence of an intermediate seed layer. The first implementation was by Char et al. (5), who observed
that SrTiO3 (001) grew on R-plane sapphire, Al2 O3 (1102), when there was no seed layer with in-plane parallel directions, SrTiO3 [110]//Al2 O3 [1120]. However, when there was an MgO(001) seed layer present, the
in-plane orientation was SrTiO3 [100]//MgO[100]//Al2 O3 [1120]. So, by patterning the MgO layer with conventional lithography, 45 grain boundaries in the SrTiO3 (001) buffer layer could be placed in arbitrary positions.
Other combinations of substrates, seed layers, and buffer layers were later found to work. All of them
had in common with the original concept that the materials were oxides with dissimilar crystal structures and
large lattice mismatches so that
times one lattice constant provided as close a match with a 45 in-plane
rotation as could be achieved by cube-on-cube growth. While the virtue of this technique compared to bicrystal
junctions is the ability to arbitrarily place grain boundary junctions on a mask, its disadvantage is that 45
is larger than the angle normally selected for bicrystal junctions. At this large angle, Ic Rn products at 77 K
were on the order of just 10 V to 20 V and critical currents per unit junction width approximately 3 A/m.
Although some multilayer SQUIDs have been based on biepitaxial junctions, there has been little work on
their integration with an HTS groundplane for digital circuits.
Step-Edge Grain Boundary Junctions. When a c-axis YBCO film is grown over a sharp step in a cubic
(or nearly cubic) substrate or deposited insulator, the YBCO grows such that the c-axis is perpendicular to the
local principal crystalline axes of the substrate material (6), forming a pair of grain boundaries (at the top and
bottom of the step), which behave electrically as one or two Josephson junctions. Figure 2(b) shows the simplest
example, where each grain boundary (GB) is of the symmetric, (103)(103) type. Figure 4 shows a transmission
electron micrograph of a cross section of such a boundary, formed at the top of a step in a single-crystal LaAlO3
substrate. In practice the lower GB often consists of a mixture of orientations, including (010)(001).
Such devices show IV characteristics with rather ideal RSJ shape, and they also show the signatures
of true Josephson behavior, including critical current modulation with field (though often nonideal), Shapiro
steps, and the emission of Josephson radiation. Values of J c are fortuitously close to those required for digital
and SQUID applications in the 65 K to 77 K range, with Ic Rn values of up to several hundred microvolts at 77
K (7). Electrical characteristics, and stability with thermal cycling, are improved by paying close attention to
forming a sharp, clean stepfor example, by the use of ion milling with a very hard mask, such as amorphous
carbon. While there have been reports of sets of such junctions with J c spreads as narrow as 5% (1), typical
spread values are 30% or above, making the long-term viability of this junction technique for complex circuits
doubtful. For SQUIDs this is not an issue; and the simplicity of fabrication, along with the low measured noise
of such junctions, makes them attractive. Their incorporation into multilayers is relatively straightforward.
The nature of the Josephson junction in this geometry is controversial. For example, GBs similar to those
observed in step edge junctions have been deliberately fabricated in planar YBCO films, by the use of seed
layers to control film orientation (see the section entitled Biepitaxial Grain Boundary Junctions), and have
been found to not exhibit weak link behavior. Also, while some of the most definitive work on these junctions
has attributed the weak link to a (010)(001) GB at the bottom of the step (6), recent measurements which probe
the GBs individually (by use of a narrow YBCO lead, formed by shadowing, along the step face) suggest that
it is the symmetric GB at the top which is the weakest.
Damage-Induced HTS Josephson Junctions.

Ion-Damaged Weak Links. Weak links based on ion damage can be classed into two categories: (i)
single-layer devices in which a focused ion beam (FIB) creates a weak link [Fig. 2(c)]; and (ii) multilayer
structures where ion surface damage of a base electrode produces a Josephson junction after deposition of a
counterelectrode [Fig. 2(d)]. FIB junctions have been produced by a number of groups including Zani et al.
(8), who used a 300 keV Si ion beam. The main attraction of the FIB approach is its relative simplicity, since
only a single HTS film is required. However, because the typical FIB spot size is much greater than HTS
coherence lengths, weak link behavior in these devices is presumably due to local variations in damage which
lead to a parallel array of filamentary connections across the damaged region. Consequently, FIB-defined weak
Fig. 4. TEM crosssection of the upper, symmetric grain boundary in a step-edge grain boundary junction formed at a
step in a LaAlO3 substrate. (Courtesy of Claire Pettiette-Hall of TRW.)
links often show nonideal IV characteristics, as well as excess SQUID inductance. While this technology has
produced high-resistance junctions and working SQUIDs up to 60 K, it has been largely superseded by other
junction processes.
The second type of HTS Josephson junctions based on ion damage rely on ion bombardment at relatively
low energies to create surface damage on a base electrode, which is then overlayed by an epitaxial counterelectrode. Work in this area has included room-temperature oxygenargon and oxygenfluorine-based plasma
treatments (9), as well as low-energy Ar and Xe ion mill processing (10). The surface ion damage approach
avoids the complication of a deposited interlayer and has produced high-quality Josephson junctions with
10
Ic Rn products up to 250 V at 77 K. However, the weak links working at 77 K also typically have current
densities and resistances which are not well-suited to digital and SQUID applications (J c > 105 A/cm2 and
Rn A < 2 10 9 -cm2 ). More recently it has been found that radio-frequency (RF) plasma exposure at 400 to
500 C can produce higher resistance weak links with J c 1 spreads as small as 8% at 4.2 K (11).
Electron-Beam-Damaged Weak Links. Electron-beam-damaged junctions are fabricated by writing a
line across a narrow, 2 m to 10 m wide bridge with an electron beam in a transmission electron microscope
with 120 keV or 350 keV beam energies (12). The electrons are thought to disorder oxygen ions located in
the chains of copper and oxygen which are present in YBCO. The higher damage energy, 350 keV, results in
a weak link that is stable at room temperature, whereas disorder caused by 120 keV electrons starts to heal
even at room temperature. In either case, the evidence for oxygen disorder is mainly found in (a) the annealing
behavior of these junctions which tends to a recovery of the initial, undamaged bridge properties over time
with temperatures <400 C and (b) energy barriers of 1.1 eV which are characteristic of the chain oxygen sites.
Targeted junction properties can be achieved by adjusting the total damage dose and the junction length.
Critical current modulation in a magnetic field indicates that the damage is uniform compared to the scale of
the junction width. The most sophisticated digital circuit demonstrations that have been performed without
integrated HTS groundplanes have used this type of junction. A possible long-term limitation of any sequential
process of writing to form junctions using either electron or focused ion beams is that writing time can become
significant as circuit complexity and junction counts increase. The special equipment required, the lack of
junction stability, and the fact that groundplane integration has not yet been demonstrated are other probable
reasons why this technique has been used in university laboratories but has not been adopted by industrial
labs.
Poisoned Weak Links. An alternative technique for weakening a small area is to dope an HTS film
with a small patch of a material that degrades the superconducting properties. In the example of Simon et al.
(13), a 10 nm thick, several-micron-wide Al stripe was patterned on a substrate before deposition of a 200 nm
thick YBCO film. A weak link formed where a bridge patterned in the YBCO film crossed the Al stripe. The Al
dopant depressed the critical current of the YBCO by two orders of magnitude.
This technique has the benefit of a simple fabrication process but has not been used in recent years. It
is included here for completeness and to illustrate two of the factors that influence junction reproducibility:
control of bridge length and control of interface thickness. Since the ideal length of a coupling region between
YBCO banks is no greater than tens of nanometers, it is preferable that the length is determined by a
reproducible scale such as a film thickness or step height rather than the width of a patterned line. The deposited
interlayer junction configurations described below are designed to use these better-controlled length scales and
to maintain a thinner interface layer between undamaged YBCO banks and a more weakly superconducting
region than one can achieve by an interdiffusion process.
Deposited-Interlayer Josephson Junctions.

Step-Edge SNS Junctions. Like the step-edge grain boundary (SEGB) junction, the step-edge SNS
junction is formed at a step, either in the substrate or a deposited insulator, but in this case the deposition is
performed directionally, from behind the step, so that the YBCO is discontinuous. This discontinuity is bridged
by an in situ deposited noble metal such as Ag (14,15), making the nominally SNS structure illustrated in
Fig. 2(g). A cross-sectional scanning electron micrograph of an actual device is shown in Fig. 5, which clearly
shows the YBCO to be discontinuous across the step. In practice the resistance of such a device is dominated
not by the normal metal itself, but by the interfaces with the YBCO. By using a directional ion mill from
behind the step, it is possible to remove most of the normal metal, leaving only that which is shadowed by the
step. In this case, Rn can be as much as 10 , and Ic Rn as high as several millivolts, at 4.2 K. Useful SQUIDs
operating at 77 K, with either flip-chip or integrated pick-up coils, are routinely fabricated, with modulation
voltage of several microvolts. Statistics for large numbers of junctions are not available to evaluate the critical
current spreads. The directionality of the fabrication process means that this junction type does not scale well
for complex circuit fabrication.
11
Fig. 5. Cross-sectional electron micrograph of a step-edge SNS junction of the type shown in Fig. 2(g), clearly showing
the discontinuity of the YBCO film (dark) over the step, bridged by the AuAg alloy (light). (Courtesy of Mark DiIorio of
Magnesensors Inc.)
Edge-Geometry Weak Links. The HTS edge-geometry weak link is also known as an edge junction or
ramp junction and is shown in Fig. 2(h). This structure typically consists of a c-axis-oriented HTS base electrode
film overlaid by a thick insulator (not necessarily epitaxial) with an edge produced in the bilayer by ion milling
or more rarely by wet etching. An epitaxial normal metal interlayer is deposited on the exposed edge followed
by growth and patterning of the HTS counterelectrode. The SNS edge junction configuration is the most widely
used HTS junction approach, because this geometry offers a number of advantages, including the fact that the
critical superconductor/normal metal (SN) interfaces are located on the longer-coherence-length surfaces of the
superconducting electrodes. The edge structure also enables the fabrication of very-small-area devices using
conventional photolithography because one of the device dimensions is determined by the thickness of the
base electrode film. Not surprisingly, however, successful fabrication of uniform sets of high-quality SNS edge
junctions requires great care in base electrode edge formation. Shallow edges (<40 ) are needed to avoid grain
boundary formation in the counterelectrod, and the edge properties must be independent of edge orientation
for ease of circuit layout. The most common approach to edge formation uses argon ion milling with rotating
substrates and a tapered photoresist mask. Base electrode edge cleaning before growth of the normal metal
and counterelectrode is also important and is usually done by low-energy ion milling or by etching in a dilute
(<1%) solution of bromine in alcohol.
As with trilayer junctions (discussed below), the edge junction approach utilizes an all-epitaxial stack
of base electrode, normal metal interlayer, and counterelectrode which places tight constraints on potential
interlayers. Suitable interlayers must be lattice-matched to the HTS electrodes, grow without pinholes, and be
chemically compatible with the superconductors at the elevated temperatures necessary for epitaxial growth.
These requirements point to the use of materials with similar structures and compositions to YBCO such as
PrBa2 Cu3 O7 (PBCO) or YBaCu3 x Cox O7 (Co-doped YBCO) (16), the two most commonly used interlayers. Note
that the interlayers are often referred to as normal metals, but, in fact range from semiconductors (PBCO)
to superconductors operating above their transition temperature (Co-doped YBCO).
IV characteristics for a set of Co-doped YBCO edge junctions produced at Northrop Grumman are shown
in Fig. 6 (17). The normal metal layer is 50 of YBaCu2.8 Co0.2 O7 . For this chip the average junction parameters,
12
Fig. 6. IV characteristics at 65 K for a chip with junction parameters suitable for SFQ logic. There are nineteen 4 m
wide junctions with 50 Co-doped YBCO interlayers and an average resistance of 0.97 (1 = 6%). The average Ic is
327 A (1 = 13%) and the average Ic Rn product is 315 V (1 = 9%).
Fig. 7. Critical current modulation at 55 K for a 4 m wide SNS junction with La-YBCO base electrode, with the magnetic
field normal to the substrate. The normal metal layer is 50 of Co-doped YBCO, and the junction resistance is 1.1 .
at 65 K, are J c = 4.1 104 A/cm2 with 1 = 13%, Ic Rn = 315 V with 1 = 9%, and Rn A = 7.7 10 9 -cm2
with 1 = 6% (Rn = 0.97 ), values which are suitable for small-scale SFQ circuit applications. Note that
this resistance value is surprisingly high for SNS junctions, as will be discussed in more detail in the section
entitled Advanced Issues. As shown in Fig. 7, the critical current modulation in edge-geometry weak links
can approach the ideally expected |(sin x)/x| behavior, indicating fairly uniform pair current transport through
the normal metal layer (17). Good results have also been obtained for junctions utilizing PBCO and Ga-doped
PBCO interlayers. For example, Verhoeven et al. (18) have produced Ga-doped PBCO edge junctions with
Ic Rn products up to 8 mV at 4.2 K. The data suggest that transport in these devices takes place by resonant
tunneling.
13
Trilayer-Geometry Weak Links. Trilayer-geometry Josephson junctions [Fig. 2(i)] consist of a layered
sandwich structure of HTS, interlayer, and HTS epitaxial films, as well as associated wiring and insulator
layers. This approach most closely resembles the highly successful LTS NbAl tunnel junction process and
consequently has attracted considerable attention. However, the trilayer process does require additional epitaxial insulator and HTS layers for wiring. Another potential disadvantage of this approach is the relatively
large area of the devices, which can lead to unacceptably low values of Rn in many SNS processes, as well as
a greater sensitivity to defects such as pinholes. Some of the earliest work in HTS deposited-interlayer weak
links utilized c-axis oriented YBCO electrodes with a PBCO interlayer (19). Researchers at Varian have used
a molecular beam epitaxy approach to engineer c-axis trilayers on a layer-by-layer basis (20). Most recent
studies of trilayer junctions have focused on a-axis-oriented or (103)-oriented trilayers due to the longer superconducting coherence length parallel to the ab planes, which in principle should lead to larger Ic Rn products.
While it is not clear how an a-axis trilayer epitaxial wiring scheme will deal with the inherently lower J c for
wiring runs along the c-axis direction, some promising results have been obtained with this approach. Sato
et al. (21) have produced (103)-oriented trilayer junctions with 350 PrBaCuO interlayers that exhibit RSJ
IV characteristics at 50 K with J c = 440 A/cm2 , 1 = 38%; Rn A = 1.2 10 7 -cm2 , 1 = 21% (1.3 for a
3 3 m2 junction), as well as sensible magnetic field modulation.
Planar or In-Line Junctions. These junctions, illustrated in Fig. 2(j), consist of a narrow gap (<1 m) in
a YBCO film, bridged by a normal conducting film either situated below the YBCO or deposited on top after
the gap is etched by, typically, a focused ion beam. The normal layer is usually Au or Ag (above the YBCO) or
PBCO (above or below). While the normal metal coherence length of Au is long enough that gaps of order 1 m
should support a critical current, the estimated values of N for PBCO and similar materials are so short that
even devices as short as 0.1 m should not exhibit a critical current. The fact that they do has been used as
evidence to suggest the existence of a long-range proximity effect in these materials, although the existence
of this is controversial. For example, it is possible that the critical current is partly due to YBCO not fully
removed from the gap.
There is little data on the reproducibility of these devices, but it is generally believed that they have little
potential for applications requiring many junctions. In addition, their characteristics are relatively nonideal,
often exhibiting large excess current for example.
Microbridges. If a superconducting bridge is made with a width comparable to the coherence length, ,
it behaves as a Josephson junction. Since is only a few angstroms in YBCO, this is not a practical way to make
a Josephson device. However, if a bridge is wider than , but still less than the penetration depth , it can behave
in many ways as a Josephson junction, for example, exhibiting constant voltage steps in response to applied
microwaves, in this case as a result of magnetic field vortices moving across the bridge in synchronization with
the microwave field. As such they may be useful for SQUID applications, but not for SFQ circuits.
These devices have been fabricated both in the ab direction [Fig. 2(k)], using focused ion beam etching
for example, and in the vertical direction in a c-axis film [Fig. 2(l)]. The latter case is interesting in that
the corresponding penetration depth is long enough (1 m at 77 K) that conventional photolithography
suffices. The c-axis microbridges have demonstrated encouraging Ic uniformity [13% for a few devices (22)],
although critical currents are typically somewhat high, and resistances low, compared to typical application
requirements, suggesting that more aggressive lithography may still be required. Low-inductance multilayer
SQUIDs have been demonstrated, and the fabrication process is quite compatible with the needs of a multilayer
process.
14
Fig. 8. Schematic cross section of a multilayer process incorporating edge SNS junctions over an HTS groundplane. The
horizontal shading of the YBCO layers indicates the direction of the copperoxide planes. All layers are expitaxial except
for the Au, used for contacts.
Circuit Integration
Multilayer Circuit Requirements. Multilayer structures are required for many important applications
of HTS junctions including digital circuits, SQUIDs with integrated pickup coils, voltage standards, and phase
shifters, among others. Junction-specific requirements on Ic and Rn for some of these applications have already
been discussed in the section entitled Overview. It is desirable that these junction properties be independent
of location in a multilevel structurefor example, on or off of groundplanes. This is a nontrivial constraint
because junction performance is often intimately related to details of film microstructure which can be affected
by growth over the underlying layers in a multifilm stack. Multilayer circuits also require good electrical
isolation (>104 -cm) between superconducting layers, using insulators with low dielectric constants and
low losses, if high-frequency applications are the objective. High-critical-current-density vias and crossovers
(>105 A/cm2 ) are essential for most HTS circuit applications. Digital circuits also call for the integration
of superconducting groundplanes to produce the low-inductance SQUIDs and interconnects needed for SFQ
logic. More complex circuits can require additional epitaxial insulators and wiring levels as well as integrated
resistors.
Fabrication Issues. A cross-sectional schematic view of the multilayer structure used to integrate edge
SNS junctions on an HTS groundplane is shown in Fig. 8. A minimum of four mask levels and six epitaxial oxide
film layers are needed for this process. Additional epitaxial film layers are often used such as buffer layers
between the substrate and groundplane, or passivation layers above the groundplane, but these additional
layers do not alter the basic processing steps.
The substrate selected for the most complex multilayer structures is single-crystal, perovskite-structure
NdGaO3 which is representative of the other candidate substrates. The (110) and (001) faces of NdGaO3 are
virtually identical in providing a square two-dimensional lattice with approximately a 1% lattice mismatch for
growth of c-axis-oriented epitaxial YBCO films.
The most important requirements for YBCO groundplane films are magnetic penetration depths, (T),
close to intrinsic values and smooth surfaces. These properties can be achieved, in principle, by films grown
by a number of different techniques. In practice, pulsed laser deposition, sputtering, and co-evaporation are
the three techniques most commonly used for multilayer film growth. The effect of the penetration depth on
circuit inductance is discussed in the next section. The smoothest films, those with approximately 1 nm rootmean-square (rms) roughness, have less than 103 /cm2 of the outgrowths that are commonly found in YBCO
films. Most of the outgrowths are second-phase copper oxide particles that are 0.5 mm to 1.0 mm in diameter
and grow higher than the film surface by a distance comparable to the film thickness.
As long as a copper oxide outgrowth does not form in a location where a junction is patterned, it does
not have a deleterious effect on YBCO film properties. For applications of YBCO films requiring a single layer,
15
such as bandpass radio frequency (RF) filters, no effort is made to minimize them. In fact, their presence is
a sign that excess copper available at the growing film surface has been consumed, leaving behind a matrix
of stoichiometric YBCO which has optimized properties. However, outgrowth density must be minimized by a
slightly copper-deficient film composition for multilayer circuits to maintain electrical isolation between layers
by the epitaxial insulators.
The epitaxial insulator that grows with the best edge coverage and smoothest surfaces is SrTiO3 . Its high
dielectric constant, on the order of = 500, is not a problem for low-frequency applications such as SQUID
magnetometers. However, there is no consensus materials choice at present for high-speed digital circuits
where a dielectric with = 10 to 30 is a practical maximum. Examples of relatively low- materials used as
epitaxial insulators are Sr2 AlNbO6 and Sr2 AlTaO6 .
Since the integration level is very low at this stage of the development of HTS circuits, demands on
lithography are minimal, and this is usually performed with contact lithography and standard resists. However,
to avoid formation of step-edge grain-boundary junctions where interconnects cross steps in underlying films,
an etch process is required that results in sidewalls that are sloped just 5 to 30 from the plane of the
substratesimilar to the angles used for edge SNS junction fabrication. Tapered edges are obtained by ion
milling with Ar or Ar/O2 mixtures with the ion beam at an angle with respect to the substrate and the substrate
rotating about its normal.
Because the surface of YBCO reacts with air and photoresist to form an amorphous layer of carbonates
and hydroxides 2 nm to 10 nm thick, the surface must be cleaned after lithography and before a subsequent
epitaxial film layer can be grown. In the case of edge SNS junctions shown in Fig. 8, the most important
exposed surface is the edge cut in the base electrode which will serve as the template for growth of an N-layer
and YBCO counterelectrode. Some combination of oxygen plasma ashing, blanket removal of the reacted layer
by ion milling, and wet chemical etching, typically with brominealcohol mixtures, is performed to prepare
surfaces for the next film deposition.
Figure 8 shows that a Au contact layer is typically used to complete this multilayer structure. The contact
layer is normally deposited in situthat is, after the YBCO counterelectrode layer has been deposited and
cooled to room temperature in oxygen, but before it has been exposed to air. In situ Au deposition lowers contact
resistance and improves adhesion. Other room-temperature film layers may be required for resistors and for
electrical isolation of the resistor layers, but they do not require the same care that must be taken for film
layers grown at high temperature.
Groundplanes. As discussed in the section entitled Application Requirements, digital circuits require
low inductances, of order a few picohenries, to ensure that the LIc product is about a flux quantum. The
standard way to keep inductances low is to incorporate a superconducting ground plane above or below the
active devices, which tends to confine the magnetic field in the relatively small volume between the two (or
more) superconductor layers. Calculation of inductance is also simplified for this geometry since the problem
is essentially reduced to counting squares and using the expression for the inductance per square of such a
microstrip line:
where d is the insulator thickness, b1 and b2 the superconductor thicknesses, and is a factor that determines
the field strength at the center of the finite width microstrip. The long penetration depth of YBCO, compared to
Nb for example, means that the inductance per square is much higher. For example at 65 K we have 1 = 2 =
0.23 nm, so for d = b1 = b2 = 0.2 nm and 1, we have Lsq 0.84 pH. Contrast this with the case for Nb at 4.2
K, where 0.04 nm and Lsq 0.35 pH. Thus the circuit-driven constraints of low inductance are particularly
16
Fig. 9. An example of a simple multilayer HTS circuit, a 1-bit analog-to-digital converter, fabricated with edge SNS
junctions over an HTS groundplane. The lightest areas are the gold-coated counterelectrode, while the darkest are the
base electrode. The process used to produce the circuit is extendible to much more complex circuits, provided that junction
critical current uniformity can be improved sufficiently.
difficult to meet for HTS materials. Given the need for high Ic for thermal stability, the requirement that LIc
be about a flux quantum is particularly difficult to achieve, suggesting the need for novel gate layouts in HTS.
The integration of a YBCO groundplane has been demonstrated for several HTS junction types, including
step-edge SNS, step-edge grain boundary, and edge SNS junctions. Measured inductances are consistent with
other measurements of the penetration depth.
Example Circuits. The key factor in determining the yield of working HTS circuits is the degree of
control over junction critical currentsespecially the on-chip spread. The more complex the circuit, the tighter
the Ic spread must be, although the numerical relationship between Ic spread and circuit yield depends on the
circuit margins. While LTS SFQ circuits often exhibit wide margins, of order 30% or more, extrapolating such
values to HTS operation at 40 K to 65 K is controversial. If these margins do hold out, possibly in cases where
bit-error rate is not a driving concern, then spreads of 15% should allow, based purely on statistical arguments,
yields of 50% for circuits with up to about twenty junctions (24). In fact a number of such circuits have been
demonstrated at up to about 65 K, at least at low speed, bearing out some optimism.
Demonstrated HTS gates or circuits include digital devices such as logic gates (OR, AND, etc.), set-reset
flip-flops, toggle flip-flops, sampling circuits, and shift registers, as well as analog devices such as SQUID
amplifiers. For digital devices, in which all of the junctions need to work for correct operation, the highest
junction count is about 30. Figure 9 shows an example of a 10-junction circuitthe first stage of a low-power
analog-to-digital converter, fabricated with edge SNS junctions over a ground plane.
So far there has been very little work to quantify the experimental margins of such HTS circuits. Should
the more pessimistic estimates of HTS SFQ margins be borne out, then it may be necessary to rely on voltagestate logic, where increased power dissipation will erase one of the major advantages of superconductivity, or
on multi-flux-quantum schemes, several of which have been proposed.
Manufacturability of HTS circuits will also require that chip-to-chip parameter spreads be well controlled.
For a single circuit it may be feasible to tune the operating temperature to, for example, adjust the average
17
critical current to the desired value, but this is clearly not practical for a system consisting of several separately
manufactured superconducting circuits. Adjustment of overall circuit biases is a more practical solution to
poorly targeted critical current values but will significantly increase costs due to the need for increased circuit
testing and qualification. Thus ultimately it will be necessary to control chip-to-chip parameter reproducibility
to the level of a few percent at most, which will be a significant challenge for these complex materials.
There is no single answer to the question of what is the best application for HTS Josephson junctions.
However, a consensus of those working in the field believes that applications that utilize the analog precision
of Josephson devices constitute the most promising niche where Josephson devices can surpass semiconductor
circuits. Thus there is an emphasis on high precision analog-to-digital and digital-to-analog converters in such
applications as radar, communications, and precision instrumentation. For example, an A-to-D converter with
20 bits of accuracy on a 10 MHz signal bandwidth would leapfrog semiconductor A-to-D converters, based on
their historical rate of progress, by about 10 years. A potential application of such a converter is in radar,
where it is desired to pick out a small target (airplane, missile) from a large clutter signal (rain, mountainous
terrain, waves). Such a circuit should contain, depending on the details of the architecture, anywhere from
several hundred to several thousand Josephson junctions. Assuming circuit margins of 30% (which may be
optimistic), the required junction critical current spread for 50% yield, based on the calculations of Ref. 23,
would be approximately 10% to 8%. Should the margins be reduced to, for example, 15% due to thermal noise
issues, then the required spread would be more like 5% to 4%.
The issue of operating temperature is also of crucial importance, and is influenced by the potential circuit
performance, as limited by thermal noise, and the availability, reliability, size, weight, and cooling power of
the cryocooler. For example, a typical Stirling-cycle cooler with 0.3 Watts of heat-lift at 4.2 K would have an
input requirement of some 1500 W, and would weigh about 250 pounds. This power and weight probably rules
out the application of LTS circuits in most airborne platforms. On the other hand a Stirling cooler with 4 W
of heat lift at 77 K would only require about 100 W of input power, and should weigh only about 10 pounds,
making airborne deployment much easier. Operating at 40 K, should thermal noise require it, might reduce
the available heat lift to 0.4 W, which should still be sufficient circuits of several thousand junctions.
Advanced Issues
Proximity Effect. When a superconductor and normal metal are brought into contact, Cooper pairs
from the superconductor can diffuse into the normal metal. Due to phonon-induced pair breaking, the pair
amplitude (also known as the superconducting order parameter or wavefunction) in the normal metal decays
exponentially over a decay length defined as the normal metal coherence length, n . In the clean limit where
ln , the mean free path in N, is much greater than n , the coherence length is given by
while in the dirty limit with ln n , the coherence length is
where the diffusion constant, Dn equals vn ln /d and d is the dimensionality. Theories describing the details
of the superconductornormal-metal proximity-effect interaction have been developed for a variety of cases
including back-to-back SN contactsthat is, the SNS weak link (24,25). In the SNS Josephson junction,
18
Fig. 10. Critical current versus temperature for a YBCO/CoYBCO/YBCO edge SNS junction. The solid line is a fit to
the proximity effect theory of DeGennes. Despite the fact that the junction is nonideal, in that it exhibits a large interface
resistance, it still appears to exhibit behavior consistent with the proximity effect.
pairs from each superconducting electrode leak into the normal metal interlayer and the overlap of the
exponentially decaying pair amplitudes determines the strength of interaction between the superconductors.
Consequently, the magnitude of the Josephson critical current scales as exp[L/n (T)], where L is the normal
metal bridge length. More specifically, in the dirty limit for long SNS bridges (L n ) relatively close to T c
(T > 0.3T c ), it is found that
where i is the superconducting gap at the superconductornormal-metal interface. This equation indicates
that the critical current of an SNS weak link should also vary exponentially with temperature, because of
the (T) 1/2 temperature dependence of the dirty-limit normal-metal coherence length. The exponential length
and temperature dependence of the critical current are the distinguishing signatures of true proximity effect
devices. Indeed, there are a number of examples of HTS SNS devices which are largely consistent with proximity
effect theory, most notably the junctions using Co- or Ca-doped YBCO as the normal metal layer. An example
of exponential critical current dependence on temperature for a Co-doped YBCO SNS edge junction is shown
in Fig. 10, along with a proximity theory fit to the data (26). However, it is often found that HTS devices
with a nominal SNS configuration do not show an exponential critical current dependence on temperature. In
fact, such devices commonly exhibit a quasilinear temperature dependence, which may indicate that pinhole
conduction through the normal metal is dominating the electrical characteristics (25).
Control of Resistance in SNS Devices. The normal state resistance of an SNS weak link is given
by the sum of the normal metal resistance plus the resistance of each of the two SN interfaces. In the ideal
SNS device the interface resistances are zero and the total device resistance is just Rn = n L/A, where n is the
normal metal resistivity, L is the normal metal thickness, and A is the cross-sectional area. For typical values
of these parameters in an SNS edge junction with a YBa2 Cu2.8 Co0.2 O7 normal metal layer at 65 K (n = 250
-cm, L = 100 , and A = 4 0.2 m2 ) we find Rn = 0.03 . In practice, such low values of resistance are
often undesirable. For example, for SFQ digital applications at 65 K junctions are biased at a fixed current
of order 500 A so that the available Ic Rn product with Rn = 0.03 is only 15 V, far less than the required
value of approximately 300 V. Increasing SNS device resistances to a practical level requires adding interface
resistance without degrading the inherent Ic Rn product. In principle, this can be done in at least two ways: (1)
19
by incorporating an inhomogeneous interface resistance to reduce the effective device area or (2) by producing
a thin insulator at one SN interface to form an SINS structure. In practice, different groups have seen widely
varying values of SNS resistance, ranging from the ideal but impractical case of very low Rn A (16) to the more
technologically interesting case of high-Rn A devices (27).
For SNS weak links using Co-YBCO as the normal metal, it has been found that the interface resistance is
sensitive to a variety of factors including the base electrode material and the normal metal and counterelectrode
deposition conditions (17). For example, SNS devices using YBa2 Cu3 O7 base electrodes grown by pulsed laser
deposition (PLD) exhibit more than an order of magnitude lower resistance than devices with La-doped YBCO
base electrodes (YBa1.95 La0.05 Cu3 O7 ), or GdBa2 Cu3 O7 or NdBa2 Cu3 O7 base electrodes. Varying the normal
metal and counterelectrode growth parameters can also have a dramatic effect on device resistance: Highpressure PLD growth in an ArO2 atmosphere results in Rn A products over a factor of 10 smaller than for
devices produced using the more conventional PLD deposition conditions in a pure oxygen background. While
the detailed nature of the interface resistance is not understood at this point, the base electrode material
dependence suggests that cation disorder (e.g., Y and Ba exchange) is affecting device resistance. The fact that
the growth conditions of the normal metal also have a strong effect on SNS resistance indicates that defects
frozen in in the early stages of normal metal growth may also play an important role in determining interface
resistances. Because SNS interface resistances are strongly affected by a number of material and fabrication
parameters, it is possible to control SNS device resistances over two to three orders of magnitude, with Rn A
products ranging from 0.03 -m2 to more than 10 -m2 . Importantly, even in the relatively high Rn A limit
required for SFQ applications (0.5 -m2 to 2 -m2 ), Co-doped YBCO SNS devices incorporating significant
interface resistance still behave like true proximity effect devices (see, for example, Fig. 10) with parameter
uniformity suitable for small-scale SFQ circuits.
Limits on Reproducibility. Speculation on the origins of parameter spreads have led to many experiments in fabrication of edge SNS junctions. Surprisingly, the fabrication parameters which result in junction
resistances greater than Rn = n L/A do not appear to systematically contribute to larger spreads in critical
currents or other junction parameters. Similarly, the uniformity of current flow through a junction which can
be inferred from Ic (B) indicates that junctions with significant interface resistance maintain uniform current
distributions.
Poor control over many fabrication parameters will certainly result in junction spreads worse than the
state of the art. A good example is the roughness of YBCO base electrodes which gets transferred into the edge
by patterning with Ar ion milling. While improvements from 10 nm to 2 nm rms surface roughness provide
a measurable benefit for junction reproducibility, further improvements in smoothness have had a negligible
effect. A second example is that junctions facing in all four in-plane directions sometimes exhibit a distribution
of critical currents that is direction-dependent. However, when all processing steps are made isotropic, stateof-the-art junction uniformity can be achieved as easily in a set of junctions facing in four directions as in a set
facing just one way.
These results have led us to examine defects that are intrinsic to YBCO. The role of oxygen disorder in
YBCO has been investigated in several types of edge junction experiments. Decreasing the number of oxygen
vacancies by plasma oxidation or annealing in ozone simply scales Ic for all treated junctions by a constant
factor as large as five. Experiments in which orthorhombic YBCO electrodes were replaced by doped YBCO
compounds which were tetragonal have been inconclusive in determining the possible role that twinning in
YBCO might have on parameter spreads. Junctions were fabricated to face in 110 in-plane directions instead
of the standard 100 directions to minimize the effects attributable to twinning, but no improvement in junction
uniformity was observed.
Finally, the fact that similar best-case critical current spreads are observed for different junction fabrication processes using the same base electrode materials suggests that microstructural defects in the base
electrode or base electrode edge are limiting Ic spreads. Further improvements in materials quality and edge
formation techniques are expected to lead to improved junction spreads.
20
Conclusions
Josephson junctions based on YBCO are the fundamental building blocks for a variety of superconducting
electronics applications operating at temperatures >50 K. The properties of individual junctions fabricated
in a variety of configurations are sufficiently close to ideal Josephson behavior to meet application requirements. However, integration of junctions into multilayer circuits and demands on reproducibility of junction
parameters when higher junction counts are needed have narrowed development efforts to a few promising
configurations. Most of the current HTS circuit fabrication effort in industrial laboratories is based on edge
SNS junctions which have been used for the most sophisticated and extendible digital circuit demonstrations. Further incremental improvements in the uniformity of these junctions to 1 Ic spreads less than 10%
will permit medium-scale integrated circuit fabrication. A parallel effort, mainly by university researchers,
is exploring higher-risk alternative junction configurations intended to circumvent some of the limitations to
junction uniformity that may exist for edge junctions.
BIBLIOGRAPHY
1. T. Van Duzer, C. W. Turner, Principles of Superconductive Devices and Circuits, New York: Elsevier, 1981.
2. D. Robbes et al., The ac Josephson effect in constrictions engraved in bulk YBa2 Cu3 O7 and dc SQUID operation at 77
K, Nature, 331: 151153, 1988.
3. B. H. Moeckly, D. K. Lathrop, R. A. Buhrman, Electromigration study of oxygen disorder and grain boundary effects in
YBCO thin films, Phys. Rev. B, 47: 400, 1993.
4. D. Dimos et al., Orientation dependence of grain-boundary critical currents in YBa2 Cu3 O7 bicrystals, Phys. Rev. Lett.,
61: 219222, 1988.
5. K. Char et al., Bi-epitaxial grain boundary junctions in YBa2 Cu3 O7 , Appl. Phys. Lett., 59: 733735, 1991.
6. K. Herrmann et al., Characterization of YBa2 Cu3 O7 step-edge junctions, Supercond. Sci. Technol., 4: 583586, 1991.
7. J. Luine et al., Characteristics of high performance YBCO step-edge junctions, Appl. Phys. Lett., 61: 11281130, 1992.
8. M. J. Zani et al., Focused ion beam high T c superconductor dc SQUIDs, Appl. Phys. Lett., 59: 234236, 1991.
9. R. B. Laibowitz et al., All high T c edge junctions and SQUIDs, Appl. Phys. Lett., 56: 686688, 1990.
10. B. D. Hunt et al., High Temperature Superconductor Josephson Weak Links, in S. I. Raider et al. (eds.), Low Temperature
Electronics and High Temperature Superconductivity, Electrochemical Soc. Proc., vol. 93-22, Pennington, NJ: The
Electrochemical Society, 1993, pp. 462472.
11. B. H. Moeckly, K. Char, Interface Engineered High T c Josephson Junctions, in H. Koch and S. Knappe (eds.), ISEC 97
6th Int. Supercond. Electron. Conf. Extended Abstracts, vol. 1, Braunschweig, Germany: PTB, 1997, pp. 810.
12. S. K. Tolpygo et al., High quality YBa2 Cu3 O7 Josephson junctions made by direct electron beam writing, Appl. Phys.
Lett., 63: 16961698, 1993; A. J. Pauza et al., Electron beam damaged high-T c junctions stability, reproducibility and
scaling laws, IEEE Trans. Appl. Supercond., 5: 34103413, 1995.
13. R. W. Simon et al., Progress Towards a YBCO Circuit Process, in R. McConnell and S. A. Wolf (eds.), Science and
Technology of Thin Film Superconductors II, New York: Plenum, 1990, pp. 549560.
14. R. H. Ono et al., High-T c superconductor-normal metal-superconductor Josephson microbridges with high-resistance
normal metal links, Appl. Phys. Lett., 59: 11261128, 1991.
15. M. S. DiIorio et al., Manufacturable low-noise SQUIDs operating in liquid nitrogen, Nature, 354: 513515, 1991.
16. K. Char, L. Antognazza, T. H. Geballe, Properties of YBa2 Cu3 O7x /YBa2 Cu2.79 Co0.21 O7x /YBa2 Cu3O7x edge junctions,
Appl. Phys. Lett., 65: 904906, 1994.
17. B. D. Hunt et al., High-resistance HTS SNS edge junctions, Appl. Supercond., 1998, in press.
18. M. A. J. Verhoeven et al., Ramp-type junction parameter control by Ga doping of PrBa2 Cu3 O7 barriers, Appl. Phys.
Lett., 69: 848850, 1996.
19. C. T. Rogers et al., Fabrication of heteroepitaxial YBa2 Cu3 O7x /PrBa2 Cu3 O7x /YBa2 Cu3O7x Josephson devices grown
by laser deposition, Appl. Phys. Lett., 55: 20322034, 1989.
21
20. J. N. Eckstein, I. Bozovic, G. F. Virshup, Atomic layer-by-layer engineering of high T c materials and heterostructure
devices, MRS Bull., 19 (9): 4450, 1994.
21. H. Sato, S. Gjoen, H. Akoh, Improvement of junction properties of YBaCuO/PrBaCuO/YBaCuO trilayer Josephson
junctions, IEEE Trans. Appl. Supercond., 7: 25102513, 1997.
22. S. W. Goodyear et al., Vertical c-axis microbridge junctions in YBa2 Cu3 O7 /PrBa2 Cu3 O7 thin films, IEEE Trans. Appl.
Supercond., 5: 31433146, 1995.
23. D. L. Miller, J. X. Przybysz, J.-H. Kang, Margins and yields of SFQ circuits in HTS materials, IEEE Trans. Appl.
Supercond., 3: 27282731, 1993.
24. P. G. DeGennes, Rev. Mod. Phys., 36: 225, 1964.
25. K. A. Delin, A. W. Kleinsasser, Stationary properties of high critical temperature proximity effect Josephson junctions,
Supercond. Sci. Technol., 9: 227269, 1996.
26. M. G. Forrester, unpublished, 1997.
27. B. D. Hunt et al., High-T c SNS edge junctions and SQUIDs with integrated groundplanes, Appl. Phys. Lett., 68:
38053807, 1996.
BRIAN D. HUNT
MARTIN G. FORRESTER
JOHN TALVACCHIO
Northrop Grumman Science and Technology Center

Copyright
HYSTERESIS AND COUPLING LOSSES IN SUPERCONDUCTORS

A key issue for most applications of superconductivity involves ac losses. Designers need to understand the
mechanisms of ac losses in order to lay out the conductors and windings correctly and to predict the performance
range in operation. In devices operating at the grid frequency (transformers, current limiters, generators,
motors, and power transmission cables) ac losses mostly affect the cryogenic load and hence the overall efficiency
of the application. In large pulsed windings [fusion magnets, (SMEs)] the coupling current loss affects the
stability, that is, the ability of the superconductor to withstand magnetic field transients. The dc magnetization
of a superconductor, which leads to hysteresis loss, is a crucial issue in applications such as imaging and
accelerator magnets, where the residual, low field magnetization of the superconductor affects the quality
(linearity and homogeneity) of the generated magnetic field.
In normal conducting materials, both dc and ac losses are due to the finite electrical conductivity. At first
glance, it appears contradictory that ac losses can occur in a superconducting material, with zero electrical
resistivity.
The nonreversible magnetization behavior of bulk superconductors was known well before a practical
superconducting wire was manufactured; afterward, it was called ac loss, which sounds much more negative
than magnetization loop. For single-core wires and tapes, the only ac loss was the hysteresis loss, with the
associated disruptive flux jumps. For multifilamentary composites, a new source of loss (filamentary coupling
loss) was identified and stimulated the development of very sophisticated strand layouts, with high resistivity
barriers and mixed matrices; on the other hand, flux jumps were no longer an issue. Later on, with large
stranded conductors, the biggest concern was cable loss (interstrand coupling loss): the effort to limit ac losses
was focused on the cable layout rather than the internal structure of the multifilamentary composite.
Because of ac losses, the competitiveness of superconductors has been limited in many fields of applied
electrical engineering. Most of the superconducting coils in use today (commercial and research) are operated
in dc mode, with limited ac loss occurring during the slow charging up. In those coils, the fastest field change
is seen during a safety discharge, when concern about heating or even quenching the conductor is small.
Both hysteresis and coupling current losses occur in a time-varying magnetic field, but their relative
weights depend on the specific application. In power transmission cables, the hysteresis loss is by far the
largest source, as well as in small, slow rate, potted windings. In large cabled conductors for big SMES and
fusion poloidal field coils, the coupling current loss dominates. The hysteresis loss per unit volume is a function
of the filament size. The coupling current loss per unit volume increases with the conductor size: with few
exceptions, the larger the conductor, the bigger the weight of the coupling current loss.
The physical mechanism of ac losses in superconductors is no longer the object of baseline research.
Most R&D activities are devoted to developing low loss, stable conductors. For design purposes, the ability
to accurately calculate (i.e. predict) the ac loss during operation is crucial for a reliable and cost-effective
engineering approach.
Hysteresis Loss
In type I superconductorsfor example, pure metals with defect-free latticesthe magnetic field does not
penetrate the bulk of the material and the superconducting shielding currents flow only at the surface. Such
materials have reversible magnetization and no hysteresis loss, with B = 0 and M = 0 H (Meissner effect)
inside the material.
In type II superconducting materials used for practical applications, both low and high T c , the surface
shielding currents have a marginal role for the magnetization and the flux penetrates the bulk of the conductor.
The nonreversible magnetization of type II superconductors is the reason for the hysteresis loss (1 2 3). The
energy loss per unit volume of superconducting material, Q, can be written in a general form for a closed cycle
of applied magnetic field B (e.g., an oscillation at grid frequency, a chargedischarge cycle of a magnet, or the
superposition of an ac field on the background field)
where M is the average value of the magnetization inside the superconductor. The explicit expression for the
local magnetization, needed to evaluate the integral, depends on the superconductor geometry and on the
model selected to describe the magnetic flux penetration and the flux profiles inside the superconductor.
The formulas quoted below are for cylindrical superconducting filaments of diameter D: whenever the
real cross section of the filaments is not round (e.g., oval, dendritic, or hollow filaments or clusters of bridged
filaments), the parameter D in the formulas should be considered as an equivalent diameter. For tapes and
flattened filaments with high aspect ratio (e.g., for some high T c conductors), the formulas for an infinite slab
can be used as a convenient approximation.
The Magnetization Curve. A dc magnetization curve for a type II superconductor is shown in Fig. 1.
After cooldown, at zero field, M = 0. As an external magnetic field is initially applied, the shielding currents at
the filament surface prevent the flux penetration into the bulk superconductor. The diamagnetism is perfect
(i.e., M = 0 H) as long as the applied field does not exceed the first critical field, Bc1 . In a type II superconductor,
total flux exclusion (the Meissner effect) occurs only at the beginning of the first, virgin magnetization: if the
superconductor is cooled down in the presence of a magnetic field, no flux exclusion occurs.
Above Bc1 , the average magnetization increases until the flux penetrates to the center of the filament, at
B = Bp1 . As the field increases, the diamagnetism decreases (upper branch of the curve) and the magnetization
eventually becomes 0 at the upper critical field, B = Bc2 (not shown in Fig. 1). When the field is decreased, the
flux profiles reverse their gradient in the filament and the average magnetization is >0 (lower branch of the
curve). At B = 0, the flux trapped in the filaments is called residual magnetization.
The magnetic flux enters the filament as discrete flux quanta. The diffusion of the flux quanta in the
bulk type II superconductor is restrained by the pinning centers, which establish field gradients (flux profiles)
inside the filaments. The pinning centers are microscopic normal zones (e.g., metallurgical or lattice defects)
that provide a potential hole to trap the flux quantum. According to the critical state model, a supercurrent
encircles each pinning center: the strength of the pinning centers (i.e., their ability to hold the flux quanta)
is a function of the fraction of the external field to the critical field, b = B/Bc2 (T, ) where is the mechanical
strain. Strong pinning centers are able to store large density of magnetic energy. The magnetization current
density J c is directly proportional to the volumetric density of the pinning forces in the superconductor.
The penetration field Bp is the field difference between the filament surface and the electrical center line
(see Fig. 2): in Beans model (4), J c is assumed constant inside the filaments for a given b, and Bp is a linear
function of the critical current density. For a round superconducting filament of diameter D, the penetration
field Bp (b), in perpendicular and parallel orientation, is proportional to the density of the magnetization
Fig. 1. Typical magnetization loop of a type II superconductor and related terminology.
Fig. 2. Penetration field in a superconducting cylinder of diameter D in parallel applied field (or infinite slab of thickness
D), without (left) and with (right) transport current. The dashed profile models the lower branch of the magnetization
curve.
currents flowing respectively in the axial (J c ) and the azimuthal (J c ) direction (5, 6)
For an infinite slab of thickness 2a, with applied field parallel to the slab surface, the penetration field is
A partial magnetization loop occurs when a small external field variation is superimposed on a background
dc field. If the amplitude of the field change, B, is larger than 2Bp , full penetration is achieved, that is, the
magnetization moves from the upper to the lower branch of the curve.
It is questionable how far Beans model for linear flux profiles is an acceptable approximation. When the
penetration field is much smaller than the applied field, Bp B, the J c variation over Bp can reasonably be
neglected. As Bp is proportional to the filament size, the range of field over which Beans model reliably applies
is larger for thin-filament superconductors. Whenever Beans model is not considered adequate, an explicit
formula for J c (b), such as the expression proposed by 7
must be substituted in Eqs. (2) and (3) (B0 and are fitting parameters). The magnetization and loss formulas
become more complex using Eq. (4, but the improvement in the accuracy of the loss results is not dramatic. At
very low field, when the linear profile approximation is rough, the parameters in Eq. (4) cannot be satisfactorily
fitted by experimental results, as a direct measurement of Ic close to 0 field is impossible due to the self-field
and instabilities. The formulas below apply only to the linear profile assumption. An example of hysteresis loss
formulas using a nonlinear profile approach can be found in 8.
Integrating in cylindrical coordinates the linear flux profiles from Eqs. (2) and (3) over the filament
volume, the upper and lower branches of the magnetization curve are obtained as explicit functions of the
critical current density and filament diameter. In perpendicular and parallel applied field orientations the
average magnetization per unit volume, according to 6, is
Basic Formulas for Hysteresis Loss. Substituting Eq. (5) into Eq. (1), the hysteresis loss for a closed
field cycle of amplitude B = Ba Bb is obtained as a function of the average critical current density or the
average penetration field, defined by
and
The results of the integration are summarized in Table 1 for the three cases of an infinite slab of thickness 2a
with field parallel to the slab surface and a cylinder with diameter D perpendicular and parallel to the applied
field. The formulas are different for partial penetration (B 2Bp ) and full penetration (B 2Bp ). A further,
easier formula is proposed for B 2Bp : this formula overestimates the loss. The shaded areas in (Fig. 1) give
a measure of the excess, which is accounted for by using the formula reported in Table 1 for B 2Bp .
For a given B, the loss maximum occurs when B = 2Bp . The loss maximum, Qmax , reported in Table 1,
is a fraction of the magnetic field energy density; it does not depend on the critical current, critical temperature,
strain, or filament diameter. In some cases, it may be useful to use Qmax to get a feeling for the worst-case loss
without performing time-consuming calculations. For a given B, the filament parameters J c and D determine
the reduced field b for which B = 2Bp (b) is fulfilled, that is, the loss is maximum (9). The loss formulas
in Table 1 may also be written as a fraction of Qmax or of the magnetic field energy density B2 /20 , thus
introducing a dimensionless loss factor, which is, for a given geometry, only a function of B/Bp (b) (10). In
Table 1, Q is the hysteresis loss per unit filament volume and J c is the filamentary critical current density.
For some superconducting strands, such as Nb3 Sn and high T c superconductors, the noncopper critical current
is referred to instead of the filamentary critical current density. Whenever the exact filament fraction is not
known, it is possible to use the loss formulas for B 2Bp , replacing J c by the noncopper critical current Ic .
The hysteresis loss is then expressed in joules per meter of conductor length.
Anisotropy and Variable Angle Orientation. Due to the integration path of the flux profiles in the
axial and radial direction of the cylindrical filament, the ratio of the magnetization in parallel and perpendicular
field orientation is, according to Eq. (5),
The difference observed in the amplitude of magnetization measurements at 0 and 90 orientation is larger
than 4/ and reveals of the anisotropy of the critical current density, that is, J c = J c .
In NbTi filaments, the largest source of pinning centers is the precipitation of -Ti and cell dislocation
(11). During the manufacturing process (drawing and annealing), the pinning centers are created and strongly
oriented in the axial direction. The pinning forces are very anisotropic, resulting in a critical current density
much larger in the azimuthal direction than in the axial direction. The ratio J c /J c in NbTi conductors is
a function of the field and also depends on the manufacturing history (filament size, alloy composition, cold
work): from experimental magnetization measurements, the critical current anisotropy is J c /J c 3 (12).
In Nb3 Sn conductors, the major source of pinning centers is the grain boundaries that form during the
reaction heat treatment. The anisotropy of the critical current density is linked to the grain orientation, which
Fig. 3. Areas of the magnetization loops versus the orientation angle between filaments and applied field. Single core NbTi
strand, D = 127 m. From 6. Reprinted from Cryogenics, 18, A. P. Martinelli and B. Turck, Some effects of field orientation
on the magnetization of superconducting wires, pp. 155161, copyright 1978, with permission from Elsevier Science.
is influenced by the heat treatment schedule. The Sn diffuses radially into the Nb filaments, and the Nb3 grains
are elongated in the radial direction, giving a higher density of the boundary lines for the axial critical current
(13). Typical values of the anisotropy in filamentary Nb3 are J c /J c 0.5 (14).
The anisotropy of the critical current density should not be confused with the variation of the transport
critical current as a function of the orientation angle of the applied field, J c (B ). The azimuthal critical current
density, J c , to be used for the hysteresis loss in parallel field orientation, is not the same as the transport
critical current measured with parallel field orientation, J c = J c (B ), but J c J c (B ). For both NbTi and Nb3
Sn conductors, a larger transport current has been observed in the parallel applied field, J c (B ) > J c (B ) (6,
14,15,16).
The orientation of the superconducting filaments in cabled conductors with respect to the cable axis
changes continuously over a broad range of angles. For large, multistage conductors, average strand angles
of 16 to 25 are commonly observed: the range of the strand angles and its statistical distribution depend
on the number of cable stages and the pitch sequence. The hysteresis loss at intermediate angles cannot be
interpolated from the formulas in parallel (0 ) and perpendicular (90 ) fields. The behavior of the loss as a
function of the angle has been observed to be not monotonic, with a peak around 30 and a minimum at small
angle (<10 ); see Fig. 3 from 6. The interference of the magnetization currents flowing in longitudinal and
azimuthal directions distorts the flux profiles and does not allow a practical definition of the penetration field.
At angles close to 0 , the flux profiles adjust themselves for consecutive field cycles and the loop area decreases
until a reproducible magnetization is obtained after 10 to 20 cycles. An attempt to model the magnetization at
intermediate angles can be found in 17.
Filament Diameter. The filament diameter is a key parameter for the hysteresis loss formulas. It can
be either estimated directly from metallographic examination of the strand cross section or deduced from the
magnetization and critical current measurements.
For most of the commercial NbTi strands, the magnetization currents are confined to the individual
filaments: permanent currents linking groups of filaments by proximity effects are observed only in very thin,
highly packed filaments, with submicron interfilament spacing (11). The critical field for proximity effects,
Bc1p , is a function of the temperature, transport current, ratio of spacing to filament size, matrix resistivity
and impurities, twist pitch, and sample length (18,19,20).
For A15 superconductors, as well as for high-temperature superconductors, the estimation of the filament
diameter from metallographic investigation is not accurate. In the Nb3 Sn strands, because of different access
to the Sn source (for both bronze method and internal Sn strands), the filaments do not all grow to the same
size. A nonreacted Nb core may be left in some region of the filamentary zone, turning the Nb3 Sn cylinders
into hollow cylinders. A major problem affecting the assessment of the filament diameter in Nb3 composites is
bridging: when the Nb filaments are tightly packed in the matrix, the Nb3 Sn layers grow during the reaction
heat treatment to build either continuous superconducting links between filaments (21) or mechanical contacts
(22), which behave like the proximity effects. The density of bridging is a function of the spacing to filament
ratio s/d (or local area ratio between CuCuSn matrix and Nb filaments) and of the heat treatment schedule
(23). The superconducting properties of the bridges linking the filaments may be different from those of the bulk
filament: at higher field, temperature, or strain, some of the links may become too weak for the magnetization
currents. In these cases, the filament diameter is a function of b. The paths of the magnetization currents in a
cluster of randomly bridged filaments cannot be analytically modeled: whenever bridging occurs and the loss
formulas for cylinders are used, the equivalent filament diameter must be determined from the magnetization
curve, preferably at different b.
Three main methods are used to derive the filament diameter from the magnetization measurement in
a perpendicular field. All methods use Beans model and assume that the filament critical current density
is directly measured in the same field range. The same methods can also be applied to deduce the critical
current density once the filament diameter is known (e.g., to estimate J c or the low field J c , when a direct
measurement of Ic is not possible).
Diameter from the Penetration Field. The minimum field change to move from the upper to the lower branch
of the magnetization curve (see Figs. 1 and 2) is Bp = 2Bp . The filament diameter can be estimated using
Eq. (2) or (3) and Bp (B) from the magnetization curve. The advantage of this method is that no calibration
of the magnetization is necessary to estimate Bp .
Diameter from the Amplitude of the Magnetization. The filament diameter can be deduced using Eq. (5)
from the amplitude of the magnetization, measured as half of the distance from the lower to the upper
branch of the curve. The accuracy of this method is limited by the calibration of the magnetization.
Diameter from the Energy Loss of a Closed Field Cycle. The hysteresis energy loss for a closed field cycle
(magnetization loop) can be estimated either by the line integral of the magnetization curve, according
to Eq. 1, or by the calorimetric method, after subtracting the coupling loss contribution, if any. If the
calorimetric method is used, the magnetization does not need to be calibrated. According to the amplitude
of the applied field B (Bp > B or Bp < B), the measured energy is compared with the formulas in
Table 1 to deduce the filament diameter.
Crossing the Zero Field. The formulas for magnetization and hysteresis loss have limited validity
at low applied fields, especially at the zero-field crossing. On one side, the linear flux profile approximation
(Beans model) is very rough at fields smaller than the first penetration field, and below Bc1 the surface screening
currents prevent any flux change inside the filament. On the other hand, non-current-carrying superconducting
materials are sometimes included for manufacturing reasons in technical superconducting strands, resulting
in low field perturbations of the magnetization curve. In soldered cables, the low field superconductivity of the
solder may also play a similar role.
Fig. 4. Magnetization loop of a Nb3 Sn multifilamentary strand with Nb + Ta diffusion barrier. The low field peak of the
magnetization is due to the pure Nb shell with diameter 0.5 mm.
In thin filament NbTi strands, a Nb shell encircles each filament to prevent TiCu intermetallic formation
during the intermediate heat treatment process, and some Nb3 Sn suppliers use a Nb layer as diffusion barrier
or include it to buffer the Ta or V barrier on the side facing the stabilizer. In both cases, a continuous Nb shell
is left on the outer side of the barrier. When a continuous Nb3 Sn ring grows from the Nb diffusion barrier, its
magnetization is as much as one order of magnitude larger than that in the filament (24).
The pure Nb behaves like a soft type II superconductor, with Bc 0.18 T. The effect of the screening
currents in the Nb layer on the outer side of the diffusion barrier, whose diameter is more than 100 times
larger than that of the filament, can be clearly recognized as a low field peak in the magnetization curve; see
Fig. 4.
Whenever an anomaly of the magnetization curve occurs at the zero crossing, large errors are likely in
the hysteresis loss calculation. If the filament diameter is derived from microscopic examinations or from the
higher field magnetization, the loss at low field will be substantially underestimated by the loss formulas. In
contrast, if the energy loss of a bipolar field cycle is used to deduce the filament diameter, this, and hence the
higher field loss, will be overestimated, because of the additional contribution of the Nb below 0.18 T. The range
of the operating conditions should dictate the decision on the criterion to be used for the filament diameter. If
necessary, a correction factor can be added in the calculation code to include the Nb magnetization contribution
at the zero-crossing field.
Hysteresis Loss with DC Transport Current. When a longitudinal current is superimposed on the
transverse field magnetization currents of a filament, the electrical center line is moved to the periphery of the
filament (or slab), the flux profiles are asymmetric, and the penetration field decreases by a factor 1 i, where
i is the ratio of the longitudinal current (also called the transport current) to the critical current
Below penetration (i.e., for B < 2Bi p ), the low B amplitude formula in Table 1 can also be used in the
presence of a dc transport current. Above penetration, the magnetization decreases as a function of the transport
current, dropping to 0 for i = 1, that is, Itr = Ic . For large B, the area of the magnetization loop (i.e., the energy
supplied by the external field change) decreases when a transport current is superimposed on the magnetization
currents [see Fig. 5(b) from 25]
Fig. 5. Magnetization loss as a function of the dc transport current for a single core NbTi conductor: (a) B = 0.25
T < 2Bi = 0 p , (b) B = 0.50 T > 2Bi = 0 p . From 25. Reprinted from Cryogenics, 25, T. Ogasawara, Y. Takahashi, K. Kanbara,
Y. Kubota, K. Yasohama, and K. Yasukochi, Alternating field losses in superconducting wires carrying dc transport currents:
Part 1. Single core conductors, pp. 736740, copyright 1979, with permission from Elsevier Science.
As a function of the transport current, the magnetization loss increases until full penetration is achieved for
B = 2Bp (1 i). A further increase of the transport current decreases the magnetization loss; see Figs. 5(a)
and 6 from 25.
A change of the filament magnetization beyond 2Bi p is opposed by the power supply, which works to
maintain the transport current, that is, the asymmetric flux profiles. A voltage appears along the filament,
and an extra energy Qd due to the dynamic resistance Rd must be added to the magnetization loss (26, 27).
Above penetration, the dynamic resistance is proportional to the amplitude of the field change and inversely
proportional to the duration t0 of the field change
10
Fig. 6. Magnetization loops at increasing transport current for a single core NbTi conductor (D = 250 m), with B = 0.25
T, from 25. The magnetization loss is maximum at F, when 2Bi = 60 p = 0.25 T; see also Fig. 5(a). Reprinted from Single core
conductors, pp. 736740, copyright 1979, with permission from Elsevier Science.
For B 2Bi p , Rd , that is, the energy loss per cycle, Qd , is proportional to i2 but is independent of the
field rate. The total loss in filaments carrying a dc transport current is
From Eq. (12), at B 2Bi p the ratio of the total loss with transport current to the magnetization loss is
2 for any transport current. For Bp B/2 > Bi p , the loss enhancement factor can be much larger than a
factor of two (10). This can be understood by remembering that the low B magnetization loss is proportional
to B3 , but the dynamic resistance loss Qd is proportional to Bi2 .
The transport current also affects the azimuthal magnetization currents, modifying the local field orientation angle. Some kind of dynamic resistance is also expected because of the interference of Itr with J c .
Equations (9) and (12) give a satisfactory and validated (25) model for a single-core strand. However, two
assumptions are required to extend them to a multifilamentary strand or to a cable of stranded wires:
Each filament of each strand carries the same fraction of critical current.
The longitudinal current is constant during the external field change.
11
Both assumptions are highly unlikely: On one hand the current distribution is not homogeneous across
either the cable or the filaments of an individual strand. On the other hand, the several coupling current loops
induced by a field change cause local, time-dependent, very large variations of the current density compared
with the average value.
Self-Field Loss. A special case of hysteresis loss occurs for a straight, solid conductor carrying a
longitudinal current in the absence of any external field. The flux penetration is due only to the self-field,
which is proportional to the longitudinal current. Full penetration and maximum loss occur for I = Ic . For ac
operation, it is convenient to write the hysteresis loss per unit length per cycle as a function of i, the ratio of
the peak current to the critical current. 28 proposed a formula for self-field loss in an isolated thin slab and a
round (or elliptical) filament
For round or elliptical cross section
For a thin slab
For conductors with the same critical current, the loss ratio at saturation (i.e., at i = 1) is Qr /Qs = 1.3. At small
fractions of the critical current, Qr /Qs = 1/i, showing that the advantage of the thin slab geometry is significant
only at very small current density.
In a round multifilamentary composite, the filaments are not transposed for self-field, and the filamentary
zone of diameter Dfz can be treated as a single core, applying a filling factor for the critical current. 10
discusses the self-field loss for a round multifilamentary composite in terms of penetration field. The complete
penetration field is Bps = 0 J c Dfz /2, and the partial penetration field is Bms = iBps . The loss per cycle per unit
filament volume is
Equation (15) can be written in terms of critical current, as a loss per unit length, and becomes
which is identical to Eq. (13) except for the filling factor . In the case of an oscillating, unidirectional current,
i is defined as the ratio of transport to critical current, and the loss formula proposed by 10 as loss per unit
length, becomes,
The use of thin filaments does not help to reduce the self-field loss, as the nontransposed filamentary
zone behaves like a single core with critical current reduced by the filling factor . When the self-field loss
12
becomes a crucial issue, it is recommended to select a transposed cable or braid, where the filamentary zone of
the individual strands is kept as small as possible.
The preceding formulas have become very popular in the high temperature superconductor community,
although both Norris and Wilson warned about the limits of their applicability. The assumption of constant
critical current may result in significant errors at low field. The twist in the multifilamentary composites
introduces a spiral component of the self-field. For cabled conductors, the field from the neighboring strands
may give rise to coupling current loss, not taken into account in the preceding formulas.
Accuracy of Hysteresis Loss Estimation. The overall accuracy of the hysteresis loss estimation is
affected by the simplifications assumed in the model, the accuracy of the conductor parameters, the local field
orientation, the distribution of the transport current (29, 30), and the nonfilamentary magnetization at low
field. The weight of the individual error sources depends on the conductor layout and operating conditions.
The use of sophisticated computer codes does not help much to improve the accuracy of the hysteresis loss
prediction, which lies, in the best cases, around 20%.
Model Accuracy. The geometrical basis for hysteresis formulas is either a cylinder or an infinite slab.
The actual filament geometry, especially in the case of bridging, is not a cylinder: even when an equivalent
diameter is defined, it does not perfectly model the real filament over the entire range of operating conditions.
Beans assumption of linear flux profile may be a source of inaccurate estimation of the magnetization
and penetration field at low magnetic field, especially for thick filaments, strands, and tapes. In addition, the
model does not account for surface screening currents below Bc1 .
The loss formulas may include the effect of the strain and temperature on J c and Bp , but the integration
of the magnetization is done under isothermal conditions. A step-by-step integration, calculating the magnetization from the instantaneous value of J c (T,), also would not be correct: an increase of T or || at constant
field decreases the magnetization, as J c decreases and more flux penetrates. However, a decrease in T or || at
constant field leaves the flux profiles, and hence the magnetization, unchanged.
Coupling Current Loss

Two filaments in a strand, as well as two noninsulated strands in a cabled conductor, constitute a loop for
induced currents under a perpendicular time-varying field, that is, they are coupled in a current loop. A large
portion of the loop is superconducting, that is, the linked area is large but the loop resistance is small. To
reduce the linked area, the filament bundle is twisted and the strands are cabled with tight pitches, leading to
transposition with respect to the perpendicular field (31).
The magnetic energy, initially stored in the coupling current loop, is released as Joule heating by the
resistive decay of the induced currents, with a time constant, , which is the ratio of the loop inductance to the
loop resistance. In a round, ideal multifilamentary strand, with the filaments homogeneously distributed over
the cross section, the time constant for the interfilament coupling currents is a function of the twist pitch lt
and the transverse resistivity
The transverse resistivity in a multifilamentary composite is a function of the bulk resistivity m , of

the matrix and the superconducting fraction in the filamentary zone . If no resistance barrier is found at the
13
interface between filaments and matrix, according to 32 the transverse resistance is
When a high resistivity barrier builds up around the filaments (e.g., in hot extruded NbTi composites), the
transverse resistance is assumed to be
In a cable of noninsulated strands, the interstrand coupling currents add to the interfilament loops. In a
multistage cable, a large variety of coupling current loops exists, each with an individual time constant i . The
size of the loops depends on the length and sequence of the cable pitches, but the exact path of the coupling
currents, and hence the transverse resistance, is hard to predict (33). As a general trend, the transverse
resistance is larger and the loss is smaller when the pitches of the different cable stages all have the same
direction (34) and their ratio is close to one, that is, short pitches are used for the higher cable stages and long
pitches for the lower cable stages.
In most cases, the loss is not homogeneously distributed over the strand or cable volume: a dimensionless
geometry factor, ni , is associated with each current loop with time constant i . The geometry factor (35,36,37)
allows for the demagnetization effects (e.g., round versus flat conductor) and normalizes the loss to the overall
strand volume (e.g., when the filament bundle is surrounded by a large normal metal shell) or to the cable
volume (e.g., when an interstrand current loop is restricted to a fraction of the cable volume).
Steady State Coupling Loss Formulas. When the time scale of a field change (e.g., the duration
of a linear ramp or the period of a field oscillation) is much larger than any of the conductor time constants,
steady state conditions are established for the coupling currents. For linear field change, with constant dB/dt,
the power loss (10, 35, 38) per unit volume of strand material is
For sinusoidal field variations B = (B/2 sin t, with frequency and = 2, the average power loss (10, 35, 39)
is
The energy loss Qc for a field cycle of amplitude B is respectively
14
For cabled conductors with multiple current loops and associated time constants, the n in the preceding
formulas is the sum of the individual terms
In steady-state conditions, where all the current loops are fully activated, it is not necessary to know the
breakdown of n into individual components. The average coupling loss can be calculated from the overall n,
obtained, for example, from measurements on a short conductor section.
The tool for experimental assessment of the coupling currents loss is the loss curve, where the energy per
cycle per unit volume of strand (or cable) is plotted as a function of the field rate, for a linear ramp, or the
frequency, for a sinusoidal field sweep. The hysteresis loss is the extrapolation of the loss curve to dB/dt = 0.
From the initial slope of the loss curve, n is derived using the steady-state formulas, Eq. (22).
Transient Coupling Loss Formulas. The energy loss per unit volume in Eq. (22) is linear in the field
rate or frequency. However, the energy loss has an obvious upper limit set by the magnetic field energy density,
B2 /20 . At > 0.3 and at ramp time t0 < 10, Eqs. (20 21 22) give a loss overestimation larger than 10% and
should be replaced by transient field loss formulas.
For multifilamentary strands in an oscillating field, the currents flowing in the outer filament layers
screen the inner volume of the conductor. In a fast ramped field, the field penetrates the innermost layers with
the time scale of the decaying screening currents , even if the duration of the applied field change is smaller.
In fully transposed cables, the mechanism of screening depends on the interstrand current loops and is hardly
predictable.
For conductors characterized by a single time constant , the transient formulas for sinusoidal oscillations
(35), linear ramp (10), and exponential decay (40) are respectively for B = (B/2) sin t,
for = B/t0
for B = B(1 e t/t 0 )
In case of multiple time constants, Eqs. (24 25 26) cannot be applied using the n defined in Eq. (23). If each
current loop behaved independently (i.e., the screening currents of the largest loops did not affect the applied
field at the other loops), the total transient loss would be the sum of the individual i-contributions. From
15
Eqs. (24) and (25)
On the assumption that the larger current loops screen the smaller current loops, which is what happens in
nontransposed conductors, a formula for the transient loss in a sinusoidal field has been proposed in 37 for a
conductor with N loops
It is hard to reliably predict the transient coupling loss in a large multistage cable. On one hand, the
overall n should be broken down into the individual ni i : the procedure to add the ns measured separately
on the lower cable stages is not satisfactory, as the transverse resistance and the current loops change when
the subcables are bundled together. On the other hand, because the current path for each loop is not known,
it is hard to decide to what extent the higher loops do screen the smaller loops, that is, to decide between Eqs.
(28) and (29).
In conductors with multiple time constants, the largest underestimation of the transient field loss occurs
when the Eqs. (24 25 26) for a single time constant are applied. The steady state formulas, Eqs. (20 21 22), give
the largest overestimation. Whenever the breakdown of the overall n is known, Eqs. (2728) give a better, but
still conservative, estimate, because they assume no screening. On the contrary, Eq. (29) is rather optimistic,
because it treats the current loops as nested shells. Equation (29) has been used to find the n components
from the experimental loss curves in 37 and 41.
An example of coupling loss prediction for a conductor with multiple time constants is shown in Fig. 7. The
steady-state, overall time constant is assumed to be n = 100 ms (e.g., drawn from an experimental loss curve).
The breakdown of the time constant is assumed to be 1 = 2 ms, 2 = 8 ms, 3 = 40 ms, and n1 = n2 = n3 = 2
(round cross section). The plot in Fig. 7 shows the loss according to Eqs. (22), (24), (27), and (29), for an applied
field oscillation B = (B/2)/sin t, with B = 2 T. At low frequencies (i.e., for < 0.2), all the formulas give the
same result, but at higher frequencies the assumptions about the current paths and the multiple screening
lead to substantially different results.
Saturation of Coupling Currents. According to the coupling loss formulas, the energy dissipation
occurs in the resistive section of the induced current loops, that is, in the composite matrix for interfilament
coupling loss and at the strand-to-strand contacts for the interstrand coupling loss. At a first approximation
level, the coupling currents are assumed not to change the filament magnetization. Actually, the coupling currents flowing in the outer filament layers create in steady state a field difference across the multifilamentary
zone. The associated magnetization loss is referred to as penetration loss and can be treated in analogy to the
hysteresis loss of a solid filament of the size of the filamentary zone, Dfz , with a critical current J c , where
16
Fig. 7. Example of prediction of coupling loss in oscillating field, for a conductor with multiple time constant, n = n1 1 + n2
2 + n3 3 = 4 + 16 + 80 = 100 ms. The plot shows the result obtained neglecting the shielding effect, Eq. (22), and with three
different approaches to taking account of shielding.
is the superconductor fraction in the filamentary zone (39). In steady state, the penetration loss per cycle Qp ,
normalized to the volume of the filamentary zone, can be easily found by replacing Bp by in the formulas
of Table 1:
The total loss in a round multifilamentary composite with n = 2 is the sum of the coupling loss, according to
Eq. (20), and the penetration loss
Whenever the loss is experimentally assessed, the penetration loss does not need to be added to the
coupling loss, because it is already buried in the n inferred from the loss curve. The penetration loss for an
oscillating field and for a transient field are discussed in 39 and 10.
At high field rates, the coupling currents may reach the critical current. The outer filament layer is
saturated, and the difference between outer and inner fields is the penetration field for the filamentary zone,
= Bfz p = 0 J c Dfz /. Saturation in a multifilamentary composite occurs whenever
17
The saturation loss is the upper limit of the penetration loss and is obtained by substituting Bfz p for in Eq.
(30):
When the condition of Eq. (32) is fulfilled (i.e., when the current loops are saturated), the coupling loss
does not increase for higher field rates. The maximum total loss in a multifilamentary composite for a long
duration (steady-state) field change is independent of and can be written by substituting Eqs. (32) and (33)
into Eq. (31):
Coupling Loss with Transport Current. As long as coupling currents and transport current use a
small fraction of the superconducting cross section, the influence of the transport current is limited to the
hysteresis loss change. At a higher field rate or higher i (ratio of transport current to critical current), all the
superconducting cross section is eventually engaged to carry either the transport or the coupling currents. The
criterion for saturation with transport current in Eq. (32) becomes
The larger i is the smaller the loop current (and hence the field rate) is to achieve saturation. Above saturation
(i.e., when 2Iloop + Itransport > Ic ), the excess of transport current must be accommodated in the superconducting
cross section carrying Iloop . The paths of the coupling currents, with the current direction reversing with the
periodicity of the transposition pitch, force the transport current (or a fraction of it) to switch continuously from
one to the other filament (interfilament coupling) or strand (interstrand coupling) to match Iloop . The energy
dissipated is at the expense of the power supply, and it is called the dynamic resistance loss (29), because of
the analogy between hysteresis and coupling loss due to transport current (see 42 for a discussion of the limits
of this analogy). Whenever a transport current is imposed, the magnetic energy density of the applied field
cannot be considered as an upper limit for the overall loss.
Above saturation, the coupling currents (and coupling loss) decrease and the dynamic resistance loss
sharply increases. The behavior of the total loss as a function of i over the full range of has been calculated
analytically for a slab; see Fig. 8 (from 29). A cylinder requires a numerical calculation for the saturation range
(43), leading to a results similar to Fig. 8. Experimental results on interfilament loss with transport current
(29, 43) confirm the behavior of Fig. 8. At a very large field rate (i.e., when saturation occurs even at i = 0), the
effect of the transport current on the overall loss is an increase by a factor 1 + i2
At an intermediate field rate (i.e., when saturation is achieved only above a certain value of transport current)
18
Fig. 8. Normalized loss, Q/( fz p /0 ), in a slab geometry as a function of the transport current fraction i for different = /Bfz p
(29). The locus of the knees of the curves shows the saturation threshold. Above saturation, the loss is dominated by the
dynamic resistance. Reprinted from Cryogenics 20, T. Osagawara, Y. Takahashi, K. Kanbara, Y. Kubota, K. Yasohama, and
K. Yasukochi, Transient field losses in multifilamentary composite conductors carrying dc transport currents, pp. 216222,
copyright 1980, with permission from Elsevier Science.
the behavior of the total loss as a function of i is complex [see (29, 42, 43)]. Using Qc (i) = (1 + i2 )Qc (0) is not
recommended and may lead to large errors in the actual loss values.
In large cable-in-conduit conductors, the occurrence of saturation cannot be exactly predicted from
Eq. (35). If the current loops (i.e., the coupling currents) are not homogeneously distributed, a redistribution of the transport current at the start of the field change may avoid the occurrence of saturation and
dynamic resistance loss in steady state. Using Eq. (35) with the average i and the overall cable time constant is
very conservative and may result in an underestimate of the field rate causing saturation. In cable in conduit
with multiple time constants, where only the overall n is known, it may be difficult to select the correct to
be used in Eq. (35). An example of saturation of coupling currents in a large cable-in-conduit conductor with
nonhomogeneous current distribution is reported in 44.
Coupling Loss in Flat Cables. In flat cables and in rectangular composites with sides a and b,
both n and are much larger for field perpendicular to the broad side a of the cable. The loss ratio for the
19
two orthogonal orientations has been calculated as a function of the aspect ratio = a/b for conductors with
homogeneous transverse resistivity. The results obtained by Murphy et al. with concentric ellipses (45), Turck
et al. with concentric rectangles (46), and Campbell with rounded-edge concentric layers (35) are respectively
In practical large flat cables, an insulating (or high-resistivity) strip is placed in the midplane to increase the
transverse resistance of the coupling currents for field orientation perpendicular to a; see for example 47. In
such cases, the 38 are no longer valid and the loss anisotropy is reduced (48).
At intermediate angles, an analytical loss formula is proposed in 45, modeling the flat cable as an ellipse
with homogenous transverse resistivity. In practical, nonhomogeneous flat cables, the coupling loss for field
orientation at an angle with respect to the broad side a can be roughly estimated by splitting the field into
the orthogonal components and adding the loss contributions
For flat cables with a large aspect ratio, the second term in Eq. (39) can be neglected over a broad range of
angles.
Coupling Loss in Spatially Changing Magnetic Field. When a superconducting cable is exposed
to a time-varying magnetic field that is not homogeneous along the conductor, the periodicity of the boundary
conditions for the coupling currents loops is affected. If the flux linked by two geometrically identical current
loops next to each other is not balanced, the coupling currents extend beyond the boundary of the pitch length.
In one-stage cables (e.g., one-layer flat cables and Rutherford cables), adjacent strand pairs may build current
loops with different flux balance in the presence of a spatial gradient of magnetic field. Because of the different
boundary conditions, the individual strand pairs carry coupling currents of different amplitude. The inductance
associated with these extended current loops is larger and the resistance is smaller, resulting in time constants
that may be orders of magnitude larger than with strictly periodic boundary conditions. The result is a strong,
quasi-steady-state current imbalance and larger loss.
49 50 first did an analysis of the coupling loss in a spatially changing magnetic field for flat cables.
The subject, later named , (BICCs), assumed a much larger relevance in the context of Rutherford cables for
accelerator magnets. Here, the spatial field gradients along the conductor at the saddles of the dipole magnets
are large and occur over a length smaller than the cable pitch. The long-lasting current imbalance across the
cable leads to field distortions (51, 52) and ramp rate limitations in the accelerator dipoles (53, 54).
The variation of the strand crossover resistance along Rutherford cables has also been shown to be a
potential reason for BICCs (53). In multistage cable-in-conduit conductors, the current loops do not have a
regular pattern, as a result of the nonhomogeneous distribution of the interstrand resistance. Flux imbalance
for current loops next to each other is expected to be frequent in large cable-in-conduit conductors, even in a
spatially homogeneous magnetic field.
Interstrand Resistance in Cable-in-Conduit Conductors. (CICCs) are a special case of multistage
cables. What makes the CICCs different is the tribological nature of the transverse resistance, which is only
marginally determined by the bulk properties of the metallic components. A database for coupling currents
loss in CICCs and its implication for stability is discussed in 55.
In CICCs with void fraction in the range of 30% to 40%, the coupling currents may follow complex, zigzag
paths through a number of good electrical contacts at the strand crossovers. Rather than the 2-D smeared
transversal resistivity, the interstrand resistance is the critical parameter, together with the pitch length, for
20
assessing and controlling the coupling loss in CICCs. The interstrand resistance in CICCs has units of ohmmeters and is measured as the dc resistance between a strand pair embedded in a cable, multiplied by the
length of the cable section, which should be longer than a pitch length. For the same conductor, the interstrand
resistance in a CICC may vary as much as 20% from piece to piece and as a function of the sample length and
homologous strand pair (56).
NbTi Strands. The interstrand resistance depends on the operating transverse load and on the conductor
history, including contact surface oxidation, heat treatment, and cycling. In CICCs made of bare NbTi strands,
the interstrand resistance, as well as the coupling loss, may have a broad range of results (57): the thin
layer of copper oxide that develops at room temperature at the strand surface provides a precarious resistive
barrier, which may either partly dissolve or break under heat treatment and/or applied load (58 59 60). The
electromagnetic load in operation can also produce a dramatic increase in the coupling loss in CICCs made
of bare NbTi strands (61, 62). Whenever long term, reliable control of the coupling loss is desirable, it is
recommended to use a surface coating for the NbTi strands.
Several coatings have been investigated on NbTi strands for contact resistance, mostly in the frame of the
Rutherford cable development for accelerator magnets, including Zn, SnAg, Ni, and Cr. The SnAg (Stabrite) soft
coatings produce very low contact resistance with applied load (59, 63) and are not recommended for low-loss
cables. A resistive barrier can be obtained by a diffusion heat treatment at 200 C of the SnAg-coated strand
before cabling, building a Sn-rich bronze shell at the surface; however, the diffusion heat treatment also affects
the (RRR) of a fraction of the stabilizer. The Ni coating, as well as the Zn, has higher contact resistance and
is not sensitive to curing heat treatment and applied load (63). The Cr coating has proved to be effective in
cutting the interstrand loss in medium-sized CICCs (64, 65). A direct comparison of contact resistance for bare
and Cr- and Ni-plated strands indicates almost one order of magnitude higher resistance for Cr than for Ni
and another order of magnitude difference between Ni and clean Cu (57).
Nb3 Sn Strands. In CICCs of bare Nb3 Sn strands, diffusion bonding (sintering) occurs at a number of
strand crossovers during heat treatment, resulting in low interstrand resistance and high coupling loss (66).
The Cr coating has been identified as a reliable, thin coating to avoid sintering during the heat treatment, with
moderate effect on the strand RRR.
The effect of the Cr plating on the coupling loss of Nb3 Sn-based CICCs has been the object of systematic
investigations. A comparison of CICC samples identical except for Cr plating by vendors (67) has found a
variation as high as a factor of four for the interstrand resistance (and the loss), depending solely on the proprietary electroplating processes. The influence of the Cr plating process parameters on the contact resistance
is discussed in 68.
The void fraction is an important parameter affecting the interstrand resistance and coupling loss in
CICCs, with a low loss range above 40% voids and a large loss range below 30% (69). The looseness of the cable
in the jacket may play an even more important role than the average void fraction: the electromagnetic forces
associated with the coupling currents tend to pull apart the strand bundle and relax the contacts at the strand
crossovers.
The most striking effect observed in the interstrand resistance of Cr-plated Nb3 Sn CICCs is the decrease
in resistance after heat treatment and the increase after mechanical (70) and/or electromagnetic loading (44,
71). The diagram in Fig. 9 shows the evolution of the interstrand resistance at different steps: the large drop
after the heat treatment may be due to a partial depletion of O from the Cr oxide at the sealed surface of
the strand crossovers. The subsequent increase of the interstrand resistance after bending loads (as happens
in the react and transfer coil manufacturing process) or transverse loads (Lorentz forces in operation) can be
understood in terms of microscopic movements of the crossovers, which partly lose their initial engagement.
An important lesson learned from these results is that most loss test results obtained for short samples in
the virgin state (i.e., as heat-treated, without any load) overestimate the loss by about one order of magnitude
compared with the actual coil operating conditions.
21
Fig. 9. Evolution of the interstrand resistance in a Cr-plated Nb3 Sn CICC (81 strands, 30%), before and after the heat
treatment and after controlled bending strain.
Target Interstrand Resistance for CICCs. The goal of the designer is to obtain in operation a value
of interstrand resistance large enough to reduce the coupling loss to an acceptable level but small enough to
allow fast and effective current redistribution among the strands and provide stability under transient local
disturbances.
The experimental results on the interstrand resistance Rc and the coupling current constant n, measured
on the same Nb3 Sn CICCs specimens, have been correlated in 55 to obtain
Once the acceptable coupling loss is assessed and the cable pitch lt is known, Eq. (40) indicates the minimum
design value for the interstrand resistance, that is, for the kind of strand coating to be selected. The maximum
allowable value for the interstrand resistance is assessed by the requirement on the interstrand current sharing
length, l0 . According to 72
where R (/m) is the normal longitudinal resistance of the strand per unit length. Combining Eqs. (40) and
(41), a criterion for interstrand resistance in CICCs summarizing the coupling loss and stability requirements
22
is obtained according to 55
In very large CICCs, Eq. (42) may need a correction if the interstrand resistance for the first cable stage, to
be used for the current-sharing length, is significantly smaller than in the cable stage where the largest loss
occurs (73).
Accuracy of Coupling Current Loss Estimation. The calculation of the coupling current loss in
steady-state conditions is based on the experimental value of n, drawn from the loss curve. In other words, to
estimate the coupling loss for a coil in operation, it is necessary to measure first the loss on a conductor specimen
under controlled conditions. The assessment of n from the pitch and the expected transverse resistivity may
lead to very rough loss estimates.
The accuracy of n depends on the loss calibration method, the resolution and reproducibility of the loss
measurements, and the number of test points in the linear range of the loss curve (n is inferred by linear
regression from the initial slope of the loss curve). In multistage, cabled conductors, the loss curve may show
slight deviations from linearity at a very low field rate, possibly due to very large time constants associated
with the very small factor n (41). In such cases, the actual operating field rate should drive the choice of the
range of the loss curve from which n must be inferred. Typical experimental values of n have an error bar
larger than 10% (74).
When the coupling currents are not in steady state (transient field change or high frequency oscillations),
the product n from the loss curve must be broken down into n and to use the transient formulas from Eqs.
(24 25 26 27 28 29). This is very hard to do, especially in the case of multiple time constants. A measurement
of from the decay time of the induced currents is discussed in 75 for conductors with a single (or dominant)
time constant.
Besides the issue of breaking down the overall n, the prediction of non-steady-state loss in conductors
with multiple time constants is made difficult by the choice between a model with independent current loops
or with nested loops (see discussion of Fig. 7). For 1 or ramp time t0 , the error in loss estimation may be
large and the transient formulas should be used only to provide the order of magnitude of the expected loss.
In large cable-in-conduit conductors, the accuracy of the coupling loss estimation is much worse than
in multifilamentary composites. On one hand, it is hard to reproduce in a short sample measurement the
actual load history for interstrand resistance and hence to obtain a reliable result for n in operation. On the
other hand, the nonhomogeneous distribution of the interstrand resistance may give rise, in a coil, to complex
patterns of BICCs, not measurable on a short sample (61, 62). In the presence of a transport current, more
incertitude is added because of the occurrence of dynamic resistance loss and the variation of current density
across the cable (44, 73).
Concluding Remarks
Although the basic mechanisms of ac losses in superconductors are well known, the optimization of low-loss
conductor design and the prediction of ac losses in operation are still the subject of several studies and much
R&D activity. The complexity of the conductor layout and the variety of the boundary conditions conclusively
show that the existing formulas and models need systematic improvement.
The challenging task for the designer consists in distinguishing, for an individual application, the relevant
from the negligible sources of ac losses and implementing measures to reduce their effect on the overall
performance of the device. Reduction of ac losses is always a welcome result, but it must be carefully balanced
23
with possible negative effects (e.g., reduction of the overall current density due to passive components or poor
stability due to impaired current sharing).
The task of calculating ac losses may be challenging, although no complex finite-element models are
required, as they are for example in thermohydraulic and mechanical analyses. For the hysteresis loss a rough
estimate is easy to make, because the results for a single filament can be reliably scaled to large conductors.
However, very accurate predictions are not easy.
Estimating the coupling current loss for large conductors may be difficult. The electrodynamic behavior
can be predicted only on the basis of several assumptions, which can hardly be verified. Small-scale experiments,
even on full-sized conductors, are mostly unable to fully reproduce the actual operating conditions.
BIBLIOGRAPHY
1. D. Saint-James E. J. Thomas, G. Sarma, Type II Superconductivity, New York: Pergamon, 1969.
2. A. M. Campbell J. E. Evetts, Flux vortices and transport currents in type II superconductors, Adv. Phys. 21 (90):
333357, 1972.
3. H. Ullmaier, Irreversible Properties of Type II Superconductors, New York: Springer-Verlag, 1975.
4. C. P. Bean, Magnetization of hard superconductors, Phys. Rev. Lett., 8(6): 250253, 1962.
5. C. Y. Pang, P. G. McLaren, A. M. Campbell, Losses in superconducting cylinders in transverse field, Proc. ICEC, 8:
739743, 1980.
6. A. P. Martinelli B. Turck, Some effects of field orientation on the magnetization of superconducting wires, Cryogenics,
18: 155161, 1978.
7. Y. B. Kim, C. F. Hempstead, A. R. Strnad, Critical persistent currents in hard superconductors, Phys. Rev. Lett., 9:
306309, 1962.
8. H. Brechna C. Y. Hwang, Hysteretic effects in pulsed superconducting magnets, Cryogenics, 19: 217223, 1979.
9. K. Kwasnitza P. Bruzzone, Hysteresis losses of multifilament superconductors in superimposed dc and ac fields,
Cryogenics, 21: 593597, 1981.
10. M. N. Wilson, Superconducting Magnets, London: Oxford Univ. Press, 1983.
11. E. W. Collings, Applied Superconductivity, Metallurgy and Physics of Titanium Alloys, New York: Plenum, 1986.
12. M. P. Mathur et al., Anisotropy of the critical current density in the NbTi filaments of round composite superconductors,
J. Appl. Phys., 43 (9): 38313833, 1973.
13. H. H. Farrell, G. H. Gilmer, M. Suenaga, Grain boundary diffusion and growth of intermetallic layers: Nb3 Sn, J. Appl.
Phys., 45: 40254035, 1974.
14. T. Schild, J. L. Duchateau, D. Ciazynski, Influence of the field orientation on the critical current density of Nb3 strands,
IEEE Trans. Appl. Supercond, 7: 15121515, 1997.
15. S. Takacs, M. Polak, L. Krempasky, Critical currents of NbTi tapes with differently oriented anisotropic defects,
Cryogenics, 23: 153159, 1983.
16. M. Jergel S. Takacs, The dependance of critical current of Nb3 Sn tapes on the direction of a transverse magnetic field,
J. Low Temp. Phys., 10 (3/4): 469, 1973.
17. J. F. Artaud, Distribution de courant dans des fils supraconducteurs soumis a` un champ magnetique dorientation
quelconque, PhD Thesis, Univ. de Aix-Marseille, France, 1994.
18. T. S. Kreilick, E. Gregory, J. Wong, Influence of filament spacing and matrix material on the attainment of high quality,
uncoupled NbTi fine filaments, IEEE Trans. Magn., 24: 10331036, 1988.
19. M. Polak et al., Anomalous magnetization behavior in fine filamentary NbTi superconducting wires, IEEE Trans. Appl.
Supercond., 3: 150153, 1993.
20. M. D. Sumption E. W. Collings, Influence of twist pitch and sample length on proximity effect coupling in multifilamentary composites described in terms of a field independent, two current region model, Cryogenics, 34: 491505,
1994.
21. A. K. Gosh, K. E. Robins, W. B. Sampson, Magnetization measurements of multifilamentary Nb3 Sn and NbTi conductors,
IEEE Trans. Magn., 21: 328331, 1985.
22. R. B. Goldfarb K. Itoh, Reduction of interfilament contact loss in Nb3 superconductor wires, J. Appl. Phys., 75: 2115
2118, 1994.
24
23. R. B. Goldfarb J. W. Ekin, Hysteresis losses in fine filament internal-tin superconductors, Cryogenics, 26: 478481,
1986.
24. S. S. Shen, Effect of the diffusion barrier on the magnetic properties of practical Nb3 Sn composites, Adv. Cryog. Eng.
Mat., 28: 633638, 1982.
25. T. Ogasawara et al., Alternating field losses in superconducting wires carrying dc transport currents: Part 1. Single
core conductors, Cryogenics, 19: 736740, 1979.
26. T. Ogasawara et al., Effective resistance of current carrying superconducting wire in oscillating magnetic field 1: Single
core composite conductor, Cryogenics, 16: 3338, 1976.
27. S. S. Shen R. E. Schwall, Interaction of transport current and transient external field in composite conductors, IEEE
Trans. Magn., 15: 232235, 1979.
28. W. T. Norris, Calculation of hysteresis losses in hard superconductors carring ac: Isolated conductors and edges of thin
sheets, J. Phys. D, 3: 489507, 1970.
29. T. Ogasawara et al., Transient field losses in multifilamentary composite conductors carrying dc transport currents,
Cryogenics, 20: 216222, 1980.
30. D. Ciazynski et al., Ac losses and current distribution in 40 kA NbTi and Nb3 Sn conductors for NET/ITER, IEEE Trans.
Appl. Supercond., 3: 594599, 1993.
31. M. N. Wilson et al., Experimental and theoretical studies of filamentary superconducting composites, J. Phys. D, 3:
15171546, 1970.
32. W. J. Carr, Jr., Conductivity, permeability and dielectric constant in a multifilament superconductor, J. Appl. Phys., 46
(9): 40434047, 1975.
33. K. Kwasnitza P. Bruzzone, Large ac losses in superconducting Nb3 Sn cable due to low transverse resistance, Proc.
ICEC11, 741745, Berlin, 1986.
34. K. Kwasnitza I. Horvath, Experimental evidence for an interaction effect in the coupling losses of cabled superconductors, Cryogenics, 23: 914, 1983.
35. A. M. Campbell, A general treatment of losses in multifilamentary superconductors, Cryogenics, 22: 316, 1982.
36. B. Turck, Effect of the respective position of filament bundles and stabilizing copper on coupling losses in superconducting composites, Cryogenics, 22: 466468, 1982.
37. A. Nijhuis et al., Coupling loss time constant in full size Nb3 Sn CIC model conductors for fusion magnets, Adv. Cryog.
Eng. Mat., 42B: 12811288, 1996.
38. H. Brechna G. Ries, Ac losses in superconducting synchrotron magnets, IEEE Trans. Nucl. Sci., 18 (3): 639642, 1971.
39. G. Ries, Ac losses in multifilamentary superconductors at technical frequencies, IEEE Trans. Magn. 13 (1): 524527,
1977.
40. J. P. Soubeyrand B. Turck, Losses in superconducting composites under high rate, pulsed transverse field, IEEE Trans.
Magn., 15: 248251, 1979.
41. A. Nijhuis et al., Electromagnetic and mechanical characterization of ITER CS-MC conductors affected by transverse
cyclic loading, part 1: Coupling currents loss, IEEE Trans. Appl. Supercond. 9, 10631072, 1999.
42. A. M. Campbell, The effect of transport current and saturation on the losses of multifilamentary superconducting wires,
Cryogenics, 21: 107112, 1981.
43. D. Ciazynski, Effect of the transport current on the losses of a superconducting composite under fast changing magnetic
field, IEEE Trans. Magn., 21: 169172, 1985.
44. P. Bruzzone et al., Test results for the high field conductor of the iter central solenoid model coil, Adv. Cryog. Eng., 45,
to be published.
45. J. H. Murphy et al., Field orientation dependence of ac losses in rectangular multifilamentary superconductors, Adv.
Cryog. Eng., 22: 420427, 1975.
46. B. Turck et al., Coupling losses in a rectangular multifilamentary composite, Cryogenics, 22: 441450, 1982.
47. E. Seibt, Investigations of a steel reinforced NbTi superconducting flat cable for toroidal field magnets, IEEE Trans.
Magn., 15: 804807, 1979.
48. P. Bruzzone, Ac losses in high current superconductors for nuclear fusion magnets, Ph. D. Thesis, ETH 8224, Zurich,
1987.
49. G. Ries S. Takacs,

Coupling losses in finite length of superconducting cables and in long cables partially in magnetic
field, IEEE Trans. Magn., 17: 22812284, 1981.
50. S. Takacs,
Coupling losses in cables and in spatially changing ac fields, Cryogenics, 22: 661665, 1982.
25
51. A. A. Akhmetov, A. Devred, T. Ogitsu, Periodicity of crossover currents in a Rutherford-type cables subjected to a
time-dependent magnetic field, J. Appl. Phys., 75: 31763183, 1994.
52. L. Krempasky C. Schmidt, Theory of supercurrents and their influence on field quality and stability of superconducting
magnets, J. Appl. Phys., 78: 58005810, 1995.
53. A. P. Verweij, Electrodynamic of superconducting cables in accelerator magnets, Ph.D. Thesis, Univ. of Twente, Enschede, The Netherlands 1995.
54. L. Krempasky C. Schmidt, Ramp rate limitation in large superconducting magnets due to supercurrents, Cryogenics,
36: 471483, 1996.
55. P. Bruzzone, Ac losses and stability on large cable-in-conduit superconductors, Phys. C, 310: 240246, 1998.
56. A. Nijhuis, P. Bruzzone, H. H. J. ten Kate, Influence of Cr plating on the coupling loss in cable-in-conduit conductors,
Appl. Supercond. 1997, Inst. Phys. Conf., 158: 921924, 1997.
57. M. D. Sumption et al., Contact resistance and cable loss measurements of coated strands and cables wound from them,
58. M. D. Sumption et al., Calorimetric measurements of the effect of nickel and Stabrite coatings and resistive cores on
ac loss in accelerator cables under fixed pressure, Adv. Cryog. Eng., 42: 13031311, 1996.
59. E. W. Collings et al., Magnetic studies of ac loss in pressurized Rutherford cables with coated strands and resistive
cores, Adv. Cryog. Eng., 42: 12251232, 1996.
60. D. Richter et al., Dc measurement of electrical contacts between strands in superconducting cables for the LHC main
magnets, IEEE Trans. Appl. Supercond., 7: 786789, 1997.
61. T. Hamajima et al., Ac loss performance of the 100 kWh SMES model coil, Proc. of Magnet Technology Conf., 16, Sept.
99, Ponte Vedra, FL.
62. A. Kawagoe et al., Increase of interstrand coupling losses of superconducting cable-in-conduit for acutal condition of
sweep rate, IEEE Trans. Appl. Supercond. 9: 727730, 1999.
63. J. M. Depond et al., Examination of contacts between strands by electrical measurement and topographical analysis,
64. T. M. Mower Y. Iwasa, Experimental investigation of ac losses in cabled superconductors, Cryogenics, 26: 281292,
1986.
65. K. Kwasnitza, A. Sultan, S. Al-Wakeel, Ac losses of a 10 kA NbTi cable-in-conduit superconductor for SMES application,
Cryogenics, 36: 2734, 1996.
66. M. Nishi et al., Test results of the DPC-TJ, a 24 kA40 A/mm2 superconducting test coil for fusion machines, Fusion
technology 1992, Proc. 17th SOFT, 1993, p. 912916.
67. P. Bruzzone, A. Nijhuis, H. H. J. ten Kate, Effect of Cr plating on the coupling current loss in cable-in-conduit conductors,
Proc. ICMC 96, Kitakyushu, Japan, May 1996, Amsterdam, The Netherlands Elsevier, p. 12431248.
68. Y. Ipatov, P. Dolgosheev, V. Sytnikov, Prospective barrier coatings for superconducting cables, Supercond. Sci. Technol.,
10: 507511, 1997.
69. A. Nijhuis et al., Parametric study on coupling loss in subsize ITER Nb3 Sn cabled specimen, IEEE Trans. Magn., 32:
27432746, 1996.
70. P. Bruzzone, A. Nijhuis, H. H. J. ten Kate, Contact resistance and coupling loss in cable-in-conduit of Cr plated Nb3 Sn
strands, Proc. MT-15, Beijing, October 1997, Science Press, 1998, p. 12951298.
71. A. Nijhuis, H. H. J. ten Kate, P. Bruzzone, The influence of Lorentz forces on the ac loss in sub-size cable-in-conduit
conductors for ITER, IEEE Trans. Appl. Supercond., 7: 262265, 1997.
72. G. Ries, Stability in superconducting multistrand cables, Cryogenics, 20: 513519, 1980.
73. P. Bruzzone, Stability under transverse field pulse of the Nb3 Sn ITER cable-in-conduit conductor, Proc. of Magnet
Technology Conf., 16, Sept. 1999, Ponte Vedre, FL.
74. P. Bruzzone et al., Coupling currents losses bench mark test of ITER subsize conductor, IEEE Trans. Magn., 32:
28262829, 1996.
75. L. Krempasky C. Schmidt, Time constant measurements in technical superconductors, Cryogenics, 39: 2333, 1999.
PIERLUIGI BRUZZONE
Centre de Recherches en Physique des Plasmas
MAGNETS FOR MAGNETIC RESONANCE ANALYSIS AND IMAGING
305
Experimentally, NMR is performed as follows (3,4): nuclei

are immersed in a static field B0, which results in the development of a net polarization along the field direction, occurring
with a time constant 1. Transitions among their spin states
are excited by a high-frequency field B1, oriented perpendicular to the static field. In a rotating frame of reference (rotating at the Larmor precession frequency), the magnetization
vector of the spin is tilted from a longitudinal direction (along
B0) toward the transverse plane. The angle of precession depends on the strength and duration of the applied RF field
and is given by
= B 1 p
MAGNETS FOR MAGNETIC RESONANCE

ANALYSIS AND IMAGING
Nuclear magnetic resonance (NMR) was discovered in 1946
by Purcell (1) and Bloch (2). Classically, it is the precession
of the spins of nuclei with magnetic moment, subjected to a
transverse radio frequency (RF) field in the presence of a longitudinal magnetic field. Nuclear species of biological interest
having nonzero magnetic moment are listed in Table 1 together with their Larmor precession frequency-field dependence.
Table 1. The Larmor Precession Constant for Various
Nuclides
Nuclide
Hydrogen
Deuterium
Carbon
Oxygen
Sodium
Phosphorus
Atomic Number
NMR Frequency
(MHz/Tesla)
1
2
13
17
23
31
42.5759
6.5357
10.705
5.772
11.262
17.236
(1)
where p is the pulse width and is the gyromagnetic ratio.

Following the pulse, the magnetization decays transversely with a time constant 2, the spinspin relaxation time.
The polarization develops (or decays) along the field with a
time constant 1, the spinlattice relaxation time.
NMR is the preeminent method for the identification of
chemical species in weak solution. It also has useful applications in solid materials. The most exacting specifications for
an NMR magnet are imposed by high-resolution NMR. The
resonant frequency of a nucleus depends not only on B0 but
also, to a small extent, on the shielding provided by the electronic structure of the chemical compound. This effect is the
chemical shift and is distinctive for each chemical species.
Thus the resonant frequency of the 1H nucleus in water is
different from that in benzene (C6H6) or in the methyl or
methylene group in alcohol (CH3CH2OH). These small differences in frequency are typically a few parts per million and
provide a means to identify the components of a complex molecule.
Early NMR spectrometers used continuous wave (CW)
methods in which the frequency of the B1 field would be
changed slowly and the absorption of a tank circuit enclosing
the sample would be recorded as a spectrum of power absorption versus frequency. At the resonant frequency a sharp increase in absorption would be observed. The width of the peak
depended, among other things, on the magnification Q of the
tank circuit.
In modern NMR, a pulse of RF of sufficient strength and
duration is applied to the sample so that all the spins are
excited. The pulse is then switched off, and the signals emitted at various frequencies by the sample during relaxation of
the spins are monitored. A Fourier analysis of the signal then
transforms the time-dependent spectrum into a frequency-dependent signal, thus revealing the resonance peaks associated
with the chemical shifts (3,4).
The uniformity of the static field B0 is the key to highresolution NMR and to sharp images in magnetic resonance
imaging (MRI). A uniform field allows large numbers of nuclei
to precess at exactly the same frequency, thus generating a
strong signal of narrow bandwidth. The underlying theory
and practice of high-homogeneity superconducting magnets is
described in this section. MRI magnets differ from those used
for NMR analysis in that spatial distribution of either signal
strength or 1 or 2 relaxation times are measured over a volume far greater than that of a sample for chemical analysis.
In MRI, the predominant nuclear species examined is hydrogen in water. Density or 1 or 2 is measured on planes
306
throughout a body and reconstructed as two-dimensional

maps.
The field strengths of NMR magnets are higher than those
of MRI magnets. From the discovery of NMR, in 1946, to
1967, magnetic fields were limited to 2 T that could be generated by electromagnets. A 5 T superconducting magnet was
introduced in 1967, and slow improvements led to the 20 T
magnetic field available in NMR magnets today. The driving
forces for the increase in field strength are the chemical shift
(separation of the nuclear species which is linearly proportional to field strength), signal strength (which is proportional
to the square of the field strength), and signal-to-noise ratio
(which is proportional to the 3/2 power of the field strength).
Although various experimental techniques have been applied
to improve signal-to-noise ratio, including tailored pulse sequences, signal averaging, cooled conventional receiver coils
and superconducting receiver coils, increased field strength is
still desirable for increased chemical shift. Magnets for highresolution NMR are now almost exclusively superconducting,
and it is only that type that is described here.
The optimal field strength for MRI is determined by a
number of factors, including reduction in imaging time, reduction in chemical shift artifacts, and reduction in cost, and
by limits to the exposure of a patient to electromagnetic radiation, as set by regulations. Even though most MRI magnets
have field strengths up to 1.5 T, a few experimental magnets
have been built or designed for functional MRI studies with
field strengths up to 8 T. In order to achieve the desired combination of field strength, working volume, and stability, superconducting magnets represent the principal type employed. However, both water-cooled resistive magnets, ironcored electromagnets, and rare earth permanent magnet MRI
systems have been used or are in use for special applications.
The superconducting magnet is here considered exclusively.
DESIGN PRINCIPLES
NMR Magnets
The analysis of weak solutions imposes several requirements
on the magnet. In order to obtain usable signal-to-noise ratios, a large volume of sample must be used. This immediately demands good field uniformity so that variations in
background field strength do not give rise to different frequencies, which would, of course, mask the small chemical
shifts being sought. High field strength is desired as detailed
above. Even if these requirements are met, the dilution of the
sample may be such that repeated pulses are required. The
final signal-to-noise that can be obtained from a number of
pulses is proportional to N. A run may take many hours or
even days to accomplish. During that time not only must the
spatial homogeneity of the background field be excellent, but
the magnitude of the field also must be constant, or at least
must change only very slightly. (The reason that any change
is permitted is that a frequency lock can be used to adjust the
frequency of the RF to match a slow and slight change in the
background field.)
To summarize, the NMR magnet should have high field
strength and great uniformity, and the field must be stable.
MRI Magnets
The essential principles for MRI magnets are identical to
those for NMR magnets, but the volumes of homogeneity are
much greater, whereas the homogeneity is somewhat lower.

Despite field strengths lower than those of NMR magnets, the
stored magnetic energies of MRI magnets are greater by reason of the large bore, which must be sufficient to house correction coils, pulsed gradient coils, and a patient. The MRI system differs significantly from that of the NMR system by
including means to superimpose linear field gradients on the
background homogeneous field. These pulsed gradients define
thin planes in which the field is known but different from that
elsewhere. Thus the frequency of nuclear magnetic resonance
is spatially encoded so that the signals generated by the relaxing nuclei have frequencies which define their position. As
in NMR for chemical analysis, the MRI signals may interrogate either the density of nuclei or the 1 or 2 relaxation
times.
THEORETICAL DESIGN
Almost all superconducting NMR and MRI magnets are solenoids. The reason for that is the relative simplicity and ease
of manufacture and design of solenoids, compared with, for
instance, extended dipoles. Although the generation of the RF
field could be simpler with a transverse background field, the
difficulty of manufacture of a high-background field magnet
would far outweigh any advantage in the RF coil. The construction of a high-homogeneity solenoid proceeds in three
parts: a winding array is designed, based solely on the analysis of the axial variation of the field of a solenoid; the magnet
is wound and the spatial variation of its actual field is measured; and the unwanted errors in the field arising from manufacturing imperfections are removed by shimming.
The center field of a solenoid is given by
B0 = 0 Ja0 ln{[ + ( 2 + 2 )1/2 ]/[1 + (1 + 2 )1/2 ]}
(2)
where J is the overall winding current density, a0 is the inner

radius, is the ratio of outer to inner radii, and is the ratio
of length to inner diameter (5). Because SI units are used
throughout, 0 is the permeability of free space, 4 107
H/m, a0 is in meters, and B0 is in Tesla.
The field strength decreases at points on the axis away
from the center of the solenoid. The axial variation of field
strength on the z axis is expressible as a Taylors series
B(z) = B0 + (d 2 B/dz2 )z2 /2 + (d 4 B/dz4 )z4 /4!

+ (d 6 B/dz6 )z6 /6! +
(3)
Only even terms appear because the center of the solenoid

coincides with the origin.
Figure 1 illustrates the geometry of a thin solenoid, of radius a0 and extending a length z0 to the right of the origin.
For such a thin solenoid, the derivatives of the field at the
origin are as follows:
B0 = 12 0 iz0 (a20 + z20 )1/2

dB/dz = 12 0 ia20 (a20 + z20 )3/2
d 2 B/dz2 = 12 0 i3z0 a20 (a20 + z20 )5/2
d 3 B/dz3 = 12 0 ia20 (3a20 12z20 )(a20 + z20 )7/2
(4)
j - (At/m)
a0
0
z0
Figure 1. Geometry of a thin solenoid showing the coordinate system

used to define current geometries.
where i is the sheet current density in amp-turns per meter

and a0 and z0 are as illustrated in Fig. 1.
The field of a solenoid symmetric about the center plane
has even symmetry and no odd derivatives. So, by evaluating
the even derivatives of the field at the center, the axial variation of field generated by a solenoid can be calculated to an
accuracy determined by the number of derivatives used and
by the distance from the center. The derivatives can be
treated as coefficients of a Cartesian harmonic series so that
Bz = B0 + b 2 z 2 + b 4 z 4 + b 6 z 6 +
in which
b2 = 0 i[3z0 a20 (a20 + z20 )5/2 ]/4

b4 = 0 i[(45a20z 60z3 )(a20 + z20 )9/2 ]/48
b6 =
0 i[(5a40z
20a20 z3
+ 8z
)(a20
(5)
z20 )13/2]/1440
For coils of odd symmetry, such as shim coils described later,

the corresponding harmonics are
b1 = 0 i[(a20 + z20 )3/2 ]/2

b3 = 0 i[(3a20 12z2 )(a20 + z20 )7/2 ]/12
(6)
Notice that the magnitude of any harmonic coefficient is mediated by the denominator of the expressions that each include the term (a02 z02)(n1/2), where n is the order of the harmonic. Thus, the generation of high-order harmonics requires
coils with large values of current (ampere-turns) or small radius. This is significant in the construction of shim coils, as is
noted later.
Associated with an axial variation of field is a radial variation, arising from radial terms in the solution of the Laplace
scalar potential equation. For instance, even-order axial variations are accompanied by axisymmetric radial variations (6)
of the form
B2 (z, x, y) = b2 [z2 12 (x2 + y2 )]

B4 (z, x, y) = b4 [z4 3(x2 + y2 ) + 38 (x2 + y2 )2 ]
307
Figure 2 illustrates a set of nested solenoids. Solenoid 1

gives rise to nonzero values of the harmonic coefficients b2,
b4, b6, etc. If dimensioned correctly, solenoid 2 by contrast can
produce equal values for some or all these coefficients but
with opposite polarity. Then at least b2 and b4 will have net
zero values, and the first uncompensated harmonic to appear
in the expression for axial field variation will be the sixth
order. A minimum, but not necessarily sufficient, condition is
that as many degrees of freedom are needed in the parameters of the coils as there are coefficients to be zeroed.
This method can be extended to as many orders as desired.
In most high-resolution NMR magnets, the required uniformity of the field at the center is achieved by nulling all orders
up to and including the sixth. That is, the solenoid is of eighth
order. In the design of the solenoids, no odd order appears, of
course. The first residual harmonic will have a very small
value close to the center, although at greater axial distances,
the field will begin to vary rapidly. Thus, the design of a highhomogeneity solenoid requires only the calculation of the field
or the field harmonics on axis, and those harmonics may be
easily calculated using only Cartesian coordinates.
MANUFACTURING ERRORS
The theoretical design of a high-homogeneity magnet can be
simple because only axial terms in the z field need to be considered. However, the manufacturing process introduces errors in conductor placement which generate both even- and
odd-order axial and, most significantly, radial field gradients.
Further, the materials of the coil forms, the nonisotropic contraction of the forms and windings during cool-down to helium temperature and the effects of the large forces between
the windings may also introduce inhomogeneity. Typically,
the homogeneity of an as-wound set of NMR solenoids is not
better than 105 over a 5 mm diameter spherical volume (dsv)
at the center. For high-resolution NMR, an effective homogeneity of 109 over at least 5 mm dsv is required. The improvement of the raw homogeneity to this level is achieved by three
steps, superconducting shim coils, room temperature shim
coils and, in NMR magnets only, sample spinning. (Additionally, in cases of poor raw homogeneity, ferromagnetic shims
may be used occasionally in NMR magnets and routinely in
MRI magnets to compensate for large errors or significant
high-order harmonics.) The presence of radial field gradients
necessitates a more comprehensive field analysis than is convenient with Cartesian coordinates.
2
1
r
(7)
These equations show that if b2 or b4 are zero there will be no

axisymmetric radial variation of field.
Figure 2. Principle of harmonic compensation. Coaxial solenoids

generating field harmonics of opposite sign.
308
The harmonic components of the z field can also be expressed in the form of associated Legendre functions of order
n, m (7). Those functions define the variation of the local z
field strength at points around the center of the magnet and
include variation of the field with azimuth . Thus,
Bz (n, m) = rn (n + m + 1)Pn,m (u)
(10)
[Cn,m cos(m) + Sn,m sin(m)]

y
x
Figure 3. The system of spherical coordinates specifying field points
and current sources.
LEGENDRE FUNCTIONS
The expression of the harmonics of the field in terms of
Cartesian coordinates provides a simple insight into the
source of the harmonics. However, as the order of the harmonic increases, the complexity of the Cartesian expressions
renders manipulation very cumbersome, and an alternative
method is needed. The Laplace equation for the magnetic field
in free space is conveniently solved in spherical coordinates.
These solutions are spherical harmonics, and they are valid
only in the spherical region around the center of the solenoid,
extending as far as, but not including, the nearest current
element. Figure 3 illustrates the coordinate system for spherical harmonics. The convention followed here is that dimensions and angles without subscripts refer to a field point, and
with subscripts they refer to a current source.
The axisymmetric z field generated by a coaxial circular
current loop can be expressed in the form of a Legendre polynomial, thus,
Bz =
gn rn Pn (cos )
(8)
where Cn,m and Sn,m are the harmonic field constants in tesla
per metern, Pn,m(u) is the associated Legendre function of order n and degree m, and u is cos(). The order n is zonal,
describing the axial variation of z field. The degree m is tesseral, describing the variation of the z field in what would be
the xy plane in Cartesian coordinates. is the azimuth to
the point at radius r from an xz plane. is the elevation of
the point from the z axis. Tables of the values of the Legendre
polynomials can be found in standard texts on mathematical
functions (8).
In Eq. (10), m can never be greater than n. For example, if
n m 0, Bz(0,0) is a uniform field independent of position.
If n 2 and m 0, Bz(2,0) is a field whose strength varies as
the square of the axial distance [i.e., B2 of Eq. (7)]. If n 2
and m 2, Bz(2,2) is a field that is constant in the axial direction but increases linearly in two of the orthogonal radial directions and decreases linearly in the other two. Figure 4
shows a map of the contours of constant field strength of a
Bz(2,2) field harmonic for which S2,2 0. The Bz(2,2) field has
zero magnitude at the origin and along the x and y coordinate
axes. Of course, the direction of the zero values of the Bz(2,2)
harmonic will not generally lie in the Cartesian x and y
planes. Depending on the relative values of Cn,m and Sn,m in
Eq. (10), the zero harmonic planes will lie at an angle other
than 0 or m/2. The constant field contours of Bz(2,2)
extend to infinity along the z axis and represent, arbitrarily
in this figure, values for Bz(2,2) of 104, 106, and 108, for
example. Within the indicated cylinder centered on the z axis,
the value of the harmonic is everywhere less than 106. For
y
8
n=0
where r and define the azimuth of the field point in spherical coordinates, and u is cos(). Pn(u) is the zonal Legendre
polynomial of order n and gn is a generation function given by
gn = 0 iPn+1 cos(0 ) sin(0 )/(2 n+1 )
(9)
where 0 and 0 define the position of the current loop in

spherical coordinates. In this text, it is the convention that
n 0 represents a uniform field. The field strength given by
Eqs. (8) and (9) is constant with azimuth at constant radius r.
Equations (8) and (9) are equivalent in spherical coordinates to those of Eqs. (4), (5), and (7) on the z axis but additionally predict the z field off axis. In the design of the main
coils Eqs. (8) and (9) offer no more information than Eq. (5).
However, in the calculation of the off-axis z fields, they provide important additional information that can be used in the
optimization of coil design when fringing fields must be considered.
4
2
Figure 4. Surfaces of constant magnitude of a B(2,2) harmonic field,

showing that the tesseral harmonic is zero when the azimuth is a
multiple of /2.

x
Optimization Methods
102
104
106
39
z
309
Figure 5. Surfaces of constant magnitude of a B(3,0) harmonic field,

showing that the zonal harmonic is zero when the elevation is 39
or 90.
higher values of m, there are more planes of zero value.

Thus, Bz(4,4) has eight planes of zero value, Bz(8,8) has 16,
and so forth. A harmonic Bz(4,2) defines a field in which the
second-degree azimuthal variation itself varies in second order with axial distance.
The zonal harmonics Bz(2,0), Bz(3,0), Bz(4,0) have conical
surfaces on which the value of the field is zero. Thus, for instance, Bz(3,0) has contours of zero value such as are shown
in Fig. 5 to lie at 39 and 90.
The four hyperbolas are actually surfaces of rotation about
the z axis, and each represents a constant value for Bz(3,0) of,
say, 106 [the uniform field, Bz(0,0) at the origin having unity
value]. Within the indicated ellipsoid, centered on the origin,
the value of the Bz(3,0) harmonic is therefore everywhere less
than 106.
The contours of the zero values of the spherical harmonics
are analogous to combinations of Figs. 4 and Fig. 5. The zero
values now lie on straight lines radiating from the origin. The
surfaces of constant value look like the spines of sea urchins.
As for the zonal harmonics, ellipsoidal surfaces roughly describe boundaries within which the magnitudes of the spherical harmonics do not exceed a given value. These error surface diagrams are often used in the design of an MRI magnet
to identify the maximum calculated field error within a central volume caused by the highest uncompensated harmonic.
Thus, in general, the deviations from the ideal uniform solenoidal field can be expressed as the sum of a large number
of harmonics each described by the associated Legendre function of order n and degree m. Although the Cartesian expressions of Eq. (4) can be used for the design of a coil system to
generate a uniform field, the associated Legendre functions
must generally be used for the analysis of the measured field
and the design of shim coils or of ferromagnetic shims to compensate for harmonics with nonzero values of m (9).
With the recent rapid increase in the speed and size of computers, an alternative technique for the design of uniform
field magnets has been developed. Not only is a uniform field
of specific magnitude required but that should be combined
with other criteria. For instance it could be accompanied by
the smallest magnet, that is, the minimum of conductor, or
by a specified small fringing field. To achieve these ideal solutions, an optimization technique is now generally used. The
field strength of a set of coils is computed at points along the
axis, and, if fringing field is a consideration, at points outside
the immediate vicinity of the system. The starting point may
be a coil set determined by a harmonic analysis as described
earlier. Now however, mathematical programming methods
are employed to minimize the volume of the windings satisfying the requirement that the field should not vary by more
than the target homogeneity for each of the chosen points.
Again, for purposes of homogeneity, only field on axis is considered because the radial variation of axisymmetric components of field is zero if the axial component is zero. The field
strengths at points outside the magnet will be minimized by
inclusion of a set of coils of much larger diameter than the
main coils but carrying current of reverse polarity.
All design techniques, but particularly that of optimization, are complicated by the highly nonlinear relationship between the harmonic components generated by a coil and the
characteristics of the coil. Thus the reversal in sign of the
harmonic components occurs rapidly as the dimensions or position of a coil are changed. In the example of an NMR magnet shown in Fig. 6, the value of the second harmonic changes
by 4 ppm for an increase in the diameter of the wire in the
small coil l of only 0.1 mm. The optimization of the ampereturns, shape and position of a coil thus affects the various
harmonics in highly nonlinear and often conflicting ways.
Design optimization involves the computation of an objective function which contains all the elements that have to be
minimized, subject to a set of constraints (10). For example,
it may be required to minimize some combination of winding
volume or magnet length subject to constraints on the field
error at a number of points within the bore and on the fringing field at some point outside the magnet. The objective function would then be of the form
N
pVi + L
(11)
i=1
s t u v
x
1060 mm
w
y
j k
h
g
k
h
fe
dc
b
a
520 mm
Figure 6. Coil profiles of an actual 8th order compensated NMR solenoid. The graded sections a through j produce axial harmonics of
which orders 2, 4, and 6 are compensated by sections k and l. Layers
x and y are shim coils.
310
where Vi is the volume of a coil, N is the number of coils, and

L is the length of the magnet. The factor p weights the relative importance of the volume and of the length. This objective function is then minimized subject to the following constraints:
N
+8 +6
+4 +2
6+
4+
2
Bi, p B0
< B2
(12)
+8
i=1
N
Bi,f < B f
+3
8+
2 3 1 3
(13)
i=1
In Eq. (12) Bi,p is the field at point p due to coil i. The equation
represents the constraint on uniformity of field. It could also
be expressed in terms of harmonic terms; for example, each
even term up to P10,0 being less than 106 B0, the center field.
[The inclusion of the squared terms in Eq. (12) allows for either positive or negative error field components.] Equation
(13) expresses the condition that the fringing field should be
less than, say, 1 mT (10 gauss) at a point, outside the magnet
system. The 10 gauss criterion frequently represents the maximum field to which the public may be exposed in accessible
areas around an MRI system.
The minimization of the objective function is performed by
a mathematical programming algorithm, whereas the solution of the constraining Eqs. (12) and (13) will require a nonlinear technique (such as NewtonRaphson), in order to deal
with the extremely nonlinear variation of the harmonics as
they change with coil geometry (11).
SHIELDING
The minimization of the external fringing field is becoming
increasingly important for the siting of MRI systems, so the
active shielding of MRI magnets with center fields up to 2 T
is now almost universal. (Active shielding of MRI magnets
with center fields above 2 T is uneconomical and is not generally attempted.) Active shielding is generally achieved by the
inclusion in the coil array of two reverse polarity coils at diameters typically twice that of the main coils. Because of the
large dipole moment of an MRI magnet, the unshielded fringing field will extend several meters from the boundary of the
cryostat. Consequently, active shielding is applied to many
MRI magnets with central fields of over 0.5 T (12). The effect
of the shielding on the harmonics of the center field must, of
course, be included in the design of the compensation coils.
SHIMMING
The harmonic errors in the field of an as-built magnet divide
into purely axial variations (axisymmetric zonal harmonics,
which are accompanied by radial variations dependent on the
elevation from the z axis, but independent of ) and radial
variations (tesseral harmonics, which depend on , where
is the angle of azimuth in the xy plane).
In order to compensate for the presence of various unwanted harmonic errors in the center field of the as-built
coils, additional coils capable of generating the opposite harmonics are applied to the magnet. For each set of n and m
in the associated Legendre functions, a current array can be
Figure 7. A set of axial shim coils for harmonic correction up to

B(3,0). These coils generate small harmonics of 4th order and higher.
designed in the form of a set of arcs of varying azimuthal

extent and symmetry and with various positions and extents
along the z axis. The magnitude of the harmonic field that an
array generates can be controlled by the current. This is the
principle of variable harmonic correction for both MRI and
NMR magnets. (Correction by means of ferromagnetic shims
is not variable.)
The shimming of the unwanted harmonics is a process in
two independent parts. First, there is the design of as many
sets of coils as are needed to generate the compensating harmonics. Second is the measurement of the actual field errors
to determine the magnitudes of the various harmonic components and the application of currents to the previously designed coils to provide the compensation. In fact, because superconducting shims must be built into the magnet prior to
installation in the cryostat and cooldown, the range of harmonic errors in the field of the as-built magnet must be
largely anticipated. Typically it might be assumed that the
level of harmonic error decreases by a factor of three for each
unit increase in n or m. Therefore, as a rough guide it has
been found that compensation of up to B(3,0) for the zonal
harmonics and up to B(2,2) for the tesseral harmonics is satisfactory in most cases for the superconducting shims of small
bore NMR magnets. There will also be a set of room temperature shims in a high resolution NMR system. Those will compensate for errors typically up to B(6,0) and B(3,3) in many
cases. Typically there may be up to 28, but exceptionally up
to 45 independent shims in all. They will be constructed according to a different principle from the superconducting
shims. The shimming of MRI magnets is accomplished by current shims, typically up to n 3 and m 2, and by ferromagnetic shims.
Superconducting Axial Shims
These will be simple circular coils combined in groups so as to
generate a single harmonic only (13). Thus, a coil to generate
B(3,0) must generate no B(1,0) nor B(5,0). Because the superconducting shim coils need to generate only a small fraction
of the field due to the main coil, they generally need only comprise one to three layers of conductor. For that reason the
harmonic sensitivities can be calculated directly from Eqs. (4)
and (5). A set of axial shims providing correction of B(n,0)
harmonics for n 1 through 3 are shown in Fig. 7. Note that,
for a fixed linear current density, only the angles defining the
start and end of each coil are needed, together, of course, with
the current polarities, either side of the center plane of the
magnet, odd for n 1, 3, 5, . . . and even for 2, 4, 6, . . ..
The set of coils illustrated in Fig. 7 generate negligible harmonics above the third order, B(3,0). The individual coils of
each harmonic group are connected in series in sets, there
being in each set enough coils to generate the required axial
harmonic but excluding, as far as is practical, those harmonics that are unwanted. Thus, in the figure, coils labeled 2 generate second-order B(2,0) but no fourth order. However, they
do generate higher orders. The first unwanted order is B(6,0)
but that is small enough that it may be neglected. So also
with all higher orders because the denominator in the expressions of Eqs. (4) and (5) strongly controls the magnitude of
the harmonic. Also illustrated in the figure is the effect on
harmonic generation of the angular position of a circular current loop. Each of the dashed lines lies at the zero position of
an axial harmonic. Thus, at an angle of 70.1 from the z axis,
the B(4,0) harmonic of a single loop is zero. Two loops carrying currents of the same polarity and suitable magnitude
may be located on either side of the 70.1 line to generate no
fourth-order harmonic yet generate a significant second order
harmonic. Similarly, a coil for the generation of only a first
order axial harmonic is located on the line for zero third order. The zero first-order harmonic line is at 90, the plane of
symmetry. In order therefore to generate a third order with
no first, two coils must be used, with opposing polarities. The
coils are all mirrored about the plane of symmetry, but the
current symmetries are odd for the odd harmonics and even
for the even harmonics. The loops may be extended axially as
multiturn coils while retaining the property of generating no
axial harmonic of a chosen order, if the start and end angles
subtended by the coils at the origin are suitably chosen.
The principles described earlier can be applied both in the
design of shim coils and in the selection of main coil sets. A
further observation from the zero harmonic lines of Fig. 7 is
that the higher harmonics reverse sign at angles close to the
plane of symmetry of the system. This implies that, to produce single, high-order harmonics, coil positions close to the
plane of symmetry must be chosen because the other coil locations where the sign of the harmonic reverses are too far from
the plane of symmetry to be usable; the coils lying a long way
from the plane of symmetry generate weak high-order harmonics.
311
radial harmonics may be generated. For the shim coil configuration of Fig. 8, the first unwanted radial harmonic is m
6. The higher tesseral harmonics are much smaller than the
fundamental because of the presence in the expression for the
field of a term (r/r0)n. Generally, the arc length is chosen to
eliminate the first higher-degree radial harmonic. As an example, if the arc length of each shim coil shown in Fig. 8 is
90 the B(6,6) harmonic disappears. The B(10,10) harmonic
is negligible.
The superconducting shims are almost invariably placed
around the outside of the main windings. Although the large
radius reduces the effective strength of the harmonics they
generate, the shim windings cannot usually be placed nearer
to the center of the coil because of the value of winding space
near the inner parts of the coil and because of the low critical
current density of wires in that region due to the high field.
A comprehensive treatment of shim coil design may be found
in Refs. 6 and 9. Those references also include details of superconducting coil construction. It should be noted, however,
that some expressions in Ref. 6 contain errors.
Ferromagnetic Shims
Ferromagnetic shimming is occasionally used in high field,
small bore NMR magnets, but its principal use is in MRI
magnets. It is in that application that it will be described.
The principle invoked in this kind of shimming is different
from that of shim coils. The shims now take the form of discrete pieces of ferromagnetic material placed in the bore of
the magnet. Each piece of steel is subjected to an axial magnetizing field at its position sufficient to saturate it. It then
generates a field at a point in space that is a function of the
mass of the shim and its saturation magnetization Bs with
little dependence on its shape. For ease of example, a solid
cylinder of steel will be assumed. The axis of the cylinder is
in line with the field, as shown in Fig. 9. (In Fig. 9 the axis
labeled z is that of the shim, not that of the MRI magnet
itself. In fact, the shim will usually be placed at the inside
surface of the bore of the MRI magnet.)
The field B, caused by the ferromagnetic shim, contains
both axial and radial components. The axial component Bz is
the correcting field required, and it adds arithmetically to the
field of the magnet. The radial component adds vectorially to
the field and produces negligible change in the magnitude of
Superconducting Radial Shims

The radial shims are more complex than those for purely
axial harmonics because the finite value of m requires a 2mfold symmetry in the azimuthal distribution of current arcs,
the polarity of current always reversing between juxtaposed
arcs in one z plane (6,9). For instance, m 2 requires four
arcs, as shown in Fig. 8. However, as for m 0, the set of
current arcs shown in Fig. 8 will generate B(n,m), where n is
2, 4, 6, etc., or 1, 3, 5, etc., depending on even or odd current
symmetry about the z 0 plane. So, the positioning of the
arcs along the z axis is again crucial to the elimination of at
least one unwanted order, n. Fortunately, the azimuthal symmetry generates unique values of the fundamental radial harmonic m. (Eight equal arcs cannot generate an m 2 harmonic.) However, depending on the length of the arc, higher
71
27
x
90
Figure 8. Schematic of a set of radial shim coils for correction of a

B(2,2) harmonic showing the positioning necessary to eliminate
B(4,2) and B(4,4).
312
In order to generate useful harmonic corrections in NMR

magnets for large n and m, electrical shims are located in the
warm bore of the cryostat. Although in older systems those
electrical shims took the form of coils tailored to specific harmonics, modern systems use matrix shims. Essentially, the
matrix shim set consists of a large number of small saddle
coils mounted on the surface of a cylinder. The fields generated by unit current in each of these coils form an influence
matrix, similar to that of a set of steel shims. The influence
matrix may be either the fields produced at a set of points
within the magnet bore, or it may be the set of spherical harmonics produced by appropriate sets of the coils.
Br
Bz
z
Figure 9. Field vectors generated by a ferromagnetic shim in the
bore of an MRI magnet. Bz adds arithmetically to the main field; Br
adds vectorially and so has negligible influence on the field.
the axial field. Therefore, only the axial component of the

shim field must be calculated. If the saturation flux density
of the shim is Bs, the axial shim field is given by
Bz = BsV [(2 tan2 )/(tan2 + 1)5/2 ]/(4z3 )
(14)
where V is the volume of the shim and z and are as shown

in Fig. 9.
The practical application of ferromagnetic shims involves
the measurement of the error fields at a number of points,
and the computation of an influence matrix of the shim fields
at the same points. The required volumes (or masses) of the
shims are then determined by the inversion of a U, W matrix,
where U is the number of field points and W is the number of
shims. In an MRI magnet, the shims are steel washers (or
equivalent) bolted to rails on the inside of the room temperature bore of the cryostat. In the occasional ferromagnetic
shimming of an NMR magnet, the shims are coupons of a
magnetic foil pasted over the surface of a nonmagnetic tube
inserted into the room temperature bore or, if the cryogenic
arrangements allow, onto the thermal shield or helium bore
tube. As in the design of the magnet, linear programming can
be used to optimize the mass and positions of the ferromagnetic shims (e.g., to minimize the mass of material).
Resistive Electrical Shims
The field of an NMR magnet for high resolution spectroscopy
must be shimmed to at least 109 over volumes as large as a
10 mm diameter cylinder of 20 mm length. If, as is usually
the case, substantial inhomogeneity arises from high-order
harmonics (n and m greater than 3), superconducting shims
are of barely sufficient strength. This arises because of the
large radius at which they are located, at least in NMR magnets (e.g., in the regions x and y of Fig. 6). In general, the
magnitude of a harmonic component of field generated by a
current element is given by
Bn r n+1 /rn+1
0
(15)
where r is the radius vector of the field point, and 0 is the

radius vector of the source. Thus, the effectiveness of a remote
source is small for large n.
FIELD MEASUREMENT
The accurate measurement of the spatial distribution of field
in the as-wound magnet is essential to shimming to high homogeneity. Sometimes, measurement of the field is possible
at very low field strengths with tiny currents flowing in the
windings at room temperature. That may allow mechanical
adjustment of the positions of the main compensations coils
(k k in Fig. 6) to reduce the B(1,0), B(2,0) and B(1,1) harmonics. Major field measurement is made with the magnet at
design field strength and in persistent mode. The methods of
measurement in NMR and MRI magnets are generally different.
In NMR magnets, because of the small bore, the field is
measured by a small NMR probe on the surface of a cylindrical region about 8 mm diameter and over a length of up to 10
mm. The measurements are made at typically 20 azimuthal
intervals. From these field measurements, the predominant
harmonics can be deduced, using a least-squares fit, and
shimmed by means of the superconducting shim coils, both
axial and radial. With subsequent measurements, as the harmonic content becomes smaller, the higher harmonics become
evident and in turn can be shimmed. The field measurements
are usually reduced to harmonic values because the shim sets
are designed to generate specific harmonics. The correcting
current required in any particular shim set is then immediately determined. Measurement and shimming is always an
iterative process, generally requiring several iterations to
achieve homogeneities of better than 109 over 5 mm dsv.
Field measurement in an MRI magnet is usually performed differently because much more space is available and
because knowledge of the magnitudes of the harmonics in associated Legendre polynomial form is an advantage in the
shimming process. In this case, the measuring points will lie
on the surface of a sphere. Typically, the diameter of this
sphere may be 500 mm. The field is measured at intervals of
, often 30, around each of the circles of intersection with
this spherical surface of several z const planes, called
Gauss planes. From these measurements and by the property
of orthogonality of the associated Legendre functions, the values of the constants Cn,m and Sn,m in Eq. (10) can be deduced
by the following methodology.
The double integral
+1
1
2
0
P(n, m)(u)[cos(m)]Pi, j (u)[cos( j)] du.d
is nonzero only if i n and j m. Then, for m 0, its value

is
2 (n + m)!/(2n + 1)(n m)!
So, if both sides of Eq. (10) are multiplied by Pi,j(u)[cos( j)]
or Pi,j(u)[sin( j)] and double integrated, the right-hand side
will be nonzero only if i n and j m. Then
+1 2
1
Bz (n, m)Pi, j (u)[cos( j)] du.d
313
sired homogeneity. For example, if the wire diameter is

very large, say, greater than 3 mm, the field will develop a fine structure away from the z axis. If the winding lay of a large diameter wire is helical in each layer,
a helical structure may arise in the amplitude of the
field with consequent problems in the correction of the
resulting high harmonics.
Nonmagnetic coil forms must be used because the presence
of discrete regions of ferro- or strong paramagnetism
will generate large harmonics of high order (large n and
possibly also m), which would be very difficult to shim.
= Cn,m rn (n + m + 1)2 (n + m)!/(2n + 1)(n m)!

and
+1 2
1
Bz (n, m)Pi, j (u)[sin( j)] du.d
= Sn,m rn (n + m + 1)2 (n + m)!/(2n + 1)(n m)!
(16)
Equation (16) are realized in practice by the measurement of

the field at each of 60 points (for example) and the multiplication of each value by the spherical harmonics, Pi,j(u) cos( j)
and Pi,j(u) sin( j) at that point. The integration is numerical.
The method usually employed is Gaussian quadrature, similar in principle to Simpsons rule for numerical integration in
Cartesian coordinates, but in which the z const planes are
the roots of the Legendre polynomial and the weights assigned to the values measured on each of these planes are
derived from the Lagrangian. Tables of the roots and weights
are found in standard texts on numerical analysis (14). For
the purposes of example, assume the number of planes p 5,
and the number of azimuthal points per plane q 12, for a
total of 60 points on the surface. The planes are at z/r 0,
z/r 0.5385, and z/r 0.9062; r is the radius of the
spherical surface. The corresponding weights are 0.5689,
0.4786, and 0.2369. The measurements are made on the circles of intersection and two numerical integrations of Eq. (16)
performed, one for the cos(m) terms and the other for the
sin(m) terms. Then the values of Cn,m and Sn,m are obtained
from
Cn,m =
p q

p
Sn,m

wq B(u p , q )Pn,m (u p ) cos(m )
[(2n + 1)(n m)!]/[2rn (n + m + 1)!]

=
wq B(u p , q )Pn,m (u p ) sin(m )
[(2n + 1)(n m)!]/[2r n (n + m + 1)!]
(17)
(18)
where B(up, q) is the field at the point p, q, the subscripts p

and q denote each of the 60 points, and wq is the Gaussian
weighting for the plane q.
NMR MAGNET DESIGN AND CONSTRUCTION
Practical issues peculiar to the design and construction of
NMR magnets include the following:
The wire diameter must be such that layers of windings
near the inner radius of the solenoids do not generate
discrete field fluctuation of a size comparable to the de-
Voltage
The index of the wire must be high. All high-resolution NMR

requires high field stability, with a decay not exceeding about
108 per hour. To achieve that, the magnets operate in persistent mode. A superconducting switch is closed across the
winding after energization so that the current flows without
loss in a resistanceless circuit. The superconducting switch
consists of a small coil (usually noninductive) of a superconducting wire equipped with a resistance heater. When the
heater is energized, the temperature of the coil is raised above
the critical value, and the coil becomes resisitive. The charging voltage applied to the magnet then causes only a small
current to flow in the switch. When the magnet has been
charged, the switch heater is turned off, the coil cools, and
the magnet current can then flow through the switch without
loss. If a magnet does not need frequent resetting, its rate of
field decay must be small. The joints between wire lengths
and the switch and the magnet must be superconducting, and
the wire must be without resistance. Although the joints can
indeed be made so that their critical currents exceed the operating current, the effective resistance of the wire, owing to
its index, may be high enough that decay in persistent mode
exceeds acceptable levels for NMR. The resistance of the wire,
manifest as a low value of the index, arises from variation in
the critical current along the length of the wire. If a short
region exists where the superconducting filaments are thin or
have low pinning strength, a fraction of the current transfers
between superconducting filaments through the copper (or
bronze) matrix, giving rise to the resistance. The voltage associated with this resistance appears in critical current measurements on samples of the wire.
Figure 10 shows the typical trace of voltage gradient along
a superconducting wire in a fixed field as a function of cur-
Current
ic
Figure 10. Voltage gradient along a composite superconductor as a

function of steady current, showing the effect of index.
314
rent. The defined critical current is that at which a voltage

gradient of, typically, 0.1 V/cm is measured. As the current
is increased beyond the critical value, that voltage gradient
increases. An approximation to the gradient near to the critical current is
v (i/ic )N
(19)
where i/ic is the ratio of actual current to critical current and

N is the index of the wire. The higher the value of N, the
sharper is the superconducting to normal transition. Clearly,
for values of i below the critical value, the voltage gradient
will be small; the larger the index, the smaller the gradient.
So, for NMR magnets, an appropriate combination of index
and the ratio of i/ic must be chosen. The index of most niobiumtitanium (NbTi) wires suitable for MRI and NMR is
typically 50. However, for niobiumtin (Nb3Sn) wires, the index is lower, typically around 30, and the matrix is the more
resistive bronze. So, for high field NMR magnets using
Nb3Sn inner sections, lower ratios of i/ic are necessary. The
concept of the index is only an approximation to the behavior
of voltage as a function of current. In fact, theory and measurement indicate that the effective index increases as i/ic decreases below 1 (15,16). Field decay arising from the index is
constant and is distinguished from that caused by flux creep.
The latter is a transient effect. It dies away with a logarithmic time dependence after a magnet has been set in persistent mode.
Protection of the magnet from the consequences of quenching must be compatible with the electrical and thermal isolation of the magnet from room temperature systems. Quenching is an spreading irreversible transition from the
superconducting to the normal resistive state in the winding.
The energy released during quenching in an NMR or MRI
magnet must be dissipated as heat in the winding. In order
to limit the energy and hence heat dissipated in any part of
the winding, the magnet must be electrically divided into sections, each of which is provided with a shunt, often in the
form of diodes. This subdivision limits the energy that can
be transferred between sections and thereby minimizes the
temperature rise and voltage generated within a section during quenching (17).
NMR Magnet Design
Figure 6 illustrates the winding array of a typical 750 MHz
NMR magnet (18). Table 2 specifies the dimensions and winding specifications of the sections.
At a current of 307.86 A, these windings generate 17.616 T
at the center; that corresponds to 750 MHz proton resonance
frequency. The total inductance is 109.2 H, and the stored
energy is 5.17 MJ. The first nonzero harmonic of the design
is the 12th. The coils s, t, u, v, and w and their mirror images
are the axial shim coils located in the annular space x. The
radial shim coils are located in the space y.
The winding of the Nb3Sn sections of high field NMR magnets presents particular problems. The wire is wound in the
unreacted state after which it must be heated at about 700C
for up to 200 h to transform the separate niobium and tin
components into the superconducting compound. The wire is
insulated with S-glass braid, with a softening temperature of
about 1000C. (An alternative is E-glass. Although the E-
glass may start to soften during the heat treatment, it is

stronger in the prefired state than S-glass and therefore better survives the exigencies of winding.) After the heat treatment the winding is consolidated by impregnation with epoxy resin.
The forms on which the Nb3Sn wire is wound must also
endure the heat treatment without distortion. Stainless steel
is the universal choice for the coil forms although titanium
alloys have been used. The alloy 316 L is generally preferred
because of its very small magnetic susceptibility. If the form
is assembled with welds, those must be made with nonmagnetic filler, if used. The inner bore of the form must be quite
thick if distortion is not to occur. The reason for that lies in
the expansion coefficients of the wire and of stainless steel.
The unreacted Nb3Sn wire consists of bronze, niobium, tin,
tantalum, and copper. During reaction, the copper and bronze
have negligible strength, and the mechanical properties of the
niobium and tantalum dominate. Their coefficients of thermal
expansion are smaller than that of stainless steel with the
consequence that, as the temperature rises during the heat
treatment, the bore of the form will expand faster than the
inner diameter of the winding. If the bore tube is thin, it can
buckle against the constraint of the winding.
The need for thick bore tubes leads to windings of several
wire diameters on one form. The thick bore tube occupies
space that could otherwise be used by field-generating winding. In order to minimize the diluting effect of the bore tube,
large winding builds are used. However, in the high field regions of Nb3Sn windings, the wire diameter must be graded
to optimize the cross section of Nb3Sn corresponding to the
local field. An alternative to the thick stainless steel bore tube
is the transfer of the reacted winding to an aluminum form
before impregnation with epoxy resin. This has been used occasionally, as in the example of Fig. 6 (18).
MRI Magnet Design
The design construction of MRI magnets follows the principles involved in the construction of NMR magnets (19). The
forces and energies are generally greater. For example, the
force tending to center each large end coil of the MRI magnet
illustrated in Fig. 11 is 1,339,000 N (150 tons). NbTi conductor is used exclusively in MRI magnets because, to date, center fields of no more than 5 T are used. The NbTi filaments
in the copper matrix of composite NbTi wires are twisted to
approximate transposition. Because of the low fields in MRI
magnets, wires with few NbTi filaments can be used. Those
filaments can then be arrayed as a single circular layer within
a copper matrix.The filaments are then fully transposed and
are magnetothermally very stable. Mechanical perturbation
is nevertheless a problem, and attention has to be paid to the
interface between a winding and the coil form against which
it presses. Because of the high stored energies, large copper
cross sections are needed in the conductor to avoid over heating during quenching. Currents are typically up to 500 A. A
common form of conductor is a composite wire embedded in a
copper carrier. The latter frequently has a grooved rectangular cross section into which the composite wire is pressed or
soldered. Insulation may be cotton or kapton wrap instead of
enamel, and wax may be used as an impregnant as an alternative to epoxy.
315
Table 2. Dimensions of the Windings of a 700 MHz NMR Magnet

Section
Number
Peak Field
(T)
Wire
Type
Wire
Diameter
(mm)
Inner
Diameter
(mm)
Outer
Diameter
(mm)
Winding
Length
(mm)
Number of
Turns
a
b
c
d
e
f
g
h
j
k
l/l
17.62
16.97
15.93
14.8
13.83
11.80
9.83
8.31
5.70
3.13
3.14
Nb3Sn
Nb3Sn
Nb3Sn
Nb3Sn
Nb3Sn
Nb3Sn
NbTi
NbTi
NbTi
NbTi
NbTi
2.4
2.22
1.84
1.83
1.63
1.61
1.41
1.30
1.14
1.30
1.30
43
52.2
68.8
83.4
99.0
121.7
141.3
158.0
176.0
194.2
194.2
52.2
65.3
79.9
92.5
115.2
134.8
149.5
171.0
186.2
212.5
212.5
600
600
650
650
700
750
800
1000
1000
88.7
377.1
1000
1620
1770
1780
3870
4194
3396
6948
6992
1020
4110 (each)
At a current of 307.86 A these windings generate 17.616 T at the center; that corresponds to 750 MHz proton resonance frequency. The total inductance is 109.2 H
and the stored energy is 5.17 MJ. The first nonzero harmonic of the design is the twelfth.
Most whole-body MRI magnets used in clinical applications have room temperature bores of between 1 and 1.3 m,
with fields up to 2 T. An example of the profile of the windings
of a whole-body MRI magnet is illustrated by the simple fivecoil system of Fig. 11.
The center field is 1.5 T and the dimensions of the windings are listed in Table 3. The compensation is to tenth order
(10 ppm over a 500 mm sphere). The current is 394 A for a 1.5
T center field. The inductance is 78 H and the stored energy 6
MJ.
The fringing field of this magnet extends a long way from
the cryostat in which the coils are housed. The 1 mT (10
gauss) line is at an axial distance of 11.3 m and at a radial
distance of 8.8 m from the center. Access to the space within
these limits must be restricted because of the dangers to the
wearers of pacemakers, the attraction of ferromagnetic objects, and the distortion of video monitors.
This may be an expensive restriction in a crowded hospital.
Therefore, methods of shielding the space from the fringing
fields are frequently used. Three methods are generally available: close iron, remote iron, and active shielding. The use of
iron close to the coils has been used in a few instances. However, the iron must be at room temperature, to avoid otherwise severe cryogenic penalties. That leads to difficulties in
2000 mm
1600 mm
Figure 11. Coil profile of a 1.5 T unshielded MRI magnet illustrating

typical coil placement.
balancing the forces between the coils and the iron in order
to minimize the loads that the cryogenic supports must resist.
Remote iron takes the form of sheet, typically several millimeters thick, placed against the walls of the MRI room. This
involves rather awkward architectural problems but is used
frequently where the restricted space can still extend several
meters from the cryostat.
The third form of shielding is by superconducting coils,
built around the main coils, operating in series with the main
coils as part of the persistent circuit, and in the same cryogenic environment. Those shield coils generate a reversed
field to cancel, or reduce, the external fringing field. Typical
of the resulting magnet is the eight-coil configuration shown
schematically in Fig. 12.
Particular aspects of the illustration follow. The outwardly
directed body forces in unshielded MRI windings are supported in tension in the conductor. However, in the shielded
version, those forces are too large to be supported by the conductor alone, and a shell is applied to the outside of the winding against which the accumulated body forces act. Thus, the
body forces on coils 3 and 4 are supported on their outer surfaces by a structural cylinder. Coil 4 provides the compensation of the dipole moment of the three inner windings so that
the fringing fields of the magnet are much reduced from those
of the unshielded magnet. The reduction in the volume of the
restricted space is about 93%. The magnet is much heavier
(and more expensive) than the simple unshielded type and
the structural design of the cryostat and the suspension system is accordingly stronger. The highest fields generated at
the center of shielded whole-body MRI magnets is 2 T. See
also Ref. 12.
Table 3. Example of a 1.5 T Whole-Body MRI Magnet

Coil
Number
Inner
Radius
(mm)
Outer
Radius
(mm)
Left
End
(mm)
Right
End
(mm)
Number
of
Turns
1
2
3
4
5
741.9
742.2
742.1
742.2
741.9
807.1
785.7
777.0
785.7
807.1
977.0
391.9
74.9
244.6
724.9
724.9
244.6
74.9
391.9
977.0
3030
1180
960
1180
3030
316
CRYOGENICS
As for NMR magnet systems, the economic operation of superconducting MRI magnets demands cryogenic systems with
low heat in-leak. The evolution of MRI cryostats has been significant over the past 15 years. They have changed from simple liquid helium, liquid nitrogen shielded reservoirs with relatively high cryogen evaporation rates to single or multicryocooled cryostats. In one embodiment, no refrigerant is
used in some types of cryocooled MRI magnets; in other examples, a combination of cryocoolers and refrigerants provide a
zero evaporation rate. Demountable current leads are an essential feature of any magnet system with low refrigerant
evaporation rate, and have been a standard feature of MRI
magnet systems since 1974. An implication of demountable
current leads is the need for the MRI magnet to be self-protecting during quenching, just as an NMR magnet must be.
PULSED GRADIENT COILS

In addition to the uniform background field, which it is the
function of the MRI magnet to generate, pulsed gradient
fields must be superimposed on that field in order to create
the spatial encoding of the resonant frequencies of the protons
(or other species) within the body. Those pulsed gradient
fields are generated by three sets of room temperature coils,
each set being driven by a powerful ramped current source.
The pulsed field gradients are linear (dB/dz, dB/dx, dB/dy),
as far as it is possible to design pure first-order gradient coils.
Two problems arise in the overall design as a consequence
of these pulsed gradient fields. The first is the effect on the
superconductor of the periodic incident fields. Although the
thermal shields and coil forms lie between the gradient coils
and the superconductor, the incident pulsed field would still
be significant there. Those small fields would cause a loss
within the conductor through the mechanisms of hysteresis
and coupling.
The second is the distortion of the gradients arising from
currents induced in adjacent structures, such as the thermal
4
1
Figure 13. Schematic of actively shielded pulsed gradient coils for

the dBz /dx gradient showing the spacing between the main and
shield coils.
shields and helium vessel, also sometimes the room temperature bore tube, if that is metallic. This distortion is minimized
by locating sets of shield coils near the room temperature bore
tube. These active shield coils are energized in opposition to
the main pulsed gradient coils. They serve to confine the return flux of the gradient coils to flow in the space between the
main gradient coils and the shield coils. The eddy currents
induced in the surrounding structures are thereby minimized.
The shield coils reduce the efficiency of the pulsed gradient
system, that effect becoming more pronounced as the diameter of the main coils becomes a large fraction of that of the
shield coils. At a diameter ratio greater than about 0.85, the
efficiency is so reduced that the driving power required for
useful gradient fields becomes prohibitively large. Figure 13
illustrates the form of the shielded dB/dx or dB/dy pulse coils.
The dB/dZ coils are simple solenoids surrounded by
shielding solenoids.
BIBLIOGRAPHY
1. E. M. Purcell, H. C. Torrey, and R. V. Pound, Resonance absorption by nuclear magnetic moments in a solid, Phys. Rev., 69: 37,
1946.
2. F. Bloch, W. W. Hansen, and M. Packard, Nuclear induction,
Phys. Rev., 69: 127, 1946.
3. E. Becker, High resolution NMR, Theory and Applications, New
York: Academic Press, 1980.
4. R. R. Ernst, G. Bodenhausen, and A. Wokaun, Principles of Nuclear Magnetic Resonance in One and Two Dimensions, Oxford:
Clarendon Press, 1987.
5. D. B. Montgomery, Solenoid Magnet Design, New York: WileyInterscience, 1969, p. 4.
Figure 12. Schematic of an actively shielded MRI magnet showing

the large coils needed to generate a main field while the shield coils
generate an opposing field.
6. M. D. Sauzade and S. K. Kan, High resolution nuclear magnetic

spectroscopy in high magnetic fields, Adv. Electron. Electron
Phys., 34: 193, 1973.
7. W. R. Smythe, Static and Dynamic Electricity, New York:
McGraw-Hill, 1950, pp. 147148.
MAJORITY LOGIC
8. M. Abramowitz and I. A. Stegun (eds.), Handbook of Mathematical Functions, Washington, DC: US Dept. of Commerce, Natl.
Bur. of Standards, 1964.
9. F. Romeo and D. I. Hoult, Magnetic field profiling: Analysis and
correcting coil design, Magn. Resonance Med., 1: 4465, 1984.
10. M. R. Thompson, R. W. Brown, and V. C. Srivastava, An inverse
approach to the design of MRI main magnetics, IEEE Trans.
Magn., MAG-30: 108112, 1994.
11. W. H. Press et al., Numerical Recipes: The Art of Scientific Computing, Cambridge, UK: Cambridge Univ. Press, 1987.
12. F. J. Davies, R. T. Elliott, and D. G. Hawksworth, A 2 Tesla active shield magnet for whole body imaging and spectroscopy,
13. E. S. Bobrov and W. F. B. Punchard, A general method of design
of axial and radial shim coils for NMR and MRI magnets, IEEE
Trans. Magn., MAG-24: 533536, 1988.
14. P. Davis and P. Rabinowitz, Abscissas and weights for Gaussian
quadrature of high order, J. Res. NBS, (RP2645), AMS (55): 35
37, 1956.
15. Y. Iwasa, Case studies in Superconducting Magnets, New York:
Plenum, 1994, pp. 306307.
16. J. E. C. Williams et al., NMR magnet technology at MIT, IEEE
Trans. Magn., MAG-28: 627630, 1992.
17. B. J. Maddock and G. B. James, Protection and stabilisation of
large superconducting coils, Proc. Inst. Electr. Eng., 115: 543
546, 1968.
18. A. Zhukovsky et al., 750 MHz NMR magnet development, IEEE
Trans. Magn., MAG-28: 644647, 1992.
19. D. G. Hawksworth, Superconducting magnets systems for MRI,
Int. Symp. New Develop. in Appl. Superconductivity, Singapore:
World Scientific, 1989, pp. 731744.
JOHN E. C. WILLIAMS
Massachusetts Institute of
Technology
MAGNETS FOR NMR. See MAGNETS FOR MAGNETIC RESONANCE ANALYSIS AND IMAGING.
MAGNETS, PERMANENT. See PERMANENT MAGNETS.

MAGNETS, SUPERCONDUCTING. See SUPERCONDUCTING CRITICAL CURRENT;
SUPERCONDUCTING MAGNETS FOR

SUPERCONDUCTING MAGNETS, QUENCH PROTECTION.
PARTICLE ACCELERATORS AND STORAGE RINGS;
MAGNETS, SUPERCONDUCTING FOR NUCLEAR

FUSION. See SUPERCONDUCTING MAGNETS FOR FUSION REACTORS.
MAINTENANCE, AIRCRAFT. See AIRCRAFT MAINTENANCE.
MAINTENANCE, JET TRANSPORT AIRCRAFT. See

JET TRANSPORT MAINTENANCE.
MAINTENANCE, SOFTWARE. See SOFTWARE MAINTENANCE.
317
POWER QUALITY
747
POWER QUALITY
Society is becoming more and more reliant on machines, devices, and processes that are sensitive to the quality of the
electrical power they use. At the same time, the generation
and distribution system for electrical power is becoming more
and more complex and vulnerable to internal and external
perturbations that disrupt the quality of power available to
the end user. For these reasons, there is a large market developing for high-quality power, and several technologies that
can provide such power have been or are being developed. In
this article, we discuss some of these power quality issues and
provide some insight into one of the technologies proposed to
provide high-quality power. The first part of the article focuses on power quality issues and how microSMES (Superconducting Magnetic Energy Storage) technology is inserted
into the power flow stream. The second part is devoted to a
discussion of the fine points of microSMES technology from a
design and performance point of view.
POWER QUALITY
The term power quality encompasses a broad range of technical issues that concern everyone, from the casual computer
user to the equipment manufacturer, the industrial plant
manager, and the utility transmission engineer. The technical
issues associated with power quality at all these levels are
complex, and they are often disguised behind misconceptions,
vague definitions, and misapplied solutions. The lack of a consistent vocabulary, difficulty in fully characterizing power
quality at the ultimate point of use, and confusion over the
applicability of available solutions often encourages an adversarial relationship between the user, the equipment manufacturer, and the utility provider.
The technical community as well as the general public is
increasingly concerned about power quality factors for a number of interrelated reasons. Some of the issues behind this
growing apprehension include:
Increasing levels of automation and flexibility have
caused a proliferation of microprocessor-based controls
and power electronic systems that are more sensitive to
power quality variations than older electromechanical
devices.
Utility operating procedures designed to minimize sustained interruptions have increased the frequency of
short-duration interruptions through wider use of line
reclosers and instantaneous breaker operations.
A growing emphasis on energy conservation has encouraged greater use of high-efficiency adjustable-speed motor drives and nonlinear power supplies that generate high
levels of harmonics and heating of neutral conductors.
The increasing sophistication of industrial processes and
their reliance on expensive raw materials have greatly
748
POWER QUALITY
multiplied the cost of recovering from disruptions caused

by poor power quality.
These factors combine to place a tremendous economic burden
on industry, particularly in high-tech operations such as
semiconductor wafer fabrication. Estimates of this economic
impact on US industry alone range from $13.3 billion to $25.6
billion per year (1).
Power Quality Standards
The definition of power quality terms and standard methods
for measuring and characterizing electromagnetic phenomena
have received considerable attention in recent years from numerous standards bodies. The principal bodies involved in establishing international power quality standards are the Institute of Electrical and Electronics Engineers (IEEE) and the
International Electrotechnical Commission (IEC). Other
North American standards organizations involved with power
quality issues include the American National Standards Institute (ANSI), National Electrical Manufacturers Association
(NEMA), Underwriters Laboratory (UL), Canadian Standards
Association (CSA), and Semiconductor Equipment and Mate-
rials International (SEMI). European organizations involved

in similar activities include the European Union Standards
Organization (CENELEC) and the International Conference
on Large High Voltage Electric Systems (CIGRE).
The electromagnetic disturbances defined in the standards
and recommended practices produced by these organizations
can be classified in several ways, depending on the perspective of the defining body. Table 1 is an example of a particularly comprehensive classification system that not only defines a generally accepted term for each type of disturbance
but also gives typical values for spectral content, duration,
and magnitude where appropriate (2).
Power Quality Data
The most frequently referenced method of representing power
quality data and equipment compatibility information is the
CBEMA curve, originally developed by the Computer Business Equipment Manufacturers Association. This organization, now known as the Information Technology Industry
Council (ITIC), approved a revised version of the CBEMA
curve in 1996. The revised curve, shown in Fig. 1, specifies an
envelope of acceptable power quality defined in terms of volt-
Table 1. Classification of Power Quality Disturbances
Categories
1.0 Transients
1.1 Impulsive
1.1.1 Nanosecond
1.1.2 Microsecond
1.1.3 Millisecond
1.2 Oscillatory
1.2.1 Low frequency
1.2.2 Medium frequency
1.2.3 High frequency
2.0 Short-duration variations
2.1 Instantaneous
2.1.1 Interruption
2.1.2 Sag (dip)
2.1.3 Swell
2.2 Momentary
2.2.1 Interruption
2.2.2 Sag (dip)
2.2.3 Swell
2.3 Temporary
2.3.1 Interruption
2.3.2 Sag (dip)
2.3.3 Swell
3.0 Long-duration variations
3.1 Interruption, sustained
3.2 Undervoltages
3.3 Overvoltages
4.0 Voltage unbalance
5.0 Waveform distortion
5.1 dc offset
5.2 Harmonics
5.3 Interharmonics
5.4 Notching
5.5 Noise
6.0 Voltage Fluctuations
7.0 Power frequency variations
Typical
Spectral
Content
Typical
Duration
5 ns rise
1 s rise
0.1 ms rise
50 ns
50 ns1 ms
1 ms
5 kHz
5500 kHz
0.55 MHz
0.350 ms
20 s
5 s
0.530 cycles
0.530 cycles
0.530 cycles
0100th harmonic
06 kHz
Broadband
25 Hz
Typical
Voltage
Magnitude
04 pu
08 pu
04 pu
0.1 pu
0.10.9 pu
1.11.8 pu
30 cycles3 s
30 cycles3 s
30 cycles3 s
0.1 pu
0.10.9 pu
1.11.4 pu
3 s1 min
3 s1 min
3 s1 min
0.1 pu
0.10.9 pu
1.11.2 pu
1 min
1 min
1 min
Steady state
0.0 pu
0.80.9 pu
1.11.2 pu
0.52%
Steady state
Steady state
Steady state
Steady state
Steady state
Intermittent
10 s
00.1%
020%
02%
01%
0.17%
POWER QUALITY
749
Percent of normal voltage (rms or peak equivalent)
500
400
300
Applicable to 120, 120/208, and
120/240 nominal voltages
200
140
120
110
100
90
80
70
Voltage-tolerance
envelope
40
0
1 s
0.001 c
0.01 c
0.1 c
1 ms
0.5 c
3 ms
1c
20 ms
10 c
100 c
1000 c
0.5 s
10 s
Duration of disturbance in cycles (c) and seconds (s)
110%
100%
Percent of nominal voltage
age magnitude and duration. The curve assumes that equipment will ride through disturbances falling within the envelope without any malfunctions. It is assumed that
disturbances that fall below the envelope may cause the load
equipment to drop out because of a lack of energy, whereas
disturbances that fall above the envelope may lead to overvoltage trips, insulation breakdown, and other problems. Unfortunately, the actual voltage tolerance of electronic equipment
varies quite widely and rarely approximates the CBEMA
curve. Nevertheless, the CBEMA curve has become a de facto
reference for evaluating the quality of the electrical supply
and the tolerance of sensitive equipment to voltage disturbances.
Figure 2 shows a plot of voltage sag data collected from
15 semiconductor manufacturing sites overlaid on the lower
portion of the CBEMA envelope. The disturbance data are a
subset of the data collected as part of the EPRI Distribution
Power Quality (DPQ) study (3). The DPQ study collected data
on a representative sample of utility distribution feeders,
with monitors placed on 100 different feeders at 24 geographically dispersed utilities over a period of approximately 2
years. The data shown in Fig. 2 was restricted to data acquired at semiconductor plants to characterize the electrical
environment faced by large industrial customers served by
dedicated substations. These sites generally have fewer disturbances than the average substation serving a mix of com-
Steady
state
Figure 1. Revised voltage tolerance envelope developed by CBEMA to characterize compatibility at the userss load in
terms of the minimum voltage magnitude
and duration of the disturbance. This de
facto standard was intended to provide a
benchmark for measuring the ridethrough capability of sensitive electronic
equipment against the quality of the
available electric supply.
90%
80%
70%
60%
50%
40%
30%
CBEMA curve
20%
10%
0%
10
100
Duration (cycles)
1000
Figure 2. Scatterplot of EPRI DPQ study data overlayed on the relevant portion of the revised CBEMA curve. The 1076 disturbance
events were acquired at 15 semiconductor manufacturing plants over
a period of approximately 2 years. The number of disturbances that
fell on or below the CBEMA curve was 166, giving an average of 5.4
events per year per site outside the voltage tolerance envelope.
750
POWER QUALITY
mercial and residential loads. Even so, the semiconductor

plants experienced 1076 disturbances, with 166 of the events
lying below the CBEMA curve. This resulted in an average of
5.4 sags and interruptions per year at each site that fell outside the CBEMA voltage tolerance envelope.
The DPQ disturbance data is shown in contour format in
Fig. 3 to display the frequency of occurrence for sags and interruptions as a function of their magnitude and duration values (4). This visualization method clearly indicates that the
vast majority of the disturbances are relatively small amplitude sags of short duration. Statistical analysis of the DPQ
data reveals:
The average interruption rate was 0.5 per month per
site;
The average sag rate was 4 per month per site (10% V
90%);
The ratio of voltage sags to interruptions was approximately 10 : 1;
Most voltage sags had a duration of less than 167 ms (10
cycles at 60 Hz).
From this data, we can conclude that power conditioning
equipment providing ride-through protection times of at least
1 s can protect sensitive industrial loads against the vast majority of the electrical disturbances they will encounter.
Superconducting Magnetic Energy Storage Configurations
The acronym SMES was coined in the 1970s at the University
of Wisconsin. It stands for superconducting magnetic energy
storage, a concept whereby energy is stored in the form of a
magnetic field. The field is created by a superconducting magnet, which has virtually no electrical resistance when cooled
to the boiling point of helium. The lack of electrical resistance
means that there are no storage losses; so in principle SMES
is a very efficient means of storing energy.
The basic energy storage element of a microSMES system

can be connected to the utility and the protected load in a
number of ways to maximize protection capabilities for different applications. The operating requirements for the magnet
subsystem are fairly similar for these various configurations,
whereas the power electronics interface is quite specific to the
target application. Several potential application classes will
be briefly discussed in the following sections.
Shunt-Connected System. A simplified block diagram of a
shunt-connected microSMES application is shown in Fig. 4.
This configuration can be used to protect virtually any type of
sensitive ac load or industrial process. The load is initially fed
from the utility source through the normally closed mechanical disconnect switches on either side of the solid-state isolation switch. When a voltage sag or interruption occurs on the
utility source, the solid-state isolation switch is opened to isolate the load from the unreliable utility power. The microSMES unit instantly begins to discharge through the inverter so that the load is seamlessly transferred from the
utility source to the energy storage system. If utility power is
restored during the discharge, the microSMES unit is synchronized to the grid, and the solid-state isolation switch
transfers the load back to the utility source. The microSMES
unit is then recharged to full-energy storage capacity in preparation for additional protection cycles.
Series-Connected System. A series-connected microSMES
unit is shown in the simplified block diagram of Fig. 5. In this
application, the microSMES unit is configured to compensate
for voltage sags by providing a voltage boost in series with
the load. Because the microSMES unit does not support the
full load as it does in the shunt-connected configuration, the
same amount of energy storage can support a larger electrical
load for an equivalent time or the same size load for a longer
time. The load is normally fed from the utility source through
the solid-state isolation switch bypassing the injection trans-
Compatibility (site events)
Nominal voltage base 480

90
2530 events
1520 events
Figure 3. Contour plot of EPRI DPQ

study data. The contour lines represent
the average number of disturbance events
per site that are more severe (having a
lower voltage and longer duration) than
the magnitude and duration values along
the contour line. For example, there were
15 events per year with a magnitude less
than 60% of nominal voltage and a duration longer than 20 ms.
Percent of nominal voltage (%)
80
70
1015 events
60
05 events per
site per year
50
40
30
510 events per

site per year
20
10
10
20
40
60
80
100
200
300
400
Duration (ms)
500
600
700
800
POWER QUALITY
Maintenance
bypass
switch
Utility
source
Disconnect
switch
Disconnect
switch
Solid-state
isolation
switch
ac
Converter
dc
V
Converter
Sensitive
load
for
industrial
process
Inverter
Chopper
Figure 4. Simplified block diagram of a

shunt-connected microSMES system. This
configuration protects sensitive ac loads
against dips and interruptions by disconnecting from the utility service during a
disturbance and supplying the load with
energy stored in the superconducting
magnet.
Superconducting
magnet
Support
system
former. When a voltage sag is detected, the solid-state switch

opens, and the load current flows through the injection transformer. The inverter then injects just enough energy into the
transformer to maintain the desired voltage level at the load.
The isolation switch closes when nominal utility voltage is
restored and the microSMES unit is recharged and placed in
standby mode in preparation for the next event.
Maintenance
bypass switch
Utility
source
751
Motor Drive System. In adjustable speed motor drive applications, the microSMES unit supplies a dc voltage rather
than the ac voltage supplied in shunt- and series-connected
applications. As shown in the block diagram of Fig. 6, the dc
voltage from the microSMES unit is fed into the dc link between the rectifier and inverter of the motor drive itself. This
configuration eliminates the need for a separate inverter for
Sensitive
load
for
industrial
process
Solid-state
isolation switch
Disconnect
switch
Disconnect
switch
Injection transformer
ac
Converter
dc
Inverter
V
Converter
Chopper
I
Support
systems
Superconducting
magnet

series-connected microSMES system. This
configuration protects sensitive ac loads
against voltage dips by boosting the voltage across the injection transformer to
compensate for low voltages at the utility
source.
752
POWER QUALITY
Adjustable speed drive
ac
Utility
source
dc
Converter
Converter
dc
ac
Critical
motor
for
industrial
process
V
Converter
Voltage regulator
I

microSMES system protecting an adjustable speed motor drive. This configuration
eliminates the inverter in the microSMES
system because the stored energy in the
magnet is supplied directly to the dc link
of the motor drive.
Superconducting
magnet
Support
systems
the MicroSMES system. The voltage regulator in the microSMES system simply ensures that the dc link capacitor
remains fully charged in the event of a loss of utility power
feeding the rectifier.
croSMES unit to support the load completely during short

disturbances and to slow down the rise time of the battery
discharge for longer disturbances. This hybrid configuration
can significantly enhance battery performance because battery lifetime is decreased by deep discharges with fast rise
times.
Hybrid UPS System. A second configuration requiring dc

output from the microSMES is shown in Fig. 7, where the
microSMES unit is integrated with a conventional battery
Uninterruptible Power Supply (UPS) system. The microSMES voltage regulator monitors the dc link of the UPS
and supplies energy from the magnet whenever the link voltage drops because of a loss of power at the input rectifier. If
the loss of input power persists, the batteries gradually begin
to supply energy to the inverter as the stored energy in the
microSMES is depleted. This configuration allows the mi-
MicroSMES
The remainder of this article concentrates on the system that
provides the energy to ride through a power disturbance. In
particular, energy stored in a magnetic field is discussed.
Magnetic energy storage systems are described in general and
compared with other forms of storage. Details of the design
Battery UPS
ac
Utility
source
dc
Converter
Converter
dc
ac

hybrid microSMES/battery UPS system.
In this configuration, the superconducting
magnet supplies energy to the load during
short-duration disturbances, and the batteries must be discharged only during
sustained outages. The fewer number of
cycles and lower discharge depth required
from the batteries significantly increases
their lifetime.
Converter
Voltage regulator
I
Support
systems
Superconducting
magnet
Critical
motor
for
industrial
process
POWER QUALITY
Unconditioned
utility power
Raw
power
Optional
fast-start
stand-by
generator
Power
electronics
Continuous
conditioned power to
customer load
Clean
power
Converter
753
I
Power
leads
Controls
and remote
monitoring
Cryocooler/
recondenser
Superconducting
toroid
Helium vessel
Thermal shield
Vacuum vessel
are discussed, with the purpose of identifying the design drivers that affect the economy of such systems. Finally, a generalized algorithm to arrive at a design that minimizes the cost
required to store a unit of energy is developed to further illuminate the design issues related to microSMES.
What is SMES?
Designs for SMES systems have been proposed at several
sizes and stored energy scales. Early SMES studies focused
on very large systems, which stored thousands of megawatthours of energy and were designed to provide peaking power
for large utility baseloads. SMES systems storing tenths of
megawatt-hours have been proposed to provide stabilization
of power distribution grids subject to transients that can disrupt the delivery of power over transmission lines. Some applications in the military and elsewhere require hundreds of
megawatt-seconds of energy delivered at very high power that
can be supplied by SMES systems. Finally, SMES systems
storing on the order of 1 MW s to 10 MW s of energy are
proposed to provide short-term high-quality power to equipment and processes sensitive to voltage sags and/or power
outages. The latter set of systems as a group is referred to as
microSMES, or SMES.
A typical block diagram of a SMES system is shown in
Fig. 8. There is a power electronics subsystem that provides
an interface between the energy storage system (the magnet)
and the power source/load. Depending on the application,
there might be an auxiliary stand-by power source included
in the system to handle long power outages. Ancillary subsystems support the magnet. It operates at liquid helium temperature and requires refrigeration and insulating vacuum
systems to maintain operating conditions. The heart of the
system is a superconducting magnet. This is an electromagnet
wound with superconducting wire. When the wire is operated
at liquid helium temperature (4.5 K, or 450F), it will carry
an electric current with no resistance. Thus there are no resistive losses, and the storage of energy in the form of a magnetic field becomes possible.
The economics of SMES are determined by these three
major subsystems. Unlike battery storage, where all the cost
Figure 8. Block diagram of a microSMES

storage system, showing major subsystems. Each subsystem is a major cost
driver for SMES technology and must provide optimum performance.
is in the power electronics, or capacitors, where all the cost is

in the energy storage system, SMES costs are split between
the power electronics and the storage, with nontrivial standby power needed to run the refrigeration system. Each of
these subsystems must be optimized for maximum performance at minimum cost for SMES to be successful in the
marketplace.
Because the power electronics are somewhat common to all
types of energy storage, the magnet system and refrigeration
system must be optimized for SMES to be competitive with
other technologies. For the magnet system, the maximum
stored energy must be obtained using the minimum amount
of material (superconducting wire). For the refrigeration system, minimizing the heat leaking into the cold space is the
key to optimum performance. These issues will be discussed
in greater detail later.
How Does SMES Compare to Other Forms of Energy Storage?
Before describing a SMES system in greater detail, some
points about the technology should be made, with the purpose
of driving out those applications where SMES makes sense
and where it doesnt make sense. A discussion of the desirable
and undesirable qualities of SMES follows.
Positive Attributes. SMES technology has many virtues. It
can provide power at an arbitrary level and do so virtually
instantaneously. Furthermore, the life of the device is not
limited by the power level or depth of discharge. A SMES system will provide power pulses throughout its design lifetime
of up to 30 years with no degradation in performance.
The conversion from magnetic energy to electrical energy
is very efficient. Because the magnetic field is created by an
electric current circulating in a magnet, it is very straightforward to extract the energy. All that is needed is a voltage
applied across the terminals of the magnet. The only loss is
in the form of heat dissipated in the leads and electronic devices in the power electronics system (PCS), which is
minimal.
A large fraction of the stored energy can be extracted from
the system. The only limitation is that the voltage must be
754
POWER QUALITY
applied to maintain the delivered power. As will be discussed

in further detail later, the voltage must rise as the current
drops during a discharge. Eventually the voltage will rise to
a level where it can break down the insulation on the coil.
The discharge must be stopped just before this happens. With
proper design, it is possible that this limit is not reached until
only a few percent of the original stored energy remains.
The storage system itself requires no maintenance. The
only moving parts are in the refrigerator. Maintenance is required on the refrigeration system, but it is possible that it
can be performed while the remainder of the system is operating. Thus SMES has the potential to have no down time
for maintenance.
SMES is generally environmentally benign (see the discussion of fringe magnetic fields). It does not employ hazardous
or toxic materials. There are no components requiring recycling or disposal during or at the end of its life.
Finally, the state of charge of the system is easily determined. A simple current or magnetic field measurement is all
that is needed to ascertain that the system is charged and
ready to deliver power.
Negative Attributes. SMES has some undesirable characteristics as well. Some are real, and some are falsely perceived by the market, but nonetheless have prevented SMES
from being an overwhelming success to date.
First, SMES is an expensive technology. The superconducting wire used in the magnets is expensive to begin with.
The cost of the labor and materials to fabricate the structure
of the magnet, the helium containment vessel, the cryostat,
and the supporting refrigeration system is and will always be
an order of magnitude greater than the cost of batteries storing the same amount of energy.
The technology is esoteric and is perceived to be risky. Superconducting magnets are esoteric for several reasons. Because they operate at such cold temperatures, vacuum insulating technology and helium refrigeration technology are a
necessary part of the design. Furthermore, it takes a trivial
amount of energy to destroy the superconductive properties
of the wire (e.g., the frictional heat generated if two adjacent
wires rub together through the distance of 4/100ths of a mil,
or the heat generated by the creation of microcracks in an
epoxy-impregnated coil when it is energized). If the superconductive property is lost, systems must be in place to protect
the wire from overheating and burning out from the excessive
resistive heat that is generated. In addition, there can be very
high internal voltages in a superconducting magnet that has
gone resistive, which can potentially arc through the electrical insulation around the wire and render the magnet useless
at worst, or have much reduced capacity at best. These problems are usually encountered when a magnet is first energized, but rarely afterwards. A magnet might require several
attempts before it can be fully ramped to its design current
and field, but once it has been operated at design conditions,
the risk of an unwanted quench virtually vanishes. The risk
inherent in the technology is to the producer because if a
magnet is not properly designed and fabricated, it becomes
apparent only after it has been built. Then it must be
scrapped. The risk to an end user is negligible once a magnet
has been operated at design conditions. Over the years, the
potential problems have come to be understood, and although
they cannot be eliminated, they can be mitigated by paying
close attention to proper design practice and quality control

during fabrication. Indeed, superconducting magnets have
reached a mature state of reliability for commercial applications: refer to magnetic resonance imaging (MRI) systems
routinely operating in hospitals throughout the world.
It is possible that a SMES system will have fairly large
fringe fields (i.e., the magnetic field will extend well beyond
the physical limits of the magnet system). This can lead to
siting problems or restrictions that can add to the cost. An
alternative is to use toroidal magnet systems, where the field
is confined totally within the bore of the toroid. This essentially eliminates the fringe field, with the penalty of requiring
more superconducting wire to produce the same amount of
energy as a solenoid.
The maintenance required for the refrigeration system can
be expensive. Depending on the size of the refrigerator required, maintenance can vary from requiring a simple seal
replacement to rebuilding a compressor periodically. The cost
of this maintenance must be factored into the life cycle cost of
the installed system. Typical practice is to provide redundant
components such as compressors, with an associated penalty
of additional capital cost.
Finally, SMES suffers from a penalty imposed by the Second Law of Thermodynamics. In order to operate at temperatures close to absolute zero, large amounts of power are required to operate the refrigerator. Typical helium
refrigerators require from 500 W to 5000 W of compressor
power to remove a single watt of heat load from a helium
bath. Thus there are large parasitic power requirements to
operate a system. In addition, the current that circulates
through the magnet must also circulate through the power
electonics system, creating waste heat that must be removed.
This waste heat becomes an air conditioning load that represents additional parasitic power.
Because SMES will always be more expensive than batteries, it should be considered only for applications where batteries cannot compete: power levels higher than 2 MW or so
and situations requiring many repetitive discharge cycles
lasting on the order of two or three seconds or less. These
conditions exploit the positive attributes of SMES to best advantage.
What Does the Storage System Look Like?
The storage subsystem of a SMES unit consists of a magnet
and a cryogenic system. The magnet is a superconducting coil
or array of coils. The cryogenic system includes a cryostat
that insulates the cold helium vessel from the ambient surroundings, and a helium liquefier.
The magnet may be a single coil or an array of coils. It
may be wound in a solenoid configuration or in a toroid. The
advantage of a solenoid is that it produces more stored energy
per unit conductor than a toroid does, and it is much simpler
to fabricate. The advantage of a toroid is that the magnetic
field is confined almost completely within the bore of the
torus.
It is not within the scope of this article to discuss the art
of designing and fabricating superconducting magnets. However, insofar as they relate to the technology of SMES, the
main components of a magnet will be discussed. A superconducting magnet consists of a conductor, wound on some struc-
POWER QUALITY
ture, with the conductor isolated from adjacent turns and the
supporting structure by some insulation system.
The conductor represents the major materials cost of a
SMES system. It will be either a single monolithic wire, or
a cable of smaller wires. At currents up to 2000 A to 3000 A,
a single wire will do the job and is easier to fabricate and
wind. At higher currents (higher powers), a cable is the preferred approach. The primary reason is that, as the size of
wire increases, the ac losses during discharge become significant. These losses are reduced in a cable because of the resistance created at the many contact points between individual
wires.
The structure must support the winding against the Lorentz loads created when a conductor is placed in a magnetic
field. In a simple solenoid, the structure is trivial because the
net force on the coil is zero (excepting its own weight). The
forces act to expand the coil radially outward and compress
the coil axially. The winding itself is the structure. In solenoidal arrays, or toroids, there are significant forces that act on
the individual coils, which must be reacted with structure.
The nature of the forces depend on the geometry and arrangement of the coil array. For a toroid, a centering force acts to
compress the toroid into as small a diameter as possible.
The insulation system is key to the economic viability of
SMES because it determines the maximum allowable voltage at which the coil can be discharged. The voltage level determines how much of the stored energy can actually be delivered at a given power level, and the higher the voltage, the
lower the current can be, which reduces the refrigeration
load. The largest voltages appear across the terminals of the
coil, and also from coil to ground. The insulation in these
areas must be designed to withstand high voltage stresses
for many cycles. The voltages between layers of the coil, and
between turns, are generally much less because the terminal
voltage gets divided essentially evenly over each layer and
turn.
The cryogenic system maintains the operating environment for the superconducting coil, which is at the normal boiling point of liquid helium, or approximately 4.5 K. The components are the helium refrigerator and the cryostat. A key part
of the cryostat that affects the economics of SMES are the
leads that connect the cold magnet to the warm power conditioning system.
The complexity, size, and cost of the refrigerator depends
on the size of the heat load to the refrigerated space. Larger
heat loads, on the order of tens of watts, require a refrigerator
based on the Collins cycle. These are commercially available
and consist of a compressor and a cold box. The compressor
produces a high-pressure stream of gas, which is passed
through the cold box. Inside the cold box are counterflow heat
exchangers, which cool the high pressure gas with cold, lowpressure gas returning from the cold space. At several temperature stages, part of the gas stream is diverted through an
expansion engine to produce additional cold cooling gas. Only
a portion of the gas that is compressed actually is used to
refrigerate the magnet. One of the problems with helium refrigerators is that any other contaminants in the gas stream
condense out on the heat exchanger surfaces. This includes
the oil that must be used to lubricate the compressor. One of
the reasons helium refrigerators are so expensive and hard to
maintain is that the gas cleaning systems require periodic
cleaning.
755
At smaller heat loads, newer cryocooler technology is preferred. These are small refrigerators based on the Gifford
McMahon cycle, which employs a regenerator heat exchanger
to cool the high-pressure gas stream. These devices are also
subject to degradation resulting from fouled heat transfer surfaces, but they have the advantage of not requiring cold
expansion engines that can wear out, and the regenerator
tends to be self-cleaning to some degree. Cryocoolers still require periodic maintenance, but the system can be designed
to accommodate a changeout of the cold head without having
to shut down the magnet. The disadvantage of cryocoolers is
that they are not as efficient as the Collins machines, which
is why they are not used for large heat loads.
The cryostat provides a vacuum enclosure for the helium
vessel and supports the vessel with low conductivity supports.
A thermal shield is placed between the vacuum vessel and
helium vessel to intercept heat radiated from the warm surface. The thermal shield is cooled either with liquid nitrogen
or by conduction to a cryocooler. Multilayer insulation is used
to further reduce the heat leak. As long as the vacuum is
maintained at a pressure of 102 Pa or lower, the heat leak
into the helium vessel can be made very low, with proper design practice.
The current leads to the coil will always be the largest single source of heat into the magnet. The connections, or power
leads, cannot be made from materials with low thermal conductivity because they would then have a high electrical resistance. More electrical resistance means more resistive heating, which then ends up in the cold magnet system. The
fundamental tradeoff for power leads is the balancing of the
heat conducted down the lead against the heat generated by
resistance within the lead. Whatever one does to make one
component smaller increases the other by roughly the same
amount. All is not lost, however; there are three things that
can be done to minimize the heat input.
The first is simply to reduce the current the leads have
to carry. The required refrigeration power is directly proportional to the amount of current in the leads. A good
magnet design, then, will use the least amount of current
possible. The drawbacks are threefold. First, less current
in the conductor means more turns of conductor are required to provide the same stored energy. Depending on
how the conductor is priced, this can mean additional
capital cost for the magnet system. (Note that the total
volume of conductor stays about the same, so if the cost
of conductor is dominated by bulk material cost, the cost
is about the same regardless of current; if the conductor
cost is primarily the result of processing, longer lengths
of smaller wire will cost more.) The other two drawbacks
are risk related. More turns of conductor make it harder
to protect the magnet from overheating in the event of a
quench. Also, as the current is decreased, it requires
more voltage across the coil terminals to extract the
same amount of power. As the voltage increases, so does
the risk of damaging the coil insulation as a result of
voltage breakdown.
The second way to reduce the heat leak is to employ a
high-temperature superconductor for a portion of the
leads. The ceramic materials are an exception to the general rule that thermal conductivity goes up as electrical
756
POWER QUALITY
resistivity goes down. Although still a developing technology, high-temperature superconductor (HTS) leads
have been built and have demonstrated a fourfold reduction in heat leak over conventional leads. They are complex and lack design maturity at the present time but
offer a promising path to reduced heat leak.
The final way to reduce the heat load caused by the
power leads is to short the coil leads with what is called
a persistent switch. This is a length of superconducting
wire stabilized with coppernickel alloy instead of pure
copper, so it is more resistive when normal. The switch is
opened by driving it normal. Allowing it to cool to helium
temperature makes it superconducting. When superconducting, the current circulates entirely within the helium
vessel. The leads only carry current during a current
pulse when the energy is extracted from the SMES device. As a consequence, the leads can be rated for transient operation only, resulting in a much smaller crosssection and a reduced heat leak.
What Drives the Design?
The design of a SMES storage system is fundamentally
driven by the need to minimize the cost necessary to provide
a specified amount of power for a specified span of time. The
cost includes the cost of materials (primarily superconductor),
the cost of labor to manufacture the magnet system, and the
life-cycle operating cost of the system (the cost of electricity
to run the refrigerator dominates). In this section, the design
issues and tradeoffs are discussed, and a generic algorithm
for producing an optimized design is outlined.
The ultimate design drivers for a SMES device are the
power P it is required to deliver and the time duration tp the
power is needed. It is then basic to SMES that the energy
stored in the magnet system be at least equal to the product
of the required power and the time duration. Other considerations force the total stored energy to exceed this amount.
There are constraints to the design as well. The fact that
the voltage must rise as energy is extracted from the magnet
has been mentioned. The maximum allowable voltage is one
constraint on how much energy can be extracted.
Another constraint to the efficiency of the design is the fact
that heat is produced in the windings when the coil is discharged. This is the result of what are commonly referred to
as ac losses. They arise when a magnetic field changes with
time in a conducting medium. If the ac losses heat the conductor to the point that superconductivity is lost, some of the
stored energy is lost to heat dissipated in the windings and
cannot be extracted as electrical power.
Another constraint is the stress level in the coil pack. The
factors that make a coil efficient in terms of cost are the current density in the conductor, the field that the coil produces,
and the size of the coil. Making any of these three larger to
improve cost efficiency also increases the stress level in the
conductor. Thus there is an upper limit to the economics resulting from the strength of the materials.
The behavior of superconducting material in a magnetic
field also places a limit on the performance of SMES. There
is a limit to the amount of current a conductor can carry at
any given field. This is called the critical current. The critical
current density in the niobiumtitanium superconducting
alloy is limited to about 3 109 A/m2 at 5 T and drops to zero

at about 9.5 T.
Finally, although one would like to make the current in
the magnet as large as possible so that the discharge voltage
is small, the heat leak associated with the current leads gets
larger as the current increases. The economics of refrigeration
places an upper limit on the value of current.
What Are the Design Tradeoffs?
The constraints on the design lead to tradeoffs between various cost and risk elements. Quantification of these tradeoffs
is the first step toward an algorithm that can be used to produce an optimized design. The ultimate goal of a SMES design is to provide the necessary power for the necessary time
at minimum cost. The cost will include materials, fabrication
labor, and life cycle support and maintenance costs. The primary material cost will be cost of the superconductor. The
main labor cost will be to wind the coils and assemble the
structure and vessels. The largest life cycle cost will be the
cost of electricity to run the helium refrigerator and cooling
for the power electronics.
Excess Storage (Cost) vs. Current. There is a tradeoff between
the cost of excess storage in the coil and the cost of refrigeration to remove the heat load resulting from the current leads.
The system cost can be represented in a simple fashion by the
expression
C = a Uo + b Io
(1)
The coefficient a represents the material and labor cost per

unit megajoule of stored energy to build the unit, and b represents the cost of electricity to refrigerate the heat load due to
the current leads.
As energy is extracted from the magnet, the voltage must
always rise to maintain constant power as the current in the
magnet drops. At some point before all the energy is extracted, the voltage reaches the maximum allowable value
Vm, and the discharge must stop. If P is the power extracted
from the unit, V is the terminal voltage, and I is the current
in the coil, then
P = IV
(2)
The energy U stored in the magnet at any time is

U=
1 2
LI
2
(3)
where L is the inductance.

During a discharge, the energy stored in the magnet is reduced from the original amount Uo by the product of power
and time Pt. Thus
U = Uo + Pt
(4)
Note that for a discharge, the sign of P is negative ( a negative

voltage must be applied to the coil to extract energy). If tf is
the duration the power is needed, then the ratio of the energy
extracted to the energy initially stored in the SMES unit can
POWER QUALITY
be expressed as

2
Ptf
P
=1
Uo
Vm Io
(5)
Equation (5) was developed by noting that at any time the

ratio of U to Uo is equal to the square of the ratio of I to Io
and that at the end of the discharge, when the current is a
minimum, the voltage is at its maximum value (in a negative
sense). Note that, for a given allowable maximum voltage,
power level, and time duration, as the initial current increases, the required initial stored energy decreases. Depending on the relative values of the cost coefficients a and
b, there will be a minimum in the total cost that should be
sought.
Excess Storage vs. Power (ac Losses). The discharge can also
be limited by ac losses. If the losses cause the magnet to
quench before the end of the discharge time tf , the energy
remaining in the magnet is lost to heat deposited in the windings. The factors that determine this limit can be determined
as follows.
is proportional to the square of the
The ac loss heating Q
field decay rate B. For the purposes of this article, this can be
written as
Q = aB 2
(6)
Now, the field is proportional to the current. If Bp is the operating field at the operating current Io, then it can be shown
that the total heat Q that is deposited in the conductor during
a discharge is equal to
tf
Q=a
0
B 2 dt = a

B2p P
Uo
ln
4 Uo
Uo + Ptf
(7)
At some value of Q, the magnet will quench. This equation

defines the parameters that can reduce Q. Operating at a
lower field will reduce Q but will mean the magnet will be
larger (stored energy is proportional to the square of the field
integrated over all space). A larger magnet in turn will most
likely mean higher costs. Increasing the stored energy also
reduces Q with more cost. The tradeoff here is to determine
the minimum size and stored energy which allows a total energy Ptf to be extracted before the magnet quenches (if it does
at all).
How Is the Design Optimized?. Given all these issues related to cost and performance, how does one go about finding
the optimum design for a SMES unit? The ideal approach
would be to set down the end requirements P and tf , derive
the required initial stored energy Uo, apply all the constraints, and end with the geometry of the coil defined. Unfortunately, the relationships between field, current density, inductance, and coil geometry do not lend themselves to closed
form solutions, which can be manipulated to provide such an
approach. Instead, an iterative process must be followed,
whereby a trial geometry is selected first and then the resulting SMES system is checked to see that it satisfies all
requirements and constraints. If not, the process is iterated
until it does. Then cost is determined as a function of the
757
geometry, and the geometry is varied again until a minimum

cost is attained. It is worthwhile to outline in a little more
detail the steps involved because it helps to illuminate a little
better the design issues that affect the cost performance of
SMES.
This algorithm finds the minimum wire cost per unit
stored energy for a SMES device. The variables that are free
to adjust are the descriptors of the coil geometry: the length,
the coil pack thickness, and the radius of the coil. It is common to use the nomenclature shown in Fig. 9, where ai is the
inner radius of the coil pack, is the ratio of outer to inner
radius, and is the ratio of coil pack length to diameter. The
optimized design will satisfy the constraints that
the hoop stress in the conductor will be equal to or less
than some allowable value w,
the absolute value of voltage at the end of a discharge of
duration tf and power level P will be equal to or less than
an allowable value Vm,
the critical current density in the superconductor will be
equal to or less than the critical current density Jc at the
point where the field is highest in the coil pack, and
the coil will not quench as a result of ac losses before the
energy extraction is complete.
The first step is to pick trial values for and . (It is assumed
that a solenoid design is being pursued. In the case of a toroid, there will be some relationship between the length of
the coil and the major radius of the torus. Therefore, the major radius becomes a free parameter instead of .)
The next step is to determine the resulting magnetic field,
with an assumed value of Jo ai. It doesnt matter what the
value is at this stage, just something convenient. One of the
reasons for computing the field is to compute the stress level
in the conductor subsequently. Exactly what detail is needed
for the field distribution is a function of the detail needed for
the stress calculation. In every case, however, the field at a
particular point can be expressed as
B = o Jo ai F (, )
(8)
The value of F depends on where it is evaluated, but the point

is that B anywhere is proportional to the overall coil pack
current density Jo and the coil radius ai.
Coil winding
ao
ai
= ao / ai
= (l/2) ai
Coil centerline
Figure 9. Identification of nomenclature used to describe magnet geometry.
758
POWER QUALITY
The stress is proportional to the product of radius, field,

and current density. It can always be written as
= o (Jo ai )2 S(, )
(9)
The function S will depend on the level of complexity in the

stress analysis. For thin solenoids with an approximately linear field profile through the winding, Wilson provides a closed
form expression (5). Given a stress allowable w, the product
of current density and radius is then determined:
Jo ai =
o w
S(, )
(10)
Now the field at any point is known. In particular, the peak

field in the coil pack is now determined, using Eq. (8).
The next step is to find the value of current density that
minimizes the amount of superconductor needed. This value
is the critical current density of the superconductor at the
peak field. In any coil winding, there is space needed for material other than superconductor, and prudent design calls for
some temperature margin for the conductor. If is the ratio
of nonsuperconductor to superconductor in the wire, is the
fraction of the coil pack actually occupied by wire, and Tm is
the temperature margin of the superconductor, then
Jo =
Jc (Bp , To + Tm )
1+
(11)
Once Jo is known, ai is known.

Given a complete description of the geometry and the current density, the stored energy can be found. Usually the
value of (L/n2) is determined, either from tables, or from a
computer code. In general, it will be proportional to ai and a
function of the shape of the coil; thus,
(L/n2 ) = ai G(, )
(12)
Once (L/n2) is known, the stored energy is found from

1 L
1
U=
(Jo A)2 = ai G(, )[Jo 2a2i ( 1)]2
2
2 n
2
to minimize the amount of conductor needed. The current is

set to the minimum value allowed by the voltage limit and
the available stored energy (for a given configuration) to minimize the heat load.
SUMMARY
We have seen how the economics of SMES are affected by
the power conditioning system, the storage system, and the
parasitic power required to maintain operating conditions.
There are certain power quality application that benefit most
from the advantages of SMES. These are high-power situations with many repetitive cycles lasting a few seconds or less.
The primary design driver for the magnet system is producing
the maximum amount of stored energy for minimum cost,
subject to constraints that available materials impose. A systematic algorithm to produce a cost-optimized design was developed. It maximizes the efficiency of the superconductor,
while satisfying the constraints of stress, ac loss heating,
maximum voltage, and allowable current density in the superconductor.
As the market for high-quality power develops and the relative benefits of SMES as an option become better understood, it is expected that magnetic energy storage will make
a major impact on the economics of power quality sensitive
processes.
BIBLIOGRAPHY
1. C. DeWinkel and J. D. Lamoree, Storing power for critical loads,
IEEE Spectrum, 30 (6): 3842, 1993.
2. R. C. Dugan, M. F. McGranaghan, and H. W. Beaty, Electrical
Power Systems Quality, New York: McGraw-Hill, 1996.
3. T. E. Grebe, An Assessment of Distribution System Power Quality,
EPRI Projects RP 3098-1 and 3797-03, April 1995.
4. Draft Standard, Electric Power System Compatibility with Electronic Process Equipment, IEEE P1346 Working Group.
5. M. N. Wilson, Superconducting Magnets, Oxford, UK: Clarendon
Press, 1983, pp. 4147.
(13)
Note that U is proportional to the fifth power of the coil

radius.
At this point, it must be determined whether (a) the energy
stored is sufficient to provide the required power for the required time and (b) the ac losses limit the energy extraction
or not. If not, a new trial geometry must be selected and the
process begun again. If both requirements are met, the algorithm proceeds. The next step is to determine the current,
from Eq. (5). Then the cost of wire is determined. Usually,
conductor can be estimated at some fixed rate per kiloamp
meter. Thus it is straightforward to scale cost data from a
single point.
Once the cost as a function of stored energy is established,
it is straighforward, at least in principle, to apply some optimization routine to the algorithm to find the optimum geometry that minimizes cost. The advantage of this algorithm is
that, to the extent possible, it automatically finds the conditions of maximum stress and current density for a given field
SCOTT D. PECK
JOHN C. ZEIGLER
Houston Advanced Research Center
POWER QUALITY. See POWER SYSTEM HARMONIC

CONTROL.
POWER, REACTIVE. See REACTIVE POWER.

POWER SEMICONDUCTOR CIRCUITS. See POWER
DEVICES.
POWER SEMICONDUCTOR DEVICES. See POWER DEVICES.
POWER SEMICONDUCTOR SWITCHES. See POWER

DEVICES.
POWER SPECTRAL DENSITY. See SPECTRAL ANALYSIS.

POWER SPLITTERS. See POWER DIVIDERS.
POWER STATION DESIGN, NUCLEAR. See NUCLEAR
POWER STATION DESIGN.
NETWORK DESIGN: ALGORITHMS AND EXAMPLES
INTRODUCTION
An interconnection network connects various sources of information using a set of point-to-point links. A link is a
connection using a copper wire or an optical ber, or it may
be wireless. The nodes are autonomous data sources and
can request to transfer any amount of information to any
other node. Figure 1 shows an example network consisting
of four nodes. Node A has a link connected to node B and
a link connected to node C. Node B is connected to nodes
A and D. Nodes C and D are connected to nodes A and B,
respectively. If node C desires to send some information to
node B, it sends it to node A, which in turn routes it to node
B. Node A thus acts as an intermediate node. The capacity of a node is the amount of information it can transmit
(also called source capacity) or receive (also called sink capacity). The capacity of a link is the amount of information
that can be transferred over the link in one unit of time.
The network design deals with the interconnection of
various nodes and how to transmit information from one
node to another. Network architecture and design both
have multiple meanings. The most commonly used interpretation relates to the decisions one needs to make to design a network. The four most important aspects of network
architecture and design are described below.
Network Topology
A topology denes how nodes are interconnected. For example, the topology of the NSF network is shown in Fig.
2. Most network topologies are hierarchical in nature. The
design involves developing the structure of the hierarchy,
structures of nodes at each level, and detailed designs of the
nodes. It also involves assigning link and node capacities
to transport the desired trafc. A hierarchical topology is
depicted in Fig. 3. We will be studying the decisioin-making
process and related algorithms and examples in detail in
this artice.
A network node is placed in a hierarchical fashion in
such a way that it is close enough to several data sources.
The closeness is described in terms of suitable performance
metrics such as physical distance and cost of connection. A
network node serves as a service point for all data sources
Figure 1. A four-node network.
Figure 2. Topology of the NSF network.
Figure 3. A hierarchical network.
connected to it. Such a node is called a gateway as it connects data sources to network nodes. Nodes A, B, C, are such
nodes. Gateways connect to data sources, such as nodes
1, 2, .... 12 at the next lower level in the hierarchy and to
routers or switches such as nodes X, Y, etc. at the next
higher level. Switches and routers route information to
other switches and routers on way to other gateways from
where the data are delivered to destinations. There may be
more levels in the hierarchy.
Node and link placement, and their capacities, in a network topology depend on the desired or required trafc ow
that is dened by the trafc characteristics. This problem
is well-studied, and more information can be found in References (17). In principle, ideal locations for both network
nodes and links may be specied using algorithms (possibly
complex) that would optimize network design using performance metrics of interest. In practice, these placements are
also governed by factors such as existing network, ease of
operation, and convenience of management, which are not
always easy to accommodate in the design process.
Transmission Technology
Physical layer transmission technologies describe the characteristics of physical medium. These technologies involve
signal processing techniques, modulation and demodulation techniques, coding and decoding of information, multiplexing and demultiplexing techniques employed to enhance the utilization of each link, and issues related to
these techniques. Physical medium can be a wire, such
as copper link, coaxial link, optical ber, or wireless link
using microwave or radio frequencies. The signals being
transmitted over the physical medium can be modulated
and demodulated using amplitude modulation, frequency
modulation, or phase modulation where the information
being transmitted modies the shape of the waveform being transmitted. Figure 4 demonstrates examples of modulation techniques. Multiplexing techniques such as time
division, frequency division, and code division multiplexing
techniques are used to mix and transmit information from
various sources on a single link. In time division multiplexing, each source is given a fraction of time in a given interval, called a frame. In frequency and code division multiplexing, frequencies and bandwidth available on a channel are effectively partitioned so that all sources can use
parts of the capacity of the channel simultaneously. Figure
5 shows different multiplexing techniques.
Network Design: Algorithms and Examples
Figure 4. Different modulation techniques.
Figure 5. Time and frequency multiplexing.
Trafc Control and Network Management Techniques

This aspect of network architecture involves the control
of the switching technology, ow control algorithms for
smooth ow of offered loads to the network, control messages ow to set up paths, connection requests and response protocols, collection of data on actual usage, fault
detection and management algorithms, and effective resource utilization algorithms.
Connection requests between nodes in a network are
realized by employing a routing algorithm. Routing algorithms are designed to use existing network capacity and
switching methodology. Either circuit switching or packet
switching can be employed for data transfer in a communication network.
In a circuit-switched network, a complete, dedicated
path from a source to a destination is established through
the network before communication begins. Dedicated physical resources are reserved for the communication to take
place. A signal is sent from the source to the destination
node, through intermediate nodes, requesting a connection.
An acknowledgment signal is sent back from the destination to the source accepting or rejecting the request. A request is accepted only if all nodes on the path have required
resources available and they can reserve these resources
statically. If no free path exists from end to end, the trafc
is blocked and has to wait for transmission. Path establishment may take substantial time. Once the path is established, information is transmitted freely from the source
node to the destination node using the path. The sender
and receiver may use any format for data transfer and
bit rate subject to the constraints of the physical channel.
When the transfer is completed, the path is removed.

In a packet-switched transmission, no physical path is
established in advance. Instead, when a source has information to transmit, it assembles it in a packet. A packet
consists of data to be sent and a header. The header contains the source and destination addresses, possibly intermediate node addresses through which the packet must
be routed, and some error correcting codes to check the
correctness of the information. The packets are forwarded
from node to node, one hop at a time. The packets are
queued in buffers at the intermediate nodes along a route
between a source and a destination, traveling from node to
node, releasing links, and switching elements immediately
after using them. A packet is received by a node, checked
for correctness, A and retransmitted, if required. During
the transmission, store and forward operation at each node
increases the overhead and time delay of data packets.
Most networks use packet switching for smaller messages. In a packet-switched network, it is possible that the
path may change from packet to packet between the same
source-destination pair. Circuit switching is used when a
source/destination has a substantial amount of information to transmit. The main advantage of circuit switching
is that the links on a path are always available for communication. The only delay is the propagation delay. However,
if no information transfer takes place for sometime during
the exislence of a connection, the capacity on the path is
wasted. To reduce this waste, it is possible to statistically
multiplex the path. A path is established in advance between a source and a destination that is used by every
packet from that source to that destination. However, actual transmission occurs as in a packet-switched network.
Such a path is called a virtual path, as opposed to a physical path that is Established in a circuit-switched network.
Such a service, in which a path is established in advance,
also called a connection-oriented service. A pure packetswitched service is called a connectionless service. To set up
a path, the network control sends messages to nodes on the
path to request a connection. If all nodes on the path from a
source node to a destination node agree, then the path is established. Flow control algorithms are employed to control
the actual ow of information so that each node on the path
is not overwhelmed with the information. If a fault occurs
in a node on the path, then the path is established again. In
case of a connectionless service, data packets are just sent
to the next node along the path. If a faulty node is encountered on the path, the routing is changed on-the-y. The
Internet datagram service uses a pure packet switching
protocol, whereas the telephone system, for the most part,
uses circuit switching. Networks employing asynchronous
transfer mode (ATM) (that are currently being developed)
use the concept of virtual path for ATM cell transmission.
An ATM cell is a small, 53-byte-long packet that includes
48 bytes of actual data and 5 bytes of control information.
In this case, a path is established that is used by all cells,
but actual transmission is cell by cell from point to point
similar to a packet-switched network.
Cost
The cost of the network is viewed differently by different
people. The cost includes parameters such as installation of
links and nodes, including the cost of the facility to house
the nodes and lay the links (copper or ber). Laying out
links is very expensive and includes buying/leasing land,
digging land, laying out conduits, cost of cables, wires, or
optical bers, end-interfaces, buffering, processing hardware at each end of a link, and management of links. Additional operational and maintenance costs are needed to
keep the hardware up and running, replacing faulty components and cables, and managing the resources.
For a network service provider, the cost consists of laying out and operating the network. On the other hand, the
users or consumers of network resources do not concern
themselves with these costs. The costs they account for are
the cost quoted to them by the network providers in terms
of tariffs for different quality of service (QoS) at different
times. These tariffs are usage sensitive and depend on the
volume of data being transported, time of day (morning,
evening, or night), priority of transmission, tolerable delay and loss of data, and several such factors. These factors together are called QoS parameters. To provide and
guarantee a specic quality of service, the network service provider has to dedicate some network resources such
as bandwidth on individual links, buffer spaces at various
nodes on the path, time slots for transmission of specic
data, and alternative resources in case of a failure for that
service. The cost of these resources forms the basis for tariffs. The development of a cost model for a link is a difcult
problem. Often good approximations and simplication of
cost structures are used by the network service providers to
keep the complexity of the network design and service tariffs under control. In our designs, we consider both models
(actual physical network cost model and consumer network
cost model) of cost in our designs.
APPROACHES TO NETWORK DESIGN
If a new network is being designed from scratch with no
existing capacity, well-dened trafc requirements (trafc
intensities), and full freedom in selecting network components, then the designers can make the best possible decisions by balancing the cost and QoS requirements, such as
throughput, delay, and other performance measures. However, more often than not, most real designs are incremental; that is, the resources are added or upgraded over the
existing capabilities as required by the new demands. The
network really evolves with the needs and, in general, is
in response to the new requirements. This restriction restricts the optimality in design as the existing design governs the nal output.
Inputs for network designs are based on the best estimates of the anticipated trafc between various sources
and destinations. Such data are available in the form of
a trafc matrix. Many networks are designed using current and additional anticipated needs and certain rules of
thumb in an incremental fashion. The decisions are based
on the experience of the designer. It is possible to make
serious mistakes as part of a new design. For example,
when the information transmission is from point to point

as in packet switching, intermediate nodes store and forward the incoming information. By not providing enough
buffer space or control for incoming trafc streams, losses
may be excessive and/or delays may exceed the acceptable
limits. A loss may or may not be tolerable. For example,
in a voice communicatioin, a small loss may almost go unnoticed, but loss of even a byte may not be tolerable in a
computer le transfer application. For a voice or real-time
video communication, any signicant delay may mean that
the information is no more relevant at the destination.
The design process could be manual or automated using
exact or heuristic-based algorithms. An automatic design
process can avoid such serious design aws in the network.
Unfortunately, most known properties and optimization
techniques, relate to networks that are designed new and
not incrementally. Heuristic algorithms are used as part of
the Automated design process to incorporate design principles used in manual algorithms. One of the most used
heuristicis is the use of a greedy algorithm.
Sometimes a greedy algorithm may nd an optimal solution. A greedy algorithm selects a feature that seems to
be of immediate benet. Consider a situation in which several nodes communicate with each other. Providing a direct
link between two nodes that have a maximum amount of
trafc owing between them is a greedy approach. This
may have other effects later on in the algorithm. Similarly,
incorporating and using the cheapest link in a network is
also a good design practice. However, this may have serious
cost implications at a later stage in the design and a greedy
algorithm may fail to account for them. A greedy algorithm
may not always yield the best result, but nonetheless it is
the most used heuristic algorithm.
To fully understand the network design process and algorithms necessary in network design, we rst develop a
graph model of the network. Graph models capture the exact behavior of a network and simplify the task of analysis.
GRAPH MODEL FOR NETWORK
A network is represented by a graph G = (V, E),where V
is a nite set of elements called nodes or vertice and E is
a set of unordered pairs of nodes called edges or arcs (8).
This graph is undirected. A directed graph is also dened
similarly except that the arcs or edges are ordered pairs.
For both directed and undirected graphs, an arc or an edge
from a node i to a node j is represented using the notation
(i, j). Examples of ve-node directed and undirected graphs
are shown in Fig. 6. In an undirected graph, an edge (i,j)
can carry data trafc in both directions (i.e., from node i
to node j and from node j to node i) Whereas in a directed
graph, the trafc is only carried from node i to node j.
Graph Representations
A graph is stored as either an adjacency matrix or an incidence matrix as shown in Fig. 7. For a graph with N nodes,
an N N 0-1 matrix stores the link information in the
adjacency matrix. The element (i, j) is a 1 if node i has a
link to node j. An incidence matrix, on the other hand, is an
N M matrix where M is the number of links numbered
Figure 8. An example of an acyclic graph.

Figure 6. A directed and an undirected graph.
Figure 7. Matrix representation for graph of Fig. 6(a).
from 0 to M 1. The element (i, j) stores the information

whether link j is incident on node i or not. Thus, the incidence matrix carries information about exactly what links
are incident on a node. If a graph has more than one link
from a node to another node, the incidence matrix will be
able to carry this information exactly, whereas the adjacency matrix will require additional information to store
the number of links.
The following terms associated with a graph are used
throughout this article:
1. The degree of a node is the number of links incident
on a node. In the case of a directed graph, we count
both the number of incoming links, or in-degree, and
the number of outgoing links, or out-degree, of a node.
For example, in Fig. 6a, node 1 has a degree of 3,
whereas in Fig. 6b, node 1 has an in-degree of one
and out-degree of two.
2. A walk in a graph G = (V,E) is a sequence
of nodes w = [v1 ,v2 , . . .vk ], k > 1, such that
( j, j + 1) E, j = 1, 2, . . . k 1 . A walk is closed if
k > 1 and v1 = vk .
3. A walk without any repeated nodes in it is called a
path.
4. A closed walk without any repeated intermediate
nodes is called a cycle. An acyclic network does not
contain any cycles as shown in Fig. 8.
5. A node s is said to be connected to node t if node s has
a path to node t in the graph. This path is called an
(s, t) path.
6. The length of a path is the number of links on the
path.
7. An (s, t) path is called the shortest path if there is no
other path of length shorter than the length of the
given path.
Figure 9. A graph and its connected components.
8. (i,j) denotes the length of a shortest path between

nodes i and j. In a network, it is a measure of the
maximum communication delay.
9. The diameter (the longest shortest path between
any pair of nodes) of a graph is given by Max
{(i, j) i, j V }.
10. A graph is said to be connected if a path exists between any of nodes, s and t.
11. A graph is said to be strongly connected if i, j, V
a path exists, from node i to node j.
12. A connected component of a graph (V,E) is a subgraph G = (V , E), V V and E E with every
(i, j) E, i, j V such that G is strongly connected.
An example of connected components of a graph is
shown in Fig. 9 where nodes 1, 2, and 3 form one connected component and nodes 4, 5, and 6 form another.
Node 7 is a component by itself.
13. The node connectivity of a graph is the minimum
number of nodes that should be removed from the
graph in order to partition it into two disjoint subgraphs, that is, the number of node-disjoint paths.
The node connectivity is a measure of the reliability
of communication.
GRAPH ALGORITHMS
Once we select the graph model of a network, various algorithms can be used to efciently design and analyze a
network architecture. Some of the most fundamental al-
gorithms among them are nding trees in a graph with

minimum cost (where cost is dened appropriately) or nding a minimum spanning tree; visiting nodes of a tree in
a specic order; nding connected components of a graph;
nding shortest paths from a node to another node, from
a node to all nodes, and from all nodes to all nodes in a
distributed or centralized fashion; and assigning ows on
various links for a given trafc matrix.
In the following discussion, we describe some useful
graph algorithms that are important in network design.
Recall that N represents the number of nodes and M represents number of links in the graph.
u1 = 0. Let node k be the last node on the shortest path

from node 1 to node j. Then we can say that u j = uk + ak j ,
which also implies that the path from node 1 to node k
with path length Uk must also be the shortest path from
node 1 to node k. Otherwise, the path we selected is not the
shortest path. This Concept is from the principle of optimality. Now, we only have a nite number of choices for k.
Bellmans equations use this principle concept to search for
shorter paths to other nodes by using the known shortest
path to node k and edge weights of direct links from node
k to other nodes for all such ks. The equations state that
Shortest Path Routing
and
Shortest path routing, as the name suggests, nds a path

of the shortest length in the network from a source to a
destination (913). This path may be computed statically
for the given graph regardless of resources being used (or
assuming that all resources are available to set up that
path). In that case, if at a given moment all resources on
that path are in use, then the request to set a path between
the given pair is blocked. On the other hand, the path may
be computed for the graph of available resources. This reduced graph will be obtained after removing all links and
nodes that may be busy at the time of computing from the
original graph. In either case, computation of the shortest
path is based on the following concepts.
Suppose for a given graph G, each arc (i, j) is also assigned a weight (or length) denoted by ai,j . We are interested in nding out a path of the shortest length from a
given source node to a given destination node. The path
will not have any node repeated. This type of problem is
fundamental in graphs, and in particular networks, as we
may be interested in searching for a path from a source to
a destination that is least expensive in terms of traversal.
The weight or length of an arc may represent, the actual
cost of traveling on that edge. The cost may be in terms of
delay, dollars, or any other metric of importance.
One key ideas in computing the shortest path is that
of dynamic programming. It is also based on the principle of optimality. For the shortest path computation, it has
been shown that, if all edge weights are positive, then an
undirected graph can be treated as a directed graph by replacing each undirected (i, j) by two directed edges (i, j) and
(j, i). With negative edge weights, this transformation introduces cycles of negative weights and the shortest path
may go through the cycle as often as necessary to bring the
total path lengths to zero or negative. Thus, it is not desirable. In the following, we will assume that all edge weights
are positive. With that assumption, the shortest path can
be computed using the following formulation.
Bellmans Equations. To compute the shortest path from
source s to destination t, it turns out that we end up computing the shortest path from the source node s to all destinations (9). Let aij be the weight of edge (i, j) if the edge exists. Otherwise, it is . Let uj be the weight of the shortest
path from origin s to node j. For simplicity we assume that
the nodes are numbered from 1 to n and the source node is
node 1. We can always renumber the nodes. It is clear that
u1 = 0
uj = min{uk + ak j }
k = j
j = 2, 3, , N
Using these equations, we can nd a shortest path to a

node as follows. First, nd a node k with edge (k,j) such that
u j = uk + ak j . Then nd an arc (l,k) such that uk = ul + alk ,
and continue in this fashion. Eventually, we would reach
node 1. Unfortunately, Bellmans equations do not lead to
a solution directly.
Shortest Path in Acyclic Network. In an acyclic network,
as shown in Fig. 8, it is easy to use Bellmans equations
to nd a shortest path. The nodes in such a network can
be renumbered in such a fashion that an edge (i, j) exists
if and only if i < j. In this case we can rewrite Bellmans
equations as
u1 = 0
and
uj = min{uk + ak j }
k< j
j = 2, 3, , N
These equations can then be solved as u1 is known, u2 only

depends on ul , u3 only depends on u1 and u2 , and so on. The
complexity of this problem is O(N2 ).
Dijkstra Method. For cyclic graphs, we need another
method given by Dijkstra (11). This method is applicable
to a graph for which edge weights are positive. This algorithm starts with labeling nodes in stages. At each stage of
computation, some labels are designated permanent and
others remain tentative. A permanent label on a node represents the true length of the shortest path from that node.
After including the new labeled nodes, distances to all other
nodes are computed again.
Let dij denote the distance from node i to node j. Let i
be the source node. Then dii is set to zero and dij , i = j is
set to a large value if j is not a neighbor of i. Otherwise,
it is set equal to the weight of the direct link aij . Next,
the algorithm nds a node j with minimum d and labels
it permanent. It then uses it to improve distances to other
nodes by computing
dik min(dik , di j + a jk )
At each stage in the process, the value of dik represents the
best known shortest distance from i to k. Using these labels
of the nodes, the algorithm then marks another unlabeled
Figure 11. Two parts of a network connected by only two links.
from node i to node j. Also,

uii = 0
and
u0i j = i = j
Figure 10. Example of execution of Dijkstra shortest path algorithm.
and
= min(um
um+1
ij
ik + ak j )
k
node with the mimimum value of dik as permanently labeled. The same computation is carried out again. As all
edge weights are positive, in the next iteration, none of the
marked nodes can have any smaller value.
An example of the execution of the algorithm is shown
in Fig. 10. Node A is the source node. A dark node is a permanently labeled node. At each step, one node is marked
labeled and the value associated with a node is its shortest distance from the source thus far with L being a large
value. The algorithm terminates in N 1 steps.
The last equation computes the shortest path lengths for

the paths that contain up to m + 1 edges given that we
know the shortest path lengths for paths that contain up to
m edges. This calculation seems to be equivalent to matrix
multiplication C = A B, where element cij is computed
using
ci j =
aik bk j
We modify the computation of cij as

ci j = min{aik + bk j }
k
Shortest Paths Between All Pairs of Nodes

Now, suppose we want to compute the shortest path between all pairs of nodes. This process may be necessary as
communication may occur between any pairs of nodes. It
is desirable to use the shortest path as this reduces the
requirements for network resources. Sometimes using the
shortest path may cause congestion as has been shown by
many researchers. For example, suppose that the network
graph is such that it can be partitioned into two parts, A
and B, and the two parts are only connected by two links,
one with a low-weight link and the other with a highweight link as shown in Fig. 11. All communication between the two halves will use the low-weight link and the
other link remains unused. The second link should not have
been included in the design, but if it exists, then its use will
reduce the congestion on the low-weight link. The shortest
path routing algorithm does not use the seconolink at all.
Coming back to the all-to-all communication problem,
we can compute paths from every node to every other node.
Thus, we need to solve the problem N times. Alternatively,
we may use an integrated procedure developed separately,
which may be more advantageous. We investigate the latter approach next. Let uij denote the length of the shortest
path from node i to node j, and let uij m be the shortest path
such that the path contains no more than m edges. It is
clear that uij N will be Uij , the length or the shortest path
by replacing multiplication by addition and summation by

minimum function. We know that A = (ai j ) is the matrix
of arc lengths, and we let U0 be the identity matrix; then
U 0 A = A. Let U m+1 = U m A. Then U N1 = AN1 gives
us the desired shorten path length matrix. It turns out
that this type of matrix multiplication is also associative.
Thus, we can compute A2k = Ak Ak , and once 2k > n 1,
we have U N1 . A single matrix multiplication has O(N3 )
complexity, and we need to perform log N matrix multiplications. Therefore, the overall complexity is O(N3 log N).
This process is more complex than Dijkstras algorithm but
in practice may run faster. An example computation for the
ve-node graph in Fig. 6a for a given edge weight matrix
A is given next:
100 40
100
0
A=
40
30
20
20
30
30
20
20 30
0
100
A2 =
40
30
70
100
0
50
130
20
40
50
0
20
30
30
130
20
0
50
70
20
30
50
0
90
0
50
70
20
90
A4 =
40
30
70
40
50
0
20
30
30
70
20
0
50
70
20
30
50
0
FloydWarshall method
Another method to compute the shortest paths between all
node pairs is from Floyd and Warshall with a computational complexity of O(N3 ). In this method, um
i j denes the
length of the shortest path from node i to j such that it
does not pass through nodes numbered greater than m
1 except nodes i and j. Then
u1i j = ai j
and
m
m
= min{um
um+1
ij
i j , uim + um j }
uN+1
is the shortest path length matrix. Also, um+1
= 0 for
ij
ij
all i and for all m.
This procedure has N(N 1)(N 2) equations, each of
which can be solved by using N(N 1)(N 2) the additions
and N(N 1)(N 2) comparisons. This order of complexity
is the same as that for Bellmans method (also known as
the BellmanFord method as it was independently discovered by two researchers), which yields the shortest path
only from a single origin. The Dijkstra method can also
be applied N times, once from each source node, to compute the same shortest path length matrix. This process
takes only N(N 1)/2 additions for each pass, for a total of
N 2 (N 1)/2 additions, but again housekeeping functions
in Dijkstras method make it noncompetitive.
The computation in the FloydWarshall method proceeds with u1 = A and Um+1 is obtained from Um by using row m and column m in Um to revise the remaining
elements. That is, uij is compared with uim + um j and is replaced if the latter is smaller. Thus, the computation can
be performed in place and is demonstrated in the following
for the graph in Fig. 6a.
0
0
100 40 30
100 40
100 0
20
100 0
140
A0 =
40 0 20 30 A = 40 140 0
30
20 0
30
130 20
20
30 0
20
30
0
100
A3 =
40
30
70
100
0
140
130
20
40
140
0
20
30
30
130
20
0
50
70
20
30
50
0
0
100
A4 =
40
30
70
100
0
140
130
20
Multiple Shortest Paths

Many times it is useful to be able to compute additional
shortest paths between a node pair, which may be longer
than the rst shortest path but still short in case the rst
shortest path is not available. The rst path may be congested or may have a failed link or a node. The problem can
be constrained by specic requirements such as allowing
30
130
20
0
40
140
0
20
30
20
30
30
130
20
0
50
0
100
A2 =
40
30
120
70
20
30
50
0
0
90
A5 =
40
30
70
100
0
140
130
20
40
140
0
20
30
90
0
50
70
20
40
50
0
20
30
30
130
20
0
30
70
20
0
50
120
20
30
70
20
30
50
0
or not allowing repeated nodes and links or specic nodes

and/or links. Specic methods exist to compute alternative
shortest paths for all cases (see reference 14). One specic
case with respect to fault tolerance is nonavailability of a
node or a link. Such a path can be computed by removing
the specic node or link in the original graph (removal of a
node also removes all associated links) and then using the
same shortest path algorithm. In another scenario, we may
want another path that is mutually exclusive of the rst
path. In that case, all nodes and links have to be removed
from the original graph before computing another shortest
path. The algorithm to be used in these cases is the same
as already stated.
Minimum Spanning Tree (MST)
The minimum spanning tree is the best tree one can identify in a given graph with edge weights. Recall that edge
weights represent some cost of communicating on that
edge. The cost may be delay or expense in terms of real
dollars to use the link.
The MST problem is to nd a set of edges with a total
minimum cost so that the nodes in the graph remain connected. A greedy algorithm can be used to nd this set of
edges, called MSTE. The algorithm starts with one edge
with minimum weight. Then it nds an edge e, the best
candidate that has not yet been considered and adds it if it
is feasible. An edge can only be added to this set if it does
not create a cycle in the graph with the same set of nodes
as the original graph and set of edges MSTE. MSTE is complete when it contains N 1 edges in an N node graph. It
is known that a greedy algorithm indeed nds an MSTE.
Several algorithms are available to nd an MST. We
will consider two algorithms here based on the greedy approach, but their complexities may differ slightly.
Kruskal Algorithm. This algorithm essentially requires
all edges to be sorted, shortest rst. Then the edges are
included in set MSTE, one at a time, in an order such that
the edges do not form a cycle. The test for forming a cycle can be efciently made by maintaining a proper data
structure of edges included thus far. The complexity of sorting is O(M log M)), the test is of complexity O(M + N) as
suggested by Tarjan (12). As the process terminates once
the set MSTE includes N 1 edges, one may not have to
sort all edges (the rst few may be sufcient). This result
be achieved by putting all edge weights in a heap that can
be created in O(M) time. An edge with the smallest weight
can be removed from the heap in O(log M) time. If k edges
have to be considered to select N 1 edges for inclusion in
MSTE, then the complexity of the selection process is O(M
+ klog M). Therefore, the total complexity is O(M + N + k
log M). An example of execution of Kruskals algorithm is
shown in Fig. 12. Each edge is labeled with its weight and
its number (shown in brackets). In each pass, the selected
edge and the included nodes are shown in the table.
Prims Algorithm. For a dense network, when M is of
O(N2 ), an alternative method to nd an MST is from Prim
(13). This algorithm maintains a tree and adds additional
nodes to the tree using minimum cost edges. For this pur-
Figure 12. Kruskals MST algorithm.
pose, the minimum distance of each node that is out of the

tree is maintained from the tree nodes. Each time a new
node is added, the distances of nodes that are not yet in the
tree from the tree change. Therefore, these distances need
to be revised. In fact, the distances of the nodes outside the
partial tree from the newly inserted node only need to be
considered as that is the only change in the tree. The algorithm has a complexity of O(N2 ). We need N passes, one
each to select N nodes to be included in the tree. Each time
we need to nd a node with minimum distance (this is an
O(N) procedure) and update distances of all other nodes
after considering the new node (another O(N) procedure if
the distances are maintained in the adjacency list). Both
O(N) procedures can be performed in O(d) if the maximum
degree of each node is only d because we only need to consider d neighbors of the new node introduced in the tree.
Thus, the overall complexity of the procedure is O(dN).
Constrained MST. The MST computation may be constrained using some optimality criteria or requirements.
In the case of constrained MST computation, the Selection
of edges is constrained using appropriate selection criteria
consistent with the specied constraints. For example, in
the previous algorithm, it is assumed that the weight of an
edge is the only criterion. But the new constraint may be
that no node can have more than a certain number of edges
connected to it. In that case, the algorithm may have to decide on a selectable candidate differently. If a node already
has a given number of edges originating from it, then no
more edges connected to that node may become part of the
solution.
Tree Traversal
For a given tree graph, one may like to visit all nodes of the
tree. Recall that a tree graph has no loops and the number
of edges is exactly equal to N 1. A node is visited after
another node along a link. We will assume that no more
Figure 13. A 13-node tree example.
than one link exists between any pair of nodes. Nodes can
be visited in two different ways. In the rst case, once we
are at a node, we visit all of its neighbors before visiting any
other (non-neighbor) nodes. This process is called breadth
rst order. For example, for a given tree graph as shown in
Fig. 13, we rst visit the root node A. After that, we visit
all its children, which are B, C, and D. Then we traverse
children of B, C, and D, which are E and F, G and H, and
I, respectively. Finally, we visit the children of these nodes
and include nodes J, K, L, and M in the list of visited nodes.
In the second case, we visit nodes in depth rst order.
In this case, when we visit a node, we immediately visit
its children rst before visiting any of its siblings. In the
example tree of Fig. 13, the nodes will be visited in order
A, B, E, F, J, C, G, K, L, H, D, I, and M.
Depending on the application, one or the other method
is used. For example, if the tree nodes represent solutions
of a problem and we are interested in one solution, depth
rst search is likely to yield the solution faster. On the
other hand, if we are interested in all possible solutions,
then breadth rst search is more appropriate.
Network (Max) Flow
In a given network, one may like to compute the available
capacity on all paths from a source to destination. In that
case we need to determine the maximum information ow
possible from the source to the destination, which is accomplished by using a network ow analysis algorithms (15).
The network graph is treated as a directed graph, and the
maximum possible ow from a source node s to a destination node t is computed. For a given directed graph, each
edge (i, j) is assigned a capacity using a nonnegative value
Cij that represents the available capacity to carry information on edge (i, j) from node i to node j. In addition, nodes
may have additional constraints in terms of amount of information they can support in terms of buffer space and
other factors from all incoming edges or links. This characteristic is the node capacity constraint. Let Xij be the
amount of actual ow through edge (i, j). At each node, information must be conserved as part of the total ow from s
to t. That means the amount of information entering a node
must be the same as the amount of information leaving
that node. This information must not exceed the capacity
of the node, or the following constraints must be satised:
0 Xi j Ci j ,
and

i
Xi j =

j
X ji
Also
Xs j is the amount of information that leaves the
j

X jt , which is the amount
source node s and is equal to
j
of information that arrives at the destination node t.
Any such set of ows Xij that satises the above constraints is called a feasible ow set. Maximizing feasible
ow by increasing ow on different links while satisfying
all constraints yields the max-ow value. For a given graph,
this result is achieved. For a given graph, this result is
achieved as follows.
First, we nd a feasible ow from node s to node t (0
ow is trivial). Now, let P be an undirected path in the
directed network from s to t. An edge on this path is called
a forward edge if it is directed toward node t. Otherwise, it
is a backward edge. A ow on this path can be augmented
or increased if Xi j < Ci j on all forward edges and Xi j > 0 on
all backward edges. The amount of increase is given by
min{ min {Ci j Xi j }, min {Xi j }}
forward
backward
If this value is greater than zero, then such a path is called

an augmentation path. The process is repeated on all possible undirected paths. A ow is maximum if no augmentation path is available.
Figure 14 demonstrates computation of maximum ow.
Figure l4a depicts a feasible ow. Each edge is marked with
its capacity and the current ow value. Figure 14b shows
an augmentation path with three forward edges and one
backward edge. Using the relationship described above,
the amount of ow that can be increased is one. Figure
14c shows the graph again with new feasible ow. Figure
14d shows another augmentation path with three forward
edges. The ow can be increased by two on this path, and
the new feasible ow is shown in Figure 14e. Figure 14f
shows another augmentation path with four forward edges
and the ow is again increased by two to obtain a maximum
ow of ten as shown in Figure 14g.
The maximum value of a s-t ow is equal to the minimum capacity of a s-t cut. A cut is dened by a set of edges
that partitions the network into two parts with s and t in
separate partitions. A minimum cut set is a cut set whose
total capacity of the edges is minimum.
Linear Programming Problems (LPPs)
In network design, we are mostly concerned with minimizing cost or delay in the network while maximizing the performance. Such problems can be expressed as optimization problems. The statements of such problems have an
objective function that is required to be minimized or maximized subject to certain constraints. In most cases, these
constraints are also linear in relation. A general linear programming problem (16, 17) is to nd values of n real variables, denoted by x1 , x2 , x3 , , xn , which will minimize or
maximize an objective function given by
z=
n
c jx j
j=1
where Cj is a cost or reward value associated with variable

Xj . The set of constraints that governs a feasible solution
may vary in numbers and includes linear combinations of
10
Figure 14. An example demonstrating feasible ow and augmentation paths.
variables xs and has a general form as in

n
ai j x j > bi
est paths have to be provided.

However, there is always a trade-off between these two
objectives. Most routing algorithms are based on assigning
a cost measure to each link in a network. The cost could be
a xed quantity related to such parameters as link length,
bandwidth of link, or estimated propagation delay. Each
link has a cost associated with it, and in most cases, it is
assumed that the links have equal cost.
An interconnection network is strictly nonblocking if a
routing algorithm exists to add a new connection without disturbing existing connections (1). A network is rearrangeable if its permitted states realize every permutation
or allowable set of requests; here it is possible to rearrange
existing connections if necessary(1). Otherwise it is blocking.
The store-and-forward operation in packet switching
incurs a time delay and causes signicant performance
degradation. If the algorithm is used in a packet switching
network, the total time delay of a data packet is obtained
by summing up the time delay at each intermediate node.
As the nonavailability of any link along a route causes the
route not to be available, the network sees a high probability of blocking under heavy trafc, which rejects the
incoming request and eventually causes data loss or delay.
The routing algorithm can be centralized or decentralized. A centralized algorithm may use a global backtracking depth rst search or any other algorithm described in
section on algorithms.
i = 1, 2, , m.
j=1
The values of variables may also be bounded by some

lower and upper bounds as parts of constraints. For example, it may be desirable that all variables are positive or do
not exceed a certain value. Various methods solve LPPs.
The most commonly used method is the simplex method,

n!
which has no more than mn =
solutions for m
m!(n m)!
n possible solutions. The simplex method systematically
searches for an optimal solution over this space.
A variation of this problem is when all variables are
restricted to be integers only. This situation is called an integer programming problem (IPP), and it makes the problem more complicated. Standard packages are available to
solve the two types of problems. The goal of a network designer is to formulate the proble as an LPP or IPP and
then solve it using a standard package or a heuristic algorithm. If the problem size (the number of variables and
constraints and therefore the number of possible solutions
to search from) becomes too large, then we use heuristic
methods only to solve the problem.
ROUTING ALGORITHM
A routing algorithm establishes an appropriate path from
any given source to a destination. The objective of network
routing is to maximize network throughput with minimal
cost in terms of path length. To maximize throughput, a
routing algorithm has to provide as many communication
paths as possible. To minimize the cost of paths, the short-
EMBEDDING ARBITRARY CONNECTION REQUESTS

The interconnection network should be able to embed arbitrary requests until resources are available in the network. If a set of requests is such that each node needs to
communicate with a unique node, then such a set of requests is called a permutation. It is desirable to be able to
satisfy this set of requests simultaneously. If each node requires communications with up to k other nodes, it may not
be possible to satisfy these requests in one round and the
communication requests may have to wait. Depending on
the application environment, either the requests are partitioned in k disjoint permutatiqns (some may be partial
permutations) or the communication needs are satised
in k rounds without any contention. Altenatively, the network is designed to satisfy all requests up to k requests
at the same time. A better solution would probably lie in
between these two extreme cases. Depending on the number of transmitters and receivers, a node should be able to
source and sink those many connections. The links in the
network should be able to support the trafc corresponding
to the requests being serviced simultaneously. The permutation routing capability of a network is extremely useful
in improving the overall performance of a system.
In a permutation routing, the messages transferred
from a source to a destination can be regarded as a commodity ow. For each commodity, the required ow of commodity is 1 for a single source and a single destination. In
a general network, the problem of solving multicommodity
integral ows is known to be NP-complete.
NETWORK TOPOLOGY DESIGN

A network can be designed using various topologies. Many
interconnection networks have been proposed by the research community; some have been prototyped, but few
have progressed to become commercial products. A network may be static or dynamic (1823). The topologies can
be divided into two categories: (1) regular and (2) irregular. The regular topologies follow a well-dened function
to interconnect nodes. The regularity, symmetry, and most
often the strong connectivity of the regular network topologies make them suitable for general-purpose interconnection structures where the characteristics of the trafc originating from all nodes are identical and destinations are
uniformly distributed over the set of nodes. Thus, the link
trafc is also uniformly distributed. The irregular topologies are optimized based on the trafc demands. If there
is a high trafc ow between two nodes, then they may be
connected using a direct link. If a direct link is not feasible,
then an alternative is to provide a short path between the
two nodes. Such designs are much more involved and need
special attention.
We will rst discuss regular topologies and then get into
the design of irregular topologies. We will also discuss some
specic regular topologies, such as a binary cube and its
variations, in greater detail.
Regular Topologies
Several regular topologies have been proposed by various
researchers in the literature. The most important among
these are complete connected graphs, star, tree, ring, multiring, mesh, and hypercube. One desirable property of a
structure is to be able to accommodate or embed an arbitrary permutation. We discuss various regular topologies
in the following paragraphs.
Completely Connected Topologies. In a completely connected topology, every node is connecled to every other node
as shown in Fig. 15a; that is, for every i, j, N, (i, j) E.
Thus, N (N 1) links exist. The routing is straightforward as a node i directly sends messages for node j on the
corresponding link (i, j). Each node has N 1 transmitters
and N 1 receivers, one for each link. The diameter of the
graph is one and the reliability of the network is very high
as, in addition to a direct link, N 2 paths of two hops
exist from a node to every other node. This topology is the
most expensive but most efcient. In practice, not many
networks are aesigned using this topology. However, in a
given network, one may set virtual topologies that are the
eqivalent of a completely connected graph.
Star and Tree Topologies. The star and the tree are two
topologies that require a minimum number of links to connect N nodes. The number of links is exactly equal to N 1.
A star topology has a central node to which all other nodes
are connected as shown in Fig. 15b. The tree topology, as
shown in Fig. 15c and d, is hierarchical where the root of
the tree at each level has to act as the intermediate node
in any communication between nodes in the two halves of
the tree (called left subtree and right subtree). In the star
11
topology, the central node communicates with every other

node using the direct link. If we consider the central node
as only an intermediate node, then the routing between
any two nodes is always through the central node and each
path is of length two. The central node may become a bottleneck in communication. Failure of this node also causes
the entire network to fail. Moreover, the central node is the
most expensive node with degree N 1 and has to support
N 1 other connections. On the other hand, the degree of
each node is bounded and that is a big advantage. For example, in a star, each node connects to only one other node,
and in a binary tree, each node only connects to three other
nodes, one link to its parent node and at most two links to
its Children nodes. The longest path in a binary tree can
be up to 2 logN. In a hierarchical structure like a tree, a
different number of parallel links can be used to connect
nodes at two adjacent levels to accommodate more trafc
near the root node. This node is called a fat tree (24).
Rings and Multirings. The rings and multiring topologies
are even simpler designs with xed node degrees. For a
simple ring, each node is connected to two other nodes. If
the connections are unidirectional, then the simplest ring
has one incoming link and one outgoing link. The diameter
of the graph is N 1. In a bidirectional ring, each node
has two incoming links and two outgoing links. A node i
has a link to node i + l and node i 1 (module N). The
diameter of the graph is N/2. The multiring architecture
has multiple links from each node to other nodes. Each set
of corresponding links from each node forms one ring. Some
examples of ring topologies are shown in Fig. 15e, f, and g.
Meshes. A node in an n-dimensional mesh structure has
2n neighbors, two in each dimension. A two-dimensional
structure is shown in Fig. 15h. Each grid point is numbered
using an n-dimensional tuple. Two-and three-dimension
meshes are most commonly used in designing interconnection structures for multiprocessor systems. A mesh can be
extended or shrunk in any dimension allowing easy reconguration and scalability required in many subsystem designs.
Hypercubes and its Variations. A hypercube is ndimensional structure as shown in Fig. 15i. Hypercubes
and its variants are popular interconnection structures because of unique properties such as symmetry, regularity,
low diameter, and good fault tolerance characteristics (25).
A Boolean n-cube Qn = (V, E) has |V | = N = 2n nodes. Each
node is numbered using an n-bit binary string. The Hamming distance between two binary strings is the number
of bit positions in which they differ. A pair of nodes in a
Boolean cube is connected by an edge providing a bidirectional communication path between them if the Hamming
distance between their binary addresses is one. An important property of an n-cube is that it can be constructed
or decomposed recursively from/to two lower dimensional
subcubes as is clear from its recursive denition as given
next.
Denition. Boolean n-cube Qn = (V, E) is dened recursively as follows.
12
Figure 15. A hierarchical network.
1. The 0-cube Q0 is dened as a single node with no

edge.
2. Qn = Qn1+ Qdn1 , where the + operation is a twofold
operation of the graph G = (V,E) denoted by Gt = G +
Gd that yields a graph Gt = (Vt , Et ), where Vt = V
Vd and Et = E Ed {(v, |V | + v)| v V }.
The degree of each node, the diameter of the graph, and

the node connectivity of the hypercube graph is n each. The
length of the shortest path between any two nodes i and j in
an n-cube is equal to the Hamming distance between their
binary representations. There are H(i,j)! shortest paths between two nodes i and j, and among them, H(i, j) paths
are independent (node-disjoint or parallel). In a Boolean
n-cube, there are no cycles of odd length. The other regular
topologies discussed above can be embedded in a binary
n-cube or its variations discussed below.
Dynamic Topologies. A dynamic topology is created by

modifying an existing topology as the need arises. This
modication is achieved by adding links between nodes to
either create more paths or point-to-point direct links to
reduce delays and congestion and improve performance.
Resulting networks usually look like random graphs with
possibly no symmetry and very little fault tolerance.
Recongurable Topologies
There are two important issues in the design of a recongurable network: ease of embedding a given permutation
and the cost of implementing the network. An N N crossbar can realize all permutations easily but has a cost that is
proportional to O(N2 ). To reduce the cost, a rearrangeable
network (26) may be acceptable. The generalized foldingcube (GFC) and the enhanced hyper-cube (EHC) are two
such topologies derived from binary cube architecture.
Generalized Folding Cube. A generalized folding cube

is obtained by folding a hypercube along any dimension
as follows. For a given a Boolean n-cube Qn = (V,E), the
folding operation of the cube Qn denoted by f(Qn ) yields
a graph Q1n1 = (V 1 , E1 ) and consists of the following two
steps. (1) Split the cube into two subcubes by removing
(n l)-dimension links from Qn (Qn1 = (V , E ) and Qn1 =

(V , E )). (2) Overlap the two subcubes Q n1 and Qn1 in
such a way that v V and v V become one and the same
node v1 V1 if v and v differ by 2n1 . v1 is numbered as
min(v , v ). Either of the two links in each dimension, corresponding to v and v , can be used by either nodes, v and
v , for communication.
The kth folding operation f k (Qn ) = f ( f k1 (Qn ))
yields a graph Qknk = (V k , Ek ). The kth unfolding operapk
tion f k (Qnp ) = f 1 ( f (k1) (Qnp )) = Qn+k = (V pk , E pk ),
where k {1, 2, . . . , p}. The GFC denoted by Qnp = (V p , E p )
for p 0, is dened from the folding operation of a hypercube: f p (Qn+ p ) = Qnp . The GFC consists of 2P pairs
of links in each dimension, and each node of the GFC
consists of 2P individual nodes of the original cube and a
(n + 1)2 p (n + 1)2 p switch. The original hypercube Qn =
(V, E) can be considered as a special case of the GFC and
denoted by Q0 n = (V0 , E0 ). Figure 15j shows 3-D GFC with
2P pairs of links in each dimension. The rearrangeability
of the GFC is shown in Reference 27.
Enhanced Hyper-Cube. If we wish to keep only one node

at each vertex position and still want to design a rearrangeable network, then by duplicating links in any one
dimension of the original hypercube, i.e., two pairs of links
are provided instead on one, we obtain a structure that can
provide conict-free routes for every permutation (28). The
EHC is shown in Fig. 15k.
A recongurable architecture, such as EHC or GFC, can
embed other structures efciently. The EHC and the GFC
concepts can be combined to design a more cost-effective
network. This design methodology has been used to design
and implement the Proteus multicomputer system (29).
Helical Cube. A binary cube grows only as an integer

power of two. To remove this deciency, several of alternatives (30) have been suggested. An attractive option is a
helical cube that removes N K nodes from a hypercube
to obtain a K node structure while preserving all advantageous properties of the binary cube such as regularity, simplicity of routing, and fault tolerance. The degree of each
node remains n = log K. Only neighbors of removed nodes
are affected and reconnected in such a fashion that the high
graph connectivity is maintained. The links connected to
nodes that are being removed are connected pairwise using
a helical connection strategy, thus the name helical cube.
An example of a helical cube is shown in Fig. 15l. The details of the actual connection scheme are given in Reference 31. It has been shown that this structure can have
any number of nodes while maintaining a high connectivity and the same level of fault tolerance as the original
cube.
13
Arbitrary Topology Design

If the graph structure is not constrained to be a regular
topology, then the design problem can be formulated as a
linear or nonlinear programming problem. Suppose we are
use certain kinds of links and are given a trafc matrix.
Here we assume that only one type of links is available and
will consider a more complex problem in the last section.
We will wish to design a network that is connected. The
cost of connecting different links is different. Let Xi,j denote
whether a link between nodes i and j exists, Xi j {0, 1}, and
suppose the cost to lay the link is denoted by Cij . Let the
original nodes be numbered as 1 to N.
One goal of the design is minimize the cost that is given
by
Cost = Min
Ci j Xi j
ij
Then the existence of links has to be subjected to conditions that the network should satisfy. For example, each
node should be connected by at least one link. This can be
specied as
Xi j + X ji 1
Then we may have constraints to specify that there is a

path from each node to another node, or the graph should
be connected. It is hard to formalize this as an equation but
can be easily checked for a given Xij conguration. After all
the constraints are specied, one solves the problem to nd
a solution that is a vector of Xij .
It may appear to be a simple problem but is generally
very hard even for a moderate number of nodes. Therefore,
it is usually solved using some heuristics. We will see a
solution technique in the example section.
QUALITY-OF-SERVICE (QOS) REQUIREMENTS
Unlike conventional packet-or circuit-switched networks,
some applications such as broadband integrated services/digital network (B-ISDN) require the network to provide not only connectionless trafc transportation but also
connection-oriented operation for real-time data transfer
between end users with multiple bit rates. Broadband
packet switching based on the asynchronous transfer mode
(ATM) that has a xed packet length has been proposed for
multimedia and multibit rate communications of end users
by using the network resources efciently. The most important aspect of these networks is to satisfy the (QoS) requirements. These features require a different approach to network design in comparison with the conventional packetswitched network design. For example, the cell loss probability has to be considered in an ATM network design. In
circuit-switched networks, the call blocking probability is
an important metric to determine the design of the circuitswitched networks.
Connection-oriented services have certain maximum
delay requirements in exchanging information between
the end users as given by the QoS requirements. The delay in a packet-switched network includes switching, queu-
14
ing, transmission, and propagation delays. Because of the

high data rate of ber-optic links, propagation delay and
node queuing delay are the dominant delay factors. In the
conventional packet-switched network design problem, the
average network delay and throughput have usually been
used as the metrics to optimize the network cost and performance. In multimedia networks, services may have critical delay requirements; instead of the average network delay requirement, the end-to-end delay must be considered
while determining the network topology.
In addition to these new requirements, the high data
rates require special attention to fault management or
fault tolerance. Compared with low-speed data networks,
it is possible to lose many data, packets if a data link fails
even for a short time. Fault management requires that the
network has a control mechanisms that ensures that the
existing trafc is affected as little as possible because of
a failed link, and the trafc on the failed link is rerouted
through the spare capacity on other links. This rerouting
of trafc from a failed link to the other linkss (32) can be
performed by a special facility such as the digital crossconnect system (DCS)(33).
Fault tolerance in high-speed networks is greatly
needed even for short-time link failures because of large
cell loss possibility. An alternative route may be longer
than the original path. If a service, such as data le transfer, is not sensitive to propagation delay, the reconguration can be done using arbitrary available spare capacity on
the other links. As voice and video service are sensitive to
end-to-end delay, the reconguration path must be selected
such that the end-to-end delay requirements are met. This
performance requirement restricts the logical reconguration that can be embedded into the physical network.
Therefore, while designing the network topology, possible
failures of network links must be considered in advance.
To maintain the QoS requirements in services, we also
have to consider the cell loss probability during a burst
transfer. Burst cell loss can occur in several stages of the
network: switch buffer overow, cells discarded for congestion control, and physical link errors. The optical ber link
has negligible physical link errors. However, the switch
buffers for each link may be of a xed size and the packet
contention for the same link may cause the output buffer
to overow in each link. Thus, we have to nd the optimal
link capacity assignment to meet the cell loss restrictions.
With this in mind, we investigate a fault-tolerant backbone network design algorithm and network resource management schemes while considering different (QoS) such
as cell loss probability and mean end-to-end delay requirements for each call request. In the design algorithm, we
ensure that these performance requirements are met even
in the presence of faults in the network. We rst introduce
some mathematical notation and then present the design
formulation, solution heuristics, and numerical examples
demonstrating the goodness of the solution.
Figure 16. A bell-shaped curve.
DELAYS AND QUEUING IN NETWORK DESIGN

Probability Distributions
We will rst describe three important probability distribution functions that are used in the analysis of network
systems. More details can be found in Reference 34.
Normal Distribution. A random variable x is normally
distributed if its probability density function is of the form
f (x) =
1
2 2
(x)2
2 2
This is a bell-shaped curve density function as shown in

Fig. 16. The peak of the bell occurs at x = , and the width
of the bell depenas on the variable . The random variable
x is completely characterized by the two variables, mean ,
and the variance 2 . The variable is the standard deviation. Three standard deviations from the mean cover about
99% of the area under the curve. That is why most of the
time we are interested in + 3 variations in the value of
random variable x.
Binomial Distribution. The number of ways in which k
out of n objects can be selected is given by
C(n, k) =
n!
k!(n k)!
If the probability of selecting a particular type of object is

p (and the probability of selecting the other object(s) is (1
p)), then the probability of selecting k such objects out of
a total of n objects is given by
P(n, k, p) = C(n, k) pk (1 p)nk
The mean value of this statistic is E(n, p) = n p and the
variance is V (n,
(1 p). The standard deviation
p) = n p
is given by V (n, p) = n p (1 p).
Exponential Distribution. A random variable x is exponentially distributed with parameter if the probability of
x t is given by
p(x t) = 1 et
t0
The mean and variance of x are 1/ and 1/2 , respectively.

An arrival of a request for service is usually modeled using a random process. Two requests are often assumed to be
independent of each other. A process in which interarrival
times between two consecutive requests are independent

and distributed according to an exponential distribution
with parameter is called a Poisson process (with parameter ).
15
last bit is transmitted by the head of the link (source),

and it is received by the tail of the link (destination).
This delay depends on the physical distance of the
link and speed of propagation and can be substantial
for a high-speed link.
Design
How we use these distributions can be demonstrated using
the following example. Suppose a network has ten nodes
that want to communicate among themselves. Assume the
probability that a node originates a data request is p=
0.1. The switching network can connect a call if all required links are free or not in use. How many links should
we provide so that a communication request can be satised with high probability? In this case, the average number of requests
is E(n, p) = 10
0.1 = 1.0 and the deviation is = 10 0.1 0.9 = 0.9 = 0.95. To satisfy most
of the requests with high probability, we may like to provide + 3 = 1 + 3 0.95 4 links.
Delays in Networks
A communication link can be viewed as a bit pipe over
which a given number of bits is transmitted over a unit
of time. This number is called the transmission capacity
of the link and depends on the physical channel and the
interface at the two ends of the link. The bit pipe (link) is
used to serve all trafc streams that need to use the link.
The trafc of all streams may be merged into a single queue
and transmitted on a rst-come-rst-serve basis. This process is called statistical multiplexing. It is also possible
to maintain several queues for a link, one for each trafc stream or one for each priority if the incoming trafc
streams have multiple priority levels assigned to them. If
a packet length is L and the link capacity is C bits/s, then
it takes L/C seconds to transmit a packet.
In case all incoming communication requests for a link
are assigned to a queue and serviced as the resources become available, there are four different kinds of delays a
packet suffers on a link. If the packet has to travel through
multiple links, then the total delays will be the sum of delays on all links
1. Queuing Delay. The queueing delay is the delay between when a packet is assigned to a queue and when
it is ready to be processed for transmission. During this time, that packet simply waits in a queue.
This time depends on the number of packets waiting
ahead of this packet in the queue.
2. Processing Time Delay. The processing time is the
time between events when the packet is ready to be
processed and the time it is assigned to the link for
the transmission. The processing delay depends on
the speed of the link processor and the actions the
processor needs to take to schedule the transmission.
3. Transmission Delay. The time difference between
the transmission of the rst and last bit of the packet
is referred to as the transmission delay. This delay
depends on the bit transmission rate of the link.
4. Propagation Delay. The propagation delay refers to
the time difference between the instances when the
Queuing Models
To compute the queuing delay for a packet, we have to understand the nature of the packet arrival process to a link,
the kind of service time it needs (amount of transmission
time), and the number of links we have from the source
to the destination. In most queuing systems (35, 36), we
assume that the arrival process is a Poisson process. We
also assume that the holding time (the amount of time a
request requires to service) follows an exponential distribution with parameter . The mean service time is then
given by l/. If two nodes i and j are connected by m links,
then m packets can be transmitted from node i to node j at
the same time. Generally m = 1 and therefore packets are
transmitted one at a time. In case of circuit switching, it
can be observed as one request being established at a time.
M/M/m Queue. A queuing system with m servers, Poisson arrival process, and exponentially distributed service
times is denoted by the M/M/m queuing system. The rst
letter M stands for memoryless. It can also be G for general
distribution of interarrival times or D for deterministic interarrival times. The second letter stands for the type of
probability distribution of the service times and can again
be M, G, or D. The last number indicates the number of
servers.
In a M/M/l queuing system, the average number of re1
quests in the system in steady state is given by
and
the average delay per request (waiting time plus service

1
time) is given by
. Utilization of the system is denoted
by = /, and the average time for a request in a system

is given by average service time/(l ). The average waiting time Tw is given by the difference of the average time
in system and the average service time. This time is equal
to 1/( ) 1/. The average number of requests in the
queue is given by Tw . Also, the probability that exactly
k requests are waiting is given by Pk = (1 )k . These
results for a queuing system will be used later on.
Performance Metrics. When a request for service arrives,
the server (link) may be busy or free. If the server is free,
the request is serviced. If the server is busy, then there
are two possibilities: 1) The request is queued and serviced
when the server becomes available. In this case, we are
interested in nding out how long, on average, a request
may have to wait before it is serviced. In other words, we
need to nd out how many requests are pending in a queue
or the average length of the queue. This has implications
in designing queues to store requests. 2) The incoming request is denied service, which is called blocking. In this we
are interested in determining the blocking probability for
an incoming request. Again, this has implications in network design. We would like the blocking probability to be
16
domain.
WDM Network Design Issues
Figure 17. A ber divided among multiple wavelengths.
as small as possible.
EXAMPLE 1: DESIGN OF A NETWORK USING WDM
FIBER OPTICS
Wavelength Division Multiplexing-Based Optical
Networking Technology
With the advent of optical transmission technology over
optical bers, the communication networks have attained
orders of magnitude increase in the network capacity.
The bandwidth available on a ber is approximately 50
THz (terahertz). Hence, wavelength division multiplexing (WDM) was introduced that divided the available
ber bandwidth into multiple smaller bandwidth units
called wavelengths. Figure 17 depicts the WDM view of
a ber link. Different connections, each between a single
source/destination pair, can share the available bandwidth
on a link using different wavelength channels. Advanced
features such as optical channel routing and switching supports exible, scalable, and reliable transport of a wide variety of client signals at ultra-high speed.
Early optical networks employed broadcast and select
technology. In such networks, each node that needs to
transmit data broadcasts it using a single wavelength and
the receiving node selects the information it wants to receive by tuning its receiver to that wavelength. To avoid
unnecessary transmission of signals to nodes that do not
require them, wavelength routing mechanisms were developed and deployed. The use of wavelength to route data is
referred to as wavelength routing, and networks that employ this technique are known as wavelength-routed networks. In such networks, each connection between a pair of
nodes is assigned a path and a unique wavelength through
the network. A connection from one node to another node
established on a particular wavelength is referred to as a
lightpath. A wavelength-routed WDM network is shown in
Fig. 18. The gure shows connections established between
nodes A and C, B and C, H to G, B to F, and D and E. The connections from nodes A to C and B to F share a link. Hence,
they have to use different wavelengths on the ber.
One alternative to circuit switching, described above,
is to use optical packet switching (OPS) or optical burst
switching (OBS) (4446)technology in the backbone. The
major advantages of OPS/OBS are the exible and efcient
bandwidth usage, which enables the support of diverse services. However, implementation technologies are not yet
there for successful deployment of them in an all-optical
WDM network design involves assigning sufcient resources in the network that would meet the projected trafc demand. Typically, network design problems consider a
static trafc matrix and aim at designing a network that
would be optimized based on certain performance metrics.
Network design problems employing static trafc matrix
are typically formulated as optimization problems. To formulate a network design problem as an optimization problem, the inputs to the problem, in addition to a static trafc demand, are some specic reguirements, e.g., required
network reliability and fault tolerance requirements, network performance in terms of blocking, and restoration
time when a failure occurs. The objective of the optimization problem is to nd a topology that would minimize the
resources, including the number of links and bers, the
number of wavelengths on each ber, and the number of
cross-connect ports, to meet the given requirements. The
outputs include the network conguration and the routes
and wavelengths that are to be used for source-destination
pairs. The network design problem can be formulated as an
integer liner programming (ILP) or mixed integer linearprogramming (MILP) problem. As the number of variables and constraints can be very large in WDM networks,
heuristics are usually used to nd solutions faster.
If the trafc pattern in the network is dynamic, i.e.,
specic trafc is not known a priori, the design problem
involves assigning resources based on a certain projected
trafc distributions. In case of dynamic trafc, the network designer attempts to quantify certain performance
metrics in the network based on the distribution of the
trafc. The most commonly used metric in evaluating a
network under dynamic trafc pattern is blocking probability. The blocking probability is computed as the ratio of
number of requests that cannot be assigned a connection to
the total number of requests. With this metric, one makes
decisions on the amount of resources that are needed to
be deployed in a network, the operational policies such as
routing and wavelength assignment algorithms, and call
acceptance criteria.
Trafc Grooming WDM Networks
Data trafc in ultra-long-haul WDM networks is usually characterized by large, homogeneous data ows. The
metropolitan-area WDM networks, on the other hand, have
to deal with dynamic, heterogeneous service requirements.
In such WAN, and MANs, equipment costs increase if separate wayelengths are used for each service. Each wavelength offers a transmission capacity at gigabit per second
rates, whereas the users request connections at rates that
are far lower than the full wavelength capacity. In addition, for networks of practical size, the number of available wavelengths is still lower by a few orders of magnitude than the number of source-to-destination connections
that may be active at any given time. Hence, to make the
network viable and cost-effective, it must be able to offer
sub-wavelength-level services and must be able to pack
these services efciently onto the wavelengths. These sub-
17
Figure 18. A wavelength-routed WDM network.
wavelength services henceforth are referred to as low-rate

trafc streams in comparison with a full wavelength capacity. Such an act of multiplexing, demultiplexing, and
switching of lower-rate trafc streams onto high-capacity
lightpaths is referred to as trafc grooming. WDM networks offering such sub-wavelength low-rate services are
referred to as WDM grooming networks. Efcient trafc
grooming improves the wavelength utilization and reduces
equipment costs.
Dynamic Trafc Grooming in WDM Network
In the future, as Internet Protocol (IP) becomes the prevailing protocol, it is the responsibility of the IP layer to
effectively multiplex trafc onto wavelengths. These IPover-WDM networks are likely to be arranged in a mesh
topology rather than a ring. The trafc requirements of IP
are bound to change much faster than the static scenario.
It is thus important that dynamic trafc grooming is employed so that the networks can efciently accommodate
changes in trafc. Minimizing equipment costs in such a
dynamic trafc grooming scenario for SONET/WDM rings
is an important consideration.
It is possible to restrict trafc grooming in such a way
that all trafc streams that are groomed on a path originate and terminate at the same node pair. For example,
Fig. 19, trafc streams between node pair (S1 ,D1 ) and trafc streams between node pair (S2 ,D2 ) are groomed on their
respective paths. In another case, it is possible that trafc
streams between different node pairs share a path. For example in Fig. 19, trafc streams between node pair ((S1 ,D1 ))
and trafc streams between node pair ((S1 ,D2 )) can share a
link. Similarly, trafc streams between node pair ((S2 ,D1 ))
Figure 19. Example of grooming streams for same node pair.
and trafc streams between node pair ((S2 ,D2 )) can share
a link.
A challenging problem for carrying IP trafc over WDM
optical networks is the huge opto-electronic bandwidth
mismatch. One approach to provisioning fractional wavelength capacity is to divide a wavelength into multiple subchannels using time-, frequency-, or code division multiplexing and then multiplex trafc on the wavelength, i.e.,
trafc grooming, However, optical processing and buffer
technologies are still not mature enough to achieve online
routing decisions at high speed. With the development of
MPLS (Multiprotocol Label Switching) and GMPLS (Generalized Multiprotocol Label Switching) standards (4749)
it is possible to aggregate a set of IP packets for transport over a single lightpath. Therefore, trafc grooming in
IP over WDM optical networks is performed at two layers,
namely IP trafc grooming and WDM trafc grooming. IP
trafc grooming is the aggregation of smaller granularity
IP layer trafc streams. It is performed at MPLS/GMPLSenabled IP routers by using transmitters and receivers.
This aggregated trafc streams are then sent to the optical layer where WDM trafc grooming (or wavelength level
trafc grooming) is performed by using optical add-drop
18
Figure 20. Illustrative example of IP trafc grooming.
multiplexors (OADMs). The two-layered grooming reduces

the workload at both IP and optical layers.
IP Trafc Grooming Issues. The main cost in IP trafc
grooming is from the transmitters and receivers at the
end nodes rather than number of wavelengths, which was
the main cost for grooming ring network design. Minimizing the number of transmitters and receivers required is
equivalent to minimizing the number of lightpaths that are
needed, because each lightpath needs one transmitter and
one receiver. Figure 20 demcts an illustrative example that
shows how IP trafc grooming helps to reduce the number
of transmitters and receivers in a three node network.
Assume that each link has a capacity of 100 units. The
matrix in Fig. 20a is the original trafc matrix. It includes
the location and capacity of three requests. Figure 20a depicts one solution in the absence of IP trafc grooming,
and simply establishes a lightpath (connection) for each sd pair. It requires one transmitter and one receiver at each
node.
Figure 20b depicts another solution based on the fact
that the capacity requested by s-d pair (1, 3) is relatively
smaller. Thus, instead of reserving a separate lightpath
for it, the spare capacity along lightpath 1 2 and 2
3 can be reused to accommodate the trafc of s-d pair (1,
3). That is, the trafc from Node 1 to Node 2 and 3 both
take the route from Node 1 to Node 2. Node 2 receives and
analyzes the trafc, drops the trafc that is destined for it,
and forwards the remaining trafc (from Node 1 to Node 3)
along with its own trafc (from Node 2 to Node 3) to Node 3.
This add-and-drop procedure is performed by transmitters
and receivers at Node 2. In this scenario, the trafc carried
by the optical layer is represented by the matrix in Fig. 20b.
The scheme shown in Fig. 20b results into one less transmitter and receiver in comparison with the scheme shown
in Fig. 20a. However, the lower size trafc request (1, 3)
takes a longer route in IP layer to avoid reserving an entire wavelength for it. This tradeoff needs to be made in
order to alleviate the wavelength underutilization in the
the optical layer.
Approach to IP Grooming Problem. Let DNN = {dst } denote the trafc matrix, where dst denotes the trafc capacity required from source node s to destination node t, and
represent the capacity requirement of the systems.
The IP trafc grooming problem is described as follows.
Given a trafc matrix for a network, how to aggregate the

trafc requests for transporting, such that the total number
of transmitters (and receivers) required in the network is
minimized.
The Physical topology is represented by a graph GP (V,
E), with V being the set of nodes and E being the set of
physical links. The Virtual topology (logical topology) is represented by a graph Gl (V, L) with nodes corresponding to
the nodes in the physical network and edges corresponding
to the lightpaths. Each lightpath may extend over several
physical links (spans). The link ow and link capacity for
link (m, n) (from node m to node n) are denoted by xmn and
umn , respectively.
Notice that, each request is assigned a dedicated lightpath, the virtual topology would be a full-connected network if there is a request for each node pair. The desired
grooming network is the one with a minimum number of
transmitters and receivers, which is a solution with a minimum set of arcs in its virtual topology that is sufcient to
carry the given trafc.
For this problem to be meaningful, it is assumed that
each request has a capacity smaller than or equal to the
full-wavelength capacity. Note that, for a capacity requirement of more than a full wavelength, there has to be some
full wavelength paths assigned to this request and its remaining capacity need would be fullled using the trafc
grooming algorithm. The terms link and arc are used
interchangeably here.
This problem is similar to a capacitated multicommodity ow design problem (50) with limited link capacities.
Therefore, this problem can be formulated as an ILP optimization problem. It is assumed that a request from the
same s-d pair will always take the same route. Also, it is
assumed that each link has the same capacity that is given
by W C, where W denotes the number of wavelengths carried by a link and C denotes the full-wavelength capacity.
1. Notations:
1.1. Parameters:
* W: Maximum number of wavelengths in each
direction in a bidirectional ber (technologydependent data).
* C: Maximum capacity of each wavelength. (It
is assumed that each wavelength has the same
capacity.)
* s, t = 1,2,. . . , N: Number assigned to each node
in the network.
* l = l,2,. . . ,L: Number assigned to each link in
the network.
k
* Lst : (data) For each s-d node pair, list all possible routes from source node s to destination
node t, excluding routes that pass through a
node more than once, and number them using
3
k as an index. That is, r1,6
indicates the third
route from Node 1 to Node 6.
l,k
* Ast : (binary data) takes the value of 1 if arc l is
on the kth from node s to t; zero otherwise.
* dst : Denotes the trafc capacity required from
source node s to destination node t.
1.2. Variables:

k
* st : Binary variable, route usage indicator,
takes a value of 1 if route rstk is taken; zero otherwise.
* ul : Integer variable, logical link usage indicator, keeps an account of the number of lightpaths on arc l in the virtual topology.
Problem Formulation
1. Objective: The objective is to minimize the number of
arcs in the virtual topology, which reects the minimum number of lightpaths in the optical layer. Recall
that variable ul counts the number of lightpatns on arc
i in the virtual topology. If the capacity carried by arc
i exceeds the full wavelength capacity, multiple lightpaths between the same node pair are required. Thus,
the number of transmitters (and receivers) increase.
min
ul
(1)
lL
2. Fiber link capacity constraint: Let TC1 be the total capacity carried by link l, which is given by equation 2.
Constraint 3 guarantees that the aggregated capacity
on any arc does not exceed the total ber capacity, which
is bounded by W C.
TCl =

(s,t),s = t
stk Al,k
st dst
(2)
TCl W C
3. Trafc routes constraint: Equations 4 and 5 ensure that

if a request from node s to t, occurs one and only one
route is
assigned to the request. In another word, dst
0; set k stk = 1. Otherwise, no trafc request occurs
from node s to node t, and none
of the routes from node
s to node t will be taken; hence, k stk = 0.
stk dst
(4)
NC
stk dst
The limitation of this exact ILP formulation is that

it enumerates all possible routers for each s-d pair and
searches for an optimal set of arcs in virtual topology. In a
N2 h
fully connected network of N nodes, up to h=0 PN2
possible routes exist for each s-d pair, where Pmn is the permutation operation. This search requires large computation time as the network size increases. The formulation
can be further simplied by adding a hop-length constraint
such that the number of possible routes is reduced to a
reasonable number; consequently, the computation time is
saved. However, this network design problem is still a special case of multicommodity ow problem, which becomes
unmanageable even for moderata-sized networks. Therefore, a heuristic approach would be desired for obtaining
good solutions in a reasonable amount of time that capture all constraints of the ILP solution.
Approximate Approach
For a network G(V, E), in the absence of IP trafc grooming,
the number of transmitters and receivers required at node
s, denoted by Txsmax and Rxsmax , respectively, can be derived
from matrix DNN .
Txsmax =
dst

C
(8)
dts

C
(9)
t:(s,t) E
Rxsmax =

t:(t,s) E
(3)
(5)
4. Arc usage constraint: Recall that the arc usage indicator

ul , counts the number of lightpaths required on arc l
(logical link l) in order to carry the aggregated trafc
TC1 ul = [TCl /C], which is obtained by using equations
6 and 7. For example, if C = 48 and TCi = 62, [62/48] =
2 lightpaths are required on logical link i from its start
node to its end node to its end node.
C ul TCl
(6)
C ul TCl + C
(7)
Notice that, from equations 3 and 6, the total number of

lightpaths on a logical link l is bounded by the number of
wavelengths on the optical ber.
Additional constraints, such as the limited number of
transmitters on each node, can be easily added to this formulation. This process helps to capture the cost on each
node in the networks.
19
where C denotes the full wavelength capacity that can

be used, because request dst requires at most [dst /C] transmitters at node s to transmit trafc dts ; likewise, it requires
at most [dst /C] receivers at node from nodes t to receive
trafc dst from node s.
From the perspective of network ows, the total
amount
of outgoing trafc ows observed by node s is t = s dst and

the total amount of incoming ows to node s is
d .
t = s ts
Hence, the minimum number of transmitters and receivers
ne network to carry the trafc in DNN can be derived using
the following two equations:
Txsmin
=
t:(s,t) E

Rxsmin =
dst
t:(t,s) E
dts
(10)
(11)
In general, Txsmin and Rxsmin are loose lower bounds. The

reason is that, to reduce the number of transmitters (and
receivers), some s-d pairs may have to take multiple hops
and hence increase the link load in the virtual topology.
This overhead load is not captured in equations 10 and 11,
and it is dependent on the trafc pattern.
Trafc Aggregation Algorithm
To develop a trafc aggregation heuristic approach, the basic idea is to merge the smaller trafc request onto bigger bundles to reduce the number of transmitters and receivers. Although the total numberof lightpaths required
in the network is reduced, the ner granularity requests
may take multiple-hop and longer routes. This process may
20
introduce delay for lower-rate requests, and it would be affordable in the future slim IP-over-WDM control plane. As
a matter of fact, this is a trade-off that has to be made to
reduce the overall network cost.
An element in trafc matrix can be reallocated by merging it with other trafc streams. Thus, no need exists to
establish a direct path for that s-d pair. An element in trafc matrix can be aggregated if it is smaller then the full
capacity, i.e., has spare capacity on a wavelength channel
and allows other trafc streams to be merged on it. Each
element in the trafc matrix can be viewed as in one of the
three states:
State 0: If it can be reallocated or be aggregated.

State 1: If it cannot be reallocated but can be aggregated.
State 2: If it cannot be eliminated or aggregated. For

example, if dst = 0, no trafc exists to be reallocated
and no need exists to allocate trafc.
The goal of the trafc aggregation algorithm is to choose
a trafc stream dst that can be merged with some other
trafc streams dsn and dnt , so that dst can be carried using
a multipie-hop path and not burden the system to establish a new path for it. After selecting dst , the basic trafc
aggregation operation on trafc matrix D consists of the
following three steps:
1. dsn dst + dsn .
2. dnt dst + dnt .
3. dst 0.
After this operation, the trafc request between s-d pair
(s, t) is aggregated on s-d pairs (s, n) and (n, t). Let TR(Ta,t,n )
be the number of transmitters (equals to the number of
receivers) needed after merging dst with dsn and dnt . TR(T )
is called the upper bound, where T is the original trafc
matrix.
The key here is to select dst and node n to reduce the
value of TR(Ts,t,n ). In experimenting with the ILP formulation, described above, it is observed that the ILP solution
uses multi-hop routes for smaller requests, whereas the
bigger requests tend to use direct single-hop paths. This
observation is used to develop a heuristic solution. Figure
21 gives the trafc Aggregation algorithm. The resulting
new trafc matrix gives the structure of a virtual topology and the required capacity on each physical link. The
idea is to integrate smaller trafc request, say dst , to those
bigger trafc requests, dsn and dnt , to saturate the existing wavelength paths before establishing a new one. This
would force some smaller granularity trafc to take longer
routes with multiple hops, while saving some lightpaths.
The algorithm starts by nding the s-d pair with minimum request capacity that is in state 0 (Step 2 in Fig.
21), say dst . Next it searches for a set of all eligible intermediate nodes, namely K (Step 4a in Fig. 21). Dene
the index value of an item v in set K as index(v) = max(dsv ,
dvt ). The intermediate node n is selected from K to saturate
some wavelengths. Hence, if K is not empty, n is chosen as
the node with the maximum index value. One could choose
Figure 21. Approximate approach: trafc aggregation.
maximum or minimum to keep an order in which nodes are

explored. We choose maximum here. The algorithm then
updates the current trafc matrix after an intermediate
node is decided (Step 4e in Fig. 21). If K is empty, no eligible intermediate node is found for this s-d pair, dst state
is changed from 0 to 1, which means request dst cannot be
reallocated, but could be aggregated. The algorithm keeps
searching for the next s-d candidate for aggregation until
no eligible s-d pairs in State 0 can be found.
Complexity Analysis
One s-d pair is changed from State 0 to either State 1 or
State 2 in each step. Thus, the algorithm terminates after
at most N2 passes. Without any complex, the run time for
searching target in each loop is up to N2 ; it takes another
N loops to nd the set K. Thus, the overall computation
complexity of this algorithm is O(N5 ). In practice one will
never see this complexity and the algorithm terminates
much faster. One way is to use effective data structures to
make the search more efcient and faster.
Example of Trafc Aggregation
Figure 22 illustrates an example of how the trafc aggregation algorithm performs. Assume that each wavelength
has a capacity of OC-48 (2.5Gbps), and the minimum allocatable unit is OC-1. Thus, C = 48. Consider trafc matrix
that is composed of random combination of OC-1, OC-3,
and OC-12. An original trafc matrix includes all possible
s-d pairs, which is shown as the top-left matrix in Fig. 22.
The algorithm starts by nding the minimum eligible sd pair that can be reallocated, which is (1, 4) with d1,4 = 2 in
this example. Next it nds the possible intermediate nodes
to include into set K. It can be observed that K = 2,3 with
index(2) = 43 and index(3) = 37. Among the candidate nodes
in K, the one with the highest index value is chosen; that is,
n = 2. Next, the current trafc matrix is updated by removing d1,4 from the original position and aggregating it with
21
Figure 22. An illustrative example of the trafc aggregation algorithm.
d1,2 and d2,4 , which results into the matrix on the top right
in Fig. 22. Next the algorithm selects d2,3 = 6 and completes
its processing by choosing n = 1. The algorithm continues
until no more relocatable s-d pair exists as shown in Fig.
22. The botton-left matrix shows the nal results. Application of equations 8 and 9 indicate that 12 transmitters
(and receivers) are required for the original trafc matrix.
After trafc aggregation, this number is reduced by 3.
Solutions and Resuits
The previously given ILP formulation is solved by using
CPLEX Linear Optimizer 7.0. The ILP formulation and the
trafc aggregation approach are applied to solve IP trafc
grooming problem for a six-node network, with W = 6, C =
48. Table 1 gives a trafc matrix with randomly generated
50 requests. The integer numbers indicates the request capacity in a unit of OC-1 (51.84 Mbps). The objective is to design a network with as few logical links as possible. Notice
that there are totally P40 + P41 + P42 + P43 + P44 = 65 routes
for each s-d pair in a six-node network, and this number
increases dramatically as the network size increases. It
would be a great burden and might be unnecessary as well
to obtain optimality by searching among all possible routes.
Experiments with different maximum hop-length as 3, 4,
and 5 are performed on this six-node network. The results
show that limiting the hop-length to 3 still yields close to
an optimal solution, whereas the number of all candidate
paths for each s-d pair is effectively reduced from 65 to
P40 + P41 + P42 = 17, which signicantly reduces the size of
the feasible region of this ILP formulation; hence, it reduces
the computation complexity of solving the ILP optimization
problem.
The results obtained from solving ILP with hop length
= 3 and the trafc aggregation approach are shown in Fig.
23a and respectively.
According to Equations 10 and 11, at least nine transmitters (receivers) are required. Figure 23a shows an optimal solution consisting of 11 lightpaths by solving, the ILP
formulation with a maximum hop-length limit of 3. Figure 23b shows a solution with 12 transmitters (receivers)
using the trafc aggregation approach. Table 2 shows the
virtual topology routing assignments obtained by solving
Figure 23. Comparison of the ILP solution and the heuristic approach: an illustrative example, (a) Results obtained by solving
the ILP optimization problem with hop-length limit 3. (b) Results
obtained from trafc aggregation approach.
the ILP formulation and the trafc aggregation heuristic

algorithm.
Observations
Figure 23 also shows the similarity between the virtual
topology design obtained from solving ILP formulation and
the heuristic approach. More specically, the ILP formulation tends to keep bigger requests on shorter paths in virtual topology and tries to integrate smaller trafc streams
onto bigger bundles. The ILP approach provides an optimal
solution by performing exhaustive search among all possible routes. The trafc aggregation heuristic algorithm also
yields a very good solution in this example by just perfonning a local search, which takes much less computation
time. However, as an approximate approach, the trafc aggregation heuristic cannot guarantee optimality.
The integration of the trafc helps to reduce the number
of transmitters and receivers. On the other hand, it also
introduces overhead trafc to the network and impacts,
the resource utilization. Besides, it adds potential delays
to the requests, which have been reallocated to take multiple hops in the virtual topology. From Table 2, it can be
observed that the average hop-length in the ILP solution is
80/50 = 1.6. The average hop-length in the trafc aggregation heuristic is 77/50 = 1.54, whereas without grooming,
given enough resource, the minimum average hop-length
is 1. The more one saves on transmitters and receivers, the
longer the average hop-length is, accordingly the longer is
the average delay. This trade-off is an unavoidable one that
would have to be faced.
The ILP approach becomes unmanageable quickly as
the size of the network increases. The reason is that the
number of all possible arcs in the corresponding fully con-
22

Table 1. Requests Matrix for A six-Node Network
1
2
3
4
5
6
1
0
12+12+12+3
3
3
3
1+3
2
3
0
1
12
3+12
12
3
3+1+1
3
0
3+12+3+3
12
0
4
12+12
1+3
12+12
0
0
3+12
5
3+1+1
0
3+1+1
1
0
0
6
12+12
1+1+12
0
3+1+1+12
3+1
0
Table 2. Resulting Routes in Virtual Topologies

Node
pair
1-2
1-3
1-4
1-5
1-6
2-1
2-3
2-4
2-6
3-1
3-2
3-4
3-5
4-1
4-2
4-3
4-5
4-6
5-1
5-2
5-3
5-6
6-1
6-2
6-4
Requested
capacity
3
5
24
5
24
39
3
4
14
3
1
24
5
3
12
21
1
17
3
15
12
4
4
12
15
ILP formulation
Route on VT
1-6-2
1-4-3
1-4
1-4-3-5
1-6
2-1
2-4-3
2-4
2-4-6
3-5-2-1
3-5-2
3-5-4
3-5
4-1
4-6-2
4-3
4-3-5
4-6
5-4-1
5-2
5-2-4-3
5-4-6
6-2-1
6-2
6-2-4
Trafc aggregation
Route on VT
1-4-2
1-4-3
1-4
1-4-3-5
1-6
2-1
2-1-4-3
2-6-4
2-6
3-4-1
3-4-2
3-4
3-5
4-1
4-2
4-3
4-3-5
4-2-6
5-2-1
5-2
5-3
5-2-6
6-4-1
6-4-2
6-4
Table 3. Trafc Matrix for a 10-Node Network

1
2
3
4
5
6
7
8
9
10
1
0
3
9
6
0
11
0
0
4
0
2
5
0
3
0
6
3
2
5
5
9
3
8
8
0
8
10
4
10
6
11
9
4
11
4
7
0
4
4
2
2
8
3
5
3
0
3
2
0
3
11
3
8
7
6
8
5
10
5
2
0
5
1
2
10
7
5
1
11
5
11
2
0
11
3
1
8
7
2
8
2
10
6
1
0
1
2
9
8
3
0
1
5
8
6
5
0
1
10
10
1
6
1
2
3
0
0
5
0
nected network increases dramatically as the number of

nodes increases. The performance of the IP trafc aggregation heuristic approach is studied in terms of wavelength
utilization in the following section.
EXAMPLE 2: LIGHT TRAIL NETWORK ARCHITECTURE

FOR GROOMING
The Light trail architecture concept has been proposed as
a novel architecture designed for carrying ner granularity IP trafc. A light trail is a unidirectional optical trail
between the start node and the end node. It is similar to a
lightpath, with one important difference that the intermediate nodes can also access this unidirectional trail. More-
23
Table 4. Resulting Light Trails Tlmax = 4

No.
1
2
3
Light Trails
2, 3, 4, 7, 9
3, 2, 6, 8, 10
4, 3, 2, 1, 5
Hops
4
4
4
4
5
4, 7, 6, 8, 10
5, 1, 2, 3, 4
4
4
6
7
8
5, 1, 6, 7, 9
5, 1, 6, 8, 10
5, 8, 7, 9, 10
4
4
4
9
10
11
9, 7, 4, 3, 2
9, 7, 6, 1, 5
10, 8, 6, 2, 3
4
4
4
12
10, 8, 6, 7, 4
13
10, 9, 7, 8, 5
Accommodated sd Pairs
(3,7) (3,4) (2,7) (2,9) (4,9)
(2,6) (2,8) (2,10) (3,6) (3,8) (3,10)
(4,1) (4,3) (4,5) (3,5) (1,5) (3,1)
(2,1)
(6,8) (6,10) (4,6) (4,7) (4,8) (4,10)
(1,2) (1,3) (1,4) (5,2) (5,3) (5,4)
(2,4)
(1,7) (1,9) (6,9)
(1,8) (1,10) (1,6) (5,6)
(9,10) (8,9) (5,9) (5,8) (5,7) (7,9)
(5,10)
(9,2) (9,3) (9,4) (7,3) (7,2) (3,2)
(7,6) (6,5) (9,1) (9,6) (6,1)
(10,3) (10,2) (8,3) (8,2) (6,3) (6,2)
(2,3)
(10,6) (10,4) (7,4) (6,4) (6,7) (8,4)
(8,6) (8,7)
(10,9) (10,8) (10,7) (10,5) (9,8) (9,7)
(9,5) (8,5) (7,8) (7,5)
Load
23
32
34
22
48
21
27
44
39
25
44
35
38
Table 5. Local Best-Fit: Resulting Light Trails Tlmax = 4

No.
1
Light Tails
3, 2, 6, 8, 10
Hops
4
10, 8, 6, 2, 3
1, 6, 2, 3, 4
1, 5, 8, 10, 9
5
6
2, 6, 8, 7, 9
3, 4, 7, 8, 5
4
4
7
8
9
10
11
12
13
14
15
16
4, 3, 2, 6, 1
4, 7, 9, 10
5, 8, 7, 4, 3
9, 7, 6, 2, 1
9, 7, 4, 3
9, 10, 8, 5
10, 8, 6, 7, 4
1, 5, 8, 6, 7
5, 1, 2
6, 1, 5
4
3
4
4
3
3
4
4
2
2
Accommodated sd Pairs
(3,10) (2,10) (2,8) (3,2) (6,10) (2,6)
(6,8) (3,8) (3,6)
(10,3) (8,6) (10,8) (6,2) (6,3) (8,2)
(8,3) (2,3) (10,2)
(1,4) (6,4) (2,4) (1,2) (3,4) (1,3)
(1,6)
(1,9) (10,9) (5,10) (1,5) (8,9) (5,9)
(1,8) (1,10)
(2,9) (2,7) (6,7) (7,9) (6,9) (8,7)
(3,5) (7,8) (4,5) (4,8) (8,5) (4,7)
(7,5) (3,7)
(4,1) (2,1) (4,6) (4,3) (3,1) (6,1)
(4,10) (4,9) (9,10)
(5,3) (8,4) (7,4) (5,4) (5,8) (7,3)
(9,1) (9,6) (7,2) (9,7) (7,6) (9,2)
(9,3) (9,4)
(9,5) (9,8) (10,5)
(10,4) (10,6) (10,7)
(1,7) (5,6) (5,7)
(5,2)
(6,5)
over, light trail architecture, as detailed later on, does not

involve any active switching components. However, these
differences make the light trail an ideal candidate for trafc grooming. In light trails, the wavelength is shared in
time by the nodes on the light trail. Medium access is arbitrated by a control protocol among the nodes that have
data ready to transmit at the same time. In a simple algorithm, upstream nodes have higher priorities over the
nodes downstream.
Current technologies that transport IP-centric trafc
in optical networks are often too expensive, because of
their reliance on an expensive optical and opto-electronic
approach. Consumers generate diverse granularity trafc, and service providers need technologies that are affordable and seamlessly upgradable. The exclusion of fast
Load
44
47
47
41
31
38
42
7
38
21
19
16
14
18
6
3
switching at the packet/burst level, combined with the exible provisioning for diverse trafc granularity, makes the
light trails an attractive option to conventional circuit-and
burst-switched architecture.
Light Trail
A four-node light trail is depicted in Fig. 24. The light trail
starts from Node 1, passes through Node 2, Node 3, and
ends at Node 4. Each of the nodes 1, 2, and 3 are allowed
to transmit data to any of their respective downstream
nodes without a need for optical switch reconguration.
Every node receives data from upstream nodes, but only
a requested destination node(s) accepts the data packets,
whereas other nodes ignore them. An out-of-band control
signal carrying information pertaining to the setup, tear
24
light trail, which mainly comes from the power splitting

at each node, the length of a light trail is limited and is
estimated in terms of hop-length. The expected length of a
light trail is 4 to 6 hops (51).
Figure 24. A light trail and possible trafc streams.
Light Trail Characteristics

As no need exists to dynamically congure any switches
when using light trails to carry IP bursts, it leads to an excellent provisioning time. Moreover, the major advantage
of using light trails for burst trafc is the improved wavelength utilization. Utilization here is dened as the ratio
of capacity used over time for actual data transmission to
the total reserved capacity.
Multicasting in the optical layer is another salient feature of light trail architecture. Nodes in a light trail can
send the same quanta of information to a set of downstream
nodes without a need for a special processing or control arbitration.
In general, a light trail offers a technologically exclusive
solution that enables several salient features and is practical. It exhibits a set of properties that distinguishes and
differentiates it from other platforms. The following four
characteristics are key properties:
Figure 25. An example node structure in the light trail framework.
down, and dimensioning of light trails is dropped and processed at each node in the light trail. As a light trail is
unidirectional, a light trail with NT nodes can be used by
NT (NT 1)
optical connections along the trail. The
up to
2
six paths for the four-node light trail are shown in Fig. 24.
Node Structure
Figure 25 provides a node structure that can be deployed in
a light trail framework. In the gure, the multiple wavelengths from the input link are demultiplexed and then
sent to corresponding light trail switches. A portion of the
signal power is directed to the local receiver, the remaining signal power passes through an optical shutter. Such
a shutter can be realized using various technologies as
an AOTF (Acousto-Optic Tunable Filter). Thus, a node receives signals from all wavelengths. If a particular wavelength is not being used by an upstream node (incoming
ber has no signal), the local host can insert its own signal; otherwise, it does not use the trail. The local signal is
coupled with the incoming signal as shown in the gure.
Figure 26 depicts a connection of a four-node light trail
in a network and the corresponding ON/OFF switch congurations. The direction of communication is from Node 1
to Node 4. The optical shutter is set to the OFF state at the
start and end nodes of the light trail such that the signal is
blocked from traveling further. For an intermediate node
along the light trail, the optical shutter is set to the ON
state to allow the signal to pass through the node. A unidirectional light trail is thereby obtained from the start node
to the end node. No switch reconguration is required after
the initial light trail setup. From the power loss within the
The light trail provides a way to groom trafc from

many nodes to share a wavelength path to transmit
their sub-wavelength capacity trafc.
The light trail is built using mature components that
are congured in such a way that allows extremely
fast provisioning of network resources, which allows
for dynamic control for the uctuating bandwidth requirements on the nodes connected to a trail.
The light trail offers a method to group a set of nodes
at the physical layer to create optical multicasting,
which is a key feature for the success of many applications.
The maturity of components leads to the implementation of the light trail in a cost-effective manner resulting in economically viable solutions for mass deployment.
Light trail architecture brings up various issues in designing optical networks for transporting IP-centric trafc.
These questions are as follows:
How to is a set of light trails identied at the design

phase for the given trafc?
How hard is this problem?

What are the new constraints introduced by the light
trail architecture?
How good can wavelength utilization be in light trail

networks?
How is survivability achieves in light trail networks?

These questions are answered in the following discussion.
25
Figure 26. An example node conguration in the light trail framework.
Light Trail Design

To identify a set of light trails to carry the given trafc is
one key issue in setting up light trails in a WDM network.
The performance of the light trail in terms of wavelength
utilization depends on the locations of the light trail. The
goal of the design problem therefore is to develop an effective method to groom trafc in a light trail architecture and
to come up with a set of light trails. The light trail design
problem is stated as follows:
Given a graph G(V, E), where |V | = N, and trafc matrix
DNN , dene a minimum number of light trails to carry the
given trafc.
The design problem is expected to be a hard problem.
The approach to identify a set of light trail to be set up in a
network presented here consists of two steps. The rst step
is called the trafc matrix preprocessing step. As stated earlier, because of the power losses on the lines, a long light
trail may not be advisable. The length of a light trail is
limited and is specied in terms of hop-length, denoted by
Tlmax . A reasonable hop-length of a light trail is set to 5.
Therefore, in the rst step, a single long hop trafc is recursively divided into multiple hops.
The second step is to formulate the design problem and
to solve it as an ILP optimization problem, for a given network topology and rened trafc matrix obtained from step
one. The objective is to nd a minimum number of light
trails that is required for the system to carry the trafc.
Step I: Trafc Matrix Preprocessing. In preprocessing of
given trafc matrix, a single long hop trafc is divided into
multiple hops to satisfy the hop-length constraint. For a
given network physical topology G(V, E), with N nodes and
E links, one can apply Dijkstra s shortest path algorithm
to nd the shortest path between all s-d pairs. This step results into a distance matrix HNN = {hst }, where hst denotes
the physical distance from node s to node t.
The length of a light trail is a main constraint from the
loss both at nodes and over the links. Let Tlmax be the maximum length of a light trail. For trafc between an s-d pair
(i,j), where hst > Tlmax , it is not possible to accommodate
this trafc on a direct light trail. Thus, this trafc needs to
go through multiple hops. Here one light trail is counted
as one hop, which necessitates the rst step in this approach, namely trafc matrix preprocessing.
Let DNN = {dst } denote the estimated trafc matrix.
Trafc matrix preprocessing returns a modied trafc matrix that satises DNN = { dst : hst Tlmax , dst > 0 }. Figure 27 provides pseudo-code for the trafc matrix preprocessing algorithm.
In this step, trafc on s-d pair (s, t) with hst > Tlmax is
reallocated on multiple hops. The goal is to nd a node n
such that path from node s to node n forms the rst hop,
which is less than Tlmax in distance. A next intermediate
node n is found recursively for a new source node. Among
all possible intermediate nodes, n is chosen to be as close to
destination node t as possible, as shown in step 1 in Fig. 27.
This is done to reduce the number of hops that the original
trafc has to take.
After the preprocessing of the trafc matrix, each
nonzero element in modied trafc matrix would have corresponding distance that is less than Tlmax , the maximum
length allowed for a light trail.
Step II: ILP Formulation: Given the network topology
G p (V, E), and modied trafc matrix obtained from Step I,
the next step is to list all possible paths within the hoplength limit for each s-d node pair, which can be accomplished by applying a breath rst search for each node.
These eligible paths form a set of all possible light trails.
Among all possible choices, the next step is to choose an optimal set of paths to form the light trail network, such that
the total number of light trails is minimized. This problem
is formulated as an ILP optimization problem. It is also assumed that each request cannot be divided into different
parts and transferred separately.
For the given directed graph G p (V, E), N = |V |, let LT be
set of all the possible light trails within hop-length limit
Tlmax and Let T = 1, 2, . . . , |LT | be the number assigned to
each light trail in the LT.
Let C denote the full-wavelength capacity, represented
as an integer that is a multiple of the smallest capacity
requests. The smallest capacity request is denoted as 1.
The integer entry in trafc matrix DNN , represented by
dst , denotes the requested capacity from node s to node t in
the units of the smallest capacity request.
A single ber network with fractional wavelength capacity is considered. Hence, dst C. In the absence of wavelength converters, the wavelength continuity constraints
must be satised for light trail networks. The grooming
helps to increase the wavelength utilization and reduces
the total number of wavelengths that is required to satisfy the trafc needs. The following notations are used in
problem formulation.
Variables.
: (binary variable) Route indicator takes the Value

st
of 1 if request (s, t) takes light trail ; zero otherwise.
This also implies that nodes s and t are on trail and
26
Figure 27. Light trail establishment step 1: Trafc matrix preprocessing.
s is t s upstream node.
: (binary variable) Light trail usage indicator takes

value of 1 if trail is used by any request; zero otherwise.
ILP Formulation.
Objective:
min
(12)
When C = 1, the objective is to minimize the number

of light trails that is required in the network. When C
is dened as the hop-length of light trail , the problem
becomes to minimize the total wavelength-links in the
networks, which represent the total reserved capacity in the networks. This can be used to optimize the
wavelength capacity utilization, although that might
consume more light trails.
Assignment constraint: Each request is assigned to
one and only one light trail.
st = 1
(s, t) : dst D, dst > 0
(13)
Light trail capacity constraint: The aggregated request capacity on a light trail should not exceed the
full-wavelength capacity.
st dst C
(14)
(s,t)
Light trail usage constraint: If any of the s-d pair is

assigned on light trail , is set to 1; otherwise, if
none of the s-d pairs picked light trail , l = 0. Recall
that is a binary variable.
st
(s, t) : dst D
{0, 1}
(15)
(16)
Solution Considerations
The light trail design is a challenging problem for the following reasons.
First, to use a wavelength fully, one would like to groom
near full-wavelength capacity trafc onto the wavelength.
This is similar to a normal trafc grooming problem, which
is often formulated as a knapsack problem and is known
to be an NP-complete problem. However, it might be infeasible to simply set up a light trail for any set of trafc requests that add up to C. For example, given that
d12 + d13 + d16 = C, it might not be possible to establish the
desired light trail because of the physical hop-length constraint. As a matter of fact, the light trail hop-length limit
introduces complexity to the problem.
Second, the ILP formulation of the light trail design
problem is similar to the bin packing problem, which is
an NP-hard problem. However, if light trails are treated
as the bins, and elements in the given trafc matrix as
the items in the bin packing problem, this problem differs
from a normal bin packing problem because of a potential
physical route constraint that an item cannot be put in
any of the given bins but only a subset of the bins. More
specically, an s-d pair can be assigned to the routes that
satisfy 1) nodes s and t belong to the route and 2) node
s is the upstream node of node t along the route. Hence,
the approximate algorithms for solving normal bin packing problems cannot be directly applied here for solving the
light trail design problem.
Light Trail Design: Heuristic Approaches
As the study of Reference 52 proves that the light trail
design problem is NP-hard, the following heuristic algorithms for light trail design is proposed. It is well known
that the rst-t and best-t are two common and effective
heuristic algorithms for solving bin packing problems. In
the following, the best-t algorithm is used to solve the
light trail design problem.
The Best-Fit Approach. Recall that, after trafc matrix
preprocessing, each request in the newly obtained trafc
matrix satises the light trail hop-length limit; that is, the
shortest hop-length for each s-d pair is no greater than
Tlmax .
The goal of the second step is to identify a set of light
trails for carrying the given trafc. To do this, rst pick
the s-d pair that has the longest distance in the distance
matrix Hst . A light trail between this s-d pair is eventually
required.
Once an s-d pair with the longest physical hop-length is
found, the head and tail of a light trail is decided. The goal
now is to nd the best eligible light trail between these two
end nodes, which is analogous to fully packing a bin in the
bin packing problem. Two subproblems need to be solved.

First, selection of a path (within the hop-length limit) between these two nodes is required. Second, assignment of
requests to this light trail needs to be identied.
To nd the best light trail between the known head and
tail nodes, an exhaustive search among all possible paths
between the two nodes is performed. Best-t here tries to
pick up the path between the head and the tail nodes that
is the best among all paths available between the head and
the tail nodes. This search; is still local therefore, the nal
results might not be globally optimal.
For each eligible path between the known head and tail
nodes, all possible s-d pairs along this path are sorted according to their required capacities, before the routing decision is made. There are two different ways of packing
them onto a path rather than doing it randomly. One is to
allocate the smallest requests rst, which is called the increasing packing order, and the other way is to allocate the
biggest requests rst, which is called decreasing packing
order.
Increasing packing order tries to allocate ner requests rst, so that the number of requests that can
be packed onto this path is maximized. Some capacity might still be left on this light trail, but that is
not sufcient for the next smallest request. This approach grooms as many requests as possible onto the
light trail, thereby leaving the rest of the network
with fewer number of requests that still need to be
allocated. The expectation is that this contributes to
the saving on the total number of light trails that
are needed in the network. However, for each light
trail, the packing might not be the most efcient or
the spare capacity might not be minimized.
Decreasing packing order tries to allocate bigger requests rst and leaves the light trail with minimum
spare capacity. However, as the big requests are allocated rst, the total number of requests that can be
carried by the light trail might be smaller than that of
the allocation in the Increasing packing order. Therefore, it could leave more requests unallocated in the
network and more light trails might need to be set up
later on in order to carry all requests. The spare capacity on each light trail is minimized in this approach
at the time of allocating the capacity.
It is not clear which approach works better and always
gives the minimum number of light trails required in the
network. It depends on the trafc patterns. A preferred
approach is to try both and choose the one that yields a
better solution for given data.
Algorithm Design. For the given graph, all possible paths
for each s-d pair can be computed. The paths information is stored appropriately. The data structure called
KSPath[N][N][NRoutemax ] contains the path information
for each route in the network.
For efcient usage, paths are sorted according to their
physical hop-length, such that KSPath[head][tail][1] contains the shortest path information (hop-length, interme-
27
diate nodes along this path) head to tail, and so on.

Figure 28 gives the pseudo-code of the best-t algorithm. In this pseudo-code, seq is used to denote a route
among all valid routes from which head and tail are chosen to be the trail. Also notice that only sub-wavelength
level requests are considered here. Therefore, by default, a
shortest path is chosen as the light trail to carry a given
request if no better path can be found. That is, initially, seq
= 1.
When there is a tie in route selection, the path that
can accommodate most requests is chosen. It is possible to
design and apply different criteria. As mentioned, sorting
AllRequest[ ] in different ways yields different algorithms,
namely, best-t decreasing packing order and best-t increasing packing order.
Discussions. The proposed heuristic algorithm has two
steps, as shown in Fig. 27 and 28. Both the rst step and the
second step would need the information of paths between
each s-d pairs. Therefore, one can rst nd out all possible paths for each s-d pairs. The worst-case complexity of
the exhaustive searching for each s-d pair is O(N3 ), The total running time for nding all possible routes is O(RN 3 ),
where R is the number of s-d pairs (requests). In fact, instead of searching for all paths, it is preferable to search
among the K-shortest path with K being big enough. This
could reduce the complexity to O(N(E + NlogN + KN)) for
all node pairs, which may be a promising choice for big
networks.
In best-t packing of Step 2, for each s-d pair, the best-t
route is chosen among all K paths. For path with n nodes,
there are a maximum of t = (n 1) + (n 2) + + 1 =
O(n2 ) s-d pairs, where n is bounded by Tlmax . Hence,
2
t = O(Tlmax
). The sorting takes O(tlogt) loops, and packing
takes another t loops. Thus the total complexity is O(tlogt)
loops for each path. There are K paths, and the same procedure is performed on the selected best-t path. Therefore,
2
a total of O(K(tlogt)) = O(K(Tlmax
logTlmax )) loops is needed
for each s-d pair. At least one s-d pair is eliminated from
matrix R in Fig. 28 in each step and the program stops
when R is empty.
Algorithm Performance. To evaluate the performance of
the above ILP formulations and heuristic algorithms, experiments are performed on a physical topology given in
Fig. 29. To simplify the problem, it is assumed that each
physical link is bidirectional with the same length.
Table 3 gives a randomly generated trafc matrix for
this example. The integer numbers indicate the requested
capacity in a unit of OC-1 (51.84 Mbps). An entire wavelength capacity is OC-48. As aforementioned, only the fractional wavelength capacity is considered for trafc grooming in light trail networks. Intuitively, if every s-d pair requires a capacity greater than half of the full wavelength
capacity, no two requests can be groomed on a light trail.
Thus, it is assumed that most s-d pairs request a small
fractional capacity of the full wavelength channel. Hence,
integer numbers between 0 and 11 are randomly generated
as requested capacities in the experiments. The resulting
trafc matrix is shown in Table 3.
28
Figure 28. Light trail design step 2: best rst approach.
Figure 29. A 10-node example network.
The CPLEX Linear Optimizer 7.0 is used to solve the

ILP formulation proposed. It is assumed that each candidate path can be used once; that is, u = 1. Assume that the
hop-length limit Tlmax = 4, from the topology it is observed
that all s-d pairs have paths within this hop-length limit.
Hence, the trafc matrix preprocessing does not make any
change in the given trafc matrix.
Table 4 presents the results obtained by solving the ILP
formulation with hop-length limit Tlmax = 4. It is observed
that W = 4 is sufcient on each link, although no constraint
is imposed on the number of wavelengths.
Table 4 shows the 13 light trails that are needed to carry

the given trafc. The trafc assignment obtained from solving the ILP formulation is also listed. For each light trail,
the summation of all trafc it carries is calculated and
shown in the right-most column in Table 4.
Table 5 depicts the results from solving the local best-t
heuristic algorithm proposed above. In this example, local best-t increasing packing approach requires 16 light
trails.
Discussions. An observation from the optimal solutions

obtained by solving ILP formations is that only the longest
candidate paths are chosen as light trails, because only
the number of light trails is being minimized. The program
stops searching further once the number of light trails does
not decrease, even though it is possible to substitute some
light trails with the other shorter paths.
The problem becomes unmanageable in case of the ILP
approach as the problem size increases. In such a scenario, the use of relaxation techniques would be a preferred choice. When the trafc is uniform or the variation
among different requests are small enough that they can
be approximately treated as uniform trafc, DNN = {ds,t =
d| (s, t)}. LP relaxation is a very effective means for obtaining fast solutions, which can be achieved by modifying
the light trail capacity constraint in the ILP formulation
as follow. The rest of the formulation remains the same.
s,t C/d
(17)
0 1
(18)
(s,t)
s,t
(19)
In this formulation, the coefcient matrix of variables is

totally unimodular. Hence, the LP relaxation still yields integer solutions. This effect is for the same reason as noted
earlier in an LP to ILP in earlier article. Thus, this formulation can be applied to solve the light trail design problem
where the trafc requests have similar capacities.
SUMMARY
The network design deals with the interconnection of various nodes and how to transmit information from one
node to another. We have addressed the issues in network
design. Four important factors, network topology, transfer technologies, network management and control techniques, and cost were identied and discussed. We also discussed analysis methods using graph theoretic models for
a network and issues in topology design and optimization.
We also introduced performance metrics such as blocking
probability, throughput, and delay and how to account for
them in the design of a network. An important issue to such
design is fault tolerance. Two example designs, a WDMbased optical ber network and a light trail network architecture, were presented considering all the factors together,
demonstrating the concepts presented, and analyzing the
trade-offs in the design, and the methods to resolve them.
BIBLIOGRAPHY
1. Benes, V. E. Mathematical Theory of Connecting Networks and
Telephone Trafc; Academic: New York, 1965.
2. Schwartz, M. Computer Communication Network Design and
Analysis Prentice-Hall: Englewood Cliffs, NJ, 1977.
3. Lawler, E. Combinatorial Optimization: Networks and Matroids; Holt, Rinehart & Winston: New York, 1985.
4. Bertsekas, D.; Gallager, R. Data Networks; Prentice-Hall: Englewood Cliffs, NJ, 1987.
29
5. Stallings, W. Data and Computer Communication, Macmillan:

New York, 1997.
6. Kershenbaum, A. Telecommunications Network Design Algorithms McGraw-Hill: New York, 1993.
7. Tanenbaum, A. S. Computer Networks; Prentice-Hall: Englewood Cliffs, NJ, 1996.
8. Harary, F. Graph Theory; Addison-Wesley, Reading, MA,
1969.
9. Bellman, R. On a Routing Problem. Q. Appl. Math. 1958, 16,pp
8790.
10. Girard, A. Routing and Dimensioning in Circuit-Switched Networks; Addison-Wesley: Reading, MA, 1990.
11. Dijkstra, E. W. A Note on Two Problems in Connection with
Graphs, Numerische Mathematik. 1959, 1, pp 269271.
12. Tarjan, R. E. Data Structures and Network Algorithms, Society for Industrial and Applied Mathematics, Philadelphia, PA,
1983.
13. Prim, R. C. Shortest Connection Networks and Some Generalizations. BSTJ, 1957, 36, pp 13891401.
14. Dreyfus, S. E. An Appraisal of Some Shortest-Path Algorithms.
Oper. Res., 1969, 17, 395412.
15. Ford, L. R.; Fulkerson, D. R. Flows in Networks; Princeton
University. Press; Princeton, NJ, 1962.
16. Hadley, G. Linear Programming; Addison-Wesley: Reading,
MA, 1962.
17. Dantzig, G. B. Linear Programming and Extensions; Princeton
Univ. Press: Princeton, NJ, 1963.
18. Frank, H.; Frisch, I. T.; Chou, W. Topological Considerations in
the Design of the ARPA Computer Network; Proc. Conf. Rec.,
1970 Spring Joint Computer Conf. AFIPA Conf.; AFIPS Press,
1970: Mantvale, NJ. 1970.
19. Gerla, M.; Kleinrock, L. On the Topological Design of Distributed Computer Networks. IEEE Trans. Commun., 1977,
COM-25, pp 4860.
20. Gerla, M.; J. A. S., Monteiro; Pazos, R. Topology Design and
Bandwidth Allocation in ATM Nets. IEEE JSAC., 1989, 7, pp
12531262.
21. Chattopadhyay, N. G.; Morgan, T. W.; Raghuram, A. An Innovative Technique for Backbone Network Design. IEEE Trans.
SMC 1989, 19 (5):pp 11221132.
22. Somani, A. K. Design of an Efcient Network; Proc. 7th International Parallel Processing Symposium, Newport Beach, CA,
April 1316, 1993, pp. 413418.
23. Feng, T.-Y. A Survey of Interconnection Networks. Computer,
1981, 14, pp 1227.
24. Leiserson, C. Fat-Trees: Universal Network for HardwareEfcient Supercomputing. IEEE Trans. Comput., 1985, C-34,
pp 892901.
25. Sullivan, H.; Bashkow, T. R. A Large Scale, Homogeneous,
Fully Distributed Parallel Machine, I; Proc. 4th Annu. Symp.
Comput. Arch.; 1977, pp 105117.
26. Valiant, L. G. A Scheme for Fast Parallel Communication,
SIAM J. Comput., 1982, 11, pp 350361.
27. Choi, S. B.; Somani, A. K. The generalized hyper-cube; Proc.
ICPP-90;August 1990. pp I/372I/375.
28. Choi, S. B.; Somani, A. K. Rearrangeable Hypercube Architecture for Routing Permutations. J. Parallel Distributed Computing. In Press.
29. Arun K., Somani et al. Proteus System Architecture & Organization, Proc. 5th International Parallel Processing Symposium; 1991, pp 287294.
30
30. Katseff, H. P. Incomplete Hypercube. IEEE Trans. Comput.,

1988, 37, pp 604607.
31. Somani, A. K.; Thatte, S. The Helical Cube Network. Networks,
1995, 26, pp 87100.
32. Gavish, B.; Neuman, I. Routing in a Network with Unreliable Components. IEEE Trans. Commun. 1992, 40, pp 1248
1258.
33. Grover, W. D. The Self-Healing Network: A Fast Distributed
Restoration Technique for Networks Using Digital CrossConnect Machines; Proc. IEEE Globecom;Dec. 1987, pp.
28.2.128.2.6.
34. Allen, A. O. Probability, Statistics, and Queuing Theory; Academic Press: New York, 1978.
35. Kleinrock, L. Queuing Systems Volume 1: Theory, WileyInterscience, New York, 1975.
36. Kleinrock, L. Queuing Systems Volume 1: Computer Application; Wiley-Interscience, New York, 1980.
37. Woodruff, G. M.; Kositpaiboon, R. Multimedia Trafc Principles for Guaranteed ATM Network Performance. IEEE JSAC,
1990, 8, pp 437446.
38. Yeh, Y. S.; Hluchyj, M. G.; Acampora, A. S. The knockout
Switch: A Simple Architecture for High Performance Packet
Switching. IEEE JSAC, 1988, SAC-5, pp 12641273.
39. Yee, J. R.; F. Y. S. Lin A Routing Algorithm for Virtual Circuit Data Networks with Multiple Sessions Per O-D Pair. Networks, 1992, 22, pp 185208.
40. Song, K.; Somani, A. K. Modeling and Design of Dependable
High Speed Information Networks. Int. Assoc. Sci. Technol.
Develop. J. 1997.
41. Nishida, T.; Miyahara, H. Fault Tolerant Packet Switched Network Design Using Capacity Augmentation; Proc. IEEE INFOCOM 88.
42. Wu,T.-H.; Kong, D. T.; Lau, R. C. An Economic Feasibility Study
for a Broadband Virtual Path SONET/ATM Self-Healing Ring
Architecture. IEEE JSAC 1992, 10, pp 14591473.
43. Lazzr, A. A.; Pacici, G.; White, J. S. Real-Time Trafc Measurements on MAGNET II. IEEE JSAC 1990, SAC-8, pp
467483.
44. Mahony, M.; Simeonidou, D.; Hunter, D. Tzanakaki, A., The
Application of Optical Packet Switching in Future Communication Networks, IEEE Commun. Mag., 2001, pp 128
135.
45. Guillemot, C. et al., Transparent Optical Packet Switching:
The European ACTS KEOPS Project Approach, J. Lightw.
Technol., 1998, 16, pp 21172134.
46. Yamada Y., et al., Optical Ouput Buffered ATM Switch Prototype Based on FRONTIERNET Architecture, IEEE J. Select.
Areas Commun., 1998, 16, pp 12981307.
47. Rekhter, Y.; Davie, B.; Katz, D.; Rosen, E.; Swallow, G. Cisco
Systems Tag Switching Architecture Overview; Network Working Group Request for Comments:2105; 1997.
48. Viswanathan, A.; Feldman, N.; Boivie, B.; Woundy, R. ARIS: Aggregrate Route-Based IP Switching; IETF Internet Draft;Mar.
1997.
49. Katsube, Y.; Nagami, K.; Esaki, H. Toshibas Router Architecture Extensions for ATM Overview, Network Design Group Request for Comments: 2098;Feb. 1997.
50. Ahuja, R. K.; Magnanti, T. L.; Orlin, J. B. Network Flows: Theory, Algorithms, and Applications; Prentice Hall, Englewood
Cliffs, NJ, 1993.
51. Clouqueur, M.; Grover, W. D. Quantitative Comparison of
End-to-end Availability of Service Paths in Ring and Mesh-
Restorable Networks; Proceedings of the 19th Annual National

Fiber Optics Engineers Conference (NFOEC 2003); Orlando,
FL, September 2003. 711.
52. Fang, J.; He, W.; Somani, A. K. IP Trafc Grooming in Light
Trail Optical Networks. IEEE J. Select. Areas Commun., In
press.
ARUN K. SOMANI
Iowa State University
RECONFIGURABLE COMPUTING
INTRODUCTION
Although recongurable fabrics can in principle be constructed from any type of technology, in practice, most contemporary designs are made using commercial eld programmable gate arrays (FPGAs). An FPGA is an integrated circuit containing an array of logic gates in which
the connections can be congured by downloading a bitstream to its memory. FPGAs can also be embedded in integrated circuits as intellectual property cores. More detailed surveys on recongurable computing are available
in the literature (25).
Microprocessors offer an easy-to-use, powerful, and exible implementation medium for digital systems. Their
utility in computing applications makes them an overwhelming rst choice, and parallel interconnections of microprocessors can be extremely powerful. Moreover, it is
relatively easy to nd software developers, and microprocessors are widely supported by operating systems, software engineering tools, and libraries. Unfortunately, their
generality does not make them the best choice for a large
class of applications that need to be optimized for performance, power, board area.
Application-specic integrated circuits (ASICs) and FPGAs are able to arrange computations in a spatial rather
than temporal fashion and greater levels of parallelism
than a microprocessor can be achieved. Thus, performance improvements of several orders of magnitude can
be achieved. Also, the absence of caches and instruction decoding can result in the same amount of work being done
with less chip area and lower power consumption (6). As an
example, in a cryptographic key search problem, a single
FPGA with 96 parallel RC4 encryption engines operating
at 50 MHz achieved a speedup of 58 over a 1.5-GHz Pentium 4 implementation (7).
An example involving the implementation of a nite impulse response (FIR) lter is shown in Fig. 1. The recongurable computing solution is signicantly more parallel
than the microprocessor-based one. In addition, it should
be apparent that the recongurable solution avoids the
overheads associated with instruction decoding, caching,
register les, and speculative execution, and unnecessary
data transfers as well as control hardware can be omitted.
Compared with ASICs, FPGAs offer very low nonrecurrent engineering (NRE) costs, which is often a more
important factor than the fact that FPGAs have higher
units costs and many applications do not have the very high
volumes required to make ASICs a cheaper proposition. As
integrated circuit feature sizes continue to decrease, the
NRE costs associated with ASICs continue to escalate, increasing the volume at which it becomes cheaper to use an
ASIC (see Fig. 2). Recongurable computing will be used
in increasingly more applications, as ASICs become only
cost effective for the highest performance or highest volume applications.
Additional benets of recongurable computing are that
its technology provides a shorter time to market than
ASICs (associated FPGA fabrication time is essentially

zero), making many fabrication iterations within a single day possible. This benet allows more complex algorithms to be deployed and makes possible problem-specic
customizations of designs. FPGA-based designs are inherently less risky in terms of technical feasibility and cost, as
shorter design times and lower upfront costs are involved.
As its name suggests, FPGAs also offer the possibility of
modications to the design in the eld, which can be used to
provide bug xes, modications to adapt to changing standards, or to add functionality, all of which can be achieved
by downloading a new bitstream to an existing recongurable computing platform. Reconguration can even take
place while the system is running, this being known as runtime reconguration [e.g., (8)]. Runtime reconguration is
explained in more detail later in this article.
In the next section, we introduce the basic architecture
of common recongurable fabrics, followed by a discussion
of applications of recongurable computing and system architectures. Runtime reconguration and design methods
are then covered. Finally, we discuss multichip systems and
end with a conclusion.
RECONFIGURABLE FABRICS
A block diagram illustrating a generic ne-grained islandstyle FPGA is given in Fig. 3 (9). Products from companies
such as Xilinx (10) Altera (11), and Actel (12) are commercial examples. The FPGA consists of a number of logic cells
that can be interconnected to other logic and input/output
(I/O) cells via programmable routing resources. Logic cells
and routing resources are congured via bit-level programming data, which is stored in memory cells in the FPGA.
A logic cell consists of user-programmable combinatorial
elements, with an optional register at the output. They are
often implemented as lookup tables (LUTs) with a small
number of inputs, 4-input LUTs being shown in Fig. 3. Using such an architecture, subject to FPGA-imposed limitations on the circuits speed and density, an arbitrary circuit
can be implemented. The complete design is described via
the conguration bitstream which species the logic and
I/O cell functionality, and their interconnection.
Current trends are to incorporate additional embedded
blocks so that designers can integrate entire systems on
a single FPGA device. Apart from density, cost, and board
area benets, this process also improves performance because more specialized logic and routing can be used and
all components are on the same chip. A contemporary
FPGA commonly has features such as carry chains to enable fast addition; wide decoders; tristate buffers; blocks
of on-chip memory and multipliers; embedded microprocessors; programmable I/O standards in the input/output
cells; delay locked loops; phase locked loops for clock deskewing, phase shifting and multiplication; multi-gigabit
transceivers (MGTs); and embedded microprocessors. Embedded microprocessors can be implemented either as soft
cores using the internal FPGA resources or as hardwired
cores.
In addition to the architectural features described, intellectual property (IP) cores, implemented using the logic cell
Recongurable Computing
Figure 1. Illustration of a microprocessor based FIR lter vs. a recongurable computing solution. In the microprocessor, operations are
performed in the ALU sequentially. Furthermore, instruction decoding, caching, speculative execution, control generation and so on are
required. For the recongurable computing approach using an FPGA, spatial composition is used to increase the degree of parallelism.
The FPGA implementation can be further parallelized through pipelining.
Figure 2. Cost of technology vs. volume. The crossover volume for which ASIC technology is cheaper than FPGAs increases as feature
size is reduced because of increased non-recurrent engineering costs.
Figure 3. Architecture of a basic island-style FPGA with four-input logic cells. The logic cells, shown as gray rectangles are connected to
programmable routing resources (shown as wires, dots, and diagonal switch boxes) (source: Reference (9) and (76)).
resources of the FPGA, are available from vendors and can

be incorporated into a design. These IP cores include bus interfaces, networking components, memory interfaces, signal processing functions, microprocessors and so on and
can signicantly reduce development time and effort.
The bit-level organization of the logic and routing resources in island-style FPGAs is extremely exible but
has high implementation overhead as a result. Tradeoffs
exist in the granularity of the logic cells and routing resources. Fine-grained devices have the best exibility; however, coarse-grained elements can trade some exibility for
higher performance and density (13). With modern technologies, the speed of the routing resource is a limiting
factor and trends have been to increase the functionality
of the logic cells (e.g., use logic cells with larger numbers
of inputs to reduce interconnect requirements). For datapath oriented applications such as in digital signal processing, coarse-grained architectures such as Pipewrench (14)
and RaPID (15) employ bus-based routing and word-based
functional units to utilize silicon resources more efciently.
Companies such as M2000 (16) and eASIC (l7) provide
recongurable fabric in the form of an IP block that can
be embedded in an ASIC design. Fully synthesizable embedded cores have also been proposed (18). Such fabrics
enable post-fabrication changes and allow the exibility
and performance benets associated with recongurable
computing to be enjoyed in ASICs. They will be increasingly important in future systems for adding functionality,
xing bugs, monitoring, and debugging after fabrication.
APPLICATIONS
Recongurable computing has found widespread application in the form of custom computing-machines for highenergy physics (19), genome analysis (20), signal processing (21, 22), cryptography (7, 23), nancial engineering (24)
and other domains (25). It is unique in that the exibility of
the fabric allows customization to a degree not feasible in
an ASIC. For example, in an FPGA-based implementation
of RSA cryptography (23), a different hardware modular
multiplier for each prime modulus was employed (i.e., the
modulus was hardwired in the logic equations of the design). Such an approach would not be practical in an ASIC
as the design effort and cost is too high to develop a different chip for different moduli. This led to greatly reduced
hardware and improved performance, the implementation
being an order of magnitude faster than any reported implementation in any technology at the time.
Another important application is logic emulation (26,
27) where recongurable computing is used not only for
simulation acceleration, but also for prototyping of ASICs
and in-circuit emulation. In-circuit emulation allows the
possibility of testing prototypes at full or near-full speed,
allowing more thorough testing of time-dependent applications such as networks. It also removes many of the dependencies between ASIC and rmware development, allowing them to proceed in parallel and hence shortening development time. As an example, it was used in Reference (28)
for the development of a two-million-gate ASIC containing
an IEEE 802.11 medium access controller and IEEE 802.1
la/b/g physical layer processor. Using a recongurable prototype of the ASIC on a commodity FPGA board, the ASIC
went through one complete pass of real-time beta testing
before tape-out.
Digital logic, of course, maps extremely well to negrained FPGA devices. The main design issues for such
systems lie in partitioning of a design among multiple
FPGAs and dealing with the interconnect bottleneck between chips. The Cadence Palladium II emulator (29) is a
commercial example of a logic emulation system and has
256-million-gate logic capacity and 74-GB memory capacity. It uses custom ASICs optimized for logic emulation
and is 10010,000 times faster than software-based register transfer language simulation. Further discussion of
interconnect time-multiplexing and system decomposition
is given later in this article.
Hoang (20) implemented algorithms to nd minimum
edit distances for protein and DNA sequences on the
Splash 2 architecture. Splash 2 can be modeled in terms
of both bidirectional and unidirectional systolic arrays. In
the bidirectional algorithm, the source character stream
is fed to the leftmost processing element (PE), whereas
the target stream is fed to the rightmost PE. Comparing two sequences of length m and n requires at least
2 max(m + 1, n + 1) processors, and the number of steps
required to compute the edit distance is proportional to the
size of the array. The unidirectional algorithm is suited for
comparing a single source sequence against multiple target sequences. The source sequence is rst loaded as in the
bidirectional case, and the target sequences are fed in one
after the other and processed as they pass through the PEs
(which results in virtually 100% utilization of processors,
so that the unidirectional model is better suited for large
database searches).
The BEE2 system (22), described in the next section, was
applied to the radio astronomy signal processing domain,
which included development of a billion-channel spectrometer, a 1024-channel polyphase lter banks, and a twoinput, 1024-channel correlator. The FPGA-based system
used a 130-nm technology FPGA and performance was
compared with 130-and 90-nm DSP chips as well as a 90nm microprocessor. Performance in terms of computational
throughput per chip was found to be a factor of 10 to 34 over
the DSP chip in 130-nm technology and 4 to 13 times better than the microprocessor. In terms of power efciency,
the FPGA was one order of magnitude better than the DSP
and two orders of magnitude better than the microprocessor. Compute throughput per unit chip cost was 20307%
better than the 90-nm DSP and 50500% better than the
microprocessor.
SYSTEM ARCHITECTURES
Recongurable computing machines are constructed by interconnecting one or more FPGAs. Functionally, we can
view FPGA-based systems as consisting of two components, reprogrammable FPGAs providing logic implementation and eld programmable interconnect chips (FPICs)
providing connectivity among FPGAs. The FPICs, in turn,
could be implemented as ASICs or using FPGAs. Most sys-
tems include other elements, such as microprocessors and

storage, and can be treated as processing elements and
memory that are interconnected. Obviously, the arrangement of these elements affects the system performance and
routability.
The simplest topology involves FPGAs directly connected in a ring, mesh, or other xed pattern. FPGAs serve
as both logic and interconnect, providing direct communication between adjacent devices. Such an architecture
is predicated on locality in the circuit design and further
assumes that the circuit design maps well to the planar
mesh. This architecture ts well for applications with regular local communications (30). However, in general, high
performance is hard to obtain for arbitrary communication
patterns because the architecture only provides direct communications between neighboring FPGAs and two distant
FPGAs may need many other devices as hops to communicate, resulting in long and widely variable delays. Furthermore, FPGAs, when used as interconnects, often result
in poor timing characteristics.
A major change in the architecture of FPGA-based systems was the concept of a partial crossbar interconnect, as
in Realizer (26) and BORG (31). This scheme is common in
logic emulation systems. Interconnection through FPICs
implies that all pairs of FPGAs are neighbors, resulting in
predictable interconnect delays, better timing characteristics, and better overall system performance (32, 33). Figure
4, from Reference (26) depicts a recongurable computing
system designed for logic emulation. Arrays of recongurable processors and FPICs, both implemented using FPGAs, reside on the emulation modules. The user inputs the
emulated design netlist and commands from the workstation. The workstation and control processor personalize the
emulation module, which are used in place of the emulated
chip. Thus, the target system can function properly before
the actual chip is available. Furthermore, testing and design change can be made by modifying software instead of
reworking hardware.
Figure 5 depicts the SPLASH 2 architecture (34). Each
board contains 16 FPGAs, X1 through X16. The blocks M1
through M16 are local memories of the FPGAs. A simplied
36-bit bus crossbar, with no permutation of the bit-lines
within each bus, interconnects the 16 FPGAs. Another 36bit bus connects the FPGAs in a linear systolic fashion. The
local memories are dual ported with one port connecting to
the FPGAs and the other port connecting to the external
bus. It is interesting to note that the crossbar was added to
the SPLASH 2 machine, the original SPLASH 1 machine
only having the linear connections. SPLASH 2 has been
successfully used for custom computing applications such
as search in genetic databases and string matching (20).
Other designs have used a hierarchy of interconnect
schemes, differing in performance. The use of multi-gigabit
transceivers (MGT) available on contemporary FPGAs allows high bandwidth interconnection using commodity
components. An example is the Berkeley Emulation Engine 2 (BEE2) (22), designed for recongurable computing
and illustrated in Fig. 6. Each compute module consists
of ve FPGAs (Xilinx XC2VP70) connected to four double
data rate 2 (DDR2) dual inline memory modules (DIMMs)
with a maximum capacity of 4GB per FPGA. Four FP-
GAs are used for computation and one for control. Each
PPGA has two PowerPC 405 processor cores. A local mesh
connects the computation FPGAs in a 2-D grid using lowvoltage CMOS (LVCMOS) parallel signaling. Off-module
communications are of via 18 (two from the control FPGA
and four from each of the compute FPGAs) Inniband
4X channel-bonded 2.5-Gbps connectors that operate fullduplex, which corresponds to a 180-Gbps off-module fullduplex communication bandwidth. Modules can be interconnected in different topologies including tree, 3-D mesh,
or crossbar. The use of standard interfaces allows standard
network switches such as Inniband and 10-Gigabit Ethernet to be used. Finally, a 100 base-T Ethernet connection
to the control FPGA is present for out-of-band communications, monitoring, and control.
Commercial machines such as the Cray XD1 (35), SRC
SRC-7 (36), and Silicon Graphics RASC blade (37), have
a similar interconnect structure to the BEE2 in that they
are parallel machines employing high performance microprocessors tightly coupled to a relatively small number of
FPGA devices per node. Nodes are interconnected via high
speed switches and for specialized applications, such machines can have orders of magnitude performance improvement over conventional architectures. Switching topologies can be altered via conguration of the switching fabric.
RUNTIME RECONFIGURATION
A recongurable computing system can have its functionality updated during execution, resulting in reduced resource
requirements. A runtime recongurable system partitions
a design temporally so that the entire design does not need
to be resident in the FPGA at any given moment (38, 39).
Conguration and execution can be overlapped to improve
performance in the presence of reconguration latency. Using this technique, designs that are larger than the physical hardware resources can be realized in an efcient manner.
Dharma, a time-sharing FPGA architecture, was proposed that contains a functional block and an interconnect
network (40). The interconnect and the logic can be timeshared. The authors proposed that emulated design topology be levelized in a folded pipeline manner; this topology
simplies the architecture and provides predictable interconnect delay (Fig. 7).
Single context, partially recongurable, and multiple
context architectures have been proposed. In a single
context system, any changes to the functionality of the
FPGA involves reloading the entire bitstream; early FPGAs were of this type. This scheme has the disadvantage of
long reconguration time. Partial reconguration, as supported by the Xilinx Virtex FPGAs (10), allows portions
of the FPGA to be changed via a memory mapped scheme,
whereas the other portions of the FPGA continue functioning. Compared with a single context scheme, area overhead
is associated in providing this feature. Multiple context
architectures, such as NECs Dynamically Recongurable
Processor (DRP) (41), allow a number of complete congurations to be stored in the fabric simultaneously and thus
reconguration can be achieved in a small number of cy-
Figure 4. Example of a logic emulation system. Arrays ofFPGAs and FPICs reside on the emulation modules. The user inputs the
emulated design netlist and commands from the workstation. The workstation and control processor personalize the emulation modules,
which are used in place of the emulated chip.
Figure 5. SPLASH2 architecture. Each board contains 16 FPGAs, XI through XI6. The blocks Ml through Ml6 are local memories of the
FPGAs. A simplied 36-bit bus crossbar, with no permutation of the bit-lines within each bus, interconnects the 16 FPGAs. Another 36-bit
bus connects the FPGAs in daisy-chain fashion. The local memories are dual ported with one port connecting to the FPGAs and the other
port connecting to the external bus.
cles. This architecture has the shortest context switch time,

however, a larger area overhead is associated with implementation of this scheme.
The logical unit of reconguration could be at a number of levels including the application, instruction, task,
block, or sub-block level. An example of application-level
reconguration could simply involve loading a runtimedependent bitstream to support a particular coding standard in a video coding application. The Dynamic Instruction Set Computer (DISC) (42) supported demand-driven
modication of the instruction set through partial reconguration. The commercial Stretch processor (43) combines
recongurable fabric with a processor to support the execution of custom instructions implemented on a recongurable fabric. Furthermore, the fabric can be recongured
at runtime and the design environment is software-centric,
with programming of the processor being in Stretch C.
An operating system for guarantee-based scheduling of
hard real-time tasks has been proposed (44). Under control
of software running on a microprocessor, task circuits can
be scheduled online and placed in a suitable free space in a
hardware task area. Communications between tasks and
I/O are done though a task communication bus, and termination of a task frees the recongurable resources used. It
Figure 6. BEE2 Compute Module block diagram. Compute modules can be interconnected via the Inniband IB4X connectors, either
directly or via a 10-Gigabit Ethernet switch. The 100-Base T Ethernet can be used for control, monitoring, or data archiving.
Figure 7. Dynamic Architecture for FPGA-based systems. The architecture contains a functional block and an interconnect network.
The interconnect and the logic can be time shared. The emulated design topology is levelized in a folded pipeline manner. The levelized
topology simplies the architecture with predictable interconnect delay.
was shown that hardware in the hardware task area can

be shared by tasks and the overheads associated with its
implementation on a partially congurable platform were
acceptably low.
A pipeline stage is often a convenient block-level unit
for reconguration. In incremental pipeline reconguration (45), an application with S pipeline stages can be implemented in an FPGA with fewer than S physical pipeline
stages. This is done by adding one pipeline stage and removing one pipeline stage in each stage of the computation.
Execution and computation can be overlapped.
Runtime reconguration can be done at even lower levels. A crossbar switch which employs runtime reconguration of the FPGAs routing resources has been described
(46). This scheme was able to achieve density, switch up-
date latency and performance higher than possible using

conventional means.
Tools have been developed to support runtime reconguration. For example, JBits (47) is a set of Java classes that
provide an application programming interface to the Xilinx FPGA bitstream. The interface operates on either bitstreams generated by Xilinx design tools or on bitstreams
read back from actual hardware and allows the FPGA logic
and routing resources to be modied.
DESIGN METHODS
Hardware description languages (HDLs) such as the Very
High Speed Integrated Circuit Hardware Description Lan-
guage (VHDL) and Verilog are commonly used to specify

the logic of a recongurable system. Descriptions in these
languages have the advantage of being vendor neutral, so
the same description can be synthesized for different targets such as different FPGA devices, different FPGA vendors, and ASICs. For this reason, these languages are often
the target language for higher level tools that offer higher
levels of abstraction.
Module generators and libraries are commonly deployed
to promote reuse. For example, vendors such as Altera and
Xilinx have parameterized libraries of components that
can be used in a design. These libraries are generated so
that a circuit optimized for the particular application can
be produced. As an example, a parameterized oating point
library might allow the wordlength of the exponent and
mantissa to be specied as well as whether denormalized
numbers are supported. The module generator then generates a netlist or VHDL-based oating point adder that can
be included in a design.
A high level language can be directly mapped to a netlist
or HDL. As an example, Luk and Page described a simple
compilation process (8, 49) from a high level language with
explicit parallel extensions to a register transfer language
(RTL) description. Parallel execution of statements is implemented via parallel processes, and these can communicate via channels through which a single-word message
can be passed. Variables in the user program are mapped
to registers, all expressions are implemented as combinational logic, and multiplexers are used in the case a register has multiple sources. A datapath that matches the
dataow graph of the input source description is generated using this strategy. The clocking scheme employed is
a global, synchronous one, and a convention that each assignment takes exactly one clock cycle is followed. A start
signal is used to feed the clock and to enable each register
that corresponds to a variable, and a nish signal is generated for the assignment in the following clock cycle. To
execute statements sequentially, the start and nish signals of adjacent statements are simply connected together,
creating a one-hot distributed control scheme. Conditional
statements and loops are formed by asserting one of several possible start signals that correspond to alternative
basic blocks in a program. Completion of conditional or loop
constructs and synchronization of parallel blocks are implemented by combining relevant nish signals using the
appropriate combinatorial logic. An example showing the
translation of a simple code fragment to control and datapath is shown in Fig. 8.
Commercial tools that can compile standard programming languages such as Java, C, or C++ [e.g., (49)] are available. Examples include Handel-C from Celoxica (50) and
Catapult C from Mentor Graphics (51). The use of traditional programming languages improves productivity as
low level details are handled by the compiler. This is analogous to C versus assembly language for software development. Another difference with potentially large implication is that, using these tools, software developers can also
design recongurable computing applications. Domainspecic languages such as MATLAB/Simulink (52) offer
even greater improvements in productivity because they
are interactive, include a large library of primitive rou-
tines and toolkits, and have good graphing capabilities.

Indeed, many designs for communications and signal processing are rst prototyped in MATLAB and then converted to other languages for implementation. Tools such
as the MATCH compiler (53) and Xilinx System Generator
can translate a subset of MATLAB/Simulink directly to an
FPGA design.
The availability of embedded operating systems such
as Linux for microprocessors on an FPGA provides a familiar software development environment for programmers, greatly facilitating program development through
the availability of a large range of open-source libraries
as well as high quality development tools. Such tools can
greatly speed up the development time and improve the
quality of embedded systems. Hardware/software codesign
tools such as Alteras Nios II C-to-Hardware acceleration
compiler enable time-critical functions in a C program to be
converted to a hardware accelerator that is tightly coupled
to a microprocessor within the FPGA (54).
Issues developing with the mapping of algorithms to
hardware are more generally discussed by Isshiki and Dai
(55), who focus on the differences between implementing
bit-serial versus bit-parallel modules (e.g., adders and multipliers) on FPGA architectures. Although latency is larger
for bit-serial modules, the reduction in area frequently
makes area-time products signicantly lower for such implementations. More specically, such advantages as the
following can be obtained: 1) For bit-parallel modules, the
I/O pin limitation is a major problem, and the large size
of the module cluster can result in unused space and underutilized logic resources; 2) bit-serial modules are easier
to partition as cell-to-cell connections are sparse and do
not cause I/O problems; and 3) high fanout nets can impair
routability of bit-parallel modules. Leong and Leong (56)
generalized further with a design methodology that can
translate a dataow description with signals of different
wordlengths to a digit serial design.
MULTICHIP SYSTEMS
Special care must be taken in the design of large and multichip recongurable systems. In this section, we describe
some theoretic results relevant to the major architectural
and issues associated with such designs.
Interconnect Organization
A classic Clos network (57) contains three stages: inputs,
intermediate switches, and outputs, as shown in Fig. 9. It
can be used to interconnect pins in a recongurable computing system, and its input and output stages are symmetric. Suppose the rst stage has r n m crossbar switches,
the second stage has m r r switches, and the third stage
has r m n switches, let us denote the network as c(n,m,r).
For any two-pin net interconnect requirement, the network
c(n, m, r) can achieve complete routability if m is not less
than n. The routing method can be described by recursive
operations (58). In the rst iteration, we reduce the network to c(n 1, m 1, r). In the ith iteration, we reduce
the network to c(n i, m i, r). When n i = 1, we have
r 1 (m n + 1) switches in the rst stage, m n + 1 r r
Figure 8. Hardware compilation example. The C program is translated into a datapath (top) and control (bottom). Execution of statements
in the while loop are controlled by s1 and s2; s0 and s3 correspond to the start signals of the statements before and after the while loop.
Figure 9. Clos network. A Clos network contains three stages: inputs, intermediate switches, and outputs. The input and output stages
are symmetric. In the gure, the rst-stage has r n m switches, the second-stage has m r r switches, and the third-stage has r m n
switches.
switches in the second stage, and r(m n + 1) 1 switches

in the third stage. In other words, only one input exists in
each rst-stage switch and one output in each third-stage
switch. In this case, one second-stage r r switch is enough
to route the r inputs of r rst-stage switches to the r outputs
of r third-stage switches, thus completing the interconnect.
The reduction from c(n i, m i, r) to c(n i 1, m i
1, r) can be derived by a maximum matching algorithm.
The matching algorithm selects disjoint signals from different input switches to different output switches. One
second-stage switch is then used to route the selected signals. From Halls theorem, the maximum matching and
routing can always reduce the network to c(n i 1, m

i 1, r).
Conceptually, the routing problem can also be formulated as edge coloring on a bipartite graph G(V1 , V2 , E) (31).
The node sets V1 and V2 represent the switches in the input
and output stages, respectively. An edge in E represents
a two-pin net interconnect requirement between the corresponding input and output switches. In Reference (31),
Chan and Schlag assigned colors to the edges of the bipartite graph. Edges of the same color are bundled into
one group and the corresponding set of nets are routed by
one switch in the second stage. The work of Reference (59)
was then used to nd a minimum edge coloring solution in

O(|E|logn).
The three-stage Clos network can be folded into a twostage network (Fig. 10) so that the inputs and outputs are
mixed in the rst stage. Thus, the corresponding bipartite
graph G(V1 , V2 , E) constructed above for edge coloring is
also folded with V1 and V2 merged into one set.
To nd the routing assignment, the folded edge coloring graph can be unfolded back to a bipartite graph using
an Euler path search. The Euler path traverses every edge
exactly once and denes the edge direction according to
the direction of the traversal. We then recover the original bipartite graph by splitting the node set back into two
sets V1 and V2 and unfold the edges such that all edges
are directed from V1 to V2 . We can nd the minimum edge
coloring solution of the unfolded bipartite graph and apply
the solution back to the folded routing problem.
In practice, the rst-level crossbar of the Clos network
is replaced with FPGAs to save board space (Fig. 11).
Routability is worse than an ideal Clos network. Even with
a true Clos network, complete routability of multipin nets
is not guaranteed, which is an important practical consideration because in microelectronic design, many multipin
nets typically exist.
In an attempt to solve the multipin net and routability
problem, we can introduce extra connections among FPIDs
as shown in Fig. 12. However, extra FPID interconnections
also incur extra delay. We can also expand the fanout width
of FPGAs so that each FPGA I/O pin is connected to more
than one FPIC (60, 61). The fanout width expansion improves routability without signicant additional delay. The
multiple appearances of I/O pins increase the probability
that a signal connection can be made in a single stage,
which is especially critical for multipin nets. However, the
additional fanouts increase the needed pin count of FPICs.
Thus, we need to nd a balanced fanout distribution that
reduces the interconnect delay with a minimal pin requirement.
A tree-structured network can simplify the mapping
process for certain applications. In Reference (62), an example of a tree-structured network is illustrated for a Very
Large Scale Simulator (VLSS). The VLSS tree structure
has all logic components located at the leaves and interconnect switches at the internal nodes. The machine covers a
capacity of eight million gates. Each branch is an 8-bit bus.
The higher up the level of the tree, the less parallelism the
signal distribution can achieve. Therefore, a partitioning
process is designed to minimize the high level interconnect
and maximize the parallel operation.
Interconnect Multiplexing
Time multiplexing is an effective method for tackling the
scalability problem in interconnecting large designs. The
time-sharing method can be extended from traditional bus
organization (27, 62) to network sharing (63) and further
to function block sharing (40).
Interconnect can be time shared as a bus (27, 62). If n
communication lines exist between two FPGAs, they can
be reduced to a single line by storing logical outputs in
shift registers and time-multiplexing the communication
in phases. Such a scheme was employed in the virtual wires

logic emulation system (27), which is efcient because interconnects are normally capable of being clocked at much
higher rates than the critical path of the rest of the system,
and all logical wires are not simultaneously active. This
scheme can greatly reduce the complexity of the interconnecting network or printed circuit board in a multi-FPGA
system.
Li and Cheng (63) proposed that a dynamic network be
viewed as overlapping L conventional FPICs together but
sharing the same I/O pins. A dynamic routing architecture can increase the routability and shorten interconnect
length. Each switching network is a full crossbar, which can
be recongured to provide any connections among I/O pins.
The select lines are used to activate only one switching network at a time; thus the I/O pins are dynamically connected
according to the conguration of this active switching network. By dynamically reconguring the FPICs, L logic signals can time-share the same interconnect resources.
Memory Allocation
Interconnect schemes should also consider how memory
is connected to the FPGAs. Although combining memory
with logic in the same FPGA is the most desirable method
for reducing routing congestion and signal delay, separate
components can supply much larger capacity at higher density and lower price. Figure 13 demonstrates three different ways of allocating the memories in a Clos network (31,
64). The memory may be attached directly to a local FPGA
(Fig. 13a), attached to the second-stage switches of the Clos
network via a host interface (Fig. 13b), or attached to the
rst-stage switches of the Clos network (Fig. 13c). The rst
method provides good performance for local memory access. However, for the case of nonlocal memory access, the
routability and delay are concerns. The second method is
slower than the rst method for local memory accesses but
provides better routability. The third is the most exible as
the memory is attached to the network and the routability is high. However, every logic-to-memory communication
must go through the second interconnect stage.
Bus Buffer Insertion
In FPGAs, signal propagation is inherently slow because of
its programmable interconnect feature. However, the delay
of long routing wires can be drastically reduced by buffer
insertion. The principle at work is that by inserting buffers
we can decouple capacitive effects of components and interconnect driven by the buffers and thereby improve RC
delay.
Given a routing topology for a net and timing requirements for its sinks, an efcient optimal buffer insertion
algorithm was proposed in Reference 65. Experimental results show dramatic improvement versus the unbuffered
solution. Thus, it is advantageous to have abundant buffers
in FPGAs. However, each possible buffer and its programmable switch adds capacitance to the wires, which in
turn will contribute to delay. Thus, a balance point needs to
be identied to trade off between the additional delay and
capacitance of the buffers versus the improvement they
can provide.
10
Figure 10. Folded Clos network. The three-stage Clos network is folded into a two-stage network so that the inputs and outputs are
mixed in the rst stage.
Figure 11. Variations of the Clos network. The rst level crossbar of the Clos network is replaced with FPGAs to save board space.
Routability is worse than an ideal Clos network.
11
Figure 12. Variations of Clos network. The fanout width of FPGAs is expanded so that each FPGA I/O pin is connected to more than one
FPIC. The fanout width expansion improves routability without signicant additional delay.
For a multisourced bus, the problem of buffer insertion

becomes more complicated, because the optimization for
one source may sacrice the delay of others. Furthermore,
the direction of the buffer needs to be arbitrated by a controller. Instead of using such a controller, a novel approach
is to use a patented open collector bus repeater (66). When
idle, the two ends of the repeater are set to high. When
the repeater senses the pull-down action on one side, it
presents the signal on the other side until the pull-down
action is released from the originated signal. The bus repeater eliminates the need for a direction control signal,
resulting in a simpler design and better use of resources.
System Decomposition
To decompose a system into multiple devices, Yeh et al. (67)
proposed an algorithm based on the relationship between
uniform multi-commodity ow and min-cut partitioning.
Yeh et al. construct a ow network wherein each net initially corresponded an edge with ow cost one. Two random
modules in the network were chosen and the shortest path
(i.e., path with lowest cost) between them was computed.
A constant <1 was added to the ow for each net in the
shortest path, and the cost for every net in the path was
incremented. Adjusting the cost penalizes paths through
congested areas and forces alternative shortest paths. This
random shortest path computation is repeated until every
path between the chosen pair of modules passes through at
least one saturated net. The set of saturated nets induces
a multi-way partitioning in which two modules belong to
the same cluster if and only if there is a path of unsaturated
nets between them.
For each of these clusters, the ux (dened as the cutsize between the cluster and its complement, divided by
the size of the cluster) is computed and the clusters are
sorted based on their ux value. Yeh et al. began with a
single cluster equal to the entire netlist, and then peeled
off the clusters with lowest ux. This approach was attractive because the saturated nets are good candidates to be
cut in a partitioning solution. As peeled clusters can be very
small, a second phase may be used to make the multi-way
partitioning more balanced. This approach, with its subsequent speedup by Yeh (68), is well-suited for large-scale
multi-way partitioning instances.
The system prototyping phase may also explore netlist

transformations such as logic replication and retiming to
minimize cut size (I/O usage) or system cycle time. Such
transformations are needed as inter-device delays can be
relatively large and because devices are often I/O-limited.
In Reference (69), Liu et al. proposed a partitioning algorithm that permits logic replication to minimize both cut
size and clock cycle of sequential circuits. Given a netlist
G = (V, E), their approach chooses two modules as seeds s
and t, then constructs a replication graph that is twice
the size of the original circuit. This graph has the special property that a type of directed minimum cut yields
the replication cut (i.e., a decomposition of V into S, T,
and R = V S T where s S, t T and R is the replicated
logic) that is optimal. A directed version of the FiducciaMattheyses algorithm is used to nd a heuristic directed
minimum cut in the replication graph. Cong et al. (70)
present an efcient algorithm for the performance-driven
multi-way circuit partitioning problem that considers the
different local and global interconnect delay introduced by
the partitioning.
Alpert and Kahng (71) survey the FPGA partitioning literature in the context of major graph partitioning
paradigms. The current partitioning problems are 1) low
usage rate of FPGA gate capacity because I/O pin limit, 2)
low clock rate because of interconnect delay between multiple FPGAs and 3) long CPU time for the mapping process.
System Planning and Design Changes
For a given system decomposition to be implemented on
a multi-FPGA prototyping architecture, all connections
within each device and between devices must be routable.
Chan et al. (72) invoke much literature on routability prediction in gate arrays, as well as theoretical concepts, such
as the Rent parameter, to obtain a fast routability estimate for arbitrary netlists and FPGA architectures. Their
method ascribes one of three levels of routable (easily
routable, marginally routable, or unroutable) to a netlist
based on various parameters. Specically, combining a
wirelength estimator due to Feuer, the average number
of pins-per-cell, and the estimated Rent parameter yields
a relatively accurate routability predictor. The utility of
these parameters is contrasted with that of other criteria
such as El Gamals channel width requirement (73) or the
12
Figure 13. Memory organization, (a) Memory is attached directly to a local FPGA. (b) Memory is attached to the second-stage switches
of the Clos network via a host interface, (c) Memory is attached to the rst-stage switches of the Clos network.
average pins-per-net ratio.

In addition to routability, connections must also
meet system timing constraints. Selvidge et al. (74) extend the original virtual wires (27) concept in their
TIERS (Topology-IndEpendent Routing and Scheduling)
approach. The problem formulation assumes that an assignment from a multiple-FPGA partitioning (i.e., a design
graph) to a target topology graph has already been made.
The objective is to assign links (i.e., signal nets) to channels between devices; as with the Virtual Wires concept,
specic timeslices for a channel can be assigned to multiple links as long as no two links need to transmit signals at
the same time. The TIERS algorithm uses a greedy method
to order the links and then routes each link in the scheduled order while reserving channel resources; factors of up
to 2.5 improvement in system cycle time are achieved.
Chang, et al. (75) address the combined issues of
routability and system timing by applying layout-driven
logic resynthesis techniques. For a given wire that cannot
be routed, alternative wires and alternative functions are
identied, such that the given unroutable wire can be removed from the circuit and replaced with a new wire (or
wires) or new logic without affecting functionality. Cheng
et al. estimate that between 30% and 50% of wires have socalled triple-wire alternatives (i.e., replacements consisting of three or fewer wires). Their method rst routes the
wires that do not have any alternatives then replaces any
unroutable wire with available alternatives. System timing can be improved by replacing long wires with shorter
alternatives.
CONCLUSION
Recongurable computing offers a middle ground between
software-based systems and ASIC implementations, and
is often able to combine important benets of both. Implementations are able to avoid overheads such as unnecessary data transfers, decoding and control mandatory in
microprocessors, and designs can be optimized on a basis
specic to an application, a problem instance or even an
execution. Using this technology, it is possible to achieve
size, performance, cost, or power improvements over more
conventional computing technologies.
ACKNOWLEDGMENTS
The authors would like to thank Y M. Lam for his help in
preparing this manuscript and Prof. Wayne Luk (Imperial
College) for his proofreading of this article.
BIBLIOGRAPHY
1. Estrin, G. Recongurable Computer Origins: The UCLA
Fixed-plus-variable (F+V) Structure computer. IEEE Ann.
Hist. Comput. 2002, 24 (4), pp 39.
2. Compton, K.; Hauck, S. The roles of FPGAs in Reprogrammable Systems. Proc. IEEE 1998, 86, (4), pp 615639.
13
3. Bondalapati, K.; Prasanna, V. Recongurable Computing System. Proc. IEEE 2002, 90 (7), pp 12011217.
4. Compton, K.; Hauck, S. Recongurable Computing: A Survey
of System and Software. ACM Comput.Surv. 2002, 34 (2), pp
171210.
5. Todman, T.; Constantinides, G.; Wilton, S.; Mencer, O.; Luk,
W.; Cheung, P. Recongurable Computing: architectures and
design methods. IEE Proc. Comput. Digit. Tech. 2000, 152, (2),
pp 193205.
6. DeHon, A. The Density Advantage of Congurable Computing.
IEEE. Computer 2000, 33 (4), pp 4149.
7. Tsoi, K. H.; Lee, K. H.; Leong, P.H.W. A Massively Parallel RC4
Key Search Engine; Proc. of the 10th Annual IEEE Symposium
on Field-Programmable Custom Computing Machines FCCM;
Washington, DC, IEEE Computer Society, 2002, pp 1321.
8. Liang, J.; Tessier, R.; Goecket, D. A Dynamicallyrecongurable, Power-efcient Turbo Decoder; 2004, pp
91100.
9. Betz, V.; Rose, J.; Marquardt, A. Eds. Architecture and CAD
for Deep-Submicron FPGAs; Kluwer Academic Publisher: Dordrecht, the Netherlands, 1999.
10. Xilinx.http://www.xilinx.com (accessed 2006).
11. Altera.http://www.altera.com (accessed 2006).
12. Actel.http://www.actel.com (accessed 2006).
13. Ahmed, E.; Rose, J. The Effect of LUT and Cluster Size on
Deep-submicron FPGA Performance and Density Proc. of the
2000 ACM/SIGDA Eighth International Symposium on Field
Programmable Gate Arrays ACM Press: New York, 2000; pp
312.
14. Goldstein, S. C.; Schmit, H.; Budiu, M.; Cadambi, S.; Moe, M.;
Taylor, R. R. Piperench: A Recongurable Architecture and
Compiler. Computer 2000, 33 (4), pp 7077.
15. Ebeling, C.; Cronquist, D. C.; Franklin, P. Rapid - Recongurable Pipelined Datapath; Proc. of the 6th International
Workshop on Field-Programmable Logic, Smart Applications,
New Paradigms and Compilers; London, UK, Springer-Verlag:
l996, pp 126135.
16. M2000. FPC06: http://www.m2000.fr (accessed 2006).
17. eASIC. http://www.easic.com (accessed 2006).
18. Wilton, S.; Kafa, N.; Wu, J.; Bozman, K.; AkenOven, V.; Saleh,
R. Design Considerations for Soft Embedded Programmable
Logic Core. IEEE Solid Circuits 2005, 40 (2), pp 485497
19. Moll, L.; Vuillemn, J.; Boucard, P.; High-energy Physics on
DECPeRLe-1 Programmable Active Memory; Proc. of the 1995
ACM Third International Symposium on Field-programmable
Gate Arrays; New York, (FPGA 95), ACM Press: 1995, pp
4752.
20. Hoang, D. T. Searching Genetic Database on Splash, IEEE
Workshop on FPGAs for Custom Computing Machines, Napa,
CA, April 1993, pp 185191.
21. Ting, L.-K.; Woods, R.; Cowan, C. F. N. Virtex FPGA Implementation of a Pipelined Adaptive LMS Predictor for Electronic
Support Measure Receivers. IEEE Trans. VLSI Syst. 2005, 13
(1), pp 8695.
22. Chang, C.; Wawrzynek, J.; Brodersen, R. W. BEE2: A high-end
Recongurable Computing System. IEEE Des. Test 22 (2), pp
114125.
23. Shand, M.; Vuillemin, J.; Fast Implementations of RSA Cryptography: Proc. 11th Symposium on Computer Arithmetic,
1993, pp 252259.
24. Zhang, G. L.; Leong, P. H. W.; Ho, C. H.; Tsoi, K. H.; Cheung,
C. C. C.; Lee, D. -U.; Cheung, R. C. C.; Luk, W. Recongurable
14
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
Acceleration for Monte Carlo based nancial simulation: Proc.
International Conference on Field Programmable Technology
(ICFPT). 2005, pp 215222.
Vuillemin, J.; Patrice, B.; Didier, R.; Shand, M.; Herve, T.;
Philippe, B. Programmable Active Memories: Recongurable
System Come of Age. IEEE Trans. VLSI Syst. 1996, 4 (1), pp
5659.
Butts, M.; Batcheller, J.; Varghese, J.; An Efcient Logic Emulation System. IEEE Trans. Very Large Scale Integr. (VLSI)
Syst. 1993, pp 171173.
Babb, J.; Tessier, R.; Dahl, M.; Hanono, S. Z.; Hoki, D. M.;
Agarwal, A. Logic Emulation with Virtual Wires. IEEE Trans.
Computer-Aided Design 1997, 16 (6), p 609.
de Souza, L.; Ryan, P.; Crawford, J.; Wong, K.; Zyner, G.; McDermott, T. Prototyping for the Concurrent Development of
an IEEE 802.11 Wireless LAN Chaipset: Proc. International
Conference on Field-Programmable Logic and its Applications
LNCS 2778, Springer, 2003, pp 5160.
Cadence.
http://www.cadence.com/datasheets/
lncisivePalladiumllds.pdf (accessed 2006). Palladium Data
Sheet, 2005, pp 18.
Bertin, P.; Roncin, D.; Vuillemin, J. Introduction to Programmable Active Memories. DEC Memo 3 1989, pp 19.
Chan, P. K.; Schlag, M. D. F. Architectural Tradeoffs in Fieldprogrammable Devices Based Computing System; EEE Workshop on FPGAs for Custom Computing Machines; l993, pp
152161
Mohsen, A. Programmable Interconnects Speed System Verication. IEEE Circuits Devices Mag, 1993, 9 (3), pp 37
42.
Slimane-Kadi, M.; Brasen, D.; Saucier, G. A Fast-FPGA Prototyping System That Uses Inexpensive High-performance
FPIC. ACM Int. Workshop on FPGAs, Berkeley, CA, 1994, pp
1.3. 111.
Arnold, J. M.; Buell, D. A.; Davis, E. G. SPLASH 2; 4th Annual ACM Symposium on Parallel Algorithms and Architectures (SPAA 9Z); 1992, pp 316322.
Cray. http://www.cray.com/downloads/Cray XDl Datasheet.pdf
(accessed 2006). Cray XDl Datasheet.
SRC Computers. http://www.srccomp.com (accessed 2006).
SGI. http://www.sgi.com/products/rasc/ (accessed 2006).
Villasenor, J.; Mangione-Smith, W. H. Congurable Computing. Scientif. Amer. 1997, pp 6771.
Becker, J.; Hubner, M. Run-time Recongurability and Other
Future Trendy; Proc. of the 19th Annual Symposium on Integrated Circuits and Systems Design (SBCCI 06); ACM Press:
New York, 2006, pp 911.
Bhat, N. B.; Chaudhary, K.; Kuh, E. S. Performanceoriented Fully Routable Dynamic Architecture for a Fieldprogrammable Logic Device. Memorandum No. UCB/ERL
M93/42, Electronics Research Lab., College of Engineering,
UC Berkeley, 1993, pp. 121.
NEC Electronics. http://www.necel.com/drp/in/index.html (accessed 2006).
Wirthlin, M. J.; Hutchings, B. L. A Dynamic Instruction Set
Computing; Proc. of the IEEE Symposium on FPGAs for Computing Machines Washington, DC, IEEE Computer Society,
1995, pp 99107.
Stretch, Inc. http://www.stretchinc.com (accessed 2006).
Steiger, C.; Walder, H.; Platzner, M. Operting System for Recongurable Embedded Platforms: Online Scheduling of Realtime Task. IEEE Trans. Comput. 2004, 53 (11), pp 13931407.
45. Schmit, H. Incremental Reconguration for Pipelined Applications; Proc. of the 5th IEEE Symposium on FPGA-Based Custom Computing Machines Washington, DC, IEEE Computer
Society, 1997, pp 4755.
46. Young, S.; Alfke, P.; Fewer, C.; McMillan, S.; Blodget, B.; Levi,
D. A High I/O Recongurable Crossbar Switch; Proc. of the
11th Annual IEEE Symposium on Field-Programmable Custom Computing Machines (FCCM 03); Washington, DC, IEEE
Computer Society, 2003, pp 310.
47. Guccione, S.; Levi, D.; Sundarajan, P. A Java-based Interface for Recongurable Computing; Second Annual Military and Aerospace Applications of Programmable Devices
and Technologies Conference (MAPLD); September 1999. citeseer.ist.psu.edu/681383.html.
48. Luk, W.; Page, I. Compiling Occam into FPGAs. FPGAs. Abingdon, EE &CS books: 1991; pp 271283.
49. Page, I. Constructing Hardware/Software System From a Single Description. VLSI Signal Processing, 1996, 12,pp 87
107.
50. Celoxica. http://www.celoxica.com (accessed 2006).
51. Mentor Graphics. http://www.mentor.com/products/esl/high
level synthesis/catapult synthesis/index.cfm
(accessed
2006).
52. Mathworks, Inc. http://www.mathworks.com (accessed 2006).
53. Haldar, M.; Nayak, A.; Choudhary, A.; Banerjee, P. A System
for Synthesizing Optimized FPGA Hardware from MATLAB;
Proc. of the 2001 IEEE/ACM International Conference on
Computer-aided Design (ICCAD) Piscataway, NJ, IEEE Press:
2001, pp 314319.
54. Lau, D.; Pritchard, O.; Molson, P. Automated Generation
of Hardware Accelerators with Direct Memory Access from
ANSI/ISO Standard C Functions, Proc. of the 14th Annual
IEEE symposium on Field-Programmable Custom Computing
Machines (FCCM06), Washington, DC, IEEE Computer Society, 2006, pp 4556.
55. Isshiki, T.; Dai, W. M. High-level Bit-serial Datapath Synthesis
for Multi-FPGA System; International Workshop on FPGAs;
1995, pp 167174.
56. Leong, M. P.; Leong, P. H. W. A Variable-Radix Digit-serial
Design Methodology and Application to the Discrete Cosine
Transform. IEEE Trans. VLSI System 2003, 11 (1), pp 90
104.
57. Clos, C. A Study of Nonblocking Switching Networks. Bell System Tech. J. 1953, 32,pp 406424.
58. Benes, V.E. Mathematical Theory of Connecting Networks and
Telephone Trafc; Academic Press: New York, 1965.
59. Cole, R.; Hopcroft, J. On Edge Coloring Bipartite Graphs.
SIAM J.Computing 1982, 11,pp 540546.
60. Richards, G. W.; Hwang, F. K.A Two-stage Rearrangeable
Broadcast Switching Network. IEEE Trans. Commun. 1985,
COM- 33(10), pp 10251035.
61. I-Cube, Using FPID Devies in FPGA-based Prototyping. Application Note 1994, pp 111.
62. Wei, Y. C.; Cheng, C. K.; Wurman, Z. Multiple-level Partitioning: An Application to the Very Large-Scale HardWare Simulator. IEEE J. Solid State Circuits 1991, pp 706716.
63. Li, J.; Cheng, C. K.. Routability Improvement Using Dynamic
Interconnect Architecture. IEEE FPGAs for Custom Computing Machines 1995, Apr, pp 13.2.7.
64. Chan, P. K.; Schlag, M. D. F.; Martin, M. BORG: A Recongurable Prototyping Board Using Field-programmable Gate Arrays. Int. Workshop on FPGA; Berkeley CA, 1992, pp 4751.
65. Lillis, J.; Cheng, C. K.; Lin, T. T. Optimal Wire Sizing and Buffer
Insertion for Low Power and a Generalized Delay Model.
IEEE/ACM Int. Conf. on Computer-Aided Design; 1995, pp
138143.
66. Hsieh, W. J.; Jenq, Y. C.; Horng, C. S.; Lofstrom, K. Input/output
I/O Bidirectional Buffer for Interfacing I/O Parts of a Field
Programmable Interconnection Device with Array Ports of a
Cross-point Switch. US Patent no 5,428,800, 1992, pp 113.
67. Yeh, C. W.; Cheng, C. K.; Lin, T.T. A Probabilistic
Multicommodity-ow Solution to Circuit Clustering Problems. IEEE Int. Conf. on Computer-Aided Design; 1992, pp
428431.
68. Yeh, C. W. On the Acceleration of Flow-oriented Circuit Clustering. IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst.
1995, pp 13051308.
69. Liu, L. T.; Kuo, M. T.; Cheng, C. K.; Hu, T.C. Performancedriven Partitioning Using a Replication Graph Approach.
ACM/IEEE Design Automation Conf; June 1995, pp 206
210.
70. Cong, J.; Lim, S. K.; Wu, C. Performance driven Multi-level
and Multiway Partitioning with Retiming; Proc. of the 37th
Conference on Design Automation (DAC 00); New York, ACM
press: 2000, pp 274279.
71. Alpert, C. J.; Kahng, A.B. Recent Directions in Netlist Partitioning: A Survey. Integration, The VLSI. 1995, August,pp
181.
72. Chan, P. K.; Schlag, M. D. F.; Zien, J. Y. On routability Prediction for Field-programmable Gate Arrays; IEEE Design Automation Conf.; Dallas, 1993, pp 326330.
73. El Gamal, A. Two-dimensional Stochastic Model for Interconnections in Master Slice Integrated Circuits. IEEE Trans.
CAS. 1981, 28 (2), pp 127138.
74. Selvide, C.; Agarwal, A.; Dahl, M.; Babb, J. TIERS: Topology
Independent Pipelined Routing and Scheduling; Int. Symp. on
FPGA; 1995, pp 2531.
75. Chang, S. C.; Cheng, K. T.; Woo, N. S.; Marek-Sadowska, M.
Layout Driven Logic Synthesis for FPGA; Proc. ACM/IEEE
Design Automation Conference. 1994, pp 308313.
76. Leong, M. P. FPGA Design Methodologies for High Performance Applications. PhD dissertation, The Chinese University of Hong Kong, 2001.
CHENG CHUNG-KUAN
ANDREW B. KAHNG
PHILIP H.W. LEONG
Dept. of Computer Science and
Engineering, University of
California, La Jolla,
California
Dept. of Computer Science and
Engineering, The Chinese
University of Hong Kong,
Shatin, Hong Kong
15
SQUIDs
311
SQUIDs
Since the development of the Superconducting QUantum Interference Device (SQUID) in the late 1960s and its commercial introduction in 1970, SQUID-based instruments have
proved to be the most sensitive measurement devices not only
for magnetization measurements but also for a number of
other electrical measurements. Their device noise (well below
1 mK), frequency response to dc, and low drift permit electromagnetic measurements at levels far below those of conventional techniques.
SQUID instruments consist of a SQUID amplifier or sensor and a detection circuit that transforms the signal of interJ. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
SQUIDs
est into a magnetic flux that is detected by the SQUID sensor.

Associated control electronics transform this signal into a
room temperature voltage that is available for additional signal processing if needed. The SQUID amplifier and the detection coils are superconducting devices. Thus some type of refrigerant (e.g., liquid helium or nitrogen) or refrigeration
device is needed to maintain the SQUID (and detection coil)
in the superconducting state. Additional signal-conditioning
electronics may be needed to improve the signal-to-noise
ratio.
THE JOSEPHSON EFFECT
SQUIDs combine two phenomenaflux quantization where
the flux B A penetrating a superconducting loop is quantized in steps of 1 0 h/2e 2.068 1015 Wb, and the
Josephson effect (1) (electrons tunneling from one superconducting region to another separated by a resistive barrier).
Currents smaller than a critical current IC, can penetrate the
barrier (usually called a weak link) with no voltage drop
(Fig. 1).
A typical weak link might have a critical current of 10 A.
If the loop has a diameter of 2 mm, this is equivalent to several flux quanta. In a superconductor loop interrupted by a
weak link Josephson junction, magnetic flux threading
through a superconducting loop sets up a current in the loop.
As long as the current is below the critical current, the complete loop behaves as if it were superconducting. Any changes
in the magnetic flux threading through the loop induce a
shielding current that generates a small magnetic field to oppose the change in magnetic flux. The weak link can be a
region in which the current flowing is greater than the current needed to drive the superconductor normal IC. Details on
the Josephson effect and the theory of SQUIDs can be found
in References 2, 3, and 4.
SQUID OPERATION
SQUIDs are operated as either RF (radio frequency) or dc
SQUIDs. The prefix RF or dc refers to whether the Josephson
junction(s) is biased with an alternating current (RF) or a dc
current.
The RF SQUID
For the RF SQUID (5), flux is normally (inductively) coupled
into a SQUID loop containing a single Josephson junction via
Feedback current
RF
coil
Input
RF
amp Det
Cryogenic region
RF
oscillator
Phase
sensitive
detector
Output
RF
oscillator
Figure 2. Block diagram of SQUID input and electronics for lockedloop operation.
an input coil (which connects the SQUID to the experiment)

and an RF coil that is part of a high-Q resonant circuit to
read out current changes due to induced flux in the SQUID
loop (Fig. 2).
This tuned circuit is driven by a constant current RF oscillator that is weakly coupled to the SQUID loop. The detected
RF output is found to be the periodic function (Fig. 3).
One way to measure the change in input coil current is to
simply count the number of periods it produces in the detected RF output. A more commonly used mode of operation
is a feedback scheme (Fig. 2), which locks in on either a peak
or a valley in the triangle pattern output from the RF peak
detector. A feedback flux is applied to the SQUID through the
RF coil that just cancels the change in flux from the input coil.
The dc SQUID
The dc SQUID (Fig. 4) differs from the RF SQUID in the manner of biasing the Josephson junction (dc rather than ac) and
the number of junctions (two rather than one).
The dc SQUID is biased with a dc current approximately
equal to twice IC and develops a dc voltage across the junctions. A change in the flux penetrating the SQUID loop enhances the current through one Josephson junction and reduces the current through the other, driving one junction
normally and the other superconducting. This asymmetry,
which is periodic in 0, is used to provide a feedback current
that nulls the flux penetrating the SQUID loop. Although total flux within the SQUID loop is in multiples of 0, by measuring the voltage drop across the feedback resistor, resolutions of external flux changes at the 106 level can be
achieved. The linearity of flux-locked loop SQUID systems are
V
S N S
Ic
+Ic
Figure 1. IV curve of a Josephson tunnel junction.
Detected RF output
312
1 0
Input flux, input current

Figure 3. Triangle pattern showing detected output voltage as a
function of flux in the SQUID.
SQUIDs
dc current
source
313
SQUID systems offer user selectable slew rates along with

high-pass and low-pass filters for noise reduction.
Feedback current
Sensitivity
ac
amp
Input
Sync
detector
Because of the varying input impedances of SQUID sensors,

the sensitivity of SQUID devices is best discussed in terms of
the energy sensitivity:
Output
Ref
Modulation
oscillator
Cryogenic region
Figure 4. Block diagram of a typical dc SQUID.
typically better than 1 ppm. Like the RF SQUID, this feedback current (presented as a voltage at the output) is a direct
measure of changes in flux applied to the SQUID.
Control Electronics. The system output voltage is the voltage drop across the feedback resistor in a negative feedback
loop controlled by the SQUID electronics. The feedback signal
is generated in response to changes in the output signal of the
SQUID sensor. The output of the SQUID sensor is periodic
in the field coupled into the SQUID loop. Negative feedback
(similar to a phase-locked loop technique) is used to maintain
the system operating point at a particular (and arbitrary) flux
quantum. When operated in this mode, the system is in a
flux-locked loop.
One important factor of SQUID design is such that the
feedback electronics be able to follow changes in the shielding
currents. If the shielding current changes so fast that the flux
in the SQUID loop changes by more than 0 /2, it is possible
that the feedback electronics will lag behind the rapidly
changing flux. When the electronics finally catch up, they
can lock on an operating point (Fig. 3) different from the original. In this case, the SQUID has lost lock because the
SQUID has exceeded the maximum slew rate of the electronics. This places an upper limit on the bandwidth of the system. The typical bandwidth of commercially available SQUID
systems is dc to 50 kHz. Custom electronics have been built
extending bandwidths above 5 MHz. Typical slew rates for
SQUIDs are in the range of 105 106 0 /s.
Even though one may not need or want to observe rapidly
changing signals, situations may arise when ambient noise
(e.g., 60 Hz) may determine the slew rate requirements of the
system. To recover a signal from such interference, the system must be able to track all signals present at the input,
including the noise. When system response is sped up to handle very fast signals, sensitivity to RF interference and spurious transients is also increased. Because the ability to remain
locked while subjected to strong electrical transients is greatest when the maximum slew rate is limited (slow), whereas
ability to track rapidly varying signals is greatest when the
maximum slew rate is greatest (fast), it is desirable to be able
to match the maximum slew-rate capability to the measuring
situation. As a matter of convenience, many commercial
2
EN = Li IN
=
2
N
Li
(1)
where Li is the input inductance of the device, IN is the current noise, and N is the flux sensitivity. EN is often expressed
in terms of Plancks constant h 6.6 1034 J/Hz.
The major limiting factor in the noise of a RF SQUID is
the bias frequency f 0 used to excite the tank circuit and that
RF SQUID noise is proportional to 1/ f 0 (6). As f 0 increases,
the complexity of the electronics also tends to increase.
The minimum noise energy for a dc SQUID is given by (7)
EN = kB T LloopC
(2)
where kB is Boltzmanns constant, Ll is the inductance of the
SQUID loop, and C is the capacitance of the junction. Substituting appropriate numbers indicates that the minimum
noise energy EN for a dc SQUID is on the order of h/2. Devices
with sensitivities of h have been constructed. These extremely low noise levels are achieved by limiting dynamic
range and avoiding feedback. The need for practical (useful)
devices requires that feedback be used and that the SQUID
have a reasonable dynamic range. Commercially available RF
SQUIDs have noise levels of 1029 J/Hz; commercial dc
SQUIDs are typically 1031 J/Hz.
In addition to the frequency independent (white) component of system noise, there exists a low-frequency contribution that increases as the frequency decreases. The onset of
this 1/f noise can be dependent on the ambient magnetic field
when the SQUID sensor is cooled. When cooled in the earths
magnetic field, the point at which the 1/f noise equals the
white (frequency independent) noise is typically 1 Hz. Cooling the SQUID sensor in low ambient magnetic fields (less
than 1 T) may improve the 1/f performance by as much as
an order of magnitude. A large contribution to this noise in
some dc SQUIDs can arise from the presence of the dc current
bias. By chopping the dc bias in combination with the conventional flux modulation techniques, it is possible to reduce this
added 1/f noise. This ac bias reversal approach (8) separates
the original signal waveform from the noise associated with
the dc bias and can reduce 1/f noise at very low frequencies.
The major difference between RF and dc SQUIDs is that
the dc SQUID offers lower noise. From a historical viewpoint,
although the dc SQUID was the first type of SQUID magnetometer made, the early development was with RF SQUIDs
because of the difficulty in fabricating two nearly identical
Josephson junctions in a single device. With modern thin film
fabrication techniques and improvements in control electronics design, the dc SQUID offers clear advantages over the RF
SQUID for many applications.
Limitations on SQUID Technology
It is important to bear in mind several fundamental limitations in designing SQUID-based mesurement systems and
data reduction algorithms.
314
SQUIDs
1. A fundamental limitation of SQUIDs is that they are

sensitive to relative (field or current) changes only. This
is a consequence of the fact that the output voltage of a
SQUID is a periodic function (Fig. 3) of the flux penetrating the SQUID loop. The SQUID is flux locked on
an arbitrary maximum (or minimum) on the V
curve, and the SQUID output is sensitive to flux
changes relative to this lock point.
2. A second limitation exists on the system bandwidth. Although the SQUID has an intrinsic bandwidth of several gigahertz, when operated with standard flux-locked
loop electronics using ac flux modulation, the maximum
bandwidth is typically 50 kHz to 100 kHz. Another limitation is the presence of 1/f noise. High Temperature
Superconducting (HTS) SQUIDs [and early commercial
Low Temperature Superconducting (LTS) dc SQUIDs]
exhibit excess 1/f noise due to critical current fluctuations of the Josephson junctions. This noise can be reduced by reversing the dc bias voltage (ac bias). This
limits the maximum bandwidths less than half the bias
reversal frequency. If the bias reversal frequency is too
high, noise can be induced due to voltage spikes in the
transformer coupled preamplifier input circuit. Because
of this, the maximum bandwidth of present day HTS
SQUIDs is 30 kHz. If megahertz bandwidths are required, the ac bias is not used; however, there will be
excess noise below 1 kHz.
3. Finally, SQUID magnetometers are vector magnetometers. For a pure magnetometer operating in the earths
magnetic field, a 180 rotation will sweep out a total
field change of 100 T. If the magnetometer has a sensitivity of 10 fT/ Hz, tracking the total field change
requires a dynamic range of 100 T/10 fT 200 dB,
well beyond the capabilities of current electronics. In
addition, the rotational speed must not cause the current flowing through the SQUID sensor to exceed its
slew rate limitations. An ideal gradiometer is insensitive to a uniform field and would not suffer this dynamic range limitation.
INPUT CIRCUITS
Whether an RF or dc SQUID, a SQUID system can be considered as a black box that acts like a current- (or flux-) to-voltage amplifier with extremely high gain. In addition, it offers
extremely low noise, high dynamic range (140 dB), excellent
linearity (1 : 107), and a wide bandwidth that can extend
down to dc.
Today, SQUIDs are fabricated as planar devices. In this
configuration, the superconducting loop, Josephson junctions
and coils (input, feedback, and modulation) are patterned on
the same device. Multilayer deposition techniques are used
(primarily in LTS devices), and coils are normally in the form
of a square washer. The planar configuration leads to quite
small devices, occupying only a few cubic millimeters compared to 5 cm3 (1.2 cm diam. 5 cm) for older toroidal RF
SQUIDs (9). Another advantage of the planar device is that
it is possible to have the detection coils as part of the SQUID
sensor, eliminating the need for separate (three-dimensional)
detection coils. Such an integrated sensor has the potential to
reduce the complexity of multichannel systems significantly.
Although it is possible to couple magnetic flux directly into

the SQUID loop, environmental noise considerations (see Fig.
9) make this difficult, if not impossible, in an unshielded environment. In addition, the area of a typical SQUID loop is
small (0.1 mm2), and its resulting sensitivity to external flux
changes ( A B) is also small. Although a larger loop
diameter would increase the SQUIDs sensitivity to external
flux, it would also make it much more susceptible to environmental noise. For this reason, external flux is normally inductively coupled to the SQUID loop by a flux transformer.
Conceptually, the easiest input circuit to consider for detecting changes in magnetic fields is that of a SQUID sensor
connected to a simple superconducting coil (Fig. 5).
Because the total flux in a superconducting loop is conserved, any change in external field through the signal coil
will induce a current in the flux transformer that must satisfy
= NA
B = (Lcoil + Li )
I
(3)
where B is the change in applied field; N, A, and Lcoil are the

number of turns, area, and inductance of the detection coil;
Li is the inductance of the SQUID input coil; and I is the
change in current in the superconducting circuit. If the lead
inductance is not negligible, it must be added to Lcoil and Li.
To calculate the sensitivity and noise level of a simple detection coil system, the inductance of the detection coil must
be known. The inductance of a flat, tightly wound, circular
multiturn loop of superconducting wire is given by (10)

8r
2
4 103 N 2 r ln
H
cm turn2
(4)
where r is the radius of the detection coil and is the radius

of the (superconducting) wire. Knowing the coil inductance
Lcoil, we can rewrite Eq. (3) as
B = (Lcoil + Li )
I/NA
(5)
Because the SQUID system has an output proportional to the

input current, maximum sensitivity is obtained by using the
input circuit that provides the maximum current into the
SQUID and satisfies all other constraints of the experimental
apparatus. For a pure magnetometer, the maximum sensitivity will occur when the impedance of the detection coil
matches that of the SQUID sensor (Lcoil Li).
Detection Coils
Several factors affect the design of the detection coils (11).
These include the desired sensitivity of the system, the size
and location of the magnetic field source, and the need to
SQUID
loop
Magnetometer coil
(N = 2 turns)
Electronics
Li
Lc
Figure 5. Schematic diagram of typical SQUID input circuit.
SQUIDs
match the inductance of the detection coil to that of the

SQUID. The ability to separate field patterns caused by
sources at different locations and strengths requires a good
signal-to-noise ratio. At the same time, one has to find the
coil configuration that gives the best spatial resolution. Unfortunately, these two tasks are not independent. For example,
increasing the signal coil diameter improves field sensitivity
but sacrifices spatial resolution. In practice, system design is
restricted by several constraints: the impedance and noise of
the SQUID sensors, the size of the dewar, and the number of
channels, along with the distribution and strength of noise
sources.
It is extremely important for dc response that the detection
coil(s) be superconducting. Resistance in the detection circuit
has two effects: (1) attenuating the signal and (2) adding Nyquist noise. Resistive attenuation is important only below a
frequency f 0, such that the resistive impedance is equal to the
sum of the inductive impedances in the circuit (e.g., f 0 R/
Lt, where Lt is the total inductive impedance of the circuit).
Resistive noise is important only if it becomes comparable to
other noise sources or the signal (1030 J/Hz for biomagnetism, 1026 J/Hz for geophysics). For a SQUID with EN
1030 J/Hz, the total resistance of the circuit, including any
joints, must be less than 1.4 1013 (12). Thus it is very
important that all solder joints, press-fits, or connections have
as low a joint resistance as possible.
Figure 6 displays a variety of detection coils. The magnetometer [Fig. 6(a)] responds to the changes in the field penetrating the coil. More complicated coil configurations provide
the advantage of discriminating against unwanted background fields from distant sources while retaining sensitivity
to nearby sources.
Because of the present inability to make flexible wire or
make true superconducting joints in HTS materials, threedimensional HTS coil structures [e.g., Figs. 6(b,df)] are not
possible. Present day HTS magnetometers are fabricated as
planar devices and are available only as pure magnetometers
[Fig. 6(a)] and planar gradiometers [Fig. 6(c)]. As a result,
commercially available HTS devices are currently in the form
of magnetic-sensing rather than current-sensing devices.
Gradiometers
(b)
(c)
(d)
(e)
(f)
Figure 6. (a) magnetometer, (b) first derivative gradiometer, (c) planar gradiometer, (d) second derivative gradiometer, (e) first derivative asymmetric gradiometer, (f) second derivative asymmetric gradiometer. Courtesy of S. J. Williamson.
than 99% of the influence of objects more than 300 b distant

(Fig. 7). In essence, the gradiometer acts as a compensated
magnetometer. It is possible to use two gradiometers connected in series opposition [Fig. 6(d)] to further minimize the
response of the system to distant sources. This can be extended to higher orders by connecting in series opposition two
second-order gradiometers, etc. Doing so, however, reduces
1
Fir
0.1
st
de
Se
Comp
coil
ati
ve
nd
gra
de
dio
riv
Baseline (b)
me
at
0.01
ter
iv
Signal
coil
e
gr
ad
m
Distance to source
io
0.001
riv
co
et
er
Magnetometers are extremely sensitive to the outside environment. This may be acceptable if one is measuring ambient
fields. If what is to be measured is close to the detection coil
and weak, outside interference may prevent measurements at
SQUID sensitivities. If the measurement is of a magnetic
source close to the detection coil, a gradiometer coil may be
preferred. The field of a magnetic dipole is inversely proportional to the cube of the distance between the dipole and the
sensor. It follows that the field from a distant source is relatively uniform in direction and magnitude at the sensor. If we
connect in series two identical and exactly parallel loops
wound in opposite senses, separated by a distance b (the baseline), we obtain a coil [Fig. 6(b)] that will reject uniform fields.
Because the response of a single coil to a magnetic dipole
goes as 1/r3, an object that is much closer to one coil than the
other will couple better to the closer coil than the more distant. Sources that are relatively distant will couple equally
into both coils. For objects that are closer than 0.3 b, the gradiometer acts as a pure magnetometer, while rejecting more
(a)
315
Source
0.0001
1b
10 b
100 b
Distance to pickup coil
Figure 7. Response of gradient coils relative to magnetometer response (1/r3 suppressed).
316
SQUIDs
the sensitivity of the instrument to the signal of interest and

may not significantly improve the signal-to-noise ratio.
Rejection of distant noise sources depends on having a precise match (or balance as it is sometimes referred to) between
the number of area-turns in the coils. A symmetric gradiometer [Fig. 6(b) requires that Ns As Nc Ac where N is the
number of turns and A is the area of the signal and compensation coils, respectively. An asymmetric design [Fig. 6(e,f)]
has the advantage that the inductance (Ls) of the signal coil(s)
is much greater than the compensation coils (Lc); greater sensitivity is achieved than with a symmetric design. Another
advantage is that the signal coil diameter is reduced, leading
to potentially higher spatial resolution. The optimum conditions for the number of turns in an asymmetric signal coil is
given by (13):
(Ls + Lc + Li + Ll ) Ns
(Ls + Lc + Li + Ll ) = 0
Ns
(6)
If the gradiometer is perfectly made (balanced), it will reject

uniform fields. However, if one coil has a larger effective diameter than the other, the response will not be that of a perfect gradiometer, but that of a gradiometer in series with a
magnetometer. Mathematically, the balance, can be defined
as Vt G H, where Vt is the system response, G is the
coils response to a gradient field, and H is the applied uniform fields. Typically, coil forms used to wind gradiometers
can be machined (grooved) to achieve balances that range
from 0.01 to 0.001. Planar devices, through photolithography,
can achieve lower levelsa factor of 10 or better. Superconducting trim tabs placed within the detection coils can improve to the parts per million level. High degrees of balance
can allow a SQUID gradiometer to operate in relatively large
(millitesla) ambient fields while maintaining sensitivities in
the tens of femtotesla.
For multichannel systems (such as are used in biomagnetism), it is not possible to use externally adjustable trim
tabseach tab tends to interfere with the others. The use of
electronic balancing (14) can provide balance ratios at the
parts per million level. In this situation, portions of (additional) magnetometer reference channel response are
summed electronically with the gradiometers input to balance out its effective magnetometer response. The use of
eight-element tensor arrays as reference channels can further
improve external noise rejection. The major advantage of electronic balancing is significant improvement in immunity to
low-frequency environment noise.
between room temperature and the cryogen chamber. Within

the vacuum space, a thermal shield (anchored to the neck
tube) acts to reduce heat transfer by thermal (black body) radiation. The thermal shield either can be vapor cooledusing
the enthalpy of the evaporating helium or nitrogen gasor
have the shield thermally connected to a liquid nitrogen reservoir. Dewars with remoable sections use liquid-nitrogencooled shields.
If the experiment involves measurements interior to the
dewar, then a metallic dewar is preferable. Metallic dewars
offer significant shielding from environmental noise at frequencies above 10 Hz to 100 Hz. If the system is to measure
magnetic fields exterior to the dewar, the dewar must be magnetically transparent, and metallic construction is not appropriate. Dewars for external field measurements are normally
constructed of nonmetallic, low-susceptibility materials to
minimize their magnetic interactions with the SQUID sensors
and detection coils. Materials used are typically glass-fiber
epoxy composites such as G-10. In an effort to get the detection coil(s) as close as possible to the object being measured,
a tailed design is often used. This decreases the forces on
the bottom of the dewar and allows the use of thinner end
pieces (closer tail spacing). Dewars for biomagnetic measurements often have curved tails to get closer to the head.
The major advantage of high-temperature superconductivity is the simplified cryogenics and reduced spacing between
cryogenic regions and room temperature. The thermal load
(due to conduction and black body radiation) is less, and the
heat capacity of what needs to be cooled is larger (implying
smaller temperature variations for a given heat load). Because the latent heat/unit volume of liquid nitrogen is 60
times larger than liquid helium, hold times become months
rather than days for an equivalently sized dewar.
Neck
tube
Vacuum
space
Thermal
shield
Outer wall
at room
temperature
REFRIGERATION
The superconducting nature of SQUIDs require them to operate well below their superconducting transition temperature
(9.3 K for Nb and 93 K for YBa2Cu3O7). The thermal environment for the SQUID sensor and detection coil has typically
been liquid helium or liquid nitrogen contained in a vacuuminsulated vessel known as a dewar (Fig. 8). The cryogen hold
time depends on the boil-off rate (heat load) and the inner
vessel volume.
The major heat load on dewars is the result of thermal
conduction down the neck tube and a magnetometer probe
along with black body radiation. The space between the inner
and outer walls is evacuated to prevent thermal conduction
Neck
baffle(s)
Magnetometer
probe
Inner wall
at liquid
helium
temperature
SQUIDs
Detection
coils
Tail spacing
Figure 8. Typical design of a fiberglass dewar used for biomagnetic

measurements.
SQUIDs
Closed Cycle Refrigeration
107
Environmental Noise
The greatest obstacle to SQUID measurements is external
noise sources. If the object being measured is within the cryostat (such as is typical in most laboratory experiments), metallic shielding can minimize external noise (e.g., act as a
low-pass eddy current shield). Superconducting shields essentially eliminate all external field variations. This assumes
that any electrical inputs to the experimental region have
been appropriately filtered. Powerline or microprocessor clock
frequencies can severely degrade performance.
When measuring external fields, the SQUID magnetometer must operate in an environmentthe magnetic field of
the earththat can be ten orders of magnitude greater than
its sensitivity (Fig. 9). The magnetic field at the surface of
the earth is generated by a number of sources. There exists a
background field of 50 T with a daily variation of 0.1
T. In addition, there is a contribution (below 1 Hz) from the
interaction of the solar wind with the magnetosphere. The
remaining contributions to external magnetic fields are primarily man-made. These can be caused by structural steel
and other localized magnetic materials such as furniture and
instruments that distort the earths field and result in field
gradients; moving vehicles that generate transient fields;
electric motors; elevators; radio, television, and microwave
transmitters; and the ever-present powerline electromagnetic
field and its harmonics.
106
Hospital
(Helsinki)
Laboratory
(Berlin)
Lab.
(Helsinki)
Low
noise
Field noise (fT/Hz1/2)
As an alternative to the use of liquid cryogens, closed cycle

refrigeration would be desirable for several reasons. These include reduction of operating costs, use in remote locations,
operation in nonvertical orientations, avoiding interruptions
in cryogen deliveries, safety, and the convenience of not having to transfer every few days. Although one mechanically
cooled SQUID system has been built (15), the inherent vibration and magnetic signature of present day closed cycle refrigerators prevent them from widespread use. The development
of pulse tube refrigerators (16) offers promise for magnetometer operation without cryogens.
Typical geomagnetic
activity
(Stanford)
104
103
Eddy current shielded
room (Tampere)
102
Magnetically
shielded room
(M.I.T.)
101
Magnetically
Shielded room
(Berlin)
1
101

/0 f
(7)
where f is the frequency of the applied field, is the electrical

resistivity, and 0 is the magnetic permeability of free space.
In situations where the wall thickness t , external fields
are attenuated by
1
Hi
=
He
1 + (2 f L/R)2
(8)
where L is the inductance of the enclosure and R is the resistance along the path of current flow. Unfortunately, induced
Sensor limit
1
101
Frequency (Hz)
102
Figure 9. Rms field noise spectra in various environments as a function of frequency. Courtesy of S. J. Williamson.
currents in the shield generate noise. For a cylindrical shape,

Noise Reduction
One method to attenuate external noise sources is with an
eddy current shield that generates fields that act to cancel
the externally applied fields within the conducting material.
The shielding effect is determined by skin depth . For a sinusoidal varying wave
High
noise
105
Brms =
317
64kB Tt
hd
(9)
where h is the length and d is the diameter of the can. The

cut-off frequency is given by f 3 dB /4dt. Because of noise
considerations, eddy current shields that are to be placed
near the detection coils should be made from relatively poor
conductors such as BeCu.
Shielded Rooms
Another approach is to use eddy current shielding to shield
the entire measurement system. An eddy current room constructed with 2 cm high purity aluminum walls can achieve
shielding 40 dB at 60 Hz with improved performance at
higher frequencies. The equivalent field noise is less than 200
f T/ Hz at frequencies above 1 Hz.
The need for shielding at lower frequencies has led to the
use of magnetically shielded rooms (MSR). In the situation
where t , the attenuation goes as (r/ )et/ . Using pure eddy
current shielding would require wall thicknesses that could
exceed 1 m or more (below 1 Hz). For a ferromagnetic mate-
318
SQUIDs
rial, the permeability of the material [ 0(1 )] replaces

0 in Eq. (8). The shielding is due to the fact that flux prefers
the path with the highest permeability. Because magnetically
soft materials (e.g., mu-metal) can have permeabilities
that exceed 104, the external magnetic flux is routed around
the walls, avoiding the interior. The use of multiple shields
can act to further shield the interior of a MSR. For the sixlayer Berlin MSR (Fig. 9), shielding factors exceeded 104 at
frequencies above 0.01 Hz with noise levels below 3 f T/ Hz.
APPLICATIONS
A large number of applications (Fig. 10) configure the SQUID
as a magnetometer. SQUIDs can also be configured to measure a wide variety of electromagnetic properties (Fig. 11).
The state of the art in materials processing limits the variety of superconducting input circuits that can be used with
HTS SQUIDs. As already mentioned, there is no existing
method for making superconducting connections to SQUIDs
with HTS wire. As a result, commercially available HTS devices are currently in the form of magnetic sensing [Fig. 6(b)]
rather than current sensing devices [Fig. 6(a,cf)].
Laboratory Applications (4,8,17)
Table 1 shows typical capabilities of commercially available
SQUID-based instruments. The number in the parenthesis
refers to the corresponding Fig. 11.
Current. One common use of a SQUID is as an ammeter
[Fig. 11(a)]. The input can be connected to an experiment at
liquid helium temperatures or to room temperature. If the
signal is to be inductively coupled to a detection coil that is
connected to the SQUID input, then the circuit must be superconducting if dc response is desired.
If the measurement is of a current that passes through the
detection coil, a toroidal geometry for the detection coil has
the advantage of extremely good coupling to the source while

rejecting contributions due to external sources. Because the
measurement is inductive, there is no loading of the currentgenerating elements.
Voltage. Typically the majority of applications use superconducting circuits. There are, however, a number of applications where resistive circuits are used. One example is the
detection of extremely small voltages or resistances (Fig. 12).
When a voltage VI is applied across the input terminals, a
current is generated in the SQUID input coil. In this situation, the feedback current (IF) that would normally be applied
to the SQUID loop is fed back via RF through rs until the
voltage drop across rs is equal to VI and there is no net current
through the SQUID. VO measures the voltage drop across RF
and rs with VI VOrs /(RF rs). The voltage gain of the system
is determined by the ratio of RF /rs. Typical values for RF and
rs are 3 and 30 , respectively, giving a voltage gain of 108.
The standard resistor rs is typically at 4.2 K. The voltage
source, however, may be at a completely different temperature.
The input noise of a SQUID picovoltmeter (1014 V) is a
function of the source resistance and temperature (R Ts),
the inherent voltage noise (due to rs), and the inherent current noise of the SQUID. Measurement of the Johnson noise
in a resistor (V2 4 kBTR f where f is the bandwidth of
the measurement) can be used to determine absolute temperature. Commercially available LTS SQUIDs have equivalent
device temperatures 1 K and are suitable for noise thermometry.
With the addition of an appropriate current source [Fig.
11(d)], it is possible to measure resistance. Resolutions of
1011 can be achieved for Rx 102 . Other applications of
picovoltmeters include measurements of thermopower, thermal electromotive forces (emfs) (thermocouples), and infrared bolometers.
yy
;;
;;
yy
108
SQUID
susceptometry
Nondestructive test
and evaluation
Magnetic field (T)
1010
Magnetocardiography
1012 Geophysical
Magnetic
anomaly
detection
1014
HTS SQUID sensors
Magnetoencephalography
Figure 10. Field sensitivities and bandwidths typical of various applications. The
lines indicate the sensitivity of commercially available SQUIDs.
1016
104
102
1
Frequency (Hz)
LTS SQUID sensors

102
104
SQUIDs
319
(a) ac and dc current
(b) Magnetic field

ac
bridge
Z rstd
rstd
(e) ac resistance/inductance bridge

RF
m
(c) dc voltage
Current
source
ac
bridge
Rx
rstd
RF
(d) dc resistance
(f) ac Mutual inductance (susceptibility bridge)
Figure 11. (a) ac and dc current, (b) magnetic field, (c) dc voltage, (d) dc resistance, (e) ac
resistance/inductance bridge, (f) ac mutual inductance (susceptibility bridge).
Ac Susceptibility (4). The SQUID can also be used as the

null detector in an ac bridge circuit (Fig. 13) to measure both
resistive and reactive components of a complex impedance.
The unknown impedance Z is excited by a current generated
by an oscillator voltage, which is attenuated by a precision
ratio transformer . The difference between the voltage developed across the unknown impedance Z and that developed in
the secondary of a nulling mutual inductor m is applied to
the input of the SQUID circuit. The primary current in m is
proportional to the oscillator voltage and defined by the set-
Table 1. Typical Capabilities of SQUID-based Instruments

Measurement
Current [Fig. 11(a)]
Magnetic fields [Fig. 11(b)]
dc voltage [Fig. 11(c)]
dc resistance [Fig. 11(d)]
Mutual/self inductance [Fig. 11(e)]
Magnetic moment [Fig. 11(f)]
SQUID
electronics
VO
Liquid helium region

RF
SQUID
RS ,TS
Sensitivity
1012
1015
1014
1012
1012
1010
A/ Hz
T/ Hz
V
H
emu
rs
Figure 12. Block diagram of SQUID picovolt measuring system.

Negative feedback is applied to the cryogenic input circuit through
the voltage divider formed by RF and rs.
320
SQUIDs
Liquid helium region
Null
detector
Control
electronics
Preamp
SQUID
Oscillator
ac bridge unit
Figure 13. Block diagram of ac bridge.
ting of the ratio transformer . An additional reactive current

is supplied by a second ratio transformer , which causes the
primary current to be passed through a capacitor rather than
a resistor, thus generating a 90 phase shift in the voltage
applied to m. The amplified off-balance signal, which appears
at the output of the SQUID control electronics, can be displayed by means of a lock-in amplifier tuned to the oscillator
frequency.
As mentioned earlier, the sensitivity is limited inherently
by Johnson noise in the resistive components of the unknown
(including the potential connections) and by the device noise
of the SQUID sensor. Assuming IN 1 pA/ Hz, such a system is capable of measuring R between 1010 to 0.5 and L
(self and mutual) between 1012 and 103 H. Using a current
comparator as the bridge excitation V, a 0.1 ppm resistance
bridge can be constructed.
SQUID Magnetometer/Susceptometers (16). Instead of using
a secondary ac excitation coil [Fig. 11(f)], a dc field can be
used to magnetize samples. Typically the field is fixed, and
the sample is moved into the detection coils region of sensitivity. The change in detected magnetization is directly proportional to the magnetic moment of the sample. Because of
the superconducting nature of SQUID input circuits, true dc
response is possible.
Commonly referred to as SQUID magnetometers, these
systems are properly SQUID susceptometers. They have a homogeneous superconducting magnet to create a very uniform
field over the entire sample measuring region and the superconducting pick-up loops. The magnet induces a moment
allowing a measurement of magnetic susceptibility. The superconducting detection loop array is rigidly mounted in the
center of the magnet (17). This array is configured as a gradient coil to reject external noise sources. The detection coil geometry determines what mathematical algorithm is used to
calculate the net magnetization. Oppositely paired Helmholtz, first and second derivatives have all been successfully
used. Coupling two axial channels of differing gradient order

can significantly improve noise rejection.
Sensitivities of 108 emu have been achieved, even at applied fields of 9 T. Placement of secondary excitation coils can
allow ac susceptibility measurements to be made in the presence of a significant dc bias field. Variable temperature capability (1.7 K to 800 K) is achieved by placing a reentrant cryostat within the detection coils.
Other Laboratory Applications. NMR signals (18) can be
measured by placing a sample (e.g., protons or 19F) in the center of SQUID detection coils and either sweeping the external
field or appying an rf excitation to the sample. The same experimental concept can be used to measure electron paramagnetic resonance (EPR) signals. SQUIDs have been used for
more esoteric applications including temperature measurements with resolution near 10 K to 12 K. SQUIDs have also
been used to measure position for gravity wave detectors with
ngstrom resolution and tests of Einsteins General Thesub A
ory of Relativity. Because SQUID magnetometers are vector
devices, they can detect rotations better than 103 arc-seconds
in the earths magnetic field. SQUIDs have been used in
searches for dark matter such as Weak Interacting Massive
Particles (WIMPs) along with attempts at detecting magnetic
monopoles and free quarks.
Geophysical Applications (19,20)
SQUID magnetometers are used to measure the earths magnetic field (Fig. 9) at frequencies ranging between 1 kHz and
104 Hz. A technique known as magnetotellurics (21) can be
used to determine the electrical conductivity distribution of
the earths crust by measuring the earths electric and magnetic field. Because the earth is a good electrical conductor
compared to the air, the electrical field generated in the ionosphere (as a result of solar wind) is reflected at the earths
surface, with components of both the electric and magnetic
field decaying as they penetrate into the earth. The decay
length or skin depth 500 , where is the electrical
resistivity of the earth and is the period of the electromagnetic wave.
In magnetotellurics, the electric field (as a function of frequency) is related to the magnetic field via an impedance tensor where E() H(). The impedance tensor contains
four complex elements Zxx, Zxy, Zyx, and Zyy and is related to
the resistivity by ij 0.2 Zij()2 where has units of mV/
km-nT.
Magnetic anomaly detection uses the five unique spatial
components of B to locate a magnetic dipole uniquely. This
has potential uses in mineralogical surveys and detection of
unexploded ordnance.
Nondestructive Test and Evaluation (22,23)
Magnetic-sensing techniques such as eddy current testing
have been used for many years to detect flaws in structures.
A major limitation on their sensitivity is the skin depth [Eq.
(7)] of metallic materials. Because SQUID sensors have true
dc response and superior sensitivity, they can see deeper
into metallic structures. dc response also means that they can
detect remnant magnetizationwithout the need for externally applied magnetic fields. Their flat frequency response
and zero phase distortion allows for a wide range of applica-
SQUIDs
Table 3. Medical Applications of SQUIDs
Table 2. NDE Measurement Techniques

Imaging
Intrinsic currents
Remnant magnetization
Flaw-induced perturbations in applied currents
Johnson noise in metals
Eddy currents in an applied ac field (flaws)
Embedded magnetic sensors
Hysteretic magnetization due to:
Cyclic stress (strain)
Simultaneous dc & ac magnetic fields
Magnetization of paramagnetic, diamagnetic and ferromagnetic
materials in dc magnetic fields
tions. One potential application of SQUIDs is in detection of

stress or corrosion in reinforcing rods used in bridges, aircraft
runways, or buildings. Table 2 shows some of the measurement techniques that can be used with SQUID sensors.
SQUID magnetometers have been used to make noncontact measurements of timing circuits (24)one instrument
has better than 10 m resolution (25). Such instruments with
MHz bandwidths could be used for circuit board and integrated circuit (IC) mapping.
Medical Applications of SQUIDs (23,2629)
The use of bioelectric signals as a diagnostic tool is well
known in medicine [e.g., the electrocardiogram (ECG) for the
heart and the electroencephalogram (EEG) for the brain]. The
electrical activity that produces the surface electrical activity
that is measured by EEG and ECG also produces magnetic
fields. The analogous magnetic measurements are known as
the magnetocardiogram (MCG) and the magnetoencephalogram (MEG). Other physiological processes also generate
electrical activity with analogous magnetic fields (Fig. 14).
Magnetic fields from active electrical sources in the body
can be measured passively and external to the body by placing the magnetometer in close proximity to the bodys surface.
It has been shown that a population of neurons in the brain
can be modeled as a current dipole that generates a well-de-
Signal strength (fT)
105
Cardiogram
Oculogram
104
Myogram
Fetal cardiogram
103
rhythm
Auditory evoked response
Retinogram
101
1
10
Frequency (Hz)
fined magnetic field profile. Mapping of these field profiles can

be used to infer the location of the equivalent active dipole
site region to within millimeters. Using evoked response techniques, the location of signal pathways and information processing centers in the brain can be mapped at different delay
times (latencies) following the stimulus.
There are also magnetic mesurements for which there are
no electrical analogs. These are measurements of static magnetic fields produced by ferromagnetic materials ingested into
the body and measurements of the magnetic susceptibility of
materials in the body. In particular, information on the quantity and depth of diamagnetic or paramagnetic materials
(such as iron stored in the liver) can be obtained by using
magnetizing and detection coils of differing sizes in the same
instrument and measuring the induced field as a function of
distance. This technique is already being used to monitor patients suffering from iron overload diseases such as thalassemia and hemochromatosis.
The development of the SQUID has allowed the development of noninvasive clinical measurements of biomagnetic
fields. The use of gradiometers can allow measurements to be
made in unshielded environments at sensitivities below 20
fT/ Hz. Typically, however, neuromagnetic measurements
are made in room-sized MSRs that will allow measurements
of the magnetic field of the brain over the entire surface of
the head (150 positions simultaneously). Table 3 gives some
of the areas in which SQUID magnetometers are currently
being used in medical research.
1. B. D. Josephson, Possible new effect in superconductive tunneling, Phys. Lett., 1: 251253, 1962.
Evoked cortical activity

10
102
Studies of the brainneuromagnetism

Epilepsy
Presurgical cortical function mapping
Drug development and testing
Stroke
Alzheimers
Neuromuscular disorders
Prenatal brain disorders
Performance evaluation
Studies of the heartmagnetocardiography
Arrhythmia
Heart muscle damage
Fetal cardiography
Other medical applications
Studies of the stomachgastroenterology
Intestinal ischemia
Noninvasive in vivo magnetic liver biopsies
Lung function and clearance studies
Nerve damage assessment
BIBLIOGRAPHY
Visual evoked response

102
321
102
103
Figure 14. Typical amplitudes and frequency ranges for various biomagnetic signals.
2. J. Clarke, SQUID fundamentals, in H. Weinstock (ed.), SQUID

Sensors: Fundamentals, Fabrications and Applications, Dordecht:
Kluwer, 1997.
3. T. Van Duzer and C. W. Turner, Principles of Superconductive
Devices and Circuits, New York: Elsevier, 1981.
4. T. P. Orlando and K. A. Delin, Foundations of Applied Superconductivity, Reading, MA: Addison-Wesley, 1991.
322
SRAM CHIPS
5. R. P. Giffard, R. A. Webb, and J. C. Wheatley, Principles and

methods of low-frequency electric and magnetic measurements
using an rf-biased point-contact superconducting device, J. Low
Temp. Phys., 6: 533610, 1972.
6. R. J. Prance et al., Fully engineered high performance UHF
SQUID magnetometer, Cryogenics, 21: 501506, 1981.
7. C. D. Tesche and J. Clarke, DC SQUID: noise and optimization,
J. Low Temp. Phys., 29: 301331, 1982.
8. M. B. Simmonds and R. P. Giffard, Apparatus for reducing low
frequency noise in dc biased SQUIDs, US Patent No. 4,389,612,
1983.
9. R. L. Fagaly, Superconducting magnetometers and instrumentation, Sci. Prog., Oxford, 71: 181201, 1987.
10. F. W. Grover, Inductance Calculations, Working Formulas and
Tables, New York: Dover, 1962.
11. J. P. Wikswo, Jr., Optimization of SQUID differential magnetometers, AIP Conference Proc., 44: 145149, 1978.
12. R. B. Stephens and R. L. Fagaly, High temperature superconductors for SQUID detection coils, Cryogenics, 31: 988992, 1991.
13. R. Illmoniemi et al., Multi-SQUID devices and their applications,
in D. F. Brewer (ed.), Progress in Low Temperature Physics, vol.
XII, Amsterdam: Elsevier, 1989.
14. J. Vrba, SQUID gradiometers in real environments, in H. Weinstock (ed.), SQUID Sensors: Fundamentals, Fabrications and Applications, Dordecht: Kluwer, 1997.
15. D. S. Buchanan, D. N. Paulson, and S. J. Williamson, Instrumentation for clinical applications of neuromagnetism, in R. W. Fast
(ed.), Advances in Cryogenic Engineering, vol. 33, New York: Plenum, 1988, pp. 97106.
16. C. Heiden, Pulse tube refrigerators: A cooling option, in H. Weinstock (ed.), SQUID Sensors: Fundamentals, Fabrications and Applications, Dordecht: Kluwer, 1997.
17. R. E. Sarwinski, Superconducting instrumentation, Cryogenics,
17: 671679, 1977.
18. R. A. Webb, New technique of improved low-temperature SQUID
NMR measurements, Rev. Sci. Instrum., 48: 1585, 1977.
19. J. Clarke, Geophysical Applications of SQUIDs, IEEE Trans.
Mag., MAG-19: 288294, 1983.
20. H. Weinstock and W. C. Overton, Jr. (eds.), SQUID Applications
to Geophysics, Tulsa: Soc. of Exploration Geophysicists, 1981.
21. K. Vozoff, The magnetotelluric method in the exploration of sedimentary basins, Geophysics, 37: 98114, 1972.
22. G. B. Donaldson, SQUIDs for everything else, in H. Weinstock
and M. Nisenoff (eds.), Superconducting Electronics, New York:
Springer-Verlag, 1989.
23. J. P. Wikswo, Jr., SQUID magnetometers for biomagnetism and
non-destructive testing: Important questions and initial answers,
24. R. L. Fagaly, SQUID detection of electronic circuits, IEEE Trans.
Magn., MAG-25: 12161218, 1989.
25. J. Kirtley, Imaging magnetic fields, IEEE Spectrum, 33 (12): 40
48, 1996.
26. S. J. Williamson and L. Kaufman, Biomagnetism, J. Magn. Mag.
Mat., 22: 129202, 1981.
27. G-L. Romani, S. J. Williamson, and L. Kaufman, Biomagnetic instrumentation, Rev. Sci. Instrum., 53: 18151845, 1982.
28. R. L. Fagaly, Neuromagnetic instrumentation, in S. Sato (ed.),
Advances in Neurology, Vol. 54: Magnetoencephalography, New
York: Raven Press, 1990.
29. C. Aine et al., Advances in Biomagnetism Research: Biomag96,
New York: Springer-Verlag, 1997.
ROBERT L. FAGALY
Tristan Technologies
SUPERCONDUCTORS, STABILITY IN FORCED FLOW
87
SUPERCONDUCTORS, STABILITY
IN FORCED FLOW
Forced-flowcooled conductors are the preferred choice for
magnets that must operate in an electromagnetic and mechanically noisy environment, when pulsed operation requires
minimization of ac losses, or whenever the operating conditions require a reliable and cost-effective design. In this article we review the guidelines that motivated the choice of
forced-flowcooled conductors to obtain an effective and stable superconductor design for large magnets, such as those
for fusion, superconducting magnetic energy storage (SMES),
particle detectors, or magneto-hydrodynamic (MHD) application. We will discuss the particular features of the stability
margin in forced-flowcooled conductors and the models commonly used to compute it.
SUPERCONDUCTOR STABILITY
Superconductors exhibit zero resistance only within relatively
narrow parameters of temperature, magnetic field, and transport current, below the so-called critical surface. When
brought outside this region by a disturbance (e.g., by energy
deposition stemming from a mechanical motion) superconductivity is lost and Joule heating is generated. If not prevented
by other mechanisms, the superconductor cascades further
from its nominal operating point into an irreversible process
leading to the complete loss of superconductivity in the magnet. This process is commonly known as a quench. Even if the
magnet is properly protected against damage, a magnet
quench is an undesirable event in terms of availability and
cost. A well-designed magnet will not quench under normal
operating conditions. The study of stability pertains to the
understanding of the processes and mechanisms whereby a
superconductor will remain (or not) within its operating region, thus ensuring magnet operation without quench. This
area of study has evolved through many years of experimentation and analysis.
88
The first superconducting magnets were cooled by immersion

in a helium bath. As we will see later, classical stability theory as derived for these bath-cooled magnets does not directly
extend to forced-flow conductors. It is nonetheless useful to
review here the oldest and simplest stability criterion developed for bath-cooled conductors, the so-called Stekly criterion
of cryostability (1). In their original development, Stekly and
Zar (1) backed a superconducting material with a low-resistance copper shunt. The cross section ACu of the shunt was
such that in the case of a transition of the superconducting
material to the normal state, the maximum Joule heating,
obtained when the current I was completely displaced from
the superconductor to the copper, was smaller than the heat
removal capability at the conductor perimeter pw wetted by
the helium. Under this condition the conductor always recovered from a perturbation, irrespective of the size of the disturbance that caused the quench. In brief, the conductor was unconditionally stable. Writing the simple power balance of
heating and cooling, they came to the following criterion for
cryostability, formulated using the so-called Stekly parameter :
=
Cu I 2
<1
hpw ACu (Tc Top )
(1)
where Cu is the stabilizer resistivity, h the heat transfer coefficient between conductor and cooling bath, Tc is the critical
temperature, and Top the bath operating temperature.
Stability versus Perturbation Spectrum
Cryostable conductors have an exceptional tolerance to energy inputs. The drawback is that the resulting operating current density is low, and thus coil size and cost are large. The
present approach is different, and consists in designing the
conductor to be stable against the spectrum of energy disturbances expected in the magnet, instead of requiring the conductor to be stable against disturbances of arbitrary nature
and intensity. This implies a comparison of the initial estimate of the energy release mechanisms and magnitude to the
so called energy margin E that we define as the maximum
energy deposition that the conductor can tolerate while recovering from the superconducting state. Let us take as an example a conductor in which mechanical energy releases dominate. Following Keilin (2,3), we compare schematically in Fig.
1 the stability margin to the disturbances that can potentially
drive the conductor normal. As shown there, the stability
margin E decreases with increasing current I, while mechanical disturbances, the curves labeled D, increase with
the current. In this case, the conductor is no longer unconditionally stable but it has an upper stability margin, traditionally expressed in mJ/cm3 of metal in the cable. The disturbance spectrum must be interpreted as the energy released in
each event. At increasing current a single event releases an
increasing energy because of the proportionality to either the
Lorentz forces (I2) or to the strain energy in the cable (I4). As
the magnet is charged, the two curves approach, until eventually the spectrum of mechanical disturbances (D) equals and
surpasses the stability margin (E). At this point the magnet
will quench as soon as a perturbation event will take place.
The most likely event during magnet charge-up will be the
Energy (arbitrary units)
Stekly Criterion for Cryostability
D1
D2 Dn
Current (arbitrary units)

Figure 1. Typical curves for the conductor stability margin (E) and
mechanical disturbance spectrum (D).
one associated with the largest energy release. Most mechanical energy inputs are associated with irreversible processes
such as stick-and-slip motions and cable compaction. Therefore, once this event has taken place, and the associated energy has been released, the energy perturbation spectrum at
the following charge-up will be diminished. We illustrate this
situation in Fig. 1 by the set of curves D1, D2, Dn that represent the perturbation spectrum at successive charge-ups 1, 2,
. . ., n. The intersection of these curves with the energy margin moves towards higher currents, and we see from this elementary example a simple explanation of the phenomenon of
training that disappointed early builders of superconducting
magnets.
We see from this simple example that we have two possibilities to guarantee the stable performance of the conductor.
The first is to decrease the energy perturbations (motions,
cracks, ac losses) as much as possible so that the highest possible operating current can be achieved. This solution can be
adopted for small- to medium-scale magnets operating in a
quiet environment where, for instance, the perturbation energy input can be limited by properly fixing the cable in the
winding pack. On the other hand, large size magnets, as typical of SMES systems, thermonuclear fusion experiments or
MHD applications, operate in a mechanic and electromagnetic noisy environment (e.g., rapidly changing magnetic
fields or large stress cycles) that per force results in a minimum value of the perturbation spectrum. In this case, the
designer must increase the stability margin to tolerate the
existing perturbation spectrum. This can be achieved by increasing the heat sink associated with the cable.
In the temperature range typical of the operation of a superconducting cable, generally from 2 to 4 K, all solid materials are known to have a very small heat capacity. In the same
temperature range, helium uniquely possesses a volumetric
heat capacity two to three orders of magnitude larger than
solids. Naturally, cable designers tend to take advantage of
this feature, trying to make an effective use of the heat sink
provided by adding a limited amount of helium to the cable.
To achieve this, it is necessary to increase the heat transfer
coefficient at the wetted surface of the conductor, at the conductor surface to volume ratio, or both. Forced-flowcooled
conductors are designed along this line to make the most effective use of the helium heat sink.
89
Stability in forced-flowcooled conductors is different from

classical stability theory in adiabatic and bath-cooled wires,
cables, or built-up monoliths mainly for three reasons:
The largest heat sink providing the energy margin is the
helium, and not the enthalpy of the strands themselves
or conduction at the end of the heated length.
This heat sink is limited in amount.
The helium behaves as a compressible fluid under energy
inputs from the strands, implying additional feedback on
the heat transfer coefficient through heating induced
flow.
Round bundle with 37 strands enclosed in rectangular

conduit: Showing transposition of strands
The main issue is, therefore, the heat transfer from the
strand surface to the helium flow and the thermodynamic process in the limited helium inventory.
Figure 2. The original concept of CICC, as presented by Hoenig,

Iwasa, and Montgomery (7). Reproduced from Ref. 7 by permission
of Servizio Documentazione CRE-ENEA Frascati. Copyright 1975
CRE-ENEA Frascati.
FORCED-FLOWCOOLED CONDUCTORS
consideration finally brought Hoenig, Iwasa, and Montgomery

(6,7) to present the first CICC prototype idea, shown in Fig. 2.
Although many variants have been considered, the basic
CICC geometry has changed little since. A bundle conductor
is obtained, cabling superconducting strands, with a typical
diameter in the millimeter range, in several stages. The bundle is then jacketed, that is, inserted into a helium-tight conduit, which provides structural support. Helium occupies the
interstitial spaces of the cable. With the cable void fractions
of about 30 to 40% commonly achieved, the channels have an
effective hydraulic diameter of the order of the strand diameter, while the wetted surface is proportional to the product of
the strand diameter and the number of strands. The small
hydraulic diameter ensures a high turbulence, while the large
wetted surface achieves high heat transfer, so that their combination gives excellent heat transfer properties.
Because of the limited helium inventory, a sufficiently
large energy input will always cause a quench in a CICC.
This behavior has been defined by Dresner (11,12) as metastable. The question is the magnitude of the minimum energy
input producing a quench in a particular operating condition.
This parameter, the stability margin E, was identified by
Hoenig (7) as fundamental to the design of a CICC. It is usually measured as an energy per unit strand volume (traditionally expressed in mJ/cm3). In its original definition, the energy input was thought to happen suddenly, and initial
experiments and theory concentrated on this assumption.
Throughout this chapter we will use the same definition of
the stability margin, extending it to an arbitrary energy deposition time scale.
The heat transfer mechanisms determining stability in supercritical He-I and superfluid He-II are different enough to
warrant a separate treatment. The phenomenology of each
case and the experimental data supporting the stability calculations are presented in the next sections.
In a forced-flow conductor, the helium and the superconductor

form a single unit, with the coolant flowing inside a pipe also
housing the superconductor, or with parallel cooling and electrical paths in close thermal contact. The most common design for this class of conductors is at present the cable-inconduit conductor (CICC), in which a superconducting cable
is placed inside a conduit that also serves as helium containment. We will use this particular type of conductor as a prototype for the general discussion on stability in forced-flow
cooled conductors.
History of Cable-in-Conduit Conductors
The CICC concept evolved from the internally cooled superconductors (ICS), which had found application in magnets of
considerable size between the late 60s and early 70s [see in
particular the work of Morpurgo (4)]. In ICS, the helium was
all contained in the cooling pipe, very much like standard water-cooled copper conductors. The conductor could be wound
and insulated using standard technology and the magnet
would be stiff both mechanically and electrically, a considerable advantage for medium and large systems requiring, with
increasing stored energy, high discharge voltages. Control of
the heat transfer and cooling conditions was achieved using
supercritical helium, thus avoiding the uncertainties related
to a flowing two-phase fluid.
A major drawback of this concept was that according to
heat transfer predictions, a large helium massflow would
have been required in order to achieve good stability and thus
high operating current density. This would require large
pumping work and eventually impair the efficiency of the
cryogenic system. Chester (5) readily recognized the advantage of the increase in the wetted perimeter obtained by subdivision of the strands. Subdivision dramatically increases
the surface-to-volume ratio, thus improving heat transfer for
a given cable cross section. Hoenig et al. (68) and Dresner
(911) developed models for the local recovery of ICSs after a
sudden perturbation, where they found that for a given stability margin the mass flow required would be proportional to
the 1.5th power of the hydraulic diameter as the fixed superconductor inventory is divided in finer and finer strands. This
STABILITY MARGIN OF CABLE-IN-CONDUIT

CONDUCTORS IN SUPERCRITICAL HELIUM
Dependence on the Mass Flow
Measurements of the stability margin of CICCs in supercritical helium started early in their history (1319). One of the
90
in the stability margin to low values, the ill-cooled regime, is

found. In this regime, the stability margin is lower than in
the well-cooled regime by typically one to two orders of magnitude, and depends on the type and duration of the energy perturbation.
The transition between the two regimes was identified by
Dresner (20) to be at a limiting operating current Ilim:
0.6
Nb3Sn at 9T
NbTi at 9T
0.5
0.4
0.3
Ilim =
hpw ACu (Tc Top )

Cu
(2)
0.2
0.1
0
0.1 (NRe = 200)
1 (NRe = 2000)
10 (NRe = 20,000) 100
Coolant flow rate (g/cm3s)
Figure 3. Stability margin of a NbTi and a Nb3Sn CICCs as a function of the steady state helium flow, measured by Hoenig, Montgomery, and Waldman (14). Reproduced from Ref. 14 by permission of
IEEE. Copyright 1979 IEEE.
aims was to study the dependence of stability on the coolant

flow, to determine the influence of the turbulent heat transfer
coefficient and the thermal coupling of strands and helium.
The first surprise came with the observation by Hoenig (3)
that the stability margin was largely independent of the operating mass flow (see the curves reported in Fig. 3), a result
soon duplicated by Lue and Miller (17). These results showed
that the heat transfer at the wetted surface of the strands
during a temperature excursion was only weakly correlated to
the steady state mass flow and the associated boundary layer.
As discussed by Dresner (20) and Hoenig (16), during a
strong thermal transient the heat transfer coefficient h at the
strand surface changes mainly because of two reasons (see
also appendix, Transient Heat Transfer, below):
(a) thermal diffusion in the boundary layer (a new thermal
boundary layer is developed and thus h increases compared to the steady state value), and
(b) induced flow (21) in the heated compressible helium
(associated with increased turbulence and thus again
with an increase in h).
The concurrence of these two effects was advocated to explain
the weak dependence of E on the steady mass flow rate.
Dependence on the Operating Current. A second parameter
of major interest in the experiments on stability was the operating current of the cable. Several experiments (see the vast
amount of data presented in Refs. 22 through 27) have revealed a characteristic behavior of the stability margin as a
function of operating current. As we show schematically in
Fig. 4, at low operating current a region with high stability
margin is observed. We name this region, following Schultz
and Minervini (28), the well-cooled regime. In this regime, the
stability margin is comparable to the total heat capacity
available in the cross section of the CICC, including both
strands and helium, between operating temperature Top and
current-sharing temperature Tcs. At increasing current, a fall
The above definition of the limiting current Ilim is obtained

equating the Joule heat generation to the removal at the
strand surface, assuming that the helium temperature is constant, and is therefore equivalent to the Stekly criterion of
Eq. (1). As discussed later, the heat transfer cofficient h is not
constant, but it is a complex function of time and several
other parameters such as heating pulse waveform and
strength, heating induced flow, and details of the cable design. Let us assume for the moment that the heat transfer
coefficient is constant in time and equal to an effective value.
As shown by Lue (29), it is possible to estimate this effective
value of h, deducing it from the location of the limiting current in several experiments. For operating currents below
Ilim (i.e., in the well-cooled regime), the heat generation is
smaller than the heat removal to the helium. A normal zone
recovers, provided that the helium is a sufficiently large heat
sink capable of absorbing the heat pulse and the subsequent
Joule heating. On the other hand, above Ilim, in the ill-cooled
regime, a normal zone generates more heat than it can exchange to the helium, and therefore recovery is not possible.
This observation indeed explains the behavior of the energy margin below and above Ilim. In the well-cooled regime,
recovery is unconditional; the cable can transfer a large heat
10000
Limiting
current
Well-cooled
regime
Stability margin (mJ/cm3)
Energy input Qimp (J/cm3 of wire)
0.7
Ill-cooled
regime
Strands+helium
heat capacity
1000
Lower
limiting
current
100
Multiple
stablility
region
Strands
heat
capacity
10
0.00
1.00
Iop/Ic
Figure 4. Schematic behavior of the stability margin as a function

of the cable operating current.
low
Ilim
=
hpw ACu (Tc Tcs )

Cu
(3)
that we call lower limiting current for analogy to Eq. (2) and
low
because Ilim
is always smaller than Ilim. For operation at (and
low
below) Ilim , the full heat sink can be used for stabilization and
the stability margin is at the upper limitthe well-cooled
low
value. Between the two values Ilim and Ilim
, the stability margin falls gradually.
Multiple Stability
Near the limiting current the balance between heat production and removal becomes critical. Indeed, in some cases, a
multivalued region can be found in the vicinity of Ilim, as schematically shown in Fig. 4. As mentioned earlier, supercritical
helium behaves as a compressible fluid in the typical range
of operation of a magnet. Therefore, any heat pulse causes a
heating-induced flow driven by the fluid expansion and proportional to the pulse power. The flow in turn modifies the
heat transfer at the wetted surface of the conductor, enhancing the heat transfer coefficient. Let us concentrate on the
close vicinity of the limiting current, just above Ilim on the illcooled side. A large heating power, above the ill-cooled stability margin, can result in a significant heating-induced flow
and thus a large enhancement of the heat transfer coefficient.
Hence the power balance can be tipped in the direction favorable to recovery, and a second stable region appears. This is
what has been observed by Lue, Miller and Dresner (18,19)
500
200
H (mJ/cm3)
pulse to the helium and still recover at the end of the pulse,
provided that the helium temperature has not increased
above Tcs. Therefore, the energy margin is of the order of the
total heat sink in the cable cross section between the operating temperature Top and Tcs, including, obviously, the helium. In the ill-cooled regime, an unstable situation is reached
as soon as the strands are current sharing, and therefore the
energy margin is of the order of the heat capacity of the
strands between Top and Tcs plus the energy that can be transferred to the helium during the pulse. As mentioned earlier,
in practical cases, the heat capacity of the helium in the cross
section of a CICC is the dominant heat sink by two orders of
magnitude and more, and this explains the fall in the stability margin above Ilim.
The transition between the well-cooled and ill-cooled regimes happens in reality as a gradual fall from the maximum
heat sink values to the lower limit [Miller, (25)]. An intuitive
explanation of this fall can be given using again the power
balance at the strand surface. For the derivation of Eq. (2) it
was assumed that the helium has a constant temperature
Top. In reality, during the transient, the helium temperature
must increase as energy is absorbed and power is transferred
under a reduced temperature difference between strand and
helium. Two limiting cases can be defined. The first is the
ideal condition of helium at constant temperature, giving the
limiting current of Eq. (2), for which, however, the energy absorption in the helium is negligible. Operation exactly at Ilim
results thus in a stability margin at the lower limitthe illcooled value. The second limiting case is found when the
Joule heat production can be removed even when the helium
temperature has increased up to Tcs. This second case is obtained for a current of (and below)
91
100
50
NbTi single triplex
w = 1.0 mm, pabs = 5.0 atm
h = 16.7 ms, vHe = 0
Lsample = 3.8 m, B = 6.0 T
20
340
360
380
400
Is (A)
420
440
Figure 5. Stability margin of a NbTi CICC as a function of the operating current, measured by Lue and Miller (19). The experiment
was performed on a single triplex CICC of 3.8 m length (Lsample), with
strand diameter of 1 mm (w), under zero imposed flow (vHe) at a helium pressure of 5 bar (pabs). The background field was 6 T (B), and
resistive heating took place in 16.7 ms (h). Reproduced from Ref. 19
by permission of IEEE. Copyright 1981 IEEE.
in experiments on single triplex NbTi cables. Figure 5 reports

one such multiple stability curve, as measured as a function
of the operating current. This situation is evidently not agreeable for reliable operation and should be avoided in a sound
design by remaining safely below the limiting current.
Dependence on the Operating Field
The stability margin is a function of the background field B
mainly through the dependence on critical and current-sharing temperatures. A higher B causes a drop both in the limiting current (through a decrease of Tc and increase of Cu)
and in the energy margin (through a decrease in Tcs). Therefore, E drops as the field increases. An interesting feature is
that the limiting current only decreases with (Tc Top)1/2, that
is, with a dependence on B weaker than that of the critical
current. At large enough B we will always have that Ilim is
larger than Ic and the cable will reach the critical current
while still under well-cooled conditions.
Dependence on the Heating Time Scale. The stability margin
depends on the duration of the heating pulse, as shown experimentally by Miller et al. (17), and reported here in Fig. 6.
A change in the heating duration for a given energy input
corresponds to a change in the pulse power. In the well-cooled
regime, that is, for low operating currents in Fig. 6, the heat
balance at the end of the pulse is in any case favorable to
recovery, and therefore the energy margin does not show any
significant dependence on the pulse duration. On the other
hand, when the conductor is in the ill-cooled regime, its temperature can increase to or slightly beyond Tcs without caus-
92

500
H T =7T
H (mJ/cm3)
200
100
Figure 6. Dependence of the stability

margin for a CICC (indicated on this plot
as H) on the heating time scale (h), as
measured by Miller et al. (17). The parameters varied in the experiment, indicated in the inset, are the transport current in the sample Is, the helium flow
velocity vHe, and the pressure p. Reproduced from Ref. 17 by permission of IEEE.
Copyright 1979 IEEE.
Is (A) vHe (g/s) p (atm)

50
20
10
ing a quench. This limits the heat flux per unit length at the
wetted surface to roughly hpw(Tcs Top). The consequence is
that energy transferred to the helium, and thus the energy
margin, will grow at increasing pulse duration, until it becomes comparable to the total heat capacity available (as in
the well-cooled regime). This effect is partially balanced for
very fast pulses, because the heat transfer coefficient can exhibit very high values at early times (see the appendix, Transient Heat Transfer, below), which could shift the well-cooled/
ill-cooled transition at higher transport currents. In principle,
higher energy margins should be expected in this range. However, the high input powers in this duration range tend to
heat the conductor above 20 K, in a temperature range where
the stabilizer resistivity grows quickly, and the power balance
is thus strongly influenced. This effect causes the saturation
of the energy margin for extremely fast pulses (well below 1
ms duration).
Dependence on Operating Temperature and Pressure
The dependence on the operating temperature and pressure
in supercritical conditions is not easily quantified. The reason
is that the helium heat capacity in the vicinity of the usual
regimes of operation (operating pressure pop of the order of 3
to 10 bar and operating temperature Top around 4 to 6 K)
varies strongly with both pop and Top. This affects both the
heat sink and the heat transfer coefficient (through its transient components). An increasing temperature margin under
constant operating pressure gives a higher E. But a simultaneous variation of pop and Top, under a constant temperature
margin, can produce variations of E as large as a factor two
in the range given above [see Miller (25) and Chaniotakis,
(30)].
STABILITY MARGIN OF CABLE-IN-CONDUIT
CONDUCTORS IN SUPERFLUID HELIUM
If the operating temperature is lowered below the so-called
lambda value T (e.g., 2.17 K at 1 atm), helium undergoes a
10
20
(ms)
300
0.37
250
0.37
250
200
250
1.3
250
1.8
250
3.0
50
100
200
500
state change and becomes a quantum fluid: superfluid helium,

or He-II. He-II has unique properties and its physical behavior is very different from that of normal helium (or He-I). For
our purposes, the most remarkable fact is that He-II does not
obey the traditional Fourier law of conduction (proportionality between heat flux and temperature gradient), but rather
follows a nonlinear law of the form
q = K(T )1/3
(4)
where q is the heat flux, K is a parameter that depends on

the thermodynamic state of the He-II, and T is the temperature gradient. For heat fluxes of practical interest, the He-II
properties are such that the equivalent thermal conductivity is extremely high, orders of magnitude higher than in HeI. Superfluid heat conduction refers to the ability of stagnant He-II to provide significant heat removal over long
lengths of narrow channels without an appreciable temperature gradient, making it an attractive alternative for superconducting magnet cooling (31,32).
Similarly to the behavior in He-I, the stability margin of a
CICC operating in He-II is determined by the balance between Joule heating and the ability of the helium to provide
enough enthalpy margin, given that metal-to-helium heat
transfer is sufficient. Lottin and Miller (27) have measured
stability margin of a CICC at different operating temperatures, both in supercritical and superfluid helium. We show
typical results of this experiment in Fig. 7. The stability margin behaves at low current in a way similar to what would be
expected in the case of He-I operation. However, at the illcooled transitionthe first drop in the stability curve taking
place at similar currents both in He-I and He-IIthe stability
margin shows a peculiar behavior. Owing to the large heat
transfer capability in He-II, the power balance at the strand
surface remains favorable for recovery as long as the bulk
helium is in the He-II phase. Therefore, in a first approximation, the full heat sink between the initial operating point and
the transition temperature T is still available at levels of the
1000
6T
He = 0.50.6 mg/sec
m
500
PHe = 23 atm.
93
surface had its temperature clamped at T. This is indeed the

basis for the simplified model used to calculate the heat flux
limits in He-II and the behavior of the stability margin in the
superfluid region.
4.2 K
3.6 K
H, mJ/cm3
200
CALCULATION OF THE STABILITY MARGIN IN HE-I
1.8 K
The calculation of the stability margin in a CICC is a difficult

task, involving accurate computation of compressible helium
flow and heat diffusion in a complex geometry. For practical
purposes, several simplified models have been developed.
These models make extensive use of the experimental evidence discussed in the previous sections as a basis for introducing and justifying several simplifications. For the purpose
of introducing the reader to the concepts involved in the calculation of the energy margin, this presentation will start
with a very simplified model (an integrated energy balance),
and then proceed to introduce refinements to the model [a
zero-dimensional (0-D) energy balance model, and a one-dimensional (1-D) flow model].
100
50
20
10
Il = 194A
Il = 210A
5
160
180
200
220
240 260
I T, A
280
300
320
340
Figure 7. Stability margin of a NbTi CICC as a function of the operating current, measured by Lottin and Miller (27), at different temperatures in supercritical and superfluid helium (filled-in symbols are
quenches, open symbols are recoveries). Reproduced from Ref. 27 by
permission of IEEE. Copyright 1983 IEEE.
operating current at which the conductor would have turned

to be ill-cooled for operation in He-I. In other words, the conductor can still be considered as well-cooled for temperature
excursions up to T. We can call this region of operation the
superfluid stability regime. As the helium undergoes a phase
transition at the temperature T, the available heat sink is
significant, of the order of 200 mJ/cm3 of helium volume. At
increasing current, finally, the power balance can eventually
become unfavorable, as soon as the heat removal capability of
He-II reaches its upper limit. There the final transition to the
ill-cooled regime of operation takes place. Unlike the He-I
case, the transition from high to low stability-margin regimes
occurs gradually without a region of multivalued stability owing, on the one hand, to a small coefficient of thermal contraction (compared to the large thermal expansion in He-I), and,
on the other hand, to a very high effective thermal conductivity of He-II, both preventing the onset of large heating induced flows.
In the superfluid well-cooled stability regime, where the
conductor takes advantage of the superfluid helium properties only, the stability margin has a distinct behavior (3337)
that, as indicated by Seyfert, Laffarranderie, and Claudet
(33), is mainly determined by the near-field effects of surface
heat transfer to He-II. Following a disturbance, the metal
temperature exceeds T almost immediately, and the helium
in contact with the strand undergoes the transition to He-I.
However, at some distance from the strand surface, there is
an He-I/He-II transition where the temperature is T. It can
be shown that, given the heat transfer properties of He-II, the
layer of He-I adjacent to the strand is negligibly thin (33).
The consequence is that, for all practical purposes, heat
transfer in the superfluid bath takes place as if the strand
Energy Balance
The simplest stability model is that considering the energy
balance for the combined helium/strand system integrated
over the duration of the disturbance. This method gives a
rough estimate of the stability margin in the well- and illcooled regimes (called here Ewc and Eic) based on the available heat capacities and the location of the well-cooled/illcooled boundary (neglecting the dual-stability region), and
has the advantage of producing easily applicable design criteria for the selection of the cable layout. We introduce the maximum heat sink in the cable cross section (referred to the unit
strand volume) Emax:
Emax =
Tcs
Top
AHe
C dT +
ASt He
Tcs
Top
CSt dT
(5)
where CHe and CSt are respectively the volumetric heat capacity of helium and of the strands, while AHe and ASt are their
cross sections. The estimate of CHe can pose some questions.
As known from thermodynamics, the volumetric heat capacity
in a compressible fluid depends on the process assumed. Two
limiting cases can be identified: a process at constant volume,
where we have that CHe cv; or the case of constant pressure, where we have that CHe cp (cv and cp represent the
helium constant-volume and constant-pressurespecific
heat). The proper selection depends on the comparison of the
characteristic times involved. In a transient where the flow
characteristic times are much longer than the heating and
recovery time (i.e., for long-heated zone or fast-heating pulse),
the process will be at constant volume. Approximate constant
pressure conditions will be found when the flow characteristic
times are much shorter than the heating time (short heated
zone or long pulse). The real process will be between these
two extremes, but generally a conservative estimate is obtained choosing the constant volume process.
In the well-cooled regime we will have
Ewc
Emax
94
that is, the energy margin is at most equal to the available

heat sink up to Tcs, and in general smaller than Emax. A first
reason is that during the heat pulse e and the recovery time
r, the Joule heat generated by the current sharing strand
consumes the available heat capacity. An approximation of
the Joule heat contribution normalized to the strand volume
is given by

QJoule =
e + r
0
CuI 2
dt
ACu ASt
(6)
This contribution increases at increasing operating current

and increasing energy deposition time, although the above approximation tends to give only an upper limit and overestimates the real contribution (the strands are assumed fully
normal for the whole transient). Still, for fast and for most
common heating pulses (typically in the 1 to 10 ms range) the
term above is small. It is then justifiable to neglect the gradual fall of E, and to take
Ewc
Emax
(7)
For operation in the ill-cooled regime at and above Ilim, the

energy margin can be approximated as the sum of the strand
heat capacity up to Tcs and the energy transferred from the
strand to the helium during the heat pulse (38), again expressed per unit of strand volume:
Eic
Tcs
Top
CSt dT +
pw
(Tcs Top )
ASt
e + r
hdt
(8)
where the second term on the right-hand side is an approximation of the energy transferred to the helium under the assumption that the strands rise instantaneously to Tcs and the
helium temperature Top does not change significantly. For
short energy pulses, the use of Eq. (8) shows that generally
Eic Emax. The energy margin given by Eq. (8) tends to
increase when the energy deposition time e increases, which
is consistent with the experimental results quoted earlier. For
very long pulses, the power input in the strand can be transferred to the helium without a significant temperature difference. At the limit of long pulse times, the whole heat capacity
is used again and we have that Eic Emax. In any case, the
value of the maximum heat sink Emax of Eq. (5) remains the
absolute upper limit of the stability margin. In summary,
Eqs. (7) and (8) give the estimated energy margin respectively
below and above the limiting current of Eq. (2).
Zero-Dimensional Model
The next level of complexity and accuracy in the calculation
of the stability margin consists of introducing time as a variable to capture the distinction between short and long duration pulses while neglecting heated-zone length effects. Maintaining the fundamental distinction between strand and
helium temperature, it is possible to write this 0-D balance
as follows:
TSt
= q Ext + q Joule pw h(TSt THe )
t
T
AHeCHe He = pw h(TSt THe )
t
AStCSt
(9a)
(9b)
The rightmost terms in Eqs. (9a) and (9b) represent the thermal coupling of strands (at temperature TSt) and helium (at
temperature THe) at the wetted perimeter pw with a heat
transfer h. In Eq. (9a) we have in addition the external and
Joule heat sources (per unit conductor length) qE xt and qJoule,
respectively. The Joule heating can be computed once the critical current dependence on the temperature Ic(T) is known.
Note that an accurate calculation of qJoule is necessary to describe the recovery phase properly. This model is attractive
because of its simplicity; it can be solved efficiently and used
routinely. It is accurate in describing the local energy balance
on the time scale of recovery, but some care must be taken in
the selection of the parameters in order to capture flow-related physical features that only a 1-D model can include.
The first parameter to be chosen properly is the volumetric
helium heat capacity, as we discussed earlier. The second parameter that requires care is the heat transfer coefficient,
changing in time during the transient. While the boundary
layer formation and the associated diffusive component of the
heat transfer coefficient can be approximated in a local treatment as a variable thermal resistance between strands and
helium, the heating-induced flow and its effect on stability
are not amenable to local treatment. An average value for this
component is a reasonable choice, but the actual modeling is
to a large extent left to empiricism [see Lue (29)]. This is, in
fact, one of the research areas on stability margin in CICCs.
The search of the stability margin with the 0-D model is
the virtual analogue of the experimental technique. A trialand-error search is done on the energy input, increasing or
decreasing it as a function of the quench or recovery result at
the end of the transient.
One-Dimensional Model
With a typical hydraulic diameter in the millimeter range,
the overall helium flow in a CICC can be expected to be onedimensional, with a good approximation already over flow
lengths of the order of 1 m. As the helium flows generally in
turbulent regime, the helium temperature is nearly uniform
in the cross section of the CICC. Therefore, the temperature
gradients in the cable cross section reduce to those across the
strand, and are negligibly small. We assume that the current
distribution is uniform in the strands. In well-designed
CICCs, the current can redistribute over typical lengths of the
order of some centimeters in times of the order and below 1
ms. In this case, the heat generation in the CICC cross section
during current sharing is also uniform. This is not the case
for CICCs with insulated strands or high transverse resistance, where the current redistribution can take several seconds over lengths of several meters. In this case, an homogenized treatment is not appropriate and the stability margin
is actually strongly degraded. We will therefore drop this case
in the following treatment. As the stability transients are fast
compared to the thermal diffusivity of the conduit materials
(e.g., steel), the conduit contribution to the energy balance is
neglected also.
These assumptions lead to a much simplified 1-D model of
the CICC, where two constituents are identified: the helium
and the strands. Both are at uniform, but distinct, temperature. The compressible flow equations in the helium (mass,
momentum, and energy balances) are written to include wall
friction, modeled using a turbulent friction factor. Strand and
helium exchange heat at the wetted surface, and the thermal

coupling is usually modeled using the correlation for the heat
transfer coefficient h discussed in Appendix. The system is
then described by the equations:
+
=0
t
x
v v2
p
v|v|
+
+
= 2 f
t
x
x
Dh
e ve pv
pw h
+
+
=
(T THe )
t
x
x
AHe St
AStCSt

TSt
T
ASt
KSt st
t
x
x
= q Ext + q Joule pw h(TSt THe )
(10a)
(10b)
(10c)
(10d)
where is the helium density, p its pressure and v is the flow

velocity, f the friction factor and Dh is the hydraulic diameter
of the conductor. The total specific energy e is defined as the
sum of the internal specific energy i and the kinetic specific
energy, that is,
e=i+
v2
2
Finally, the strand heat balance of Eq. (10d) takes into proper
account the contribution of the heat conductivity KSt along the
cable length.
The 1-D model introduced above is widely used for detailed
calculations of stability margin. When the numerical solution
technique to account for the different time scales involved is
properly selected, the model can predict the heating-induced
flows responsible for multivalued stability, and can be
adapted directly to follow the evolution of the normal zone
when the energy input is large enough and the coil quenches.
The only significant modification in this case is the need to
take into account the additional heat capacity of the conduit
material. This modification is straightforward and consists of
adding a temperature diffusion equation to the system. Because of the level of fine details, even within the simplification
of the 1-D assumption, this model gives the possibility of wide
parametric analysis. Its main drawback is that, dealing with
largely different time scales, it is slow and not easy to handle.
CALCULATION OF STABILITY MARGIN IN HE-II

A CICC operating in superfluid helium is most efficient if it
is designed to take advantage of the large heat removal capability at the strand wetted surface. As we discussed in the
description of the general features of the stability margin in
He-II, it is possible to operate the cable at a current density
significantly higher than in the case of supercritical helium.
However, we recall that in the superfluid well-cooled regime,
the upper stability margin is determined by the helium enthalpy available between operating temperature and the
lambda point, of the order of 200 mJ/cm3 of helium volume.
Therefore, in the design for operation in He-II, we assume
implicitly that if the Joule heating is enough to drive the bulk
temperature above T, cooling in He-I is so reduced in relation
to superfluid cooling that the conductor will not recover.
95
As for operation in supercritical He-I, the stability margin

of a CICC in He-II can be computed at different levels of approximation and complexity. We need in this case to modify
the heat transfer coefficient (see the Appendix) and the helium energy balance to take into account the equivalent thermal conductivity given by Eq. (4). However, the model used
in practice to design CICCs for stable operation in He-II is
different from that discussed in the previous section. In this
case, we concentrate on the heat removal capability, with the
aim of maintaining it sufficiently high so that the full helium
heat sink up to the lambda point is available for stabilization.
The essence of the simplified model, due to Dresner
(36,37), consists in solving the 1-D heat transport equation in
stagnant He-II in a channel in order to obtain the effective
cooling capacity. The model implies that even if the helium
goes through the lambda transition at the conductor surface,
the bulk of the coolant remains in the superfluid state and
cooling is determined by conduction along the channel (in
the case of a CICC, the channel is the imaginary annulus
of helium surrounding each strand). This is consistent with
Seyferts observation that the layer of He-I around each
strand is negligibly thin (33). Dresners model is based on the
analytical solution of the nonlinear heat conduction in the
annulus of He-II around the strands, and the ability of the
helium to absorb the heat flux stemming from the Joule heating (for details of the derivation see Refs. 36 and 37). In that
context, it is useful to define the following two quantities:
ACu
E
pw

Eo = hHe (T ) hHe (Top ) L
E=
(11a)
(11b)
where the quantity E is just a different scaling of the stability

margin (the quantity of interest in this calculation), hHe is the
helium enthalpy per unit volume, and L is the effective channel length, so that Eo represents the maximum enthalpy
available between the operating temperature and the lambda
temperature in the annulus of helium around each strand
(the total heat sink). The equivalent channel length L depends on the helium cross section as follows:
pw L = AHe =
f He
A
1 f He St
(11c)
where f He is the CICC void fraction. The analytical solution of

the Dresner model leads to a relationship between the stability margin and the design current density that can be expressed using the nondimensional groups E/Eo and qj /q*, with
the latter terms defined as follows:
qj =
q =
CuI 2
pw ACu
1/3
KCHe
(T Top )2/3
(4Eo )1/3
(11d)
(11e)
in which qj represents the Joule heating in the cable per unit

of cooling surface. The quantity q* is a fiducial heat flux that
depends on the effective thermal conductivity of superfluid
helium (K), the volumetric specific heat of the helium (CHe),
the temperature difference between the operating point and
the lambda transition, and Eo (defined above). This quantity
96
E/Eo
0.1
0.01
0.1
1.0
qj/q*
10.0
Figure 8. Semiempirical representation of the stability margin for a

CICC operating in He-II.
stems from the solution of the nonlinear heat conduction

equation for He-II. The ratio qj /q* determines the severity of
the Joule heating pulse that needs to be transported and absorbed by the superfluid helium.
The model predicts an expression relating E/Eo as a function of qj /q*, which is shown in Fig. 8. The results of this
stability model indicate that, as in He-I operation, there are
two distinct regimes: at low currents the conductor is wellcooled and essentially the entire helium enthalpy margin (up
to the transition to He-I) is available to stabilize the cable; at
higher currents the cable becomes ill-cooled and the stability
margin is significantly lower because of the inability of the
He-II to conduct the Joule heat flux over the length of the
equivalent channel.
The criterion used in practice then stems directly from the
results of this model. To ensure the highest possible stability
margin (the entire helium enthalpy from the operating point
to the lambda point), the conductor needs to be designed so
that qj 0.2q*, for at higher fractions of q* the margin decreases rapidly. Note that as a result, the conductor design
(and the stability margin) depends only on the physical properties of the He-II and on the current density in the stabilizer.
The current sharing temperature does not enter into the determination of stability as a result of assuming that the conductor cannot recover once the lambda transition is crossed
during a disturbance. The present model has been successfully used in the design of a large (200 kA) CICC for application in SMES (39,40).
CONCLUSIONS AND RESEARCH DIRECTIONS
This article has presented the basic considerations and models that go into the design of CICCs for stable operation.
Enough is known of the mechanisms determining stability so
that, in conjunction with other constraints, CICCs can be designed successfully and even optimized (41). However, this
does not mean the field is not open to new areas of research.
As new magnet designs are proposed, and as more stringent
requirements are imposed on the designer, areas of further
study continue to openin particular, work toward improving our understanding of stability under transient operating
conditions (e.g., ac losses and stability) or for more complex
cable geometries (e.g., CICC with central channels).
As may be clear after review of the literature, stability depends in a synergistic manner on the dc and ac operating conditions of the cable in the coil (4244). This is the main direc-
tion of the actual research in the field of CICC stability. In

particular, in view of the applications to pulsed magnets, the
interaction of stability, current distribution, and ac losses in
the cable is one of the main topics. The so-called ramp-rate
limit of operation for pulsed magnets (4345) (a decrease in
the maximum achievable current at increasing field change
rate) is an outstanding example of this synergistic interaction. The appearance of such a phenomenon, explained so far
in terms of nonuniform current distribution and a degradation of the stability margin of the cable, has alerted us to
the difference between dc stability, with constant operating
current and background field, and ac stability of the cable.
However, while dc operating conditions are easier to produce and simulate, ac stability is difficult to measure and
poses some basic problems in the interpretation of the data.
The simulation and prediction of ac stability are therefore object of an intense activity in the field of transient electromagnetics in superconductors and thermohydraulics.
APPENDIX: TRANSIENT HEAT TRANSFER
The main issue in heat transfer is fast transients, for their
relevance to stability. A strong variation of the transient heat
transfer to a flow of supercritical helium was demonstrated
by Giarratano (46) and Bloem (47) in dedicated measurements on short test sections. The experiments showed an initial peak in the heat transfer coefficient at early times, below
1 ms. At later times, in the range of some ms to about 100
ms, the initial peak decreased approximately with the inverse
of the square root of time. This behavior could be explained
in terms of the diffusion of heat in the thermal boundary
layer. Using the analytical solution of diffusion in a semi-infinite body (the helium) due to a heat flux step at the surface,
the effective heat transfer coefficient can be computed as
[Bloem (47)]:
1
ht =
2
KHe cp
t
(12)
where KHe is the heat conductivity of helium. The expression

above is shown to fit properly the experimental data for times
longer than 1 ms and until the thermal boundary layer is
fully developed. At earlier times, Eq. (12) would tend to predict an exceedingly high heat transfer coefficient, consistent
with the assumptions of the analytical calculation. In reality,
the early values of h are found to be limited by the Kapitza
resistance (48) at the contact surface of strand and helium,
which gives a significant contribution only when the transient
heat transfer coefficient is in the order of or larger than 104
W m2 K (or in the case that the wetting helium is in the
superfluid state). A suitable expression for the heat transfer
coefficient in the Kapitza resistance can be obtained using
2
2
hK = 200(TSt
+ THe
)(TSt + THe )
(13)
which, in fact, approximates a radiation-like phenomenon at

the conductor surface with an equivalent heat transfer coefficient (TSt and THe are the strand and the helium temperature, respectively).
At later times, usually around 10 to 100 ms, the thermal
boundary layer is fully developed and the steady state value
of h is approached. Its value appears to be well approximated

by a correlation of the Dittus-Boelter form, as shown by
Yaskin (49) and Giarratano (50). A best fit of the available
data is obtained with the following expression (neglecting corrections due to large temperature gradients at the wetted surface):
hs = 0.0259
KHe 0.8 0.4

Re Pr
Dh
(14)
An empirical expression for the heat transfer to supercritical helium during a transient finally can be obtained modeling the Kapitza resistance and the helium boundary layer as
series thermal resistances, and taking

ht h K
hs hK
h = max
,
(15a)
ht + h K h s + h K
This expression is in good agreement with the experimental
results quoted above, and shows that, for short pulses, the
heat transfer coefficient only depends on the helium state and
not on the flow conditions. At temperatures below the lambda
point (superfluid helium), the Kapitza resistance is the only
limit to the heat transfer at the strand wetted surface. In this
case, we approximate the heat transfer coefficient simply as
h = hK
(15b)
Equations (15a) and (15b) can then be used to approximate

the heat transfer coefficient at the strand-helium interface for
the entire transient in the supercritical regime, or to cover
the entire operating temperature range, from below the
lambda point to 4.2 K and above. It should be noted, however,
that the expressions above are only a convenient tool to implement in stability models covering most regimes of interest,
and do not capture all the complexities encountered, for instance, when operating in atmospheric superfluid, when
phase changes may be taking place at the interface during
a transient.
BIBLIOGRAPHY
1. Z. J. J. Stekly and J. L. Zar, Stable superconducting coils, IEEE
Trans. Nuc. Sci., 12: 367, 1965.
2. V. E. Keilin, Explanation of main features of superconducting
windings training by balance of acting and permissible disturbances, IEEE Trans. Appl. Supercond., 3: 297300, 1993.
3. V. E. Keilin, An approach to deduce quench fields of superconducting magnets from model coils testings, IEEE Trans. Magn.,
30: 19621965, 1994.
4. M. Morpurgo, The design of the superconducting magnet for the
Omega project, particle accelerators, vol. I, Glasgow, UK: Gordon and Breach Publishers (printed by Bell and Bain Ltd.), 1970.
5. P. F. Chester, Superconducting magnets, Reports on Progress in
Physics, XXXII: 561614, 1967.
6. M. O. Hoenig and D. B. Montgomery, Dense supercritical helium
cooled superconductors for large high field stabilized magnets,
IEEE Trans. Magn., 11: 569, 1975.
7. M. O. Hoenig, Y. Iwasa, and D. B. Montgomery, Supercriticalhelium cooled bundle conductors and their application to large
superconducting magnets, Proc. 5th Int. Conf. Magn. Tech., Rome,
Italy, 2125 Apr. 1975, in N. Sacchetti, M. Spadoni, and S. Stipcich (eds.), Frascati Lab. Naz. del CNEN, 519, 1975.
97
8. M. O. Hoenig et al., Supercritical helium cooled cabled, superconducting hollow conductors for large high field magnets, Proc. 6th
Int. Cryo. Eng. Conf., Grenoble, France, 1114 May 1976, in K.
Mendelssohn (ed.), Guilford, Surrey 1PC Science and Technology
Press, 310, 1976.
9. L. Dresner, Stability-optimized, force-cooled, multifilamentary
superconductors, IEEE Trans. Magn., 13: 670, 1977.
10. L. Dresner and J. W. Lue, Design of forced-cooled conductors for
large fusion magnets, Proc. 7th Symp. on Eng. Probs. of Fus. Res.,
I: 703, 1977.
11. L. Dresner, Stability of internally cooled superconductors: A review, Cryogenics, 20: 558, 1980.
12. L. Dresner, Superconductor stability 1983: A review, Cryogenics,
24: 283, 1984.
13. M. O. Hoenig and D. B. Montgomery, Cryostability experiments
of force cooled superconductors, Proc. 7th Symp. on Eng. Probs. of
Fus. Res., I: 780, 1977.
14. M. O. Hoenig, D. B. Montgomery, and S. J. Waldman, Cryostability in force cooled superconducting cables, IEEE Trans. Magn.,
15: 792, 1979.
15. M. O. Hoenig, Internally cooled cabled superconductorsPart I,
Cryogenics, 20: 373389, 1980.
16. M. O. Hoenig, Internally cooled cabled superconductorsPart II,
Cryogenics, 20: 427434, 1980.
17. J. R. Miller et al., Measurements of stability of cabled superconductors cooled by flowing supercritical helium, IEEE Trans.
Magn., 15: 351, 1979.
18. J. W. Lue, J. R. Miller, and L. Dresner, Stability of cable-in-conduit superconductors, J. Appl. Phys., 51: 772, 1980.
19. J. W. Lue and J. R. Miller, Parametric study of the stability margin of cable-in-conduit superconductors: Experiment, IEEE
Trans. Magn., 17: 1981.
20. L. Dresner, Parametric study of the stability margin of cable-inconduit superconductors: Theory, IEEE Trans. Magn., 17: 753,
1981.
21. L. Dresner, Heating induced flows in cable-in-conduit conductors,
Cryogenics, 19: 653, 1979.
22. J. R. Miller et al., Stability measurements of a large Nb3Sn forcecooled conductor, Adv. Cryo. Eng., 26: 654, 1980.
23. J. W. Lue and J. R. Miller, Performance of an internally cooled
superconducting solenoid, Adv. Cryo. Eng., 27: 227, 1982.
24. J. V. Minervini, M. M. Steeves, and M. O. Hoenig, Experimental
determination of stability margin in a 27 strand bronze matrix,
Nb3Sn cable-in-conduit conductor, IEEE Trans. Magn., 21: 339,
1985.
25. J. R. Miller, Empirical investigation of factors affecting the stability of cable-in-conduit superconductors, Cryogenics, 25: 552,
1985.
26. T. Ando et al., Investigation of stability in cable-in-conduit conductors with heat pulse duration of 0.1 to 1 ms, Proc. 11th Int.
Cryo. Eng. Conf., 756, 1986.
27. J. C. Lottin and J. R. Miller, Stability of internally cooled superconductors in the temperature range 1.8 to 4.2 K, IEEE Trans.
Magn., 19: 439, 1983.
28. J. H. Schultz and J. V. Minervini, Sensitivity of energy margin
and cost figures of internally cooled cabled superconductors
(ICCS) to parametric variation in conductor design, Proc. 9th
Magn. Tech. Conf., 643646, 1985.
29. J. W. Lue, Review of stability experiments on cable-in-conduit
conductors, Cryogenics, 34: 779786, 1994.
30. E. A. Chaniotakis, Energy margin of cable-in-conduit conductor
as a function of operating pressure and initial heated zone, IEEE
Trans. Magn., 32 (4): 29662969, 1996; also E. A. Chanionakis,
98
SUPERCONDUCTORS, STABILIZATION AGAINST FLUX JUMPS

Effect of operating pressure and heated length on the stability of
CICCs, J. Fusion Energy, 14 (1): 6973, 1995.
31. G. Claudet et al., Superfluid helium for stabilizing superconductors against local disturbances, IEEE Trans. Magn., 15: 340,
1979.
32. S. W. Van Sciver, Stability of superconductors cooled internally
by He II heat transfer, Cryogenics, 31: 516, 1991.
33. P. Seyfert, J. Laffarranderie, and G. Claudet, Time-dependent
heat transport in subcooled superfluid helium, Cryogenics, 22:
401, 1982.
34. C. Meuris et al., Transient stability of superconductors cooled by
superfluid helium at atmospheric pressure, Proc. Int. Inst. Refrigeration, 215223, 1981.
35. C. Meuris, Experimental study of the stability of a superconductor cooled by a limited volume of superfluid helium, IEEE Trans.
Magn., 19: 272, 1983.
36. L. Dresner, A rapid, semiempirical method of calculating the stability margins of superconductors cooled with subcooled He-II,
IEEE Trans. Magn., 23: 918921, 1987.
37. L. Dresner, Stability of Superconductors, New York: Plenum
Press, 1995.
38. C. Schmidt, Stability of superconductors in rapidly changing
magnetic fields, Cryogenics, 30: 501, 1990.
39. D. L. Walker et al., SMES conductor design, IEEE Trans. Magn.,
25: 1596, 1989.
40. S. D. Peck, J. C. Zeigler, and C. A. Luongo, Tests on a 200 kA
Cable-in-Conduit Conductor for SMES Application, IEEE Trans.
Appl. Supercond., 4: 199, 1994.
41. L. Bottura, Stability, protection and ac loss of cable-in-conduit
conductorsA designers approach, Fus. Eng. Des., 20: 351
362, 1993.
42. E. Tada et al., Downstream effect on stability in cable-in-conduit
superconductor, Cryogenics, 29: 830, 1989.
43. N. Koizumi et al., Experimental results on instability caused
by non-uniform current distribution in the 30 kA NbTi
Demo Poloidal Coil (DPC) conductor, Cryogenics, 34: 155,
1994.
44. N. Koizumi et al., Current imbalance due to induced circulation
current in a large cable-in-conduit superconductor, Cryogenics,
36: 409418, 1996.
45. S. Jeong et al., Ramp-rate limitation experiments using a hybrid
superconducting cable, Cryogenics, 36: 623629, 1996.
46. P. J. Giarratano and W. G. Steward, Transient forced convection
heat transfer to helium during a step in heat flux, Trans. ASME,
105: 350357, 1983.
47. W. B. Bloem, Transient heat transfer to a forced flow of supercritical helium at 4.2 K, Cryogenics, 26: 300308, 1986.
48. S. Van Sciver, Helium Cryogenics, New York: Plenum Press,
1979.
49. L. A. Yaskin et al., A correlation for heat transfer to superphysical (sic) helium in turbulent flow in small channels, Cryogenics,
17: 549552, 1977.
50. P. J. Giarratano, V. D. Arp, and R. V. Smith, Forced convection
heat transfer to supercritical helium, Cryogenics, 11: 385393,
1971.
LUCA BOTTURA
CERN
CESAR LUONGO
Bechtel
98
SUPERCONDUCTORS, STABILIZATION AGAINST

FLUX JUMPS
INTRODUCTION
Superconducting conductors used for ac and dc magnets are
usually composite conductors consisting of many intrinsically stable fine superconducting filaments in a normal
metal matrix.
It is well known that a simple twisting of the wire as a
whole cannot ensure an equal current distribution between
each filament or each layer of filaments. It appears that during energization the current fills first the outer layers of filaments up to the critical current density, whereas hardly any
current flows in the inner layers.
It has already been shown that this particular current distribution with the field totally shielded from the interior of
the wire could result in instabilities and premature quenching of large multifilamentary wires as it does in a large monofilament. It has already been observed and explained how a
highly conductive matrix can slow the process and eventually
totally damp the flux jump. Several laboratories have also observed how highly resistive matrix wires could hardly be stable. This was not very surprising and could be explained in
agreement with the theories of stability of wires made with
many filaments in homogeneous matrix (copper or cupronickel).
This work was mainly devoted to superconducting wires
made of filaments embedded in a resistive matrix (such as
CuNi to reduce coupling current losses) with an outer shell of
low resistivity (such as Cu) to provide enough stability. In
order to help our understanding, this study was carried out
in line with previous work on stability (16). It concerns the
so-called intrinsic stability. We have shown how most of the
classical criteria (usually unrelated) can be integrated into
one unique and general expression. In particular, in the case
of superconducting composites with a highly resistive matrix,
we investigated how the different physical parameters can be
optimized for achieving stabilization.
Intrinsic stability of a superconducting composite refers, in
this approach, to hampering the development of flux jumpings, the origins of which lie in the nonuniform current distribution inside the wire. This means that a wire is intrinsically
stable if a small perturbation does not lead the conditions
to enter in a diverging spiral, T Jc B E Q
T, caused by the decrease of critical current Jc with respect
to temperature T.
The sources of perturbations are mainly of internal origin
(change in current, mechanical hysteresis) or, by extension,
any small change in temperature due to an external source
(field change, for instance).
In the last section of this article, a more complete discussion is given about what can be considered to be a small
perturbation with respect to the actual relationship between
Reprinted from P. Maccioni and B. Turck, A general criterion for intrinsic stabilization of superconducting composites with highly resistive matrix, 00112275/91/03019212. 1991 ButterworthHeinemann Ltd, Cryogenics, 1991 March, pages 192203, with permission from Elsevier Science.
electric field and current density. Thus, intrinsic instability is

directly ascribed to the nonuniform current distribution inside the wire (a direct consequence of superconductivity),
which stores a source of magnetic energy, and the decrease in
critical current density Jc with respect to temperature T.
Two ways of preventing this phenomenon are known: by
absorbing heat generation itself (adiabatic stabilization) and
by removing it by using the enthalpy of external materials or
coolants (dynamic stabilization). In fact, the second process is
different from the first only in that the flux jump has slowed
enough to provide time for heat to be transferred to the coolant. This results only in obtaining more available enthalpy
than the enthalpy of the wire itself, to absorb the magnetic
energy stored in the system.
FUNDAMENTAL EQUATIONS AND CALCULATIONS
For a given transport current It less than the critical current
Ic, the current density and field profile are shown in Fig. 1.
(The transport current It and the background field Ba are kept
constant.) The critical current density is a unique function of
temperature. For simplicity, the dependence on temperature
is taken to be linear:
Jc = Jc0
Tc T
Tc Tb
Jc = 0
for T Tc
for T > Tc
99
In previous work (devoted to copper matrix composites),

the resistivity of the normal metal matrix was taken into account by the expression (5)
E = m (J s Jc )
where m is the effective or average resistivity of the wire in
the critical state, J is the average current density in the wire,
and s is the superconductor space factor.
In contrast to the previous work, we now consider the case
of a multifilament wire made of two different regions: a central region consisting of the superconducting filaments in a
highly resistive matrix (CuNi), for which a justified assumption is made in a later step in the calculations that the resistivity of this central zone is infinite, and an outer normal region with good electrical conductivity (Cu).
Calculations are derived for a slab model. Owing to the
self-field effect, for a given transport current It, the field penetrates to radius as as shown in Fig. 1, with
i=
It
as
=1
Ic
af
where af is the limit of the filamentary region and a is the

outer dimension of the wire (a R/2 for equivalence
with a round wire).
According to Fig. 1, there are three zones of study:
0 x as
In the frame of a simplified critical model, Jc is a step function versus electric field:
as x af
af x a
J(E = 0) = 0
J(E = 0) = Jc
The effects of more realistic expressions of current density,
such as Jc0 Jc0(E/E0)1/n, are discussed in the final section.
nf Jco
nf Jco
A small temperature perturbation dT in the composite results

in a magnetic flux motion together with an induced electric
field. The deviations from the steady state are related by two
electromagnetic equations and one thermodynamic equation
in each region.
In the central region (filamentary zone with no current)
H
=0
t
J=0
Cf
as
as
a af
as
as
af a
T
= div(K grad T )
t
In the central region carrying the current,
H
t
Jc
H
= f
rot
t
t
T
Cf
= f Jc E + div(K grad T )
t
rot E = 0
In the outer copper shell,
Without Cu shell
(homogeneous composite)
With Cu shell
(shell effect)
Figure 1. Schematic representation of current distribution and field

profiles in a composite before and after a temperature perturbation.
rot E = 0
rot H = Jn
Cn
H
t
T
= n Jn2 + div(Kn grad T )
t
100
For simplicity, the physical properties of materials are assumed to be independent of temperature. Equivalent and average characteristics are derived for each zone after a homogenization that takes into account the local structure of the
composite strand.
In each zone, according to the differential equations, the
solution for the temperature can be developed in the form of
a sum of terms with time and space, for separate variables:

T (x, t) = Xi (x) exp(i t/ )
(1)
where
Cf a2f
Kf
(2)
A dimensionless differential equation with respect to space

can be derived in the filamentary region:
X
(4)
(1 )X
(2)
+ ( )X = 0
(3)
must be equal to zero. From Eq. (1), it can be inferred that

any positive value leads to an irrevocable increase of temperature with time. The first positive value for determines
the limit between stability and instability (see also Refs. 2
and 3).
Our calculations were carried out in the same way as in
previous papers (27). One of the goals of this work was to
find an analytical expression for the stability conditions that
included the parameter and other physical characteristics.
As stability is violated for the very first positive value, a
power-series expansion of the hyperbolic functions can be
made. Let us set the dimensionless parameters
=
KnCf
,
KfCn
2 2
a
0 s2 Jc0
Cf (Tc Tb )
(4)
Kf 0
Cf f
(5)
=
4(1 + h )
4h

4
1
3
i3
en
,
af
h=
ht af
Kf
1+
h
2

(9)
From our final expression Eq. (9), we can find a condition for
when an instability occurs ( 0). Conversely, a general
stability criterion can be written in the form
D f
Dmf
<
is the ratio of the thermal diffusivity over the magnetic diffusivity.

In composites with a highly resistive matrix, is much less
than 1, which allows a further simplification of the differential equation
(6)
In the outer copper shell, if the Joule heating is neglected, the

heat equation becomes
X =0
GENERAL STABILITY CRITERION
Note that sa f af and that
X
nCf
,
0 Kf
4
2 4
2 2
+ 2 +
i
i
+h 1 +
X (4) X (2) X = 0
When tends to zero, it can be written in the form
with the fundamental parameters
Kn
,
Kf
(7)
A + (A2 + 4hB)1/2
2B
(10)
with
A= 4 1+
B=
4
h
1+
3
+2
h
(2i + ) + 2h + 4
i3 +
1+
h
2
h 4
i
2
(11)
(12)
For greater clarity, it is convenient to present criterion Eq.

(10) in terms of an energy balance. The magnetic energy released by the flux jump should be less than the available enthalpy in the system (composite and exchange to the surrounding helium layer)
0 (s Jco a)2 3
i < Cf (Tc Tb )[1 + f (ht , K, C, , . . .)]
3
with
=
Kn Cf
Cn Kf
(8)
For a given constant transport current, there are eight boundary conditions among the three zones concerning the electric
field, magnetic field, temperature, flux, and heat transfer to
the outer coolant. The general solutions of the differential
equations are combinations of eight hyperbolic functions, the
coefficients of which can be derived from the set of eight linear equations.
In order to ensure self-consistency and to obtain a nonunique trivial solution for , the determinant of the system
where f is a function containing all the extra terms issuing

from Eq. (10). To confirm the general character of this criterion for composites with a highly resistive matrix, we can
show how some usual criteria can be found, at the cost of a
few approximations.
We must bear in mind in any case that the first assumption that has been made in our calculations is 0 in the
filamentary zone (f ).
Let us recall a few typical orders of magnitude for physical
properties (Table 1). It can be seen that thermal and electrical
diffusivities are exactly permutable for the inner region and
the outer shell.
101
Table 1. Physical Properties of the Materials of Superconducting Composites

Zone
K (W mK1)
C (J m3 K1)
( m)
D (m2 s1)
Dm (m2 s1)
1.0
300
1500
1000
3 107
3 1010
6 104
0.3
0.25
2.5 104
NbTi, CuNi
Cu
It takes 0.5 s to diffuse heat over 0.5 mm in the copper

or magnetic flux in the CuNi matrix, whereas it takes a much
longer time (0.3 ms) to diffuse magnetic flux in the copper or
heat in the CuNi matrix. This means that the process in the
filamentary region is almost locally adiabatic. There is hardly
any current generated in the inner core due to the fact that
at the initiation of the flux jump the electric field in the inner
core is zero when it is at a maximum at the interface between
the filamentary region and the outer copper shell. The selffield effect tends to expel the excess current to the periphery.
APPLICATIONS TO A FEW SIMPLIFIED OR USUAL CRITERIA
Adiabatic Criterion
If one assumes ht 0 (no heat exchange with the helium
bath), Eq. (10) becomes
<
3
i3
1+
enCn
af Cf
and using Eq. (11) we obtain
(13)
C = Cf 1 +
C(T
Tb )
0
1/3
enCn
< Cf (Tc Tb ) 1 +
af Cf
enCn
af Cf
(s Jc )1/3 a2/3
A1 =
(14)
enCn
1+
af Cf
Dynamic Criterion
Several workers have proposed analytical expressions for stability criteria of homogeneous composites (68). To our knowl-
af Cf
enCn
4 1+
af Cf

(16)
e
2
enCn
4 1+
af Cf
38 i
(17)
2
enCn
af Cf

0 Kn en
enCn
+ 2i
2+
nCf af
af Cf
i3
3

enCn
+ 2i +
af
e C af Cf e C 2
n n
n n
+2 + 3
+
af Cf
af Cf
Kn
2+ 0
nCf
4 1+
Whereas using a wire with the highest possible critical current density seems important, the gain in stable density is
not so significant, however. On the other hand, the stability
is not as dependent on the size a as is often considered. Although it is true that the stability parameter varies as a2,
the maximum stable average current density varies as a2/3.
For instance, doubling the thickness a results in a stable average current density multiplied by 0.63.
ha
h en
1 + A1 t f + A2 t
Kf
Kn
e C
A2 =
(15)
with
yields
J<
2+3
This criterion indicates the role of the enthalpy of the outer

normal metal shell. With i 1 and en 0 (no shell), we obtain
the usual so-called adiabatic criterion (14). The stable parameter varies with i as 1/i3. A direct application of this
expression is the evaluation of the maximum average current
density J that can be carried in a wire of half-dimension a.
Letting
and
0 (s Jc0 a)2 3
i
3
enCn
0 (s Jc0 a)2 3
i < Cf (Tc Tb ) 1 +
3
af Cf
J = s Jc i
edge, they have always been presented for i 1 (critical current) and for nonzero values for in the case of exact
solutions obtained along a similar approach.
Let us consider a typical composite wire as a guide to justify some approximations, with af 0.7 103 m, en 0.3
103 m, s 0.2, Cu 0.5, CuNi 0.3, and ht 103 W m2
K1. The basic assumption ( 0) is satisfied by f /0
Kf /Cf . To derive a dynamic criterion, it is necessary to assume
that the heat conductivity in the inner region is not negligible (f is kept infinite). The two terms htaf /Kf and hten /Kn are
considered to be much less than 1.
The general criterion can then be written as follows:
2
(18)
Equation (16) emphasizes the respective influences of the inner region and of the outer shell. From Eq. (16), two particular cases can be derived as follows.
1. Letting en 0 (no outer normal shell) and i 1, we find
5 ht af
0 (s Jc0 a)2
< Cf (Tc Tb ) 1 +
3
24 Kf

(19)
This expression is similar to the expression obtained

previously (6) in the case of a homogeneous multifilamentary composite for low values of the ratio :
7 ht a
0 (s Jc0 a)2
< Cm (Tc Tb ) 1 +
3
20 Km

(20)
102
The small discrepancy between the two expressions can

be mainly ascribed to the fact that the calculations performed previously (6) were only correct for 0,
whereas the present calculations were carried out with
0.
2. If we keep en 0 and assume a very low resistivity for
the outer normal metal shell (made of copper, for instance), Eq. (16) becomes
enCn
0 (s Jc0 a)2 3
i < Cf (Tc Tb ) 1 +
3
af Cf
with
Kn
A2 = 0
nCf
e
3+2
n
+ 2i

a
f
h en
1 + A2 t
Kn
(21)
e C
4 1+
af Cf
enCn
af Cf
2
Similarity with the Cryogenic Criterion

Equation (23) contains two terms: the enthalpy of the composite and the enthalpy absorbed by the helium. If the resistivity
of the shell is small enough or the heat transfer large enough,
the enthalpy of the composite becomes negligible, and stability is completely ensured by the transfer to helium.
The criterion becomes
30 ht en
0 (s Jc0 a)2 3
i <
3
4n
(22)
Again we find a simplified criterion that emphasizes the

beneficial influence of a low resistivity n as in the case
of homogeneous composites. This is already visible in
Eqs. (21) and (22). It can be pointed out more clearly
for en /af 1, which permits another simplification step.
It follows that
0 (s Jc0 a)2 3
i
3

enCn
3 0 h t en
< Cf (Tc Tb ) 1 +
+
af Cf
4 n Cf
The enhancement of stability with increase in the Cu shell

thickness is clear. Evidently a good heat transfer is of no help
when no copper can damp or slow the development of the flux
jump. In contrast, with a sufficient copper thickness, time is
provided to take advantage of the heat transfer to the coolant
and therefore for accounting for the enthalpy absorbed by helium in the heat energy balance.
af
+ 2i (Tc Tb )
(25)
Under these conditions, the criterion can be written, to a first

approximation, in the form of generated and exchanged heat
fluxes. For a thin outer copper shell
4n (s Jc0 i)2 a2
< ht (Tc Tb )
9en
(26)
Effect of Thermal Conductivity
e
af
+ 2i

(23)
In the general expression Eq. (10), assuming now a very high

heat exchange to the helium (ht ) and good thermal and
electrical properties of the shell (high Kn and low n), we obtain
We can see that the simplified Eq. (23) of our general

criterion Eq. (10) can also be put in a form similar to
more conventional criteria derived for homogeneous
composites. For instance, for i 1 (and in the frame of
the particular assumptions), we find stability for
3 0 h t en
0 (s Jc0 a)2
en
< (Tc Tb ) Cf + Cn
+
3
af
2 n
3 0 h t a
0 (s Jc0 a)2
< (Tc Tb ) Cm +
3
10 m
(24)
(exact solution in Ref. 6) and

4 ha
0 (s Jc0 a)2
< (Tc Tb ) Cm + 2 0 t
3
40 ht en
nCf
ht af 4
B
i
2Kf
A
when it was determined for homogeneous composites in

previous work
e
(approximate solution in Ref. 8). Both expressions were

given with no outer copper shell.
We can see that the criterion given in Eqs. (23) and (24)
indicates the enhancement of stability by the heat removal to
the coolant due to the shell effect. In addition to the enthalpy of the composite, the third term in Eq. (24) represents
the enthalpy transferred to the helium during the diffusion
time of the current in the copper shell.
hence
A2 4
ht af
B
Kf
In particular, when i 1, it can be found that the condition

for stability becomes
n (s Jc0 a)2 af
<1
8Kf (Tc Tb ) en
(27)
This expression can be compared with the expression that

was regarded as being the dynamic stability criterion:
m (s Jc0 a)2
<1
3Km (Tc Tb )
This criterion, developed by several groups, quoted in Ref. 8,
assumes an infinite heat transfer coefficient to the coolant. As
a result, it is too optimistic in most practical cases. However,
it is correct in that it determines the limit for the existence of
stationary solutions for temperature-dependent critical cur-
pression Q(i) as a function of the current i in the form of a

boundary line between stability and instability for any given
value of the pair (, i).
In order to compare with theoretical and numerical calculations performed previously (6), we give here an analytical
expression for the particular case of a highly resistive matrix
superconducting composite without a stabilizing shell (en 0).
In the general criterion Eq. (10), let us set en 0 and define h (htaf )/Kf . We find
100
Q(i)
Unstable
50
h = 0.0001
h = 1.0
Stable
0.5
i
3 2(h + 2) + [2h2 (2 + i4 ) + 16h(1 + i3 /3) + 16]1/2

i3
8 + 3hi
(28)
This expression can be compared with the expression obtained for the homogeneous composite studied previously (6)
in the case of a low ratio of thermal diffusivity to magnetic
diffusivity ():
Figure 2. Comparison of stability curves for composites with no

outer shell with usual curves given for homogeneous composites.
Dashed lines, this work ( 0) ; solid lines, Ref. 5 ( 0.1).
rent profiles inside the filamentary region. It points out

clearly, in this case, the effect of poor thermal conductivity.
Simplified Analytical Expression for Homogeneous Composites
In the general criterion Eq. (10) written in a dimensionless
form, stability is ensured when Q(i), where
Q(i) =
103
A + (A2 + 4hB)1/2
2B
Figure 2 shows the stability function Q(i) for two values of

reduced heat transfer h. In the present approach 0 because f is infinite. In the previous approach (6), the limiting
case 0.1 was considered. It can be seen that the results
are similar if it is remembered that the theory of the homogeneous composite was developed for a round wire (a correction
coefficient of /4 was applied to the curves plotted previously
(6) to obtain the curves plotted in Fig. 2.
Developing Eq. (28) for i 1 and for small values of the
reduced heat transfer h leads to
The coefficients A and B include all the physical parameters,

the heat transfer, and the reduced transport current. For a
given set of physical parameters, it is possible to plot the ex-
D
Dm
s = 3 1 +
5h
24
[another form of Eq. (19)].
General criterion
(10)
ht
ht = 0
ht af
<<1
Kf
Adiabatic criterion
(14)
Thermal conductivity effect

(27)
and
ht en
<<1
Kn
Dynamic criterion
(16)
en = 0
n 0
en 0
Simplified dynamic criterion
for homogeneous composite
(19)
Simplified dynamic
criterion with shell effect
(23)
Enthalpy of composite << energy absorbed by helium

Cryogenic criterion
(26)
Figure 3. Summary of conditions to deduce usual criteria from the general stability criterion
100
90
80
70
Q(i)
60
50
40
30
20
10
0.00
0.20
0.40
0.60
0.80
1.00
i
(a)
200
100
90
80
70
Q(i)
Q(i)
60
100
50
40
30
20
10
0.00
0.20
0.40
0.60
0.80
1.00
0.00
(b)
(c)
100
90
90
80
80
70
70
60
60
50
40
30
30
20
20
10
10
0.40
0.60
0.80
1.00
0.60
0.80
1.00
0.60
0.80
1.00
50
40
0.20
0.40
i
100
0.00
0.20
Q(i)
Q(i)
104
0.00
0.20
0.40
(d)
(e)
Figure 4. Stability is improved when both copper shell thickness and the heat transfer coefficient attain significant values for copper shell thickness en /af : (a) 0; (b) 0.3; (c) 0.6; (d) 1.0;
(e) 1.3. Lines from top to bottom: h 10, 1, 0.1, 0.01.
Our expression especially developed for the case of an

outer good conducting shell is in perfect agreement with previous models devoted to homogeneous composites.
DETERMINATION OF THE STABLE CURRENT Is IN A

PARTICULAR EXAMPLE
For any given composite there is no special difficulty in plotting the general curve Q(i) of criterion Eq. (10). Then, the
evaluation of the parameter
SUMMARY
At the cost of some particular assumptions, it has been shown
that the general criterion Eq. (10) encompasses the usual criteria that have been already developed in various particular
cases. This can be represented schematically as in Fig. 3.
INFLUENCE OF THE OUTER SHELL THICKNESS

AND OF THE HEAT TRANSFER COEFFICIENTS
A set of curves have been plotted in Fig. 4 using the more
general criterion Eq. (10) directly. The particular case under
investigation corresponds to typical orders of magnitude of
200, 200, and 1.
The thickness of the shell and the heat transfer coefficients
are given in the form of the dimensionless parameters
en /af and h htaf /Kf .
It is clearly seen in Fig. 4(a), for 0, that because of the
low thermal conductivity of the filamentary region, little is
expected in improving the stability from a good heat transfer
to the coolant. Figures 4(a)4(e) show the significant role of
the heat transfer coefficient when is not negligible (for low
values of h, improvement is only provided by the increase in
the equivalent heat capacity of the composite). A noticeable
improvement in stability is provided when both and h simultaneously attain significant values.
The copper fraction in the conductor is given by Cu
/(l ). This expression means that the amount of copper is
totally concentrated in the outer shell. It is, for instance, 23%
for 0.3 and 56% for 1.3. The reduced heat transfer is
about h 1 for ht 103 W3 m2 K1 and af 0.6 103 m.
The influence of the shell thickness is summarized in one
set of curves for h 1 in Fig. 5.
150
100
0 (s Jc0 a)2
Cf (Tc Tb )
leads immediately to the determination of the maximum stable current is.

As a practical example, let us consider a multifilamentary
composite with a cupronickel matrix surrounded by a copper
shell with the two fixed parameters 2R 1.35 mm (diameter)
and s 0.20 (superconductor volume fraction).
In order to point out again the influence of the outer copper, let us present the curves in terms of copper volume fraction (the CuNi is entirely located in the filamentary region)
R2f = (1 Cu )R2
and CuNi = 1 s Cu
The equivalent dimensions used in the slab model are given

by
a=
R and af =
Rf
2
2
respectively. All the parameters intervening in the expressions for Q(i) can be derived as equivalent physical properties.
The expressions Q(i) are plotted in Figs. 6(a)6(e) with the
copper fraction as a parameter for given values of the actual
heat transfer coefficient ht. Up to a heat transfer of 102 W
m2 K1, the composite can be considered to behave as in adiabatic conditions.
In order to determine the stable operating current, we can
proceed as follows. Calculate the parameter
=
0 (s Jc0 a)2
Cf (Tc Tb )
SMALL PERTURBATION AND THE CRITICAL STATE MODEL
50
0
0.0
and then, for a given copper fraction and a given heat transfer
coefficient, determine the maximum operating current. For
instance, Fig. 6(d) shows the evaluation of the stable current
for two amounts of copper for a particular case. In a field of
11 T and if one assumes the strand to be cooled in superfluid
helium to 1.8 K (ht 5 103 W m2), 90, which means
that the transport current can be increased safely up to 60%
of the critical current for 60% of copper, whereas it is 77% of
the critical current for 70% of copper.
200
Q(i)
105
0.2
0.4
0.6
0.8
1.0
i
Figure 5. Influence of the copper shell thickness en /af on stability
curves for h 1. Lines from left to right: 0, 0.3, 0.6, 1.0, 1.3.
The theoretical calculations that have been discussed in the

previous sections assume an ideal critical state model, that
is, the electric field and the current density in the superconductor are related simply by E 0 J Jco. Actually, in
composites made up with a large number of fine filaments, a
continuous dependence of E on J, caused mainly by inhomogeneities, can be observed.
Q(i)
200
100
0.00
0.20
0.40
0.60
0.80
1.00
i
(a)
200
Q(i)
Q(i)
200
100
0.00
0.20
0.40
0.60
0.80
1.00
0.00
0.20
0.40
i
(b)
(c)
0.60
0.80
1.00
0.60
0.80
1.00
Q(i)
200
100
0.00
100
200
Q(i)
106
0.20
0.40
0.60
0.80
1.00
100
0.00
0.20
0.40
(d)
(e)
Figure 6. Stability curves for a 1.35 mm diameter wire with 20% of superconductor. ht (a) 0;
(b) 102; (c) 103; (d) 5 103; (e) 104 W m2 K1. Lines from left to right: Cu 0.3, 0.5, 0.6, 0.7.
107
One of the expressions that is most often accepted is
J n
c
Jc0
where n is of the order of 10100 depending on the field level

and on the quality of manufacture. Under these conditions,
the critical current is defined for a given electric field. For
instance, the conditions E0 105 V m1 and Jc0 2 109
A m2 result in another definition of critical current for E
104 V m1: Jc 2.16 109 A m2 for n 30 and Jc
2.046 109 A m2 for n 100.
One of the consequences is that the concept of a small perturbation is more difficult to define, since there is no step
function in the E(J) relationship. Another way of pointing out
this effect is to compare the apparent resistivity in this resistive critical state with the resistivity Cu of the stabilizing copper.
The differential resistivity is
E
E
=n 0
J
Jc0
J
Jc0
n1
The effective critical current J*c for which this value equals
the copper resistivity can be estimated. For E/J Cu,
Jc
=
Jc0
Cu
Jc0
nE0
1/n1
Assuming Cu 5 1010 m and E0 105 V m1, for n

30
E (V m1)
E = E0
n = 100
n = 30
E0
0
J
Figure 7. How the determination of Jc is affected by the n value.
ent from the distribution given by the ideal critical state

model. The spiral aspect of the filament enables some current
to flow in the unsaturated region (5); after a current increase,
the current decays in the outer layers of filaments because of
the resistive effect (n value), and in a conductor with limited
length, the current distribution is mainly imposed by the
transfer resistances from layer to layer.
To conclude with these considerations, we can describe a
typical perturbation that can obviously give rise to the required electric field for transferring enough current in the
copper (Fig. 8). If we consider that the current-sharing temperature is given by
Tcs Tb = (Tc Tb )(1 i)
Jc = 1.32 Jc0
Jc = 1.07 Jc0
This shows that the critical current density Jc0 defined for E0
has to be noticeably exceeded in order to obtain a stabilizing
effect by the copper matrix.
In order to establish the full relevance to our model, the
perturbation has to generate enough electric field so that J*c
is locally exceeded. This is hardly to be expected from the ac
losses produced by an external changing field or by the selffield of the increasing current itself, except in case of very
high field or current change rate (in less than 10 ms).
A way to cope with this problem is to evaluate the stability
through and i, not in using Jc0 to define but with J*c taken
as 10 to 30% more than Jc0 depending on the n value (Fig. 7).
This will result in a larger value for (20 to 60%) and a
smaller value for i. However, the stable transport current I*s
will be larger
Is = is Ic
As a result, one can see that a low n value could suggest that
the conductor is more stable, when actually it is only the
true critical current that has been underestimated.
Some other factors can give rise to a more stable current
than given in the calculations. They arise from the existence
of a steady-state current distribution (for E 0) that is differ-
and if we consider that a shift in critical current of typically

10% is at least necessary, because of the shape of the E(J)
curve, a small perturbation T 0.1(Tc Tb) is typical for
initiating a flux jump in a wire whose operating parameters
(, h, i, ) just lie on the theoretical curves between stable
and unstable behavior [in other words, Jc0(E0, Tb) can also be
considered as J*c at the temperature Tb T; see Fig. 7].
CONCLUSION
The general analytical criterion that has been presented concerns the intrinsic stability of superconducting composites
E (V m1)
and for n 100
J*
Jc01 Jc02 c
Tb + T
E0
0
J
Jc0
Tb
Jc*(Tb)
Figure 8. Effect of a small temperature perturbation on the E(J)

curve.
108
SUPERCONDUCTORS, TYPE I AND II
with a highly resistive matrix. It has shown consistency and

continuity with previous theoretical results.
Apart from the general character of this criterion, the influence of an outer copper shell on intrinsic stability conditions has been clearly emphasized with the help of some analytical expressions.
The critical stability curve Q(i) for a given composite under precise operating conditions acts as a useful basis to predict its behavior and to determine a safe current for intrinsic stability.
Conversely, our criterion can be used in a straightforward
manner for a composite design. It becomes possible to consider how a modification of physical characteristics (ratios,
relative positions of the materials, etc.) can improve the stability conditions. The general criterion Eq. (10), given in an
analytical form, can be of great help in the optimization of the
structure of low-loss composites.
We have also pointed out the great advantage of locating
the copper at the periphery of the composite in order to take
full advantage of the appearance of a maximum electric field
to help current flow in the copper. On the other hand, copper
around the axis is of almost no help, as the electric field is
always negligible in the unsaturated region. However, the debate is not totally settled. In fact, the self-field effect is reduced and the conditions for a flux jump are less likely when
the filaments are distributed in a concentric layer with no
superconducting region close to the axis (9). The theory has
to be developed for this case. Also undetermined is whether a
good compromise exists with some copper both in the center
and at the periphery.
NOMENCLATURE
a
af
as
Ba
C
Cf
Cm
Cn
Dm
D
E
en
h
ht
i
Ic
Is
It
J
Jc
Jc0
Jn
K
Kf
Composite external radius (slab model) (m)

Multifilamentary zone radius (m)
Saturated zone radius (m)
Background magnetic field
Specific heat of the composite (J m3 K1)
Average specific heat in the filamentary zone
(J m3 K1)
Average specific heat in homogeneous composites (J
m3 K1)
Average specific heat in the normal metal shell (J
m3 K1)
Magnetic diffusivity (m2 s1)
Thermal diffusivity (m2 s1)
Electric field (V m1)
Normal metal shell thickness (m)
htaf /Kf , reduced heat transfer coefficient
Heat transfer coefficient (W m2 K1)
It /Ic, reduced transport current
Critical current (A)
Stable transport current (A)
Transport current (A)
Current density in the composite (A m2)
Critical current density at T for the given B (A m2)
Critical current density at Tb and B (A m2)
Current density in a normal metal region
Thermal conductivity in the composite (W m1 K1
Average thermal conductivity in filamentary zone
(W m1 K1)
Km
Kn
Tb
Tc
Average thermal conductivity in homogeneous

composites (W m1 K1)
Average thermal conductivity in normal metal shell
(W m1 K1)
Helium bath temperature (K)
Critical temperature for a given field (K)
Greek Letters
f
s
Cu
CuNi
f
m
n
Kn /Kf , dimensionless ratio

Dimensionless stability parameter
Critical value of bounding stable and unstable
domains
(Kn /Kf )(Cf /Cn), dimensionless ratio
(n /0)(Cf /Kf ), dimensionless ratio
en /af , reduced thickness of normal metal shell
Superconductor ratio in the filamentary zone
Superconductor ratio in the composite
Copper ratio in the composite
Cupronickel ratio in the composite
Eigenvalue of the system
Magnetic permeability of vacuum
(Kf /Cf )(0 / f ), dimensionless ratio of the diffusivities
Matrix resistivity ( m)
Average composite normal resistivity ( m)
Normal metal shell resistivity ( m)
BIBLIOGRAPHY
1. P. S. Swartz and C. P. Bean, A model for magnetic instabilities in
hard superconductors: adiabatic critical state, J. Appl. Phys., 46:
49914996, 1968.
2. M. N. Wilson et al., Experimental and theoretical studies of superconducting composites, J. Phys. D, 3: 15171532, 1970.
3. M. G. Kremlev, Stability of critical states in type II superconductors, JETP/Lett., 17: 312318, 1973.
4. M. G. Kremlev, Damping of flux jumping by flux flow resistance,
Cryogenics, 13: 132137, 1974.
5. J. L. Duchateau and B. Turck, Self field degradation effect in adiabatic conditions, Cryogenics, 14: 481486, 1974.
6. J. L. Duchateau and B. Turck, Dynamic stability and quenching
currents of superconducting multifilamentary composites under
usual cooling conditions, J. Appl. Phys., 46: 49894995, 1975.
7. R. G. Mints and A. L. Rokhmanov, On the theory of flux jumps in
hard superconductors, J. Phys. D, 9: 22812287, 1976.
8. M. N. Wilson, Superconducting Magnets, Oxford, UK: Clarendon
Press, 1983.
9. B. Turck, Self field effect in round and rectangular multifilament
composites and stability of superconducting coils, Proc. Conf.
ICEC, 6, IPC Science and Technology Press, 497500, 1976.
B. TURCK
P. MACCIONI
Euratom-CEA Association
SUPERCONDUCTING MAGNETS FOR PARTICLE ACCELERATORS AND STORAGE RINGS
743
more elementary the particles, the higher the energy needed

to smash them. Experiments at the proton scale require beam
energies of the order of 1 TeV or more.
The beams of charged particles are produced by accelerator
systems made up of several stages, which progressively raise
the energy. In the largest machines, the last stage of the accelerator chain, usually referred to as main ring, can have a
circumference of several tens of kilometers and is installed in
an underground tunnel. Such a ring is operated in three
phases: (1) injection, during which the beam, which has been
prepared in various preaccelerators, is injected at low energy,
(2) acceleration, during which the beam is accelerated to nominal energy, and (3) storage, during which the beam is circulated at nominal energy for as long as possible (typically up
to 24 h) and is made available for physics experiments. As
mentioned above, there are two types of experiments: (1)
fixed-target experiments, for which the beam is extracted from
the main ring to be blasted against a fixed target, and (2)
colliding-beam experiments, for which two counterrotating
beams are blasted at each other. The breakage products are
analyzed in large detector arrays surrounding the targets or
collision points.
A main ring of a large accelerator system is designed as a
synchrotron-type accelerator, and the beam is circulated on
an ideally circular orbit, which remains the same throughout
injection, acceleration, and storage (1). The charged particles
are accelerated by means of electrical fields and are guided
and focused by means of magnetic fields. The electrical fields
are provided by RF cavities. In large machines, the bending
and focusing functions are separated; the former is provided
by dipole magnets, whereas the latter is provided by pairs of
focusing/defocusing quadrupole magnets (see the discussion
that follows). The magnets are arranged around the ring in
a regular lattice of cells, which are made up of a focusing
quadrupole, a set of bending dipoles, a defocusing quadrupole,
and another set of bending dipoles. During acceleration, the
field and field gradient of the magnets are raised in proportion to particle momentum to maintain the beam on the design orbit and to preserve its size and intensity.
Bending and Focusing Magnets
SUPERCONDUCTING MAGNETS FOR PARTICLE

ACCELERATORS AND STORAGE RINGS
TYPES OF PARTICLE ACCELERATOR
Accelerator Systems
One of the main activities in nuclear and high-energy physics
is the study of internal structures of charged particles. The
research is carried out by smashing particles into pieces and
then analyzing the nature and characteristics of the pieces.
The particles are broken by accelerating them to high moments and either blasting them against a fixed target or colliding them among themselves. To increase the event rate,
the particles are bunched into a high-intensity beam. The
Coordinate System Definitions. Let (O, u, v, w) designate a

rectangular coordinate system, and let (C) be a circle of center O, located in the (u, v) plane and representing the design
orbit of an accelerator ring. Furthermore, let P be a given
point of (C), and let (P, x, y, z) designate a rectangular coordinate system associated with P, such that x is a unit vector
parallel to (OP), y and w are one and the same, and z is tangent to (C) at P. The x axis defines the horizontal direction,
the y axis defines the vertical direction and the z axis corresponds to the main direction of particle motion.
Normal Dipole Magnet. A normal dipole magnet is a magnet, which, when positioned at P, produces within its aperture
a magnetic flux density parallel to the (x, y) plane and such
that
Bx = 0 and By = B1
(1)
where Bx and By are the x and y components of the flux density and B1 is a constant.
744
According to Lorentz law, a charged particle traveling

along the direction of the z axis through the aperture of such
a magnet is deflected on a circular trajectory parallel to the
horizontal (x, z) plane. The trajectory radius of curvature
can be estimated from
E
0.3 q B1
(2)
Here, is in meters, B1 is in teslas, q is the particle charge in

units of electron charge, and E is the particle energy in gigaelectron-volts (GeV). The effect of a dipole magnet on a
beam of charged particles is similar in some respects to that
of a prism on a light ray.
Equation (2) shows that, to maintain a constant radius of
curvature as the particle is accelerated, the dipole field must
be ramped up in proportion to particle energy.
Normal Quadrupole Magnet. A normal quadrupole magnet
is a magnet, which, when positioned at P, produces within its
aperture a magnetic flux density parallel to the (x, y) plane
and such that
Bx = gy
and By = gx
(3)
where g is a constant referred to as the quadrupole field gradient (in teslas per meter).
According to Lorentz law, a beam of positively charged
particles traveling along the direction of the z axis through
the aperture of such a magnet is horizontally focused and vertically defocused when g is positive, and vertically focused
and horizontally defocused when g is negative. In reference to
its action along the x axis on a beam of positively charged
particles traveling in the z direction, a magnet with a positive
gradient is called a focusing quadrupole, while a magnet with
a negative gradient is called a defocusing quadrupole. To obtain a net focusing effect along both x and y axes, focusing
and defocusing quadrupoles must be alternated in the magnet
lattice. For either type of quadrupole magnets, the focal
length f can be estimated from
E
f
0.3qglq
(4)
Here, f is in meters, E is in GeV, q is in units of electron

charge, g is in teslas per meter, and lq is the quadrupole magnetic length in meters. The effect of focusing/defocusing quadrupoles on a beam of charged particles is similar to that of
convex/concave lenses on a light ray.
Equation (4) shows that to maintain f constant as the particle beam is accelerated, the quadrupole field gradient must
be ramped up in proportion to beam energy.
PARTICLE ACCELERATORS AND SUPERCONDUCTIVITY
Why Superconductivity?
Throughout the years, the quest for elementary particles has
promoted the development of accelerator systems producing
beams of increasingly higher energies. Equation (2) shows
that, for a synchrotron, the particle energy is directly related
to the product B1. Hence, to reach higher energies, we must
increase either the accelerator radius or the dipole field (or

both). Increasing the accelerator radius means a bigger tunnel. Increasing the dipole field above 2 T implies the use of
superconducting magnets. The trade-off between tunneling
costs, magnet development costs, and accelerator operating
costs is, since the late 1970s, in favor of using superconducting magnets generating the highest possible field and field
gradient (2).
Superconductivity is a unique property exhibited by some
materials at low temperatures where the resistivity drops to
zero. As a result, materials in the superconducting state can
transport current without power dissipation by the Joule effect. This offers at least two advantages for large magnet systems such as those needed in accelerator main rings: (1) significant reduction in electrical power consumption and (2) the
possibility of relying on much higher overall current densities
in the magnets coils. There are, however, at least three drawbacks in using superconducting magnets: (1) the superconductor generates magnetization effects that result in field distortions that have to be corrected (see section on field quality),
(2) the magnets must be cooled down and maintained at low
temperatures, which requires large cryogenic systems (see
section on magnet cooling), and (3) it may happen that an
energized magnet, initially in the superconducting state,
abruptly and irreversibly switches back to the normal resistive state in a phenomenon referred to as a quench (see
section on quench performance).
The occurrence of a quench causes an instantaneous beam
loss and requires that all or part of the magnet ring be rapidly
ramped down to limit conductor heating and possible damage
in the quenching magnet (see section on quench protection).
Once the quenching magnet is discharged, it can be cooled
down again and restored into the superconducting state, and
the machine operations can resume. A quench is seldom fatal
but is always a serious disturbance. All must be done to prevent it from happening, and all cautions must be taken to
ensure the safety of the installation when it does happen.
Review of Superconducting Particle Accelerators
Tevatron. The first large-scale application of superconductivity was the Tevatron, a proton synchrotron with a circumference of 6.3 km built at Fermi National Accelerator Laboratory (FNAL) near Chicago, IL, and commissioned in 1983 (3).
The Tevatron now operates as a proton/antiproton collider
with a maximum energy of 900 GeV per beam. It relies on
about 1000 superconducting dipole and quadrupole magnets,
with a maximum operating dipole field of 4 T (4).
HERA. The next large particle accelerator to rely massively
on superconducting magnet technology was HERA (Hadron
Elektron Ring Anlage) built at DESY (Deutsches ElektronenSYnchrotron) near Hamburg, Germany, and commissioned in
1990 (5). HERA is an electron/proton collider with a circumference of 6.3 km. It includes two large rings: (1) an electron
ring, relying on conventional magnets (maximum energy: 30
GeV), and (2) a proton ring, relying on superconducting magnets (maximum energy: 820 GeV). The maximum operating
field of the superconducting dipole magnets is 4.7 T (6).
UNK. Since the early 1980s, the Institute for High Energy
Physics (IHEP) located in Protvino, near Moscow, Russia, has
745
Table 1. Selected Parameters of Major Superconducting Particle Accelerators

Laboratory
Name
Circumference (km)
Particle type
Energy/beam (TeV)
Number of dipoles
Aperture (mm)
Magnetic length (m)
Field (T)
Number of quadrupoles
Aperture (mm)
Magnetic length c (m)
Gradient (T/m)
Commissioning
FNAL
Tevatron
6.3
pp
0.9
774
76.2
6.1
4
216
88.9
1.7
76
1983
DESY
HERA
6.3
ep
0.82
416
75
8.8
4.68
256
75
1.9
91.2
1990
IHEP
UNK
21
pp
3
2168
70
5.8
5.0
322
70
3.0
97
undecided
SSCL
SSC
87
pp
20
7944
50
15
6.79
1696
50
5.7
194
cancelled
BNL
RHIC
3.8
heavy ions
up to 0.1 a
264
80
9.7
3.4
276
80
1.1
71
1999
CERN
LHC
27
pp
7
1232 b
56
14.2
8.36
386 b
56
3.1
223
2005
Per unit of atomic mass.

Two-in-one magnets.
c
Quadrupoles come in several lengths.
b
been working on a proton accelerator project named UNK

(Uskoritelno-Nakopitelniy Komplex). The circumference of
the UNK main ring is 21 km for a maximum energy of 3 TeV
in a fixed target mode (7). The maximum operating dipole
field is 5 T (8). A number of superconducting dipole and quadrupole magnet prototypes have been built and cold-tested,
and the tunnel is almost completed, but, given the present
(1998) economic situation in Russia, the future of the machine
is undecided.
SSC. In the mid 1980s, the United States started the Superconducting Super Collider (SSC) project, a giant proton
proton collider with a maximum energy of 20 TeV per beam
(9). The last stage of the SSC complex would have been made
up of two identical rings of superconducting magnets installed
on top of each other in a tunnel with a circumference of 87
km. The maximum operating dipole field was 6.8 T. The project was eventually canceled in October 1993 by decision of the
US Congress, after 12 miles of tunnel had been dug near Dallas, TX, and a successful superconducting magnet R&D program had been carried out (10).
RHIC. Brookhaven National Laboratory (BNL), located on
Long Island, NY, will complete the construction in 1999 on
its site of the Relativistic Heavy Ion Collider (RHIC). RHIC
is designed to collide beams of nuclei as heavy as gold, accelerated in two identical rings to energies between 7 and 100
GeV per beam and per unit of atomic mass (11). Each ring
has a circumference of 3.8 km; the maximum operating dipole
field is 3.4 T (12).
LHC. In December 1994, the European Laboratory for Particle Physics (CERN) approved the construction of the Large
Hadron Collider (LHC) in its existing 27-km-circumference
tunnel located at the Swiss/French border, near Geneva,
Switzerland (13). LHC will be a proton/proton collider with a
maximum energy of 7 TeV per beam. It will have a single ring
of so-called twin-aperture superconducting magnets, housing
within the same mechanical structure, the pipes for two
counterrotating proton beams (14). The maximum operating
dipole field is set at 8.36 T. Commissioning is planned for
2005.
Prominent Features of Superconducting Accelerator Magnets

Selected parameters of the major projects of superconducting
particle accelerators are summarized in Table 1, whereas Fig.
1(ae) presents cross-sectional views of the Tevatron, HERA,
SSC, RHIC, and LHC dipole magnets (15).
The magnets rely on similar design principles, which are
detailed in the following sections. The field is produced by
saddle-shaped coils that, in their long straight sections, approximate cos n distributions of conductors (with n 1 for
dipole magnets and n 2 for quadrupole magnets). The coils
are wound from Rutherford-type cables made of NbTi multifilamentary strands and are mechanically restrained by
means of laminated collars. The collared-coil assembly is
placed within an iron yoke providing a return path for the
(a)
(b)
(d)
(c)
(e)
Figure 1. Cross-sectional views of superconducting dipole magnets

for large particle accelerators (15): (a) Tevatron, (b) HERA, (c) SSC,
(d), RHIC, and (e) LHC.
746
magnetic flux. In the case of the Tevatron, the collared-coil

assembly is cold, whereas the iron yoke is warm. Starting
with HERA, the iron yoke is included in the magnet cryostat,
and the cold mass is completed by an outer shell delimiting
the region of helium circulation. The cold mass of the LHC
magnets includes two collared-coil assemblies within a common yoke. Tevatron, HERA, UNK, SSC, and RHIC magnets
are cooled by boiling helium at 1 atm (4.2 K) or supercritical
helium at 3 atm to 5 atm (between 4.5 K and 5 K), whereas
LHC magnets are cooled by superfluid helium at 1.9 K
(1 atm 0.1 MPa).
(a)
(b)
Superconducting Accelerator Magnet R&D

A number of laboratories are presently involved in R&D work
on high field or high field gradient accelerator magnets.
Among them is Twente University, located near Enschede in
the Netherlands, which, in 1995, cold-tested at CERN a short
model dipole magnet (made with Nb3Sn cable), which reached
11 T on its first quench at 4.4 K (16). Soon after, in early
1997, Lawrence Berkeley National Laboratory (LBNL), located in Berkeley, CA, cold-tested a short dipole magnet
model (also made with Nb3Sn cable), which, after a number
of training quenches, reached a record dipole field of 13.5 T
at 1.8 K (17).
CONDUCTOR AND CONDUCTOR INSULATION
Superconducting Material
The most widely used superconducting material is a metallic
alloy of niobium and titanium (NbTi), with a Ti content between 45% and 50% in weight (18). NbTi is easy to massproduce and has good mechanical properties. It is a type-II
superconductor, with a coherence length of 5 nm, and a London penetration depth , of 300 nm (chapter 2 of Ref. 2).
The upper critical magnetic flux density of NbTi, BC2, can
be estimated as a function of temperature T using

T 1.7
(5)
BC2 (T ) = BC20 1
TC0
where BC20 is the upper critical magnetic flux density at zero
temperature (about 14.5 T) and TC0 is the critical temperature
at zero field (about 9.2 K).
The critical current density of NbTi, JC, can be parametrized as a function of temperature, magnetic flux density, B,
and critical current density at 4.2 K and 5 T, JCref , using (19):
T 1.7

JC (B, T )
B
B
C0
1
=
1
JCref
B BC2 (T )
BC2 (T )
TC0
(6)
where C0, , , and are fitting parameters. (Typical values
for LHC strands are C0 30 T, 0.6, 1.0, and
2.0.) Since the time of the Tevatron, a factor of about 2 has
been gained on the critical current density at 4.2 K and 5 T,
and values in excess of 3000 A/mm2 are now obtained in industrial production (20).
The highest dipole field reached on a NbTi magnet is 10.53
T at 1.77 K (21). Magnet designers consider that this is about
the limit for NbTi and that to produce higher fields, it is nec-
Figure 2. Rutherford-type cable for accelerator magnet: (a) cable

sketch and (b) cross-sectional view of a cable strand.
essary to change the material. The only other material that is

readily available at (small) industrial scale is an intermetallic
compound of niobium and tin (Nb3Sn) belonging to the A15
crystallographic family (18). Nb3Sn presents interesting superconducting properties (e.g., its upper critical field at zero
temperature and zero strain is in excess of 25 T) (22). However, its formation requires a heat treatment at temperatures
up to 700C for times up to 300 h. Furthermore, once it is
reacted, it becomes very brittle, and its superconducting properties are strain-sensitive. Hence, Nb3Sn calls for special fabrication techniques, which, so far, have limited its use. In recent years, significant R&D work has been carried out to
improve the performance of Nb3Sn wires, thanks to the International Thermonuclear Experimental Reactor (ITER) program (23,24).
Although great progress has been made in the development of so-called high-temperature superconductor (HTS),
such as bismuth copper oxides, Bi2Sr2CaCu2Ox and (Bi,Pb)2
Sr2Ca2Cu3Ox, and yttrium copper oxides, YBa2Cu3O7, these
materials are not ready yet for applications requiring lowcost, mass-production, and high-critical current density (25).
Rutherford-Type Cable
Superconducting accelerator magnet coils are wound from socalled Rutherford-type cables. As illustrated in Fig. 2(a), a
Rutherford-type cable consists of a few tens of strands,
twisted together, and shaped into a flat, two-layer, slightly
keystoned cable (26). The strands themselves consist of thousands of superconducting filaments, twisted together and embedded in a matrix of normal metal (18). Except for the cables
used in a few R&D model magnets, the filaments are made of
NbTi, and the matrix is high-purity copper. The strand diameter ranges from 0.5 mm to 1.3 mm, and the filament diameter ranges from 5 m to 15 m. Figure 2(b) presents a crosssectional view of a typical SSC strand.
The small radii of curvature of the coil ends preclude the
use of a monolithic conductor because it would be too hard to
bend. A multistrand cable is preferred to a single wire for at
least four reasons: (1) it limits the piece length requirement
for wire manufacturing (a coil wound with a N-strand cable
requires piece lengths which are 1/N shorter than for a similar coil wound with a single wire), (2) it allows strand-tostrand current redistribution in the case of a localized defect
or when a quench originates in one strand (27,28), (3) it limits
the number of turns and facilitates coil winding, and (4) it

limits coil inductance (the inductance of a coil wound with a
N-strand cable is 1/N 2 smaller than that of a similar coil
wound with a single wire). A smaller inductance reduces the
voltage requirement on the power supply to ramp-up the magnets to their operating current in a given time and limits the
maximum voltage to ground in case of a quench (see quench
protection section). The main disadvantage of using a cable is
the high operating current (over a few thousand amperes),
which requires large current supplies and large current leads.
The main issues for strand design and manufacturing are
(1) copper-to-superconductor ratio, which should not be too
small to limit conductor heating in case of a quench while
achieving a high overall critical current; (2) filament size, to
limit field distortions resulting from superconductor magnetization at low field (see section on field quality); (3) superconductor critical current density, which can be improved by improving pinning and filament uniformity (18) and (4) piece
length.
The main issues for cable design and fabrication are (1)
compaction, which should be large enough to ensure good mechanical stability and high overall current density while leaving enough void (typically a few percent in volume) for liquid
helium cooling; (2) control of outer dimensions to achieve suitable coil geometry and mechanical properties; (3) limitation
of critical current degradation due to strand and filament deformations at the cable edges (29,30), and (4) control of interstrand resistance, which should not be too small to limit
field distortions induced by coupling currents while ramping
(see section on field quality) and should not be too large to
allow current redistribution among cable strands.
The interstrand resistance can be modified by oxidizing or
by coating the strand surface (31,32). Also, a thin, insulating
foil (such as stainless steel) can be inserted between the twostrand layers of the cable (33). The strands used in HERA
and LHC cables are coated with a silver-tin solder, called stabrite. Half of the strands of the Tevatron cable are coated with
stabrite, whereas the other half are insulated with a black
copper oxide, called ebanol. UNK, SSC, and RHIC cables rely
on natural oxidation. Up to now, no foiled cable has been used
in a magnet.
Note that at the end of cabling, the high purity copper of
the strand matrix is heavily cold-worked and that it may require an annealing procedure.
747
K is far worse than that of liquid helium and that it degrades

significantly with increasing temperature (34).
The insulation of Tevatron, HERA, and UNK magnets, of
most SSC magnets, and of the early LHC models is made up
of one or two inner layers of polyimide film, wrapped helically
with a 50% to 60% overlap, completed by an outer layer of
resin-impregnated glass tape, wrapped helically with a small
gap. The inner layer is wrapped with an overlap for at least
two reasons: (1) the polyimide film may contain pinholes that
must be covered (the probability of having two superimposed
pin holes in the overlapping layer is very low) and (2) the
Tevatron experience has shown that it was preferable to prevent the resin impregnating the glass wrap from entering in
contact with the NbTi cable (the energy released by cracks in
the resin is believed to be sufficient to initiate a quench) (Ref.
4, p. 784). The outer layer is wrapped with a gap to set up
helium cooling channels between coil turns. The resin is of a
thermosetting type and requires heat to increase cross-link
density and cure into a rigid bonding agent. The curing is
realized after winding completion in a mold of very accurate
dimensions to control coil geometry and Youngs modulus
(35).
RHIC magnets and the most recent LHC models use a socalled all-polyimide insulation where the outer glass wrap is
replaced by another layer of polyimide film with a polyimide
adhesive on its surface (36). The all-polyimide insulation has
a better resistance to puncture, but the softening temperature
of the adhesive can be higher than the temperature needed to
cure a conventional resin (225C for RHIC-type all-polyimide
insulation compared to 135C for SSC-type polyimide/glass
insulation).
MAGNETIC DESIGN
Field Produced by Simple Current-Line Distributions
Single Current-Line in Free Space. Let (O, x, y, z) designate
a rectangular coordinate system, and let (I, R, ) designate
a current-line of intensity (I), parallel to the z axis, and located at a position s R exp(i) in the complex (O, x, y) plane,
as represented in Fig. 3(a). (The current-line intensity is chosen to be negative to end up with a positive factor in the right
member of Eq. (8).) The magnetic flux density B, produced by
this current-line in free space, can be computed using Biot
and Savarts law. It is uniform in z and parallel to the (x, y)
Cable Insulation
The main requirements for cable insulation are (1) good dielectric strength in a helium environment and under high
transverse pressure (up to 100 MPa), (2) small thickness (to
maximize overall current density in the magnet coil) and good
physical uniformity (to ensure proper conductor positioning
for field quality), (3) retention of mechanical properties over
a wide temperature range (from helium temperature to coil
curing temperaturesee the discussion that follows), and (4)
ability to withstand radiation in an accelerator environment.
In addition, the insulation system is required to provide a
means of bonding the coil turns together to give the coil a
semirigid shape and facilitate its manipulation during the
subsequent steps of magnet assembly. It is also desirable that
the insulation be somewhat porous to helium for conductor
cooling. Note that the dielectric strength of helium gas at 4.2
Iron
Im
I
R
Ry
(a)
(b)
Figure 3. Representations of a single current-line: (a) in a vacuum

and (b) inside a circular iron yoke.
748
The first term (k 0) of the series corresponds to a pure normal dipole field parallel to the y axis. The B2k1 coefficients
are called the allowed multipole field coefficients of this current distribution.
y
y
+I
+I
I
x
I
+I
+I
I
+I
(a)
Octuplet of Current-Lines with Quadrupole Symmetry. Similarly, the magnetic flux density produced by the octuplet of
current-lines represented in Fig. 4(b) is given by
+I
By + iBx =
Figure 4. Examples of current-line distributions with selected symmetries (a) quadruplet of current-lines with an even symmetry about
the x axis and an odd symmetry about the y axis and (b) octuplet of
current-lines with even symmetries with respect to the x and y axes
and odd symmetries with respect to the first and second bisectors.
plane. It can be expanded into a power series of the form (37)

+

(Bn + iAn )zn1
for z = x + iy, |z| < R
for z = x + iy, |z| < R
(12)
where
B4k+2 =
40 I
cos[(4k + 2)]
R4k+2
(13)
The first term (k 0) of the series corresponds to a pure normal quadrupole field whose axes are parallel to the first and
second bisectors. For this current distribution, the allowed
multipole field coefficients are the normal (4k 2)-pole field
coefficients.
(7)
n=1
Two-Dimensional Geometry
where Bx and By are the x and y components of B, and An and

Bn are constant coefficients, referred to as skew and normal
2n-pole field coefficients, given by
Bn + iAn =
0 I
[cos(n i sin(n )]
2Rn
(8)
Single Current-Line within a Circular Iron Yoke. Let us now

assume that the current-line of Fig. 3(a) is located inside a
circular iron yoke of inner radius Ry, as represented in Fig.
3(b). The contribution of the iron yoke to the magnetic flux
density can be shown to be the same as that of a mirror current-line, of intensity (Im), and position sm in the complex
plane, where
Im =
1
I
+1
and sm =
R2y
s
(9)
Here designates the relative magnetic permeability of the

iron yoke, and s* designates the complex conjugate of s. Note
that the mirror image method is applicable only if the iron
yoke is not saturated and as long as its permeability is
uniform.
Quadruplet of Current-Lines with Dipole Symmetry. Using
these expressions, the magnetic flux densitity produced by the
quadruplet of current-lines (I, R, ), (I, R, ), (I, R,
) and (I, R, ), represented in Fig. 4(a), can be estimated from the power series expansion
By + iBx =
B4k+2z4k+1
k=0
(b)
By + iBx =
+

+

Symmetry Considerations. The field computations presented

in the previous section showed that current distributions with
the symmetries of Fig. 4(a) (i.e., even with respect to the x
axis and odd with respect to the y axis) were suitable for generating dipole fields, whereas current distributions with the
symmetries of Fig. 4(b) (i.e., even with respect to the x and y
axes and odd with respect to the first and second bisectors)
were suitable for generating quadrupole fields. Starting from
these premises, the coil geometry can be optimized to obtain
the required dipole or quadrupole field strength within the
magnet aperture. In addition, in most accelerator designs, it
is required that high-order multipole field coefficients be as
small as possible. Hence, the coil geometry optimization is
also carried out to minimize the contributions from nondipole
or nonquadrupole terms.
cos n Coil Designs. The coil geometry most commonly used
for a dipole magnet is composed of the cylindrical current
shells shown in Fig. 5(a). The magnetic flux density produced
by such shells can be computed by dividing them up into
quadruplets of current-lines having the symmetry of Fig. 4(a)
and by summing their contributions over a shell quadrant. It
follows that the magnetic flux density is again given by Eq.
+J
J
J
60
30
B2k+1z
2k
for z = x + iy, |z| < R
+J
(10)
J
k=0
+J
where
(a)
20 I
B2k+1 =
cos[(2k + 1)]
R2k+1
(11)
(b)
Figure 5. Current shell approximations for the generation of

multipole fields: (a) dipole field and (b) quadrupole field.
;y;;
;
y
yy
;
yy
;
y
y;y;y;y;
(10), but the expressions of the multipole field coefficients become

B1 =
20 J
(Ro Ri ) sin 0
and
B2k+1 =
20 J
(2k + 1)(2k 1)
R2k1
i
1
R2k1
o
for k, k 1
(14a)

sin[(2k + 1)0 ]
(14b)
Here, Ri and Ro are the inner and outer radii of the shells, 0
is the pole angle, and J is the overall current density, which
is assumed to be uniform. Note that B3 (first allowed
multipole field coefficient after B1 in a current distribution
with a dipole symmetry) is nil for 0 /3.
Similarly, the coil geometry most commonly used for a
quadrupole magnet is made up of the cylindrical current
shells shown in Fig. 5(b). The magnetic flux density is here
given by Eq. (12), where
B2 =
and
B4k+2
0 J
=
k(4k + 2)
20 J
ln
R
1
1
4k
4k
Ri
Ro
Ri
sin 20
(15a)

sin[(4k + 2)0 ] for k, k 1
749
Figure 6. Conductor and Lorentz force distributions in a quadrant

of a 50-mm-aperture SSC dipole magnet coil (38).
Fig. 6, which shows the conductor distribution in a quadrant

of a 50-mm-aperture SSC dipole magnet coil (the vectors represent the components of the Lorentz force discussed in the
section on mechanical design).
Note that the magnetic flux density produced by the coil of
Fig. (6) can be accurately computed by dividing each turn into
two rows of elementary current-lines parallel to the z axis and
approximately equal in number to the number of cable
strands (Ref. 39, p. 226).
(15b)
Note that B2 corresponds to the quadrupole field gradient g

and that B6 (first allowed multipole field coefficient after B2
in a current distribution with a quadrupole symmetry) is nil
for 0 /6.
By reference to the conductor distribution around the circular inner bore, such coil geometries are referred to as cos
and cos 2 designs. They are very compact and make the most
effective use of conductors by bringing them close to the useful aperture.
Current Shell Approximations. In practice, the current shells
of Figs. 5(a) and 5(b) are approximated by stacking into an
arch the slightly keystoned cables described in the conductor
section. High-field dipole or high-field-gradient quadrupole
magnets usually rely on two coil layers whose contributions
add up. Also, wedges are introduced between some of the coil
turns to separate the conductors into blocks. The blocks
angles are then optimized to eliminate high-order multipole
field coefficients (37).
In the case of Tevatron, HERA, and UNK magnets, the
cable keystone angle is large enough to allow the formation
of an arch with the desired aperture. Furthermore, each coil
turn lies along a radius vector pointing towards the aperture
center. In the case of SSC and LHC magnets, the coil aperture
is reduced to minimize the volume of superconductor. This
results in a keystone angle requirement deemed unacceptable
from the point of view of cabling degradation. Hence, in these
magnets, the cables are not sufficiently keystoned to assume
an arch shape, and the wedges between conductor blocks
must be made asymmetrical to compensate for this lack (38).
Also, the coil turns end up being nonradial, as illustrated in
Iron Yoke Contribution. The coils of particle accelerator

magnets are usually surrounded by an iron yoke, which provides a return path for the magnetic flux while enhancing the
central field or field gradient.
As an illustration, let us place the cylindrical current
shells of Fig. 5(a) within a circular iron yoke of inner radius
Ry. The contribution of the iron yoke to the normal (2k 1)y
pole field coefficient B2k1
can be estimated as (Ref. 2, p. 53)
By2k+1
1
=
+1
Ri Ro
R2y
2k+1
Bs2k+1
(16)
where is the relative magnetic permeability of the iron

yoke, Ri and Ro are the current shell inner and outer radii,
and Bs2k1 is the (2k 1)-pole field coefficient produced by the
current shell alone.
Equation (16) shows that the smaller Ry, the larger the
field enhancement. However, there are two limitations on
how close the iron can be brought to the coils: (1) room must
be left for the support structure, and (2) iron saturates for
fields above 2 T, resulting in undesirable distortions (see section on field quality).
As already mentioned, the Tevatron magnets use a warm
iron yoke (i.e., placed outside the helium containment and
vaccum vessel), but starting with HERA magnets, the iron
yoke is included within the magnet cold mass. For SSC dipole
magnets, the field enhancement due to the iron yoke is of the
order of 20%. In LHC magnets, two coil assemblies (powered
with opposite polarity) are placed within a common iron yoke.
This twin-aperture design results in leftright asymmetries
in the yoke surrounding each coil assembly taken individu-
750
ally. These asymmetries must be taken into account when calculating the field quality.
Operating Margin. Equations 14(a) and 15(a) show that, to
achieve high fields and high field gradients, it is desirable to
maximize the overall current density in the magnet coil. This
can be done by three means: (1) maximizing the superconductor performance, (2) minimizing the copper-to-superconductor
ratio in the cable strands, and (3) minimizing the turn-to-turn
insulation thickness. As explained in other sections, there are
lower bounds on the values of copper-to-superconductor ratio
and insulation thickness in order to limit conductor heating
in case of quenching and to ensure proper electrical insulation. As for the superconductor, the upper limit is the critical
current density at the given temperature and magnetic flux
density.
The magnetic flux density to which the conductor is exposed is nonuniform over the magnet coil, but the maximum
current-carrying capability of the conductor is determined by
the section where the magnetic flux density is the highest. In
most cases, this corresponds to the pole turn of the innermost
coil layer. Let Bp f(I) designate the peak magnetic flux density on the coil as a function of supplied current I, and let
IC f(B) designate the supposedly known cable critical current as a function of applied magnetic flux density B. The
intersection between these two curves determines the maximum quench current of the magnet Iqm.
In practice, magnets must be operated below Iqm so as to
ensure that the superconductor is in the superconding state
and to limit the risks of quenching. Let Iop designate the operating current. Then, the current margin of the magnet mI
is defined as
mI = 1
Iop
Iqm
(17)
The excellent quench performance of the HERA magnets (6)

suggests that the current margin can be set to as little as
10%, but it is safer to aim for 20%.
In comparison to other superconducting magnets, such as
solenoids for magnetic resonance imaging, a current margin
of 10% to 20% is quite small. This implies that accelerator
magnets are operated very close to the superconductor critical
surface and that they are very sensitive to any kind of disturbances that may cause the magnet to cross the critical surface
and lead to a quench.
A peculiarity of a two-layer, cos dipole magnet coil design
is that the peak field in the outermost layer is quite a bit
lower than in the innermost layer. Hence, when using the
same cable and current for both layers, the outer layer is operated with a much higher current margin than the inner
layer, which can be considered as a waste of costly superconductor. SSC and LHC dipole magnet coils use a smaller conductor for the outer layer than for the inner layer. This results in a higher overall current density in the outer layer
and reduces the difference in current margins. Such action is
referred to as conductor grading. The main disandvatage of
grading is that it requires splices between inner and outer
layer cables (which, of course, are connected electrically in
series and only require one power supply).
Helium
containment
shell
Iron yoke
380 mm OD
Collars
45 mm soil
aperture
Bus
work
Helium
passage
Coils
Figure 7. Conceptual block design developed at BNL for a high field,
twin-aperture dipole magnet (41).
Limits of cos n Design. The cos n coil design has been very
successful until now, with a record dipole field of 13.5 T
reached by the LBNL short dipole magnet model (using
Nb3Sn cables at 1.8 K). However, it has two main drawbacks:
(1) the coil ends are difficult to make (see the section on coil
ends), and (2) due to the Lorentz force distribution, a stress
accumulation in the azimuthal direction results in high transverse pressures on the midplane conductors (see Fig. 6). For
very high field magnets, requiring the use of A15 (or even
possibly HTS) superconductors, which are strain-sensitive,
these high transverse pressures can result in significant critical current degradation (40).
Alternative coil designs, which may allow better management of the Lorentz stresses within the magnet coil, are being
investigated. As an illustration, Fig. 7 presents a conceptual
block or window-frame design developed at BNL for a twinaperture dipole magnet relying only on simple, racetrack coils
(41). Note, however, that such designs make less effective use
of superconductor.
Coil End Design
One of the main difficulties of the cos n design is the realization of coil ends. In the coil straight section, the conductors
run parallel to the magnet axis, but, in the coil ends, the conductors must be bent sharply with small radii of curvature to
make U-turns over the beam tube that is inserted within the
magnet aperture. This confers to the coil a saddle shape as
illustrated in Fig. 8.
Beam
B
x
Vacuum pipe
Figure 8. Perspective view of a saddle-shaped coil assembly for a

dipole magnet.
Sophisticated algorithms have been developed to determine the conductor trajectories that minimize strain energy
(42). These algorithms, which often require winding tests to
determine correction factors, are coupled with electromagnetic computations to minimize field distortions. SSC and
LHC magnets use precisely machined end spacers, designed
by the optimization programs, which are positioned between
conductor blocks (43). In addition, the iron yoke does not extend over the coil ends, to reduce the field on the conductors
and ensure that the peak field is located in the coil straight
section where the conductors can be better supported.
Sagitta
To limit the number of coil ends and of magnet interconnects
around the accelerator ring, the arc dipole and quadrupole
magnets are made as long as possible. The circulation of a
charged beam in a dipole magnet, of magnetic length ld, results in an angular deflection of the particle trajectory,
which can be estimated as
l
0.3qB1ld
= d
E
x
(18)
Here, is in radians and ld is in meters, B1 is the dipole magnetic flux density in teslas, q is the particle charge in units of
electron charge, and E is the particle energy in GeV.
As a result, long dipole magnets must be slightly bent to
accompany the particle trajectory. This bending, which is implemented in the (x, z) plane, is referred to as sagitta.
FIELD QUALITY
Multipole Expansion
Except near the short coil ends, the magnetic flux density produced in the bore of a particle accelerator magnet can be considered as two-dimensional. The power series expansion of
Eq. (7) is usually rewritten in the more convenient form
By + iBx = Br 104
+

(bn + ian )
n=1
z
Rr
n1
(19)
for z = x + iy, |z| < Ri

where Bx and By are the x and y components of the magnetic
flux density, Rr is a reference radius representative of the
maximum beam size (Rr was 10 mm for the SSC and is now
17 mm for the LHC), Br is the absolute value of the dipole or
quadrupole component at Rr, an and bn are the dimensionless
skew and normal 2n-pole coefficients, and Ri is the coil inner
radius. Note the presence of the 104 scale factor.
Given the symmetries of current distributions in magnet
coil assemblies, and as explained in the previous section, only
selected normal multipole coefficients are expected to be nonzero. These allowed multipole coefficients can be tuned up by
iterating on the electromagnetic design. In practice, however,
nonuniformities in material properties and manufacturing errors result in symmetry violations that produce unallowed
multipole coefficients. For instance, a top/bottom asymmetry
in a dipole magnet produces a nonzero skew quadrupole coefficient (a2), whereas a left/right asymmetry produces a non-
751
zero normal quadrupole coefficient (b2). These unwanted coefficients can be eliminated only by improving material selection, tooling, and assembly procedures.
Field Quality Requirements
From the accelerator point of view, the beam optics is primarily governed by integrated field effects over the magnet ring.
The main field quality requirements are (1) suitable dipole
field integral and small dipole field angle variations [the former to ensure that the integrated bending angle over the
magnet ring is (2) and the latter to ensure that the particle
trajectory is planar], (2) accurate quadrupole alignment and
suitable quadrupole field integral (the former to avoid coupling of particle motions along the x and y axes and the latter
to ensure proper focusing), and (3) small high-order multipole
coefficients (to ensure large beam dynamic aperture). In the
case of high-order multipole coefficients, it is customary to
specify tables of mean values and standard deviations over
the entire magnet population (44). The tables of mean values
are referred to as systematic multipole specifications, whereas
those of standard deviations are referred to as random
multipole specifications. The specified values are all expressed at the reference radius Rr.
In large machines such as SSC or LHC, the dipole and
quadrupole field integrals must be controlled with a relative
precision of the order of 103. The variations in dipole field
angles must be kept within a few milliradians, and the tolerance on quadrupole alignement is of the order of 0.1 mm. Systematic and random multipole specifications are given up to
the 18th or 20th pole and get tighter with increasing pole
order. For SSC magnets at 10 mm, the specifications went
from a few tenths of a unit for low-order coefficients to a few
thousandths of a unit for higher-order coefficients.
Geometric Errors
Types of Geometric Errors. The specifications on multipole
coefficients require that the individual conductors and the
yoke surrounding the coil assembly be positioned with a very
good accuracy (typically a few hundredths of a millimeter in
the two-dimensional cross section). Improper positioning results in geometric errors that distort the central field and produce unwanted multipole coefficients.
The geometric errors can be classified in at least five categories: (1) errors in coil inner and outer radii and in yoke
inner radius; (2) errors in coil pole angle, wedge angle, and
conductor angular distribution; (3) symmetry violations in coil
assembly; (4) centering errors with respect to the iron yoke;
and (5) residual twist of magnet assembly.
Effects of Azimuthal Coil Size Mismatch. A common cause of
geometric error is a mismatch between the azimuthal sizes of
the various coils constituting a coil assembly. Such mismatch
results in displacements of the coil assembly symmetry
planes that produce nonzero, low-order unallowed multipole
coefficients (45). For instance, a mismatch between the azimuthal sizes of the top and bottom coils used in a dipole magnet coil assembly causes an upward or downward displacement of the coil parting planes, which produces a nonzero
skew quadrupole coefficient a2. Similarly, a systematic mismatch between the left and right sides of the coils used in a
dipole magnet coil assembly causes a rotation of the coil part-
752
ing planes, which produces a nonzero skew sextupole coefficient a3. A systematic a2 can be limited by randomly mixing
coil production, whereas the occurrence of a systematic a3 can
be avoided only by correcting tooling.
Iron Saturation
When the field in the iron yoke is less than 2 T, the relative
magnetic permeability of the yoke can be considered as very
large and uniform, and the iron contribution to the central
field increases linearly as a function of tranport current in the
magnet coil. For fields above 2 T, parts of the iron start to
saturate, and their relative magnetic permeability drops. As
a result, the iron contribution becomes a less-than-linear
function of transport current. This relative decrease in iron
contribution appears as a sag in the magnet transfer function
(38). (The transfer function is defined as the ratio of Br to the
transport current.) The transfer function sag can exceed a few
percent in dipole magnets but is usually negligible in quadrupole magnets.
In the case of a single-aperture magnet with a symmetrical
iron yoke, the saturation first occurs in the pole areas producing a positive shift in normal sextupole coefficient b3. At
higher currents, the saturation reaches the midplane areas,
producing a negative shift in b3, which partially compensates
for the effects of pole saturation. The midplane saturation can
be forced to occur sooner by punching notches (i.e., removing
matter) at appropriate locations in the yoke. As an illustration, Fig. 9 presents measurements of b3 as a function of current in the central part of a SSC dipole magnet prototype.
The measurements above 3 kA clearly show the effect of pole
saturation at high currents (the origin of the hysteresis is explained in the next section).
In the case of a twin-aperture dipole, the central part of
the yoke saturates before the outer parts, resulting in left/
right asymmetries in the yoke contributions to each aperture,
Normal sextupole coefficient, b3 (units)
2000
4000
Current (A)
6000
Figure 9. Measurements of normal sextupole coefficient b3 as a function of current in the central part of a SSC dipole magnet showing
the hysteresis resulting from superconductor magnetization and the
distortions at high currents resulting from iron saturation.
which affect the normal quadrupole coefficient b2. The saturation effects in b2 are of opposite sign in the two apertures.
In any case, the iron contribution depends on the packing
factor of the yoke laminations, which must be tightly controlled over the magnet length. Also, the iron yoke must be
carefully aligned to limit magnet assembly twist.
Superconductor Magnetization
Critical State Model. According to the so-called critical state
model, bipolar magnetization currents are induced at the periphery of the superconducting filaments in the cable strands
each time the field to which the filaments are exposed is varied (46). The magnetization currents distribute themselves
with a density equal to the superconductor critical current
density at the given temperature and field JC, in order to
screen the filament cores from the applied field change. Unlike regular eddy currents, the magnetization currents do not
depend on the rate of field variations. Also, because they can
flow with zero resistance, they do not decay as soon as the
field ramp is stopped. They are called persistent magnetization currents.
Effects of Superconductor Magnetization. When an accelerator magnet is cycled in current, the bipolar shells of magnetization currents induced in the filaments behave as small
magnetic moments which contribute toand distortthe
central field. The magnetic moments depend on JC and are
proportional to filament diameter. Their distribution follows
the symmetries of the transport-current field (i.e., the field
produced by the transport current in the magnet coil), and, if
the superconductor properties are uniform, only the allowed
multipole coefficients are affected. Computer models that can
accurately predict the field distortions resulting from superconductor magnetization have been developed (47).
The field distortions are the most significant at low transport current, where the transport-current field is low and JC
is large. They are progressively overcome as the transportcurrent field increases and JC diminishes and become negligible at high transport current. They change sign and regain
influence as the transport current is ramped down. As a result, the allowed multipole coefficients exhibit sizable hystereses as a function of transport current, which depend on magnet excitation history. This is illustrated in Fig. 9, which
shows measurements of b3 as a function of current in the central part of a SSC dipole magnet. In Fig. 9, the magnetization
effects can be seen at currents below 3 kA (as explained in
the previous section, the distortions at high field result from
iron yoke saturation).
The field distortions resulting from superconductor magnetization are one of the major drawbacks of using supeconducting magnets in a particle accelerator. They can be reduced by
reducing filament size (typically, to 5 m for SSC and LHC
strands), but they cannot be eliminated. The powering cycle
of the magnets must be adapted to avoid brutal jumps between the two branches of the multipole coefficient hystereses
while the beam circulates. Also, elaborate beam optics correction schemes must be developed. This can include superconducting, high-order multipole magnets (chapter 9 of Ref. 2).
Time Decay. In addition, the effects of superconductor
magnetization are not indefinitely persistent, but exhibit a
Coupling Currents
Skew sextupole field, A3 (104T)
As described in the conductor section, accelerator magnet

coils are wound from Rutherford-type cables, which consist of
a few tens of strands twisted together and shaped into a flat,
two-layer, slightly keystoned cable. The cable mid-thickness
is smaller than twice the strand diameter, which results in
strand deformation and large contact surfaces at the crossovers between the strands of the two layers. Furthermore,
and as explained in the mechanical design section, the coils
are precompressed azimuthally during magnet assembly.
Large pressures that keep the strands firmly in contact are
thus applied perpendicularly to the cable. The large contact
surfaces and the high pressures can result in low contact resistances at the strand crossovers.
In the steady state, the transport current flows in the superconducting filaments, which offer no resistance. When the
cable is subjected to a transverse varying field, the network
of low interstrand resistances allows the formation of current
loops, which are superimposed on the transport current. The
loop currents, referred to as interstrand coupling currents, circulate along the superconducting filaments and cross over
from strand to strand through the interstrand resistances.
Unlike persistent magnetization currents, the interstrand
coupling currents are directly proportional to the rate of field
variations, and they start to decay as soon as the field ramp
is stopped.
DCA312 4 A/s
DCA312 8 A/s
DCA312 16 A/s
DCA312 32 A/s
2
1
0
1
2
3
2000
4000
Current (A)
(a)
6000
Interstrand coupling currents have three main effects on

magnet performance (39): (1) quench current degradation (for
they are superimposed on the transport current), (2) heat dissipation (when crossing the interstrand resistances), and (3)
field distortions. This last issue is the most critical for accelerator magnet applications.
The coupling current contribution to the central field does
not depend on transport current and increases linearly as a
function of current ramp rate. If the interstrand resistance is
uniform throughout the coil assembly, the coupling current
distribution follows the symmetries of the transport-current
field, and only the allowed multipole coefficients are affected.
In practice, however, there can be large coil-to-coil differences
as well as large nonuniformities within the coils themselves,
which result in sizable effects in the unallowed multipole coefficients. This is illustrated in Fig. 10 (a, b), which presents
plots of the skew and normal sextupole field coefficients (A3
and B3) as functions of current, measured at various ramp
rates in the central part of a SSC dipole magnet prototype.
(Note that the transport-current contribution has been substracted from the data.) No particular treatment (such as
stabnite) was applied to the strands of the cable used in this
prototype.
The effects of interstrand coupling currents can be limited
by ensuring that the interstrand resistances are not too low.
However, and as mentioned in the conductor section, the interstrand resistances should not be too large either to allow
some possibility of current redistribution among cable
strands.
Longitudinal Periodicity
When measuring the field with fine spatial resolution along
the axis of an accelerator magnet, all multipole coefficients
appear to exhibit periodic oscillations (53,54). The amplitude
of the oscillations varies as a function of space, transport current, excitation history, and time, but the wavelength is always approximately equal to the twist pitch length of the cable used in the innermost coil layer.
The longitudinal periodic oscillations are believed to result
from imbalances in the current distribution among cable
Normal sextupole field, B3 (104T)
slow time decay, which, at low transport current, can result in

significant drifts of the allowed multipole coefficients (48,49).
These drifts are particularly disturbing during the injection
phase of machine operation, where the magnet current is
maintained at a constant and low level for some period of
time (50). Also, they complicate the early stages of acceleration, for, as the current is increased at the end of injection,
the drifting multipoles rapidly snap back to values on the hysteresis curves (51). Part of the observed time decay can be
attributed to flux creep in the superconductor (52), but flux
creep cannot account for the large drifts obverved after a high
current cycle (49). The nature of the other mechanisms that
may be involved is not well understood.
753
DCA312 4 A/s
DCA312 8 A/s
DCA312 16 A/s
DCA312 32 A/s
4
2
0
2
0
2000
4000
Current (A)
6000
(b)
Figure 10. Effects of interstrand coupling currents on multipole field coefficients as measured
as a function of ramp rate in the central part of a SSC dipole magnet (39): (a) skew sextupole
field coefficient A3 and (b) normal sextupole field coefficient B3. The transport-current contribution is subtracted from the data.
754
strands. The current imbalances may have at least three origins: (1) nonuniformities in the properties of cable strands,
(2) nonuniformities in the solder joints connecting the coils
in series to the current leads, and (3) large and long-lasting
interstrand coupling current loops superimposed on the
transport current (55). Such current loops could be induced
by spatial variations in the time-derivative of the field to
which the cable is exposed as it turns around the coil ends or
exits toward the current leads (5658).
The oscillation wavelength is too short to affect beam optics but may be an issue for magnetic measurements. It is
recommended that the measurements be averaged over an integer number of cable pitch lengths. Also, the slow decay of
the large interstrand coupling current loops associated with
these periodic oscillations may contribute to the drifts of the
allowed multipole coefficients observed at low and constant
transport current (see the section on superconductor magnetization) (59).
MECHANICAL DESIGN
Support Against the Lorentz Force
Components of the Lorentz Force. The high currents and
fields in an accelerator magnet coil produce a large Lorentz
force on the conductors. In a dipole coil, the Lorentz force has
three main components, which are represented in Fig. 6
(38,60): (1) an azimuthal component, which tends to squeeze
the coil toward the coil assembly midplane [which, in the coordinate system defined previously, corresponds, for a dipole
magnet, to the horizontal (x, z) plane], (2) a radial component,
which tends to bend the coil outwardly, with a maximum displacement at the coil assembly midplane (along the horizontal
x axis for a dipole magnet), and (3) an axial component, arising from the solenoidal field produced by the conductor turnaround at the coil ends and which tends to stretch the coil
outwardly (along the z axis).
axial direction, the coils either are free to expand or are restrained by means of stiff end plates.
The use of laminated collars, pioneered at the Tevatron,
was a real breakthrough in achieving a rigid mechanical support while keeping tight tolerances over magnet assemblies,
which are a few meters in length and which must be massproduced. The laminations are usually stamped by a fine
blanking process allowing a dimensional accuracy on the order of one hundredth of a millimeter to be achieved.
Azimuthal Precompression
Preventing Collar Pole Unloading. As described previously,
the azimuthal component of the Lorentz force tends to
squeeze the coil toward the midplane. At high fields, it may
happen that the coil pole turns move away from the collar
poles, resulting in variations of the coil pole angle that distort
the central field and creating a risk of mechanical disturbances. To prevent conductor displacements, the collars are
assembled and locked around the coils so as to apply an azimuthal precompression. The precompression is applied at
room temperature and must be sufficient to ensure that, after
cool-down and energization, there is still contact between coil
pole turns and collar poles.
Precompression Requirement. To determine the proper level
of room temperature azimuthal precompression, at least three
effects must be taken into account: (1) stress relaxation and
insulation creep following the collaring operation, (2) thermal
shrinkage differentials between coil and collars during cooldown (if any), and (3) stress redistribution resulting from the
azimuthal component of the Lorentz force. In addition, the
collaring procedure must be optimized to ensure that the peak
pressure seen by the coils during the operation (which may
be significantly higher than the residual precompression)
does not overstress the insulation (Ref. 60, p. 1326).
The precompression loss during cool-down, , can be estimated from
Ecl (cl cr )
Stability against Mechanical Disturbances. Because accelerator magnets are operated close to the critical current limit of
their cables, their minimum quench energy (MQE), defined as
the minimum energy deposition needed to trigger a quench,
is very small. As a matter of fact, the MQE of accelerator
magnets is of the same order of magnitude as the electromagnetic work produced by minute wire motions in the coil (61).
If the motions are purely elastic, no heat is dissipated, and
the coil remains superconducting, but if the motions are frictional, the associated heat dissipation may be sufficient to initiate a quench. This leaves two possibilities: either to prevent
wire or coil motion by providing a rigid support against the
various components of the Lorentz force or to reduce to a minimum the friction coefficients between potentially moving
parts of magnet assembly.
Conceptual Design. The mechanical design concepts used
in present accelerator magnets are more or less the same and
were developed at the time of the Tevatron (4,62). In the radial direction, the coils are confined within a rigid cavity defined by laminated collars, which are locked around the coils
by means of keys or tie rods. In the azimuthal direction, the
collars are assembled so as to precompress the coils. In the
(20)
where Ecl is the coil Youngs modulus in the azimuthal direction, and cl and cr are the thermal expansion coefficients of
the coil (in the azimuthal direction) and of the collars, integrated between room and operating temperatures. Note that
Eq. (20) is derived with the assumptions that Ecl does not depend on temperature and that the collars are infinitely rigid.
Choice of Collar Material. To limit cool-down loss, it is preferable to use for the collars a material whose integrated thermal expansion coefficient matches more or less that of the
coil. For NbTi coils with polyimideglass or all-polyimide insulation, this suggests aluminum alloy (see Table 2). HowTable 2. Integrated Thermal Expansion Coefficients between
4.2 K and Room Temperature (103 m/m)
Low carbon steel
Stainless steel (304/316)
Copper (OFHC)
Aluminum
Insulated cable (polyimide/glass)
Insulated cable (all polyimide)
a
Transverse direction; SSC inner cable.
2.0
2.9
3.1
4.2
5.1 a
5.6 a
ever, and as will be described in the next section, it is also

desirable that the collars be as rigid as possible or have an
integrated thermal expansion coefficient approaching that of
the low carbon steel used for the yoke. This favors austenitic
stainless steel, which has a lower integrated thermal expansion coefficient and whose Youngs modulus is 195 GPa at
room temperature and 203 GPa at 4.2 K, compared to 72 GPa
at room temperature and 80 GPa at 4.2 K for aluminum
alloy.
When assessing the respective merits of austenitic stainless steel and aluminum alloy, note that austenitic stainless
steel presents a better resistance to stress cycling at low temperature (63), but that it has a higher density (7800 kg/m3
compared to 2800 kg/m3 for aluminum alloy) and is more expensive.
There is no ideal choice between stainless steel and aluminum alloy, and magnets with both types of collar materials
have been built: HERA dipole magnets and most LHC dipole
magnet prototypes use aluminum alloy collars, whereas Tevatron dipole magnets and most SSC dipole magnet prototypes
rely on stainless steel collars. In any case, and whichever collar material is chosen, a thorough mechanical analysis of the
structure under the various loading conditions is required.
Radial Support
Limiting Radial Deflections. As described previously, the radial component of the Lorentz force tends to bend the coil
outwardly, with a maximum displacement at the coil assembly midplane. At high fields, this bending results in shear
stresses between coil turns and in an ovalization of the coil
assembly (along the horizontal x axis for a dipole magnet),
which generates field distortions. To prevent displacements
or deformations, the radial deflections of the coil assembly
must be limited to, typically, less than 0.05 mm.
Seeking Yoke Support. The main support against the radial
component of the Lorentz force is provided by the collars,
whose stiffness and radial width must be optimized to limit
collared-coil assembly deflections. However, in the magnetic
design of high field magnets, the field enhancement provided
by the iron yoke is maximized by bringing it as close as possible to the coil. This reduces the space left for the collars,
whose rigidity then becomes insufficient to hold the Lorentz
force, and the yoke and helium containment shell must also
be used as part of the coil support system.
The mechanical design of magnets where the yoke is
needed to support the collared-coil assembly is complicated by
the fact that the collar material (stainless steel or aluminum)
shrinks more during cool-down than the low carbon steel used
for the yoke (see Table 2). This thermal shrinkage differential
must be compensated for to ensure that, when the magnet is
cold and energized, there is a proper contact between the collared-coil assembly and the yoke along the axis of maximum
potential displacements. Such contact limits the deformations
of the collared-coil assembly and allows a partial transfer (up
to 50% in some LHC dipole magnet prototypes) of the radial
component of the Lorentz force to the yoke and the shell.
The aforementioned thermal shrinkage differential r can
be estimated as
r = Rcr (cr yk )
(21)
755
where Rcr is the collar outer radius and yk is the thermal

expansion coefficient of the yoke, integrated between room
and operating temperatures.
To limit contact loss due to thermal shrinkage differential,
it is preferable to use for the collars a material whose integrated thermal expansion coefficient approaches that of low
carbon steel. This suggests the use of austenic stainless steel
(see Table 2). However, and as was described in the section
on choice of collar material, it is also desirable to limit the
cool-down loss of coil precompression, which favors the use of
aluminum alloy.
Mechanical Design with Fully Mated Yoke Assembly. To facilitate assembly, the yoke of dipole magnets is usually split into
two halves, which are mounted around the collared-coil assembly. The shell, which is also made up of two halves, is
then placed around the yoke and welded. If the thermal
shrinkage differential between collar and yoke is not too large
(as in the case of stainless steel collars), it can be compensated for by designing and assembling the structure so that
the two yoke halves apply a compressive load over selected
areas of the collared-coil assembly. This compressive load is
obtained by introducing a shrinkage allowance into the geometry of either the collars or the yoke and by welding the shell
so as to press radially onto the two yoke halves and as to force
them to mate at room temperature. During cool-down, the collared-coil assembly shrinks away from the two yoke halves,
which remain fully mated. This results in a progressive decrease of the compressive load exerted by the yoke, but a suitable contact can be maintained over the designated areas of
the collared-coil assembly.
In practice, the compressive load provided by the yoke is
directed along a given axis. The choice of the axis drives the
choice of yoke split orientation. The SSC dipole magnet prototypes built at BNL use a horizontally split yoke with a yoke
collar compressive load directed along the vertical y axis as
shown in Fig. 11(a), while the SSC dipole magnet prototypes
built at FNAL use a vertically split yoke with a yokecollar
compressive load directed along the horizontal x axis as
shown in Fig. 11(b) (64). Both types of magnets performed
very well.
Mechanical Design with Yoke Midplane Gap at Room Temperature. For large thermal shrinkage differentials (as in the case
of aluminum collars), the yokecollar compressive load required at room temperature for a full compensation would
overstress the collared-coil assembly, and a more sophisticated mechanical design must be used. The twin-aperture
LHC dipole magnet prototypes with aluminum collars rely on
a two-piece, vertically split yoke with an open gap at room
temperature and a welded outer shell made of a material
(stainless steel or aluminum) that shrinks more during cooldown than the low-carbon steel yoke (65).
In these magnets, the yoke is designed so that, when
placed around the collared-coil assembly at room temperature
with no pressure applied to it, there remains an opening between the two yoke halves of the order of the expected thermal shrinkage differential. The yoke midplane gap is then
closed in two stages: (1) during shell welding, as a result of
the compressive load arising from weld shrinkage, and (2)
during cool-down, as a result of the compressive load arising
from thermal shrinkage differential between yoke and shell.
756
Electric
bus
Stainless steel
outer shell
Cooling
passages
Laminated
Stainless iron yoke
steel
laminated
collars
Yoke
keyway
Tapered
key
Superconducting
coil
Beam tube
Electric
bus
Stainless steel
outer shell
Cooling
passages
Laminated
Stainless iron yoke
steel
laminated
collars
Yoke
keyway
Tapered
key
Superconducting
coil
(a)
Beam tube
(b)
Figure 11. SSC dipole magnet cross-sections (64): (a) BNL-style with horizontally split yoke and
(b) FNAL-style with vertically split yoke.
The initial gap closure during shell welding is limited to avoid

overstressing the collared-coil assembly. The closure is completed during cool-down thanks to the radial pressure exerted
by the shell, which forces the two yoke halves to follow the
shrinkage of the collared-coil assembly and to maintain contact along the horizontal x axis. The yoke midplane gap must
be fully closed at the end of cool-down to ensure that the
structure is very rigid and to avoid any risk of oscillation during energization.
A crucial issue in such a design is the ability to perform
the shell welding operation in a reproducible way during
mass production so as to achieve the desired yoke midplane
gap value at room temperature and to keep a tight tolerance
on this value (of the order of 0.1 mm). As we have seen, a gap
too close may result in coil overstressing at room temeprature, whereas a gap too open may result in contact loss during
cool-down.
In some LHC prototypes, the closure of the yoke midplane
gap is controlled by means of aluminum spacers located between the two yoke halves (66). The spacers are dimensioned
to have a spring rate similar to that of the collared-coil assembly, and they prevent the gap from closing at room temperature. During cool-down, however, they shrink more than the
yoke and cease to be effective.
RHIC Magnets. In RHIC magnets, collar and yoke designs
are altogether simplified by replacing the collars by reinforced
plastic spacers and by using directly the yoke to precompress
the one-layer coils (67). It remains to be seen if this structure
could be scaled up to higher-field magnets.
End Support
As described previously, the axial component of the Lorentz
force tends to stretch the coil outwardly along the z axis. In
magnets where the yoke is not needed to support the collaredcoil assembly, a clearance can be left between the two. If the
axial stresses resulting from the Lorentz force do not exceed
the yield stress of the coil, it is possible to let the collared-coil

assembly expand freely within the iron yoke. This is the case
of the quadrupole magnets designed at Commissariat a` lEnergie Atomique/Saclay for HERA, SSC and LHC (68). However, in magnets where there is contact between collar and
yoke, it is essential to prevent stick/slip motions of the laminated collars against the laminated yoke and to provide a stiff
support against the axial component of the Lorentz force
(60,69). The ends of SSC and LHC dipole magnet coils are
contained by thick stainless steel end plates welded to the
shell.
MAGNET COOLING
Superconductor Critical Temperature
The superconducting state exists only at temperatures below
the so-called critical temperature TC. For NbTi, TC can be estimated as a function of applied magnetic flux density B using
TC (B) = TC0 1
B
BC20
1.7
(22)
where TC0 is the critical temperature at zero field (about 9.2

K) and BC20 is the upper critical magnetic flux density at zero
temperature (about 14.5 T).
Boiling and Supercritical Helium Cooling. To achieve low
temperatures and ensure stable operations against thermal
disturbances, the accelerator magnet coils are immersed in
liquid helium (70). Helium is a cryogenic fluid whose pressure-temperature phase diagram is presented in Fig. 12. Its
boiling temperature is 4.22 K at 1 atm (1 atm 0.1 MPa).
Small superconducting magnet systems usually rely on
boiling helium at 1 atm (71). Boiling helium offers the advantage that, as long as the two phases are present, the temperature is well determined. However, in large-scale applications,
10,000
Solid
Pressure (kPa)
1000
Liquid I
Supercritical
Liquid II
Critical point
100
Saturated He I
Pressurized He II
Vapor
10
Saturated He II
1
4
5
Temperature (K)
Figure 12. Pressure-temperature phase diagram of helium (71).
such as superconducting particle accelerators, the fluid is

forced to flow through numerous magnet cryostats and long
cryogenic lines, where heat leaks are unavoidable. The heat
leaks result in increases in vapor contents and create a risk
of gas pocket formation that may block circulation.
The aforementioned difficulty can be circumvented by taking advantage of the fact that helium exhibits a critical point
at a temperature of 5.2 K and a pressure of 0.226 MPa (see
Fig. 12). For temperatures and pressures beyond the critical
point, the liquid and vapor phases become indistinguishable.
The single-phase fluid, which is called supercritical, can be
handled in a large system without risk of forming gas pockets.
However, its temperature, unlike that of boiling helium, is
not constant and may fluctuate as the fluid circulates and is
subjected to heat losses.
The cryogenic systems of the Tevatron, HERA, and RHIC,
and that designed for the SSC, combine single-phase and twophase helium (71). In the case of the Tevatron and HERA, the
insides of the magnet cold masses are cooled by a forced flow
of supercritical helium, while two-phase helium is circulated
in a pipe running at the cold mass periphery (around the collared-coil assembly for Tevatron magnets, in a bypass hole in
the iron yoke for HERA magnets). In the case of the SSC, it
was planned to only circulate supercritical helium through
the magnet cold masses, while recoolers, consisting of heat
exchangers using two-phase helium as primary fluid, would
have been implemented at regular intervals along the cryogenic lines. The cryogenic system used for the RHIC is inspired by that of the SSC. In all these schemes, the boiling
liquid is used to limit temperature rises in the single-phase
fluid.
Superfluid Helium Cooling
A peculiarity of helium is the occurrence of superfluidity (70).
When boiling helium is cooled down at 1 atm, it stays liquid
until a temperature of the order of 2.17 K, where a phase
transition appears. For temperatures below 2.17 K (at 1 atm)
helium loses its viscosity and becomes a superconductor of
heat. This property, unique to helium, is called superfluidity.
Superfluidity is very similar to superconductivity, except that,
instead of electrical conductibility, it is the thermal conductibility that becomes infinite. The transition temperature be-
757
tween the liquid and superfluid phases depends on pressure.

It is called the lambda temperature T.
The LHC magnets are cooled by superfluid helium, and
their operating temperature is set at 1.9 K (72). Decreasing
the temperature improves the current-carrying capability of
NbTi dramatically and allows higher fields to be reached. (For
NbTi, the curve of critical current density as a function of
field is shifted by a about 3 T when lowering the temperature from 4.2 K to 1.9 K.) The feasibility of a large-scale cryogenic installation relying on superfluid helium has been demonstrated by Tore Supra, a superconducting tokamak built at
Commissariat a` lEnergie Atomique/Cadarache near Aix en
Provence in the South of France and operating reliably since
1988 (73).
Magnet Cryostat
To maintain the magnet cold masses at low temperature, it is
necessary to limit heat losses. There are three mechanisms of
heat transfer (74): (1) convection, (2) radiation, and (3) conduction. The convection losses are eliminated by mounting
the cold masses into cryostats, which are evacuated (71,75).
The radiation losses, which scale in proportion with the effective emissivities of the surfaces facing each other and with
the fourth power of their temperatures, are reduced by surrounding the cold masses with blankets of multilayer insulation and thermal shields at intermediate temperatures. The
main sources of conduction losses are the support posts, the
power leads, and the cryogenic feedthroughs, which are designed to offer large thermal resistances.
QUENCH PERFORMANCE
As explained in the operating margin section, the maximum
quench current Iqm of a magnet at a given operating temperature can be estimated from the critical current of the cable
and the peak field on the coil. It corresponds to the ultimate
current-carrying capability of the cable and can be raised only
by lowering the operating temperature.
When energizing a superconducting magnet, the first
quenches usually occur at currents below Iqm (chapter 5 of Ref.
76). In most cases, however, it appears that, upon sucessive
energizations, the quench currents gradually increase. This
gradual improvement is called the magnets training. The
training often leads to a stable plateau corresponding to the
maximum quench current.
Quenches below the expected maximum quench current
have at least four origins: (1) energy deposition in the magnet
coil resulting from frictional motions under the Lorentz force,
(2) energy deposition from synchrotron radiation and beam
losses, (3) heat dissipation from coupling currents in the cable, and (4) current imbalances among cable strands.
Quenches of the first origin reveal flaws in the mechanical
design or in the assembly procedures that must be analyzed
and corrected. The effects of synchrotron radiation can be reduced by implementing an intercepting screen within the
beam tube. Coupling losses and current imbalances are only
of concern for fast current cycles.
When operating an accelerator made up of several hundred
or even several thousand superconducting magnets, it cannot
be tolerated that magnets quench at random. Hence, the magnets must be designed with a safe margin above the maxi-
758
mum operating current of the machine. In addition, systematic tests must be carried out before installing the magnets
in the tunnel to ensure that their quench performance is adequate and does not degrade upon extended current and thermal cycling (77).
QUENCH PROTECTION
Conductor Heating
Although most R&D programs have been successful in developing magnet designs that can be mass-produced and meet
accelerator requirements, quenches do occur in accelerator
operations. These quenches must be handled in order to avoid
any damage of the quenching magnet, to ensure the safety of
the installation, and to minimize down time.
The most damaging effect of a quench is that, once a volume of conductor has switched to the normal resistive state,
it dissipates power by the Joule effect (Chapter 9 of Ref. 76).
Most of this power is consumed locally in heating up the conductor. In a very short time (typically a few tenths of a second), the conductor temperature can reach room temperature,
and, if the magnet is not discharged, keep on increasing.
Maximum-Temperature Requirement. The temperature rise
subsequent to a quench must be limited for at least three reasons: (1) to restrict the thermal stresses induced in the
quenching coil, (2) to prevent degradation of superconductor
properties, and (3) to avoid insulation damage.
For most materials, thermal expansion starts to be significant for temperatures above 100 K. The critical current density of NbTi is affected by exposure to temperatures above
250C. The extent of degradation depends on the temperature
level and on the duration of the exposure: at 250C, it takes
of the order of 1 h for significant degradation to occur, while
it may take less than a minute at 400C to 450C (78). Finally,
the polyimide materials used to insulate NbTi cables lose
most of their mechanical properties for temperatures above
500C.
It follows that an upper limit for conductor heating subsequent to a quench is 400C. Most magnets are designed not
to exceed 300 K to 400 K, and whenever possible, the limit
should be set at 100 K.
Protecting a Quenching Magnet
The source of conductor heating in a quenching magnet is
power dissipation by the Joule effect. Power keeps being dissipated as long as there is current in the magnet coil. To eliminate the heat source and limit the temperature rise, it is thus
necessary to ramp the current down.
To discharge a quenching magnet, all its stored magnetic
energy must be converted into resistive power. If the zone
where the conductor has switched to the normal state remains confined to a small volume, there is a risk that a large
fraction of the stored energy will be dissipated in this small
volume. In the case of a string of magnets connected electrically in series, it may even happen that the energy of the
whole string will be dissipated in the quenching magnet.
Hence, to prevent burnout, it is necessary to ensure that the
normal resistive zone spreads rapidly throughout the quenching coil. This can be done by means of heaters, implemented
near the magnet coils and fired as soon as a quench is detected.

These heaters are referred to as quench protection heaters.
In comparison with other superconducting magnets, most
accelerator magnets require an active quench protection system because of the rapidity of the temperature rise resulting
from the high current density and the low fraction of stabilizing copper in the cable strands. One notable exception is the
RHIC dipole magnets, whose one-layer coil assemblies are
wound from a cable with a high copper-to-superconductor ratio (2.25 to 1), and which do not rely on quench protection
heaters.
Hot Spot Temperature
Estimating Hot Spot Temperature. The volume of conductor
that heats up the most significantly during a quench is the
spot where the quench first originated. It is called the hot
spot. An upper limit of the hot spot temperature, Tmax, can be
determined by assuming that, near the hot spot, all the power
dissipated by the Joule effect is used to heat up the conductor.
Integrating the heat balance equation yields

S
Tmax
T0
C(T )
=
dT
(T )
dt I(t)2
(23)
t0
where C is the overall specific heat per unit volume of conductor, is the overall conductor resistivity in the normal state,
S is the conductor cross-sectional area, I is the current, t0 is
the time of quench start, and T0 is the coil temperature at t0.
The left member of Eq. (23) depends only on conductor
properties whereas the right member depends only on the
characteristics of current decay. The right-hand side integral,
divided by 106, is called the MIIT integral (Mega I times I
versus Time integral) and its value is refered to as the number of MIITs. The maximum temperatures computed from the
numbers of MIITs have been shown to be in fairly good
agreement with actual measurements of hot spot temperatures on quenching magnets (79).
Limiting Hot Spot Temperature. The hot spot temperature
can be limited by acting on either member of Eq. (23). Regarding the left member, the only conceivable action is to reduce
the overall conductor resistivity by increasing the copper-tosuperconductor ratio. However, and as explained in the conductor section, the copper-to-superconductor ratio must also
be optimized to ensure a high overall critical current. Regarding the right member, the MIIT integral can be minimized by
(1) detecting the quench as soon as possible, (2) turning off
the power supply (case of a single magnet) or forcing the current to bypass the quenching magnet (case of a magnet
string), (3) firing the quench protection heaters, and (4) discharging the quenching magnet or the magnet string.
Quench Detection
The magnets are connected to quench detection systems that
monitor the occurrence of a resistive voltage in the coil windings or the coils leads. The resistive voltage must be discriminated from inductive voltages arising from magnet ramping.
The inductive components are cancelled out by considering
voltage differences across two identical coil assemblies or two
identical parts of a given coil assembly (e.g., the upper and
lower half coils in a dipole magnet). When the resistive volt-
age exceeds a preset threshold over a time exceeding a preset

duration, the detection system generates a trigger that signals the occurrence of a quench.
Protection of a Single Magnet
Current Decay. Let us first consider the case of a single
magnet, and let us assume that, once a quench is detected,
the power supply is turned off and the magnet is switched to
an external dump resistor, Rext. The current decay is determined by
Lm
dI
+ [Rq (t) + Rext ] I = 0
dt
(24)
where Lm is the magnet inductance and Rq(t) is the developing

resistance in the quenching coils. Furthermore, the total voltage across the magnet Vm is given by
Vm = Re I(t)
(25)
where I is the current intensity.

To limit the number of MIITs, it is desirable to have a fast
current decay. Equation (24) shows that fast decay rates are
obtained either by means of a large Re or by ensuring that
Rq(t) increases rapidly. For some magnets, an external resistor can be used to extract a significant fraction of the stored
magnetic energy. However, it is also required to keep Vm to a
reasonable level (typically less than 1 kV) to avoid insulation
breakdown. Given the order of magnitude of I (up to 15 kA),
this imposes a small Re (typically a few hundredths of an
ohm), which, during a quench, is soon overcome by Rq(t).
Hence, for accelerator magnets, the current decay is largely
dominated by the resistance development in the quenching
759
coils, and the decay rate can be increased only by speeding

up Rq(t).
Maximum Voltage to Ground. The developing resistance in
the quenching coil separates the coil impedance into several
parts (Ref. 2, p. 137): unquenched parts across which the voltage is mainly inductive and quenched parts across which the
voltage is mainly resistive. The resistive and inductive voltages compensate each other partially so that their sum
equals Vm. The voltage distribution with respect to ground depends on the respective sizes and locations of these various
parts. The more uniform the quench development, the lower
the maximum voltage to ground. As an illustration, Fig. 13
shows the voltage distribution in a quenching magnet. Here,
Vm is assumed to be nil, and Rq is assumed to be concentrated
near two-thirds of the magnet length.
Quench Protection Heaters. As described earlier, to speed
up and uniformize quench development, most accelerator
magnets rely on quench protection heaters, which are fired as
soon as a quench is detected. The heaters are usually made
of stainless steel strips, which are copper clad at regular intervals along their lengths and which are placed on the outer
surface of the coil assemblies. Note, however, that the heater
firing unit relies on a capacitor bank and that it takes some
time for the energy to be released. Note also that the heaters
must be electrically insulated from the coil and that this electrical insulation introduces a thermal barrier. As a result,
there is a nonneglegible delay between the firing of the heaters and their effect on the coils, during which we must rely
on natural quench propagation (80). The heaters and their
implementations in the magnet assembly are optimized to reduce this delay.
Voltage
Inductive
voltage
Resistive
voltage
Rq
I
Current source, Vm 0
Figure 13. Voltage distribution in a

quenching magnet. The total voltage
across the magnet is assumed to be nil
and the developing resistance is assumed
to be concentrated near two-thirds of the
magnet length (2).
760
I Iq
Rb
Iq
Lm
Rq(t)
Lm
Lm
Figure 14. Electrical circuit of a quenching magnet in a magnet string (2).
In an accelerator, the magnet ring is divided into several sectors made up of series-connected magnets. The sectors are
powered independently and are electrically independent.
Once a quench is detected in a magnet, the power supply of
the sector to which the magnet belongs is turned off and the
sector is discharged over a dump resistor.
Unlike in the case of a single magnet, the current decay
rate in the sector must be limited for at least two reasons: (1)
to prevent the induction of large coupling currents in the
magnet coils (which may quench the remaining magnets in
the sector, resulting in general warming and significant helium venting) and (2) to avoid the occurrence of unacceptable
voltages to ground (because of the large overall inductance of
the sector). A too slow decay rate, however, creates the risk
that a significant fraction of the total energy stored in the
sector be dissipated in the quenching magnet, resulting in destructive overheating.
These contradictory considerations can be reconciled by
forcing the current to bypass the quenching magnet and by
ramping the current down at the desired rate in the remaining unquenched magnets. The bypass elements consist
of diodes (or thyristors) connected in parallel to individual or
small groups of magnets, as shown in Fig. 14. As long as the
magnets are superconducting, the current flows through the
magnets. Once a magnet has quenched and starts to develop
a resistive voltage, the main current is bypassed through the
diode connected in parallel, and the quenching magnet is discharged over the diode circuit. The current decay is determined by an equation similar to Eq. (24), except that Re must
be replaced by the resistance associated with the bypass element Rb.
HERA, RHIC, and LHC rely on silicon diodes that are
mounted inside the helium cryostats and operate at cryogenic
temperatures. The main requirements for these cold diodes
are (81): (1) small forward voltage and low dynamic resistance
(to limit power dissipation in the diodes), (2) good radiation
hardness, and (3) large backward voltage. In the case of the
Tevatron, which has a short current ramp time resulting in
large inductive voltages across the bypass elements, the diodes are replaced by thyristors operating as fast switches
(82). The thyristors are located outside the magnet cryostats
and require additional power leads and cryogenic feedthroughs.
Lm
Lm
I
Protection of a Magnet String
Lm
Current source
The protection system of the magnet ring must be carefully

designed and thoroughly tested before starting up the machine. The system tests are usually carried out on a cell or a
half-cell representative of the magnet lattice, and all failure
modes are investigated (8385).
SUMMARY
As of today, two large superconducting accelerator rings, Tevatron and HERA, have been built and are reliably operating,
and work is under way on two other superconducting collidersRHIC and LHC. The construction of RHIC is near
completion, and the industrial contracts for the mass production of LHC magnets will be awarded in 1999.
Since the time of the Tevatron (late 1970s), a factor of
about two has been gained on the critical current density of
NbTi at 4.2 K and 5 T, and a dipole field of 10.5 T has been
reached on a short magnet model relying on NbTi cables at
1.8 K. In recent years, encouraging results have been obtained on a couple of short dipole magnet models relying on
Nb3Sn cables, which may open the range 10 T to 15 T.
BIBLIOGRAPHY
1. D. A. Edwards and M. J. Syphers, An Introduction to the Physics
of High Energy Particle Accelerators, New York: Wiley, 1993.
2. K.-H. Mess, P. Schmuser, and S. Wolf, Superconducting Accelerator Magnets, Singapore: World Scientific, 1996.
3. H. T. Edwards, The Tevatron energy doubler: A superconducting
accelerator, Annu. Rev. Nucl. Part. Sci., 35: 605660, 1985.
4. A. V. Tollestrup, Superconducting magnets, AIP Conf. Proc., 87:
1982, pp. 699804.
5. B. H. Wiik, The status of HERA, Conf. Rec. 1991 IEEE Part. Accel.
Conf., IEEE Catalogue 91CH3038-7, 1991, pp. 29052909.
6. R. Meinke, Superconducting magnet system for HERA, IEEE
Trans. Magn., 27: 17281734, 1991.
7. V. I. Balbekov and G. G. Gurov, IHEP accelerating and storage
complex: Status and possibility of B-factory, Nucl. Instrum. Meth.
Phys. Res., A 333: 189195, 1993.
8. A. V. Zlobin, UNK superconducting magnets development, Nucl.
Instrum. Meth. Phys. Res., A 333: 196203, 1993.
9. R. F. Schwitters, Future hadron collider: The SSC, AIP Conf.
Proc., 272: 1993, pp. 306320.

10. R. I. Schermer, Status of superconducting magnets for the superconducting super collider, IEEE Trans. Magn., 30: 15871594,
1994.
761
33. J. D. Adam et al., Rutherford cables with anisotropic transverse

resistance, IEEE Trans. Appl. Supercond., 7: 958961, 1997.
34. R. A. Haarman and K. D. Williamson, Jr., Electrical breakdown
and tracking characteristics of pulsed high voltages in cryogenic
helium and nitrogen, Adv. Cryog. Eng., 21: 102108, 1975.
11. H. Foelsche et al., The relativistic heavy ion collider, RHIC, in H.

Schopper (ed.), Advances of Accelerator Physics and Technologies,
vol. 12, Adv. Ser. Directions High Energy Phys., Singapore:
World Scientific, 1993, pp. 104131.
35. P. Vedrine et al., Mechanical tests on the prototype LHC lattice

quadrupole, IEEE Trans. Magn., 30: 24752478, 1994.
12. A. Greene et al., The magnet system of the relativistic heavy ion
collider (RHIC), IEEE Trans. Magn., 32: 20412046, 1996.
36. M. Anerella et al., Improved cable insulation for superconducting

magnets, Proc. 1993 Part. Accel. Conf., 1993, pp. 27902792.
13. L. R. Evans, The large hadron collider project, in T. Haruyama,

T. Mitsui, and K. Yamafuji (eds.), Proc. 16th Int. Cryogenic Eng.
Conf./Int. Cryogenic Mater. Conf., London: Elsevier, 1997, pp.
4552.
14. R. Perin, Status of LHC programme and magnet development,
15. M. N. Wilson, Superconducting magnets for accelerators: A review, IEEE Trans. Appl. Supercond., 7: 727732, 1997.
16. A. den Ouden et al., Application of Nb3Sn superconductors to
high-field accelerator magnets, IEEE Trans. Appl. Supercond., 7:
733738, 1997.
17. R. M. Scanlan et al., Preliminary test results of a 13 tesla niobium tin dipole, Inst. Phys. Conf. Ser., 158: 15031506, 1997.
18. E. Gregory, Conventional wire and cable technology, AIP Conf.
Proc., 249: part 2, 1992, pp. 11981229.
19. L. Bottura, private communication, 1997.
20. D. C. Larbalestier and P. J. Lee, New development in niobium
titanium superconductors, Proc. 1995 IEEE Part. Accel. Conf., vol.
2, 1995, pp. 12761281.
21. D. Leroy et al., Design features and performance of a 10 T twin
aperture model dipole for LHC, Proc. 15th Int. Conf. Magn. Technol. (MT-15), Beijing, China, 1997, submitted.
22. L. T. Summers et al., A model for the prediction of Nb3Sn critical
current as a function of field, temperature, strain and radiation
damage, IEEE Trans. Magn., 27: 20412044, 1991.
23. R. Aymar, Overview of the ITER project, in T. Haruyama, T. Mitsui, and K. Yamafuji (eds.), Proc. 16th Int. Cryogenic Eng. Conf./
Int. Cryogenic Mater. Conf., London: Elsevier, 1997, pp. 5359.
24. P. Bruzzone et al., Testing of industrial Nb3Sn strands for high
field fusion magnets, IEEE Trans. Magn., 30: 19861989, 1994.
25. D. C. Larbalestier, The road to conductors of high temperature
superconductors: 10 years do make a difference!, IEEE Trans.
26. J. M. Royet and R. M. Scanlan, Manufacture of keystoned flat
superconducting cables for use in SSC dipoles, IEEE Trans.
Magn., MAG-23: 480483, 1987.
27. M. N. Wilson and R. Wolf, Calculation of minimum quench energies in Rutherford-type cables, IEEE Trans. Appl. Supercond., 7:
950953, 1997.
28. A. K. Ghosh, W. B. Sampson, and M. N. Wilson, Minimum
quench energies of Rutherford cables and single wires, IEEE
29. T. Shintomi et al., Development of large keystone angle cable for
dipole magnet with ideal arch structure, Adv. Cryog. Eng.
(Mater.), 36 (A): 323328, 1990.
30. R. M. Scanlan and J. M. Royet, Recent improvements in superconducting cable for accelerator dipole magnets, Conf. Rec. 1991
IEEE Part. Accel. Conf., 1991, pp. 21552157.
37. G. H. Morgan, Shaping of magnetic fields in beam transport magnets, AIP Conf. Proc., 249: part 2, 1992, pp. 12421261.
38. R. C. Gupta, S. A. Kahn, and G. H. Morgan, SSC 50 mm dipole
cross section, in J. Nonte (ed.), Supercollider 3, New York: Plenum, 1991, pp. 587599.
39. A. Devred and T. Ogitsu, Ramp-rate sensitivity of SSC dipole
magnet prototypes, in S. I. Kurokawa, M. Month, and S. Turner
(eds.), Frontiers of Accelerator Technology, Singapore: World Scientific, 1996, pp. 184308.
40. H. H. J. ten Kate et al., The reduction of the critical current in
Nb3Sn cables under transverse forces, IEEE Trans. Magn., 28:
715718, 1992.
41. R. C. Gupta, A common coil design for high field 2-in-1 accelerator magnets, Conf. Rec. 1997 IEEE Part. Accel. Conf., in press.
42. J. M. Cook, Strain energy minimization in SSC magnet winding,
IEEE Trans. Magn., 27: 19761980, 1991.
43. J. S. Brandt et al., Coil end design for the SSC collider dipole
magnet, Conf. Rec. 1991 IEEE Part. Accel. Conf., 1991, pp. 2182
2184.
44. T. Garavaglia et al., Application of the SSCRTK numerical simulation program to the evaluation of the SSC magnet aperture, in
M. MacAshan (ed.), Supercollider 2, New York: Plenum, 1990,
pp. 5976.
45. T. Ogitsu and A. Devred, Influence of azimuthal coil size variations on magnetic field harmonics of superconducting particle accelerator magnets, Rev. Sci. Instrum., 65 (6): 19982005, 1994.
46. C. P. Bean, Magnetization of high-field superconductors, Rev.
Mod. Phys., 36 (1): 3139, 1964.
47. H. Bruck et al., Field distortions from persistent magnetization
currents in the superconducting HERA magnets, Z. Phys. C
Part. Fields, 44: 385392, 1989.
48. R. W. Hanft et al., Studies of time dependent field distortions
from magnetization currents in Tevatron superconducting dipole
magnets, IEEE Trans. Magn., 25: 16471651, 1989.
49. A. Devred et al., Time decay measurements of the sextupole component of the magnetic field in a 4-cm aperture, 17-m-long SSC
dipole magnet prototype, Conf. Rec. 1991 IEEE Part. Accel. Conf.,
1991, pp. 24802482.
50. F. Willeke and F. Zimmermann, The impact of persistent current
field errors on the stability of the proton beam in the HERA proton ring, Conf. Rec. 1991 IEEE Part. Accel. Conf., 1991, pp. 2483
2487.
51. L. Bottura, L. Walckiers, and R. Wolf, Field errors decay and
snap-back in LHC model dipoles, IEEE Trans. Appl. Supercond.,
7: 602605, 1997.
52. P. W. Anderson, Flux creep in hard superconductors, Phys. Rev.
Lett., 9 (7): 309311, 1962.
31. D. Richter et al., DC measurements of electrical contacts between

strands in superconducting cables for the LHC magnets, IEEE
53. H. Bruck et al., Observation of a periodic pattern in the persistent-current fields of the superconducting HERA magnets, Conf.
Rec. 1991 IEEE Part. Accel. Conf., 1991, pp. 21492151.
32. J. M. Depond et al., Examination of contacts between strands by

electrical measurement and topographical analysis, IEEE Trans.
Appl. Supercond., 7: 793796, 1997.
54. A. K. Ghosh et al., Axial variations in the magnetic field of superconducting dipoles and quadrupoles, Proc. 1993 Part. Accel. Conf.,
1993, pp. 27422743.
762
55. A. Akhmetov et al., Current loop decay in Rutherford-type cables,

in P. Hale (ed.), Supercollider 5, New York: Plenum, 1994, pp.
443446.
56. L. Krempanski and C. Schmidt, Influence of a longitudinal variation of dB/dt on the magnetic field distribution of superconducting accelerator magnets, Appl. Phys. Lett., 66 (12): 15451547,
1995.
57. A. P. Verweij, Boundary-induced coupling currents in a 1.3 m
Rutherford-type cable due to a locally applied field change, IEEE
58. L. Bottura, L. Walckiers, and Z. Ang, Experimental evidence of
boundary induced coupling currents in LHC prototypes, IEEE
59. R. Stiening, private communication, 1991.
60. A. Devred et al., About the mechanics of SSC dipole magnet prototypes, AIP Conf. Proc., 249 (2): 1992, pp. 13091374.
61. T. Ogitsu, K. Tsuchiya, and A. Devred, Investigation of wire motion in superconducting magnets, IEEE Trans. Magn., 27: 2132
2135, 1991.
62. K. Koepke et al., Fermilab doubler magnet design and fabrication
techniques, IEEE Trans. Magn., MAG-15: 658661, 1979.
63. D. Perini et al., Measurements of the resistance to stress cycling
at 4.2 K of LHC dipole collars, IEEE Trans. Magn., 30: 1754
1757, 1994.
64. A. Devred et al., Review of SSC dipole magnet mechanics and
quench performance, in J. Nonte (ed.), Supercollider 4, New York:
Plenum, 1992, pp. 113136.
65. D. Leroy et al., Structural analysis of the LHC 10 T twin-aperture dipole, in T. Sekiguchi and S. Shimamoto (eds.), Proc. 11th
Int. Conf. Magn. Technol., London: Elsevier Applied Science,
1990, pp. 159164.
66. E. Acerbi et al., Development and fabrication of the first 10 m
long superconducting dipole prototype for the LHC, IEEE Trans.
Magn., 30: 17931796, 1994.
67. P. Wanderer et al., Construction and testing of arc dipoles and
quadrupoles for the relativistic heavy ion collider (RHIC) at BNL,
Proc. 1995 Part. Accel. Conf., 1995, pp. 12931297.
68. J. M. Baze et al., Design and fabrication of the prototype superconducting quadrupole for the CERN LHC project, IEEE Trans.
Magn., 28: 335337, 1992.
69. J. Strait et al., Tests of full scale SSC R&D dipole magnets, IEEE
Trans. Magn., 25: 14551458, 1989.
70. S. W. Van Sciver, Helium Cryogenics, New York: Plenum, 1986.

71. P. Lebrun, Cryogenic systems for accelerators, in S. I. Kurokawa,
M. Month, and S. Turner (eds.), Frontiers of Accelerator Technology, Singapore: World Scientific, 1996, pp. 681700.
72. P. Lebrun, Superfluid helium cryogenics for the large hadron collider project at CERN, Cryogenics, ICEC Suppl., 34: 18, 1994.
73. G. Claudet and R. Aymar, Tore supra and He II cooling of superconducting magnets, Adv. Cryog. Eng., 35 (A): 5567, 1990.
74. W. Frost (ed.), Heat Transfer at Low Temperatures, New York:
Plenum, 1975.
75. T. H. Nicol, Cryostat design for the superconducting super collider, AIP Conf. Proc., 249: part 2, 1992, pp. 12301241.
76. M. N. Wilson, Superconducting Magnets, Oxford: Clarendon,
1983.
77. P. Wanderer et al., Partial lifetime test of an SSC collider dipole,
IEEE Trans. Magn., 30: 17381741, 1994.
78. T. Schneider and P. Turowski, Critical current degradation of a
NbTi-multifilament conductor due to heat treatment, IEEE
Trans. Magn., 30: 23912394, 1994.
79. A. Devred et al., Investigation of heater-induced quenches in a
full-length SSC R&D dipole, in T. Sekiguchi and S. Shimamoto
(eds.), Proc. 11th Int. Conf. Magn. Technol., London: Elsevier,
1990, pp. 9195.
80. A. Devred et al., Quench characteristics of full-length SSC R&D
dipole magnets, Adv. Cryog. Eng., 35 (A): 599608, 1990.
81. L. Coull et al., LHC magnet quench protection system, IEEE
Trans. Magn., 30: 17421745, 1994.
82. R. Stiening et al., A superconducting synchrotron power supply
and quench protection scheme, IEEE Trans. Magn., MAG-15:
670672, 1979.
83. K. Koepke et al., Two cell doubler system test, IEEE Trans.
Magn., MAG-17: 713715, 1981.
84. W. Burgett et al., Full-power test of a string of magnets comprising a half-cell of the superconducting super collider, Part. Accel.,
43: 4175, 1993.
85. F. Rodriguez-Mateos et al., Electrical performance of a string of
magnets representing a half-cell of the LHC machine, IEEE
Trans. Magn., 32: 21052108, 1996.
ARNAUD DEVRED
Atomic Energy Commission at
Saclay
SUPERCONDUCTING CRITICAL CURRENT
In the application of superconductors, the superconducting

critical current density is often the most important parameter
in the design and engineering of practical devices. The reason
is that the majority of applications for superconducting wires
involve building electromagnets, which develop their magnetic field by virtue of the number of ampere-turns in the
magnet winding. Examples of electromagnets presently in the
marketplace include magnetic resonance imaging (MRI) magnets; high field research magnets; beam-bending, focusing,
and detector magnets for high-energy physics research; superconducting energy storage systems; and superconducting
motors and generators (1). As the critical current density of
the superconducting wire increases, the achievable magnetic
field increases, and the amount of superconductor needed for
constructing the magnet decreases. Thus, there is a direct
driving force to increase the critical current density of superconductors for magnet applications.
As with the critical temperature (TC) and critical magnetic
field (HC2), the critical current density (JC) marks the limit of
the superconducting state. For direct current (dc) densities
less than the JC, the current is carried without resistive
losses, and thus no power input. For current densities greater
than JC, a voltage develops along the superconductor, and the
zero resistance condition breaks down.
The TC and HC2, which are both determined by the chemistry and physics of the superconducting system, are relatively
unaffected by the processing of the superconductor. The same
is not true for the JC, which can be radically changed within
a given superconducting material by varying the fabrication
process and therefore the materials microstructure. For example, within the NbTi alloy system, once the composition
of the alloy has been chosen, the TC and HC2 are essentially
determined. However, by varying the metallurgical treatments of the alloy as it is processed into wire, it is possible to
vary the JC by factors of 1000 or more (2,3). The potential for
controlling the critical current density through processing
provides materials science researchers with hope for improving the properties of technical superconductors.
In describing the theory of critical current density in superconductors, it will be useful to consider two length scales.
The first is the London penetration depth , which is the distance over which an externally applied magnetic field penetrates into a superconductor. This is essentially the distance
over which we expect to see large changes in the magnitude
of the magnetic fields inside the superconductor. The second
length scale is the coherence length , which is the distance
over which the superconducting order parameter (or alternatively, the density of superelectrons) varies.
In the GinzburgLandau theory of superconductivity, the
ratio of the penetration depth to the coherence length is called
the GinzburgLandau parameter , where / . The GinzburgLandau parameter distinguishes between the two broad
classes of superconductors; type I, for which 1/2, and
type II, for which 1/2. The high critical current density
superconductors are all type II materials, and is quite large
for many of these materials, on the order of 50 or so.
The remainder of this article describes the measurement

of JC, the basic theory of critical currents in type I and type
II superconductors, the flux-line lattice and flux pinning, and
thermal effects in determining the JC. As an overview of a
large and complex subject that has occupied many researchers, this article cannot be complete. For a more detailed discussion of general superconductivity, see texts by Rose-Innes
(4), Tinkham (5), and especially Orlando (6) for an excellent
presentation from an electrical engineering perspective. Detailed surveys of flux pinning and critical current density may
be found in the monographs by Campbell and Evetts (7), and
Ullmaier (8), which form the basis of much of the discussion
following. Although the majority of the references and discussion uses examples from low-temperature superconductivity
(LTS), the principles described are also equally applicable to
high-temperature superconducting (HTS) materials.
THE RESISTIVE MEASUREMENT OF CRITICAL
CURRENT DENSITY
The critical current is usually found by a simple four-point
measurement, using the change in the resistance of the sample to determine the transition between superconductivity
and normal conductivity (see Fig. 1). The sample, a long coil
Sample Voltage
V
Multiturn
superconducting
sample
DC current
supply
Solenoid
Voltage, V
659
Current, I
IC
Figure 1. Schematic of the resistive measurement of the critical current density. The superconducting sample in the center of a high
magnetic field solenoid produces a voltage when the current exceeds
the critical current. The plot shows the V(I) characteristic of the superconducting wire. The critical current IC is defined at the appearance of a measurable voltage.
of wire, is placed in the bore of a high field electromagnet,

and a dc current is passed through it. The voltage along the
length of the superconducting wire is measured. A zero voltage indicates superconductivity, whereas a nonzero measurement indicates resistive dissipation and loss of superconductivity. This resistive measurement technique may be used
to measure critical temperatures by passing a small constant
current, varying the sample temperature, and measuring the
sample voltage. It can also be used to measure the critical
magnetic field HC2 of type II superconductors by varying the
magnetic field on the sample. Of more concern to the present
discussion is measuring the critical current IC. This is done
by holding the sample in a constant magnetic field and increasing the current through the superconductor until a voltage increase is measured (see Fig. 1). In this way the critical
current is determined.
As the current is increased through the superconducting
wire, the voltage along the wire slowly increases from zero
until a rapid increase occurs near the critical current. If the
V(I) characteristic is measured over a large enough voltage
range, the curve looks like that shown in Fig. 2. Early experiments (9) showed that the V(I) characteristic at high currents
becomes linear, and the resistivity depends on the applied
magnetic field, roughly following
ff =
H
HC2
r
No
al
a
st
te
re
s
si
ta
nc
(1)
Increasing H
Increasing = ff
Slope = ff
Lin
r
ea
flu
xf
l ow
reg
im
Current, I
Figure 2. Schematic of the voltage versus current measurement of
the critical current of a superconducting wire as a function of the
applied magnetic field. The zero voltage points have been offset for
clarity. As the field increases, the critical current decreases, and the
high current slope ff increases.
Thermal instability
to normal conductivity
Resistive transition
to flux flow state
Current, I
Figure 3. Schematic of the full V(I) characteristic of a superconducting wire. At low currents the superconductor exhibits zero resistance
and zero voltage. At the critical current there is a transition into the
flux-flow regime, and the voltage increases. At larger currents, ohmic
heating causes the temperature to rise above the critical temperature, and the sample thermally runs away to the normal state resistance behavior.
can be described by a power law

V (I) = K
where ff and n are the high current resistivity (flux flow resistivity) and normal state resistivity of the superconductor,
respectively. This linear region at high currents is called the
flux-flow regime for reasons that are described following.
Eventually, the increasing power dissipation due to the
flux flow resistivity causes the temperature of the superconductor to rise above TC and there is a phase transition to the
normal state (Fig. 3). An important point to note is that
within the flux flow regime the sample is still superconducting, though it is no longer in the zero dissipation (or zero resistance) condition.
Empirically it has been found that the shape of the V(I)
curve at low voltages (well below the linear flux-flow regime)
Voltage, V
Voltage, V
660
I n
IC
(2)
so that a plot of logV versus logI from the IC measurement

yields a straight line whose slope is sample-dependent and
also depends on the applied magnetic field (10). Recent measurements on high temperature superconducting (HTS) materials also show this power law behavior. Typically HTS materials are measured over a much larger range of current and
voltage than is usual in the LTS materials (11,12).
Over the years of development of LTS wires, it has been
empirically determined that larger slopes of the log V versus
logI plot were correlated with higher current density or
higher quality superconducting wires. This slope has come
to be known as the n-value of a wire and is commonly reported by LTS wire manufacturers. Values of n between 40
and 100 are generally considered to indicate good quality
wires. From Eq. (2) one can see that, as the n-value for a wire
increases, the transition from the zero resistance state to the
flux flow state becomes steeper and narrower.
Because of the gradual transition from zero voltage to the
linear flux-flow regime, the critical current is usually determined by using a standard measurement criterion. For many
years magnet designers preferred a constant resistivity criterion, for example, 1014 -m. A line is drawn on the V(I)
data plot with a resistive slope corresponding to 1014 -m,
based on the dimensions of the sample. The intersection of
this line with the V(I) measurement is the critical current IC.
Because the V(I) characteristic is curved, the measured value
of the IC depends on the criterion used, so this information
must be provided along with the measurement value. Historically, the 1014 -m criterion has been used by magnet designers because it was found that early superconducting magnets
thermally ran away to the normal state (quench) at a magnet
resistivity of about 1014 -m. A second commonly used criterion is the constant electric field criterion, in which the IC is
given by the current at which the V(I) measurement exceeds
a constant electric field value, for instance, 10 V/m.
661
It is important here to draw a distinction between the critical current IC, which has units of amperes, and the critical
current density JC, which has units of amperes per unit crosssectional area. The fundamental property of the superconducting state is the JC, which is the maximum current per
unit cross-sectional area of superconductor that is carried
without resistive losses. The JC is determined by measuring
the critical current IC of the specimen and dividing by the
cross-sectional area of the superconductor: JC IC /A.
An additional definition of importance is the engineering
JC which is defined as the maximum transport current per
unit cross section of superconducting wire. Most technological
superconductors are fabricated as a composite of superconducting and normal metal for thermal and mechanical stability (1,13). For the magnet designer, the engineering JC (sometimes abbreviated as JE) determines the available current in
the magnet windings. The distinction between JC and JE is
especially important when magnet designers are working
with superconducting composites in which the superconducting area is a small fraction of the total wire cross section, as
in tape composites and many early HTS wires.
It is also worthwhile at this point to describe the difference
between transport currents and shielding currents. Superconductors placed in a magnetic field exclude some or all of the
magnetic flux from the bulk of the superconductor (known as
the MeissnerOchsenfeld effect). For this exclusion to occur,
shielding currents flow on the surface of the superconductor
such that the magnetic field produced by the shielding currents opposes the applied field and cancels it out. In general,
the shielding currents flow in loops that are closed entirely
within the superconductor. In contrast, transport currents are
those currents applied from outside the superconductor using
external current sources. The transport currents are the currents used to produce magnetic fields in the superconducting
magnets and to make the resistive measurements of the critical parameters of superconductivity.
With this basic understanding of how the JC is typically
measured we can begin to discuss the physical mechanisms
limiting JC in practical materials.
superconducting states:
ULTIMATE LIMITS TO JC: THE DEPAIRING CRITICAL

CURRENT DENSITY
where and are the penetration depth and coherence

length, respectively, and o is the magnetic flux quantum. Using either Eqs. (7) or (8) we can calculate the depairing critical current density shown in Table 1 at zero kelvin and zero
magnetic field for several superconducting systems.
As can be seen from Table 1, the depairing critical current
densities are quite large, especially when compared with the
typical current density in standard copper household wiring
of about 107 A/m2. The superconducting materials shown here
have theoretical critical current densities at least 104 times
In the BardeenCooperSchrieffer theory of superconductivity, the charge carriers are pairs of electrons bound together
by a positive electronphonon interactive force. The bonding
energy of the superelectron pair at zero kelvin is denoted as
the energy gap (0). The critical temperature can be determined from the energy gap as the temperature at which the
thermal excitation energy kT is equal to the energy gap bonding the superelectron pair together. More rigorously this relationship is
2(0) = 3.5 kTC
(3)
The superconductivity stops because the thermal energy is

sufficient to depair or decouple the superelectrons.
Similarly, the thermodynamic critical magnetic field HC at
zero kelvin can be determined from the energy gap by using
the magnetic free energy difference between the normal and
1
1
0 Hc2 = N(F )[(0)]2
2
2
(4)
where N(F) is the density of states at the Fermi energy and

o is the permittivity of free space. Thus, the critical magnetic
field is that magnetic field for which the magnetic energy is
large enough to break apart the electron pair.
In a similar approach to the arguments used to estimate
TC and HC, one would expect that there is an ultimate critical
current density limited by the kinetic energy of the superelectron pairs in a transport current. When the kinetic energy of
the electrons exceeds the energy gap, the pairs break apart
and become normal (resistive) charge carriers. The kinetic energy of the superelectrons can be written as
KE =
p2
(m v2F )
= F 2(0)
2
2m
(5)
where m* is the effective mass of the charge carriers (in this

case m* 2me, two times the mass of an electron), and vF
and pF are the Fermi velocity and momentum, respectively.
The current density produced by these charge carriers is
given by
J = qns v = 2ens vF
(6)
where q is the charge of the current carrier ( 2e for electron

pairs) and nS is the density of superelectron charge carriers.
The depairing critical current density can be found by combining Eqs. (5) and (6):
JD
10 ens (0)
pF
(7)
Orlando (6) derives an equivalent form of Eq. (7) from GinzburgLandau theory as
JD = 0
3 30 2
(8)
Table 1. Theoretical Depairing Critical Current Densitya

Superconductor
NbTi
Nb3Sn
YBa2Cu3O (ab plane)
YBa2Cu3O (c plane)
a
, nm
, nm
300
65
30
200
4
3
3
0.4
Calculated using Eq. (8) for several important superconductors.
JD , A/m2
2
8
4
6
1011
1012
1013
1012
larger than copper. It is because of these large values of critical current density with no resistive losses (and therefore no
power dissipation) that superconductors are so important in
large electromagnet applications (1).
It should be remembered that the values of JD listed in
Table 1 are calculated for zero kelvin and zero magnetic field,
and in practice these conditions do not hold. In fact, these
values of current density have never been reached because of
other practical limitations. One of these limitations is the
self-field produced by a wire carrying a transport current. As
the transport current through the wire is increased, the selffield at the surface of the superconductor increases. At some
point, the magnetic field due to the transport current becomes
equal to the critical magnetic field of the wire, and the superconductivity breaks down. This model of the practical limit to
JC is know as Silsbees hypothesis (14), and is usually applied
to find the critical current limit of type I superconductors.
Of greater technological importance than the Silsbee limit
in type I materials is the limitation of the JC in type II superconductors because type II superconductors display superconductivity up to larger magnetic field values than type I superconductors. To understand the factors limiting the JC in type
II materials, it is necessary to review the magnetic properties
of these superconductors and introduce the concept of the
flux-line lattice.
where both H and J are vector quantities. For the one-dimensional case of a semi-infinite slab of type I superconductor in
the yz plane, and an applied field H, parallel to the slab in
the z-direction, Amperes law becomes
dHz /dx = Jy
Because the magnetic field decays over a length into the

superconductor, we know that the applied field and the
shielding current density are related approximately as
HA
= Jy
The principal difference between type I and type II superconducting materials lies in their response to an applied magnetic field. Type I superconductors exclude an applied magnetic field from the body of the superconductor up to the
thermodynamic critical field HC. To exclude this magnetic
flux, a shielding current is established on the surface of the
superconductor that flows in a direction so as to produce a
flux density equal and opposite to the applied field. This surface current flows in a surface layer whose thickness is equal
to the magnetic penetration depth . The magnitude of the
surface current can be found by using Amperes law which
states that the spatial variation in the magnetic field is proportional to the current density flowing:
(9)
Magnetization, M
Magnetization, M
H
Applied field, HA C
(a)
(11)
As the magnitude of the applied field increases, the magnitude of the current density increases to shield the superconductor from the field. The maximum current density is obtained when the applied field at the surface of the type I
superconductor is equal to HC, in which case JMAX HC / .
This shielding current density is the same as the depairing
current density described by Eqs. (7) and (8).
The shielding currents in the type I superconductor effectively provide a diamagnetic magnetization, M HA, as
shown in Fig. 4(a), called the MeissnerOchsenfeld effect.
For type II superconductors, the magnetic response is
somewhat different. Up to a lower critical field HC1, the magnetic response of type II superconductors is the same as that
of type I and shows a full flux expulsion, with M HA (see
Fig. 4b). In this region, the superconductor is said to be in the
Meissner state. For magnetic fields higher than HC1, the magnetic free energy balance of the superconductor makes it energetically favorable for the magnetic field to enter the bulk of
the superconductor. As the magnetic flux enters the bulk superconductor, it breaks into quantized units of flux 0, variously called the flux quantum, fluxon, fluxoid, flux vortex, or
flux line. The flux quantum has a magnitude of 0, 2.0679
1015 T-m2.
The individual flux quanta, or flux lines, orient themselves
parallel to the applied field and effectively reduce the magnetization of the type II superconductor below that of the perfect
diamagnetism of the Meissner state. This state of lower magnetization is called the mixed state. As the applied field increases, the number of flux lines per unit area increases in
the superconductor and M approaches zero. Eventually the
THE MAGNETIC FLUX LINE LATTICE
H = J
(10)
Meissner
state
HC1
Mixed state
Applied field, HA
(b)
Normal state
662
HC2
Figure 4. The magnetic behavior of type I and type II superconductors. Type I superconductors
exclude the applied field from the bulk of the superconductor by producing a supercurrent on the
surface to cancel the applied field, yielding the magnetization versus field plot shown in (a). Type
II superconductors exclude the applied field up to a lower critical field HC1 and then allow the
field to enter the bulk as flux quanta o, until the upper critical field HC2 is reached (b). At this
point the superconductivity is destroyed by the applied field.
flux lines touch one another, and the field inside the superconductor becomes equal to the applied magnetic field, driving the magnetization to zero at the upper critical magnetic
field HC2. The superconducting material remains superconducting up to large values of the applied magnetic field
(HC2), and this is one of the main reasons that the type II
materials are used for electromagnet applications.
The interaction of the individual flux lines with one another is similar to that of two parallel bar magnets. Because
the orientation of the field in the flux lines is the same, they
repel one another strongly. This causes the flux lines to distribute themselves in a periodic lattice to minimize the interflux line interactions (Fig. 5). This periodic structure,
called the flux line lattice (FLL), was theoretically predicted
by Abrikosov (15) using extensions of the GinzburgLandau
theory of superconductivity. Abrikosov found that the lowest
free energy configuration for the FLL is a triangular or hexagonal crystal. The FLL has been experimentally verified in
several ways, including magnetic particle decoration techniques (16) and diffraction from the flux line crystal by using
the magnetic moment of neutrons (17). The magnetic decoration technique, in particular, provides a striking visualization
of the periodicity of the flux line lattice, as shown in Fig. 6.
We can model an isolated flux line as a cylindrical core of
normal-phase material in which the superconductivity has
been destroyed by the magnetic field and which is surrounded
by a circulating supercurrent. The magnetic flux resides
within the core and decays into the bulk of the superconductor over a distance of the magnetic penetration depth (Fig.
7). Within this range, the local magnetic field strength H is
changing, and therefore, by using Amperes law [Eq. (9)],
there is a current flowing in the superconductor. This current
is analogous to the shielding current that flows on the superconductor surface to exclude the magnetic field. In this case
it is a circulating current that flows around the flux-line core
and has an orientation and magnitude needed to produce the
0 of magnetic flux in the core (Fig. 7). This circulating current is the origin of the name flux vortex.
The cylindrical core has a diameter twice the coherence
length (2). The coherence length is the distance over which
the superconducting order parameter 2 (or the density of
superconducting electron pairs nS) changes from its maximum
value at the core radius to zero in the center of the core (Fig.
663
Figure 6. The magnetic flux line lattice imaged using a magnetic

particle decoration technique. The small scale magnetic particles are
attracted to the regions of large flux gradient at the flux line cores
and then can be imaged by transmission electron microscopy. The
periodic triangular flux line lattice can be clearly distinguished. Reprinted with permission from H. Trauble and U. Essmann, J. Appl.
Phys., 39(9): 40524056, 1968. Copyright 1968, American Institute
of Physics.
7). The field strength in the core can be estimated as the magnetic flux divided by the cross sectional area of the flux line:
HCORE =
0
= HC2
0 2
(12)
As we introduce more flux lines into the interior of the superconductor, the circulating supercurrents of the neighboring
flux lines begin to interact and repel one another, leading to
the periodic structure of the flux-line lattice. The flux density
0
Flux line
J, circulating
current density
ns, density of
superelectrons
Fluxon
core
B, local field
Superconductor
Applied field, HA
Figure 5. Schematic of the magnetic flux line lattice in a type II

superconductor for applied fields between HC1 HA HC2. The flux
lines arrange themselves in a triangular or hexagonal lattice due to
the inter-flux-line magnetic repulsive forces.
Figure 7. Model of an isolated flux line as a core of normal material

containing the magnetic flux quantum 0. The magnetic field falls off
over a distance of , the penetration depth. The core has a radius of
, the superconducting coherence length, and the density of superelectrons falls to zero at the center of the flux line core.
664
at any point in the superconductor is found from the number

density of flux lines as B n o /A, where n is the number
of flux quanta in the cross-sectional area A. In a homogeneous
type II superconductor in the mixed state (i.e., in an applied
field between HC1 and HC2), the magnetic flux breaks up into
flux lines, each containing one quantum (o), of magnetic flux
that are periodically arranged in this two-dimensional crystal lattice.
If a transport current is applied to such a superconducting
wire, where the current flow is along the axis of the wire and
the applied magnetic field is perpendicular to the axis (as
shown in Fig. 8), there is an interactive force between the flux
lines and the transport current. Lorentzs law for the force on
a charged particle moving through a magnetic field is given
by
FL = J B
(13)
where FL is the Lorentz force density acting between the current of J and the flux density B. The FL has units of Newtons
per cubic meter and acts in a direction perpendicular to both
the flux density B and the transport current density J (Fig.
8).
The result of the Lorentz force acting on the FLL is to push
the flux lines across the superconductor. The movement of the
flux lines corresponds to a change in the flux density within
the superconducting circuit with time (Fig. 1), and from Maxwells equations for such a case (6),
B
dB
E
=
dt
(14)
In other words, the moving magnetic flux lines produce an

electric field gradient (or voltage) in the direction of the transport current flow. As the FLL moves, a voltage is generated
that must be supplied by the external power supply. The consequence of this flux motion is that the superconductor no
longer supports a transport current with zero dissipation, and
therefore the zero resistance state no longer exists.
It is important here to draw a distinction between the loss
of the superconducting state and the loss of the zero resistance condition. As long as the superconductor remains in
Applied field, HA
fields less than the upper critical field HC2 at temperatures

lower than TC and carries transport currents less than JD, it
is in the superconducting state. The nonzero resistance occurs
only because the FLL is moving under the Lorentz force produced by the transport current and changing the magnetic
flux linked by the superconductor. This is the flux-flow regime described earlier. If we could prevent the FLL from moving because of the Lorentz force, the zero resistance condition
would persist to higher transport currents, effectively increasing the critical current density.
The bulk pinning force density FP (N/m3), is defined for
samples carrying a transport current in a transverse field as
the critical Lorentz force:
F LC | = J C B
F P = |F
(15)
where JC is the current density at which voltage losses occur

in the superconductor.
In essence, holding the FLL against the Lorentz force adds
a transition line for the change from flux pinning to flux flow
to the phase diagram of the type II superconductor in Fig.
9. The solid lines represent thermodynamic phase transitions
between the superconducting (Meissner and mixed) states
and the normal state, and the dashed line shows the transition from the flux-pinning (zero resistance) condition to the
dissipative flux-flow condition.
Ideally, one would like to move the dashed line up in current density as close to the phase transition line (determined
by JD) as possible. This is the goal of the flux pinning discussed in the next section.
FLUX PINNING
To increase the current that a superconductor may carry
without power dissipation, it is necessary to restrain the FLL
against the Lorentz force by pinning it in place. There are
several mechanisms by which the FLL may be pinned, and
generally these rely on developing microstructural features
that interact with the individual flux lines. Examples of microstructural features that provide pinning resistance to the
Lorentz force include normal conducting precipitates, inclusions, voids, and grain boundaries.
The basic theory of flux pinning in type II superconductors
is conveniently broken into three sections. These are basic interactive forces, summation theory, and scaling laws (7,8).
Basic Interactive Forces
Transport
current, J
Lorentz force on flux lines, FL

Figure 8. The orientation of the magnetic field, transport current,
and Lorentz force acting on the flux line lattice. The Lorentz force
between the flux lines and the transport current causes the flux lines
to move across the superconductor.
The basic interactive forces are the forces between single, isolated flux lines and individual pinning centers. The usual
model for the basic interactive force is that the pinning center
must provide a spatial variation of the thermodynamic free
energy of the flux line. This can be visualized as either an
energy well (Fig. 10) or an energy hill. In the case shown in
the upper part of Fig. 10 the flux line has a lower free energy
when it sits in the energy well of the pinning center than it
does in the bulk superconductor, and thus there is a pinning
force holding the flux line in the well. The pinning force is
related to the free energy by the first derivative with respect
to position, so that the pinning force curve looks like that
shown in the lower part of Fig. 10. The deeper the potential
665
JD (0)
Current density
Applied magnetic field
Normal state
HC2(0)
Mixed state
Mixed state
JC(0)
HC1(0)
Meissner state
Temperature
Normal state
TC (0)
Meissner
state
Flux
f l ow
Flux
pinn
ing
HC1(0)
HC2(0)
Applied magnetic field
Figure 9. The HT and JH phase diagrams for type II superconductors. At low applied fields
the superconductor is in the Meissner state. At higher fields the superconductor enters the mixed
state with the creation of the flux line lattice (FLL). As the transport current is increased from
zero, the Lorentz force on the FLL eventually causes it to move, causing flux-flow dissipation
and a resistive voltage, shown as the dotted line. The superconducting state does not end until
the current density exceeds the depinning current density or the temperature rises above TC. As
flux pinning increases, the transition to flux flow occurs closer to the depairing critical current
density limit.
well, the steeper the energy profile, and the larger the pinning force.
If a Lorentz force is applied to a flux line trapped in this
potential well, the flux line moves in the direction of the Lorentz force until it is balanced by the oppositely directed pinning force. Thus the flux line is held in place, there is no flux
movement, and Eq. (14) shows that there is no dissipation.
The transport current is carried without power dissipation,
and the zero resistance condition is in effect. Superconducting
materials that pin magnetic flux are sometimes called hard
superconductors analogous to engineering alloys that have
been mechanically hardened by treatments to pin the movement of dislocations.
One type of basic interactive force between a single flux
line and a single pinning center is called the core interaction.
To nucleate a flux line within the superconductor, the system
Flux line
Flux line
energy
Bulk
superconductor
Pinning
center
Pinning
force
Figure 10. The variation in the free energy of the flux line in the
vicinity of a pinning center. The energy well produces a net force on
the flux line centering it in the pinning center and constraining it
against the Lorentz force of the transport current.
must provide enough energy to convert the core of the flux

line to the normal state. This energy (per unit length of flux
line), called the condensation energy, is given by the volumetric free energy due to the magnetic field within the flux line
and the cross-sectional area of the fluxon core as
ECOND =
H2
0
(16)
where HC is the thermodynamic critical field and is the superconducting coherence length.
Imagine that the superconductor contains a cylindrical
void of diameter 2 and its axis is oriented parallel to the
flux-line axis. If the flux line were centered on this void, the
condensation energy needed to produce the normal core of the
flux line would be saved, and the flux line would see a lower
free energy at the location of the void than it would in the
bulk, similar to Fig. 10. The result of this free energy change
is that the flux line requires an increase in its energy per unit
length equal to the condensation energy, Eq. (16), to move
away from the void. Thus the void acts as a pinning center
holding the flux line in place.
As the current density is increased, the Lorentz force on
the pinned flux line increases until it exceeds the maximum
gradient of the free energy versus position curve (Fig. 10). At
this point the flux line is free of the pinning center and moves
under the Lorentz force, creating a dissipative loss due to
Eq. (14).
There are many different interactions between the flux line
and microstructural defects that lead to basic interactive
forces and pinning. The core interaction may be applied to
voids and also to normal conducting precipitates (as in the
NbTi system) or weakly superconducting inclusions, for
which there is a spatial dependence of the superconducting
condensation energy. A different interaction that is thought
to be important in flux pinning in single-phase superconduc-
666
Grain boundary
l, electronic
mean free
path
reduced
bulk
Flux line core
bulk
Figure 11. The superconducting coherence length is reduced within

an electron mean free path of a scattering defect, such as a grain
boundary. This causes the flux line core to distort so that the volume
of the flux line changes as it approaches the grain boundary. The
variation of the flux-line energy with distance from the grain boundary causes a pinning interaction between the flux line and the grain
boundary.
tors, such as Nb3Sn, is the grain boundary interaction, first

proposed by Zerweck (18).
In the grain boundary interactive model, the grain boundaries are viewed as strong scattering centers for the normal
electrons in the metal, thereby reducing the mean free path
of the electrons near the grain boundary. When the electron
mean free path l is less than the coherence length, the coherence length depends on the mean free path as
dirty = 0.85(0l)1/2
(17)
This is often referred to as the dirty limit since the mean

free path of the electrons is much shorter than that of a
clean high-purity metal (5).
From Eq. (17), the coherence length is reduced within an
electron mean free path of the grain boundary. The effect of
the change in the coherence length is that as the flux line core
moves closer to the grain boundary, it becomes deformed (Fig.
11) and changes its volume so that the total energy (condensation energy times the core volume) changes with distance
from the grain boundary. The free energy difference with position leads to a pinning force, as with the core interactive
model.
A more general approach to basic interactive forces derives
from the GinzburgLandau theory of superconductivity,
which can be written to show that the variation in the free
energy of a flux line depends on spatial variations of the critical field and the GinzburgLandau parameter . One version
of this derivation (7) is to write the variation in the free energy of the flux line due to pinning defects as

E =

0 HC2
H
C2
HC2
||2 +

1 2
2

||4 dV
(18)
where is the unperturbed order parameter of the superconductor.

From this perspective, any spatial variation in either critical field (dHC2 /HC2) or (d / ) produces a change in the free
energy of the flux line that leads to a basic interactive force
for pinning. Examples of pinning defects of this sort include
normal and weakly superconducting inclusions and precipitates, dislocation clusters (subgrain boundaries), and chemical inhomogeneities, which produce pinning interactions
through changes in the electron mean free path and therefore

affect through the coherence length.
A class of basic interactive forces that can be modeled using image vortices to calculate the pinning forces are
grouped together as magnetic interactions. In these cases the
interaction between the circulating supercurrents and microstructural defects leads to pinning forces, rather than interactions involving the normal core. An example is the pinning
force between a flux vortex and an electrically insulating
plane. The interactive force is calculated by introducing an
identical image vortex on the opposite side of the insulating
plane. The overlapping supercurrents of the real flux line and
the image vortex repel one another and produce a force between the flux line and the insulating plane.
A recent addition to the theory of basic interactive forces
is that of the Josephson vortex (19,20). The Josephson vortex
model accounts for the large pinning forces found in optimized NbTi alloys for which the pinning centers are thin
sheet-like ribbons of normal conducting -Ti. These ribbons
are much thinner than the flux line core, so that the core
interaction does not accurately describe the pinning interaction. At the same time, the ribbons are not insulators, so that
the magnetic interaction also does not apply. The model estimates the basic interactive force by considering what happens
to the circulating supercurrent around the flux line, as it approaches a normal conducting planar pinning center. Given
that the supercurrent cannot readily penetrate the (normal)
pinning plane, the supercurrent spreads out along the planar
defect, slowly tunneling through the pinning plane as a superconducting Josephson tunneling current to complete the current loop on the other side of the pinning plane (Fig. 12). The
effect is that the flux line becomes distributed over a broad
area of the pinning plane and produces a pinning force due to
the Josephson current interactions.
In summary, basic interactive forces can arise from many
different physical mechanisms, of which only a few have been
described here. By providing a spatial variation in the free
energy of the individual flux lines, the pinning centers produce a pinning force that holds the flux lines stationary
against the Lorentz force, thus increasing the critical current
density of the superconductor.
Summation Theory
The second key part in the discussion of flux pinning is summation theory. Given a model for the basic interactive forces
between individual pinning centers and individual flux lines,
now we must consider the effect of large numbers of flux lines
Normal conducting
sheet pinning center
Flux core
Supercurrent
vortex
Pinned Josephson
vortex
Figure 12. In the Josephson vortex model the supercurrents surrounding a flux line are distorted as they approach a normal conducting sheet pinning center. The current spreads to the left and right of
the flux line to tunnel through the sheet and complete the circuit on
the opposite side. When the flux line rests on the sheet, it spreads
out along the length of the pinning center, losing the normal core and
distributing the flux quantum over a large area.
105
T= 4.2K
C44
Cij(Nm2)
interacting with large numbers of pinning centers. The principle complication of summation is that the flux lines interact
repulsively with one another and, in the absence of a pinning
force, order themselves in the flux line lattice. Thus, the flux
line lattice acts as a two-dimensional, elastic, crystalline
solid.
If the inter-flux-line forces are weak compared to the basic
interactive forces with the pinning centers, then the individual flux lines move out of the periodic FLL and arrange themselves so that as many flux lines as possible are located on
the pinning centers. If the number density of flux lines is less
than or equal to the number of pinning centers (for instance,
at small applied fields), then each flux line is individually
pinned, and the bulk pinning force is large. This is called direct summation, and the bulk pinning force density is just the
number density of pinning centers times the basic interactive force.
At the other extreme in which the interaction between the
flux lines in the flux line lattice is infinitely strong, the FLL
is completely rigid, and there can be no bulk pinning force
due to a collection of randomly distributed pinning centers
because, for any position of the FLL relative to the random
array of pinning centers, there will be as many basic interactive forces pulling the FLL to the left as to the right, and
the bulk pinning force density averages to zero. Even though
the basic interactive forces are very large, if the FLL acts as
a rigid solid because of interfluxon forces, there will be no
bulk pinning force, and the FLL will move under the Lorentz
force due to the transport current, yielding a low JC.
The correct description of pinning certainly lies somewhere
between these two extremes of direct summation and the
rigid FLL lattice. There are several models proposed to account for the summation of the basic interactive forces, and
all of them depend strongly on the elastic properties of the
FLL as determined by the inter-flux-line forces. In essence,
the FLL is a crystalline solid that is placed under an external
load by the competition of the transport-current-induced Lorentz forces and the restraint of the pinning forces. As the
Lorentz force loading increases, the FLL elastically distorts
until either the pinning forces are exceeded, at which point
the entire FLL breaks free and moves in unison through the
superconductor, or the load overcomes the inter-flux-line
forces, and the periodic FLL breaks apart.
This is analogous to mechanically loading a tensile specimen above its elastic limit and into the plastic deformation
region. As the mechanical test specimen is plastically distorted, crystalline defects in the specimen are created (dislocations), and the mechanical properties depend strongly on
the presence of these crystal defects. In the superconductor
with pinning centers, the increasing Lorentz force begins to
introduce crystal defects which fragment the FLL into a polycrystalline FLL. The crystalline nature of the FLL has been
experimentally observed, as has the polycrystalline and defective FLL, by using both magnetic particle decoration techniques and neutron scattering (21,22). It is also the case that
the presence of FLL crystal defects strongly affect its mechanical properties and response to Lorentz force loading (7).
There have been several calculations of the elastic behavior of the FLL. An example is shown in Fig. 13 (8,23) for a
NbTa alloy superconductor. C11 is the elastic modulus in the
plane normal to the flux line axes. This is a measure of the
stiffness of the FLL while pushing the flux lines closer together. C44 is the elastic tilt modulus which describes the
667
104
C11
103
C66
C66
102
0.2 0.4 0.6 0.8
0.0720.10.2 0.4
0.6 0.8
1
1
b
h
Figure 13. Labusch calculation of the elastic constants of the flux

line lattice in NbTa. Notice that the C11 and C44 elastic constants
increase with magnetic field, whereas the shear modulus C66 decreases with field at high fields. The b and h are the reduced fields
B/BC2 and H/HC2, respectively. Reprinted with permission from Ref. 8.
bending of flux lines along their axis. C66 is the shear modulus, which describes the resistance of the FLL to the shear of
flux lines past one another. An excellent discussion of the effect of the FLL elastic constants on the deformation of the
flux line lattice may be found in Ref. 8.
When the number density of flux lines is larger than the
number density of pinning centers (as is often the case), there
are two primary models of the summation behavior of the
FLL. These are the flux line lattice shear model and the collective-pinning model.
In the FLL shear model, individual flux lines are strongly
pinned on individual pinning centers, and the excess flux
lines not directly pinned are held in place against the Lorentz
force due to the interfluxon forces. At large enough transport
currents, the Lorentz force becomes larger than the elastic
shear modulus (C66) can support, and the FLL shears,
allowing unpinned flux lines to flow between those strongly
pinned by the pinning centers. Therefore the critical current
is determined, not by the strength of the pinning forces, but
by the shear stiffness of the FLL, given by C66.
In the Brandt model of the flux line elastic constants, the
compressive modulus C11 and the tilt modulus C44 both depend on the magnetic field roughly as
C11 C44
H2
4
(19)
whereas the shear modulus near HC2 is approximated by
C66 K 1
H
HC2
2
(20)
668
where K is a proportionality constant. At high magnetic

fields, the FLL is very stiff in bending and compression, but
becomes softer and softer in shear as the field approaches
HC2. Therefore from this summation model one would expect
that the high field critical current density would be determined by the FLL shear and would be somewhat independent
of the basic pinning interaction. A detailed model of the FLL
shear mechanism developed by Kramer (24,25) predicts a different high field behavior of the bulk pinning force density
compared to the predictions of the simple direct summation
model. Both kinds of behavior have been experimentally observed.
An alternative model for summation that applies in the
limit of a larger number density of flux lines than pinning
centers is the collective-pinning model (26,27). In collectivepinning theory, the FLL is thought to consist of a polycrystalline collection of grains in which the periodic order of the
flux line lattice is reasonably well maintained by inter-fluxline forces, but the neighboring grains are uncorrelated with
one another. Within a flux lattice grain, the flux lines are
strongly pinned to one or more pinning centers, and the correlated group is held in place by the combined action of the
basic pinning forces and the inter-flux-line forces.
In the collective-pinning model, the size of the correlated
FLL grains is determined by the number density of pinning
centers and the relative strength of the pinning force and the
inter-flux-line forces. In one limit, where the inter-flux-line
forces are large compared to the pinning forces, the flux lattice is elastically stiff and the correlated flux grain size is
large and involves many pinning centers. In this limit, the
bulk pinning force is small, as in the infinitely stiff FLL limit
described previously. As the pinning force increases relative
to the inter-flux-line forces, the correlated grains become
smaller and approach a limiting grain size equal to the mean
pinning center spacing, so that each flux lattice grain interacts with only one pinning center. In this limit, the bulk pinning force is at its maximum, because each pinning center is
applying a maximal constraint on the FLL. The bulk pinning
force is close to the direct summation pinning force in this
limit.
In summary, then, the central problem of summation theory is how one combines the effects of the individual fluxonpinning center interactions and the interfluxon forces to produce a bulk pinning force to hold the FLL against the Lorentz
force. Although the limiting cases of direct summation and
the rigid FLL are well understood, the behavior of real materials is less clear. Several models for summation have been
proposed, primarily the FLL shear and the collective-pinning
models, and validation of them with experimental measurements shows that they all have some merit, but none are capable of a complete description of the origins of the bulk pinning force. A final tool for understanding the physical
mechanisms behind the pinning force and JC is scaling laws.
Scaling Laws for Flux Pinning
Experimentally it has been found that as the testing temperature is varied, many superconductors exhibit scaling of the
bulk pinning force density versus applied magnetic field (28).
This is observed by first measuring the JC as a function of
magnetic field and developing a curve of the bulk pinning
force density versus magnetic field using Eq. (15): FP JCB.
This curve has a characteristic shape. It is zero at zero applied field, increases through a maximum with increasing
field, and drops to zero again as the Jc drops to zero at H
HC2 (Fig. 14).
As the test temperature changes, the HC2 of the sample
changes, and as a result, the bulk pinning force changes at a
given applied field. By scaling the ordinate using the reduced
field, h H/HC2, and scaling the abscissa as the reduced pinning force density, f P FP /FP MAX, where FP MAX is the maximum measured bulk pinning force density, the experimental
data frequently collapse onto a single line for all test temperatures (Fig. 14).
Although the majority of the experiments on scaling JC
have been performed in low temperature superconductors, a
large literature has also developed for scaling behavior in
high temperature superconductors. The terminology used in
HTS materials has evolved along different lines than those
used here, but the basic result is that temperature scaling of
the pinning force is also a common feature of these materials (29,30).
It is important that the pinning force follows a scaling law
with changes in magnetic field and temperature because scaling implies a single mechanism for flux pinning in the material as a function of temperature, which should be amenable
to theoretical prediction. Additionally, if scaling holds for a
given material, one only needs to measure the critical current
at one temperature and field to estimate the performance at
FP MAX
Decreasing
test temperature
FP/ FP MAX
Fp
Figure 14. To determine whether a superconducting material displays scaling of the flux pinning curve, the pinning force density versus applied field for several different test temperatures
is measured (a). The data are scaled using h
H/HC2 (b) and f P FP /FPMAX (c). If the sample
displays scaling, the different temperatures collapse onto a single plot (c).
1.0
Fp
H
(a)
h = H/HC2
(b)
1.0
0
0
h = H/HC2
(c)
1.0
other temperatures, which can be useful for magnet designers.

In general terms, scaling follows an equation, such as
2
h)
s (1
ic a
h
(a)
Fp
d ra t
which displays a peak pinning force density at h 0.25 and

a strong quadratic curvature as the applied field nears HC2.
The quadratic behavior is usually associated with Nb3Sn and
other single-phase superconductors (33).
Because many superconductors exhibit scaling, it is interesting to see if a theoretical model can predict the measurements and provide some insight into flux pinning behavior.
Because the theoretical picture of summation is somewhat diverse, it is not surprising that the theory of pinning force density scaling is not completely clear. However, there are qualitative models that explain some of the experimental behavior.
Campbell and Evetts (7) examined the low field region of
the pinning force curve, which is roughly linear with field for
nearly all of the scaling models and experiments. In this region, there are a small number of flux lines compared to the
number of pinning centers. Campbell and Evetts propose that
direct summation should apply because the spacing between
flux lines is large enough that the interactive forces between
them are weak. As the field is increased from zero, the bulk
pinning force density increases linearly because of the increased number of flux lines being pinned. This is often referred to as the partial synchronization range of fields because the flux lines become synchronized with the pinning
center array.
At some magnetic field, the number of flux lines is equal
to the number of pinning centers, and the maximum bulk pinning force density is reached. Therefore, the field of the pinning peak depends on the number of pinning centers, and the
magnitude of the peak depends on the strength of the basic
interactive forces (Fig. 15).
At higher fields, there are more flux lines than pinning
centers, and one expects a crossover from synchronization to
where the bulk pinning force is limited by other effects. In
the Campbell and Evetts model the high field pinning force
falls off because of the variation of the basic interactive force
with field, which falls as (1 h) for the core interaction. Thus,
Campbell and Evetts predict a linear low field region, a pin-
Fp
h)
s (1
ar a
(23)
Line
(22)
This is a symmetrical function of reduced field that has a

peak in the pinning force density at h 0.5. Often, optimized
Nb-Ti superconductors follow a linear scaling behavior (32).
The quadratic scaling field function is expressed by
f (h) = h1/2 (1 h)2
Increasing
basic interaction
force
High field
saturation
(21)
where K and m are empirically determined constants, f(h) is

a function only of reduced applied field, and the temperature
dependence is carried in the variation of HC2. Typically in LTS
materials, m varies between 1.5 and 2.5. The field function
f(h) may display many different kinds of scaling behavior
with magnetic field (31). The two most common are the linear
scaling and the quadratic scaling functions.
In linear scaling, the field function is given by
f (h) = h(1 h)
Increasing
number of
pinning centers
Qua
FP = K(HC2 )m f (h)
Partial
synchronization
669
h
(b)
Figure 15. In the summation model of Campbell and Evetts, the

pinning force curve is linear at both low fields and high fields, and
the position of the peak pinning force density shifts depending on the
number of pinning centers in the superconductor (a). The Kramer
model, on the other hand, shows a quadratic high field behavior, and
the pinning force density saturates at high fields, as the strength of
the basic interaction increases (b).
ning peak at a field determined by the number of pinning

centers, and a linear high field region. As the basic interactive
force increases, the pinning force curve increases in all fields.
An alternative model of summation due to Kramer (25) describes the low field portion of the pinning force curve by partial synchronization, as in the Campbell and Evetts model.
However, the high field behavior is determined by the shear
of the flux lattice past strongly pinned individual fluxons.
Therefore, the high field pinning force has a magnetic field
dependence determined by the C66 elastic constant, which
from Eq. (20) decreases as (1 h)2. Kramer also predicts that
the high field behavior will exhibit saturation such that variations in processing leading to changes in the basic interactive
forces will affect the magnitude and position of the peak but
will not affect the high field pinning force behavior (Fig. 15).
In most flux pinning theories the basic interactive force is
a function of temperature. For instance, in the core pinning
model, the basic interactive force depends on the condensation energy and therefore on HC. As the temperature changes,
so does HC. This leads to the observed temperature scaling of
FP.
However, the temperature dependence is more complicated
for some basic interactions. For example, if the pinning center
were a superconducting precipitate with TC and HC2 below
that of the bulk material, one would expect a difference in the
strength of the core pinning interaction as the temperature is
varied above and below the pinning center critical temperature and as the field moves above and below HC2 of the pinner.
This effect is normally observed as a lack of scaling and commonly as a shift of the peak in the pinning force curve as a
function of temperature.
A second example of a lack of scaling is a superconductor
in which the pinning force on the FLL is a combination of
several different basic interactive mechanisms. Such a superconductor might be a two-phase alloy in which pinning results
from both core interactions with normal precipitates and
grain boundary pinning. The different temperature and field
dependencies of the two operating pinning mechanisms lead
to a lack of scaling (34).
670
The main points to understand from this overview of the

flux pinning mechanism are the following:
Bulk flux pinning depends on the basic interactive forces
between individual flux lines and individual pinning centers.
Bulk flux pinning depends on the relative strength of the
basic interactive forces and the fluxon-fluxon forces,
which affects the summation of the individual interactions into the bulk pinning force acting on the FLL.
Scaling, or lack of scaling, provides a tool for understanding the pinning mechanisms operating in different field
and temperature regions. This understanding can help
direct modifications of the pinning microstructures by using suitable processing to optimize the pinning force and
JC of hard superconductors.
Superconducting
slab
z
HA
x
w
HA
Hz
0
The previous discussion has centered on the electrical behavior of the superconductor, but magnetic behavior is also an
important aspect of many applications. The magnetic response of the superconductor can be a valuable tool for measuring the critical current density.
An important consequence of pinning the magnetic FLL is
that the magnetic behavior of hard superconductors is
strongly hysteretic (Fig. 16). To understand the development
of the hysteretic magnetization curve, a simple but powerful
model was proposed by Bean (35) and has since been modified
and further refined (36).
To simplify discussion of the model, we use a sample geometry of an infinite superconducting plate of thickness W. The
applied magnetic field will be parallel to the surfaces of the
plate, as in Fig. 17. As the magnetic field is increased from
zero, superconducting magnetization currents develop on the
surfaces, so that they shield the interior of the superconductor from the applied field (the MeissnerOchsenfeld effect).
These shielding currents flow only within a distance (the
penetration depth) of the surface and fall off exponentially
into the superconductor. This condition persists until the external applied field exceeds the lower critical magnetic field
HC1. For fields larger than HC1 the superconductor is thermodynamically more stable if the applied field enters the super-
HA
Jc
Jy
THE CRITICAL STATE MODEL OF MAGNETIZATION
HA
Figure 17. In the Bean critical state model, the semi-infinite slab
develops the flux and current density profiles shown on the application of a magnetic field larger than HC1.
conductor as the flux line lattice. For simplification, the Meissner state below HC1 is ignored in the critical state model.
This is not a bad approximation, especially for the technological superconductors which have large values of the Ginzburg
Landau parameter and therefore small values of HC1.
If the sample has pinning centers to hold the entering FLL
in place, a magnetic field gradient is established at the surface of the superconductor which falls off into the body of the
sample. From earlier discussions of Amperes law, we know
that the current density flowing in a superconductor is directly related to the magnetic field gradient, Eqs. (9) and (10).
The situation looks schematically like that shown in Fig. 17,
where the z-directed applied field falls off into the sample
with a gradient in the x-direction that produces a current density flowing in the y-direction (into the page).
The current density flowing is equal to the critical current
density. If the flux gradient were steeper, the current density
would be larger than the JC, and the FLL would not be fully
pinned, leading to flux flow. The flux motion lowers the field
gradient until the FLL is pinned by the pinning centers, leaving a critical flux gradient and a current density equal to JC.
For this reason the model is known as the critical state
model.
In Beans original version of the critical state model, the
JC is assumed to be a constant, independent of applied field
from HC1 HA HC2. This assumption makes the flux profile
in the sample linear such that
dHz
Hz
=
= Jy = constant = JC
dx
x
Figure 16. Schematic of the hysteretic magnetization curve in strong

pinning superconductors. The arrows indicate the direction of travel
around the hysteresis loop during a typical magnetization measurement.
(24)
As the applied field is increased from HA 0, the field penetrates the sample from both sides, and generates a circulating
shielding current equal to JC (Fig. 17). The magnetization of
the slab can be found from examination of the field versus
position plot. From the definition of magnetization, we know
M(x) = HI (x) HA
(25)
where HI(x) is the local internal magnetic field. To find the

bulk magnetization, we must integrate the local magnetization over the sample volume. Because the sample is infinite
in the y- and z-directions, we can turn this into a one-dimensional integral over x, such that
1
1
Mbulk =
HI (x) dx
HA dx
W
W
1
[HI (x) HA ] dx
(26)
=
W
Comparing Eq. (26) with Fig. 18 shows graphically that the
second term is the area of the entire rectangular region,
whereas the first term is given by the area of the two darker
triangular regions. The bulk magnetization is the volume averaged difference between these, or the light gray trapezoidal
area of Fig. 18, divided by the sample width W.
Using this simple model we can determine the behavior of
the superconductor during a half magnetic field cycle used to
generate a magnetization loop of M versus HA. The process is
shown schematically in Fig. 19. For small applied fields
(points a, b) the field penetrates, and the magnetization increases rapidly with applied field. At point c the applied field
is large enough to push the magnetic flux line lattice all the
way to the center of the sample. This field is called the full
penetration field HP.
For applied fields larger than the full penetration field, the
magnetization does not change, even though the internal and
external fields increase. The magnetization curve remains flat
(points d, e).
Now, if the applied field were to be reduced, the flux lattice, which is being pinned in place by the pinning centers,
responds only near the surface region, as shown at point f.
The magnetization becomes rapidly smaller with decreasing
field. Now, the circulating supercurrents flowing in the sample have the spatial dependence shown in Fig. 20. Both the
positive and negative flowing currents are assumed to be
flowing at the critical current density. Recall that the magnitude of the critical current density in the Bean critical state
model is constant with magnetic field.
As the applied field is further reduced, the current density
profile eventually inverts, and the magnetization becomes
(h)
(g)
M, Magnetization
that the local magnetization response of the superconductor

to the applied field can be written as the difference between
the applied field and the local internal magnetic field:
671
HA, applied field

(a)
(f)
(b)
(c)
H
HA
HA
(d)
H
(e)
H
HA
HA
(a)
(b)
(c)
(d)
HA
H
HA
HA
H
HA
(e)
(f)
(g)
(h)
Figure 19. Schematic of a half hysteresis loop measurement of the

magnetization in a Type II superconductor. The flux profiles at eight
points around the magnetization loop are derived from the Bean critical state model and illustrate the source of the dependence of the
magnetization on the pinning of magnetic flux within the superconductor.
positive and constant (g). Finally, at HA 0, the magnetization is positive because of the magnetic fields trapped in the
body of the superconductor by the pinning forces acting on
the FLL.
Because the flux gradient is a constant, the full penetration field HP, varies with the width of the sample. Examina-
HA
HI(x)
Hz
0
HA
JC
Hz
Jy 0
0
Figure 18. The flux profile in the superconductor on applying a field

HA. This is useful for understanding the origin of the terms in the
integral of Eq. (26). The magnetization response to the applied field
HA is proportional to the area of the light gray trapezoid.
JC
Figure 20. The flux profile and the accompanying current density
profile for point (f) of Fig. 19. The current density profile matches the
profile of the magnetic flux at all points in the superconductor.
672
tion of Fig. 19 at the full penetration field (point c) shows that

the magnetic flux gradient
dH
HP
=
= JC
dx
(W/2)
(27)
We can also see that the magnitude of the magnetization at

this field is given by

J
2
dH
M(HP ) =
HP + x
HP dx = C x dx (28)
W
W
dx
2
integrated from x 0 to W/2. This gives the constant value
of the magnetization with applied field (points c, d, e) as
M = JC
W
2
(29)
Figure 19 also illustrates that, when one considers the entire

hysteresis loop of the magnetization measurement, the distance between the increasing field and decreasing field magnetization at any applied field is twice the result of Eq. (29),
or the more usual result from the critical state model,
M = JCW
(30)
The Bean critical state model has been modified to account

for the fact that the critical current density is not a constant
with applied field (36). These modifications lead to curved
magnetic field and current density profiles (Fig. 21) in addition to more realistic magnetization loops (Fig. 16). It is still
Hz
the case, however, that these models predict a direct relationship between the height of the magnetization loop at a given
field and the critical current density multiplied by the sample dimension.
Other modifications to the Bean Critical State Model have
incorporated the change in magnetization due to finite sample
sizes (37) and demagnetization factors for non-spherical samples (38).
The most important result of the Bean model is that the
magnetization behavior can be used as a probe of the critical
current density within the superconductor by using a technique different from the four-point resistive measurement.
For many emerging superconducting materials, magnetization measurements have allowed measuring the critical current density before long lengths of wire were available for resistive testing. It is also possible to measure JCs that would
be difficult to measure with a conventional resistive technique. An example is a cabled conductor with an IC of thousands of amperes. Multikiloamp power supplies are expensive
to purchase and operate, and the high current significantly
complicates the experimental design. Magnetization measurements of JC are not limited by the need for high current power
supplies. The magnetization measurement of JC continues to
be an important tool for the materials engineer in optimizing
the flux pinning process.
DISSIPATION EFFECTS IN HIGH CURRENT
DENSITY SUPERCONDUCTORS
The movement of the FLL within the superconductor has
many consequences for the applications of superconductors.
Examples include flux jumping (the rapid movement of magnetic flux within the superconductor which leads to localized
heating effects and the loss of the superconducting state), flux
flow near the JC, flux creep (the slow movement of the FLL
caused by random thermal jumping of flux lines out of the
pinning potentials), and magnetic hysteresis (which causes an
additional heating effect and resistive loss in ac applications
of superconductors). Because of their importance to applications, the dissipative effects have been carefully studied and
have led to some useful insights into the flux pinning process.
Flux Flow and Resistive Transition Analysis
Jy
0
Figure 21. The Bean critical state model assumes that the critical
current density is constant with the magnetic field. In modified versions of the critical state model, the critical current density is allowed
to vary with the magnetic field. The effects on the magnetic flux profiles and accompanying current density profiles are shown here for
four different applied fields. As the applied field becomes larger, the
critical current decreases. The slope of the flux profile changes with
the magnetic field, and it is also no longer linear with position in
the superconductor.
As we have seen previously, the transition from the flux pinning to the flux flow state is generally not sharp but occurs
over a range of currents during a resistive critical current
measurement. Several models for the shape of the resistive
transition have been developed to account for this behavior
(3941), but they all assume a distribution of pinning center
strengths within the wire. The idea was first proposed by
Baixeras and Fournet (42) but was not fully developed and
applied to technological superconductors until the 1980s. If
one assumes that the superconducting wire is made of an assortment of independent, current-carrying segments in series,
each with its own value of critical current (as determined by
the flux pinning defects within each segment), then the equation for the V(I) curve can be written as

V (I) = A (I I
) f (I
) dI
(31)
where I is the local critical current of an individual wire segment, A is a constant, f(I) is the critical current distribution
of the segments in the wire, and the integral is performed
from zero current to I.
In essence this states that the voltage at any transport

current I results from the segments with IC I and are therefore in flux flow. Variations in the critical current of each segment may occur because of variations in processing, chemical
inhomogeneities, or changes in geometry. For wires that are
very homogeneous along their lengths, the distribution in
critical currents f(I) is very narrow, and the V(I) characteristic is quite sharp and steep, leading to large n-values and the
associated measure of high-quality in technological superconductors (43). At the other extreme, a large variation in the
critical current distribution leads to a broad V(I) transition
and a low n-value.
The importance of the model for helping the development
of high JC superconductors is that it can be used in reverse to
determine the critical current distribution function from a
measurement of the V(I) curve. By taking the second derivative of both sides of Eq. (31),
d 2V
= A f (I)
dI 2
(32)
The second derivative of the experimentally determined V(I)

curve yields the distribution f(I) of critical currents within the
sample. In practice it is found that the resistive critical current measurement is always at a current well below the peak
of the critical current distribution f(I). Technological superconductors are limited by the weakest flux pinning segment
along the sample length.
Temperature Dependence: Flux Creep and FLL Melting
As the temperature of a strongly pinned superconductor is
increased, the thermal energy available to the pinned flux
lines increases, allowing the possibility of thermally induced
flux depinning. The higher temperatures lead to thermal excitation of the flux lines within the pinning potential wells. In
low temperature superconductors where the thermal energies
are low, the thermal energy is typically much smaller than
the depth of the pinning potential. This led early investigators to name the effect flux creep because of the similarity
to mechanical creep of crystalline materials at stresses much
lower than the yield stress.
The theory and model for flux creep was developed by Anderson (44) and elaborated by Kim and Anderson (45). The
basic premise is that flux lines at any finite temperature, experience a thermal excitation due to the thermal energy kT.
The excitation energy has a statistical probability of causing
a flux line pinned by a pinning center to become unpinned.
The probability of this thermal depinning occurring is given
by the Arrhenius factor
p = exp
U
kT

(33)
where U is the depth of the energy well of the pinning center

(Fig. 10).
For an isolated flux line, assuming no temperature gradient, there is no preferred jump direction because both the pinning potential and the thermal excitation are spatially symmetrical. Thus the flux line would execute a random walk as
the thermal excitation allowed it to leave the pinning potentials, and no net change in flux would occur.
For a flux line in a magnetic field gradient, the flux motion
associated with the thermal activation of the flux line derived
673
from a diffusion argument is
dB
U
= Aw0 exp
dt
kT

(34)
where A is a factor that depends on the local magnetic field

and field gradient and wo is the frequency with which the flux
line tries to jump out of the well.
The flux gradient provides a driving force that leads to diffusion of the flux line preferentially down the flux gradient.
This produces a changing magnetic flux density according to
Eq. (34) which can be experimentally measured. The flux gradient can exist because of the presence of either a transport
current or a magnetization current in the sample.
In the case of a magnetization-induced flux gradient, the
flux gradient would decay, eventually disappearing, even
though the flux gradient (or magnetization current density) is
less than the critical current density. This is in contrast to
the assumptions of the Bean critical state model, in which the
flux gradient adjusts itself to match the critical current density at all points in the sample. In the original experimental
work on flux creep (46), the measured decay of the magnetization currents translated into a decay time of 1092 years, effectively infinite time, so that, even with flux creep, the persistent magnetization supercurrents that flow as a result of the
MeissnerOchsenfeld effect are truly persistent.
The situation is somewhat different when the flux gradient
driving force is produced by an externally applied transport
current. In this case, the flux gradient crosses the superconductor, and the effect of the flux creep is to move flux lines
across the sample in a manner identical to the flux depinning
process at JC. In the same way that the FLL motion causes a
resistive-like dissipation in the superconductor, the flux creep
motion also contributes to the dissipation. This results in a
power loss in the superconductor that must be supplied by
the external power supply and a consequent heating of the
superconductor caused by the flux motion. If the flux motion
is too great, the temperature increases, which increases the
probability of thermally activated flux motion, and a thermal
runaway to the normal state ensues.
At higher temperatures, the thermal excitation of the flux
lines becomes comparable to the pinning potential, and the
FLL is expected to move more easily. This, in fact, was one
early argument against the possibility of technologically important superconductors at high temperatures. The argument
was that no known basic pinning interactions are strong
enough to prevent thermal excitation of the FLL at high temperatures (e.g., at 77 K, liquid nitrogen), even at very small
transport currents. Fortunately for high temperature superconductors, this has not been the case. However, the concerns
about the higher thermal energy available for excitation of
the FLL out of the HTS pinning centers led along a path different from the original flux creep models into the theories of
flux lattice melting.
Much of the theoretical picture of thermal effects on the
FLL in HTS materials is at heart the same as in low temperature materials. The basic interests are in the possibility of
scaling the pinning force with temperature, magnetic field dependence of the JC, and the importance of the higher thermal
energy in the depinning and motion of the FLL. The terminology has developed differently, but the physical mechanisms of
flux pinning and depinning do not differ substantially.
674
SUPERCONDUCTING CYCLOTRONS AND COMPACT SYNCHROTRON LIGHT SOURCES
SUMMARY AND CONCLUSIONS

The critical current density of superconducting materials is a
technologically important parameter, often more important in
engineering design than the TC or the HC2. The theory of the
JC is broad and has provided useful insights into the physical
mechanisms responsible for the large values of critical current density found in technological superconductors. This article has described the standard resistive and magnetic measurements of critical current. The theories of superconducting
critical current range from the ultimate limits to the JC, as
described by the depairing current density, to the limits imposed by the thermal excitation of flux lines out of their pinning centers. The development of a bulk pinning force from
the summation of individual pinning center interactions leads
to the empirically observed scaling laws for the JC that are
important both as a tool for understanding the flux pinning
processes and as an extrapolating technique useful for technological design.
Ultimately, this article has attempted to show the connection between the measured values of the JC and the microstructural defects responsible for pinning the magnetic fluxline lattice in place. Understanding the physical processes
underlying the development of the JC has led materials engineers to design fabrication and processing that produces the
desired microstructural features necessary for high JC. In the
past 10 to 15 years, this has led to massive increases in the
performance of superconducting materials for applications requiring high current density and promises additional improvements in the future.
BIBLIOGRAPHY
1. M. N. Wilson, Superconducting Magnets, Oxford, UK: Oxford
University Press, 1983.
2. D. C. Larbalestier, in S. Foner and B. B. Schwartz (eds.), Superconductor Materials Science: Metallurgy, Fabrication and Applications, New York: Plenum Press, pp. 133200, 1981.
3. M. Suenaga, in S. Foner and B. B. Schwartz (eds.), Superconductor Materials Science: Metallurgy, Fabrication and Applications,
New York: Plenum Press, pp. 201274, 1981.
4. A. C. Rose-Innes and F. H. Rhoderick, Introduction to Superconductivity, Oxford, UK: Pergamon, 1969.
5. M. Tinkham, Introduction to Superconductivity, New York:
McGraw-Hill, 1975.
6. T. P. Orlando and K. A. Delin, Foundations of Applied Superconductivity, Reading, MA: Addison-Wesley, 1991.
7. A. M. Campbell and J. E. Evetts, Adv. Phys., 21: 199428, 1972.
8. H. Ullmaier, Irreversible Properties of Type II Superconductors,
Berlin: Springer-Verlag, 1975.
9. Y. B. Kim, C. F. Hempstead, and A. R. Strnad, Phys. Rev., 139
(4A): 11631172, 1965.
10. F. Volker, Particle Accelerators, 1: 205207, 1970.
11. H. K. Olsson et al., Phys. Rev. Lett., 66 (20): 26612664, 1991.
12. N.-C. Yeh et al., Phys. Rev. B, 45 (10): 57105713, 1992.
13. T. S. Kreilick, Niobium-Titanium Superconductors, in Metals
Handbook, Vol. 2, 10th edition, Materials Park, OH: ASM International, pp. 10431059, 1990.
14. F. B. Silsbee, J. Wash. Acad. Sci., 6: 597, 1916.
15. A. A. Abrikosov, Sov. Phys. JETP, 5: 1174, 1957.
16. U. Essman and H. Trauble, Phys. Lett., 24A: 526527, 1967.
17. J. D. Cribier et al., Progress in Low Temperature Physics, Vol. V,

C. J. Gorter (ed.), Amsterdam: North-Holland, 1967, p. 161.
18. G. Zerweck, J. Low Temp. Phys., 42 (1): 19, 1981.
19. A. Gurevitch and L. D. Cooley, Phys. Rev. B, 50 (18): 13563
13576, 1994.
20. L. D. Cooley, P. J. Lee, and D. C. Larbalestier, Phys. Rev. B, 53
(10): 66386652, 1996.
21. H. Trauble and U. Essmann, Phys. Stat. Sol., 25: 395402, 1968.
22. J. Schelten, H. Ullmaier, and G. Lippmann, Phys. Rev. B, 12 (5):
17721777, 1975.
23. E. H. Brandt, J. Low Temp. Phys., 21: 709, 1977.
24. E. J. Kramer, J. Appl. Phys., 44 (3): 13601370, 1973.
25. E. J. Kramer, J. Electron. Mater., 4 (5): 839881, 1975.
26. A. I. Larkin and Yu. N. Ovchinikov, J. Low Temp. Phys., 34: 409
428, 1979.
27. R. Labusch, Cryst. Lattice Defects, 1: 116, 1969.
28. W. A. Feitz and W. W. Webb, Phys. Rev., 178 (2): 657667, 1969.
29. M. P. A. Fisher, Phys. Rev. Lett., 62 (12): 14151418, 1989.
30. B. Brown et al., Phys. Rev. B, 55 (14): 87138716, 1997.
31. D. Dew-Hughes, Philos. Mag., 30: 293305, 1974.
32. D. G. Hawksworth and D. C. Larbalestier, Proc. 8th Symp. Eng.
Probl. Fusion Res., 1979, Vol. 1, p. 245.
33. R. J. Hampshire and M. T. Taylor, J. Phys. F, 2: 89106, 1972.
34. C. Meingast, P. J. Lee, and D. C. Larbalestier, J. Appl. Phys., 66
(12): 59625970, 1989.
35. C. P. Bean, Phys. Rev. Lett., 8: 250253, 1962.
36. D.-X. Chen and R. B. Goldfarb, J. Appl. Phys., 66 (6): 2489
2500, 1989.
37. R. B. Goldfarb and J. V. Minervini, Rev. Sci. Instrum., 55 (5):
761764, 1984.
38. M. Daumling and D. C. Larbalestier, Phys. Rev. B, 40 (13): 9350
9353, 1989.
39. J. E. Evetts and C. J. G. Plummer, Proc. Intl. Symp. Flux Pinning
Electromagnetic Properties Superconductors, T. Matsushita, (ed.),
Fukuoka, Japan: Matsukama Press, 1986, p. 150.
40. D. P. Hampshire and H. Jones, Proc. 9th Intl. Conf. Magnet Technology, C. Marinucci (ed.), Swiss Institute for Nuclear Research,
Geneva, Switzerland, 1985, p. 531.
41. W. H. Warnes and D. C. Larbalestier, Appl. Phys. Letts., 48 (20):
14031405, 1986.
42. J. Baixeras and G. Fournet, J. Phys. Chem. Solids, 28: 1541, 1967.
43. W. H. Warnes and D. C. Larbalestier, Cryogenics, 26 (12): 643
653, 1986.
44. P. W. Anderson, Phys. Rev. Letts., 9 (7): 309311, 1962.
45. P. W. Anderson and Y. B. Kim, Rev. Mod. Phys., 36: 39, 1964.
46. Y. B. Kim, C. F. Hempstead, and A. R. Strnad, Phys. Rev. Lett.,
9 (7): 306311, 1962.
WILLIAM H. WARNES
Oregon State University
710
SUPERCONDUCTING FAULT CURRENT LIMITERS

The control of fault currents in electrical transmission and
distribution networks is a major challenge. Short-circuit current levels are considerably larger than the rated current. All
electrical equipment exposed to the short-circuit current must
be designed to withstand in particular the mechanical forces
under fault conditions, which are generally proportional to
the square of the current. Circuit breakers that isolate the
fault must be capable of absorbing the fault energy, as well
as be proportional to the square of the fault current at the
instant of breaker operation. Consider an infinite bus feeding
a load through a transmission line and transformer with 8%
combined impedance. In the event of a short circuit across the
load, the short-circuit current is limited only by the line and
transformer impedance. The steady-state short-circuit current will be 12.5 times larger than the rated current, and the
initial transient current can reach a value about twice the
peak steady-state fault current, resulting in short-circuit electromagnetic forces that are 625 times higher than the peak
forces under rated conditions. To survive such a fault, all circuit components must be designed to withstand the peak
forces.
A fault current limiter restricts the current fed to the fault
to a lower value than with the limiter absent. Unlike mechanical breakers, where the mechanical inertia in the mechanism
delays any effect on the fault current during the first few cycles, fault current limiters insert an impedance already in the
first half cycle, thus limiting the peak current transient of the
short-circuit event. Then all the network electrical equipment, such as lines, transformers, and circuit breakers, could
be designed for smaller fault current amplitudes, which would
result in a reduction of material and cost. In many transmission and distribution networks, load growth, distributed generation, and interconnections have produced increased fault
levels that exceed the capability of the installed circuit
breaker base to interrupt prospective fault currents. In some
situations, the fault levels exceed the available ratings for circuit breaker replacement or upgrading, and economically or
operationally attractive interconnections cannot be made.
With the introduction of a fault current limiter, a system
could use breakers with a smaller interrupting rating than
the system fault level. Thus electrical grids can be interconnected to produce stiffer, more reliable systems without upgrading existing breakers to higher interruption ratings.
No fault current limiter technology is widely accepted or
used today. Superconducting fault current limiters (SFCL)
form one class of fault limiting technology under active development. After several technical generations of development,
niobium-based low temperature superconductor (LTS) technology is mature enough for consideration by the electric utility industry. However, market penetration of LTS devices has
been impeded by high cost and the use of liquid helium (LHe)
cooling. The 1986 discovery of a new class of ceramic high

temperature superconductors (HTS) (1) and the subsequent
active development of HTS raised hopes for inexpensive superconductor (SC) applications due to cheaper HTS conductor
and significantly reduced refrigeration requirements. The discovery stimulated reevaluation of all previously proposed SC
applications, including several SFCL embodiments, and
prompted many innovative SFCL concepts and designs.
Many SFCL programs are active, ranging from the proofof-concept stage all the way to precommercial prototype development. These programs are pursuing several distinct types
of SFCLs, and it is not yet clear which approaches will ultimately be both technically and economically viable (2). At
present there is no off-the-shelf SFCL product available in the
marketplace, and even the most advanced of the several
SFCL commercialization programs are not expected to be
ready to offer utility-qualified products for some years to
come. Here we outline the key issues of SFCL technology and
application to give a basis for evaluating present and future
developments.
Almost all SFCL development is targeted to power frequency grid applications. Conceptual studies of utility generation and transmission systems with a large proportion of SC
generators, transformers, and transmission cable have looked
to SFCLs to limit the higher fault levels inherent in the lower
fault impedances of the SC system components. SFCLs would
also be useful in high voltage direct current (HVDC) transmission, where currently the challenges of dc fault current
interruption have not been solved. More specialized applications can be considered in high-power radio-frequency systems, magnet energization circuits associated with high-energy physics, and aerospace and shipboard systems. This
article will focus on the electric utility, power-frequency type
of application. Several surveys have been carried out of the
applicability of SFCL technology for electric utility needs
(3,4), and some studies have projected SFCL costs (5).
The most important characteristics for the user is the limiting current level. As discussed later, for most utility grid
applications, a fault level reduction of 10% to 50% is needed
to coordinate with existing protection, whereas for industrial
end users more aggressive reduction may be appropriate.
Other parameters are the trigger level for the initiation of
limiting and the reset time for the limiter to re-arm after limiting operation.
Most SFCLs may be classified into (1) quenching SFCLs,
which rely on the superconducting-normal phase transition to
switch from a low to a high impedance; (2) desaturating
SFCLs where ferromagnetic material switches between saturated and unsaturated states and the SC biases the quiescent
operating point to the low-impedance unsaturated state; and
(3) semiconductor SFCLs where the semiconducting elements
switch to a high-impedance operating point from a low-impedance point defined by bias applied from the SC. Superconductivity is essential only to the quenching SFCL, but it is useful
in desaturating and semiconductor fault limiters to reduce
the bias system losses and improve system compactness.
ELECTRIC UTILITY GRID APPLICATION

To be useful, the SFCL must operate quickly enough to limit
the first half-cycle of the fault so that the interrupter peak
711
Transformer A
Transformer B
A
Secondary breaker
B
Secondary breaker
Bus tie
Bus A
;
;
Bus B
Fault FB
Fault FF
Breaker
SFL
Figure 1. Superconducting fault limiter application in a substation.
current duty is reduced. This imposes a fast SFCL triggering

requirement. Most concepts are self-triggering, with the instantaneous fault current driving the SFCL transition to high
impedance. Otherwise, external triggering driven by state-ofthe-art relaying (detection time cycle) is used. A tunable
self-triggering level is a very attractive SFCL feature. After
SFCL triggering, a series power circuit breaker interrupts the
fault at one of the line current zeros, with an operating delay
from 1 to 5 cycles. The SFCL must endure the limited fault
current until the maximum backup breaker opening time, as
determined by the utilitys protection coordination. Triggering
selectivity should also be considered to ensure that the SFCL
does not trigger on transformer inrush and motor-starting
transients.
We can distinguish between a trimming SFCL (where the
effective SFCL impedance ZL Zs, the system source impedance), which reduces the prospective fault current by 050%,
and a dominant SFCL (ZL Zs), which achieves reductions
significantly over 50%. In existing networks that have incrementally grown to high fault levels, utilities require only
trimming limiters so that the existing selective relay equipment can be retained. In fact, dominant limiters would require major reconfiguration of system protection coordination,
which is always unattractive. Limiting the short-circuit current to values three to five times the rated current is a typical
requirement in many cases. In some specific applications, a
dominant SFCL, such as to protect a complete new installation for an industrial customer, may be used.
The possible locations for a fault current limiter in a substation are shown in Fig. 1. Two transformers feed two separate buses, providing power to several feeders. Interconnecting the two buses by closing the bus tie breaker will improve
load voltage regulation and overall system reliability, at the
cost of increased fault levels, because both transformers can
now feed into a fault. Fault current limiters are installed in
712
series with all breakers. In the case of a bus fault (shown as

FB), the fault current limiter in series with the B transformer
secondary side breaker limits the short-circuit current. In the
case where the bus tie breaker is closed, the fault current
limiter in the bus tie reduces the short-circuit current contribution from the A transformer. Should a feeder experience a
fault (shown as FF), the limiter in the feeder circuit reduces
the short-circuit current. If the A transformer was a later installation and the parallel connection of both buses is desirable, the installation of the fault current limiter in the bus tie
connection could enable continued safe operation of the system with the original breakers.
For an SFCL device to be accepted in an electric utility
system, it must meet general safety, reliability, availability,
basic insulation impulse level (BIL), and overload requirements and ratings comparable to other utility components
like transformers or circuit breakers. Currently there are no
standards for SFCL application to utility systems. Pertinent
sections of standards for other system components can be
adapted to meet the fault current limiter design requirements. The efficiency should be above 99%. The SFCL must
operate for reclosure duty and must be fail safe. Under all
operating conditions, stray field levels accessible to personnel
should be no higher than levels generated by other equipment. The exposed field levels must be limited to avoid personnel harm and projectile effects on ferromagnetic tools (and
also typical debris such as discarded nuts and bolts).
In addition, the utility user must be reassured about new
issues raised by the introduction of SC technology. Operational experience with utility system integration of SC is very
limited. A 30 MJ/10 MW superconducting magnetic energy
storage (SMES) system was installed in 19831984: Failures
in the LHe refrigeration took the system down, but the SC
coil did not fail. The operation of a small number of commercial LTSLHe 13 MJ micro-SMES units in 480 V uninterruptable power supply systems in industrial power system environments since 1993 has demonstrated reliability matching
battery systems. In addition, reliable cryogenic referigerators
have been developed for commercial application in medical
magnetic resonance imaging (MRI). All of the above experience points to the technical feasibility for an LTSLHe SC
system to achieve the same reliability as a conventional system in an electric utility environment, but such reliability has
not yet been demonstrated. The ongoing testing of a 1.2 MVA
HTSliquid nitrogen (LN2) SFL system at a Swiss hydro
electric power station (6), followed by the planned 1998 demonstration of a 21 MVA HTS SFCL system at a distribution
substation (2), will provide much-needed utility environment
operational experience and reliability data, particularly for
HTSLN2 systems.
The cryogenic system of the SFCL may present the utility
with a new unfamiliar consumable, the liquid cryogen. It is
very unlikely that LHe will be acceptable, because of its cost
and specialized handling procedures. Liquid nitrogen (LN2)
might be accepted because it is much cheaper, widely available, and already in utility use in some transmission cable
applications, where it is used to freeze the oil dielectric in the
course of maintenance procedures. Closed cryogenic systems
are the most attractive, where limited amounts of cryogen
boiloff are reliquefied, or where no liquid cryogen is used at
all.
QUENCHING TYPES
The size of the SC element, which changes from a very low
impedance state to a finite resistance (a quench), is based on
the material properties of critical current Jc and normal resistivity n, and the system requirements of crest trigger current
Ic0 and normal state SFCL resistance R, to give a conductor
cross-sectional area of Ic0 /Jc, length RIc0 /Jcn, and volume
2
RIc0
/Jc2n. In general, high Jc2n is desirable to minimize superconductor requirement.
The fault transient is divided into three time regimes: a
prelimit regime (period I) where I Ic0 and the limiter is not
yet active, a resistance growth regime (period II) where the
SFCL is in a partially resistive state below the critical temperature Tc and the limiter effective resistance Re VL /I (VL
is limiter voltage) is rising from 0 to R, and a full-resistance
regime (period III) where the SFCL temperature is above Tc
and all superconductivity has quenched. Period I behavior depends on Vs, L, and Ic0 alone.
The electrical performance in period III is analyzed in terms
of an RL circuit transient, and the average temperature transient follows from the resulting energy absorption RI2 dt divided by the SFCL mean heat capacity C (including all material
coupled by thermal diffusion to the SC element over the limiting time). The behavior in period II is more complex and is
best solved numerically (7). It is useful to describe the SFCL as
weak or strong, based on whether R is less or more than Vs /Ic0.
With a strong limiter, the limiter voltage will overcome the
source voltage during period II at which time the fault current
will peak. It is still possible that there will be a subsequent
higher current peak due to the power-frequency variation of
Vs. It is also useful to describe the SFCL resistance growth as
fast or slow, depending on whether the limiting current at the
end of period II is more or less than the quasiequilibrium fault
current Ic Vs /R. The fast weak SFCL will not force a current
peak in period II, because always VL Vs. The slow weak SFCL
will force a current peak in period II because period II extends
so long that the limiter current grows until the voltage across L
reverses and limiter dI/dt changes sign. It should be noted that
the relative strength and speed of the SFCL depend on the
power system parameters as well as the SFCL parameters R,
C, Ic0 and the difference between critical and quiescent temperatures.
Four simplified cases are considered in Fig. 2 with low
(weak) and high (strong) R and slow and fast dR/dt. The fast
strong limiter gives the lowest peak let-through current and
I2dt to interruption, but it also will produce SFCL terminal overvoltage. Weakening a fast limiter will reduce the overvoltage
at the cost of higher let-through current and I2dt. Slowing down
a fast strong limiter will increase the let-through current and
I2dt and may increase the overvoltage. Clearly SFCL strength
and speed are both desirable for the primary function of current
limiting, but they will produce overvoltage. If only weak limiting is desired, the limiter overvoltage can be controlled by
speed adjustment, and in fact all overvoltages can be suppressed by keeping speed below a particular threshold.
Period III continues until either the primary or backup circuit breaker interrupts the fault. Here the normal state SC
element carries a large current and dissipates high power,
which dominates total SFCL heating. For example, a resistive
quenching SFCL that trims the fault level of a 345 kV transmission system (with X/R ratio 15) by 10% from 70 kA to 63
30000
25000
Slow, weak SFCL
20000
Current
Phase I
Phase II
Phase III
Fast, weak SFCL
15000
Slow, strong SFCL
10000
5000
0
0.000
Fast, strong SFCL
0.002
0.004
0.006
0.008
0.010
0.012
Time
30000
25000
Phase II
Phase III
Fast, strong SFCL
Voltage
20000
15000
Slow, strong SFCL
10000
5000
0
0.000
Slow, weak SFCL

Fast, weak SFCL
0.002
0.004
0.006
0.008
0.010
0.012
Time
Figure 2. Four types of quenching SFCL performance depending on
limiter strength and speed. Direct source voltage is applied; hence,
power frequency effects are absent.
kA will dissipate 4.8 GW per quenching SFCL phase during

period III. The main challenge in quenching SFCLs is SC energy management under fault conditions.
Quenching SFCLs are in-line (line current flows in the SC
element) or transformer-coupled. In all in-line quenching
SFCLs, SC heating (and thus the amount of SC and cryogenic
system size and capacity) is reduced by a parallel fault current bypass resistor (8), inductor (9), or varistor (10) with
lower effective impedance than R. The SFCL-limiting impedance is of course reduced, but the symmetrical SC current is
also reduced. The bypass element is large with significant
(but not prohibitive) cost. Both SC and conventional (copper
conductor) bypass inductors have been proposed. There is little to recommend the SC variant because it would need to be
rated to the fault current level for both currents and forces.
A conventional reactor could be quite modestly sized to
achieve the necessary transient duty, and a bypass resistor
would also be reasonably sized. In weak current limiting applications, when designing for the same peak SC element
temperature limit, significantly more SC material will be required for the resistive bypass design. A bypass varistor
would have the additional benefit of limiting the SFCL transient overvoltage, but it has a high-energy capability requirement.
All in-line SFCLs require cryogenic current leads to connect the line current to the cryogenic system. These leads
713
must be thermally rated for the SC element fault transient

current and will become the dominant heat leak into the cryogenic system. Respectably rated distribution-level in-line
SFCLs have been built with both inductive (9) (6.6 kV/1.5 kA)
and resistive (8) bypass (7.2 kV/1 kA). Both used relatively
short lengths (on the order of 100 m) of expensive power-frequency capable ultrafine filament NbTi superconductor with
high-resistivity cupro-nickel stabilizer matrix, operating with
LHe cryogen at 4 K. Important design issues include ac derating of multistranded conductor from single-strand dc values
of Ic, immobilization of the conductors to avoid spontaneous
quench due to conductor movement, and use of cryocoolers to
produce zero-helium-boiloff systems. Heat losses from induced currents in the metallic cryostat and ac losses in nonstranded current leads and lead-to-SC wire joints were a
problem. It is expected that limiting will deposit several
megajoules in the cryostat, leading to significant LHe vaporization, which must be vented or captured and reliquefied.
Open-cycle operation (venting the boiloff) will drive the LHe
replenishment requirement, whereas the reliable reliquefaction required for a closed system is technically challenging.
HTS materials have been applied to in-line SFCLs, but at
a more modest scale than the NbTi devices because of the
difficulty in producing transposed multifilamentary conductors for low ac loss. An advantage of HTS materials is the
relatively high n, which can exceed 10 m. Modeling studies (10) suggest that Jc of at least 10100 MA/m2 (110 kA/
cm2) is required. The high n drives designs to relatively low
SC volumes and high peak temperatures. Extra heat capacity
may be added (11) by using thin (0.22 m) HTS films deposited onto sapphire or YSZ (yttria-stabilized zirconia) substrates, which provide much-needed heat capacity. It remains
to be seen whether the preferred in-line HTS element geometry will be wound multifilament or thin film.
In the shielding SFCL, the SC is coupled to the line current by transformer action [Fig. 3(a)] and forms a shorted
winding. A preferred geometry is a single turn of HTS material: Bi-2212 material is commercially available in tubular
sizes up to 40 cm diameter (12). The thermal transient can be
analyzed by the previously derived equations after the device
circuit is transformed by the turns ratio to a T-equivalent circuit. The insertion impedance will be the leakage inductance
between the line and SC windings (13), and the limiting impedance will be the magnetizing inductance in parallel with
the SC winding normal R. Desirable features of the shielding
SFCL include the isolation of line current and potential from
the SC element and no requirement for cryostat current
leads. Several groups (1416) are actively pursuing this
promising approach and 6 months of operating experience
protecting a hydroelectric plant auxiliary transformer circuit
has been reported for a 1.2 MVA 10.5 kV Bi-2212 LN2-cooled
shielding SFCL (6). Design challenges include the reduction
of cryostat circulating currents by the use of nonmetallic cryostats, thus minimizing circulating currents in the metalized
multilayer insulation; the mechanical support of transient
bursting or compressive forces on the HTS shield; the amelioration of transient thermal inhomogeneity; and optimization
of recool time. Most designs utilize nonlaminated ferromagnetic cores. Aircore devices will have larger leakage impedance. Control of shielding SFCL trigger level has been proposed by using a tertiary winding (16).
714
Ferromagnetic core
SC
limiting
coil
ac line in
SC
winding
Line
winding
ac line out
Bias
power
supply
(b)
Figure 3. Principles of shielding and
semiconductor SFCLs: (a) shielding
SFCL, (b) semiconductor bridge SFCL.
(a)
A quenching-type SFCL operation may be self-initiated by

the fault current as discussed previously, or some quenching
mechanism (17) may be externally triggered by a fault detection relay. Digital microprocessor-based relaying is now capable of giving subcycle fault detection, so that detector delay is
no longer a barrier. Several trigger methods have been proposed, including discharging a capacitor to give a magnetic
field pulse from an auxiliary coil, discharging a capacitor to
add an extra overcurrent component, triggering heaters on
the SC, or sending a warm cryogen burst into the cryostat.
Currently, most SFCL development is aimed at self-triggering
using fault overcurrent alone. However, in later generations,
commercial SFCLs will be differentiated by the triggering
system to maintain trigger level control independent of SC
property variations and cryogenic operating point variability
and to give flexibility in setting the SFCL trigger level.
Under fault limiting, some parts of the SC element will
enter the lossy period II regime before others as a result of
local variations in material quality, temperature, and magnetic field. The concern is whether such locations will bear a
disproportionately large part of the complete limiting burden,
in terms of temperature, electric field, or mechanical stress.
This has been observed in both analysis and experiment
(7,18), and significant thermal damage has been reported
(14). For HTS systems, quench propagation is so slow (centimeters per second) as to be ineffective during power-cycle
time periods, whereas LTS systems faster propagation (meters per second to kilometers per second) may overcome inhomogeneous initiation. This inhomogeneity is most critical
when the peak fault current is close to the trigger current,
leading to a prolonged period II with incomplete fault current
commutation to the bypass element (if present). Thus every
quenching SFCL will be vulnerable when the fault occurs at
the system phase corresponding to the lowest prospective
fault current and should be designed to be capable of withstanding inhomogeneous initiation until the interrupter
opens on a current zero.
After operation, a quenching SFCL must recool to its operating temperature before becoming available again. The
cryogenic system temperature profile is inverted, with a hot
SC and cool thermal shield. A high heat capacity thermal
ballast to level the recool system thermal load is attractive.
No matter what the recool system, the SC element reset time
is seconds or longer, a major shortcoming for reclose-type
fault relaying systems. Multiple parallel SFCL elements are

a common solution, so that a cold element is available immediately after SFCL operation; the drawback is increased system complexity and cost from duplication of the SC element,
thermal decoupling between elements, and a switching subsystem.
FERROMAGNETIC TYPES
An SFCL using the principle of ferromagnetic desaturation
was built in the early 1980s (19). A coil wound on a closed
iron core will have a very high ac impedance when unsaturated. If, however, the iron core is saturated, the same coil
will exhibit a very low impedance. Two cores are used per
phase, each with an ac winding carrying the line current and
also an SC dc bias winding. The dc bias windings are connected to saturate the two cores in opposite directions. The ac
windings are series connected so that instantaneous line current adds to one cores bias MMF and substracts from the
MMF of the other core. Under normal operation, both cores
are so heavily saturated by the dc bias current that the MMF
contribution of the ac does not bring either core out of saturation, and the impedance of the core is small. Under shortcircuit conditions, however, each half wave of the higher ac
fault current brings each core out of saturation in turn, thus
inserting a high impedance into the circuit and limiting the
current. At the fault current peak, the cores resaturation (in
the opposite polarity to the quiescent state) must be limited.
In a three-phase system, six cores are necessary, two for each
phase, but a single SC dc bias winding (and cryostat enclosure) passing through all six core windows can be used. A
single phase, 3 kV, 556 A prototype using an LTS coil was
built and tested with peak short-circuit currents of close to 15
kA. Although the device performed as predicted, no commercial units were ever built because of the large size of a transmission system sized unit. The total core volume required
corresponds to a transformer rated at twice the fault volt-amperes. The desaturating SFCL does not necessarily quench
during limiting, so that it has no reset time delay, and the
limiting level can be varied by adjusting the dc bias current.
The impedance of a series line reactor can also be changed
by moving armature methods that vary the effective gap of
the reactor magnetic circuit. One approach under construc-
tion (20) uses a mechanically balanced disk (hypercooled to

5070 K for enhanced magnetic and conductivity properties)
to rotate magnetic and nonmagnetic sectors in and out of the
gap, in times below cycle. The rotation is induced by inverter-fed HTS coils, controlled by line overcurrent relaying.
Using SC in the rotation control coils significantly reduces
device quiescent losses.
SEMICONDUCTOR TYPES
The nonlinear resistance of semiconductor junctions is used
in the bridge-type fault current limiter, which combines
power electronics and an SC coil (21). Figure 3(b) shows the
circuit of a single phase. A bias power supply provides a current in all four semiconductors. As long as the load current is
less than the bias current, all four semiconductors are forward biased, and the ac load current flows unimpeded by the
bridge circuit, assuming negligible losses in the semiconductors. In the quiescent condition, each thyristor conducts half
the bias current superimposed with half of the line current.
When a short circuit occurs, one pair of semiconductors is
turned on, and the other pair is turned off in each half cycle,
automatically inserting the bridge inductor into the circuit.
The inductor limits the fault current. Because the inductor
carries a bias current under normal condition, use of an SC
coil reduces the overall system losses. The bridge-type fault
current limiter has several attractive features, such as automatic insertion of the current-limiting reactor, reduction of
the first half-cycle short-circuit current, precise control of the
amplitude of the short-circuit current, complete current interruption in less than a cycle if desired, operation with multiple
fast reclosures, and high efficiency. In addition, it is conceivable to use the controlled thyristor bridge for other transmission or distribution network functions such as var control. For
distribution and transmission system semiconductor SFCLs,
the equivalent thyristors indicated in Fig. 3(b) are series
strings of thyristors, similar to those used in static var compensators or electronic transfer switches. A 2.4 kV, 2.2 kA
single-phase fault current limiter was successfully tested in
1995 and a 15 kV, 800 A, three-phase unit is being designed
to be tested in 1998 (2).
OTHER TYPES
SFCL designs where the limitation arises from magnetic circuit unbalance under fault conditions have been proposed
for example, with a three-phase SC winding on a common
core (22). The positive and negative sequence inductances
equal the device leakage inductance, whereas the zero-sequence inductance equals three times the self-inductance.
Single line-to-ground faults, which are the most common, are
then limited by the magnetics alone, without quench or thermal recovery delay, whereas the rarer two- and three-phase
faults will be limited by in-line SC winding quench. Thus, the
SC winding thermal performance must be designed as for all
in-line quenching SFCLs. An HVDC transmission SFCL (23)
has balanced positive and negative line currents canceling in
the magnetic circuit, whereas ground faults will unbalance
the SFCL. Under these conditions, the SFCL inductance limits the fault current dI/dt until the core saturates.
715
SUPERCONDUCTOR REQUIREMENTS
Either niobium-based LTS or ceramic HTS materials may be
considered. LTS conductor technology is relatively mature,
with kiloampere-rated commercial conductor optimized for either ac or dc operation available in long lengths from several
vendors. Operating Jcs are in the gigampere per square meter
range and operating temperatures are up to 6 K (NbTi) or 14
K (Nb3Sn). To produce the required normal-state resistance
in quenching SFCLs, the LTS conductor must have high values of effective n, which is governed by the stabilizer resistivity. The highest values (in available conductors) of about 30
nm (300 K) are achieved with CuNi alloy. Elementary power
frequency conductor strands have 10100A Ic and pass
through two levels of cabling and twisting to achieve the required kiloampere-class Ic. At each level of cabling, the
strands are grouped around a copper core, which enhances
quench propagation. Current sharing is a problem that is further accentuated when kiloampere-class conductors must be
paralleled to operate at 10100 kA-class transmission network-rated currents. Such an ac conductor is more expensive
than a dc conductor. Desaturating and semiconductor SFCL
types do not need the same high ac performance and use more
economical dc LTS conductors. The major drawback of LTS
conductors is the refrigeration and cooling required to maintain the required operating temperature.
Bulk, film, or wire/tape HTS conductors can be used in
SFCLs. Bulk rings or tubular cylinders, particularly useful
for shielding SFCLs, have been manufactured from Bi-2212
(12), Bi-2223 (24), and YBCO (9). All the materials have Tc
above 85 K, effective Jc in the 0.10.5 GA/m2 range at 77 K,
and n around 10 m. Even though bulk Bi-2212 parameters
are overall no better than the competing materials, Bi-2212
has the most promise because there are practical manufacturing techniques to produce technically useful cylinders with up
to 8 mm wall thickness and 40 cm diameter. Bulk forms are
also used as current leads operating between Tc and T0. HTS
materials show significantly higher Jcs in thin films. For example, 200 nm-thick YBCO on a CeO2-buffered sapphire substrate exhibited Jc 30 GA/m2 and was investigated for
quenching SFCL designs (11). The substrate heat capacity
and heat transfer to LN2 cryogen limits the temperature rise.
Such high Jc requires using a stabilizer parallel layer to protect against catastrophic local overheating, so that the high
intrinsic n is not used. YBCO thin film panels up to 20 20
cm have been tested, which could be used in arrays to produce
quenching SFCLs. Thick Bi-2212 films are also useful: an
array of MgO substrate cylinders (45 cm diameter, 12 cm
long), each supporting an 0.5 mm Bi-2212 layer, has been
used in a 6.6 kV, 400 A shielding SFCL (25). Thick film Jcs
are similar to bulk HTS material.
The thrust of HTS wire and tape development is to develop
stabilized conductor for magnet applications. Bi-2223 conductor, produced by the powder-in-tube (PIT) process, is currently closest to commercialization but limited to low operating magnetic field levels. The silver stabilizer of PIT
conductor (with around 4 nm at 100 K) bypasses the much
higher intrinsic Bi-2223 n. Multifilamentary twisted conductors are under development to produce lower ac losses. The
manufacture of wire from other HTS materials such as Bi2212 and YBCO is also under active development, and it is
716
premature to predict which material will be most applicable

to SFCLs.
The main attraction of HTS materials over LTS for SFCLs
is higher operating temperatures and lower refrigeration costs.
Present HTS Tcs point to useful operating temperatures in the
3060 K range. As in other HTS applications, tradeoffs must be
made between the advantages of using cheap LN2 cryogen (77
K operation), larger amounts of HTS, more exotic cryogens
(e.g., liquid neon, boiling at 27 K), or none coupled with better
use of the HTS (lower operating temperatures).
CRYOGENICS
Cryostat thermal design is of paramount importance because
the required cooling power is an important part of the operating cost. The quiescent cryostat heat budget consists of
SC ac losses, ambient-to-cryogenic current lead heat leak, dielectric losses, and other conduction and radiation to the cryostat. Even though the design and selection of superconductor
for low losses have been already discussed, and cryostat and
current leads design is well understood, dielectric ac losses
present a less-appreciated heating mechanism. The heat dissipation for insulation deployed in high ac electric field regions should be determined from cryogenic loss tangent.
Transmission and distribution-rated in-line SFCLs using
HTS may have quiescent heat loads 300 W and up as a result
of ac loss and current lead penetrations. In all cases, a closed
system needs refrigeration capacity to extract the significant
heat leak. When refrigerator power is interrupted, the cryostat temperature will rise at a rate governed by ac losses,
resistive losses, heat leak, and heat capacity, and ultimately
the SC system will lose superconductivity. Auxiliary power
interruptions are quite likely in an electric utility setting, and
an SFCL must have some thermal ride-through capability.
Electric utilities will find the cooldown time to operating temperature an important parameter. Liquid cryogen-based cooldowns may take as little as a few hours, whereas refrigerator
cooldown may take many hours to days.
SFCL cryostats may use liquid or gaseous cryogen for heat
transfer and storage or may operate under vacuum. Cryogen
consumption must be kept low to avoid replenishment servicing any more often than annually, so refrigeration must have
the capacity to meet the quiescent heat leak. In the typical
power and temperature range for LTS systems, two-stage Gifford-McMahon (GM) cycle cryocoolers [with possibly a JouleThompson (JT) valve] are typical, whereas for HTS systems,
one- or two-stage GM or Stirling cycle cryocoolers are candidates. The energy efficiencies of these refrigerators are well
below the Carnot cycle limit, and typically a 7 kW helium
compressor will be required for 23 W cooling at 4 K or 100
200 W at 80 K. The maintenance interval of cryocoolers depends on internal seal wear, oil mist freeze-out, and JT valve
clogging. Use of a good oil-free high-pressure ambient temperature helium compressor gives maintenance intervals longer
than one year.
ties to reap the economic benefits of operating transmission

lines at the highest possible power transfer levels and the
operational benefits of transmission system interconnection.
High-temperature superconductor developments may soon become commercially available to enable productization of some
of the many SFCL approaches under development at present.
The utility requirements of equipment robustness and reliability present a challenge for SFCLs which can be technically
met. A number of utility demonstration projects of SFCLs are
either underway or close to being underway, which will provide much-needed operational, robustness, and reliability
data. A yet unresolved issue is ultimate SFCL cost resulting
from the inherent costs of high-technology cryogenic and superconducting systems. There is scope for significant inventions and improvements in fundamental SFCL concepts and
their implementation and integration into electric transmission and distribution systems and components.
BIBLIOGRAPHY
1. J. G. Bednorz and K. A. Muller, Possible high-Tc superconductivity in the BaLaCuO system. Z. Phys. B, 64: 189193, 1986.
2. E. Leung, Surge protection for power grids. IEEE Spectrum, 34
(7): 2630, 1997.
3. R. F. Giese and M. Runde, Assessment study of superconducting
fault-current limiters operating at 77 K. IEEE Trans. Power Del.,
8: 11381147, 1993.
4. G. C. Damstra et al., Superconducting technology for current limiters and switchgear, CIGRE, 1990.
5. L. Salasoo et al., Comparison of superconducting fault current
limiter concepts in electric utility applications. IEEE Trans. Appl.
Supercond., 5: 10791082, 1995.
6. W. Paul et al., Test of 1.2 MVA high-Tc superconducting fault
current limiter, Applied Superconductivity 1997, Proc. EUCAS
1997, Third European Conference on Applied Superconductivity,
The Netherlands, June 30July 3, 1997, 2: IOP, Bristol UK,
1997.
7. M. Lindmayer, High temperature superconductors as current
limitersAn alternative to contacts and arcs in circuit breakers?,
Proc. 39th IEEE Holm Conference on Electrical Contacts, Pittsburgh 1993, pp. 110.
8. T. Verhaege et al., Experimental 7.2 kV rms/1 kA rms/3 kA peak
current limiter system. IEEE Trans. Appl. Supercond., 3: 574
577, 1993.
9. T. Hara et al., Development of a new 6.6kV/1500A-class superconducting fault current limiter for electric power systems. IEEE
Trans. Power Deliv., 8: 182192, 1993.
10. M. Lindmayer and M. Schubert, Resistive fault current limiters
with HTSCMeasurements and simulation. IEEE Trans. Appl.
Supercond., 3: 884888, 1993.
11. B. Gromoll et al., Resistive current limiters with YBCO films.
12. J. Bock, S. Elschner, and P. F. Herrmann, MCP-BSCCO 2212
tubes for power applications up to 10 kA. IEEE Trans. Appl. Supercond., 5: 14091412, 1995.
CONCLUSION
13. K. Kajikawa et al., Design and current-limiting simulation of

magnetic-shield type superconducting fault current limiter with
high Tc superconductors. IEEE Trans. Magn., 32: 26672670,
1996.
SFCL is an emerging technology that is being driven by the

confluence of superconductor capability developments and
electric utility need. Fault current limiters would enable utili-
14. J. R. Cave et al., Testing and modelling of inductive superconducting fault current limiters. IEEE Trans. Appl. Supercond., 7:
832835, 1997.
SUPERCONDUCTING FILTERS AND PASSIVE COMPONENTS

15. L. S. Fleishman et al., Design consideration for an inductive highTc superconducting fault current limiter. IEEE Trans. Appl. Supercond., 3: 570573, 1993.
16. M. Joo and T. K. Ko, The analysis of fault currents according to
core saturation and fault angles in an inductive high-Tc superconducting fault current limiter. IEEE Trans. Appl. Supercond.,
6: 6267, 1996.
17. H. J. Boenig et al., Anisotropic high temperature superconductors
as variable resistors and switches. IEEE Trans. Appl. Supercond.,
5: 10401043, 1995.
18. V. Meerovitch et al., Development of high-Tc superconducting inductive current limiter for power systems. Cryogenics, 34 (ICEC
supplement): 757760, 1994.
19. B. P. Raju, K. C. Parton, and T. C. Bartram, A current limiting
device using superconducting dc bias. IEEE Trans. Power Appar.
Syst., PAS-101: 31733177, 1982.
20. S. B. Kuznetsov, Superconducting fault current limiters. Superconductor Industry, 10 (1): 3039, 1997.
21. H. J. Boenig and D. A. Paice, Fault-current limiter using a superconducting coil. IEEE Trans. Magn., 19: 10511053, 1983.
22. S. Shimizu et al., Equivalent circuit and leakage reactances of
superconducting 3-phase fault current limiter. IEEE Trans. Appl.
Supercond., 3: 578581, 1993.
23. T. Ishigohka and N. Sasaki, Fundamental test of new dc superconducting fault current limiters. IEEE Trans. Magn., MAG-27:
23412344, 1991.
24. M. Ichikawa and M. Okazaki, A magnetic shielding type superconducting fault current limiter using a Bi-2212 thick film cylinder. IEEE Trans. Appl. Supercond., 5: 10671070, 1995.
25. H. Kado and M. Ichikawa, Performance of a high-Tc superconducting fault current limiter. IEEE Trans. Appl. Supercond., 7:
993996, 1997.
LEMBIT SALASOO
General Electric Company
HEINRICH J. BOENIG
Los Alamos National Laboratory
717
674
SUPERCONDUCTING CYCLOTRONS AND

COMPACT SYNCHROTRON LIGHT SOURCES
Superconducting magnets are the enabling technical component of the highest-energy particle accelerators (atomsmashers) constructed and planned. These accelerators require both brute force and precision in the magnetic fields,
which guide the particle beam on a roughly circular orbit,
sometimes many kilometers in diameter, in order to return it

to a radio-frequency accelerating structure for repeated energy boosts. The tera-electron volt (TeV) energy regime would
not have been reached without a technology able to produce
magnetic fields in the 5 T to 10 T range, accurate to about 1
part in 10,000, and economical enough to build in kilometer
lengths. In this regime, superconducting magnets are the
clear technology of choice.
The ability of superconducting magnets to produce strong,
accurately shaped magnetic fields over large volumes also motivates their use in lower-energy accelerators, namely cyclotrons and electron synchrotrons, where normal magnets are
also an option. Cyclotrons, because of their continuous beams,
are usually the technology of choice for nuclear research in
the hundred mega-electron volts per nucleon range. Such cyclotrons involve large magnets, in the 1000 ton range for magnets of room temperature design. Cyclotrons are also widely
used in the commercial production of radionucleides and are
used to a small degree for cancer therapy. The weight-reducing, cost-saving advantage of superconducting magnets (to
the 100 ton range) has made them the design of choice in
most recent nuclear physics projects. Superconducting magnets have also been used in a few situations for isotope production and for cancer therapy but have thus far not been
widely adopted for either of these applications.
The short wavelength optical radiation emitted by electron
synchrotrons has in the past two decades become a centrally
important element in many research fields and in potentially
important industrial applications. The need for sources compatible with on-line use in production facilities has led to the
development of compact synchrotron light sources, whose
compactness is enabled by the strong magnetic fields available from superconducting magnets.
SUPERCONDUCTING CYCLOTRONS
The cyclotron was the earliest of the circular particle accelerators, earning a Nobel prize for E. O. Lawrence in 1936. First
the cyclotron and then the synchro-cyclotron provided the
beams for energy-frontier physics studies in the 1930s and
1940s. In the late 1950s, the intrinsic cost advantage of the
ring geometry of the synchrotron (versus the pancake geometry of the cyclotron) shifted energy-frontier studies to the
newly evolving synchrotron. However, because of a series of
significant technical innovations (sector and spiral-ridge focusing and superconducting magnets), the cyclotron has retained several important regions of superiority and has not
faded from the scene in the fashion of a number of other pioneering accelerator systems (e.g., the synchro-cyclotron and
the betatron).
Figure 1 shows the key features of the classic cyclotron,
the essence of which is quite eloquently summed up by a
childs description: they (the ions) start at the center and go
round and round and come out at the edge. The round and
round is the result of the transverse force exerted on the ions
by the magnetic field, and the spiral path is the result of acceleration provided by a radiofrequency voltage applied to the
dees. The cyclotron equation qB/m 2f gives the
orbital frequency f of a particle in a magnetic field B, where
q and m are the charge and mass of the accelerated ion. The
cyclotron equation reveals a gift of nature: taking q, B, and m
Dee
External beam
Vrf
Dee
675
qB
m
Ion
source
Figure 1. Conceptual drawing of the original cyclotron. The (constant) magnetic field is into the page. (Whereas the beam spirals outward in a cyclotron, in the synchrotron discussed later the beam stays
at a constant radius and the energy increases as the magnetic field
increases.)
as constant, the orbital frequency f does not depend on the

speed of the ion because the radius of the orbit increases just
in proportion to the speed, so that the time per rotation is
constant. Limits to the cyclotron process resulting from frequency errors from the relativistic increase in ion mass as the
speed approaches the speed of light as well as deviations from
the constant frequency (isochronous) magnetic field shape
in order to provide stabilizing (focusing) forces to keep ions
on the design trajectory were realized at an early time, and
modifications were introduced. First the radio frequency (RF)
was modulated to give the frequency modulated or synchrocyclotron. Later strong and weak azimuthal regions in the
magnetic field (hills and valleys or sectors) were introduced
to provide both focusing to hold the ions near the design path
and isochronism so that the ions remain in step with the accelerating voltage and can be accelerated to high energies limited only by the bending power of the magnetic field. The energy limit of the cyclotron thus evolved into a cost limit, and
a very natural further evolution of the cyclotron developed.
As superconductors lowered the cost of magnetic bending
power, the superconducting cyclotron pushed the cost limit to
higher energies.
The start-in-the-center-and-go-to-the-outside character of
the cyclotron leads, in fact, to an exceptionally large benefit
from the use of high magnetic fields. The field must be provided over the complete pole of area R2, where R is the maximum radius. The magnetic flux which determines the
cross-sectional area of the return path steel, is field area
(i.e., BR2. Expressing the flux in terms of the bending
power BR then gives
= BR2 = (BR)2 /B
(i.e., for fixed bending power, total magnetic flux varies as
1/B). The length of the magnetic return path also varies as
1/B, giving 1/B2 scaling for the total volume of steel. Superconducting cyclotrons are then lighter by a very large factor
(10 to 20) than room temperature cyclotrons with the
same maximum bending power.
Table 1 lists the present superconducting cyclotrons of the
world, including several now (1998) under construction. Accelerators of this type have assumed a centrally important
role in medium-energy nuclear physicsthe 5 T m bending
power of the K1200 cyclotron at the National Superconduct-
676
Table 1. Superconducting Cyclotrons

Magnet
Bending
Power
(T m)
First
Beam
Sector #
Nuclear research
3.2
8/82
Compact
Compact
Compact
Compact
Name
Location
K500
Nuclear research
Nuclear research
Nuclear research
3.3
3.2
4.9
9/85
6/88
6/88
4
3
3
K100
OSCAR
AGOR
K800
Triton
NSCL, Michigan State University,

East Lansing MI
Chalk River, Ontario Canada
Texas A&M, College Station TX
NSCL, Michigan State University,
East Lansing MI
Harper Hospital, Detroit MI
Oxford Instruments, Oxford, UK
Groningen, Netherlands
Carania, Italy
Munich, Germany
Cancer therapy
Isotope production
Nuclear research
Nuclear research
Prototype SOC
1.4
0.6
3.5
4.2
1.3
4/89
6/90
4/94
5/95
3
3
3
3
12
SRC
Riken, Japan
Nuclear research
6.4
TASSC
K500
K1200
Purpose
ing Cyclotron Laboratory at Michigan State University, for

example, makes it currently (1998) the worlds highest-energy
cyclotron, whereas its 300 ton weight is minor compared to
the 4000 tons of the largest room temperature cyclotron.
[Many beams from the K1200 reach energies in excess of 10
GeV/ion making it also the highest-energy continuous wave
(CW) accelerator of any type.] The large superconducting cyclotrons at Catania, Chalk River, Groningen, and Texas
A&M form, together with the K1200, a central core of overall
world research capability in medium-energy nuclear physics.
The 1.5 T m cancer therapy cyclotron at Detroits Harper
Hospital and the twelve 0.5 T m PET (positron emission tomography) cyclotrons manufactured by Oxford Instruments
for leading medical centers in many countries have also demonstrated the feasibility and effectiveness of the superconducting cyclotron in situations where sophisticated technical
support is much less available than in a major physics research center.
The coils used in most superconducting cyclotrons are cylindrically symmetric so that the very high hoop stress
caused by the high field density and high current density are
largely uniform with azimuth except for variations caused by
the sector structure of the iron. Earlier superconducting cyclotrons used cryogenically stable helium-bath-cooled coils (i.e.,
coils with liquid helium in direct contact with each strand of
conductor and with sufficient copper in the conductor so that
if a section of conductor ceases to superconduct, the resistive
power loss in the copper is nevertheless low enough for the
conductor to recool as a result of the presence of the liquid
helium rather than go into a runaway thermal excursion).
Two of the more recent superconducting cyclotrons, at Harper
Hospital and at Groningen, in contrast, use an intrinsicallystable, epoxy-potted winding where cooling is supplied only
at the external surfaces of the coil. In this type of coil, it is
very important to have a tightly packed, fully impregnated
winding because sudden wire motion at the scale of 20 m to
40 m will release heat sufficient to drive the conductor
normal.
The last two cyclotrons in the Table 1 list (Munich and
Riken) are under construction; both attempt to carry superconducting technology an additional step by introducing noncircular coils, which preferentially strengthen the hills in the
Type
Compact
Compact
Compact
Compact
Separated
sector
Separated
sector
magnet compared to the valleys. Such a coil configuration is

necessary if the cyclotron is to operate at particle velocities
v/c 0.6, where axial focusing becomes a critical limiting
parameter and open, low field regions are necessary between
the hills (separated sectors to obtain sufficient axial focusing. The cyclotron at Riken is to produce beams up to 400
MeV/nucleon (v/c 0.7); therefore, the technology of superconducting sector magnets is a central element of the project.
The superconducting cyclotron project at Munich also involves the feature of noncircular coils, but the dominating
novel characteristic of this cyclotron is its superconducting RF
system (in addition to superconducting sector magnets). Superconducting RF is not only very attractive conceptually because the RF system is normally the dominant power dissipation element in a cyclotron but also very difficult because
superconducting RF currents will quench in the presence of
dc magnetic fields at the level of about 104 T. It is very difficult to achieve an almost field-free region in close proximity
to magnets of the field strength needed in cyclotrons. The Munich project thus uses an unusual cyclotron structure known
as a separated orbit cyclotron, which has been discussed in
the literature for many years but previously never built. In
such a cyclotron, sufficient RF voltage is provided to achieve
some stipulated separation between the last and next-to-last
orbits so that extraction of the beam is facilitated. In the Munich example, superconducting cavities provide sufficient energy gain per turn to give a 4 cm separation between final
turns, and the cavity system has performed impressively. The
magnet in such a cyclotron is, however, quite complicated,
with a series of N separate magnets on every turn, where N
is the sector number. This gives 240 magnets for the case of
the 12 sector, 20 turn Munich prototype. The problem of appropriately adjusting the current in all these magnets without inducing quenches caused by misaligned beam, has lead
to a prolonged, presently incomplete period of debugging of
the Munich system.
APPLICATIONS AND CHARACTERISTICS
OF SYNCHROTRON RADIATION
A relativistic charged particle beam deflected by a magnetic
field emits electromagnetic radiation known as synchrotron
radiation. The radiation from a light electron is more intense

than from a heavier proton. Therefore practical synchrotron
light sources are electron (or positron) storage rings. These
machines are the brightest continuous sources of vacuum ultraviolet and X-ray radiation available.
The total power P radiated by an electron beam of energy
E and current I when deflected in a magnetic field B is, in
practical units,
P[kW] = 26.6E 3[GeV]B[T]I[A]
A most useful attribute of synchrotron radiation is its high
degree of natural collimation. The power is emitted within a
cone of width 1/ around the direction of the electron beam,
where is the energy of the electron divided by its rest mass
0.511 MeV. Hence for a typical source with energy 1 GeV, the
synchrotron radiation is confined to a fan in the plane of the
electron orbit, diverging with an opening angle of less than 1
mr. Practical devices that store 1 GeV electron beams of
100 mA and deflect them in magnetic fields of 1 T therefore produce tens of kilowatts of collimated radiation, emanating from a line source that can be less than a millimeter in
diameter. Large facilities for the production and use of synchrotron radiation have been constructed at research centers
in most of the industrialized countries of the world. Compact
machines allow use of this radiation to be extended to industrial applications.
Different applications of synchrotron radiation require flux
from different parts of the electromagnetic spectrum. The
spectrum emitted by a synchrotron source is characterized by
the critical energy c. Half of the power radiated from the
source is within a broad spectrum below the critical energy,
and half is within an exponentially falling spectrum above
c. The critical energy is related to the electron energy and
deflecting field by
c [keV] = 0.665E 2[GeV]B[T]
Thus, through proper selection of the electron beam energy
and deflecting field strength, the output of a synchrotron
source can be centered in the spectral range of interest.
The bend radius of an electron in a magnetic field is
given by
[m] = E[GeV]/(0.3B[T])
Combining these expressions leads to an expression for the
bend radius in terms of the critical energy and magnetic field
[m] = 4.09c1/2[keV]/B3/2 [T]
This formula makes explicit the utility of choosing a magnet
technology that can produce a high value of B in order to produce a compact synchrotron light source with output of kilovolt X rays. Good-quality normal conducting electromagnets
constructed from copper windings and iron pole pieces are
limited to about 1.5 T by the saturation properties of the iron.
Superconducting magnets using niobiumtitanium alloy superconductors operated at liquid helium temperature can produce accurate fields several times larger, and reduce the radius of the synchrotron source by an order of magnitude.
Among the industrial and medical applications of synchrotron radiation for which dedicated machines have been pro-
677
posed are coronary angiography, protein crystallography, and

X-ray lithography. Coronary angiography with synchrotron
radiation uses an abrupt change in the absorption of X rays
by iodine at 33.17 keV. By injecting an iodine-rich contrast
agent, exposing the subject to two beams that straddle this
energy, and subtracting the images generated, vascular structure can be clearly visualized. The diffraction of X rays has
long been used to determine crystal structure. The superior
brightness of synchrotron sources compared to alternative Xray sources have made crystallographers major users of large
synchrotron research centers. Despite continuing interest, as
of this writing no dedicated compact synchrotron has been
constructed for either angiography or crystallography. However, several have been made to enable the development of Xray lithography.
X-ray lithography is a candidate to become the successor
to optical lithography for the patterning of semiconductors in
the highest-density integrated circuits. For 30 years, advances in optical lithographys ability to print fine features
has led to geometrically increasing circuit densities and chip
performance, a phenomenon characterized by Moores Law,
which observes that the number of transistors that can be put
on a chip has been doubling about every 18 months. However,
as the size of circuit features is reduced to values comparable
to the wavelength of the ultraviolet radiation used to print
the pattern (i.e., to 248 nm and below), diffraction effects
limit the resolution, and the challenge of maintaining this
trend becomes more difficult. X rays with energy near 1 kV,
wavelength near 1 nm, diffract much less, and have been used
to fabricate transistors with dimensions of 100 nm and below.
Lithography of complex, high-density circuits using X rays
has progressed to the pilot line stage, using compact synchrotron X-ray sources developed for this purpose. A related technology using higher-energy X rays to fabricate micromachine
parts also has much promise.
TECHNICAL DESCRIPTION OF A COMPACT SYNCHROTRON
Compact synchrotrons have the same major subsystems and
are designed according to the same principles as larger synchrotrons, although innovation has occurred in the design of
the superconducting magnets for compact light sources. Figure 2 shows a photograph of the Helios 1 compact synchrotron
manufactured by Oxford Instruments. Other machines include Aurora by Sumitomo Heavy Industies, COSY by COSY
Microtech, MELCO by Mitsubishi, and SuperALIS by NTT.
The major subsystems on all these machines are
an RF accelerating structure,
the magnet system,
an injector and pulsed injection magnets on the ring,
a vacuum system,
beam diagnostics and controls, and
ancillary power supplies and utilities.
The accelerating structure is a resonant radiofrequency cavity

producing a longitudinal electric field imparting energy to the
beam particles with each revolution. The energy gain serves
both to raise the average beam energy during the acceleration
cycle and to compensate for energy losses to synchrotron radiation. Synchrotron operation is based on the principle of
678
SUPERCONDUCTING ELECTROMAGNETS
Figure 2. The first Helios compact synchrotron, built by Oxford Instruments. The two superconducting magnets can be seen at either
end of the structure. X-ray beams emerge from the ports on the sides
of the magnets.
phase stability. Particles with small deviations from the central beam energy follow slightly different orbits and arrive at
the cavity at slightly different times. They automatically receive more or less energy according to their phase. The result
is effectively a restoring force causing the beam to coalesce
into bunches, with individual particles executing longitudinal
synchrotron oscillations about the center of the bunch. Phase
stability puts a constraint on the circumference of the orbit,
which must be an integer multiple of the radiofrequency
wavelength. The choice of radiofrequency is largely dictated
by convenient RF sources. Klystron-based sources at 500 MHz
have been used, as have solid-state sources at 50 MHz.
The most common compact light source layout has been a
racetrack using a pair of superconducting magnets each producing a vertical magnetic field to deflect the beam by 180.
This configuration allows other machine functions (e.g., the
RF structure, optical elements, pulsed magnets for injection,
and vacuum components) to be distributed in the two straight
sections. It still takes advantage of the small bend radius that
can be achieved with a superconducting magnet to produce a
device that can be factory assembled and delivered to a user
as a complete unit. The magnetic field is increased during the
acceleration cycle because the beam maintains a constant orbit in a synchrotron, unlike the outward spiraling orbit of the
beam in a cyclotron. In the case of the Helios system, the
superconducting magnets produce a field of 4.5 T at full energy, which allows a 700 MeV beam to be bent into a semicircle of radius 51 cm. The entire device, exclusive of the injector
and ancillary systems, weighs about 20 tons and comfortably
fits on the back of a truck.
In order to allow a beam particle with small deviations
from the ideal orbit to survive for perhaps 1012 revolutions
without being lost, some form of focusing is required. All compact synchrotrons built save one have been strong focusing,
alternating gradient machines. (The Aurora 1 compact light
source built by Sumitomo Heavy Industries was a weak focusing machine.) Focusing in one of the transverse planes can
be accomplished by a magnetic field gradient. However, if a
magnetic field gradient focuses in the horizontal plane, it inevitably defocuses in the vertical plane, and vice versa. The
alternating gradient concept uses the principle that pairs of
lenses acting together, one focusing and one defocusing, have

a net focusing effect. The gradients necessary for producing
stable orbits in a compact synchrotron can be produced by
independent quadrupole lenses, as in a separated function
machine, or by gradients built into the deflecting field, as in
a combined function machine. The Helios source shown uses
both. Precise placement of the superconducting wires is necessary to achieve the field quality required.
Most compact light sources have been designed with lowenergy injectors. The purpose of the injector is to deliver to
the ring an electron beam that is already relativistic and that
is high enough in energy to limit the dynamic range required
of the synchrotrons magnet system. Because the injection
process is inefficient and produces radiation from particles
that are not captured by the synchrotron, the energy of injection also determines the thickness of shielding required. Both
linear accelerators and racetrack microtrons are in use as injectors for compact light sources. Rings designed for operation
at 700 MeV to 1 GeV have successfully used injection energies
from 50 MeV to 200 MeV. Pulsed magnets within the ring are
required to deflect the incoming beam from the injector onto
a stable orbit.
Ultra-high vacuum technology is required in a compact
synchrotron because the lifetime of the beam can otherwise
be limited by losses from collisions between the electrons and
residual gas atoms, especially those liberated by collisions between synchrotron radiation and the vacuum chamber walls.
The presence of a cryogenic system for the superconducting
magnets makes cryopumping a natural choice. The Helios
source has achieved beam lifetimes in excess of 50 h with 200
mA of circulating electrons. Because the injection and acceleration cycle require only about 10 min, a high utilization factor
is possible.
BIBLIOGRAPHY
J. Cornell, Cyclotrons and Their Applications, Singapore: World Scientific, 1996.
F. M. Russel, A fixed frequency, fixed field, high energy accelerator,
Nuclear Instrum. Methods, 23: 229, 1963.
E. Weihreter, Compact Synchrotron Light Sources, Singapore: World
Scientific, 1996.
J. M. Warlaumont et al., IBM J. Res. Dev., 37 (3): 288448, 1993.
DAVID E. ANDREWS
Oxford Instruments, Inc.
HENRY BLOSSER
Michigan State University
SUPERCONDUCTING DEVICE RELIABILITY. See

SUPERCONDUCTING MAGNETS, QUENCH PROTECTION.
717
Ground plane
Patterned HTS
Substrate
Microstrip
Stripline
Coplanar
waveguide
Figure 1. HTS technology lends itself to the fabrication of microwave

planar devices such as those schematically depicted here in crosssection. One advantage of coplanar waveguide (and its variant slotline, not shown) is that only one patterned HTS film is needed. One
disadvantage of stripline is that it is difficult to package. HTS devices
have been demonstrated in all three structures.
superior to conventional planar technology and with the attractive feature that many well-established design techniques
can be used for HTS circuits as well.
The discussion in this article will focus on HTS microwave
technology with the understanding that conventional lowtemperature superconductors (LTS), for example, Nb or NbN,
can also be used in the same fashion. Practical LTS materials
operate typically at 4.2 K, the boiling temperature of liquid
He. Furthermore, it should be kept in mind that a larger variety of substrates can be used in LTS technology because it
does not require single-crystal epitaxial films. For example,
Nb microwave and digital circuits have been demonstrated on
Si and single-crystal sapphire (1).
HIGH-TEMPERATURE SUPERCONDUCTING
FILM PROCESSING TECHNOLOGY
SUPERCONDUCTING FILTERS AND

PASSIVE COMPONENTS
The advent of high-temperature superconducting (HTS) materials has enabled a number of applications in passive microwave electronics. Superconductors exhibit very low losses at
microwave frequencies and, although finite, at the practical
operating temperature of 77 K (boiling point of nitrogen),
these losses are more than two orders of magnitude lower
than normal conductors at frequencies of 10 GHz and below.
This has allowed the possibility for high-performance planar
microwave components since high-quality epitaxial films can
be deposited on both sides of low microwave loss, single-crystal substrates allowing the fabrication of components in planar configurations such as microstrip, stripline, and coplanar
waveguide. These configurations are widely used throughout
the microwave community in a variety of technologies ranging
from GaAs microwave monolithic integrated circuits (MMIC)
to integrated circuits using ceramic and laminated substrates. Figure 1 shows a schematic cross-section of all three
most common planar microwave structures. The use of HTS
in these circuit configurations results in passive devices such
as resonators, filters, and delay lines with performance far
The basic elements of HTS thin-film technology from the

point of view of microwave applications will be discussed here
briefly. Only epitaxial thin films will be addressed because,
to date, significantly better properties can be obtained from
epitaxial material than from polycrystalline, nonepitaxial
techniques. Most, if not all, of the developments covered in
this article, however, are valid for the case of polycrystalline
materials. The main advantage of these materials is cost and
coating of nonplanar surfaces like the inside of a cylinder to
form a high-Q cavity resonator (2).
Among the various high-temperature superconductors
there are two that have achieved a level of maturity and acceptance in the industry: (1) YBa2Cu3O7, usually referred to
as YBCO, and (2) Tl2Ba2CaCu2O8 or TBCCO. High-quality epitaxial thin films of these materials can be deposited on both
sides of a variety of low microwave loss, single-crystal substrates by physical vapor deposition techniques, mainly sputtering (3) and laser ablation (4) and also by metal-organic
chemical vapor deposition (MOCVD) (5) and coevaporation
(6). The two most common substrates to date are LaAlO3
(LAO) and MgO. Circuits on sapphire have been demonstrated but the crystal lattice mismatch between HTS and
sapphire is large enough to restrict the film thickness to below desirable values of at least 500 nm. Both LAO and MgO
718

Table 1. Basic Characteristics of the Most Used HTS Thin Films and Dielectric Substrates for
Microwave Applications
HTS Films
Critical Temperature (T c )
Surface resistance (R s ) at 77 K and
10 GHz (f 2 dependence)
Film thickness
Critical dc current density (J o ) at
77 K
Penetration depth (nm)
YBCO
90 K
0.20.5 m
105 K
0.20.5 m
400600 nm
106 A/cm 2
8001000 nm
10 6 A/cm 2
200 nm
200 nm
Substrates
Relative dielectric constant ( r ) at
77 K
Dissipation factor (tan ) at 77 K
Typical dimensions
TBCCO
LaAlO 3 (LAO)
MgO
23.4
9.7
10 5
5, 7.5, and 10 cm diameter;
250 and 500 m thick
10 5
5 and 7.5 cm diameter;
250 and 500 m
thick
Both types of HTS Films can be grown in either substrate.
substrates are available commercially from a variety of suppliers around the world and can be readily obtained in circular wafers up to 7.6 cm in diameter and 250 m to 500 m in
thickness. At 77 K and for frequencies between 1 GHz and 10
GHz the loss factor (tan ) of LAO and MgO is less than
105. This is 10 to 100 times smaller than most practical microwave substrates and is compatible with the low conductor
loss of HTS.
YBCO is typically grown in situ as an epitaxial single-crystal film with thickness ranging from 400 nm to 600 nm. This
means that the desired crystalline phase of the material is
formed as the film grows because the growth conditions can
be adjusted to obtain such results. In contrast, an amorphous
film of TBCCO is deposited first and then the film is subjected
to a postdeposition annealing treatment to form the right
crystalline phase. One advantage of the in situ deposition of
YBCO films is the ability to deposit other crystalline layers
such as insulators or normal conductors, needed for the fabrication of Josephson junctions. HTS microwave technology can
leverage the research on Josephson junctions by making use
of insulating layers in other ways such as lumped element
capacitors, for example. TBCCO films have the advantage of
a higher critical temperature (Table 1) and are typically
grown with thickness close to 1 m. These are advantages
because practical devices must be made with thicknesses two
or three times greater than the London penetration depth
(200 nm to 300 nm) at 77 K. TBCCO films thus offer a greater
operating margin than YBCO films.
The most important characteristics or these two HTS materials and their substrates are given in Table 1. The properties listed are those a microwave designer would want to
know if engaged in an HTS device design project.
Device Processing
Fabrication of microwave devices using either YBCO and
TBCCO follows relatively straightforward photolithographic
techniques. Patterning of the superconducting layer is typically accomplished by Ar-ion milling. The processing must include the deposition and patterning of low-resistivity contacts
for interfacing with other devices or instrumentation. These
are typically made by depositing a thin (200 nm to 300 nm)

layer of gold or silver followed by an annealing step at 500C
to 600C (7). Interfacing with other devices via coaxial connectors or directly to other substrates, superconducting or otherwise, can be accomplished using gold wire or ribbon attached
to the low-resistivity contacts by ultrasonic thermal compression bonding or gap-welding (ribbons). Fabrication details of
filters and delay lines made at Northrop Grumman can be
found in (8). Other institutions follow similar procedures.
Microwave Packaging of High-Temperature
Superconducting Devices
Proper packaging of cryogenic microwave devices is critical to
the success of the technology. In general, planar microwave
devices are made up of a dielectric or semiinsulating substrate, typically mounted on a metallic package or carrier by
means of conducting epoxy or soldering. The mounted device
then interfaces with other devices or measurement instrumentation most commonly via coaxial connections. In this
case, the metallic carrier forms part of the ground terminal
by connecting to both the outer jacket of the coaxial interconnection and the ground plane defined on the substrate. This
means that microwave currents flow into the planar device
through the center conductor of the coaxial interface and back
out through the ground plane on the substrate, the metallic
carrier, and its connection to the coaxial connector outer
jacket. This current flow into the packaging assembly must
occur without appreciable impedance mismatch or discontinuity. Figure 2 illustrates these points schematically with an
example for a microstrip device.
The criticality of these connections between substrate, carrier, and connectors is exacerbated when cryogenic cooling of
the microwave package is required because a large thermal
mismatch between the various components may cause cracking of the substrate or degradation of the interface between
substrate and carrier or connector. Furthermore, the quality
of the interface between the cryoelectronic substrate and the
outside world must be preserved through many temperature
cycles between ambient and cryogenic temperature to allow
for repeated testing of the device and an operational environ-

Package
wall
10
HTS
pattern
Coaxial
connector
Carrier
Substrate-carrier
conducting interface
(e.g. conducting
epoxy or solder)
Conducting interface
(e.g. metallic springs or
single-piece housing)
Figure 2. Microwave packaging of HTS devices is challenging because mechanical and electrical integrity must be maintained when
the device is cycled from ambient to cryogenic temperatures. Special
attention must be paid to the ground-current return path so as not
to introduce parasitic reactances that could severely affect the performance.
Thermal expansion ( m/cm)
ment, which may require that the device warm up to room

temperature while not in use. Suitable microwave package
designs must therefore include one or more of the following
elements: (1) the use of thermally matched materials, (2) adhesives that remain sufficiently pliable at cryogenic temperatures, and (3) configurations that allow the various parts of
the package to contract and expand freely while maintaining
good electrical contact.
When considering materials that are thermally matched to
HTS substrates the important parameter is the total contraction between room temperature and the operating temperature, for example, 77 K, and not the thermal coefficient of
expansion. Figure 3 shows the relative thermal contraction
for several materials, including LAO and MgO, in m/cm (i.e.,
at a given temperature, a 1 cm-long piece of material shrinks
so many m from its length at 300 K). As can be seen, the
slope of this curve, that is, the thermal coefficient of expansion, varies greatly with temperature over the range of interest, making it a practically useless parameter for the selection
of appropriate materials. Although there are some differences
in the rate of contraction as the materials shrink from room
0
Si
2
4
6
8
GaAs
10
12
LaAIO3
14
16
50
MgO
100 150
200
Temperature (K)
250
300
Figure 3. Measured relative thermal contraction of several materials of interest in HTS technology, including Si and GaAs. These or
similar data should be used in the selection of carrier materials and
the packaging of hybrid HTS-semiconductor components. If a good
thermal match between parts that must remain in intimate mechanical and electrical contact cannot be obtained, sufficiently pliable conductive adhesives may be adequate for some applications, particularly if the mismatched parts are small.
Cu/laminate
microstrip
Au/LAO
microstrip
at 77 K
Ground
plane
Attenuation (dB/m)
HTS
substrate
719
0.141 Semirigid
coaxial
HTS/LAO
microstrip
0.1
X-band
waveguide
0.01
0
10
15
20
Frequency (GHz)
Figure 4. Calculated attenuation comparison for various transmission line types including HTS and gold microstrip on 500 m-thick
LAO substrates. The line impedance for each line was 50 . All microstrip substrates were assumed to be 500 m thick. Notice that Xband waveguide has lower loss than HTS microstrip (for the parameters chosen). However, HTS microstrip has broader bandwidth and
the potential for smaller volume because it facilitates the integration
of several microwave components.
temperature, the key parameter is the total contraction at 77

K. For example, Nb and LAO are fairly well matched at 77 K,
even though their rate of contraction as a function of temperature is slightly different. This is borne out by extensive experimentation (8). Other substrate/carrier material pairs have
been successfully used as well.
PASSIVE SUPERCONDUCTING MICROWAVE
DEVICE FUNDAMENTALS
The key reason for developing a microwave HTS technology
is the exploitation of the low loss afforded by HTS compared
with conventional metals like gold and copper. High-performance, low-loss devices using conventional materials can generally be made at the expense of high volume, usually in the
form of hollow or partially dielectric-filled waveguide components. The potential of HTS is to enable components with the
same or better performance in a much reduced volume which
must include the cryocooler. Figure 4 shows a comparison between calculated losses in several common types of transmission line, including HTS microstrip on LAO, for the parameters listed in Table 1. For calibration, included in this
comparison is Au microstrip at 77 K, also on LAO. For the
microstrip cases only the conductor and dielectric losses were
calculated; radiation losses or coupling to spurious surface
modes were ignored. Also included are the losses in X-Band
waveguide, which are lower than HTS for the parameters chosen, although HTS microstrip offers the advantage of much
wider bandwidth and ready integration of several components
into a small volume. Notice that the useful frequency range
for X-band waveguide is typically only 8 GHz to 12 GHz. Integration capability combined with low loss are key elements of
HTS technology because most systems insertion opportunities
will arise for applications offering significant size advantages
with respect to conventional approaches.
720
A different and also useful way of getting insight into the

advantages of HTS planar circuits is from the point of view of
resonant structures, which form the basic building block of
passband filters. This can best be discussed in terms of quality factor, Q, which is the ratio between stored and dissipated
energy:
Q=
Stored energy
Dissipated energy
h
d
106
For empty electromagnetic cavity-type resonators, this is, in

essence, a figure of merit measuring the degree of compromise
between volume (stored energy) and surface area (microwave
losses on the conducting surfaces). In general, both stored and
dissipated energy depend on the dielectric constant and the
geometric configuration used, the dissipated energy depending also on the surface resistance of the (super)conducting surfaces and the losses in the dielectric. In practice, there
could be other types of losses such as radiation, which, for
simplicity, will be neglected in this discussion. In principle,
the performance of any passive device can be projected from
the Q of the type of structure used to make up the device. For
example, a resonator made up of a section of microstrip line
can be calculated from well-known expressions (9). The insertion loss of a filter can, in turn, be estimated from the Q of
the resonators that make up the filter (10).
Figure 5 is a plot comparing the Q and volume, as a function of frequency, of resonators made up of a microstrip line
section and an empty metallic cube, respectively. Since the
dimensions of the resonator are specified at each frequency,
the volume calculated is that of the smallest cube capable of
resonating at a given frequency. The microstrip HTS resonator volume was calculated assuming it is in an enclosure with
cross-section as shown in the figure, where the walls and the
lid are sufficiently far away from the superconducting strip
that their contribution to the loss is negligible. The Q of HTS
microstrip, although higher, is within the same order of magnitude as the Q of the cube resonator. However, the estimated
volume can be two or more orders of magnitude smaller, especially at the lower microwave frequencies.
Although to first order the main feature distinguishing
HTS from conventional planar microwave passive devices is
low loss, other important differences exist and are discussed
in the following sections.
105
Quality factor
(1)
104
HTS/LAO microstrip
Metallic cube
103
102
0.1
Au/LAO microstrip
10
100
104
Volume (cm3)
103
Metallic cube
102
10
1
101
0.1
Microstrip on LAO
1
10
Frequency (GHz)
100
Figure 5. Calculated comparison of quality factor (Q) and volume for

two types of resonators: 50 HTS microstrip on 500 m-thick LAO
and a metallic cube. The Q of a gold-on-LAO microstrip at 77 K was
also calculated as a reference. The metallic cube is representative of
a simple cavity resonator and was chosen because its Q is easy to
calculate (12). The volume for the microstrip was chosen as a device
with the cross-section shown in the figure with a/h 10, d/h 20,
and a length of ( /2 4h). This assumed cross section is independent
of wavelength and loses its meaning at the higher frequencies plotted,
where the substrate would, in practice, be thinner, making the microstrip always smaller than the waveguide.
Surface Impedance and Penetration Depth

The surface or internal impedance of a conductor is the characteristic impedance seen by a plane wave incident perpendicularly upon a planar (super)conducting surface. For both normal (e.g., copper, gold) conductors and superconductors, the
surface impedance per unit length and width is given by
(11,12):
ZS = RS + jXS =
r j
(2)
where is the angular frequency, is the permeability, and

is the conductivity. is real for normal conductors but is
complex for superconductors. In both cases the RF fields decay exponentially inside the material, defining a field penetration depth. In the case of superconductors, however, this
parameter is independent of frequency and is orders of magnitude smaller than the normal conductor penetration depth
(usually referred to as the skin depth). The reasons are derived from the perfect diamagnetism of superconductors, the
so-called Meissner effect, and are explained by the GorterCasimir and London two-fluid model of superconductivity
(11,13).
Table 2 summarizes the differences between normal and
superconductors from the point of view of their microwave
surface impedance. Notice that the surface resistance of superconductors, Rs, has a frequency-squared ( f2) dependence.
In contrast, normal conductors depend on the square root of
frequency (f). Figure 6 shows the difference between copper
at 300 K and 77 K, and HTS at 77 K. This must be taken into
consideration, especially when designing wide-band compo-
Table 2. Microwave Surface Impedance Comparison

Between Normal Conductors and Superconductors
Z S R S jX S
Normal Conductors
Superconductors
Real
Complex
Penetration (skin) depth:

1
Penetration depth:
L (0)
L (T)
T 2
1
TC
L (0) 150 nm for YBCO
RS XS
1
f
RS f 2
X S 2f L
Copper at 300 K:
2.1
m
f
R S 8.24 f m
critical temperature. For YBCO, L(0) 150 nm, which results in L(77 K) 214 nm. The HTS film must be at least
500 nm to 600 nm thick for operation at 77 K, in order for the
kinetic inductance effects to be negligible with respect to the
total inductance of the circuit.
For practical microwave design purposes, this allows treating the superconductor as a normal conductor with a surface
resistance that can be obtained from measured values and a
frequency-squared scale factor. It has become customary for
workers in the field to normalize the surface resistance to 10
GHz and 77 K, even though measured data may have been
taken at a different frequency. Devices where the kinetic inductance is allowed to dominate have been demonstrated (14).
However, they are lossy, difficult to fabricate, and quite dependent on temperature because of the strong temperature
dependence of the penetration depth.
YBCO at 77 K:
L (77) 0.2 m
LaAIO3 (LAO) Substrate Properties
R S 5f 2
The relative dielectric constant of LAO, one of the preferred

HTS substrates, is r 23.4 at 77 K. This is a higher value
than most common microwave substrates whose dielectric
constants usually do not exceed 10 (r 9.7 for MgO, another
preferred HTS microwave substrate). The significance of this
is that common planar component design techniques are
based on empirical circuit models whose validity may not extend to the relatively high dielectric constant of LAO. For example, the design of microstrip parallel-coupled-line filters involves the use of quarter-wave coupled-section models that
generally are not valid for r 18. This requires that more
sophisticated design techniques be developed, making use of
electromagnetic field solvers, for example, or empirically extending the range of existing models.
LAO has a cubic crystal structure above about 450C. Below that temperature it transitions to a rhombohedral structure, which is only a slight distortion from the cubic structure
consisting of a very small stretching of the cubic unit cell
along its diagonal. In order to release stress, the crystal will
form twin structures, symmetrically related regions oriented
in different directions. Noncubic crystals are anisotropic and,
as a result of this twinning, the LAO substrate is made up of
slightly anisotropic regions randomly distributed throughout
the substrate (11). The net result is an average dielectric constant with a uniformity of approximately 1%. This means a
2% variation in the resonant frequency of a planar resonator
and there are many filter applications, for example, where
this is not acceptable. In contrast, MgO, which is cubic, has a
uniformity of 0.1% (15).
Copper at 77 K:
0.9
m
f
R S 3.4 f m
nents. The figure also highlights the large difference between

copper and HTS at frequencies below 1 GHz.
As with planar microwave devices using normal conductors, best performance control is obtained when the geometric
inductance of the circuit dominates the kinetic (or internal)
inductance of the superconductor. That is, from a practical
point of view, the thickness of the superconductor must be at
least two to three times larger than the penetration depth at
the temperature of operation. The penetration depth is a
strong function of temperature and for HTS it is given approximately by (11,13).
L (T ) =
L (0)
T 4
(3)
TC
where L(0), the penetration depth at 0 K, is a fundamental

parameter of the material, T is the temperature and Tc is the
Surface resistance ()
721
102
1
Dynamic Range Considerations: Noise Figure

Cu (300 K)
102
104
Cu (77 K)
HTS (77 K)
106
108
0.1
10
100
Frequency (GHz)
Figure 6. Surface resistance of HTS at 77 K and copper at 77 K and
300 K as a function of frequency. The surface resistance of copper
scales with frequency as f 1/2; for HTS it scales as f 2.
An important consideration for any electronic device is its dynamic range, or range of signal power levels over which the
device will operate properly. In the case of passive HTS devices, they are expected to be linear over a certain dynamic
range, limited below by noise and above by the onset of nonlinear behavior.
Starting at the lower end, the noise generated in a passive
device will generally be of a thermal nature. A measure of
how much noise any device generates is given by the noise
figure (16), which is, by definition, related to the excess noise
generated in the device when a matched resistor at 290 K
(ambient temperature) is placed at the input. Thus the noise
722
figure would be equal to 1 (or, equivalently, 0 dB) if the device

were perfectly noiseless or if it were an ideally lossless passive device. The accepted noise figure definition as a function
of device temperature is (16).

T
(4)
FdB = 10 log 1 + (L 1)
290
Here, L is the insertion loss of the device as a number greater
than or equal to 1 (i.e., 10log(L) 0 dB) and T is the temperature in degrees Kelvin. For a passive, lossy device at 290 K,
the noise figure turns out to be equal to its insertion loss, a
rule that system designers commonly use when dealing with
passive components such as filters or lengths of transmission
line. HTS devices, however, because they operate at cryogenic
temperatures (77 K, typically), will have a lower noise figure,
according to the accepted definition (16). Figure 7 shows this
expression graphically as a function of the insertion loss of
the device for 77 K and 290 K (ambient temperature). Thus,
in considering the dynamic range of HTS devices, the lower
end of the range will tend to be lower than for conventional
devices, not only because of their inherent low loss, but also
because they operate at cryogenic temperatures. Measurements reported in the literature (17) confirm, to first order at
least, that the noise in HTS passive devices is indeed thermal
in nature.
Dynamic Range Considerations:
Nonlinearity and Power Handling
The dynamic range of HTS passive devices is limited above
by nonlinearities in the superconductor. This is in contrast to
conventional technology, for which this upper limit could be
orders of magnitude higher, generally limited by such phenomena as the voltage breakdown of air or the dielectric used,
or melting of the metallic pattern due to high currents. HTS,
on the other hand, is fundamentally limited by the critical
magnetic field, above which the material loses its superconducting properties. Nonlinear behavior occurs as the RF magnetic field approaches its critical value. Furthermore, in practical devices the microwave fields will generally tend to be
nonuniformly distributed so that critical values are exceeded
first in selected areas of the device. For example, a microstrip
line has much higher current density near the edges of the
line than along the center. Thus, as the signal power level is
increased, the current density at the line edges will approach
Noise figure (dB)
2
1.5
T = 290 K
1
T = 77 K
0.5
0
0.5
1.5
Insertion loss (dB)

Figure 7. Noise figure of a cryogenically cooled passive device as a
function of insertion loss. Notice that the noise figure equals the loss
at 290 K (by definition), but it is lower for devices operating at a
lower temperature.
critical state first, generating nonlinearities and increased

losses, degrading the performance of the device.
When a device is nonlinear it produces intermodulation
distortion; that is, two signals of different frequencies applied
to the device will generate mixing products. In general, the
largest mixing products are those of the third order. The upper end of the dynamic range is then reached when the power
level of the applied signals is such that the third-order products rise above the noise floor and can be mistaken and processed by the system as real signals. The nonlinear behavior
in a device is characterized by the third-order intercept point
(TOI) (18).
A system dynamic range can, in turn, be determined from
that of its components. An important example is that of a
microwave receiver front-end, usually consisting of a lownoise amplifier (LNA) placed after the antenna, which is then
followed by one or more downconversion stages. The dynamic
range of the receiver is greatly determined by the noise figure,
gain, and TOI of the LNA, with the components that follow
having much less influence. Many applications demand a preselector filter between the antenna and the LNA to reject
strong interfering signals that could generate unwanted mixing products due to the nonlinearity of the LNA (17). The preselector filter must not significantly degrade the receiver dynamic range and so it must have low insertion loss (i.e., low
noise figure) and a TOI sufficiently higher than the LNAs
with respect to both in-band signals and the rejected out-ofband interfering signals. This is an important example because HTS filter technology is a strong candidate for this type
of preselector in some applications like wireless communications base-station receivers (19).
The linearity condition, which may be a relatively minor
issue with conventional filter technology is, in contrast, very
key in HTS filter technology. Specific developments in this
area will be discussed in the section on HTS Filter Technology.
A fundamental characterization of the nonlinear behavior
and power handling in HTS materials is through the surface
impedance and its dependence on the RF magnetic field HRF
(2025).
Zs (HRF ) = Rs (HRF ) + jXS (HRF )
(5)
The essence of the nonlinear dependence of the surface impedance on signal power level or, equivalently, HRF is readily
understood by observing the response of a microstrip resonator, shown in Figure 8. As the input power is increased the
resonator Q degrades (Rs dependence on HRF) and the resonance shifts to lower frequencies (Xs dependence on HRF). Several regimes have been identified in the study of nonlinear
phenomena in HTS (22). A linear region at sufficiently low
power levels, a weakly nonlinear region where nonlinear behavior is dominated by grain-boundary weak links (Josephson-junction-like defects in the crystalline make-up of the
HTS film) and a strongly nonlinear region dominated by hysteretic vortex penetration. Above this regime breakdown of
the superconducting state occurs, with the surface resistance
increasing abruptly due to heating and the formation of normal-state domains (22).
If the magnetic field exceeds its critical value the material
becomes a normal conductor and dissipates heat which must
be removed by the cryocooling system and can even damage

25
Transmitted power (dBm)
30
T12Ca2Ba2Cu3Ox resonator
T = 7.4 K
15 dBm
straightforward case of two in-band signals undergoing the

same group delay.
HIGH-TEMPERATURE SUPERCONDUCTING FILTERS
35
40
45
50
55
10 dBm
60
1,835
723
1,840
1,845
1,850
1,855
1,860
1,865
1,870
Frequency (GHz)
Figure 8. Effect of increasing the input power level on a superconducting microstrip resonator. This measurement (courtesy of Dr. M.
Golosovsky, Hebrew University of Jerusalem) captures the essence of
the nonlinear RF power dependence of the surface impedance Zs
Rs jXs. As power level increases so does Rs and the resonance Q
decreases. On the other hand, the effect of increased power level on
Xs manifests itself on a shift of the resonance toward lower frequencies (22).
the device. The device ceases to operate as a superconducting

device and, if no damage has occurred, must recover after the
high-power source has been removed. The related topic of intentionally provoking a superconducting-to-normal transition
as a switching mechanism has been studied extensively
(2628).
Systematic studies of intermodulation distortion in HTS
samples and devices have been reported in the literature
(21,2932). TOI values in excess of 70 dBm at 1.3 GHz and
80 K have been obtained (29) for HTS planar transmission
lines a few millimeters long. This is well above most semiconductor low-noise amplifiers, for example. However, it must be
kept in mind that intermodulation distortion is a function of
the stored energy in the device, that is, the group delay. This
is important when considering the performance of HTS passband filters because filters have a delay characteristic that is
lower near the center of the passband and higher toward the
edges, depending on the filter order and type of response.
Thus, in a practical situation, if a passband filter is intended
to protect the system from a relatively high-power interfering
signal, this signal may lay on the filter skirts at a given rejection level in a region of relatively high delay. The intermodulation of this interferer with a desired signal in the middle of
the passband (relatively low delay) will produce spurs that
define the dynamic range of the filter from that specific systems perspective. To give a quantitative example, such dynamic range might be specified as a third-order intermodulation spur level of 90 dBm for an interfering signal of 40
dBm maximum power at 15 MHz from a desired signal at
passband center that has a maximum power level of 5 dBm.
In a case like this it would be difficult to talk about TOI,
which is usually defined as resulting from the intermodulation of two signals of the same power level. Such a definition
would apply to a spur-free dynamic range specification for the
One of the most important applications of HTS microwave

technology are high-performance passband filters because
they can be made in planar configurations. Filters are often
the dominant contributor to system volume, in particular
when banks of low-loss filters are required. As discussed earlier, high-Q structures can be obtained at the expense of high
volume. HTS planar configurations like microstrip or coplanar waveguide have Q comparable to cavities at a much
smaller volume (see Fig. 5) and so HTS is an attractive approach to reducing the volume of high-performance filters.
A straightforward way of thinking of bandpass filters is as
coupled resonators. The performance of a resonator is characterized by its quality factor Q, defined in Eq. (1). When the
resonator forms part of an electrical circuit, the circuit delivers and takes back energy from the resonator, affecting its
characteristics. The unloaded Q of a resonator, Qu, is its intrinsic quality factor, without the effects of an external circuit. A first-order filter consists of a single resonator. Its
bandwidth can be adjusted by the degree of coupling into the
resonator by the external circuit. In a lightly coupled resonator, little disturbance is introduced by the input and output
circuits and its resonant conditions and bandwidth are close
to those of an ideal, unloaded resonator. When coupling into
the resonator is strong, the disturbance is large and the Q is
now dominated by both the resonator and the external circuit,
making the total or loaded Q lower than the ideal, unloaded
Q, and therefore the filter bandwidth is now wider. For a
higher-order filter made up of various resonators coupled together, the idea is the same. Narrow-band filters require that
the resonators be loosely coupled to each other and the minimum bandwidth is limited by the unloaded Q of the resonators. Wider bandwidth filters will have tighter coupling
among resonators. Clearly, then, narrow-band filters are a desirable application for HTS because its inherent high Q enables narrow bandpass filters with low loss and small volume.
This was illustrated in Fig. 5, which helps to understand the
significance of the lossvolume trade-off from the point of
view of using HTS and metallic cavity resonators to make
filters. There is also a trade off between bandwidth, filter order (number of resonators), and insertion loss. The following
is an approximate expression for the mid-band insertion loss
of a filter (10), in dB, which reflects this trade-off:
n
gk
4.34
LdB,n
=
B k=1 Quk
(6)
Here, n is the filter order, gk are the normalized series inductance and shunt capacitance values of the low-pass prototype
filter (10), B is the filter bandwidth as a fraction of the center
frequency, and Quk is the unloaded Q of the kth resonator. For
the purposes of estimation, it is reasonable to assume that all
the resonators in the filter will have the same Qu. Figure 9
illustrates the trade-off between insertion loss, bandwidth,
and filter order as a function of resonator Qu. It shows how
expression (6) can be used to estimate the potential of a certain filter technology, in this case HTS, and understand its
724
5th order
4th order
L0
Insertion loss (dB)
5
4
0.1%
bandwidth
L1
L2
L3
S3
S1
S2
2
1
1%
bandwidth
0
102
103
104
105
Unloaded Q
Figure 9. Estimated insertion loss of fourth- and fifth-order Chebychev passband filters of 1% and 0.1% fractional bandwidths as a function of the unloaded Q of the resonators that make up the filter. It
was assumed that all the resonators have the same Q. The chart
shows the increase in insertion loss caused by increasing the filter
order by one and by reducing the fractional bandwidth by a factor
of ten.
limitations. The insertion loss was estimated for Chebychevtype filters (10) of the fourth and fifth orders, respectively,
and for 1% and 0.1% fractional bandwidths. The purpose of
this chart is to point out the difference in loss caused by increasing the filter order by one and by increasing the fractional bandwidth by a factor of ten. Figure 9 complements
Fig. 5 by helping to make a connection between the insertion
loss of a filter of a given order and bandwidth and a specific
structure and its volume. The information provided by these
two figures can readily be extended to cover other structures
and technologies.
Design Considerations
As discussed above, some of the most important applications
of superconductors are in narrow passband filters because
they can be realized in planar technology, which lends itself
to small structures that can be readily integrated with other
filters and circuitry. It was also discussed that the coupling
between the resonators that make up a narrow-band filter
needs to be weak. The mechanism for implementing weak
coupling between resonators must allow for its control and
predictability so that robust filter designs which are relatively
intolerant of external spurious coupling mechanisms can be
implemented. An example illustrating this point can be found
in the parallel-coupled line filter topology. Figure 10 shows
this topology, which is well known as being suited for microstrip filters with relative bandwidths below 15% (10). An analysis based on Figs. 5 and 9, however, shows that if HTS is
used then bandwidths below 1% are possible from the loss
standpoint. Indeed, this structure was used by several research groups to make initial HTS filter demonstrations with
1% to 2% fractional bandwidths (33,34). Table 3 shows the
couplings required to achieve a 1.25% bandwidth, fourth-order Chebychev filter with 0.1 dB ripple at 4 GHz (8), as well
as the distance between resonators (see Fig. 10). This distance was calculated using commercial software based on coupled microstrip line circuit models and a simple look-up table
technique (8) generated using a two-dimensional electromag-
Figure 10. Parallel coupled microstrip or stripline filter topology.

The lines are /2 resonators coupled as /4 backward-coupled sections. This is a well-known configuration suitable for fractional bandwidths of less than 15%; early demonstrations of HTS filters made
use of it. However, control of the weak coupling required for bandwidths less than 1%, which the low loss of HTS allows, is very difficult and results in designs that are very sensitive to material and
geometrical tolerances.
netic field solver as a tool to refine the results of the circuitmodel-based software. The effectiveness of this technique was
demonstrated experimentally (8). Notice in Table 3 that two
of the three required couplings are less than 30 dB and that
the error in estimating the coupled-line distances given by the
circuit-model software increases as the coupling gets weaker.
For narrower bandwidths, weaker couplings are needed,
which would result in larger separation between resonators
and hence increased difficulty in accurately predicting and
controlling the required coupling. Bandwidths of less than 1%
with this filter topology could probably be achieved with great
difficulty and very low yield because of the practical issues
associated with controlling the weak couplings required (35).
Workers in this field have realized this fundamental problem and have identified structures which allow significantly
better control of weak interresonator coupling. Recognizing
that in microstrip backward-coupled resonators, such as those
in Fig. 10, the problem is compounded by the presence of spurious forward coupling, researchers have demonstrated good
coupling control using forward-coupling alone in microstrip
(36) and backward coupling alone in stripline (37). Also, the
use of planar lumped elements (38), and inductive coupling
in coplanar line (39,40) have been demonstrated successfully.
Today, HTS filters are being made with bandwidths of 1% or
less by dedicated commercial companies for the base-station
wireless market as preproduction prototypes (19,41).
Complex Structures
The potential for filters with performance similar to bulkier
waveguide components but at significantly smaller sizes can
be most readily fulfilled for the case of banks of filters,
Table 3. Comparison Between Conventional and Look-Up

Table Approaches (see Fig. 10)
Parameter
Required
Coupling
(dB)
Conventional
(mm)
Look-Up Table
(mm)
S1
S2
S3
17.6
35.8
37.8
0.572
2.367
2.772
0.530
1.931
2.161
f o 4 GHz; W 0.176 mm; L 1 L 2 L 3 4.788 mm.
725
90 Hybrid couplers
Input
Identical f1
filters
f1
f2 f2
fn fn
Termination
T-Au pad
Output
channel n
Figure 11. Multiplexer architecture used to demonstrate a fourchannel HTS microstrip device. This configuration has the advantage
of allowing as many channels as the bandwidth covered by the 90
hybrid. Each filter is terminated in 50 and is essentially isolated
from the others. Other schemes require that the impedance termination in each filter be adjusted to account for the presence of all the
filters in the multiplexer, practically limiting the maximum number
of channels to ten or twelve.
whether switched or multiplexed. Because HTS technology is

planar, a relatively high level of integration is possible so
that, as opposed to waveguide or dielectric resonator filter
technology, a bank of N filters each occupying a volume V
occupies a volume N V, where V is significantly smaller
than for conventional technologies for the same performance.
HTS filter banks have been demonstrated by several groups
in the form of bandpass multiplexers (8,4244) or banks of
switched band-reject filters (45).
Example: Four-Channel Pass-Band Multiplexer. Figure 11
shows a diagram of a multiplexer architecture (8,42). It can
accommodate as many channels as the bandwidth of the 90
hybrid coupler covers. Input microwave energy is equally split
at the first coupler. If the frequency is within the passband of
the two identical filters connected to the coupler, it passes
through the filters and adds in phase at the output port of the
output hybrid coupler for Channel 1. If the frequency is not
within the passband of the Channel 1 filters, the signal is
reflected back to the input coupler where it recombines such
that it is out-of-phase at the input port and in-phase at the
input of the second channel hybrid coupler. The process then
repeats itself until the signal exits the device through the appropriate channel port. Figure 12 presents details of one implementation of this device (8,46) showing one input, four outputs, and a through port terminated in an external (coaxial)
load. Additional channels could be connected to this port provided they are still within the bandwidth (about 10%) of the
hybrid coupler used in this demonstration. Figure 12 also
shows a detail of the assembly, which includes the internal
HTS interconnections between filters and the integrated thinfilm resistive terminations at the out-of-phase port of the output hybrid in each channel.
The HTS material used for this work was YBCO thin film
deposited on 500 m-thick LAO substrates. The package included an aluminum frame holding the external coaxial connectors and niobium carriers onto which the LAO substrate
pieces were mounted. Niobium is a good thermal expansion
Mo-ti
resistor
VBCO
Figure 12. Photo montage of a four-channel YBCO-on-LAO microstrip multiplexer demonstrated under the US Navys High-Temperature Superconductivity Space Experiment II (HTSSE-II). Details of
the design, fabrication, and assembly of this device can be found in
Refs. 8 and 46.
match to LAO, so the electrical and mechanical integrity of

the device was preserved when cycling from ambient temperature to near 77 K. The substrate-carrier assemblies were
mounted on the aluminum frame using a beryllium-copper
spring arrangement. Figure 13 shows the measured performance. The low-frequency skirts of Channels 2, 3, and 4 show
some level of interaction between channels that can be eliminated using a wider guard-band between channels. Refs. 8
and 46 include a full discussion of the design, fabrication, and
measurements on this device.
This unit was one of a series of demonstration devices delivered to the US Navys High-Temperature Superconductivity Experiment II program by several contractors for inclusion
into the space package (47).
High-Power-Handling Designs
The promising applications of HTS require that HTS filters
handle sufficient signal power levels as to maintain linearity
over a significant dynamic range. As explained earlier, current crowding along the edges of typical planar transmission
0
Insertion loss (dB)
Output
channel 1
50
100
3.5
4.0
Frequency (GHz)
4.5
Figure 13. Measured response of the four-channel HTS multiplexer.

Further details can be found in Refs. 8 and 46.
726

Conductive
heat load
Control lines (dc)
Vacuum
housing
H
J
RF line
Radiation
shield
RF line
HTS device
Thermal
connector
Cold
head
(a)
(b)
Figure 14. Diagram (courtesy of Dr. Z-Y Shen, E. I. du Pont de Nemours and Co.) (29), showing the magnetic field and current distribution in a /2 microstrip resonator (a) and a TM010 printed disk resonator (b). In the disk the magnetic field lines are circular and remain
on the plane of the disk, so the current is not highly nonuniform as
is the case of the regular microstrip resonator. The advantage of the
disk is that it can handle much higher power levels (29,49,50).
lines (e.g., microstrip, stripline, and coplanar waveguide) ultimately limits the maximum power level that can be handled.
Increasing the quality of the material and improving the design of filter structures has been a major endeavor at several
institutions. Improved filter designs are based on planar
structures which avoid the effects of significant current
crowding at the edges, as is the case of low-impedance microstrip lines (48). Most significant is the work employing planar
resonator structures based on the circular TM010 mode
(29,49,50). The most salient features of this approach are
shown in Figs. 14(a) and (b), which show the electromagnetic
fields and current profile in a microstrip and a disk resonator,
respectively. In the latter the RF magnetic fields do not close
above the substrate but within it, under the disk. Thus the
current density does not peak at the edges of the resonator
and its distribution is more uniform. The only possible drawback of this approach is that the fields are more confined to
the disk resonator, and intercoupling between resonators to
form a filter may be more difficult, perhaps requiring threedimensional structures for proper control of the coupling. This
Cryocooler
engine
Radiated
heat load
Figure 15. Schematic diagram of the cryogenic package for a hypothetical HTS device showing conducted heat inputs through input/
output RF and dc control lines and the mechanical support of the cold
head, as well as the radiated heat input from the (warm) wall of the
vacuum housing. The purpose of this diagram is to show the main
elements that affect the design of the cryogenic package.
would eliminate some of the planar integration advantages.

Fully planar filters using this concept, however, have been
successfully demonstrated (29).
HIGH-TEMPERATURE SUPERCONDUCTING DELAY LINES
Work on superconducting delay lines started at Lincoln Laboratory well before the advent of high-temperature superconductivity, and concentrated mostly on linearly dispersive delay lines for analog signal processing. Linearly dispersive
delay lines have delay characteristics which vary linearly
with frequency over a certain operating bandwidth and can
be used to perform pulse compression, a technique to process
and detect small signals which may be below the receiver
noise floor (1). The pioneering work at Lincoln Laboratory in
this area using LTS and, more recently, HTS thin-film technologies has been extensively documented in the literature
(1,51).
Table 4. Sample System Requirements That Will Affect the Choice of Cooler and Cryogenic Packaging Approach
Requirement
Size and weight
Cool-down time
Vibration
Power consumption and
power supply type
Mode of operation
Temperature stability
and control
Unattended lifetime
Vacuum lifetime
Comments
Stringent in almost all applications
Some applications may require very fast turn-on time (e.g., a few minutes). They would be a driver toward higher cooler power and lower HTS device thermal mass
For example, a minute amount of mechanical distortion on a circuit caused by vibration from the cooler
may generate a phase modulation that degrades the circuit performance
E.g., 120 V ac
E.g., continuous, intermittent, short missions and then mostly idle, etc.
While any fine temperature feedback control loop (0.01 K) tends to be done using heaters and a temperature sensor, some applications may require a certain degree of cooling engine control (0.5 K)
Some applications (e.g., space) may require a lifetime on the order of 10 years or more
All-welded construction; use of getters in a clean, well-conditioned (baked) system
727
Table 5. Cooling Requirements That Will Influence the Cooling Power (Heat Lift) Required for a Given Application
Requirement
Comments
Power dissipated in the

device
A filter with a 0.5 dB insertion loss that must pass a 20 W signal will dissipate 2 W of heat that must be removed by the cryocooler. Also, semiconductor devices such as low-noise amplifiers, which improve in noise
and gain performance when cooled, always dissipate a certain amount of heat which must be taken into
consideration
These are the electrical interface between the cryocooled device and the outside world. For example, a filter
might require two microwave leads (input and output) and two pairs of dc control lines for the heat sensor
and a small heater to keep the temperature constant. These conductors represent a heat loss that the
cooler must overcome because they connect the outside ambient temperature with the cold device. While
the dc control lines are typically made of thin low-thermal-conductivity, high-resistivity wire (e.g., gauge
32 manganin), the microwave leads must achieve a compromise between insertion and thermal loss
Radiation loss is another form of heat loss that the cooler must overcome and therefore must be minimized.
The total surface area and their infrared radiation emmisivity are important design parameters. Low-emmisivity radiation shields are typically used between the warm vacuum vessel wall and the cold device
For those applications that have a cool-down time requirement, the thermal mass of the device to be cooled
is important and will be affected by the microwave packaging material and its shape
Number of microwave
and dc control leads
Surface area
Thermal mass
Nondispersive delay lines have a constant delay-versusfrequency characteristic and are typically used as analog
memory elements that can store a signal for, say, up to a few
hundred nanoseconds while the system is engaged in other
processing steps. Work on HTS nondispersive delay lines has
also been significant (5255). Including two recent instantaneous frequency measurement subsystems based on banks of
delay lines (52,55). Clearly, the advantages of superconductivity are that a long length of line can be fabricated in a small
volume by defining a long, planar transmission line on a wafer. Ref. 54 compares conventional nondispersive delay lines,
which require amplifiers between sections of transmission
line (e.g., coaxial), with HTS delay lines using projections
based on measurements made on relatively short (22 ns) delay lines. Key delay-line parameters are delay, bandwidth, insertion loss, and third-order intercept point. Conventional delay lines that must resort to amplification to boost the signal
are limited in dynamic range by the amplifiers.
CRYOGENIC PACKAGING
Key to the insertion of superconducting microwave circuits
into electronic systems is the integration of the HTS components with a cryogenic refrigerator and its associated control
electronics. Clearly, for HTS technology to be ultimately successful, the user must be rendered able to ignore the fact that
cryogenics are used at all, by providing long-lifetime cryocoolers and optimally small cryogenic packages with standard envelop characteristics and interfaces (e.g., 19 in rack mounts
and back-plane blind-mate connectors).
Many important considerations enter into the design of a
cryogenic package suitable for a microwave HTS subsystem.
Figure 15 is a schematic representation of this package, showing its main elements and the various heat inputs that must
be considered for an appropriate thermal design. Ref. 41 provides specific details on the cryogenic package for a communications filter subsystem.
The choice of a cryocooler will depend on the system and
the cooling requirements. An airborne military application
may require the use of a small Stirling-cycle cooler because
of volume restrictions. On the other hand, a communications
ground station in a remote location that needs to operate unattended for a long time may require a larger, more reliable
refrigerator of the GiffordMcMahon type. Cooling requirements are imposed by the component or subsystem to be
cooled and will determine the amount of cooling power required at the operating temperature. Typical sample system
and cooling requirements and some comments as to their significance are given in Tables 4 and 5, respectively.
Table 6. Some Cryogenic Refrigerator Types Likely to Be Used in HTS Technology

Cooler Type
Heat-Lift Range
Available at 80 K
Split Stirling
0.53 W
Integral Stirling
0.55 W
GiffordMcMahon
2200 W
Throttle-cycle
4 W
JouleThomson
0.52 W
Pulse tube
0.52 W
Comments
Available from many manufacturers; used primarily in the tactical military infrared detector industry. Has a cold head separated from a compressor by a metallic transfer line up to 15 cm
long
Also used in infrared detectors; at least one version is being used in an HTS development prototype. The compressor and cold finger are integrated into one unit
Widely used in the support of vacuum systems for semiconductor industry; highly reliable and
versatile. The compressor and cold head are separate units connected by fluid lines that can
be several meters long
Reliable and low cost. The compressor and cold head are separate units connected by fluid lines
that can be several meters long
Generally used as an open-cycle cooling system for short tactical missile IR detector applications
Emerging technology, low cold-head vibration and long lifetime potential
728
infrared imaging detector technology and perhaps new developments of cooled semiconductor components for fast computer workstations, cryocooler technology is progressing to
the point where long lifetimes and small-size, low-weight coolers are now widely available.
BIBLIOGRAPHY
1. R. S. Withers and R. W. Ralston, Superconductive analog signal
processing devices, Proc. IEEE, 77: 12471263, 1989. This paper
contains many references to earlier work by the authors.
Figure 16. Photograph of a HTS filter assembly for commercial wireless applications (courtesy of Superconductor Technology, Inc.).
Cryocoolers likely to be used in microwave HTS technology

will typically have from 1 W to 5 W of cooling capacity. A
primary concern systems designers have is the reliability of
cryogenic refrigerators, which varies greatly depending on
their type and size. Leveraging developments in other fields,
such as infrared detectors, the reliability of small, military
tactical cryocoolers has steadily increased in recent years,
with some manufacturers claiming up to 20,000 h of meantime to failure (MTTF). On the other hand, larger laboratory
or industrial units and specialized coolers for aerospace applications operate for 5 years to 10 years and require minimal
servicing. Table 6 lists some of the cryocooler types of interest. The intent here is not to be all-inclusive but to provide a
basic reference to the type of coolers most likely to be employed in HTS microwave technology. Reference 56 as a good
source of the latest developments in cryocooler technology.
Figure 16 is a photograph of a commercial HTS filter subsystem, showing the cryocooler and associated electronics in
their open package.
CONCLUSION
High-temperature superconductor microwave technology offers unique advantages derived from the low microwave loss
of HTS materials and the inherent low thermal noise in cryogenically cooled components. The main applications to date
are related to increased microwave receiver sensitivity, and
this is most likely to have an impact on wireless military and
commercial communications systems. The reason is that receiver sensitivity and dynamic range must be preserved in
the presence of a large number of spurious signals which, if
unfiltered, degrade receiver performance. Generation of clean
transmitted signals requires filtering in the transmitter and
this, coupled with the need to reject unwanted high-power
signals at the receiver, has spurred work on high-power handling in HTS filters. Great interest in the United States and
abroad exists in the wireless commercial communications
market and several companies are testing base-station receiver front-ends consisting of cryogenically-cooled filter-LNA
subassemblies.
HTS microwave filters are therefore a promising technology, especially at frequencies below 3 GHz where the loss in
conventional microwave materials force high-performance
filters to be very large in order to achieve the required low
insertion losses and selectivity. Leveraging developments in
2. P. A. Smith et al., YBCO thick films for high Q resonators, IEEE

3. D. W. Face et al., Large area YBa2Cu3O7 films for high power
microwave applications, IEEE Trans. Appl. Supercond., 5: 1581
1586, 1995.
4. A. Pique et al., Microwave compatible YBa2Cu3O7x films on ferrimagnetic garnet substrates, Appl. Phys. Lett., 67: 17781780,
1995.
5. E. J. Smith, J. Musolf, and E. Soares, Composition controlled
MOCVD as a route to high Q HTS thin film devices, Proc. 3rd Jt.
ISTEC/MRS Int. Workshop Supercond.: Suitable Mater. Proces.
HTS Appl. Towards Next Decade, 1997, p. 145.
6. B. Utz et al., Deposition of YBCO and NBCO films on areas of 9
inches in diameter, IEEE Trans. Appl. Supercond., 7: 1272
1277, 1997.
7. J. W. Ekin, A. J. Panson, and B. A. Blankenship, Method for
making low-resistivity contacts to high Tc superconductors, Appl.
Phys. Lett., 52: 331333, 1988.
8. S. H. Talisa et al., High-temperature superconducting spacequalified multiplexers and delay lines, IEEE Trans. Microw. Theory Tech., 44: 12291239, 1996.
9. R. K. Hoffmann, Handbook of Microwave Integrated Circuits, Norwood, MA: Artech House, 1987.
10. G. L. Matthaei, L. Young, and E. M. T. Jones, Microwave Filters,
Impedance-Matching Networks, and Coupling Structures, Dedham, MA: Artech House, 1980.
11. Z.-Y. Shen, High-Temperature Superconducting Microwave Circuits, Norwood, MA: Artech House, 1994.
12. S. Ramo, J. R. Whinnery, and T. Van Duzer, Fields and Waves in
Communication Electronics, New York: Wiley, 1965.
13. I. Vendik and O. Vendik, in E. Kollberg (ed.), High Temperature
Superconductor Devices for Microwave Signal Processing, St.Petersburg, Russia: Scladen, 1997, 3 parts.
14. K. R. Carroll, J. M. Pond, and E J. Cukauskas, Superconducting
kinetic-inductance microwave filters, IEEE Trans. Appl. Supercond., 3: 816, 1993.
15. Unpublished data obtained jointly by Northrop Grumman, NASA
Lewis Research Center and Superconductor Technology, Inc.
16. W. W. Mumford and E. H. Scheibe, Noise Performance Factors in
Communication Systems, Dedham, MA: Horizon House-Microwave, 1968, UMI Out-of-Print Books on Demand.
17. S. H. Talisa et al., Dynamic range considerations for high-temperature superconducting filter applications to receiver frontends, IEEE MTT-S Int. Microw. Symp. Dig., 1994, pp. 9971000.
18. J. B.-Y. Tsui, Microwave Receivers with Electronic Warfare Applications, Malabar, FL: Krieger, 1992.
19. G. Koepf, Superconductors improve coverage in wireless networks, Microw. RF, 37 (4): 6374, 1998.
20. D. E. Oates et al., Nonlinear surface resistance in YBa2Cu3O7x
thin films, IEEE Trans. Appl. Supercond., 3: 11141119, 1993.
SUPERCONDUCTING LEVITATION
729
21. C. Wilker et al., Nonlinear effects in high-temperature superconductors: 3rd order intercept from harmonic generation, IEEE
43. C. Rauscher, J. M. Pond, and G. B. Tait, Cryogenic microwave

channelized receiver, IEEE Trans. Microw. Theory Tech., 44:
12401247, 1996.
22. M. Golosovsky, Physical mechanisms causing nonlinear microwave losses in high-Tc superconductors, 8th Workshop RF Supercond., Abano Terme, Italy, 1997, Invited Paper.
23. S. Sridhar, Non-linear microwave response of superconductors
and ac response of the critical state, Appl. Phys. Lett., 65: 1054
1056, 1994.
44. S. J. Fiedziuszko et al., Low loss multiplexers with planar dual

mode HTS resonators, IEEE Trans. Microw. Theory Tech., 44:
12481257, 1996.
24. J. H. Oates et al., A nonlinear transmission line model for superconducting stripline resonators, IEEE Trans. Appl. Supercond., 7:
1722, 1993.
25. D. E. Oates et al., Microwave power dependence of YBa2Cu3O7
thin-film Josephson edge junctions, Appl. Phys. Lett., 68: 705
707, 1996.
26. I. Vendik and O. Vendik, High Temperature Superconducting Devices for Microwave Signal Processing, St. Petersburg, Russia:
Skladen, 1997, Part 2, Chap. 7.
27. B. S. Karasik et al., Subnanosecond switching of YBaCuO films
between superconducting and normal states induced by current
pulse, J. Appl. Phys., 77: 40644070, 1995.
28. I. Vendik et al., The superconducting microwave devices based
on S-N transition in HTS films, 27th Eur. Microw. Conf. Proc.,
1997, pp. 909914.
29. Z.-Y. Shen et al., Power handling capability improvement of hightemperature superconducting microwave circuits, IEEE Trans.
Appl. Supercond., 7: 24462453, 1997.
30. Z. Ma et al., RF power dependence study of large area YBCO thin
films, IEEE Trans. Appl. Supercond., 7: 19111916, 1997.
31. T. Dahm and D. J. Scalapino, Theory of intermodulation in a superconducting microstrip resonator, J. Appl. Phys., 81: 2002
2009, 1997.
32. O. G. Vendik, I. B. Vendik, and T. B. Samoilova, Nonlinearity of
superconducting transmission line and microstrip resonator,
IEEE Trans. Microw. Theory Tech., 45: 173178, 1997.
33. S. H. Talisa et al., Low- and high-temprature superconducting
microwave filters, IEEE Trans. Microw. Theory Tech., 39: 1448
1454, 1991.
34. W. G. Lyons et al., High-Tc superconductive microwave filters,
IEEE Trans. Magn., 27: 25372539, 1991.
35. W. G. Lyons and L. H. Lee, Accuracy issues and design techniques for superconducting microwave filters, Comput.-Aided Des.
Supercond. Microw. Components Workshop, IEEE Int. Microw.
Symp., 1994. The authors are with Lincoln Laboratory, Massachusetts Institute of Technology.
36. D. Zhang et al., Compact forward-coupled superconducting microstrip filters for cellular communications, IEEE Trans. Appl.
Supercond., 5: 26562659, 1995.
37. G. L. Matthaei and G. L. Hey-Shipton, Novel staggered resonator
array superconducting 2.3-GHz bandpass filter, IEEE Trans. Microw. Theory Tech., 41: 23452352, 1993.
38. D. G. Swanson, R. Forse, and B. J. L. Nilsson, A 10 GHz thin
film lumped element high temperature superconductor filter,
IEEE MTT-S Int. Microw. Symp. Dig., 1992, pp. 11911193.
39. D. G. Swanson and R. Forse, An HTS end-coupled CPW filter at
35 GHz, IEEE MTT-S Int. Microw. Symp. Dig., 1994, pp. 199202.
40. A. Vogt and W. Jutzi, An HTS narrow bandwidth coplanar shunt
inductively coupled microwave bandpass filter on LaAlO3, IEEE
Trans. Microw. Theory Tech., 45: 493497, 1997.
41. M. J. Scharen et al., Filter subsystems for wireless communications, IEEE Trans. Appl. Supercond., 7: 37443749, 1997.
42. R. R. Mansour et al., Design considerations of superconductive
input multiplexers for satellite applications, IEEE Trans. Microw.
Theory Tech., 44: 12131228, 1996.
45. N. O. Fenzi et al., An optically switched bank of HTS bandstop

filters, IEEE MTT-S Int. Microw. Symp. Dig., 1994, pp. 195198.
46. S. H. Talisa et al., High-temperature superconducting four-channel filterbanks, IEEE Trans. Appl. Supercond., 5: 20792082,
1995.
47. T. G. Kawecki et al., The high temperature superconductivity
space experiment (HTSSE-II) design, IEEE Trans. Microw. Theory Tech., 44: 11981212, 1996.
48. G.-C. Liang et al., High-temperature superconducting microstrip
filters with high power handling capability, IEEE MTT-S Int. Microw. Symp. Dig., 1995, pp. 191194.
49. H. Chaloupka et al., Superconducting planar disk resonators and
filters with high power handling capability, Electron. Lett., 32:
17351736, 1996.
50. S. Kolesov et al., Planar HTS structures for high-power applications in communication systems, J. Supercond., 10: 179187,
1997.
51. W. G. Lyons et al., High temperature superconductive wideband
compressive receivers, IEEE Trans. Microw. Theory Tech., 44:
12581278, 1996.
52. G. C. Liang et al., Space-qualified superconductive digital instantaneous frequency-measurement subsystem, IEEE Trans. Microw. Theory Tech., 44: 12891299, 1996.
53. N. Fenzi et al., Development of high temperature superconducting 100 nanosecond delay line, SPIE Proc., 2156: 143151, 1994.
54. S. H. Talisa et al., High-temperature superconducting wide band
delay lines, IEEE Trans. Appl. Supercond., 5: 22912294, 1995.
55. M. Biehl et al., A 4-bit instantaneous frequency meter at 10 GHz
with coplanar YBCO delay lines, IEEE Trans. Appl. Supercond.,
5: 22792282, 1995.
56. R. G. Ross, Jr. (ed.), Cryocoolers 9, A Publication of the International Cryocooler Conference, New York: Plenum, 1997.
SALVADOR H. TALISA
Northrop Grumman Corporation
SUPERCONDUCTING HIGH-ENERGY PARTICLE DETECTOR MAGNETS. See HIGH-ENERGY PHYSICS PARTICLE DETECTOR MAGNETS.
729
The levitation of a permanent magnet over a superconductor
is one of the basic tests of superconductivity, and it is a sight
that has inspired the investigation of applications not possible with any other technology (15). In this article, I briefly
review the fundamental physics of superconductor levitation,
discuss basic levitational phenomena and the features of superconductor levitation pertinent to bearings, and mention
some possible applications of superconductor levitation, with
emphasis on high-efficiency flywheel energy storage.
730
STABLE LEVITATION
In its simplest form, a superconducting levitational system
comprises a permanent magnet levitated in a stable position
over a superconductor. This stability is in stark contrast to
most magnetic systems, which are statically unstable. Earnshaw showed that there is no stable, static 3-D arrangement
of a collection of poles (electric, magnetic, or gravitational)
whose magnitudes do not change and which interact via a
1/r2 force law (6). Braunbek extended the result to show that
no stable static configuration exists when paramagnetic or
ferromagnetic material is included in the arrangement (7).
These results collectively are often referred to as Earnshaws
theorem. Earnshaws theorem is grasped intuitively by most
people when they release a permanent magnet next to the
ferromagnetic door of their refrigerator. The magnet moves to
stick to the door, or it falls on the floor. It does not hover in
space near the point where it was released.
Diamagnetic materials, such as superconductors, are not
governed by Earnshaws theorem, and they enable the possibility of creating stable levitation systems. One of the fundamental properties of superconductors is their tendency to exclude magnetic flux from their interiors. This exclusion of
magnetic flux (the Meissner effect), makes them behave like
strong diamagnets. Accordingly, a superconductor with a permanent magnet positioned close above it, as shown in Fig.
1(a), develops a shielding current, which excludes flux so that
the actual magnet sees its mirror image.
More specifically, if the magnetization of a permanent
magnet is in the vertical direction, with its north pole down,
the image is also vertically magnetized, but with its north
pole up, exerting a repulsive force on the real magnet. The
closer the magnet moves to the superconductor, the stronger
the repulsive force. The farther away the magnet moves, the
weaker the force. In fact, if the magnet flips over so that the
opposite (south) pole faces the superconductor, the screening
currents in the superconductor will reverse so as to reverse
the magnetization of the image in the superconductor, maintaining the repulsive interaction. This setup yields levitation
stable in the vertical direction. Horizontal stability is obtained if the superconductor is given a concave shape, so that
vertical superconducting walls are formed around the magnet, as first demonstrated by Arkadiev using lead, a Type-I
superconductor, in which magnetic flux is totally excluded
from the interior (8).
From a technological viewpoint, the most useful superconductors are usually Type-II superconductors, in which, above
a first critical field, Hc1, it is energetically favorable for magnetic flux to enter the interior of the superconductor in discrete localized regions that become normal (i.e., not superconducting) with each region of flux surrounded by a small vortex
of superconducting shielding current. In Type-II superconductors, the stability of the levitational phenomena resulting
from the diamagnetic response is greatly enhanced by the additional phenomena resulting from flux pinning, shown in
Fig. 1(b). A flux pinning center is a nonsuperconducting region, such as an inclusion, crack, or other crystalline defect.
Because the superconducting region surrounding the nonsuperconducting center is strongly inclined to exclude magnetic
flux, a magnetic flux line through the center often becomes
trapped there. When a sufficient number of flux lines is
trapped in the superconductor, the permanent magnet re-
Permanent
magnet
Shielding
current
Superconductor
Image of
permanent
magnet
(a)
Permanent
magnet
Superconductor
Pinning centers
(b)
Attractive
lateral force
Permanent
magnet
s
n
Repulsive
levitation force
Diamagnetic
shielding currents
Superconductor
Trapped flux
(c)
Figure 1. Schematic diagrams of levitation basics: (a) diamagnetic
response; (b) flux pinning; (c) flux trapping.
mains levitated in position, even over a flat surface. The flux

lines between the permanent magnet and the superconductor
act in an imperfect analogy to mechanical springs with attachments on the permanent magnet and in the superconductor. If the magnet moves a small distance laterally, so that
the flux lines remain in their pinning centers, the flux lines
bend and produce a laterally restoring shear force, according
to the Maxwell electromagnetic stress tensor. If the magnet
moves vertically or horizontally, the springs pull the magnet back to its equilibrium position. If the flux pinning is sufficiently strong, the magnet is stably suspended below the superconductor (9) or even along its side (10).
If the magnet moves far enough laterally that the flux lines
move from their original pinning centers to new ones, then an
additional stabilizing force, involving trapped flux, begins to
act. Trapped flux consists in regions of induced magnetization
in the superconductor of the same pole orientation as the levitated magnet and results from movement of flux lines from
their pinning centers that decreases the local flux. As shown
in Fig. 1(c), this results in an attractive interaction that reduces the levitation force but provides a lateral restoring
force.
Interest in the potential of superconductor levitation in various applications greatly increased with the discovery of superconductors whose critical temperatures (i.e., temperatures
at which they transit from the normal state to the superconducting state) exceeded the boiling point of nitrogen. Although one could create a superconducting wire magnet for
levitation, most of the present efforts involve the use of bulk
superconductors or thin-film superconductors. Unlike superconducting wire applications, in which the supercurrent must
pass from grain to grain along quite a distance, the supercurrent for levitation applications needs to circulate only within
individual grains.
The present material of choice for superconducting levitation is Y-Ba-Cu-O (YBCO) because it exhibits a high magnetic
irreversibility field at liquid nitrogen temperatures and has
the ability to grow large grains. In addition to the two temperature-dependent phase-transition fields, Hc1 and Hc2, all superconductors have a magnetic irreversibility field, Hirr, that
lies between Hc1 and Hc2. Hirr is the field at which the magnetization M as a function of applied field H is no longer doublevalued (11). For the low-temperature superconductors NbTi
and Nb3Sn, Hirr is extremely close to Hc2, and there is no important distinction between them. At higher temperatures,
thermal activation is much greater, which leads to easier flux
motion near Hc2 for HTSs. Hirr marks a phase transition between the region where magnetic flux is solidly pinned in the
superconductor and the region where flux may move. Sometimes the curve is said to denote the boundary between the
region where flux is frozen and the region where flux is
melted. Of all the known HTSs, YBCO has a relatively low
critical temperature of 92 K, but it has the highest irreversibility curve at 77 K and lower temperatures. For stable levitation, it is important that the flux be frozen in the superconductor. Otherwise, the permanent magnet would slowly lose
levitation height.
The magnetization of the superconductor is proportional to
the product of the critical current density and the grain diameter. Large grain diameters are important to achieve sufficiently large magnetizations for useful levitation forces. In
bulk materials, the grains grow to diameters of several centimeters when made by a melt-texturing process (12). In the
present state of the art, the upper limit of the grain diameter
produced by this process is about 10 cm. The ability to produce good-quality YBCO thin films is also limited to about
this size.
LEVITATIONAL PHENOMENOLOGY
If the permanent magnet is pulled hard enough to the side or
vertically, it is possible to move one or more of the trapped
flux lines into new pinning centers and so change the equilibrium position. Such a change results in a hysteretic effect in
the levitational force and an associated energy loss if it occurs
in a cyclic pattern. To explore the hysteretic effect, it is convenient to divide the behavior into two processes. The first is
called field-cooled. It occurs when the superconductor is
cooled below its critical temperature while there is a substantial magnetic field present, that is, the permanent magnet is
close to the superconductor. Field cooling produces less repul-
sive levitation force but can be used to make an attractiveforce bearing. The second is called zero-field-cooled. It occurs
when the superconductor is cooled below its critical temperature in the absence of a magnetic field, that is, when the permanent magnet is far from the superconductor. Zero-field
cooling results in the largest repulsive force but may be practically inconvenient, because it requires the cooling of the superconductor prior to the assembly of the bearing.
The hysteretic nature of the levitational phenomenon for
movements in the vertical direction is illustrated in Fig. 2. In
this example, a cylindrical, vertically magnetized permanent
magnet was kept with its bottom surface at a height of 10.0
mm above the top surface of a cylindrical YBCO superconductor while the YBCO was cooled, essentially a zero-field-cooled
condition. Then the magnet was slowly brought down to a
position 1 mm above the superconductor, and the force was
measured at various points along this first descent. Then the
magnet was moved away from the superconductor, back to its
original zero-field-cooled position. Then it followed a second
descent during which a minor reversal of 0.4 mm was made
at a distance of 2.0 mm. A second ascent was identical with
the first, and a third descent was identical with the second,
etc. As seen in Fig. 2, the force during the first descent is
always larger than the force during the second. Upon reversal
from 1 mm during the first ascent, the force drops very
quickly and even becomes negative, indicating an attractive
force. A combination of the first ascent and second descent
forms a major hysteretic loop, and the area under the curve
is equivalent to the hysteretic energy loss. The width of the
minor loop is much smaller than that of the major loop. Thus,
the superconductor acts as a nonlinear damper, and the
damping coefficient increases with amplitude. From Fig. 2, it
is also clear that the slope of the minor loop, which represents
the magnetomechanical stiffness of the system, is considerably higher than that determined from the major loop.
A feature of the first descent is that over several millimeters above the superconductor surface, the force is exponential with distance, as shown in Fig. 3. In practice, it is difficult
to measure the force immediately above the superconductor,
partly because surfaces are not flat, but mainly because the
surface is usually covered with liquid nitrogen. The exponential behavior shown in Fig. 3 allows extrapolating the force to
35
1st descent
1st ascent
2nd descent
minor loop
30
Levitation force (N)
SUPERCONDUCTOR LEVITATORS
731
25
20
15
10
5.0
0.0
5.0
8
6
Distance (mm)
10
12
Figure 2. Levitation force versus distance between the permanent

magnet and superconductor.
732
10
Levitation force (N)
important, the superconductor is a pure diamagnet and the

magnet is a set of magnetic dipoles (16). A frozen mirror
image may be used in conjunction with the diamagnetic mirror image to describe the elastic properties and energy loss in
a field-cooled system (17). The Bean model, which assumes
that the current circulating in the superconductor is either at
its critical value or zero, is often used in efforts to address the
hysteretic behavior (18).
Fo(N)
13.2
15.5
9.3
10.7
10.5
15.3
1.0
SUPERCONDUCTING BEARINGS
0.10
0.0
5.0
10.0
15.0
Height (mm)
Figure 3. Levitation force versus distance on first descent between
reference magnet and several superconductors, showing the levitation
force extrapolated to zero height.
the surface, and various superconductors can be compared

this way by using a permanent magnet of known strength.
In the examples shown in Figs. 2 and 3, the superconductor was cooled while the permanent magnet was far from its
surface, that is, zero-field-cooled. In the field-cooled case, the
levitational force is approximately zero after the superconductor is cooled below its critical temperature. However, the magnetomechanical stiffness is approximately the same as for the
zero-field-cooled case at the same separating distance. The
stiffness is dependent on the amplitude of vibration (13,14).
For small amplitudes, the stiffness is constant, but, after
some critical amplitude that depends on height, the stiffness
begins to decrease as the amplitude increases. The higher the
current density of the superconductor, the higher the critical
value for the onset of stiffness decrease.
The levitational force that the superconductor provides is
proportional to its average magnetization, which is proportional to the product of its grain diameter times its critical
current density. The critical current density in typical melttextured YBCO samples at 77 K is about 40 kA per cm2,
which, together with a diameter of several cm, allows the levitational pressures between the YBCO and an NdFeB permanent magnet to be as high as about 280 kPa (40 psi). Such
pressures have been measured in the authors laboratory with
the very best YBCO samples. However, this pressure occurs
at zero separation distance. In a practical system with a finite
separation, the pressure will be at least a factor of 2 to 3
times lower.
Thin-film YBCO often has critical current densities that
exceed 1 MA per cm2. Because the thickness of these films is
only about 1 micron, they do not provide much levitational
force. However, the stiffness of such films is often of the same
order as that produced by bulk materials of greater thickness (15).
THEORETICAL MODELS
Various theoretical models account for different aspects of superconductor behavior when a permanent magnet approaches. In many cases in which hysteretic behavior is not
If the azimuthal homogeneity of the magnetic field of the permanent magnet is high, for example, if the magnet is a cylinder with uniform magnetization throughout, the levitated
magnet rotates freely above the superconductor. As long as
the distribution of magnetic flux in the superconductor does
not change, rotation encounters no resistance. If the magnet
is spinning, the hysteretic loss in the superconductor decreases the rotational rate.
In an electromechanical system, such as a magnetic bearing, the parameters of interest are the levitational force, stiffness, damping, and rotational loss. The 280 kPa levitational
pressure is lower than that achievable in a conventional electromagnetic bearing (1 MPa) and significantly lower than
that typically achieved in mechanical roller bearings (10
MPa). The amount of mass levitated directly depends on the
number and size of permanent magnets and superconductors
available. In the present early period of technological development for superconducting bearings, several laboratories have
stably levitated masses greater than 100 kg.
In practical superconductor bearings, the low levitational
pressure available in the interaction between the permanent
magnet and the superconductor is often augmented by various hybrid schemes in which interactions between pairs of
permanent magnets provide the bulk of the levitational force.
These interactions are unstable, as Earnshaws theorem predicts, but the inclusion of a properly designed superconducting component in the bearing is sufficient to stabilize the
complete bearing. Augmentation takes the form of an Evershed-type design, in which a pair of permanent magnets is in
attractive levitation, employs permanent magnets in repulsive levitation, or uses active magnetic bearings (5).
The hysteretic nature of a superconducting bearing also
makes damping of translational motion amplitude-dependent.
For low-amplitude vibrations, damping is small, but quickly
increases as the vibrational amplitude increases. This hysteretic nature of the HTS bearing thus contributes to the robustness of the system. The hysteretic nature also results in
a larger uncertainty of the equilibrium position of the rotor
than is typical in most rotating machinery. This uncertainty
requires larger running gaps between moving and stationary
parts.
The ease with which a permanent magnet spins, when levitated over a superconductor, and the absence of contact between the surfaces, produce the illusion that the rotation is
frictionless. In reality, small magnetic losses gradually slow
the rotation. The losses are primarily the result of azimuthal
inhomogeneities in the magnetization of the permanent magnet, which produce hysteretic loss in the superconductor. Typically, in permanent magnets with the best homogeneities, at
a fixed radius above the rotating surface, the amplitude of the
ac component of the magnetic field is of the order of 1% of the

average field at that radius. Although small, this inhomogeneity is sufficient to cause a detectable decay in rotational
rate when the magnet spins in vacuum.
A figure of note for the rotational decay of a bearing is the
coefficient of friction (COF), defined as the rotational drag
force divided by the levitational force (weight of the levitated
rotor). The COF for a mechanical roller bearing is of the order
of 103. The COF for an active magnetic bearing is about
104 when parasitic losses for the feedback circuits and power
for the electromagnets are factored in. Measured COFs for
simple superconductor bearings are as low as 107. The parasitic losses of a superconducting bearing are the power required to keep the superconductor cold. For refrigerators that
operate at about 30% of Carnot efficiency (the theoretical
maximum) about 14 W of electricity are required to remove 1
W of heat at liquid nitrogen temperatures. Thus, the equivalent COF for an HTS bearing is about 2 106, about two
orders of magnitude lower than the best alternative bearing.
Two magnetomechanical resonances occur in a magnetically suspended rotor with a polar moment of inertia greater
than the transverse moment (i.e., a disk geometry): one is vertical and one is radial. In practice, the vertical resonance has
a minimal effect on the COF in most superconducting bearing
systems. The radial resonance occurs when the rotational frequency is close to that of the natural radial frequency of the
rotors vibration. If the rotor has a transverse moment of inertia greater than its polar moment, then there is an additional
resonance, having the form of a conical vibration, that is, with
the top of the rotor moving to one side while the bottom of
the rotor moves to the opposite side. Because superconducting
bearings have low stiffness, the resonances occur at low frequencies on the order of several hertz. This, together with
the large clearances possible with superconducting bearings,
leads to a robust bearing system.
Figure 4 shows the COF as a function of rotational frequency for a cylindrical permanent magnet levitated 10 mm
above a single YBCO cylinder and spinning in a vacuum
chamber. One may divide the behavior into three regions: below the resonance, the radial resonance, and above the resonance. Below the resonance ( f 3 Hz), the losses are caused
Coefficient of friction
0.0010
0.00010
105
106
107
20
40
60
80
100
120
Rotational frequency (Hz)
Figure 4. Coefficient of friction versus rotational frequency for a 25.4

mm dia., 9.6 mm high cylindrical permanent magnet levitated 10 mm
above a YBCO cylinder.
733
by the inhomogeneity of the permanent magnets field. The

resonance region (320 Hz), shown in Fig. 4, is relatively
broad. In some systems, especially those with well-balanced
rotors, the resonance is very narrow. Above the resonance
(20 Hz), the losses are affected by an additional factor,
which is caused by the rotation of the magnet about its center
of mass rather than its center of geometry or center of magnetism. As shown in Fig. 4, above and below the resonance, the
losses are mostly velocity-independent. However, detailed
studies of losses in superconducting bearings show that small
velocity-dependent effects are present which are intrinsic to
the superconductors (1921).
Because of size limitations encountered when high-performance bulk superconductors are produced, a large bearing
system needs an array of superconductors. In the case of a
single magnet levitated over an array of superconductors, an
additional loss arises from magnetization of the individual superconductors upon levitation. The magnetization of the array
leads to an ac magnetic field seen by the rotating permanent
magnet and eddy current losses that depend on the electrical
conductivity of the permanent magnet.
Some alternative bearing concepts that involve superconductors exhibit even lower COFs. The Evershed-type hybrid
has a COF of just over 108 (22). The velocity-dependent
losses associated with superconducting arrays are also greatly
reduced with this bearing design. A mixed- (where is the
magnetic permeability) bearing (23) has a COF of just over
109. In this bearing, a soft ferromagnetic cylinder ( 1) is
levitated in attractive levitation between two permanent
magnets and stabilized by a superconductor ( 1) placed
between the rotor and each of the permanent magnets.
APPLICATIONS
The availability of superconducting bearings that are so
nearly friction-free naturally leads to their consideration for
flywheel energy storage. Flywheels with conventional bearings typically experience high-speed idling (i.e., no power input or output) losses of the order of about 1% per hour. With
superconducting bearings, it is believed that losses as little
as 0.1% per hour are achievable. When coupled with efficient
motors/generators and power electronics (capable of losses as
low as 4% on input and output), the potential exists for constructing flywheels with diurnal storage efficiencies of 90%.
Probably only one other technology is capable of achieving
such high diurnal storage efficiencies: large superconducting
magnetic energy storage, which employs superconducting
coils hundreds of meters in diameter.
Electric utilities have a great need for inexpensive energy
storage, such as flywheels, because their inexpensive baseload
capacity is typically underutilized at night and they must use
expensive generating sources to meet their peak loads during
the day. A distributed network of diurnal-storage devices
could also make use of underutilized capacity in transmission
lines at night and add robustness to the electric grid. These
factors are expected to become more important in the coming
deregulation of the electric utility industry. Efficient energy
storage would also be beneficial to renewable energy technologies, such as photovoltaics and wind turbines.
With modern graphite fiber/epoxy materials, the inertial
section of a flywheel rotates with rim speeds well in excess of
734
SUPERCONDUCTING MAGNETS FOR FUSION REACTORS
1000 m/s and achieves energy densities greater than those of

advanced batteries. The kinetic energy in a (large) Frisbeesized flywheel with this rim speed is about 1 kWh, and a person-sized flywheel could store 2040 kWh. Although design
concepts for flywheels that employ superconducting bearings
with up to 10 MWh have been proposed, the most advanced
experimental versions at present are in the 100 Wh to 1
kWh class.
Because superconducting levitation is versatile over a wide
range of stiffness and damping, it has been suggested for numerous applications. Superconducting bearings, like magnetic
bearings, do not require a lubricant, which could be a major
advantage in harsh chemical or thermal environments. Superconducting bearings are particularly interesting for cryogenic turbopumps. The low friction of the superconducting
bearing allows its use in high-precision gyroscopes and gravimeters. The hysteretic nature of superconducting levitation
has suggested its use in docking vehicles in space. As one
vehicle approaches another it would experience a decelerating
repulsive force. After the relative velocities have disappeared
at a small separation distance, the vehicles would experience
an attractive force if their distances tend to separate. The stable levitational force suggests application in magnetically levitated conveyor systems in clean-room environments where
high purity requirements mandate no mechanical contact.
Trapped-field HTSs have been suggested for constructing vehicle magnets to be used in electrodynamic levitation of highspeed trains (see MAGNETIC LEVITATION).
BIBLIOGRAPHY
1. F. C. Moon, Superconducting Levitation, New York: Wiley 1994.
2. J. R. Hull, guest editor, Applied Superconductivity, 2 (7/8): 1994.
Contains several state-of-the-art papers on superconducting levitation and HTS flywheels.
3. T. D. Rossing and J. R. Hull, Magnetic levitation, The Physics
Teacher, 29 (9): 552562, 1991.
4. E. H. Brandt, Rigid levitation and suspension of high-temperature superconductors by magnets. Am. J. Phys., 58: 4349, 1990.
5. J. R. Hull, Flywheels on a roll, IEEE Spectrum, 2025, July 1997.
6. S. Earnshaw, On the nature of molecular forces which regulate
the constitution of luminous ether, Trans. Cambridge Philos. Soc.
7: 98112, 1842.
7. W. Braunbek, Freischwebende Korper im elektrischen und magnetischen Feld, Zeit. Physik, 112: 753763, 1939.
8. V. Arkadiev, A floating magnet, Nature, 160: 330, 1947.
9. P. N. Peters et al., Observation of enhanced properties in samples
of silver oxide doped YBa2Cu3Ox, Appl. Phys. Lett. 52: 2066
2067, 1988.
10. W. G. Harter, A. M. Hermann, and Z. Z. Sheng, Levitation effects
involving high Tc thallium based superconductors, Appl. Phys.
Lett. 53: 11191121, 1988.
11. T. P. Sheahen, Introduction to High-Temperature Superconductivity, Chap. 14, New York: Plenum, 1994.
12. M. Murakami, Melt Processed High-Temperature Superconductors, Singapore: World Scientific, 1992.
13. S. A. Basinger, J. R. Hull, and T. M. Mulcahy, Amplitude-dependence of magnetic stiffness in bulk high-temperature superconductors, Appl. Phys. Lett., 57: 29422944, 1990.
14. J. R. Hull et al., Magnetic levitation and stiffness in melt textured Y-Ba-Cu-O, J. Appl. Phys. 72: 20892091, 1992.
15. P. Schonhuber and F. C. Moon, Levitation forces, stiffness and

force-creep in YBCO high-Tc superconducting thin films, Appl.
Supercond. 2: 523534, 1994.
16. Z. J. Yang, Lifting forces acting on magnets placed above a superconducting plane, J. Supercond., 5 (3): 259271, 1992.
17. A. A. Kordyuk, Magnetic levitation for hard superconductors, J.
Appl. Phys. 83 (1): 610612, 1998.
18. A. B. Riise, T. H. Johansen, and H. Bratsburg, The vertical magnetic force and stiffness between a cylindrical magnet and a highTc superconductor, Physica C, 234: 108114, 1994.
19. Z. J. Yang and J. R. Hull, Energy loss in superconducting bearing
systems, IEEE Trans. Appl. Supercond. 7: 318321, 1997.
20. A. A. Kordyuk and V. V. Nemoshkalenko, High-speed magnetic
rotor with HTS bearings for precision energy losses investigations, IEEE Trans. Appl. Supercond. 7: 928931, 1997.
21. C. E. Rossman, J. I. Budnick, and B. R. Weinberger, Correlation
of frictional losses of spinning levitated magnets with ac susceptibility in high-temperature superconductors, Appl. Phys. Lett. 70:
255257, 1997.
22. J. R. Hull, T. M. Mulcahy, and J. F. Labataille, Velocity dependence of rotational loss in Evershed-type superconducting bearings, Appl. Phys. Lett. 70: 655657, 1997.
23. J. R. Hull et al., Low friction in mixed-mu superconducting bearings, J. Appl. Phys. 78: 68336838, 1995.
JOHN R. HULL
Argonne National Laboratory
SUPERCONDUCTING MAGNETS. See MAGNETIC REFRIGERATION;

AND IMAGING.
MAGNETS FOR MAGNETIC RESONANCE ANALYSIS
734
SUPERCONDUCTING MAGNETS
FOR FUSION REACTORS
The magnetic confinement of plasma is the most promising
option to use controlled nuclear fusion as a power source for
future generations. A number of different magnetic field configurations have been proposed to achieve plasma ignition, all
requiring high field strength over a large volume. Most of the
experimental machines use conventional, copper windings operated in pulsed mode, to investigate the plasma physics. The
advanced plasma experiments, as well as the future fusion
reactors, call for long confinement time and high magnetic
field, which can be reasonably maintained only by superconducting coils.
Unlike other applications of superconductivity, for fusion
magnets there is no normal conducting alternative: whenever a magnetic confinement fusion power plant will operate,
it will have superconducting windings. For this reason, fusion
magnets are an important, long-term factor in the market of
superconducting technology. Today, for NbTi-based conductors, fusion is a nonnegligible share of the market, with over
50 t of strand recently used for the LHD and about 40 t committed for W7-X. For Nb3Sn technology, two large devices, the
T-15 tokamak and the ITER model coils, have used most of
the conductor ever produced (each about 25 t of strand), being
the driving input for the development of high performance
Nb3Sn strands.
735
Table 1. Summary of Superconducting Magnet Systems for Fusion Devices
Tokamak T-7
Tokamak T-15
MFTF (all coils)
TRIAM
Tore Supra
LHD-Helical (2 coils b )
LHD-Poloidal (6 coils)
Wendelstein 7-X
Strand
Weight (t)
Conductor/
Cooling a
Stored
Energy (MJ)
Peak
Field (T)
Operating
Current (kA)
1
25
74
2
43
10
43
37
NbTi/FF
Nb 3 Sn/FF
Nb 3 SnNbTi/pool
Nb 3 Sn/pool
NbTi/pool 1.8 K
NbTi/pool 4.5(1.8)K
NbTi/FF
NbTi/FF
20
795
1 000
76
600
930 (1 650)
1 980
600
5
9.3c
212.75
11
9
6.9 (9.2)
56.5
6
6
5.6c
1.55.9
6.2
1.4
13 (17.3)
20.831.25
16
FF forced flow.
Operation at superfluid helium is planned at a later stage.
c
Design values, achieved on single coil test.
a
b
The first use of superconducting coils in experimental fusion devices dates back to the mid-1970s. In the last twentyfive years, six sizable devices for magnetic plasma confinement have been built with superconducting coils (see Table
1): T-7 and T-15 in the former Soviet Union, MFTF in the
United States, TRIAM and LHD in Japan, and Tore Supra in
France. In Germany, Wendelstein 7-X is under construction.
Moreover, a number of developmental and prototype coils
have been tested in the scope of large international collaborations (large coil task, demonstration poloidal coils, ITER
model coils).
The operating requirement for fusion magnets may vary
over a broad range, depending on the kind of confinement and
the size of the device (1), for example, from medium-field,
pure dc mode in the helical coils of the stellarators, to the
high-field, fast rate in the central solenoid of the tokamaks.
There is no general recipe for the magnet design, but a few
common issues can be identified. Long-term reliability calls
for a conservative component design and generous operating
margins. The maintenance by remote handling in a nuclear
environment imposes strong restrictions to either repair or
replacement of individual parts. Safety regulations are also a
major issue for superconducting magnets in a fusion reactor:
the design must account for any likely or less likely failure
mode of the coil system and provide that it will not turn into
a nuclear-grade accident. Last but not least, the cost of the
magnets, which is a large fraction of the reactor cost, must be
contained to be commercially competitive with other power
sources.
Only low-temperature superconductors have been considered to date for use in fusion magnets at field amplitudes up
to 13 T. A substantially higher field, which would make attractive the use of high-temperature superconductors, is not
likely to be proposed as the electromagnetic loads, roughly
proportional to the product of field, current and radius, already set a practical limit for structural materials. It may
sound surprising that the actual superconducting material
cross-section is mostly smaller than 5% of the overall coil
cross-section. The choice between NbTi and Nb3Sn conductors
is dictated by the operating field. The upper critical field of
NbTi conductors is 10 T at 4.5 K and 13 T at 1.8 K.
According to the design current density and the temperature
margin, the operating field is set at least 3 T to 4 T below
the upper critical field. In the conservatively designed fusion
magnets, the peak field for NbTi conductors is up to 9 T for
coils cooled by a superfluid helium bath (e.g., Tore Supra and

LHD helical coils) and up to 6 T for supercritical helium
forced flow (e.g., W7-X and LHD poloidal coils). At a higher
operating field, the choice of Nb3Sn conductors is mandatory
to obtain adequate temperature margins and high current
density. The increasing confidence in Nb3Sn technology, as
well as its slowly decreasing cost, tends to move down the
field threshold for the NbTi versus Nb3Sn. Conductors based
on Nb3Al are in a developmental stage and may become an
alternative to Nb3Sn for selected high-field magnets (e.g., the
D-shaped toroidal field coils), because of the better tolerance
to bending strain.
The winding packs may be either potted in epoxy resin or
laid out as a spaced matrix of noninsulated conductors in a
liquid helium bath. This last option offers the advantages of
constant operating temperature and potential high stability
due to the bath-cooled conductor surface. The drawbacks are
the poor stiffness of the winding and the limited operating
voltages (the insulation relies on the helium as dielectric).
The potted coils with forced-flow conductors have superior
mechanical performance and may operate at higher voltage:
as a rule of thumb, they become a mandatory option for stored
energy in excess of 1 GJ to 2 GJ. The cable-in-conduit conductors became, in the last decade, the most popular option for
forced-flow conductors because of the potential low ac loss and
the good heat exchange due to the large wet surface. To withstand the mechanical and electromagnetic loads, the coils are
fitted in thick-walled steel cases, either welded or bolted.
More structural material may be added, if necessary, both in
the conductor cross-section and in winding substructures, for
example, plates and cowound strips.
The magnetic stored energy is very large, up to 130 GJ for
the proposed magnet system of ITER. In case of a quench
(local transition from superconducting to normal state), the
stored energy must be dumped into an outer resistor to avoid
an overheating and damage of the winding. A large operating
current is needed to reduce the number of turns, that is, the
winding inductance, and extract quickly the stored energy at
a moderatly high voltage (up to 10 kV to 20 kV). The operating current density in the superconducting cross-section
(NbTi or Nb3Sn filaments), Jscop, is selected according to the
specific design criteria to be a fraction of the critical current
density, Jc, at the highest operating field. Typically, Jscop is in
the range of 200 A/mm2 to 700 A/mm2, and Jscop /Jc 0.3
736
REVIEW OF SUPERCONDUCTING COILS FOR FUSION
a square monolithic NbTi conductor, 6.35 6.35 mm2: the

peak field in the pool boiling cooled winding is 7.5 T at 2.4
kA, with a stored energy of 12 MJ.
The Mirror Fusion Test Facility (MFTF) (see Fig. 1) was
assembled at Lawrence Livermore National Laboratory in
1985 (3). It is the largest set of superconducting magnets for
fusion, with a total mass 1200 t and about 75 t of strand. It
consists of 8 C-type coils, 12 low-field, and 4 high-field solenoids, including the A2 coils with a Nb3Sn insert (see Fig. 2).
All magnets are pool cooled at 4.5 K by natural convection,
with two-phase coolant outlet (5% gas). The coils are all
wound in the coil case, which acts as a cryostat, with ground
insulation applied before winding. In the Yin-Yang coils, the
winding form is fitted into the thick case by a copper bladder
filled with urethane. The turn insulation is provided by G10
spacers, the layer insulation by perforated G-10 plates. The
conductors for the solenoids (two types of NbTi and one
Nb3Sn) consist of a thick multifilament composite soldered to
the copper stabilizer (see Fig. 3). The joints are made by coldwelding of the composite. For the Yin-Yang coils, the conductor is a square NbTi monolith with a perforated Cu strip
wrapped around and soldered to increase the wetted surface.
The conductor for the pancake-wound Nb3Sn insert, A21, is a
flat multifilamentary composite soldered in the Cu housing
after heat treatment (react and wind): the outer copper surface is oxidized to prevent solder wetting. All the conductors
are designed to be cryostable. The fraction of Iop /Ic is always
smaller than 2/3. After cool-down and successful commissioning of the magnet system, the project was discontinued in
1985. In 1990, the A2 coils were extracted and reassembled
in the FENIX conductor test facility, which operated for three
years at field levels as high as 13 T.
Mirror Devices
Tokamaks
In the last decade interest in the magnetic confinement in

linear machine (mirror fusion) has strongly declined, but they
were very popular in the 1970s. At the Kurchatov Institute in
Moscow, a plasma trap named LIN-5 was built in 1970 by a
split solenoid system with 0.2 m inner diameter and 5.8 T
peak field at 1 kA. In 1975, for the LIN-5B machine, a 5
tonne, bath-cooled baseball type coil was wound with 6 km of
square, monolithic NbTi conductor, 6.2 6.2 mm2 (2): at an
operating current of 2 kA (75% of the short sample current),
the peak field at the conductor is 5.6 T. The basic coil and
conductor design is very similar to the US Baseball II-T winding (see below).
In the United States, a superconducting magnetic mirror
apparatus (SUMMA) consisting of four coils with 0.9 m outer
diameter, 8.8 T peak field, and 18 MJ stored energy was built
by NASA in the early 1970s. At Livermore, the first superconducting baseball coil (Baseball II-T) was wound in 1971 with
The toroidal arrangement of plasma is the most promising

confinement geometry, with the largest number of experimental devices, including the pinch and reverse pinch, the stellarator group of machines, and the tokamaks, where a toroidal
plasma current is initiated and sustained by a pulsed ohmic
heating coil (central solenoid). Most of the superconducting
magnets for fusion belong to the tokamak family, including
four plasma experiments (T-7, T-15, TRIAM, Tore Supra) and
a number of sizable technology demonstration devices (LCT,
TESPE, DPC, Polo, ITER Model Coils).
The very first superconducting tokamak, named T-7, was
built at the Kurchatov Institute, Moscow in 19751976 and
first cooled down in 1977 (see Fig. 4). The toroidal winding
system (4) consists of 48 circular double pancakes, in aluminum case, with 60 turns in each coil. The average coil diameter is 1 m, the overall cold mass is 12 t and the stored energy
20 MJ at the nominal operation point. The torus was preas-
Figure 1. The Yin-Yang coils being assembled at one end the mirror
fusion test facility (courtesy of C. H. Henning, Lawrence Livermore
National Laboratory).
A/mm2 to 0.6 A/mm2. The current density over the coil crosssection is over one order of magnitude smaller.
In the nonsteady-state tokamak machines, the normal operating cycles and the occasional plasma disruption set additional, challenging requirements, in terms of mechanical fatigue of the structural materials and pulsed field loads on the
superconductors.
A1
West
Figure 2. The Axicell coil configuration

for the MFTF, with the Yin-Yang coils at
the ends.
East
A20
T1
S1 S3 S5
S2 S4 S6
A21
T2
M1
M2
737
12.4
3
11
1.8
5
2.1
12.5
6.5
sembled into eight segments, individually tested before final

assembly. The forced-flow conductor (see Fig. 5), is made from
a strip of nine copper pipes, 2 mm inner diameter, with 16
multifilamentary and 32 single-core NbTi strands sitting in
the grooves between the pipes and bonded to them by electroplating a 0.6 mm copper layer up to the final size of 28
4.5 mm. The cooling is by two-phase helium at 4.5 K, with all
the pancakes connected in parallel. A 15-mm-thick copper
shell sorrounds each coil and acts as an eddy currents shield
for the poloidal field variations. The strands are not transposed and the conductor suffered from severe flux jumps, triggering quenches, during ramp up and ramp down. However,
80% of the design current (6 kA at 5 T peak field) was
achieved. Operation of T-7 was discontinued at the Kurchatov
Institute in 1987. Later, T-7 was transferred to the Chinese
Institute of Plasma Physics, where it has been operating
since 1996.
The toroidal field coils of the T-15 tokamak, first operated
in 1988 at the Kurchatov Institute in Moscow, are the largest
worldwide application of Nb3Sn conductors (see Fig. 6) (5).
The 24 circular coils, with average diameter 2.4 m, consist
each of 12 single pancakes, cooled in parallel, with the He
inlet at the inner radius joints. The turn insulation is obtained by wet winding to balance the uneven conductor contour. Two stacks of six pancakes are vacuum impregnated in
two-halves steel cases, eventually bolted together. The react
and wind method was applied. The forced-flow conductor (see
Figure 4. The 24 double pancakes of T-7 in the final assembly (courtesy of V. Keilin, Kurchatov Institute).
5.2
6.9
Figure 3. Conductors for the MFTF magnets, from left to right: NbTi square conductor with enhanced wet surface for the M, T,
A1, and A20 coils, NbTi conductor for the S
solenoids, react and wind Nb3Sn conductor
for the A21 insert. (See Fig. 2 for coil identification.)
Fig. 5), is a flat cable of 11 nonstabilized Nb3Sn strands,

bonded after heat treatment to two copper pipes by an electroplated Cu layer, 1.2 mm thick. Over 100 km of conductor have
been manufactured, in units of 200 m. Each coil was tested
before assembly and achieved the specification, although a
steady-state voltage was observed, by far larger than the joint
voltage, in the range of 2.5 mV to 10 mV/coil. The design current in the tokamak is 5.6 kA at 9.3 T peak field. The limited
size of the cryoplant (allowing a mass flow rate of only 0.38
g/s conductor) and the large radiation loss limited the operating temperature to the range of 9 K to 10 K. The highest
operation point was 3.9 kA at 6.5 T, 7 K to 8 K, in agreement
with the single-coil test and in excess of the original strand
specification. The field transients due to plasma disruption,
up to 40 T/s, were withstood without quench, with an increase of the outlet temperature by 0.25 K. The design ground
voltage is 1.5 kV. For fast discharge, 250 V was applied at the
terminals, with a time constant of 104 s.
The TRIAM device at the University of Kyushu, Fukuoka
(Japan), is a compact, high-field tokamak, first operated in
1986. The superconducting 16 D-shaped toroidal field coils,
with 3 m average perimeter, are cooled by a pressurized
liquid helium bath at 4.5 K. The poloidal field coils, wound
with normal conductor, are placed inside the TF coils. The
magnet cold mass is 30 t. The toroidal field conductor (see
Fig. 7), consists of a large Nb3Sn bronze composite (10.5 3.3
mm for high grade, over a half million filaments) soldered
after heat treatment in a copper housing with roughened side
surfaces to improve the heat exchange. Beside the copper
housing (RRR 90), a Cu-clad high-purity Al profile (RRR
3000) is used as a stabilizer. Each coil is a stack of six double
pancakes, housed in a steel case. Three conductor grades,
with the same width and decreasing height, are used with
soldered joints for the double pancake. The conductor is designed to be cryostable and can withstand an energy input up
to 7.8 J/cm3 at the operating conditions: in case of plasma
disruption, a normal zone may locally occur but is recovered
within 0.3 s. At 6.2 kA, 11 T, the ratio Iop /Ic is 0.6. No quench
event has been reported after three years of operation, including plasma disruptions (6).
The tokamak Tore Supra has been assembled at Cadarache, France, in 1987 (7). The poloidal coils are wound from
copper conductors. The 18 pool cooled circular TF coils, wound
from NbTi superconductor operating at superfluid helium, are
the largest magnet mass, 160 ton, cooled at 1.8 K. Each coil
is made out of 26 double pancake, with an average diameter
of 2.6 m (see Fig. 8). A cowound prepreg tape, 0.15 mm thick,
is used for the turn insulation. The pancake spacers, 2.2 mm,
are built by a perforated prepreg thin plate with glued glass
738

Electroplated Cu
Nb3Sn strand, = 1.5 mm
Cu pipe, = 2 mm
Cu pipe, = 4 mm
28 mm
6.5
mm
4.5 mm
Figure 5. The forced flow conductors for

T-7 (left) and T-15 tokamaks, bonded by
copper electroplating.
Single core NbTi

17.4 mm
NbTi multifilament strand, = 0.85 mm
epoxy bottoms. The ground insulation is obtained by overlapped prepreg plates. A 2-mm-thick steel case is shrink-fitted to the winding and contains the atmospheric, 1.8 K He
bath. A thick steel case, with thermal insulation, is shrinkfitted and cooled at 4.2 K. The conductor is a rectangular
NbTi/Cu/CuNi multifilament composite, 2.8 5.6 mm,
wound on the short edge to minimize the ac loss from the
poloidal field variation. The temperature margin is 2.5 K,
with Tcs 4.25 K at 1400 A, 9 T peak field. Little copper cross
section is used in the conductor: for stability, the He bath
enthalpy up to the point is available, due to the very high
thermal conductivity of He II. The heat exchanger is placed
underneath the coil case, open only at the bottom. In case of
quench, a He gas pressure builds on the top of the case/cryostat, and siphons the whole He volume within 3 s through the
bottom opening, providing a very fast quench propagation and
limiting the hot spot temperature below 80 K. In 1988, about
six months after first operation, an interpancake short occurred at one coil during a fast discharge, with 1.5 kV across
the coil and 60 V across pancakes, well below the expected
Paschen minimum for helium. The damaged coil was later
replaced with a spare coil and the dump voltage was decreased to 500 V in order to limit the pancake voltage to 20
V, which was experimentally assessed as the safe threshold
to avoid interpancake discharge. The poloidal field variations
and the plasma disruption result in a temperature increase
in the He II bath as small as 0.01 K.
Besides the four above described tokamaks, a number of
prototype superconducting coils have been built under na-
Figure 6. The 24 Nb3Sn coils of T-15 assembled with the 12 horizontal ports (courtesy of V. Keilin, Kurchatov Institute).
tional or international auspices. The demonstration coils provide valuable opportunities to learn about magnet and conductor technology. In such projects, the pressure for a
conservative design is less strong and the performance margins can be better explored than in a plasma experimental
device. In the IEA Large Coil Task at the Oak Ridge National
Laboratory (8), six large D-shaped magnets, 3 3.5 m bore,
have been built to the same common specification using substantially different design approaches (see Table 2). The coils,
assembled as a tokamak (see Fig. 9), operated (19841985) at
the same design point (8 T peak field) with margins ranging
from 120% to 140%. For the first time, a cable-in-conduit
Nb3Sn conductor (react and wind coil manufacture) was used
in a large-scale application and, despite the broad resistive
transition observed in selected coil sections (similar to T-15
behavior), the coil reached 8 T with Tcs 8 K. The cryostable
conductors for the bath-cooled coils (GD, GE, JA) could be easily graded (both layer and pancake windings) and, using soldered copper profiles as stabilizers, achieved impressive results in terms of effective use of strand: from 1.4 t of strand
in the GE conductor to 8.2 t in the forced-flow conductors of
EU (NbTi) and WH (Nb3Sn), cooled by supercritical helium at
3.8 K, 10 bar to 15 bar. However, two out of three pool cooled
coils could not be dumped to the design voltage of 1 kV. The
nuclear heat load was simulated by heaters on the inner radius. The poloidal field coil variations were reproduced by a
pulsed coil traveling inside the torus: the pulsed field test,
with B 0.1 T, B 0.14 T, to 1 s, could be completed
only for three coils (JA, CH, EU). Over 90% of the stored energy could be dumped in the external resistors, except for the
WH coil, with short circuited radial plates.
At the same time of the LCT project, two small-size experiments with D-shaped coils were carried out at the Forschungszentrum Karlsruhe (Germany) and at Toshiba (Japan).
The six toroidal coils of TESPE at Karlsruhe (9) have a 0.5
0.6 m bore and are pool-cooled at 4.2 K (8 t total cold mass).
The coils are wound as double pancakes, shrink-fitted in steel
housings insulated by glass epoxy laminate. The steel case is
electron-beam welded and serves both as liquid helium container and mechanical reinforcement. The conductor is a soldered flat cable of 24 multifilament NbTi strands 1.45
mm, operating at 7 kA, with a peak field of 7 T. The TESPE
torus was first operated in 1984 with a test program focused
on mechanical load and high-voltage safety issues. The double
pancake built by Toshiba (10) in 1983 had a Nb3Sn cable-inconduit conductor, 18.3 15.7 mm, with 486 strands, encased
into a 1-mm-thick 316L steel jacket. The D-shaped coil, 1.1
0.9 m, was wound after heat teatment and tested at 10 kA,
EB welded,
Cu housing
Al stabilizer
Nb3Sn composite
Cu housing
739
CuNi cladded
Al stabilizer
18
14
7.1
3.3
12.5
10.5
NbTi strands
in a peak background field of 10 T, provided by a small split

coil. The test included hydraulic friction factor and stability.
The most recent development project for tokamaks toroidal field coil is the ITER TF Model Coil, to be tested at the
Forschungszentrum Karlsruhe by the end of 1999 in the background field of the EU-LCT coil (11). The race-track-shaped
coil, with 2.5 1.4 bore, has an overall weight of 31 t and a
stored energy of 60 MJ at 70 kA (compared to 400 t and 4
GJ in an individual full size ITER coil). The peak field is
about 9 T, compared to 12.5 T in ITER. The main goal of the
ITER TF Model Coil is to demonstrate the winding technique,
which uses precision-machined steel radial plates where the
pancake wound Nb3Sn cable-in-conduit conductor is encased
after the heat treatment (react and transfer method).
The Demonstration Poloidal Coil (DPC) project at JAERI,
Naka (Japan), aimed at comparing conductor design options
Channels 4.2 K
Thick
case
Polyimide-alumina
chocks
Thin case, 1.8 K
Ground insulation
Glass-epoxy chocks
Perforated pancake spacers

Superconducting pancake
Figure 8. Winding pack layout for the pool-cooled, NbTi toroidal field
coils of Tore Supra (courtesy of B. Turck, Tore Supra).
PbSn solder
Figure 7. Soldered monolithic conductors, stabilized with high-purity aluminum profiles, for
TRIAM (left) and the helical coils of LHD (right).
for pulsed-field coils. Three Nb3Sn react and wind winding

models with av 1.3 m, DPC-EX (12), US-DPC (13) and
DPC-TJ (14) were tested in the DPC facility in pulsed mode
up to about 7 T to 8 T, sandwiched between two pulsed NbTi
solenoids (DPC-U), connected in series. The conductors are all
forced-flow cable-in-conduit (see Fig. 10). The jacket material
is Incoloy for the US-DPC (used for the first time). The DPCTJ had a double jacket: the outer one is 3-D machined without
bending and fitted by spot-welding to the conductor after the
heat treatment. The strand surface is bare copper for the
DPC-TJ, and the coupling loss is 1000 times larger compared
to the US-DPC and DPC-EX conductors, with Cr plated
strands (2 ms). Ramp rate limitation was observed in the USDPC, probably due to transposition errors in the cable. The
NbTi strands in the cable-in-conduit of the background coils
were insulated to avoid interstrand coupling loss: the conductor turned to be unstable due to the inability to redistribute
effectively the current among the strands.
The Polo coil, a NbTi circular winding with 3 m, has
been tested in 1994 at the Forschungszentrum Karlsruhe
(Germany) (15). The cable-in-conduit conductor (see Fig. 10),
has two separate hydraulic circuits: stagnant, supercritical
He at 4 bar in the annular cable region and forced flow, 2 g/
s, two-phase He at 4.5 K in the central pipe. This design
allows a homogeneous temperature along the conductor with
a small pressure drop. The strand is a NbTi/Cu/CuNi composite and the subcables have CuNi or insulating barriers, resulting in very low coupling loss, 210 s. Four stainlesssteel corner profiles are laser-welded around the cable. Phase
resolved partial discharge was first used at 4 K to assess the
integrity of the glass-epoxy insulation. A midpoint electrical
connection in the winding enables to create very-high-field
transients in a half coil by a fast discharge of the other half
coil. The coil has been tested up to 15 kA, 3.6 T. A degradation of Ic by 30% in dc operation has been observed compared
to the strand performance. Polo does not have ramp rate limitation, the stability criterion being the only limiting criterion.
Very-fast field transient, up to 1000 T/s are withstood without
quench. High-voltage operation, up to 23 kV, has been demonstrated.
The most recent pulsed-coil development for tokamak is
the ITER CS Model Coil (16), a layer-wound, two-in-hand solenoid to be operated at 46 kA, 13 T with 0.4 T/s field rate,
scheduled to be tested at Naka (Japan) in 1999. The stored
energy is 641 MJ, compared with 13 GJ in the full size central
solenoid. The conductor is a Nb3Sn cable-in-conduit with a
740
Table 2. Summary of the LCT Coil Characteristics

GD
GE
Winding type
14 layers
3 grades
7 double pancake, 3 grades
Cooling method
Conductor
Pool boiling
Soldered flat
cable, on edge
Pool boiling
Divided flat cable
Non-Cu J op
Winding J op
He inventory
Total weight
SC strand
Test voltage
586 A/mm 2
27.4 A/mm2
1320 1
43.9 t
2t
2 kV
525 A/mm2
24.7 A/mm2
1735 1
38.6 t
1.4 t
2.5 kV
WH
CH
EU
JA
4-in-hand, 12
double pancakes
Forced flow
Nb 3 Sn cable-inconduit
22 pancakes
2-in-hand, 7 double pancakes
20 double pancake, 2 grades
Forced flow
Square, soldered
cable
Pool boiling
Soldered flat
cable, on edge
265 A/mm2
20.1 A/mm2
440 1
33.7 t
8.2 t
9.2 kV
302 A/mm2
30.1 A/mm2
110 1
41.7 t
3.5 t
10 kV
Forced flow
Divided, flat
cable-inconduit
393 A/mm2
25.7 A/mm2
663 1
39 t
8.2 t
12 kV
thick-walled Incoloy 908 jacket, manufactured as extruded

and drawn-down bars, assembled by butt-welding into a oversize pipe, up to 250 m long: the strand bundle is pulled into
the jacket and eventually rolled to the final size. The conductor is insulated by prepreg glass fabric with interleaved kapton foil, applied after the heat treatment by controlled unspringing of the individual layers.
Stellarators
The plasma confinement can be achieved in stellarators by a
number of winding configurations. The two large projects using superconducting coils are the Large Helical Device (LHD)
at Toki, Japan, and the Wendelstein VII-X (W7-X), to be assembled at Greifswald, Germany. The superconducting magnet system of LHD operates in dc mode and consists of a two
helical coils (17) wound around the toroidal vacuum vessel
and three sets of circular poloidal coils (18). The helical coils
are pool-cooled: initial operation will be at 4.2 K, 6.9 T, 13 kA
and, at a later stage, at 1.8 K, 9.2 T, 17.3 kA. A precision tool
with 13 numerically controlled driving axes has been used to
wind in situ the two helical coils (see Fig. 11). The turn and
327 A/mm2
26.6 A/mm2
1425 1
39 t
2.6 t
3 kV
layer insulation is provided by glass epoxy spacers graded

across the winding pack to provide the best mechanical support in the stressed area and the largest wet conductor surface at the peak field (the highest field and the highest mechanical stress are not at the same winding location). The
conductor (see Fig. 7), is a NbTi flat cable soldered to a CuNicladded Al stabilizer into a copper housing, eventually sealed
by two electron beam welds. The conductor is designed to be
cryostable, with Iop /Ic 0.55. The forced-flow conductor for
the poloidal coils (see Fig. 10), is a NbTi cable-in-conduit with
486 strands ( 0.76 mm or 0.89 mm), 38% void fraction. To
improve the current sharing among strands, the surface is
not coated, with a coupling loss constant of 300 ms. The temperature margin is 1.2 K to 1.6 K, Iop /Ic 0.33. The joints
between the conductor sections are realized by filament joining, resulting in 0.14 n resistance at full current. The OV
coil, 11.5 m (see Fig. 12), has been wound on the site,
with prepreg turn insulation. The first operation of the LHD
was successfully started in March of 1998.
The magnet system of the W7-X torus consists of 50 nonplanar and 20 planar coils assembled in five modular segments (19). A first nonplanar model coil was completed in
1997 and tested in 1998. The completion of the machine is
scheduled by the year 2002. The forced-flow conductor to be
used for all the coils (see Fig. 10), is a NbTi cable-in-conduit
with 243 strands, 0.57 mm, 37% void fraction. The square
jacket is made of a hardenable Al alloy, coextruded around
the cable: it is soft after extrusion and during the winding
process. After hardening at 170C, it provides the required
stiffness to the winding pack. The temperature margin is
1 K and Iop /Ic 0.5.
CRUCIAL ISSUES FOR THE DESIGN OF FUSION MAGNETS
Electrical insulation
Figure 9. The six D-shaped LCT coils assembled as a torus in the

vacuum tank (courtesy of M. Lubell, Oak Ridge National Laboratory).
The very high level of neutron and gamma radiation at the

vacuum vessel of a fusion reactor must be screened to limit
the nuclear heat load and the radiation damage at the winding components. The size of the shield (up to 1 m thick in the
ITER project) may have a substantial impact on the size and
cost of the superconducting magnets. After screening, the radiation damage on the metallic components of a superconducting coil (steel, copper, NbTi and Nb3Sn), is not critical
and partly recovers (e.g., for copper) upon warming up at
741
DPCEX (1988)
LCTWH (1981)
USDPC (1988)
DPCTJ (1988)
Toshiba (1983)
Polo (1987)
W7X (1996)
20 mm
LHDOV (1994)
ITER CS (1995)
room temperature. The actual weak link for radiation damage

is the organic fraction of the electrical insulation. In potted
windings, the glass-epoxy is broadly used, either as laminates
or prepreg wraps or vacuum-impregnated fabrics, to bond together the winding turns and to provide the required dielectric strength. The neutron and gamma act on the long molecular chain of the resin, irreversibly breaking the atomic links.
The mechanical strength of the composite, mostly the shear
strength, is affected and macroscopic cracking may occur under operating loads, eventually leading to a short circuit between winding sections. To limit this risk, the magnets must
be designed to have low stress in the insulation, that is, limit
the risk of crack propagation. Another design approach is to
separate the mechanical and electrical functions, for example,
including a redundant electrical insulation layer, either interleaved or overlapped to the glass-epoxy, to stop the crack
propagation in the resin. The free radicals originated from the
broken organic polymers are chemically active and evolve into
gaseous molecules. The most severe consequence of radiation
induced chemical reactions at 4 K is the accumulation of frozen gas bubbles (mostly hydrogen). Upon warming-up of large
windings, the internal pressure of the evolved gas increases
dramatically due to the little permeation and may lead eventually to swelling in the insulation (20). A possible cure
against postirradiation hazards of organic insulation is to re-
Figure 10. Selection of cable-in-conduit superconductors, drawn to the same scale. The strands
of Polo, LHD-OV and W7-X are NbTi, all the
other are Nb3Sn strands. The jacket material is
steel except for ITER (Incoloy) and W7-X (aluminum alloy).
duce the resin volume fraction and select the resin composition to minimize the gas evolution rate. On the other hand,
there is a broad reluctance to start an expensive and timeconsuming task for the industrial development of innovative
insulation systems, which will be actually needed only when
a fusion reactor will work at full power on a time scale of
several years. The full replacement of organic insulation systems by ceramic materials with adequate mechanical properties may be the ultimate, long-term goal to solve the issue of
the electrical insulation in the heavily irradiated fusion
magnets.
Quench Protection
In case of quench, the huge amount of energy stored in a fusion magnet must be actively dumped in an outer resistor. If
a quench fails to be detected, the ohmic power locally dissipated in the slowly expanding normal zone is sufficient within
one minute or less to melt the conductor and start a chain of
serious failures (vacuum break, electric arc, mechanical collapse). A number of quench detectors have been developed
and are currently applied in superconducting magnets, from
the easy ones (voltage balance of different winding sections,
monitoring of outlet mass flow rate) to the most sophisticated,
including the laser interference on optical fibers used as distributed thermometer, transmission, and reflection of super
high-frequency waves in the coolant channel, acoustic emission, magnetization change at the normal zone (21). However,
a redundant and intrusive instrumentation is not welcome in
a fusion reactor, as it may increase the risk of leaks and insulation failure, due to the large number of feedthrough required. Whatever the quench detector is, the ultimate question always arises: What happens if the active quench
protection fails? The design approach for an actual fusion
magnet (i.e., not for an experimental device) will need to offer
both a reliable and robust quench-protection system and a
conductor/magnet layout that intrinsically limits the damage
in case of failure of the protection system, for example, enhancing the quench propagation and the enthalpy at intermediate temperature.
Cost Optimization
Figure 11. Winding tool with 13 numerically controlled axes for the
helical coils of the LHD (courtesy of K, Takahata, NIFS).
In several applications of the superconducting magnets (e.g.,

accelerators, detectors, high-field magnets, prototypes), the
achievement of the technical goal is the main care of the de-
742
Figure 12. The OV poloidal field coil of

the LHD (courtesy of K. Takahata, NIFS).
signer, while the cost of the device does not play a major role.
However, after completion of the demonstration phase for the
fusion magnets, the cost optimization will be a key issue for
the commercial success of fusion. On one side, the behavior of
the superconductor needs to be mastered by the designer (e.g.,
ac loss, stability, mechanical properties), in order to set the
design margins at a safe but realistic level and make effective
use of the expensive superconductors. On the other hand, the
choice of the manufacturing methods and tooling may have a
very strong impact on the cost of the coil and should be included as a driving factor in the design. Two examples are
given to show how a design choice may affect the cost.
A high electrical conductivity material (stabilizer) needs to
be added to the superconductor cross-section, to allow effective current-sharing and fast recovery for small thermal disturbances. The required stabilizer cross-section may be much
larger than the superconductor. In cable-in-conduit conductors, the straight choice is to equally distribute the stabilizer
cross-section in each superconducting strand, specifying a
high Cu : non-Cu ratio. However, the cost of the Nb3Sn strand
is independent of the copper ratio. If the designer masters the
mechanism of the current-sharing among strands and knows
the operating values of the interstrand resistance, he or she
may select a much smaller Cu : non-Cu ratio in the Nb3Sn
strand and add extra copper wires in the strand bundle.
Keeping the same superconductor cross-section, that is, without affecting the operating margins, the amount of Nb3Sn
strand can be significantly reduced with a large cost saving.
A Nb3Sn conductor needs a heat treatment at 650C to
700C to form the brittle intermetallic composite by solidstate diffusion. If the designer does not master heat resistant
electrical insulation systems, he or she will conservatively
choose to first heat-treat the conductor and then insulate it
and wind in the final shape (e.g., react and wind or wind and
react and transfer methods). As the Nb3Sn after heat treatment is degraded for permanent deformation as large as 0.2%
to 0.3%, the handling for post-heat treatment insulation and
final assembly requires sophisticated tooling and continuous
adjustment (e.g., shimming of each turn) to achieve the required tolerance with minimum strain on the conductor. If a
reliable insulation system is selected, compatible with the
heat treatment procedure, the coil can be wound in the final
form and to the final tolerance before the heat treatment
(wind and react method), saving the cost of a large number of

tools and manufacturing steps and avoiding the risk associated to the post-heat-treatment handling.
Risk and Quality Assurance
Large superconducting magnets are usually unique items for
which a thorough quality assurance program cannot be conveniently established in advance, as is the case for series production, due to the lack of iterative improvements in the manufacturing procedures. The global acceptance tests of the
magnets tend to replace the quality assessment of the individual procedures, achievable only on the basis of a broad statistical database. However, an individual acceptance test of a
fusion magnets cannot reproduce all the actual operating conditions, for example, mechanical load and peak field from
other coils, nuclear radiation, mechanical and thermal cycling. Moreover, a global acceptance test should not be pushed
to the failure limit, that is, the operating margin cannot be
assessed. The lack of confidence may push the designer in a
circle of overconservative choices, for example, assuming minimum performance for material properties, welds, assembly
tolerances, which adversely affect the cost and the effectiveness of the design. To avoid this trend, the designer should
identify the critical area for quality assurance and select a
low-risk design and procedure. For example, the resistance of
a joint between conductor sections cannot be checked during
the manufacture. In this case, the designer should aim for a
joint layout where the resistance performance is only marginally affected by incorrect assembly procedure.
BIBLIOGRAPHY
1. E. Teller (ed.), Fusion, New York: Academic Press, 1981.
2. N. A. Chernoplekov, Superconducting magnet systems for plasma
physics research in the USSR, Proc. Magnet Technology Conf. MT6, 3 Bratislava, Czechoslovakia, 1977.
3. T. A. Kozman et al., Magnets for the mirror fusion test facility:
Testing of the first Yin-Yang and the design and development of
the other magnets, IEEE Trans. Magn., 19: 859, 1983.
4. D. P. Ivanov et al., Test results of tokamak-7 superconducting
magnet system (SMS) sections, IEEE Trans. Magn., 15: 550,
1979.

5. E. N. Bondarchuk et al., Tokamak-15 electromagnetic system:
Design and test results, Plasma Device Oper., 2 (1), 1992.
6. Y. Nakamura et al., Reliable and stable operation of the high
field superconducting tokamak TRIAM-1M, Proc. Magnet Technol. Conf. MT-11, 767 Tuskuba; New York: Elsevier, 1990.
7. B. Turck and A. Torossian, Operating experience of Tore Supra
superconducting magnets, Proc. 15th IEEE/NPSS SOFE, Hyannis, MA, 1994, p. 393.
8. The IEA Large Coil Task, Fusion Eng. Des., 7 (12): 1228, 1988.
9. K. P. Juengst et al., Superconducting torus TESPE at design
values, Proc. Magnet Technol. Conf. MT-9, 36 Zurich, 1985.
10. M. Yamaguchi et al., Development of a 12 T forced-cooling toroidal field coil, Proc. ICEC10, Helsinki, 1984, p. 169.
11. E. Salpietro et al., Construction of a toroidal field model coil
(TFMC) for ITER, Proc. Magnet Technol. Conf. MT-15, Beijing,
1987.
12. Y. Takahashi et al., Experimental results of the Nb3Sn demo poloidal coil (DPC-EX), Cryogenics, 31: 640, 1991.
13. M. M. Steeves et al., Test results from the Nb3Sn US-demonstration poloidal coil, Adv. Cryog. Eng., 37A: 345, 1992.
14. M. Ono et al., Charging test results of the DPC-TJ, a high current-density large superconducting coil for fusion machines,
15. M. Darweschsad et al., Development and test of the poloidal field
prototype coil POLO at the Forschungszentrum Karlsruhe, Fusion Eng. Des., 36: 227, 1997.
16. N. Mitchell et al., ITER CS model coil project, Proc. 16th ICEC/
ICMC, Kitakyushu; New York: Elsevier, 1997, p. 763.
17. S. Imagawa et al., Helical coils for LHD, Proc. Symp. Cryogenic
Syst. for Large Scale Superconducting Applications, Toki, NIFSPROC-28, 1996, p. 112.
18. K. Takahata et al., Lopoidal coils for the Large Helical Device
(LHD), Proc. Symp. Cryogenic Syst. for Large Scale Superconducting Applications, Toki, NIFS-PROC-28, 1996, p. 116.
19. J. Sapper, The superconducting magnet system for the Wendelstein 7-X stellarator, Proc. Annu. Meeting Amer. Nuclear Soc.
(ANS), Reno, NV, 1996.
20. D. Evans, R. P. Reed, and N. J. Simon, Possible hazards following
irradiation of superconducting magnet insulation, Proc. 16th
ICEC/ICMC, Kitakyushu; New York: Elsevier, 1997, p. 2017.
21. A. Anghel et al., The quench experiment on long length, Final
report to ITER, Villigen PSI, 1997.
PIERLUIGI BRUZZONE
Centre de Recherches en Physique
du Plasma
743

Copyright
SUPERCONDUCTING MOTORS, GENERATORS, AND ALTERNATORS

Superconductors are very promising and exciting materials for electric power engineering in general and for
electric machines in particular. By allowing very high current densities and by suppressing the Joule losses,
superconductors improve the performance of electric machines by reducing weight, improving efficiency and,
to a lesser degree, by increasing compactness. The reduction of losses results in long-term savings in capital
cost, making superconducting machines attractive from an economic point of view. Since a cryogenic system
is required to maintain the superconducting state, these advantages appear only above a critical (breakeven)
size or rating, such that where the refrigeration penalty is negligible. Superconductivity is thus attractive
only for large electrical machines. Small motors (up to a few kilowatts) will not be superconducting except if
room-temperature superconductors are discovered, which is highly improbable.
The critical size is reduced when the operating temperature increases. Devices with high-criticaltemperature superconductors will be attractive for lower ratings than systems cooled with liquid helium.
Electric MachinesGeneral Structure(1)

An electric machine is reversible. It can operate as a motor, converting electrical into mechanical power, or as a
generator, converting mechanical input into electricity. The electromagnetic force or torque is produced in two
ways. The first one is the interaction between currents, called armature currents, and a variable-reluctance
structure (variable-reluctance machines). The second and more widely used way is the interaction between
currents and a magnetic field called the excitation field. The mobile part can be either the armature or the
excitation. But as the energy transfer to a moving part creates losses except by electromagnetic way, the
armature is preferably stationary.
The torque per unit volume is proportional to the excitation field component perpendicular to the current
times the armature ampere-turn loading, as shown below. The armature ampere-turn loading is the total
armature current divided by the mean armature circumference. The expression for the maximum torque (max )
for a three-phase machine is simply:
where
Bo = excitation field component
K = armature ampere turn loading
ro = mean armature radius
L = active length
N s = total series turns per phase
kd = winding factor
1

I = rated armature current
mach = approximate machine volume
The excitation field is created either by permanent magnets or by current flow in a winding. Permanent
magnets are limited in magnetic induction and are made of very expensive materials, but they are not dissipative. They are well suited particularly for small machines (kilowatt range). The currents in a conventional
conductor produce heat through the Joule effect (R i2 where R is the resistance and i the current), dissipating
energy. The current capacity is hence limited by the ability to remove heat. Better cooling conditions increase
the current capacity but reduce the efficiency. The current density (current per unit cross-sectional area) is then
limited by thermal and economic factors. The allowable current density in copper is on the order of amperes
per square millimeter (5 MA/m2 to 10 MA/m2 ). With such values the amounts of conductor required to produce
magnetic fields without magnetic materials are large, leading to huge Joule losses. Thanks to the peculiar
properties of soft magnetic materials (high relative permeability), the total current (ampere turns) required to
produce a given magnetic induction is greatly reduced. For this reason practically all electric machines have a
magnetic circuit with slots where the windings are embedded.
The armature ampere-turn loading is limited by Joule losses and by the current density allowable in the
conductors, because the space they can occupy is limited.
The magnetic circuit has other advantages than the reduction of the excitation current. It confines the flux
within the machine and reduces the stray field to negligible levels. It also prevents magnetic disturbances to
other equipment. The magnetic circuit is also very useful from a mechanical point of view. When the conductors
are inserted into slots they are subjected only to a reduced electromagnetic force, since the field concentrates
itself in the teeth. The electromagnetic force is mainly applied at the interface between the slots and the
magnetic teeth. The torque is then essentially supported by the magnetic circuit and not by the conductors. The
reduced mechanical stresses on the conductors are an important advantage, because the mechanical strength
of copper is low. In a slotted structure, no special care need be taken in order to reduce the eddy-current losses
in the conductors, since they only see low fields. Without magnetic teeth, a strong mechanical support structure
must be provided in order to sustain all the electromagnetic torque, and the conductors should follow the finely
divided Litz wire configuration to avoid large eddy-current losses. However, the magnetic circuit is heavy, the
increased magnetic induction it provides is limited by saturation, and it creates pulsating torques, because the
alternation of magnetic teeth and slots produces local magnetic variations. The magnetic teeth also reduce the
space available for conductors and thus the armature ampere-turn loading. The slotted structure is also not
convenient for insulation, so that the maximum voltage is limited (to about 30 kV).
Superconducting materials show promise for electric machines because they offer the possibility to increase both the excitation field and the armature ampere-turn loading (2,3). Superconductors are particularly
convenient for producing magnetic fields that are constant in time. Since the current densities in superconductors can be very high (up to a hundred times the allowable value in copper, i.e., hundreds of megamperes per
square meter), the required quantity of conductor to produce a given field is greatly reduced from that with
conventional conductors, even without the help of magnetic materials. The magnetic circuit is usually nearly
removed when using superconductors. Magnetic materials are in general used only to form a magnetic shield
in order to avoid large stray fields outside the machine. Current maintenance in a superconducting winding
does not cost any energy, due to the absence of losses for constant current and constant external field. The
disappearance or large reduction of the magnetic circuit leads to a light and saturation-free structure with
Fig. 1. Schematic cross sections of ac generators. (a) classical; (b) superconducting field winding; (c) fully superconducting.
more active space for conductors and insulation materials. The ampere-turn loading and the voltage can then
be increased. The absence of iron teeth will decrease vibration by suppressing torque ripples. Acoustically very
quiet electric machines can be designed.
However, the torque is applied directly to the conductors. They must therefore be supported by a suitable
structure. An armature without magnetic teeth subjects the conductors to large forces at twice the frequency
of rotation, which must be restrained by novel means of support for which high reliability must be maintained. Figure 1 shows the main differences between a conventional machine and superconducting ones (for
synchronous machines).
The weak point of a conventional machine is in general its insulation, which degrades badly with time. It
is very sensitive to thermal cycling, and overheating strongly affects its lifetime. A cryogenic system is hence
very favorable from this point of view: it almost completely avoids thermal cycling in operation. Moreover, at
low temperatures all aging process are slowed down. The cryogenic components of superconducting machines
should thus last longer, particularly if the machine remains at low temperature. Numerous thermal cycles
from room temperature to cryogenic temperature must be avoided. Furthermore, they are costly in time and
energy.
The very high current densities in superconductors make them very attractive for the armature by
increasing the ampere-turn loading. However, the armature currents are in general alternating, so that losses
appear in the superconductors. This is an important disadvantage in a cryogenic environment. In order to
Fig. 2. Engineering critical characteristics of superconducting materials and wire cross sections.
discuss this point and for the sake of completeness, some information about superconducting wires (materials
and ac losses) will be given in the following sections.
Superconducting Materials for Electric Machines

Classes of Materials. Superconductors divide themselves into two major groups: the low-T c and the
high-T c materials. The low-T c superconductors (LTSs), the earlier ones, are industrial materials with high
performance in terms of current capacity under high fields (1 T5 T, Fig. 2). However, they operate in general
only at temperatures near 4.2 K, which require a complex but well-controlled cryogenic system. The most
common low-T c superconducting multifilamentary composites use niobium titanium (NbTi) with typical cost
of about 1 $/kAm to 2 $/kAm. This figure of merit, the cost of 1 m of wire carrying 1 kA, enables comparisons
between materials. The compound niobium tin (Nb3 Sn) is only used for very high-field applications (> 8 T)
and is less used for electric machines. Its use is more complicated than that of NbTi, due to the long thermal
treatments required after winding it, and its cost is higher (5 $/kAm to 10 $/kAm).
High-T c superconductors (HTSs), assuming similar costs and performance to those of NbTi, will lead to a
reduction in cryogenic costs (capital and especially operating costs), but the main advantage is the improvement
in the stability of the superconducting state, which leads to higher reliability. At 20 K the specific heats are
200 times higher than at 4 K. The specific heat, being the amount of heat input necessary to raise the
temperature, represents the materials inherent brake on temperature rise. HTSs are thus less sensitive to
thermal disturbances.
Even with the large research effort focused on HTSs, these materials have not yet achieved the state of
development stage of NbTi. HTSs are very complex anisotropic ceramic materials, difficult to fabricate in a
conventional wire or cable. Intrinsically brittle, they are sensitive to mechanical stresses, and their transport
properties under fields are still much poorer than those of NbTi (Fig. 2), except for highly oriented, essentially
epitaxial films. They are also very expensive materials, and the current price is the main barrier to their
economic development. Their cost must be lowered to 10 $/kAm to be competitive (4). At present it is nearly
50 times higher.
There are two main routes to fabricate HTS wires. The more advanced one is the (powder-in-tube) (PIT)
technique (5,6) based on bismuth-compound filaments embedded in a silver or silver alloy matrix (Fig. 2).
Lengths of BiPIT tapes as long as 1 km are produced routinely by several companies throughout the world,
and their typical critical current densities are shown in Fig. 2. Still higher critical current densities are obtained
on small samples (J c = 760 MA/m2 at 77 K, 0 T; J e 250 MA/m2 ). Some specialists think nevertheless that the
limits have almost been reached. The pure silver matrix unfortunately is not suitable for ac applications, due
to the high ac coupling losses, and new PIT wires are under development for ac applications (5) (silver alloys,
resistive barriers, etc.).
The second route consists of so-called coated conductors (7) and has much potential. Yttrium compounds
are deposited in thick films (a few micrometers) on industrial flexible textured metallic substrates through a
buffer layer. Very good performance has been obtained with these coated conductors, but only for short lengths.
The engineering current density (overall current density including substrate) is large in liquid nitrogen (on
the order of 200 megamperes per square meter at 77 K at present), and its decrease under field is small. A
lot of difficulties must be overcome to fabricate long, high-performance coated conductors, and there is now no
low-cost industrial deposition technique. High quality Y superconductor bulk pellets, up to 100 mm in diameter
(8), have been processed, and they can be used in some special machines (hysteresis, reluctance, trapped-field,
etc.).
Ac losses. One of the most spectacular properties of a superconductor is its absence of resistive losses.
This is true, however, only for non-time-varying electromagnetic quantities (dc conditions). As soon as the
magnetic induction varies with respect to time, ac losses appear in superconducting wires. The magnetic
induction can be external or due to the current in the wire (self-field). The ac losses have two main consequences.
On the one hand, they induce a temperature rise in the superconductor. Since the temperature margin is very
small ( 1 K) for low-T c materials (NbTi for example) such a rise can easily quench the superconducting coil,
that is, destroy its superconductivity. On the other hand, ac losses are very expensive energetically, since they
are dissipated at low temperatures. They therefore greatly reduce the advantage of using superconductors.
From the second law of thermodynamics, the removal of energy at a cold temperature (T c ), requires work at a
high temperature (T 0 ), usually room temperature. For an ideal closed cycle the ratio of the minimum required
work (W min ) at T 0 to the energy (Q) to be removed at T c is given by Carnots expression
(Fig. 3). As shown in Fig. 3, this minimum work increases rapidly at low temperatures. In order to take
into account the real cycle and the imperfections of the thermodynamic transformations, this ratio should be
divided by the efficiency factor of the refrigeration system:
This depends mainly on the cold power and little on the cold temperature (Fig. 3, Ref. 9).
The ratio W min /Q in real conditions [Eq. (3)] is called the specific work, and its reciprocal the coefficient
of performance. To calculate the cost of refrigeration, the losses at low temperature must be multiplied by the
specific work. For an efficiency factor of 10%, it amounts to 740 W/W and 29 W/W for cold temperatures of 4
Fig. 3. (a) Carnots specific work and (b) efficiency factor as functions of the cold power (9).
K and 77 K, respectively. These two figures illustrate the advantage of operating at high temperatures from a
cryogenic point of view and again underlines the interest in using HTSs. The ac loss cost is especially high for
LTSs, and it must be reduced to an ultralow level for the system efficiency to be acceptable.
A simple way to understand the ac losses is to consider the MaxwellFaraday law (curl E = B/t). This
shows that an electric field appears as soon as the magnetic induction varies with time. The induced electric
field associated with a current density (transport current or persistent currents) results in losses. The losses
per unit volume are the scalar product of these two vectors.
If it is not possible to suppress the ac losses, it is possible to reduce them by a suitable multifilament
structure. This will depend on the field configuration (self-field, transverse or axial field), but ultralow-ac-loss
superconducting strands are generally achieved with very fine twisted filaments embedded in a high-resistance
matrix or with resistive barriers between filaments. The strand diameter should be low as well. Ac NbTi wires
have small (< 0.2 mm) elementary strands with hundreds of thousands of filaments ( 0.2 m) in a CuNi
resistive (0.4 m) matrix (Fig. 2). The first NbTi low-ac-loss composites were developed only in the eighties
when the technology for fine filament fabrication was sufficiently developed (10). Those strands have greatly
extended the potential range for superconductivity (11). For high-T c materials the requirements are less
severe, since the cost of removing the ac-loss heat is reduced (29 W/W at 77 K compared to 740 W/W at 4 K).
Nevertheless, no oxide superconducting tape actually fulfils them with present HTS wire technology.
The ac losses explain why superconducting devices are confined to applications with dc current and
without or with time-varying fields, but in the latter case, protected from them.
Classes of Electric Machines and Their Superconducting Versions

There are many different classifications of electric machines, but the primary one is between dc and ac. Dc
machines have been widely used for adjustable-speed drives because the speed can be varied with a simple
electronic power supply. But due to the very large advances in easily controllable power electronics [silicon
switches as gate turn-off (GTO) thyristor and the insulated-gate bipolar transistor (IGBT)], ac machines can
now run at variable speed with good dynamic performance and with high efficiency and reliability. Ac machines
are simpler to build and require much less maintenance than dc machines. Dc motors are nevertheless still
common in the industrial world for adjustable-speed operations.
Dc machines are homopolar or heteropolar. In a homopolar machine neither the current nor the magnetic
field changes in direction (polarity), contrary to the heteropolar machines, where the magnetic field has an odd
number of poles. Dc machines are complicated by difficulties of current collection from the rotating part. The
problem is especially difficult for heteropolar machines, where complex commutation problems occur. A superconducting field winding suits homopolar dc machines well (12). Efforts have been devoted to superconducting
homopolar machines for marine propulsion in the United Kingdom (12), the United States (13), and China. A
program is still being carried on in the United States with HTS coils (13).
Ac machines are synchronous or asynchronous. Except for very special cases (hysteresis motors, for
example), the asynchronous type is not well suited to superconductive technology, since the torque in an asynchronous motor is directly proportional to the armature losses, whereas the main interest of superconductivity
is precisely the absence of losses. The use of external room-temperature resistances may be imagined, but the
structure then becomes complex and of dubious utility. Moreover, the air-cored structure of such superconducting machines results in a very low power factor, another disadvantage.
In contrast, superconductivity offers advantages of reduced weight and improved efficiency in building
synchronous ac machines. A lot of work has been devoted, throughout the world, to these machines, especially in
the form of large generators with a superconducting field winding and a room-temperature resistive armature
(14). The excitation is powered by a dc current and is subjected to a time-constant field under balanced
synchronous operation. The operation is called balanced when the currents in the phases are the same in
amplitude. Synchronous machines work better in balanced conditions and therefore are usually operated in
such conditions, which are very favorable for superconductors. The superconducting winding is protected by
an electromagnetic shield from time-varying fields during transient or unbalanced operations, to avoid quench
initiation and to reduce the cold-power requirements. The resistive armatures are unconventional in that their
structure is air-cored, with only an external iron shield instead of a slotted magnetic circuit.
New concepts of armature windings have been proposed (15). A superconducting armature is unfortunately problematic due to ac losses (ac currents and rotating fields). It was only in the eighties that superconducting ac armatures were successfully designed and built, thanks to the emergence of ultralow-loss ac NbTi
strands (11). It is still difficult to design an armature using present HTS wires.
All electric machines are based on the formula for the electromagnetic force: F = I l B (I = current,
B = magnetic field, l = elementary wire length). There are then two main configurations:
Radial field and axial current (drum type)

Axial field and radial current (disk type)
Homopolar Superconducting Machines. The homopolar machine is one of the oldest types of electric
machines. Faraday built one in 1831. Figure 4 shows disk- and drum-type homopolar machines. If the current
and magnetic field are constant in time, the torque is also constant. The absence of torque ripple makes the
machine acoustically very quiet. Torque or speed control is also easily achievable. Nevertheless, in a homopolar
structure it is not possible to put turns in series as in heteropolar machines, and the current must therefore be
very high to produce a large torque. By the same token, the voltage can be inconveniently small, especially for
low-speed machines. As an example with a rating of 10 MW at 100 rpm, the current reaches about 1,000,000 A
in a disk 2 m in diameter submitted to a field of 2 T (upper limit of conventional machines) and the voltage only
reaches 10 V, neglecting any losses (disk and brushes). The high currents must be transferred to the rotating
part, and that is the main problem occurring in homopolar machines, leading to large losses and short lifetimes.
Their development has consequently been limited, although some advances have been made (12), with the use
of solid brushes as well as liquid-metal contacts.
Fig. 4. Essential schemes of (a) disk-type and (b) drum-type homopolar machines.
Superconductors are convenient for the excitation of homopolar machines, since high fields (5 T to 6 T) are
achievable. The operational conditions are moreover very favorable for superconductors. The electromagnetic
torque is not supported by the excitation winding, and the field imposed by the armature current on the
excitation wire is reduced. Nevertheless, due to current-collection problems, little development has been carried
out on such machines, except for marine propulsion.
Several such machines were built at the International Research and Development (IRD) Company in the
United Kingdom from the end of the sixties until the beginning of the seventies (12). They were disk machines,
and the rating of the Fawley motor reached 2.4 MW. More recently, the Naval Surface Warfare Center in the
United States developed a 300 kW superconducting (NbTi) homopolar machine of drum type. It was installed
in a boat, the Jupiter II, (13), which in 1980 successfully demonstrated through sea operations the feasibility
of marine propulsion by superconducting machines. The NbTi coils were recently replaced by HTS coils thanks
to advances in those materials.
Ac Synchronous Superconducting Machines. An ac synchronous machine (Fig. 1, Table 1) consists
of a polyphase ac armature interacting with an excitation field. As in other machines, the field can be radial
(drum classical structure) or axial (disk structure). The latter configuration is very little used. Due to serious
problems with current transfer at high ratings, the armature is in general stationary (the stator) whereas
the excitation is the rotating part (rotor). The rotor is nearly always inside the stator. The static polyphase
armature with polyphase balanced currents produces a rotating field, which rotates at the same speed (the
synchronous speed) as the rotating part. The rotor then only sees a dc field under balanced steady-state
conditions. The excitation field is produced either by a winding fed by dc currents or by permanent magnets. In
the latter case no power supply is needed, but the excitation field cannot vary appreciably, and that is a limiting
problem for generators supplying a network with a constant-amplitude voltage. However, permanent magnets
are particularly convenient for motors, since controllable variable frequency power converters are available.
Most electric energy is produced by ac synchronous generators directly connected to the network through
a step-up transformer. The stability of the generator on the power system plays a very important part and
requires a lot of attention. In a network the generators are subject to various disturbances such as changes
of load, and they should rapidly recover stable operation. The stability is partly determined by intrinsic
parameters of the machine, but can be improved by external actions like excitation current control.
For motors supplied by a converter, the stability is determined essentially by the converter and only to a
small extent by the machine itself.
Since synchronous machines play a key role in the production of electricity, they are a subject of constant
innovations and improvements in manufacturing and operation. They have attained very high performance as
electromechanical converters. As an example, the efficiency has been raised to the remarkable level of 99% for
a 1500 MVA rating. The technological limits in terms of power density and efficiency using conventional means
are nearly reached, and only superconductivity can bring a major technological leap.
A superconducting winding can increase not only the excitation field but also the armature ampere-turn
loading, even if this remains resistive. Since the iron structure is no longer necessary with a superconducting
excitation, the iron teeth disappear and more space is available for the armature conductors. This increases
the armature loading by a factor of approximately 2 even with resistive wires. On using superconductors for
the armature instead of resistive wires, the enhancement is still much higher. The following figures give some
orders of magnitude:
Conventional machines: K = 100250 kA/m,

Air gap armature: K = 300350 kA/m,
Superconducting armature: K = 600800 kA/m.
10
It must be remembered, though, that unfortunately air-gap armatures subject the conductors to the entire
torque.
A superconducting armature is therefore very attractive, but unfortunately it develops ac losses due to
the alternating currents and the rotating field. With present HTS wire technology these losses are too high.
Only ultrafine NbTi filament ac wires are now available, and they impose a large refrigeration penalty (740
W/W) for operating at 4 K. The relatively recent availability of ac NbTi strands (at the end of the eighties) and
the wish to avoid liquid-helium cryogenics have resulted in only a few successful developments (1619). The
main work has thus been carried out on a superconducting field winding associated with a resistive armature.
Superconducting-Field-Winding AC Machines. The gain in these machines [Fig. 1(b)] is essentially
brought about by the increase of the armature loading and, surprisingly, only a little by the enhancement of
the excitation field. In an air-core configuration the field decreases rapidly with increasing radius. In a 2p-pole
cylindrical structure with an internal excitation concentrated at a radius rf , the field decreases as (rf /r) p+1 (r >
rf ). The field winding and the armature (radius r0 ) cannot be very close, due to the cryogenics vessels and the
thermal and electromagnetic shields. The ratio r0 /rf thus can easily reach 2, leading to a decrease of the field by
a factor of 8 for a four-pole machine. The field on the excitation winding is limited because the electromagnetic
stresses are proportional to the field squared. For this reason the excitation field at the armature is kept at
approximately the same value (around 1 T) in a superconducting synchronous machine as in conventional
machines. The power per unit volume is nevertheless increased by at least a factor of two, thanks to the higher
armature loading. The improvement in weight is larger, since the air-core structure avoids the heavy magnetic
circuit. There is only a small ring of laminated steel around the armature winding in order to reclose the flux
within the machine. People have imagined replacing this magnetic shield by a conducting shield in order to
decrease the armature weight even more (20). Such a shield is based on the eddy currents induced in it. This
solution, favorable from the weight point of view, would decrease the efficiency because the Joule losses of the
induced currents are higher than the iron losses in a magnetic shield.
Savings in volume and weight are not always relevant for stationary applications, and they do not justify
a complete technology change. Indeed, in a power plant the turbine driving the generator is much heavier and
bigger than it. However, a higher power density makes it possible to build machines with higher rating than
conventional technology. This was the main reason for the early investigations into superconducting generators
(see the section Historical Background and Current Trends below).
The 99% efficiency of a 1500 MVA generator is enhanced by about 0.1% to 0.2% due to the absence of
excitation Joule losses. The cryogenic losses even at 4 K are very small, and negligible for high ratings. An
upper limit for the cryogenic losses in a 1500 MVA machine is 200 W (although a reasonable value would be
lower than 100 W). Even with a specific work of 1000 W/W, they represent only 0.013% of the rating. This
improvement may appear very small; nevertheless the savings capitalized over the generator lifetime can
match the initial cost of the machine. However, these savings must not be overwhelmed by higher maintenance
and operating costs. Economic benefits are achievable only if reliability and maintenance are at least as good
as with conventional generators. Indeed, nonoperation during one single day per year will completely cancel
the gain due to lower losses. The reliability question and the uncertainty of a completely new technology are
the two main reasons that have largely deterred the use of superconductivity in electric machines. Reliability is
a key problem for superconducting machines, though numerous superconducting systems have demonstrated
great reliability. The BEBC bubble chamber at CERN has worked 24,000 h with rated current of 5700 A and
has stayed cold during nine years without any intervention. Stationary cryogenics is now a proven technology.
However, present cooling systems for machines, using water and hydrogen, are not simple. Hydrogen requires
in particular a rather complicated seal system and is dangerous.
The loss savings may be much higher for some special drives with low weight requirements, as in mobile
applications. In order to decrease the weight of the conventional electric machines, the current densities are
increased using sophisticated cooling technologies. A consequence is a reduction in efficiency. Electric machines
for mobile systems (space, planes, trains, etc.) are lighter but less efficient than their homologs for stationary
11
applications. As an example let us consider transformers for high-speed trains. Their efficiency reaches in some
conditions only 90%, compared to 98% for conventional transformers, but their weight is reduced by a factor of
two. Superconducting machines are therefore very attractive solutions for mobile applications.
The magnetic circuit close to the conventional armature conductors limits the voltage in general to around
30 kV, much below the level of a transmission network (200750 kV). The generators are hence connected to the
power grid through a step-up transformer. Due to the elimination of the magnetic teeth, it is possible to extend
the voltage upward. A direct connection of the generator to the transmission system can be considered (21).
The elimination of the step-up transformer would result in a simplification of the power generation system,
and in a still higher efficiency. Researches have been carried out along the same lines for the conventional
generators, and concepts for conventional machines operating at transmission-level voltages have been known
for decades, but technical problems are numerous. Recently ABB (Asea Brown Boveri) has presented a new
high-voltage conventional generator offering a direct connection to the power network (22).
Electromagnetic Shields. The superconducting winding is in general surrounded by an electromagnetic
shield that protects the superconducting wire from time-varying fields under transient or unbalanced operations. As soon as it is subjected to external time-varying fields, eddy currents are induced and cancel those fields
(23). The superconducting winding only experiences very attenuated time-varying fields, leading to ultralow
ac losses.
The electromagnetic shield is a conducting cylinder or a squirrel cage. The lower the resistivity of the
shield is, the better its shielding performance is. For this reason shields are often cold, to benefit from the
reduced resistivity of metals at low temperatures.
A very effective shield limits the rate at which the excitation field can be changed, whereas quick changes
are required for transient stability. Compromises are necessary. Also, electromagnetic shields are subject to
very large electromagnetic stresses under severe transients, such as sudden short circuits, and they should be
mechanically designed with great care to withstand those stresses.
For a generator, the electromagnetic shield plays an important part in stability. To damp mechanical oscillations after a perturbation, losses are required and superconductors are not convenient. So the electromagnetic
shield acts also as a damper. Unfortunately, a good shield for the superconducting winding is a bad damper,
because the resistivity must be very small to shield very small field variations. Good damping is achieved by
a shield with a large resistivity, operating in general at room temperature. Consequently the electromagnetic
shield system often consists of several shields at different temperatures. The multishield system protects the
superconductor and damps the oscillations.
Fully Superconducting Machines. A fully superconducting ac generator [Fig. 1(c) potentially surpasses
the advantages of the superconducting-field-winding generator in mass, size, and efficiency by increasing the
armature ampere-turn loading. By introducing the superconducting stator and ac wires for the excitation (24),
the cryogenic rotor can be simplified with the elimination of the electromagnetic and thermal shield system
[Fig. 1(c)]. The magnetic shield remains outside around the armature winding so as to avoid iron losses. The
magnetic shield can operate at an intermediate temperature between 4 K and 300 K (80 K, for example), since
the specific work decreases rapidly as the temperature increases (Fig. 3). It may act as a thermal shield as well.
The superconducting armature increases the freedom to select electrical characteristics and thus to benefit
from optimized system performance. For example, the synchronous reactance can be selected over a very wide
range of values. It will nevertheless be higher than in the superconducting-field-winding generator, since it is
proportional to the armature ampere-turn loading.
Steady-State and Transient Stabilities. Better power density and efficiency are not the only advantages
of superconducting machines. Superconducting generators improve the steady-state and transient behavior of
the network (25,26,27). The reactive load capacity is enhanced as well. These advantages are due to the lower
value of the synchronous reactance.
In steady-state synchronous balanced operation, the electrical diagram of one phase of a linear and
isotropic synchronous machine is very simple. It consists of an electromotive force E in series with a reactance
12
Fig. 5. Balanced steady-state equivalent circuit for one phase (star representation) of an ac generator.
X d called the synchronous reactance (Fig. 5). The latter is the inductance of one phase, taking into account
the two other phases (for a three-phase machine), multiplied by the electric pulsation rate. A resistance may
be added to take losses into account but it is generally negligible compared to the synchronous reactance. E is
proportional to the excitation current. The expressions for the power in a three-phase synchronous machine,
neglecting the losses, are
Here is called the internal angle and plays an important part in steady-state stability. In steady-state
operation where the generator is connected to the network, the voltage V is constant. Neglecting the regulation
of the excitation current, the electromotive force E is constant as well. The steady state stability limit for the
internal angle [Eq. (4)] is /2. This equation also shows that the lower the synchronous reactance is, the higher
the static stability is, since the internal angle is lower for a given power.
A magnetic circuit increases the inductances. For this reason, a superconducting generator with an aircore armature has a lower synchronous reactance than a conventional one with a iron-core structure (Table 1).
Roughly speaking, the reactance reaches 1/5 to 1/3 that of the conventional machines. Moreover, the voltage
regulation of the generator is simpler with a low synchronous reactance. Figure 5 shows in fact that the
electromotive force varies less with respect to the armature current when the synchronous reactance is low.
Figure 5 is no longer valid in the presence of transients. The machine is then represented by two axes,
with the d axis (excitation-field axis) perpendicular to the q axis. Depending on the rate of the disturbance, the
transient (X d ) and/or subtransient (X d ) reactance will have to be considered. These reactances are linked to
the synchronous reactance by dispersion coefficients, which depend only on the geometry:
A parameter commonly used to characterize the transient capability is the critical fault-clearing time
(CFCT). It is the maximum delay during which the generator can be subject to a fault (short circuit) and still
remain in synchronism after the fault has been cleared. Figure 6 shows that the subtransient reactance is the
key parameter in determining the CFTC. The synchronous reactance plays a limited part, though a low value
is slightly favorable. However, low values of the subtransient, transient, and synchronous reactances results
in very high short-circuit currents and huge electromagnetic stresses on the machine. The machines must be
mechanically designed not only for rated operation but also to withstand torques during fault conditions.
New systems have been considered to improve the stability, especially for fully superconducting generators, which are inherently unstable due to the absence of loss for damping. The control of energy transfers
13
Fig. 6. Influences of xd and xd ) on the critical-fault clearing time (24).
between the machine and a superconducting magnetic energy storage (SMES) provides good stability (28,29).
The SMES can also absorb a part of the energy during a fault and thus increases the CFCT. Stability problems
still require a lot of work, and new systems afford challenging opportunities.
Cryogenics
Cryogenics comprises all the techniques related to low temperatures. These are complicated but well mastered
now. Nevertheless, superconducting machines pose an additional difficulty due to the need to rotate one cryostat
(30), sometimes at high speed (3600 rpm). The mechanical stresses are important, and the effect of centrifugal
forces should be considered with great care. In particular, the liquid flows radially outward and is compressed,
resulting in a temperature rise. This effect is not negligible for liquid helium. However, better heat exchange
between the fluid and the superconducting coil is obtained due to the rotation.
The mechanical structure must not only withstand the electromagnetic forces of the winding itself as
in other superconducting devices, but also the entire torque of the machine during normal operation and,
above all, severe transients. The machine must not be destroyed by a sudden short circuit, for example. Huge
overtorques (up to 10 times the rated value) can then be experienced. The resulting stresses are very severe and
can be extreme on the electromagnetic shields. But the thermal losses through the mechanical structure should
be kept to a very low level because of their very high cost at 4 K (700 W/W). Special machine structures make
it possible nevertheless to suppress the torque on the field winding (31). The vacuum vessel should be very
tight, since dynamic pumping is problematic due to the rotation. The isolating vacuum should be maintained
at a cryogenic level over a long period.
Two components require special attention for a rotating cryostat: the helium transfer coupling and the
differential contraction system, which has to cope with the rotation. The transfer of liquid into a rotating
cryostat with low losses is not a simple problem, but it has been solved. The classical solution is a bayonettype connection (Fig. 7) with a rotating seal system operating in general at room temperature. Ferrofluidic,
mechanical, or labyrinth seals can be used.
The cryostat for a superconducting armature is special as well. It is stationary, but it is subjected to
rotating fields. All the central parts of the cryostat thus require electrically insulating materials to avoid eddy
currents. Fiber-glassepoxy composites have suitable electrical, mechanical, and thermal properties, but their
14
Fig. 7. Very simple example of a helium transfer system for a rotating cryostat.
use poses some difficulties in a helium cryogenic environment. Attention should be focused on helium tightness
(particularly at room temperature), reliability of the adhesive joints, and resistance to thermal shocks.
Table 1 (24) gives some characteristics of a conventional and different superconducting generators (32,
33,34) with a rating of 300 MVA at 3000 rpm.
Historical Background and Current Trends

The first superconducting machines were constructed and tested (35,36) as soon as superconducting wires with
high current capacity under field were available in the sixties. For experimental reasons, these first superconducting field windings were stationary. The first machine with a rotating superconducting field winding was
built at MIT (37). A 50 kW synchronous machine with a superconducting field winding and a superconducting
400 Hz armature was built by Dynatech (38), but the large ac losses in the armature precluded satisfactory
operation. Fully superconducting machines were abandoned up to the eighties. With the Dynatech exception,
up to the eighties, only the field winding was superconducting; the armature remained resistive and operated
at room temperature. Considerable research and development on these machines was carried out (14,39,40),
because there was then a large motivation. The rapid growth of the electric energy demand was requiring generators with ever higher ratings, but the resistive generators showed technological limits (1,800 MW), which
had been rapidly reached. Only superconductivity pushed the limits and brought the possibility of building
generators with higher ratings in order to answer the electrical demand. The numerous programs throughout
the world have led to large-scale experiments for electrical and also cryogenic investigations, demonstrating
the possibility of building superconducting generators. The main technical problems were solved, though a lot
of R&D is still required to meet the conditions necessary for building industrial products. Nevertheless, since
the eighties, only generators with medium ratings (600 MW) have been required, due to the moderation of the
electric energy demand. Most of the programs for ac generators were then reduced or stopped. The reasons
were not technical. The market conditions are unfavorable to superconducting generators only because their
critical size (between 500 MW and 800 MW for 4 K superconducting generators) is around or even above the
market rating. Only the Japanese have continued their effort, in particular with an ambitious program called
Super GM with three 70 MW superconducting model rotors (41).
The critical (breakeven) size is difficult to give with accuracyfirstly because some data are difficult to
evaluate (reliability, investment and operating costs for completely new devices, etc.) and secondly because the
critical size strongly depends on the application. Weight saving for a generator is not posted in the same way
for a stationary application (power plant) and for a moving system (train, ship, plane, etc.). A weight reduction
of 1 kg represents savings of about $700 to $800 for a plane, and $15,000 to $20,000 for a satellite (cost of
15
launching), but much less for a power plant. The critical size of a superconducting generator will consequently
be lower for an airplane than for a stationary power plant.
Superconducting ac generators have mainly been developed for large electric networks. Nevertheless,
generators for airborne applications have been studied and designed (42,43) because their light weight is then
of great interest.
Though the main developments have concerned ac generators, work has also been done on motors, especially in the United Kingdom (12), in the United States (13,44), in Japan (45), in Finland, and in China, where
homopolar machines for marine propulsion have been studied.
The emergence of superconducting strands able to operate under magnetic fields at industrial frequencies
(50 Hz or 60 Hz) with ultralow losses changed the technical situation in the eighties (11). The Dynatech
experiment of 1967 could then be reconsidered with confidence, and some small-scale fully superconducting
machines experimentally proved the possibility of designing satisfactory 50 Hz armature windings (16,17,
18,19). Due to the absence of industrial motivation to develop superconducting generators, these successful
experiments stayed at the laboratory scale.
On the other side, superconductors could be an attractive option for high-performance lightweight electric
drives. The discovery of high-T c materials has reinforced this interest because cryogenic systems at higher
temperatures than 4 K are much simpler and more practical than to those at 4 K. The use of electric drives is
continually increasing, and weight constraints are becoming more and more severe. In high-speed trains, for
example, increasing speed requires more powerful motors. Their weight should be reduced as much as possible
in order to limit the load per axle tree to avoid rapid track degradation.
Super-GM (46)). The Japanese are the only ones who have not only continued on working on superconducting machines but also reinforced their programs since the seventies. In 1988 the Ministry of International
Trade and Industry (MITI) launched a very ambitious project called Super-GM (an engineering research
association for Superconductive generation equipment and materials). The objectives were to develop a superconducting technology for electric power generation, including the design of 200-MW-class pilot generators and
the construction of 70-MW-class models. This large project is commissioned by NEDO (New Energy and Industrial Technology Development Organization). It involves 16 members from industry and research institutes.
The research and development program includes work on superconducting materials (both LTS and HTS),
structural materials, the machines themselves, and helium refrigeration systems. Three different large rotors
with NbTi field windings were built and successfully tested using a common resistive stationary armature
(Table 2). Through the three rotors numerous critical issues were investigated and tested (field winding, NbTi
conductor, excitation control, damper structure, thermal contraction system, etc.). The experimental programs
were carried out in a special test facility (Fig. 8). The superconducting windings are cooled by liquid helium
using a 100 l / h liquefier through a transfer coupling. Basic steady-state operations were performed, but also
severe tests such as three-phase sudden short-circuit and excessive negative phase. The tests began in 1997
and were completed in 1999. Machine A was connected to the 77 kV power grid and supplied 40 MVar. A
power of 79 MW was achieved with machine A, which has operated with 82 MVar. Machine B reaches the
remarkable power of 79.7 MW. It operated more than 1500 h with 44 successive starts and stops. It ran 814
h at its rated capacity of 79 MW. The quick-response machine (machine C) supported a current rise of 3200
A/s (3.8 T/s) without quenching. The refrigeration system ran 9320 h without failure, and the estimated mean
time between failures (MTBF) reached 14637 h (about 20 months). Super-GM was a very successful project,
which demonstrated the satisfactory and reliable operation of superconducting generators with significant
power even during severe conditions. The technology was developed and experimentally qualified for 200 MW
rating. Japanese workers are analyzing all the results, but design and construction of the 200-MW-class pilot
generator have been deferred.
Permanent-Magnet Superconducting Machine (47). Superconducting armatures could be designed
in the eighties thanks to the emergence of ultralow-ac-loss NbTi strands at that period. They brought a large
benefit by strongly increasing the armature ampere-turn loading with low losses even when referred to room
16
Fig. 8. Superconducting generator on the stage of the Super-GM testing center (from left to right: liquid-helium dewar,
superconducting generator, driving motor). (By courtesy of Super GM.)
temperature using the high specific work (740 W/W). In view of their reduced size and especially weight,
superconducting drives are particularly attractive for mobile systems with low-weight requirements. However,
these applications subject the drives to severe mechanical stresses due to shock and vibration. Therefore
the fully superconducting design with cryogenic field winding and armature has not been thought to be the
best suited one for those applications, especially with a NbTi field winding. Rotating 4 K cryogenics is very
complicated and appears unlikely to withstand a severe mechanical environment. It is much simpler to cope
with high stresses with a stationary cryostat. There are a lot of NbTi windings in magnetic resonance imaging
(MRI) apparatus, used on trucks, that nevertheless operate safely and reliably. These observations have led
to a hybrid structure with a permanent-magnet rotor and a NbTi armature. The rotating part is very simple
and robust, and high performance is obtained with the superconductor in the stationary armature. Permanent
17
Fig. 9. Permanent-magnet (NdFeB) superconducting (NbTi) motors: 15 kW, 750 rpm (foreground) and 150 kW, 400 rpm
(background).
magnets obviously produce a slightly lower field than a superconducting field winding, but this is acceptable
for the NbTi armature in terms of critical current density and ac losses. Two machines were built (Fig. 9) with
the support of the Delegation Generale de lArmement (DGA), France. The 15 kW, 750 rpm model successfully
experienced a large number of tests in steady state and transient operation. Its was powered by a pulse width
modulation (PWM) voltage inverter for variable speed that was easily controlled. However its size was too
small to be representative of a real machine because the critical size is greater than a megawatt, and a 150
kW, 400 rpm demonstrator followed. The latter passed successfully its first electrical tests in long-term and
constant-speed operation.
The permanent-magnet rotor might be replaced with a HTS field winding (48). Operation at around 30
K enables one to cope with a severe mechanical environment much more easily that at 4 K. The field winding
can be cooled by helium gas consisting of vapors from the liquid in the armature vessel.
High-Temperature-Superconductor Motors and Generators. A 1500 W HTS motor was built
using BiPIT racetrack coils operating at 20 K (49). US researchers from Reliance Electric and American
Superconductor Corporation (ASC), with support from the US Department of Energy, are working on motors
using HTS rotating field windings and a conventional resistive armature (50). The rotor is wound with a Bi2223PIT tape from ASC and cooled to about 30 K by helium gas circulation. The superconductor is protected
by an electromagnetic shield. The latter is composed of a cold copper shield and an external rotating vessel that
acts as a damper. The present structure of PIT with a silver alloy matrix results in ac losses. Several motors
were built. A 100 kW, 1800 rpm designed prototype was successfully tested in 1996, and it delivered 150 kW
continuously at the rated speed with the superconducting field winding operating at 27 K. The same year the
design and construction of a second motor was launched. Its rating of 750 kW at 1800 rpm (4160 V, 104 A)
corresponds to about the critical (break even) size for superconducting motors, though this parameter is difficult
18
Fig. 10. 19 MW HTS motor concept (from Ref. 51).
to evaluate. The air-core armature is cooled by water, and the supporting structure uses G-10 fiber-glassepoxy
composites. A Brayton closed-cycle helium refrigerator produces He at about 24 K to cool the HTS winding
through forced circulation. The estimated thermal loads reach 24 W for the rotating cryostat and 5 W for the
field winding. The motor is now at the final assembly stage. The next step will be a 3.7 MW precommercial
prototype.
The US Navy is interested in HTS large drives for surface combat ships or submarines. The considered
rating is 19 MW. The HTS motor is expected to be one-fifth the size and the weight of conventional machines
(51). The HTS motor design was based on a 12-pole synchronous machine with a HTS field winding. The HTS
coils are not cooled by a cryogen but are conduction-cooled with cryocoolers at a temperature in the range from
20 K to 40 K. The Gifford Mac Mahon cryocooler cold heads used are put directly on the support of the HTS
coils (Fig. 10) and are supplied with high-pressure (about 1.6 MPa) helium from an external compressor. These
cryocoolers have reliability in accordance with the requirements for those motors. The armature is air-cored
and resistive. Another advantage for such applications is the acoustical quietness of superconducting machines
due to their air-cored armature. The design of a 20 MW HTS motor was recently funded by US Navy.
Work is also being done on HTS generators, especially for mobile systems such as airborne applications
where the reductions of weight and losses are of prime interest. The US Air Force has a program for megawattclass generators using HTS field windings (52).
Special Machines
Reluctance, Hysteresis, and Trapped-Field Motors (53). Superconductors for motors can be either
wires (for windings) or bulk. YBCO pellets have large sizes [up to 100 mm in diameter now (8)] and good electromagnetic properties in liquid nitrogen (77 K), and consequently represent a new and attractive opportunity
for motors. A variable-reluctance structure containing such a rotor made of materials with different magnetic
permeabilities can produce a torque. The larger the difference of permeability is, the larger the expected torque
is. Superconductors, with permeability near zero (diamagnetic behavior) are therefore potentially attractive.
Several reluctance motors up to 10 kW have been built with bulk YBCO and tested in liquid nitrogen (54).
19
The torque in an asynchronous motor is created by the losses in the rotor. The rotor is subjected to
a rotating field. Ac losses appear in superconductors as soon as the external field is time-varying. This is
considered in general as a drawback, but it can be used in a hysteresis motor. The rotor consists of a bulk
pellet or an assembly of bulk YBCO pieces. The ac losses in the superconductors, called also hysteresis losses,
produce the torque. This is an attractive principle for small motors in special applications (cryogenic pumps,
for example). It has been studied experimentally in models with ratings up to 4 kW (55).
Rare-earth permanent magnets such as NdFeB have magnetization about 1.5 T at 300 K. Materials
with higher magnetization would result in better performancein particular, a larger torque per unit volume.
The attainment of trapped fields as high as 11.4 T at 17 K in bulk YBCO samples (56) opens new and
attractive opportunities for electric machines (57). Bulk superconducting pieces with a trapped field could be
used in synchronous machines. The operation is slightly different than that of a permanent-magnet motor,
since a HTS sample works at constant magnetic field and flux, whereas a permanent magnet works at constant
magnetization. Moreover, the magnetization is different for the two cases: it is constant in a permanent magnet,
and conical in a cylindrical HTS material. The main problem is the magnetization of the HTS samples. The
fields required are very high: 6 times the mean magnetization for zero-field cooling and cylindrical conditions.
Supersat (58). A superconducting machine of a new type, called Supersat, has been developed. Supersat
is based on rotating ferromagnetic plugs saturated by an external static field produced by a superconducting
solenoid. The FeCo plugs act as permanent magnets with a magnetization of 2.4 T, and they are all magnetized
in the same direction. The rotor is bulk and is convenient for very high rotation speeds. This is a disk-type
machine with an axial field and radially distributed armature conductors.
Conclusions
Though the technical problems are not completely solved, the further development of superconducting machines
is not determined by the technology but by the market. Superconducting machines offer better characteristics
and performance (weight, efficiency, electric behavior, etc.) than any conventional technology. The deregulation
of the electricity market and the progress in high-T c materials present a good opportunity for superconducting
devices. The cost of HTSs is at present one of the main economic limitation on those devices.
Acknowledgments
The author is pleased to thank L. Porcar, A. Foggia, and C. Chillet for their careful rereading and discussions.
BIBLIOGRAPHY
1. H. W. Beaty L. Kirtley Jr. Electric Motor Handbook, New York: McGraw-Hill, 1998.
2. H. H. Woodson Z. J. J. Steckly E. Hallas A study of alternators with superconducting field winding: Ianalysis, IEEE
Trans. Power Appar. Syst., 85: 264280, 1966.
3. H. H. Woodson et al. The application of superconductors in the field windings of large synchronous machines, IEEE
Trans. Power Appar. Syst., 90: 620627, 1971.
4. P. M. Grant T. P. Sheahen Cost projections for high temperature superconductors, presented at the 1998 Applied
Superconductivity Conf., Palm Desert, CA, 1998.
5. R. Flukiger
et al. Phase formation and critical current density in Bi,Pb(2223) tapes, Supercond. Sci. Technol., 10: 6892,
1997.
20
6. L. Masur et al. Long length manufacturing of BSCCO-2223 wire for motor and cable applications, presented at Int.
Cryogenic Conf., Montreal, 1999.
7. D. K. Finnemore et al. Coated conductors development: An assessment, Physica-C, 320: 18, 1999.
8. T. Fujimoto et al. Fabrication of 100 mm-diameter YBaCuO bulk QMG superconductors with larger levitation forces,
Inst. Phys. Conf. Series N 167, 1: 7982, 1999.
9. T. R. Strobridge D. B. Chelton Size and power requirements of 4.2 K refrigerators, in Advances in Cryogenic Engineering,
New York: Plenum, Vol. 12: 1967, pp. 576584.
10. P. Dubots et al. Behaviour of multifilamentary NbTi conductors with very fine filaments under a. c. magnetic fields,
J. Physique, 45: 467470, 1984.
11. A. Fevrier Y. Laumond Prospective uses of superconductors for 50/60 Hz applications, Proc. ICEC 10 (Int. Conf. on
Cryogenic Engineering), London: Butterworth, 1986, pp. 139152.
12. A. D. Appleton D. H. Prothero Direct current machines, in B. Seeber (ed.), Handbook of Applied Superconductivity,
Philadelphia: Inst. of Physics Publishing, 1998, pp. 15791612.
13. M. J. Superczynski D. J. Waltman Homopolar motor with high temperature superconductor field winding, IEEE Trans.
Appl. Supercond., 7: 513518, 1997.
14. J. L. Smith Overview of the development of superconducting synchronous generators, IEEE Trans. Magn., 19: 522528,
1983.
15. C. Flick New armature winding concepts for EHV and high CFCT applications of superconducting turbine generators,
IEEE Trans. Power Appar. Syst., 98: 21902200, 1979.
16. P. Tixador et al. Electrical tests on a fully superconducting synchronous machine, IEEE Trans. Magn., 27: 22562259,
1991.
17. O. Tsukamoto et al. Development of 30 kVA class fully superconducting generator, IEEE Trans. Magn., 28: 283286,
1992.
18. E. U. Klimenko et al. A superconducting field and armature windings turbogenerator model, Proc. Low Temperature
Engineering and Cryogenic Conf., Southampton, 1990.
19. I. Muta et al. Electrical characteristics of fully superconducting synchronous generator in persistent excitation mode,
IEEE Trans. Magn., 28: 434437, 1992.
20. R. E. Dodge E. P. Coomes J. L. Kirtley Design of a superconducting alternator for space-based power generation, 21st
Intersociety Energy Conversion Engineering Conf., Vol. 3, Amer. Chem. Soc., 1986, pp. 18691874.
21. M. Rabinowitz Cryogenic power generation, Cryogenics, 17: 319330, 1977.
22. M. Leijon et al. Powerformer(R): A giant step in power plant engineering, Proc. IEEE Int. Electric Machines and Drives
Conf. (IEMDC99), 1999, pp. 830832.
23. T. J. E. Miller P. J. Lawrenson Penetration of transient magnetic fields through conducting cylindrical structures with
particular reference to superconducting ac generators, Proc. IEE, 123: 437443, 1976.
24. P. Tixador Fully superconducting machines, in B. Seeber (ed.), Handbook of Applied Superconductivity, Philadelphia:
Inst. of Physics Publishing, 1998, pp. 15531577.
25. M. Furuyama J. L. Kirtley Transient stability of superconducting alternators, IEEE Trans. Power Appar. Syst., 94:
320328, 1995.
26. M. Ashkin et al. Stability criteria for superconducting generatorselectrical system and cryostability considerations,
IEEE Trans. Power Appar. Syst., 101: 45784586, 1982.
27. P. W. Eckels J. L. Smith Jr. Superconductor stability in the power system environment, Cryogenics, 29: 651654, 1989.
28. J. L. Sabrie Feasibility of large ac superconducting equipment, J. Physique, 45: 717720, 1984.
29. Y. Mitani K. Tsuji Power system stabilisation by superconducting magnetic energy storage connected to rotating exciter,
30. Z. J. J. Stekly G. Y. Robinson Cryogenic aspects of rotating superconducting field coils for superconducting machines,
Proc. Fourth Int. Conf. on Cryogenic Engineering, IPC Science & Technology Press, 1972, pp. 243249.
31. C. Pinet Y. Brunet A 500 kW, 3000 rpm cryoalternator of a new type, Electric Machines Electromech., 3: 171175, 1979.
32. M. Ashkin et al. Superconducting generator field winding design for high tolerance fault, IEEE Trans. Magn., 19:
10351038, 1983.
33. R. Gillet et al. Electricite de France Asthom Atlantique superconducting turbo generator development program, IEEE
Trans. Magn., 17: 890893, 1981.
21
34. I. A. Glebov V. N. Shaktarin High efficiency and low consumption material electrical generators, IEEE Trans. Magn.,
19: 541544, 1983.
35. M. L. Claude Sur letude dun alternateur a` bobinage supraconducteur, C. R. Acad. Sci. Paris, 265: 349351, 1965.
36. Z. J. J. Steckly et al. A study of alternators with superconducting field winding: IIexperiment, IEEE Trans. Power
Appar. Syst., 85: 274280, 1966.
37. P. Thullen J. L. Smith The design of a rotating dewar and field winding for a superconducting alternator, in Advances
in Cryogenic Engineering, Vol. 15, New York: Plenum, 1970, pp. 132140.
38. C. J. Oberhauser H. R. Kinner Some considerations in the design of a superconducting generator, in Advances in
Cryogenic Engineering, Vol. 13, New York: Plenum, 1967, pp. 161165.
39. D. Lambrecht Superconducting turbogenerators: Status and trends, Cryogenics, 25: 620627, 1985.
40. C. Flick W. R. Cown J. H. Jr. Parker General design aspects of a 300 MVA superconducting generator for utility
application, IEEE Trans. Magn., 17: 873879, 1981.
41. T. Shimada et al. Recent progress on 70 MW class superconducting generators, IEEE Trans. Appl. Supercond., 9:
11891192, 1999.
42. B. B. Gamble T. A. Keim Superconducting generator design for airborne applications, in Advances in Cryogenic Engineering, Vol. 25, New York: Plenum, 1979, pp. 127136.
43. R. D. Blaugher J. H. Parker J. L. McCabria High speed superconducting generator, IEEE Trans. Magn., 13, 755758,
1977.
44. H. O. Stevens et al. Superconducting machinery for naval ship propulsion, IEEE Trans. Magn., 13: 269274, 1977.
45. J. Sakuraba et al. Superconducting dc motor for ship propulsion, Proc. ICEC 10 (Int. Conf. on Cryogenic Engineering),
London: Butterworth, 1984, pp. 723725.
46. T. Ageta Recent R&D status on 70 MW class superconducting generators in Super-GM project, Supercond. Sci. Technol.,
13: 510515, 2000.
47. P. Tixador et al. Electrical tests on a fully superconducting synchronous machine, IEEE Trans. Appl. Supercond., 9:
12051208, 1999.
48. P. Tixador H. Daffix Conceptual design of an electrical machine with both low and high T c superconductors, IEEE
49. J. T. Eriksson et al. A HTS synchronous motor at different operating temperatures, IEEE Trans. Appl. Supercond., 7:
523526, 1997.
50. D. Aized et al. Status of the 1,000 hp HTS motor development, IEEE Trans. Appl. Supercond., 9: 11971200, 1999.
51. S. Kalsi B. Gamble D. Bushko HTS synchronous motors for navy ship propulsion, presented at Naval Symposium on
Electric Machines, Annapolis, 1998.
52. A. J. Rodenbush S. J. Young Performance of high temperature superconducting coils for implementation into megawatt
class generators, IEEE Trans. Appl. Supercond., 9: 12331236, 1999.
53. L. K. Kovalev et al. Hysteresis and reluctance electric machines with bulk HTS elements, recent results and future
development, Supercond. Sci. Technol., 13: 498502, 2000.
54. B. Oswald et al. Superconducting reluctance motors with YBCO bulk material, IEEE Trans. Appl. Supercond., 9:
12011204, 1999.
55. T. Habisreuther et al. Magnetic process in hysteresis motors equipped with melt-textured YBCO, IEEE Trans. Appl.
Supercond., 7: 900903, 1997.
56. S. Gruss et al. Trapped field beyond 11 T in bulk YBaCuO material, Inst. Phys. Conf. Series N 167, 1: 115118, 1999.
57. J. R. Hull S. SenGupta J. R. Gaines Trapped-flux internal-dipole superconducting motor/generators, IEEE Trans. Appl.
Supercond., 9: 12291232, 1999.
58. A. Rezzoug A. Mailfert P. Manfe Cryogenic supersaturated synchronous-Machine optimization and first experiment
results, IEEE Trans. Magn., 20: 17951797, 1984.
READING LIST
R. D. Blaugher Superconducting electric power applications, in Advances in Cryogenic Engineering, Vol. 42, New York:
Plenum, pp. 883898.
R. D. Blaugher Low-calorie, high-energy generators and motors, IEEE Spectrum, 34 (7): 3642, 1997, pp. 3642.
22
J. R. Bumby Superconducting Rotating Electrical Machines, Monographs in Electrical and Electronic Engineering, Oxford:
Clarendon, 1983.
P. M. Grant Superconductivity and electric power: Promise, promise . . . past, present and future, IEEE Trans. Appl.
Supercond., 7: 112133, 1997.
H. Kofler Generators with superconducting field windings, in B. Seeber (ed.), Handbook of Applied Superconductivity,
Philadelphia: Inst. of Physics Publishing, 1998, pp. 14971547.
H. Kofler Motors with superconducting field windings, in B. Seeber (ed.), Handbook of Applied Superconductivity, Philadelphia: Inst. of Physics Publishing, 1998, pp. 15481552.
PASCALTIXADOR
CNRS-CRTBT/LEG
646
SUPERCONDUCTING CAVITY RESONATORS

A key component of the modern particle accelerator (59) is
the device that imparts energy to the charged particles. This
is an electromagnetic radio frequency (RF) cavity resonating
at microwave frequencies typically between 50 MHz and 3000
MHz. Traditionally, accelerating devices are normal conducting cavities typically made from copper (59). One of the main
incentives for using superconducting cavities is that the dissipation in the walls of the copper structure can be substantially reduced. This is especially beneficial for accelerators
that operate in a continuous wave (CW) mode or at a high
duty factor (e.g., 1 percent). Superconducting cavities economically provide high CW operating fields. Another benefit
is that superconducting cavities can be designed to have a
large beam aperture which reduces the beam cavity interactions, allowing higher beam quality and higher beam current.
There are two distinct types of superconducting cavities,
depending on the velocity of the particles. The first category
is for accelerating charged particles that move at nearly the
speed of light, such as electrons in a high-energy linear accelerator [e.g., at TJNAF (1) at Jefferson Lab in Newport News,
VA] or a storage ring [e.g., LEP (2) at CERN in Switzerland].
The second type is for particles that move at a small fraction
(e.g., 0.01 to 0.3) of the speed of light, such as the heavy ions
emerging from a dc high-voltage Van de Graaff accelerator.
ATLAS (3) at Argonne National Lab, Argonne, IL is the longest-running heavy ion accelerator facility. Figure 1(a) is a
sketch of the typical superconducting accelerating structure
of the first type, and Fig. 2(a) is a corresponding photograph
(4). There are five accelerating cells that resonate in the
TM010 mode of the cylindrical cavity. As the particle traverses
each half-wavelength ( /2) accelerating gap in half a radiofrequency (RF) period, it sees a longitudinal electric field
pointing in the same direction for continuous acceleration.
Figure 1(b) is a sketch for a structure for low velocity particles, and Fig. 2(b) is a corresponding photograph (5). A coaxial
transmission line a quarter wavelength long resonates in the
TEM mode. A drift tube is suspended from the end of the
hollow center conductor. The structure has two accelerating
cells between the ends of the drift tube and the beam hole
openings located in the outer conductor of the coax. The accelerating gap is /2, where v/c. Since is small, must
be chosen to be large, to achieve a useful acceleration. Therefore a low resonant frequency is chosen, typically 100 MHz.
The wavelength also sets the height of the quarter-wave resonator. The example of Fig. 2 has a manageable height of less
than one meter.
Large-scale application of superconducting cavities to electron and ion accelerators is now established at many laboratories around the world (6). These accelerators provide highenergy electron and positron beams for elementary particle
research, medium-energy electron beams for nuclear physics
research, low-energy, heavy ion beams for nuclear research,
and high-quality electron beams for free electron lasers. Altogether more than 500 meters of superconducting cavities have
been installed worldwide and successfully operated at accelerating fields up to 6 MV/m to provide a total of more than 2.5
GV for a variety of accelerators.
The two most salient characteristics of an accelerating cavity are its average accelerating field, Eacc, and the quality factor Q0. The typical accelerating field at which 1 supercon-
RF power in
Beam-induced power out
(a)
/4
(b)
Figure 1. (a). An accelerating structure for velocity of light particles.
The resonant frequency for superconducting structures is typically
between 350 MHz and 3000 MHz. The cell length is half a wavelength
( /2) long. The phase of the electric field on the axis of each cell is
shown for the accelerating mode. Ports outside the cell region are for
input power couplers and higher-order mode power output couplers.
In many applications, the power induced by the beam in higher order
modes has to be removed by output couplers. (b) An accelerating
structure for low-velocity particles, such as heavy ions. The resonant
frequency is typically between 50 MHz and 150 MHz. The accelerating gaps are 12 in length, where v/c and v is the velocity of the
heavy ions.
ducting cavities are now operated routinely is Eacc 5 MV/m,

and the typical Q0 value is 2 109. The corresponding numbers for low-velocity structures are 3 MV/m and 109. Accelerating fields as high as 40 MV/m and Q0 values as high as
1011 have been reached in high-performance 1 superconducting test cavities.
The strongest incentive to use superconducting cavities is
in accelerators that operate in a continuous-wave (CW) mode,
or at a high duty factor (1%). For CW operation, the power
dissipation in the walls of a structure built from normal conducting material (such as copper) is substantial. Therefore
the typical CW operating field for a copper cavity is usually
kept below 1 MV/m. The microwave surface resistance of a
superconductor is typically five orders of magnitude lower
than that of copper, and therefore the Q0 is five orders of magnitude higher. For applications demanding high CW voltage,
such as increasing the energy of electron storage rings, the
advantage of superconducting cavities becomes clear. Since
the dissipated power increases with the square of the operating field, only superconducting cavities can economically
provide the needed voltage. For example, LEP requires 2.5
647
Apart from the general advantages of reduced RF capital

and reduced RF associated operating costs, superconductivity
offers certain special advantages that stem from the low cavity wall losses. Because of the low power dissipation at high
accelerating field, one can afford to make the beam hole of
superconducting cavity much larger than for a normal conducting cavity. The large beam hole substantially reduces the
beamcavity interaction [or wake fields (59)], allowing better
beam quality and higher current for improving the precision
and reaction rates of physics experiments. For the intense
proton linacs, where scraping of the proton beam tails is a
major worry because of radio-activation of the accelerator, the
wide beam hole greatly reduces the risk of beam-loss-induced radioactivity.
RF Superconductivity Basics
(a)
Figure 2. (a) Five-cell 1.5 GHz niobium cavity developed at Cornell,

now used at TJNAF. (b) A quarter-wave resonator from niobium developed for the JAERI (Tokai, Japan) heavy ion linac.
GV to double its energy from 50 GeV to 100 GeV per beam. If

copper cavities were to be used, both the capital cost of the
klystrons and the ac power operating cost would become prohibitive at the higher accelerating field. Several MW/m of ac
power would be required to operate a copper cavity at 5 MV/
m. There are also practical limits to dissipating high power
in the walls of a copper cavity. When more than 100 kW is
dissipated in a copper cell, the surface temperatures exceeds
100C, causing vacuum degradation, stresses, and metal fatigue due to thermal expansion. High accelerating fields
(100 MV/m) can be produced in copper cavities, but only for
microseconds, and the peak RF power needed (59) becomes
enormous (many hundreds of megawatts).
The remarkable properties of superconductivity are attributed to the condensation of charge carriers into Copper pairs,
which move frictionlessly. At T 0 K, all charge carriers are
condensed. At higher temperatures, some carriers are unpaired; the fraction of unpaired carriers increases exponentially with temperature, as e/kT, until none of the carriers
are paired above Tc. Here 2 is the energy gap of the superconductor, the energy needed to break up the pairs. In this
simplified picture, known as the London two-fluid model,
when a dc field is turned on, the pairs carry all the current,
shielding the applied field from the normal electrons. Electrical resistance vanishes.
In the case of RF currents, however, dissipation does occur
for all T 0 K, albeit very small compared to the normal
conducting state. While the Cooper pairs move frictionlessly,
they do have inertial mass. For high-frequency currents to
flow, forces must be applied to bring about alternating directions of flow. Hence an ac electric field will be present in the
skin layer, and it will continually accelerate and decelerate
the normal carriers, leading to dissipation proportional to the
square of the RF frequency. The two-fluid model provides a
simple explanation for the quadratic frequency and the exponential temperature dependence of the RF surface resistance.
The power dissipated is proportional to the internal electric
field (proportional to the RF frequency) and to the normal
component of the current. The normal component of the current, being proportional to the interior electric field, gives another factor proportional to the RF frequency. The normal
component of the current also depends on the number of carriers thermally excited across the gap 2 and is given by the
Boltzmann factor e/kT.
Besides the phenomenally low RF surface resistance, other
important fundamental aspects are the maximum surface
fields that can be tolerated without increasing the microwave
surface resistance substantially or without causing a breakdown of superconductivity. The accelerating field, Eacc, is proportional to the peak surface RF electric field (Epk), as well as
the peak surface RF magnetic surface field (Hpk).
The ultimate limit to the accelerating field is the RF critical magnetic field. Theoretically, this is equal to the superheating critical magnetic field. In the GinzburgLandau phenomenological theory of superconductivity (7), surface energy
considerations lead to estimates for superheating critical field
in terms of the thermodynamic critical field, Hc, and the Ginz-
648
burgLandau parameter, , as follows:
0.89
Hsh Hc
Hsh 1.2Hc
for 1
Hsh 0.75Hc
for 1
for 1
clock. This requires rapidly tuning the cavity to cancel the

effects of acoustically induced mechanical distortions (13).
Figures of Merit for a Superconducting Cavity
(1)
For the most commonly used superconductor, niobium, Hsh, is

about 230 mT, which translates to a maximum accelerating
field of 55 MV/m for a typical 1 niobium structure and
roughly 30 MV/m for a 1 niobium structure.
Typically, cavity performance is, however, significantly below the theoretically expected surface field. One important
phenomenon that limits the achievable RF magnetic field is
thermal breakdown of superconductivity, originating at submillimeter-size regions of high RF loss, called defects. When
the temperature outside the defect exceeds the superconducting transition temperature, Tc, the losses increase, because
large regions become normal conducting. Several measures
have been developed to overcome thermal breakdown, such as
(a) improving the thermal conductivity of niobium by purification or (b) using thin films of niobium (or lead) on a copper
substrate cavity.
In the early stages of the development of superconducting
cavities, a major performance limitation was the phenomenon
of multipacting. This is a resonant process in which a large
number of electrons builds up within a small region of the
cavity surface due to the fact that the secondary electron
emission coefficient of the surface is greater than unity. The
avalanche absorbs RF power, making it impossible to raise
the fields by increasing the incident RF power. The electrons
impact the cavity walls, leading to a large temperature rise
and eventually to thermal breakdown. With the invention of
the spherical cavity shape (8) [and later the elliptical cavity
shape (9)], multipacting is no longer a significant problem for
velocity-of-light structures. Multipacting is still an impediment for structures for low-velocity particles but can be reduced by long periods of exposure to high RF power, called
conditioning, during which the secondary electron emission is
reduced by long-term electron bombardment.
In contrast to the magnetic field limit Hsh , there is no
known theoretical limit to the tolerable surface electric field.
Continuous-wave electric fields up to 145 MV/m (10) and
pulsed electric fields up to 220 MV/m (11) have been imposed
on a superconducting niobium cavity surface without any catastrophic effects. However, at high electric fields, an important limitation to the performance of superconducting cavities
arises from the emission of electrons from high-electric-field
regions of the cavity. Power is absorbed by the electrons and
deposited as heat when electrons impact the cavity walls. If
the emission grows intense, it can even initiate thermal
breakdown. There have been extensive studies about the nature of field emission sites as well as development of techniques to avoid emission sites and to destroy them (12).
For low-velocity accelerators, there is an important additional performance consideration. Ambient acoustic noise (microphonics) excites mechanical vibrational modes of the cavity, causing the resonant frequency to vary. The resonant
cavities are extended, loaded structures (e.g., drift tubes supported by pipes) and generally have reduced mechanical stability. The cavity RF phase must be synchronized with an RF
We show how to calculate the important physical quantities,

such as resonant frequency, accelerating field, peak electric
and magnetic fields, power dissipation, quality factor Q0, and
shunt impedance for a simple cavity, the cylindrically symmetric pillbox. The treatment is basic to both normal conducting and superconducting cavities (see CAVITY RESONATORS). We
also work out illustrative values. Similar analytic calculations
can be carried out for a coaxial TEM quarter wave resonator,
as illustrative of an accelerating structure for low-velocity
particles. Only simple structures can be calculated analytically. For real structures with beam holes, it is necessary to
use field computation codes, such as (a) URMEL (14) for cylindrically symmetric structures and (b) MAFIA (15) for 3-D
geometries.
For a cylinder of length d and radius R, the electric (E)
and magnetic (H) fields for the standing wave TM010 mode are
Ez = E0 J0
2.405
R
eit ,
H = i
r
E0 J1
2.405
R
eit
(2)
where all other field components are 0. J0 and J1 are Bessel

functions of the radial coordinate. The angular resonant frequency 2f is given by
010 =
2.405 c
R
(3)
Note that the resonant frequency, f, is independent of the cavity length.

Assume an electron traveling nearly at the speed of light
(c). It enters the cavity at time t 0 and leaves at a time
t d/c. To receive the maximum kick from the cavity, the
time it takes the particle to traverse the cavity is to equal
one-half of an RF period, that is,
t=
1
d
= TRF =
c
2
(4)
Under this condition, the electron always sees a field pointing

in the same direction. The accelerating voltage (Vacc) for a cavity is

z=d

Vacc =
Eel dz
z=0

(5)
For an electron accelerator with energy 10 MeV, it is sufficiently accurate to use v c, so that t(z) z/c. Thus
Vacc
Vacc

z=d

iz/c
=
Ez ( = 0, z)e
dz
z=0

(6)
d

sin

z=d

2c
= E0
eiz/c dz = dE0

d
z=0
2c

= dE0 T
(7)
Here T is referred to as the transit time factor. At 1.5 GHz

we have d c/ 10 cm, and Eq. (7) simplifies to
cavity, we obtain
Vacc = 0.064 m E0
The average accelerating electric field (Eacc) is given by
Eacc =
2E0
Vacc
=
d
(8)
U=
0 E02 2
J1 (2.405) dR2
2
(15)
Pc =
RS E02 0 2
J1 (2.405)R[R + d]
0
(16)
G=
453Rd
(R2 + Rd)
(17)
Combining Eqs. (3) and (4), we find that in order to obtain

the maximum accelerating voltage from the cavity, we require
Here E has the dimensions of V/m.

Peak Surface Fields
To achieve a high accelerating field in a cavity, it is important
to minimize the design ratios of the peak fields to the accelerating field. For the TM010 mode in a pillbox cavity we have
Epk = E0 ,
Hpk =
r
649
J1 (1.841)E0 =
E0
647
2.405
R
=
d
(18)
so that G 257 . A typical observed surface resistance for

a well-prepared superconducting Nb cavity is Rs 20 n.
Thus we have a Q0 value of
(9)
Q0 =
G
= 1.3 1010
Rs
(19)
Thus we obtain the following ratios:

Epk
Eacc
= 1.6,
=
2
Hpk
Eacc
mT
A/m
= 3.05
= 2430
MV/m
MV/m
(10)
Power Dissipation and Q0

In order to support the electromagnetic fields, currents flow
within a thin surface layer of the cavity walls. If the surface
resistance is Rs, the power dissipated/unit area (Pa) due to
Joule heating is
1
Rs H 2
2
(11)
The quality, Q0, is related to the power dissipation by the

definition of Q0:
Q0 =
U
Energy stored
=
Power dissipated
Pc
(12)
where U is the stored energy and Pc is the dissipated power.

The total energy in the cavity and the power dissipated are
U=
2 0

|H|2 dv,
Pc =
1
Rs
2

|H|2 ds
(13)
where the integral is taken over the volume of the cavity.

Thus

0 v |H|2 dv

,
Q0 =
Rs s |H|2 ds
G
Q0 =
,
Rs
Vacc
= 10 MV/m
d
Epk = E0 = Eacc = 15.7 MV/m

2
A/m
Eacc = 24.3 kA/m = 30.5 mT
Hpk = 2430
MV/m
2
0 E0 2
J1 (2.405) dR2 = 0.54 J
U=
2
U
Pc =
= 0.4 W
Q0
Eacc =
The units for magnetic field used are teslas.
Pa =
For a typical cavity length of d 10 cm (at 1.5 GHz), we

obtain R 7.65 cm. For an accelerating voltage of 1 MV, we
obtain the following results:
|H|2 dV
|H|2 ds
s
(14)
0

G=
Here G is called the geometry factor. It only depends on the

cavity shape and not its size. For the TM010 mode in a pillbox
(20)
The performance of a superconducting cavity is evaluated by

measuring the Q0 as a function of the cavity field level. This
gives information on the average behavior of the RF surface.
Thermometry Based Diagnostics
To resolve the local distribution of RF losses from various
mechanisms described above, temperature mapping is used
as a diagnostic technique. A chain of rotating carbon thermometers, or an array of fixed thermometers, samples the
temperature of the outer wall of the cavity. Temperature
mapping with carbon thermometers has played a key role in
improving the understanding of mechanisms that lead to residual resistance, multipacting, thermal breakdown, and field
emission. Carbon makes a sensitive thermometer at liquid helium temperatures because, as a semiconductor, its resistance
increases exponentially with decreasing temperature. Above
the superfluid temperature (2.17 K), temperature increments
of the cavity wall of a few mK can be easily detected. A single
rotating arm bearing 10 to 20 thermometers per cell is appropriate for locating stable field emitters or thermal defects in
sizable structures, such as a multicell cavity. For temperature
mapping in superfluid helium, thermometers need to be isolated from the superfluid bath so that movable elements do
650
(a)
(b)
(c)
Figure 3. (a) A single-cell niobium cavity surrounded by an array of 700 carbon thermometers
that make close contact with the outer wall of the cavity. (b) There are 19 thermometers placed
on each individual board that is contoured to closely follow the cavity profile. (c) A single thermometer consists of a 100 carbon resistor embedded in an epoxy housing. It is held by a springloaded pin inserted into holes in the board. The surface of the thermometer is ground so that the
carbon element is exposed and subsequently covered with a thin layer of varnish to provide
electrical insulation. The leads are made of a low-thermal-conductivity alloy, such as manganin.
not provide good sensitivity. A large array of fixed thermometers is preferred. These are brought in intimate contact with
the cavity wall by the use of spring loaded contacts. Grease
applied between the cavity wall and the thermometer element
improves heat transfer and keeps the superfluid away. Due
to the large number of thermometers and leads, the fixed
method is suitable for investigations with single cell cavities.
An example of a fixed thermometry system is shown in Fig.
3, and a typical temperature map is shown in Fig. 4 (16).
Refrigerator Requirements
Although the power dissipated in the superconducting cavity
is very small, the losses will be dissipated in the liquid He
bath. Together with the static heat leak to the cryostat, these
losses comprise the cryogenic loss. Typically the ac power
needed to operate the refrigerator is larger than the dissipated power in 2K liquid He by a factor of 750. One part of
this factor comes from the technical efficiency () of the refrigerator, typically 0.2 for a large system, and the other part
Figure 4. Temperature map at 40 mT of a single-cell 1.5 GHz cavity

showing heating at a defect site near the cavity equator (labeled 1)
and field emission sites (labeled 2, 3, and 4) near the cavity iris.
comes from the Carnot efficiency c, which at 2 K is

1
300 2
=
c
2
103
Rs ()
105
106
107
108
109
1.0
Shunt Impedance
An important quantity used to characterize the losses in a
cavity at a given accelerating voltage is the shunt impedance
(Ra) as typified by a parallel RLC circuit:
2
Vacc
Pc
(22)
in which case Pc power dissipated and Vacc is the acceleration voltage. Hence the shunt impedance is in ohms.
Ideally the shunt impedance should be large for the accelerating mode so that the dissipated power is small. This is
particularly important for copper cavities, where the wall
power dissipation is a major issue and we wish to have as
large an accelerating field as possible. For the TM010 mode
pillbox cavity and Rs of 20 n we have
Ra =
3R
40 d 2
2
s 0 J1 (2.405)R[R +
d]
= 2.5 1012
(23)
Note that the ratio of Ra /Q is given by

V2
Ra
= acc
Q0
U
(24)
which is independent of the surface resistance. For the pillbox TM010 mode we have
Ra
d
= 150
= 196
Q0
R
(25)
By applying computer codes to determine electromagnetic

fields, the computed figures of merit for the Cornell/TJNAF
5-cell cavity are given in Table 1. Note that due to the presence of the beam holes the shunt impedance is reduced and
the peak surface fields are enhanced, relative to the pillbox
Table 1. Figures of Merit for the Cornell/CEBAF

5-Cell Cavity
G
R/Q (per 5-cell cavity)
Epk /Eacc
Hpk /Eacc
a
Data taken from Ref. 4.
290
480
2.6
4.7 MT/(MV/m)
Niobium
Lead
Nb3Sn
104
(21)
At 10 MV/m the required refrigerator ac power due to the RF

loss would be 300 W for the case of a single-cell 1.5 GHz cavity. For a copper cavity of the same geometry, with a typical
Rs 3 m, the RF power dissipation in the cavity would be
60 kW for an accelerating field Eacc 10 MV/m. Furthermore,
the ac wall power will be a factor of 2 higher because of the
typical klystron efficiency. Thus the ac power cost of running
a copper cavity in CW mode would be several hundred times
higher than the cost for an Nb cavity.
Ra =
651
2.0
3.0
Tc/T
4.0
5.0
Figure 5. Theoretical surface resistance at 1.5 GHz of lead, niobium,

and Nb3Sn as calculated from Halbritters program (20). The values
used for the material parameters are given in Table 2.
case. For a realistic cavity shape, R/Q0 is lowered due to the

presence of the beam holes, typically by a factor of 2.
RF Surface Resistance
Based on the very successful BCS theory (17), expressions for
the superconducting surface impedance have been worked out
by Mattis and Bardeen (18). These expressions involve material parameters, such as the London penetration depth L, the
coherence distance 0, the Fermi velocity VF, and the electron
mean free path l. They are in a rather difficult form to obtain
general formulas to work with. Computer programs have
been writtenfor example, by Turneaure (19) and Halbritter
(20). Figure 5 gives the results from Halbritters programs for
niobium and lead and Nb3Sn. Table 2 gives the material parameters used for the calculations. Calculations from the theory agree well with experimentally measured Rs for T/Tc
0.3. At lower temperatures the residual resistance term dominates.
A simplified form of the temperature dependence of Nb for
Tc /T 2 and for frequencies much smaller than 2/h 1012
Hz is
Rs = A(1/TJ) f 2 exp((T )/kB T ) + R0
(26)
Here kB is Boltzmanns constant. The second term, R0, is

called the residual resistance. Typical R0 values for Nb cavities fall in the range from 107 to 108 . The record for the
lowest surface resistance is 12 109 (21). For comparison, the surface resistance of copper at 1.5 GHz is 3 m.
Table 2. Material Parameters Used for the Calculations of

Fig. 5
Material Parameter
Tc [K]
Energy gap, /kTc
]
Penetration depth [A
]
Coherence length [A
]
Mean free path [A
Pb
Nb
Nb3Sn
7.19
2.10
280
1110
10,000
9.20
1.86
360
640
500
18.00
2.25
600
60
10
652
The operating temperature of a superconducting cavity is

usually chosen so that the first term in Eq. (26) is reduced to
an economically tolerable value. R0, referred to as the residual
resistance, is influenced by several factors. Some of the
sources are extraneous to the superconducting surfacefor
example, lossy joints between components of the structure.
Other factors originate at the superconducting surface. A
well-understood and controllable source of residual loss is
trapped dc magnetic flux from insufficient shielding of the
earths magnetic field, or other dc magnetic fields in the vicinity of the cavity. To get the highest Q0, a superconducting
cavity must be well-shielded from the earths field. Typically,
at 1 GHz, R0 is 10 /mT (22). Another important residual
loss mechanism arises when the hydrogen dissolved in bulk
niobium precipitates as a lossy hydride at the RF surface (23).
This residual loss is a subtle effect that depends on the rate
of cooldown and the amount of other interstitial impurities
present in niobium. The effect can be severe enough to lower
the Q0 to 108 depending on the amount of hydrogen dissolved
and the cooldown rate of the cavity. More than 2 ppm wt of
hydrogen can be dangerous.
Cavity Fabrication and Surface Preparation
Niobium cavities can be constructed from sheet niobium using
the techniques of forming (e.g., deep drawing or spinning) followed by electron beam welding (24). Another method is to
deposit a thin niobium film onto a preformed copper cavity
substrate (25). The copper cavity is made in essentially the
same way as the sheet niobium cavity, except for surface
preparation before film deposition. If the cavity has more
than one cell, the cells need to be tuned relative to each other,
by adjusting the dimensions, so that the accelerating field is
the same for each cell. Dimensional variations between cells
are sufficient to alter the field profile substantially. Typical
fabrication tolerances are in the range of a few tenths of a
millimeter.
The purity of niobium used is important, both in terms of
bulk impurity content and in terms of inclusions from manufacturing steps, such as rolling. Inclusions on the RF surface
play the role of normal conducting sites for thermal breakdown. Dissolved impurities serve as scattering sites for the
electrons not condensed into Cooper pairs. These impurities
lower the thermal conductivity, impede the heat transfer to
the helium, and limit the maximum tolerable surface magnetic field before the onset of thermal breakdown. The accompanying decrease in electrical conductivity, or the RRR value,
serves as a convenient measure of the purity of the metal.
The formal definition of RRR is
RRR =
resistivity at 300 K
residual resistivity at low
temperature (normal state)
(27)
Here low temperature means the temperature at which the dc

resistivity in the normal state becomes residual. A convenient
relationship between thermal conductivity and RRR for niobium is
k 0.25 RRR
W
mK

(28)
This relationship can be derived from the WiedemannFranz

law (26) and from the ratio of the superconducting to normal
conducting state thermal conductivities (27). To achieve the
optimum RF performance, the surface of the cavity must be
prepared to approach as close as possible the ideal. Microscopic contaminants can limit the performance, either by
thermal breakdown or by field emission (28). A clean RF surface is achieved by chemically etching away a surface layer,
rinsing thoroughly with ultraclean water, and then taking
precautions so that no contaminants come in contact with the
clean RF surface. The resistivity of the water should be close
to theoretically pure (18 M-cm), and the water should be
filtered to eliminate particles larger than 1 m. After etching,
water is recirculated for several hours through the cavity in
series with the water purification system so as to continuously and thoroughly remove any chemical and particulate
residue from the niobium surface. For a review of cavity fabrication and preparation procedures, see Ref. 28.
Many laboratories have found that the RF surface can be
made even cleaner if chemistry is followed by high-pressure
rinsing (HPR) of the cavity with ultrapure water (29). At
TJNAF for example, water at a pressure of 70 bar to 80 bar
is sprayed through stainless steel nozzles each having a 0.3
mm diameter orifice (30). The potent jets of water are scanned
across all parts of the RF surface to dislodge and sweep away
microscopic contaminants that have adhered to the surface.
After rinsing, the cavity is transported into a dust-free
clean room where the water is drained. The cavity surface
thus only comes in contact with filtered air. The level of cleanliness required is comparable to that in the semiconductor
industry where a clean room environment of Class 10100 is
routine. Class 10 refers to the number of particles of size 0.5
microns or larger in one cubic foot of air. The surface of the
cavity must be dried before the cavity is evacuated, placed
inside a cryostat, and cooled down for RF tests. During final
assembly, the laboratory workers in the vicinity need to wear
special particulate-free clothing and follow strict protocols to
reduce particulate generation.
During these various stages of cavity production and preparation, there are many opportunities for defects to enter the
cavity. Therefore great care must be exercised during the
manufacture of sheet metal, deep drawing of cups, electron
beam welding, chemical etching, rinsing, drying, and insertion of coupling devices, as well as in the final attachment
of the cavity to the vacuum system of the test stand or the
accelerator. Two examples of defects that caused thermal
breakdown are shown in Fig. 6 (31). On a statistical basis,
we expect that the number of defects increases with cavity
surface area.
Overcoming Field Emission
The temperature mapping diagnostic technique for superconducting cavities shows that emission arises from particular
spots, called emitters, located in high-electric-field regions.
The electrons that emerge from the emitters travel in the RF
fields of the cavity and impact the surface (Fig. 7). Some electrons may be captured in the axial fields and accelerated
along with the beam. These produce unwanted dark current, which may spoil the beam quality or impact the walls
of adjacent cavities. The pattern of temperature rise as a
function of position along a given meridian contains implicit
653
Figure 6. SEM micrographs of defects that

caused thermal breakdown. (a) A chemical
or drying stain 440 m in diameter. The
small crystal on the right side contains K,
Cl, and P. This defect quenched at Eacc 3.4
MV/m. (b) A 50 m crystal containing S, Ca,
Cl, and K. This defect quenched at Eacc
10.7 MV/m. These defects were located by
temperature maps. (Courtesy of CERN.)
information about the location and characteristics of the

source. The power deposited by the impacting electrons depends on the trajectory as well as on the intrinsic properties
of the emitter.
In their basic theory of field emission (32), Fowler and
Nordheim (FN) showed that in the presence of an electric
field, electrons tunnel out of the metal into the vacuum because of their quantum wave-like nature. However, a comparison with the observed currents reveals that, at a given field,
emission is substantially higher than the FN predictions. Traditionally, the excess has been attributed to a field enhancement factor, which is believed to be related to the physical
properties of the emitter discussed below. Both RF and dc
studies reveal that emitters are micron- to sub-micron-size
contaminant particles (13). Figure 8 shows an example of a
region of emitting particles found in a niobium cavity (33).
The properties of the emitter that lead to enhanced emission
are (a) the microgeometry of the particle (34), (b) the nature
of condensed gases or adsorbates on the surface of the particle
(35), and (c) the interface between the particle and the underlying metal RF surface (36). Accordingly, a high level of cleanliness is necessary for cavity surface preparation. Field emission free performance has been achieved with HPR (30).
Recently, many 9-cell 1.3 GHz structures were prepared at
Emitter
Figure 7. Calculated electron trajectories in a 3-cell 1.5 GHz cavity
operating at Epk 50 MV/m. The emitter is located in the end cell,
where the surface electric field is 44 MV/m. Note that a significant
number of field-emitted electrons bend back and strike the wall near
the emitter. Others are accelerated through the cavity structure and
could produce unwanted dark current that may be accelerated in
adjacent cavities.
Figure 8. (a) SEM micrograph of field emitting particles. Note the

cluster of small spherical balls which indicate that a part of the site
melted. EDX analysis shows that the particles are stainless steel.
Note also the jagged microgeometry of the particles believed responsible for field enhancement. (b) The melted cluster is expanded.
654
Figure 9. SEM pictures of the processed site found at the location

predicted via temperature maps. (a) Low magnification; (b) high magnification of crater region within the starburst of (a). The molten
splashes in the crater region were found to contain indium, presumably from the indium wire seals used to make vacuum joints.
DESY by using HPR (37). A sample of their results is shown

in Fig. 11.
When raising the RF electric field in a superconducting
cavity for the first time, the field emission often decreases
abruptly; the cavity is said to process or condition. There
has been much progress in characterizing processed emitters
at a microscopic level using techniques such as SEM, EDX,
Auger, and AFM. These studies reveal that emitter processing is an explosive event that accompanies what we usually refer to as a spark or a discharge, or the electrical
breakdown of the insulating vacuum (38). Figure 9 shows a
typical SEM micrograph of an exploded emitting site (39).
To reach the highest accelerating fields, the highest thermal conductivity is essential to avoid thermal breakdown, and
a high level of cleanliness is essential to avoid field emission.
High-pressure water rinsing is a very successful cleaning
technique to avoid field emission. In multicell structures with
large surface area, there is always a significant probability
that a few emitters will eventually find their way on to the
cavity surface. There is also the danger of dust falling into
cavities during installation of power coupling devices as well
as during installing of the structure into the accelerator.
A technique that eliminates field emitters in situ is high
pulsed power RF processing (HPP) (40). The essential idea is
to raise the surface electric field at the emitter as high as
possible, even if for a very short time ( milliseconds). As the
field rises, the emission current rises exponentially to the
level at which melting, evaporation, gas evolution, plasma
formation, and ultimately a microdischarge (RF spark) take
place. The ensuing explosive event destroys the emitter. An
important benefit of HPP is that the technique can be applied
to recover cavities after their final installation. It can also
be used to recover the performance of cavities which may be
accidentally contaminated, as, for example, in a vacuum mishap. To achieve emission-free performance at a desired Eacc,
processing must be carried out at 2 Eacc. Figure 10 shows
the improvement in performance achieved by HPP (41). [Recently, many 9-cell 1.3 GHz structures were prepared at
DESY by using HPR techniques (37). A sample of their results is shown in Fig. 11. Occasionally it is possible to achieve
field emission free performance, as shown by the best curve
of Fig. 11.]
After HPP at 1 MW
Epk pulsed = 85 MV/m
Figure 10. Performance of a 5-cell 1.3

GHz niobium cavity improved by HPP.
Before HPP, the maximum field was limited by heavy field emission to Eacc 22
MV/m. After applying 1 MW of power and
reaching Epk 90 MV/m in the pulsed
mode, the field emission was processed
away and Eacc 28 MV/m was possible in
the CW mode.
1010
Tesla goal
1011
Before HPP
109
10
15
Eacc (MV/m)
20
25
30
magnetic surface field is given by
1011
Q0
Hmax =
1010
109
655
10
15
Eacc (MV/m)
20
25
30
Figure 11. High performance of several 9-cell 1.3 GHz cavities

achieved by high-pressure rinsing.
Overcoming Thermal Breakdown

The most effective cure for thermal breakdown caused by millimeter- to submillimeter-size defects is to (a) use better quality material that is free of such defects or (b) to raise the
thermal conductivity of the niobium so that remaining defects
will be able to tolerate more power before driving the neighboring superconductor into the normal state (42). A simple
analysis of the thermal breakdown shows that the maximum
4k(T T )
c
aRn
i.e., Hmax
k RRR
(29)
Here k is the thermal conductivity, Tc is the superconducting

transistor temperature, Tb is the bath temperature, a is the
radius of the defect, and Rn is the surface resistance of the
defect. This dependence on RRR is supported by detailed numerical simulations of thermal breakdown, as well as by experiments on cavities made from Nb of different RRR (Fig.
12).
Figure 13 shows the thermal conductivity of three samples
of niobium that have different histories of heat treatment
(43). The common feature of all three curves is the sharp drop
below Tc 9.2 K, as more and more electrons condense into
Cooper pairs. At the higher temperatures (4 K T Tc), a
significant, though small, fraction of electrons is not frozen
into Cooper pairs and can carry heat effectively, provided that
the electron-impurity scattering is low. Since the temperature
in the neighborhood of the defect is between the bath temperature and Tc, the high-temperature thermal conductivity is
the most important and has the strongest effect on thermal
breakdown. The higher the RRR, the higher the thermal conductivity in this temperature range.
Below 4 K, as electrons condense into Cooper pair, electronphonon scattering also decreases. As a result, the thermal conductivity from phonons begins to increase, leading to
the phonon peak near 2 K. With decreasing temperature, the
number of phonons decreases T3, The value of the phonon
conductivity maximum is limited by phonon scattering from
Eacc (MV/m)
100
Saclay 5-cells
CEBAF 5-cell
KEK 9-cell
Quench-limited
DESY 9-cell
Cornell 9-cell
Cornell 5-cell
DESY 9-cell Quench-limited
CEBAF 5-cell Power-limited
Cornell 5-cell FE and quench limit
10
RRR
World data on multicells
1
10
100
RRR
1000
Figure 12. A summary of the results of

multicell cavities [(3943) showing the
importance of high RRR coupled with
emission reduction techniques such as
HPP and HPR. The line shows a RRR
dependence expected from the simple theory of thermal breakdown.
656
1000
W
mK
500
RRR = 400
RRR = 250
RRR = 90
200
100
50
20
10
10
T (K)
Figure 13. Thermal conductivity () of niobium with RRR 90 (as
received), RRR 400 after post-purification with yttrium, and
RRR 250 after annealing the post-purified sample for 6 hours at
1400C. (Courtesy of Wuppertal.)
lattice imperfections, of which the grain boundary density is

the most important. If the crystal grains of niobium are very
large (e.g., because of annealing at high temperature), one
observes a large phonon peak, as shown in the thermal conductivity behavior of the sample with RRR 250, which was
annealed at 1400C. Since the phonon peak is at about 2 K,
it does not help to thermally stabilize defects that heat up in
the RF field.
The light, interstitially dissolved impurities have the
strongest effect on the RRR. Among these, oxygen is dominant. The other interstitials are carbon, nitrogen, and hydrogen. Among the metallic impurities, tantalum is found in the
highest concentration (typically 500 ppm by weight) since all
naturally occurring ores contain some tantalum. This impurity level is not a problem since tantalum is a substitutional
impurity and does not substantially affect the electronic properties. However, Ta can become a problem if clustering occurs.
The Nb used to fabricate a cavity has been checked, for example, by an eddy current scanning technique (44) to look for
large defects such as Ta clusters. Next in abundance are the
higher-temperature, refractory elements, such as tungsten,
zirconium, hafnium, and titanium, usually found at the level
of 1050 ppm wt. The electron-scattering effectiveness of the
various impurities are shown in Table 3 in terms of their effect on the RRR (45).
To obtain the net RRR, one must add the resistance contributions for each impurity element in parallel. The contributions of the phonons is always present, so that the highest
theoretical RRR for niobium is 35,000 (46). Experimentally,
the highest RRR ever achieved in a niobium sample was
28,000 (47).
The most convenient method to obtain high-purity niobium

for superconducting cavities is to remove the interstitials during the electron-beam melting stages of the ingot. Multiple
melts and progressive improvements in the furnace chamber
vacuum have led to a steady increase in the RRR of commercial niobium over the last decade from 30, typical of commonly
available reactor grade niobium, to 300 (48). The RRR of
commercially available Nb continues to improve. Recently, niobium sheet of RRR 500700 became available from a Russian source (49).
If RF surface magnetic fields higher than 50 mT are desired on a consistently reproducible basis, the thermal conductivity of the niobium must be improved to RRR 300. In
one method called post-purification, the purity of the niobium
is increased by solid-state gettering of oxygen using yttrium
(50) or titanium (51) at high temperature. The foreign metal
is vapor-deposited on the niobium surface. In the same step,
the high temperature decreases the diffusion time of the oxygen in niobium. Over a few hours, oxygen is trapped in the
deposited getter layer. If yttrium is used, the best temperature is 12001250C because both the vapor pressure of yttrium and the diffusion rate of oxygen in niobium are sufficiently high. If titanium is used, temperatures of
13501400C are required because of the lower vapor pressure of titanium. Typically during post-purification, the RRR
improves by a factor of two in a few hours. An important disadvantage of the post-purification is that the yield strength of
niobium falls substantially due to the high-temperature treatment. Also, titanium diffusion into the bulk along grain
boundaries demands additional etching.
Nb/Cu Cavities
As we mentioned, thin Nb films on higher-thermal-conductivity copper is another way to avoid thermal breakdown. The
technique of sputter coating niobium has been developed by
CERN for 350 MHz structures and applied successfully to
hundreds of structures (25). In the most successful coating
method to date, thin film deposition is carried out by cylindrical magnetron sputtering. Before the coating stage, the copper cavity is degreased, chemically polished ( 20 m), rinsed
with high-purity, dust-free water and alcohol, and dried under clean laminar air flow. After bakeout of the copper cavity
to reach a good vacuum, a typical coating time is 4 hours. The
coating thickness is a few microns at a substrate temperature
of 180200C. The RRR of the deposited niobium serves as
one of the monitors of film quality. The sputtering rate and
substrate temperature are optimized to reach an RRR greater
than 20. Note that the low RRR relative to bulk niobium is
not a problem because the film is very thin. The rod-like
grains of the niobium film are up to 1 m long and 10150
Table 3. Expected RRR for 1 ppm wt of Major Impuritiesa
Element
RRR
H
N
C
O
Ta (1000 ppm wt)
2640
4230
4380
5580
1140
Note that the effect of Ta is given in terms of 1000 ppm wt.
Future Directions, New Materials

Based on the fundamental aspects, for a material to be useful
in accelerators, the primary requirements are a high transition temperature and a high superheating critical magnetic
field. Among the elemental superconductors, niobium has the
highest Tc. While lead, coated on to a copper cavity, has been
very useful in early studies and heavy-ion accelerator applications, the higher Tc and Hc has made niobium the more attractive choice. Technical considerations, such as ease of fabrication and the ability to achieve uniformly good material
properties over a large surface area, have also proven favorable for niobium. The realm of superconducting compounds
has been much less explored because of technical complexities
that govern compound formation. In looking at candidates,
such as Nb3Sn, NbN, and the new high-temperature superconductors (HTS), such as YBa2Cu3O7, it is important to select
a material for which the desired compound phase is stable
over a broad composition range. With this criterion, formation
of the compound may prove more tolerant to variations in experimental conditions, which in turn would make it possible
to achieve the desired single phase over a large surface area.
With a Tc of 18 K, Nb3Sn is the most successful compound
explored to date (54). At low fields, residual resistance values
comparable to niobium have been achieved. However, the
maximum fields reached to date are far lower than those for
sheet niobium cavities. The new HTS are even further from
the performance level desired for application to accelerators.
Figure 14 compares the measured RF surface resistance at
low fields of several superconductors: HTS, Nb3Sn, and Nb
(55). The surface resistance of Nb at 1.3 K, Nb3Sn at 4.2 K,
and copper at 77 K are included for comparison.
Early enthusiasm over the remarkable strides made in the
transition temperature of HTS are now tempered with difficulties in achieving useful properties, such as a high critical
current density. The coherence lengths of the cuprates are
within the copperoxygen planes and 3 A
very short (17 A

perpendicular to the planes, respectively). There is also a
large anisotropy of the magnetic and electrical properties between the c axis and the ab planes, with superior behavior
when the current flow is in the ab plane. To produce goodquality HTS films, it is therefore necessary to orient the
grains so that the c axis is normal to the RF surface everywhere. This restriction will be a significant challenge for realizing HTS in existing accelerating cavity shapes. It is also
essential to have the right stoichiometry and oxygen content.
Because of the short coherence length, transport properties
1
YBa2Cu3O 77K-20K single
Crystalline films
Nb3Sn
Nb
101
102
103
Rs ()
nm in diameter. When studied with transmission electron microscopy, the individual grains show a high density of defects,
consisting of dislocations and point defect agglomerates (52).
The distance between two defects varies from 2 to 20 nm. The
onset Tc of as-deposited films is 9.6 K, but the transition
width is larger than for bulk niobium (typically a few tenths
of a kelvin). The large transition width (5 K in some cases) is
indicative of poor film quality.
Although Q0 values 1010 are obtained at low fields, the
RF losses of Nb/Cu cavities increase steadily with field. This
effect is attributed to intergrain losses in the niobium films,
which become more severe at higher frequency. Recently (53),
there is evidence to show that impurities buried in the films
can also account for increased losses at high fields.
657
Cu 77 K
104
105
Nb3Sn
4.2K
106
107
108
109
Nb 1.3K
0.1 0.2
0.5 1 2
5 10 20
RF frequency (GHz)
50 100
Figure 14. Measured surface resistance of HTS compared to the

same for Nb (1.3 K) and Nb3Sn (4.2 K). In the case of HTS the resistance is quoted at 77 K if it is already residual; if the resistance is
still decreasing with temperature, the residual is obtained at 20 K.
The solid lines show the calculated surface resistance of copper at 77
K, Nb3Sn at 4.2 K and Nb at 1.3 K.
are extremely sensitive to minute defects, such as grain

boundaries and their associated imperfections. Decoupling of
superconducting grains is believed to occur because the coherence lengths approach the scale of the grain boundary thickness, forming only weak links between individual grains. As
a result, the intergrain critical current is two to three orders
of magnitude lower than intragrain critical current. Even at
a clean grain boundary, the scale of the disorder that exists
from breaking up of a unit cell can exceed the coherence
length, especially in the direction of the c axis (56).
CONCLUSION
Even at the modest fraction of the ultimate potential, many
attractive applications are now in place, and new ones are
forthcoming. As our understanding of field limiting mechanisms continues to improve, new techniques emerge to further advance gradients, such as high-purity niobium to raise
the thermal conductivity, high-pressure rinsing to provide
cleaner, field emission free surfaces, and high pulsed power
processing to destroy residual emitters. The new techniques
for bulk niobium cavities have demonstrated that gradients
can be improved to between 20 and 30 MV/m in multicell
658
structures. If such gradients can be reliably achieved, exciting

new applications are on the horizon, such as the TeV electronpositron linear collider (57) or a multi-TeV muon collider (58).
BIBLIOGRAPHY
1. J. Preble, in B. Bonin (ed.), Proc. 7th Workshop RF Supercond.,
Gif-sur-Yvette, France, CEA/Saclay 96 080/1, 1995, p. 173.
2. G. Geschonke, in B. Bonin (ed.), Proc. 7th Workshop RF Supercond., Gif-sur-Yvette, France, CEA/Saclay 96 080/1, 1995, p.
143.
3. L. M. Bollinger, Annu. Rev. Nucl. Particle Sci., 36: 475, 1987.
4. R. Sundelin, IEEE Trans. Nucl. Sci., NS-32: 3570, 1985.
5. S. Takeuchi, in K. W. Shepard (ed.), Proc. 3rd Workshop RF
Supercond., Argonne National Laboratory, Argonne, IL, ANLPHY-88-1, 1988, p. 429.
6. H. Padamsee, K. Shepard, and R. Sundelin, Annu. Rev. Nucl. Particle Sci., 43: 635, 1993.
7. V. L. Ginsburg and L. D. Landau, Zh. Eksperim. i. Theor. Fizike,
20: 1064, 1950.
8. U. Klein and D. Proch, in J. S. McCarthy and R. R. Whitney (eds.),
Proc. Conf. Future Possibilities Electron Accelerators, Charlottesville, University of Virginia, P. N1-17, 1979.
9. P. Kneisel, R. Vincon, and J. Halbritter, Nucl. Instrum. Meth., 188:
669, 1981.
10. D. Moffat, in Y. Kojima (ed.), Proc. 4th Workshop RF Supercond.,
KEK, Tsukuba, Japan, Rep. 89-21, 1990, p. 445.
11. J. Delayen and K. W. Shepard, Appl. Phys. Lett., 57 (5): 514, 1990.
12. H. Padamsee, in High Voltage Vacuum Insulation, Latham (ed.),
New York: Academic Press, 1995, p. 431.
13. J. Delayen, in Y. Kojima (ed.), Proceedings of the 4th Workshop on
RF Superconductors, KEK, Tsukuba, Japan, Rep. 89-21, 1990, p.
249.
14. U. Laustroer, U. van Rienen, and T. Wieland, DESY M-87-03,
1988.
15. R. Klatt, DESY M-86-07, 1987.
16. J. Knobloch and H. Muller, Rev. Sci. Instrum, 65 (11): 3521, 1994.
17. J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev., 108:
1175, 1957.
18. D. C. Mattis and J. Bardeen, Phys. Rev., 111: 412, 1958.
19. J. P. Turneaure, Ph.D. thesis, Stanford University, 1967; see also
J. M. Pierce, in L. Marton (ed.), Methods of Experimental Physics,
Vol. 11, 541, 1974.
20. J. Halbritter, Z. Physik, 238: 466, 1970.
21. See the survey of low field residual resistance values in J. P. Turneaure, Proc. Appl. Supercond. Conf., Annapolis, 1972, p. 621.
22. C. Vallet et al., in E. H. Henke et al. (eds.), Proc. 1992 European
Particle Accelerator Conf., Editions Frontieres, 1992, p. 1295.
23. B. Bonin and R. Roeth, in D. Proch (ed.), Proc. 5th Workshop RF
Supercond., DESY, Hamburg, Germany, DESY-M-92-01, 1991,
p. 210.
24. J. Kirchgessner, in K. W. Shepard (ed.), Proc. 3rd Workshop RF
Supercond., Argonne National Laboratory, Argonne, IL, ANLPHY-88-1, 1988, p. 533.
25. C. Benvenuti, in D. Proch (ed.), Proc. 5th Workshop RF
Supercond., DESY, Hamburg, Germany, DESY-M-92-01, 1991,
p. 189.
26. Aschroft and Mermin, Solid State Physics, W. B. Saunders, 1976,
p. 20.
27. L. P. Kadanoff and P. C. Martin, Phys. Rev. 124: 670, 1961.
28. H. Padamsee, J. Knobloch, and T. Hays, RF Superconductivity for
Accelerators, Wiley, 1998.
29. P. Bernard et al., in H. Henke et al. (eds.), Proc. 3rd Eur. Particle
Accelerator Conf., Editions Frontieres, 1992, p. 1269.
30. P. Kneisel, B. Lewis, and L. Turlington, in R. M. Sundelin (ed.),
Proc. 6th Workshop RF Superconductivity, CEBAF, Newport
News, VA, 1994, p. 628.
31. H. Padamsee, J. Tuckmantel, and W. Weingarten, IEEE Trans.
Magn., Mag-19: 1308, 1983.
32. R. H. Fowler and L. Nordheim, Proc. R. Soc. London, A119: 173,
1928.
33. J. Knobloch et al., in Proc. 1995 Particle Accelerator Conf., Dallas,
1995, p. 1623.
34. M. Jiminez et al., J. Phys. D: Appl. Phys. 27: 1038, 1994.
35. Q. S. Shu et al., in IEEE Trans. Magn. 25: 1868, 1989.
36. N. S. Xu, in R. V. Latham (ed.), High Voltage Vacuum Insulation,
Academic Press, 1995, p. 116.
37. W. D. Moeller and M. Pekeler, in S. Myers et al. (ed.), Proc. 5th European Particle Accelerator Conf., Barcelona, Spain, IOPP Publishing, Bristol, 1996, p. 2013.
38. D. Moffat et al., Particle Accel., 40: 85, 1992.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
J. Graber et al., Nucl. Instrum. Meth. Phys. Res., A 350: 582, 1994.
J. Graber et al., Nucl. Instrum. Meth. Phys. Res., A 350: 572, 1994.
C. Crawford et al., Particle Accel., 49: 1, 1995.
H. Padamsee, Proc. 1st Workshop RF Supercond., KFK, Karlsruhe,
Germany, KFK-3019, 1980, p. 145.
G. Mueller, in K. W. Shepard (ed.), Proc. 3rd Workshop RF Supercond., Argonne National Laboratories, Argonne, IL, ANL-PHY88-1, 1988, p. 331.
W. Singer, in E. Palmieri (ed.), Proc. 8th Workshop on RF Supercond., Abano Terme, Italy, 1997, in press.
K. K. Schulze, J. Metals, 33: 33, 1981.
K. Schulze, Niobium, in H. Stuart (ed.), Proc. Int. Symp., San Francisco, The Metallurgical Society of AIME, 1981, p. 163.
A. Gladun et al., J. Low Temp. Phys., 27: 873, 1977.
H. Padamsee, in R. M. Sundelin (ed.), Proc. 6th Workshop RF Supercond., CEBAF, Newport News, VA, 1994, p. 515.
A. V. Elyutin, et al., in D. Proch (ed.), Proc. 5th Workshop RF Supercond., DESY, Hamburg, Germany, DESY-M-92-01, 1991, pp.
354, 426.
50. H. Padamsee, in H. Lengeler (ed.), Proc. 2nd Workshop RF Supercond., CERN, Geneva, Switzerland, CERN, 1984, p. 339.
51. P. Kneisel, J. Less Common Metals, 139: 179, 1988.
52. C. Durand and W. Weingarten, IEEE Trans Appl. Supercond., 5:
1107, 1995.
53. S. Calatroni, in E. Palmieri (ed.), Proc. 8th Workshop RF Supercond., Abano Terme, Italy, 1997, in press.
54. M. Peiniger, in K. W. Shepard (ed.), Proc. 3rd Workshop RF Supercond., Argonne National Laboratory, Argonne, IL, ANL-PHY88-1, 1988, p. 503.
55. D. Busch et al., in R. M. Sundelin (ed.), Proc. 6th Workshop RF Supercond., CEBAF, Newport News, VA, 1994, p. 173.
56. G. Mueller, in Y. Kojima (ed.), Proc. 4th Workshop RF Supercond.,
KEK, Tsukuba, Japan, Rep. 89-21, 1990, p. 267.
57. R. Brinkmann, in Proceedings of the 1995 Particle Accelerator Conference, Cat. No. 95CH35843, 1995, p. 674.
58. Collider, A Feasibility Study, BNL-52503, 1996.
59. P. Wilson, in R. A. Carrigan, F. R. Huson, and M. Month (eds.),
Physics of High Energy Particle Accelerators (Fermlab Summer
School, 1981), AIP Conf. Proc., no. 87, American Institute of
Physics, 1982.
HASAN PADAMSEE
Cornell University
SUPERCONDUCTING COILS. See SUPERCONDUCTING

MAGNETS, QUENCH PROTECTION.
659
SUPERCONDUCTING TRANSFORMERS
39
; ;;
;; ;
Primary
Core
Secondary
In an electrical power system, power is generated far away

from the consuming areas. The electric power is transmitted
from generating locations to consuming locations through
transmission lines. A high voltage is desirable for transmitting large amounts of power in order to minimize the current
and the associated I2R losses, and reduce the amount of conductor used in transmission lines. A much lower voltage, on
the other hand, is required for distribution, for various reasons connected with safety and convenience. The transformer
makes the reduction in voltage easy and economically possible. Generally, electricity is transformed three or four times
between the location of generation and the location of consumption making transformers one of the basic elements of
an electric power system.
The physical basis of the transformer is mutual induction
between two circuits linked by a common magnetic field, as
shown in Fig. 1. The power transformer transfers electrical
energy from one circuit to another, via the medium of the
pulsating mutual magnetic field. Magnetic iron enhances the
flux linkage between the circuits. The transformer coils are
therefore made to embrace an iron core, which serves as a
conduit for the mutual magnetic flux, ensuring that the flux
links each coil fairly completely. The use of an iron core permits greater freedom in shape and relative position of the primary and secondary coils (Fig. 2), since the majority of the
mutual flux is conveyed by the core regardless of the relative
positions of the two sets of coilsprimary and secondary.
Since transformers are employed extensively in a power
system, their efficiency and losses are considered a serious
issue. The transformer design selection is normally made on
the basis of its lifetime cost which consists of the initial cost
plus the cost of operating it over its lifetime. The lifetime cost
; ;
; ;
1
Primary
T1
T2
1M
Secondary
Figure 1. Flux linkage between primary and secondary in air.
1M
Figure 2. Flux linkage between primary and secondary with iron

core.
of even a small loss could be significant. In addition, most

larger transformers employ oil for cooling the windings and
the iron core. Although the oil is an excellent cooling medium
and is a good high voltage insulator, it has attracted the ire
of environmentalists and fire departments due to the possibility of oil spills and fire hazards. Moreover, larger units are too
heavy and bulky for normal transportation channels. Space in
an urban environment is quite valuable. A more compact,
light-weight transformer could more easily be sited, possibly
even above the ground floor or basement levels. This could be
a significant advantage. Some utilities have indicated that
they might be willing to pay a premium for such an advantage. These difficulties have inspired designers and users to
look for alternative transformer solutions. Superconducting
transformers appear to offer a solution to most of these
problems.
A superconductor only operates within a space bounded by
three parameters: current density in the superconductor,
magnetic field experienced by the superconductor, and its operating temperature. The maximum operating current of a
given superconducting wire is a function of these parameters.
If any of these parameters is violated, the superconducting
wire loses its superconducting property and becomes resistive. Once in the resistive state, it generates joule heating.
This heating must be limited to a safe value in order to prevent permanent damage to the windings. The circuit breaker
feeding the transformer could be used to disconnect it if the
winding temperature exceeds a given upper limit. Since superconductors can sustain large current densities with potentially low losses, a superconducting transformer is expected
to be smaller, lighter, and more efficient. In addition, since
the superconducting transformer uses cryogenic liquids as dielectric and coolant, it is also free of environmentally unacceptable oil. During the eighties, several groups designed,
built, and tested small transformers employing low temperature superconductors (LTS) NiobiumTitanium (NbTi) superconducting windings (16) cooled with liquid helium to
around 4 K. Several problems were observed:
First, the superconductor must operate in the presence
of fluctuating ac currents in a moderately high magnetic
field resulting in decreased stability and increased ac
losses.
40
Second, under system fault conditions, if the critical current of the conductor is exceeded, recovery to the superconducting state is too slow to allow automatic circuit
reclosure.
Third, the need to cool with liquid helium reduces economic benefits of such a device.
The discovery of high temperature superconductors (HTS)
has revived interest in superconducting transformers. Presently, attempts are being made to design, build, and test
transformers with HTS windings cooled with liquid nitrogen
at 77 K. These HTS transformers are less likely to have the
problems associated with NbTi transformers. For example,
the use of HTS material greatly improves stability; the transformer could be designed to prevent quenching (i.e., transitioning to the normal state), and the consequences of ac losses
could be potentially overcome by developing low-ac-loss HTS
conductors and operating it at 77 K. References 79 have
summarized the evolution of the superconducting transformers. Recently two HTS prototype transformers have been
testeda 630 kVA, three-phase transformer by ABB (10) and
a 500 kVA, one-phase transformer by Kyushu University (11).
This article discusses potential system benefits associated
with superconducting transformers, and reviews and summarizes the design requirements for such a device. It also compares and contrasts the requirements for transformers utilizing LTS and HTS conductors.
SUPERCONDUCTING TRANSFORMER CONFIGURATION
A superconducting transformer operates using the same principles and constituent parts as found in a conventional transformer. Both employ an iron core to contain magnetic flux,
and primary and secondary windings to exchange power.
However, the construction of the two types of transformers
is quite different. For example, in conventional transformers
primary and secondary coils are directly wound on the iron
core and both iron core and coil assemblies are immersed in
a tank filled with oil that cools both the iron core and the
coils. On the other hand, in superconducting transformers,
the iron core is usually maintained at room-temperature
while the superconducting coils operate at cryogenic temperatures. The decision to maintain the iron core at room temperature is dictated by the fact that the iron core losses, due to
hysteresis and eddy-currents, are substantial (almost 1 watt
per pound) and they go up when the iron core is operated at
cryogenic temperatures (12). The iron core permeability also
goes down at low temperature which means more iron core is
required to carry the same flux at low temperature than at
room-temperature. The core losses would represent a major
load on the refrigerator if the iron core were operated at cryogenic temperature. On the other hand, the superconducting
windings must be cooled to cryogenic temperatures (between
4.5 K and 77 K) which necessitates that these windings be
enclosed in containers which could hold vacuum or cryogen or
both. These containers surround the iron core limbs and take
the shape of hollow donuts. Since they surround the iron core,
they must be constructed from nonmetallic material lest they
form a closed circuit around the iron core and thus form a
shorted secondary for the transformer, making the transformer unworkable. It is possible to employ metallic contain-
ers but a dielectric break must be included in the circumferencial direction to prevent flow of current in the container
walls. The dielectric break makes these metallic cryogen containers more expensive and less reliable. The superconducting windings must be cooled with suitable cryogen (liquid
helium for NbTi windings, liquid nitrogen for HTS BiPbSrCaCuO-2223 (BSCCO-2223) windings, or an intermediate temperature for Nb3Sn or BSCCO-2212 windings). Reference 13
describes the status of the HTS conductor technology and (14)
discusses coils made from BSCCO-2212 material. Since these
containers hold windings at low temperatures, their walls
must be thermally insulating. For low temperature operation
at around 4 K, a double wall construction is employed with
multi-layer-insulation (MLI) insulation in the vacuum space
between the walls. Additionally, an intermediate temperature
(77 K) shield is also inserted between warm and cold walls of
the cryostat. This makes the container design complex and
expensive. On the other hand, if the windings operate at 77
K then single wall construction could be used for these containers. This makes design and construction simpler and the
cost is substantially lower as compared to those of the low
temperature coils. The cost of refrigerator (both capital and
lifetime) is also much lower for devices operating at 20 K to
77 K than those operating at 4 K.
TRANSFORMER DESIGN AND ANALYSIS
The transformer design is obtained with an optimization process which involves varying several significant parameters
which are interrelated in complex ways. A transformer is
sized on the basis of its power rating, voltage, number of
phases, frequency, and short-circuit reactance. It is also necessary to pay close attention to transformer type, service conditions, cost of losses, and the relative costs of conductor, iron,
insulation, labor, machinery, and configuration. All these factors are considered when designing a superconducting as well
as normal transformer.
Design Issues
It is possible in principle to construct a superconducting
transformer without an iron core. Such transformers are
characterized by a larger reduction in losses, size, and weight
than those employing iron core but they require much larger
excitation current (1516). On the other hand, an iron core
offers the following benefits:
The core contains the mutual flux between the windings,
and thus, reduces stray field which has adverse impacts
on people and other equipment in the vicinity of the
transformer
The magnetic field experienced by the superconducting
windings is reduced, thus reducing the amount of superconductor required
The ac losses in the windings are reduced, which reduces
the size of the refrigeration system
Since losses in the iron core are large, it is normally preferable to keep the iron core at room-temperature. This requirement to operate the iron core at room-temperature forces superconducting transformer configurations which differ from
those of conventional transformers.
Iron core
Outer wall of
cryostat
Cold wall of
cryostat
Superconducting
windings
Figure 3. HTS transformer configuration.
A generic three-phase superconducting transformer is

shown in Fig. 3. It has a set of concentric primary and secondary windings surrounding each leg of a three-leg transformer
core. Since the windings operate at cryogenic temperatures,
primary/secondary winding pairs are enclosed in individual
cold containers (identified as cold wall of cryostat in Fig. 3).
This cold wall must be made of nonconductive materials in
order to prevent a shorted turn. A common warm wall encloses all three cold wallsthis wall could be metallic.
Conductor Concepts
Superconductors being considered for transformers could be
divided into two broad categories on the basis of their operating temperatures. LTS operate at liquid helium temperature (4.2 K) and HTS operate at temperatures ranging from
20 K to 77 K (the temperature of liquid nitrogen). Two LTS
conductor options are available; NiobiumTitanium (NbTi)
and NiobiumTin (NB3Sn). Likewise, two HTS conductor
choices are available in long lengths; BSCCO-2212 for operation at 20 K to 30 K, and BSCCO-2223 for operation 50 K
to 77 K. This section discusses characteristics of these conductors and their pros and cons for application in a superconducting transformer.
Niobium-Titanium Conductors. Superconducting NbTi wires
for 50 Hz to 60 Hz applications require very small diameters
in order to minimize ac losses and improve intrinsic stability
(1719). These NbTi wires are characterized by very fine filaments (0.1 m or less), high resistance CuNi matrix, and
small wire diameter (0.2 mm). A practical size conductor,
capable of carrying hundreds of amperes, consists of many
such wires.
Primary and secondary coils made with this conductor are
generally housed in a common container filled with liquid helium. Since it is expensive to remove heat generated at liquid
helium temperature (4 K) with a refrigerator, designs are always optimized to minimize heat generated at low temperatures. Since the specific heat of metals is very low at the liquid helium temperatures, a small heat input forces the
conductor into its normal conducting state. Protection against
the consequences of an unexpected quench is one of the most
significant challenges for superconducting coilsthe intensive and localized Joule heating can produce catastrophic
41
damage (20). Moreover, the protection of ac windings is more

critical than of dc windings, because of high matrix resistivityit is not permissible for wire to carry its nominal current
longer than a few milliseconds, otherwise permanent damage
could occur. Thus, transformers employing NbTi windings
must be designed carefully to avoid these problems. Similar
design approaches must be used for Nb3Sn conductors. These
conductors have higher critical temperature than NbTi and it
is therefore possible to operate them at higher temperatures
(10 K) as compared to 4 K for NbTi. Nevertheless, most
problems associated with low temperature persist. The lack
of economic feasibility and high cost of refrigeration stopped
LTS transformer activities.
BSCCO Conductors. Although BSCCO conductors have a
potential of operating at 77 K and being cooled with environmentally friendly liquid nitrogen, no suitable design of an
ac conductor exists at this time. The major problem is posed
by high aspect ratios (1 : 10) of such conductors as shown in
Fig. 4. Since hysteresis losses are directly proportional to filament diameter (17), losses induced by magnetic field perpendicular to the wide surface of the conductor are high and they
pose coil cooling and overall transformer efficiency challenges.
Furthermore, since a single strand of BSCCO wire could only
carry limited current, it is necessary to wind several strands
in parallel. If these strands are not transposed they can cause
significant coupling losses. Iwakuma (21) has proposed a
transposition scheme to minimize these coupling losses.
Losses due to field parallel to the wide face of the conductor
are likely to become acceptable.
The perpendicular field losses only occur in the end regions
(10% of the coil axial length) of coils and with clever winding
schemes, it might be possible to minimize these losses. Research is continuing at ASC and other places around the
world but no one has yet published a credible conductor design. Until a credible HTS conductor is developed for ac application, the HTS transformer will remain elusive.
Since the specific heat of metals is high at higher temperatures, it is possible to absorb larger amount of heat at the
higher operating temperatures of HTS for a modest temperature rise of the conductor. Nevertheless, ac losses in presently
available conductors are still unacceptable due to cost/efficiency considerations. ABB has made a 630 kVA transformer
(10,22) which employed BSCCO-2223 wire from ASC but its
ac losses are unacceptably high for commercial deployment.
On the other hand, this device has demonstrated that if a
suitable HTS wire was available then it would be possible to
Silver
sheath
Superconducting
filaments
Silver
matrix
Wire
Figure 4. Highly aspected HTS conductor.
42
a radius of filament (m)

f frequency (Hz)
The largest loss component is usually hysteresis loss in the
superconductor filaments. One way to reduce these losses is
to make the superconducting filament diameter (2 a) as small
as possible.
Another component of ac losses is due to coupling between
filaments. This coupling takes place when the electric field
between adjacent filaments is sufficiently large to cause a current flow between filaments through the matrix. Filaments
are tightly twisted in a helical fashion and are surrounded by
high resistivity matrix in order to reduce these coupling currents and the associated losses. The coupling losses are given
by the following equations taken from (25):
2 2 2
c
l B
t
d
e

2
c
lt2 B2
1
Pc =
d
e
16
Figure 5. Model coil made by ABB, from wire provided by American

Superconductor Corporation, for the 630 kVA transformer project in
Geneva, Switzerland.
construct a practical transformer. Figure 5 (taken from Ref.

22) shows a Model coil made by ABB, from wire provided by
American Superconductor Corporation, for the 630-kVA
transformer project in Geneva, Switzerland.
Coated conductors employing YBCO films on a substrate
are emerging as an alternative to BSCCO conductors. High
critical current in HTS films deposited on crystellographically
oriented substrates has been demonstrated recently by Los
Alamos National Laboratory (LANL) (23) and independently
by Oak Ridge National Laboratory (ORNL). Joint industrylaboratory programs are underway to scale up this coated
conductor technology. A number of technological problems
must be solved before a practical conductor could emerge. Production conductors are expected by a 2002 time frame.
The coated conductor promises significantly higher performance than the BSCCO, with projected cost/performance below the much discussed $10/kA-m commercialization benchmark. Overall strand current densities of up to 50,000 A/cm2
are expected to be achievable. In addition, the coated conductor films may ultimately be engineered to optimize filament
dimensions and to eliminate filament coupling through the
careful selection of layers within the conductor architecture.
Ac Losses
Losses in a superconducting coil are quite small under dc operation. However, these losses become significant if the coil
current is ramped rapidly or if it carries ac as in a transformer. The ac losses are generally quite small, but the refrigeration penalty amplifies their effect. When the applied
field is low, superconductor tends to screen penetration of
field into interior of the conductor. However, higher fields
fully penetrate a conductor. Under such a condition, hysteresis loss (watt) is given by the following formula (17,24) for
decoupled filaments:
Qh = BJc af
where
B field variation (peak-to-peak) (T)
Jc critical current density of superconducting filament
(A/m2)
Pc =
144
(1)
(2)
where
Pc loss per unit volume of the conductor (W/m3)
c and d conductor cross-sectional dimensions (width and
thickness) (m)
lt twist pitch length (m)
B rate of change of magnetic field (T/s)
e matrix resistivity (-m)
Equation (1) is used when the field is parallel to the wider
face Eq. 2 of the conductor d c. Eq. (2) is for the case when
c d.
It has been shown (26,27) that NbTi coils operating at liquid helium temperature must have a filament diameter on
the order of 0.1 m or less to make these coupling losses comparable to the losses in copper windings of a conventional
transformer.
Both of these ac loss phenomena also apply to HTS conductors. HTS conductors are currently made in highly aspected
tape shapes (1 : 10) as shown in Fig. 4. The hysteresis losses
due to magnetic field parallel to the broad face of the conductor are acceptable since the dimension transverse to the field
is small, but losses due to magnetic field perpendicular to the
broad face of the conductor are excessively large. Although
several industrial groups are attacking this problem, no
method has yet been published for making a conductor capable of carrying large currents (comparable to those of LTS)
while keeping hysteresis and coupling losses low. HTS conductors, however, have a couple of significant advantages over
the LTS conductorsthe temperature rise due to a transient
heat input (i.e., by conductor movement) is lower at the
higher HTS operating temperatures than when operating at
liquid helium temperature, and they have a slow transition
from superconducting to normal state which makes them inherently more stable. This advantage may translate into relaxed requirements on the acceptable filament size and other
conductor configuration parameters. However, the ac losses in
these wires must be low in order to make HTS transformers
economically acceptable.
Another significant loss component is the heat conduction

through current leads. One end of a current lead is at room
temperature and the other end is at low temperature, and the
heat is conducted along the length of the lead from warm to
cold regions. In case of LTS transformers, conduction heat can
be intercepted at the intermediate thermal shield, which is
usually kept at 70 K. The heat conduction between the 70
K and 4 K winding regions could be controlled by employing
HTS current leads. However, in case of HTS transformers operating at the liquid nitrogen temperature (77 K), all of the
lead conduction must be cooled by the liquid nitrogen coolant.
A typical 100 A pair of leads conducts 8 W of heat load into
the cold region.
Cryostat
To maintain the low temperature environment essential for
operation of LTS and HTS magnets, they must be placed in
special vessels or cryostats. These are vacuum insulated containers. Designs for LTS and HTS could be widely different;
HTS cryostats are likely to be easier to design and fabricate
than LTS cryostats. Since the cost of removing losses from
low temperatures (4.2 K) is very high, usually a double or
triple wall construction is employed. The innermost space
contains the liquid helium and outermost wall operates at
room temperature. An intermediate wall is normally employed which operates at an intermediate temperature such
as liquid nitrogen temperature (77 K). In some applications,
even another wall is introduced at 20 K to 30 K in order to
minimize total refrigeration load.
On the other hand, a HTS cryostat operating at 77 K could
employ a simple double wall construction. The outer wall is
at room-temperature and the inner wall is at liquid nitrogen
temperature. The space between the two walls is filled with
multi-layer-insulation (MLI) or some other suitable thermal
insulation such as various types of foam. Normally this simple construction reduces heat leak to an acceptable level.
Cooling System
Superconducting magnets can be cooled with a pool of liquid
cryogen or cooled by conduction with a cryocooler. The majority of LTS magnets are cooled with liquid helium. Helium has
the lowest boiling point (4.2 K) at atmospheric pressure of
any known cryogen and has been the cryogenic fluid used in
the LTS magnets. On the other hand, cost of helium is very
high ( $5/liter). Because of its cost, most facilities install
a recovery and reliquefying system. A variety of devices are
available worldwide. The choice of device is application specific as the designer must evaluate trade-offs. Primary among
these are first-cost versus operating cost; first-cost versus reliability and ease-of-use, and dollar per watt of cooling required. Devices are basically two types; open-cycle or closedcycle. The former are the simplest, lowest cost form of refrigeration available today and are quite simply open top
bucket dewars. Closed-cycle systems do not require use of
liquefied cryogens but rely on the refrigeration capacities of
the gas and the cycle design to achieve the cryogenic temperatures desired. Helium is the predominant gas used in these
devices. For larger applications such as a transformer, a
closed cycle helium system is substantially more economical
in long run than purchased liquid helium. With a closed cycle
system, the warm helium returning from the cryostat is reliq-
43
uified in a refrigerator and returned to the cryostat. A number of conduction cooled magnets which use no liquid cryogen
have been built for operation at 4 K and higher temperatures
using both HTS and LTS wires.
Economic Considerations
From a utility perspective, a transformer must have low initial and operating costs, and be light weight, compact, and
environmentally benign with a lifetime of typically 30 years.
To a great extent, a HTS transformer does have a potential
to offer these advantages.
Conventional transformers are highly reliable and flexible
in terms of their use in an electrical system. However, the
dominant component of losses is the I2R loss in the windings.
The capitalized cost of these losses over the life of a transformer could easily exceed its initial cost. The superconducting transformers are attractive because of the potentially
lower winding losses. However, there is an energy penalty associated with the input power consumed by the refrigeration
system. This energy penalty can be substantial. To remove 1
W of losses at 4 K, 500 W of refrigeration power is required.
However, only 20 W is required to remove 1 W from 77 K.
This reduction in the refrigeration power has generated a lot
of interest in transformers employing HTS conductors. An optimally designed HTS transformer is likely to have lower
losses and lower life cycle cost than a conventional unit.
The higher current density capacity of superconductors
compared to copper leads to a more compact and lighter design of transformers. Even for the identical core diameters,
the core window width (space between iron legs) could be reduced in proportion to the space saving due to the utilization
of superconducting windings. This reduces the iron core
weight. Lighter core size also leads to lower core losses. A
compact and light weight transformer might see new applications which were not feasible with the conventional transformers. Lower weight and compact size would make them
acceptable for more urban applications. Smaller core windows
also lead to lower leakage inductance which helps to improve
dynamic stability of a power system. The low leakage inductance also improves the voltage regulation to the load, and
therefore, it might eliminate complex and expensive tap
changers.
The environmentally friendly aspect of a superconducting
transformer gives additional impetus to application of these
devices. They do not have environmental, health, and safety
concerns associated with conventional transformers. In urban
settings, most transformers are installed in the basement of
high rise buildings. Environmental concerns are forcing utilities to employ oil free transformers. In a superconducting
transformer, oil is replaced with liquid helium or liquid nitrogen. They are much more benignnontoxic, nonflammable,
and noncarcinogenic. However, in the event of a quench, a
large quantity of helium or nitrogen gas could be released
which could displace oxygen in the surrounding air and present a personnel risk. This risk can be mitigated by controlled
release and installation of exhaust fans.
STATE-OF-THE-ART OF SUPERCONDUCTING TRANSFORMERS
With the advent of HTS conductors, the low temperature
transformer design efforts have essentially been abandoned.
44
Several HTS transformers programs are currently active in

Japan, Europe, and North America for operation at temperatures ranging from 20 K to 77 K. Both air-core and iron-core
options are being pursued. The major HTS programs being
pursued are listed below.
ABB has built (2830) and has been testing a 630 kVA,
3-phase transformer since March 1997. This is the
worlds first demonstration of an HTS transformer, one
that was designed and built by ABB with HTS wire that
was developed and manufactured by ASC. Following the
success of this transformer, the Electricite de France
(EDF), ABB, and ASC consortium is now developing a 10
MVA transformer (29). The 10 MVA transformer is a crucial next step on the path to a commercial-scale 30 MVA
transformer. The 10 MVA unit will be built by ABB and
will be tested by EDF in its grid by the end of 2000.
Waukesha Electric and Intermagnetic General Corporation (IGC) is building a 1 MVA transformer (8). Waukesha will build the unit using HTS wire supplied by IGC.
This unit is planned for testing in 1998.
Kyushu University consortium designed, built, and
tested a 500 MVA-class single phase HTS transformer
(11) in 1996. This was the first transformer cooled by liquid nitrogen at 77 K and was operated at a steady state
with a 500 kVA secondary inductive load.
Although HTS conductors are much more tolerant of transient heat input (primarily due to increased heat capacity of
materials at higher temperatures than at 4.2 K), the ac losses
are still significant and an attractive ac conductor configuration is still illusive. The highly aspected configuration of HTS
wire (width 10 thickness) generates excessive hysteresis
losses caused by the magnetic field component perpendicular
to the surface of the conductor. These losses are sufficiently
high that the total refrigeration power needed to remove them
from 77 K environment is comparable to the I2R loss in the
windings of a conventional transformer. Thus, the efficiency
(or life cycle cost) advantage is lost. How the ac loss problem
will be solved and how the market will value the benefits of a
superconducting transformer over conventional transformers
is not yet clear. By the year 2002, a clearer picture might
emerge about the feasibility of HTS transformers.
BIBLIOGRAPHY
1. H. Riemersma et al., Application of superconducting technology
to power transformers, IEEE Trans. Power Appar. Syst., PAS100: 33983405, 1981.
2. Westinghouse Electric Corporation, Application of low temperature technology to power transformers, US Department of Energy, DOE/ET/29324-1, February 1982.
3. A. Fevrier et al., Preliminary tests on a superconducting power
transformer, IEEE Trans. Magn., MAG-24: 14771480, 1988.
4. S. Hornfeldt et al., Power transformer with superconducting
windings, IEEE Trans. Mag., MAG-29: 35563558, 1993.
5. M. Iwakuma et al., Quench protection of superconducting transformers, IEEE Trans. Magn. MAG-27: 20802083, 1991.
6. E. S. Yoneda et al., Tests on a 30 kVA class superconducting
transformer, Cryogenics, 31: 655659, July 1991.
7. R. F. Giese, The status of progress towards high-temperature superconducting transformers, Work sponsored by the Interna-
tional Energy Agency, Argonne National Laboratory, November

1996.
8. S. P. Mehta et al., Superconductors transforming transformers,
IEEE Spectrum, 4349, July 1997.
9. R. C. Johnson, Status of superconducting power transformer development. Proc. Amer. Power Conf., Chicago, April 1996, 58-1:
89, subject 05-6, Illinois Institute of Technology.
10. D. Mikulis, HTS transformer for utility powering, Modern Power
Syst., June 1997.
11. K. Funaki et al., Preliminary tests of a 500 kVA-class oxide superconducting transformer cooled by subcooled nitrogen, IEEE
12. P. K. Mukhopadhyay et al., Unexpected behaviour of core materials of electrical coils at low temperatures, Cryogenics, 34 (3):
241, 1994.
13. A. P. Malozemoff et al., Progress in BISCCO-2223 tape technology, M2S Conf., Beijing, China, February 28March 3, 1997.
14. M. S. Walker et al., Performance of coils wound from long lengths
of surface coated, reacted. BSCCO-2212 conductor, IEEE Trans.
Appl. Supercond., 7: 889892, 1997.
15. M. Yamamoto, A Study on a Coreless Superconducting Transformer, IEEE Trans. Appl. Supercond., 3: 889, 1993.
16. N. Okada et al., Fabrication and test of superconducting air-core
autotransformers, IEEE Trans. Magn., MAG-28: 430433, 1992.
17. M. N. Wilson, Superconducting Magnets, Clarendon Press, Oxford, 1983.
18. T. Verhaege et al., A new class of ac superconducting conductors,
19. K. Kamisada et al., Research and development of superconductors for commercial frequency, IEEE Trans. Appl. Supercond., 28:
291, 1992.
20. T. Veraege et al., Protection of superconducting ac windings,
IEEE Trans. Applied Supercond., 28: 751, 1992.
21. M. Iwakuma et al., Electromagnetic properties in parallel conductors composed of Bi2223 multifilamentary wires for power
transformer windings, Appl. Supercond. Conf., August 25, Pittsburgh, PA, 1996.
22. A. M. Wolsky, IEA implementing agreement for a co-operative
programme for assessing the impacts of high-temperature superconductivity on the electric power sector, Argonne National Laboratory, January 1997.
23. S. Foltyn et al., 95 Spring MRS Meeting in San Francisco, California.
24. H. Brechna, Superconducting Magnet System, Berlin: SpringerVerlag, 1973.
25. K. Kwasnitza, AC losses of superconducting high-Tc multifilamentary Bi-2223/Ag sheathed tapes in perpendicular magnetic
fields, Physica C 233: 423435, 1994.
26. J. H. Murphy, Advanced applications of superconductors, Advances in Cryogenic Engineering, 29: 133, 1984.
27. Cha Gueesoo et al., Effect of the longitudinal magnetic field in
a 2nd level superconducting cable, Trans. KIEE, 46 (1): 5257,
January 1997.
28. ABB and American Superconductor Corporation team up to demonstrate the worlds first installation of a high temperature superconductor transformer, an ASC press release, Westborough,
MA, March 12, 1997.
29. Electricite de France, ABB, and American Superconductor sign
$15 million pact to accelerate HTS wire development for next
generation power transformers, an ASC press release, Westborough, MA, January 12, 1998.
SUPERCONDUCTORS, CRYOGENIC STABILIZATION

30. P. G. Thesond et al., High temperature 630 kVA superconducting
transformer, Report 12-302, CIGRE Meeting, Aug. 30Sept. 5,
1998, Paris.
SWARN S. KASI
American Superconductor
Corporation
45

Copyright
TUNNELING AND JOSEPHSON JUNCTIONS

Tunneling devices incorporating superconducting materials can exhibit what are known as Josephson effects.
A typical example is two superconducting materials separated by a thin dielectric barrier. Remarkably, such
a tunneling device can exhibit both a zero-voltage trace for currents below some critical value and a second
trace that has tunneling characteristics akin to a forward-biased diode. Traversing the entire currentvoltage
characteristic of the device will give a hysteretic currentvoltage trace. The scale of these phenomena is roughly
1 mV for low-temperature superconductors (such as Pb, Pb-alloys, and Nb) and 10 mV for high-temperature
superconductors (such as YBaCuO, BiSrCaCuO, and TlBaCaCuO).
Low-temperature superconductors used for Josephson applications have transition temperatures, T c , from
the normal to the superconducting state of approximately 1 K to 20 K, while high-temperature superconductors
have T c 90 K.
The zero-voltage leg of the currentvoltage characteristics of a Josephson junction is highly sensitive
to magnetic fields. This property is exploited by superconducting quantum interference devices (SQUIDs) to
measure magnetic fields down to the 10 fT range, making them the most sensitive detectors of field/flux in
existence. Superconducting quantum interference devices have been employed in prototype diagnostic systems
for noninvasive mapping of heart and brain function, but must typically be operated within the confines
of a magnetically shielded room. Prototype devices using SQUIDs for nondestructive evaluation of metal
components are currently being tested in the field. SQUIDs are also employed in the detection of undersea
anomalies (submarine and ordinance detection) and in a variety of scientific investigations such as earthquake,
gravity-wave, free-quark, monopole, and dark-matter detection. Some practical SQUID devices have also used
high-temperature superconductors.
The nonzero voltage leg of a Josephson junction current-voltage characteristic is also useful because it can
be highly nonlinear. As such, these devices are employed for mixing in the 100 GHz region, where low noisetemperatures are a must (such as in radio astronomy). This work is restricted to low temperatureson the
order of 1 Kand the use of low-temperature superconductors. High-temperature-superconductor tunneling
characteristics are not currently of sufficiently high quality for this application.
Another important potential application of Josephson junctions is their use in digital applications. The
reason is that switching with Josephson-junction-based logic involves small voltages and currents and can
occur over very short time scales. Intrinsic power and switching speed are in the 0.1 W and 1 ps range.
Josephson junctions employed for this purpose, based on Nb, are a mature technology. For a number of reasons,
current technology has focused on ultrahigh-speed analog/digital conversion rather than computation; one
issue is the current lack of a suitable memory format compatible with Josephson technology. Nevertheless,
high-speed Josephson processing circuitry continues to have promise for such targeted applications as video
image processing. A number of books and edited works are available on the subject of the Josephson effect and
its applications (1,2,3,4,5,6,7,8).
Ironically, for digital and SQUID applications the intrinsic hysteresis of the junctions must be suppressed.
For junctions that employ a thin, insulating oxide barrier, this implies the use of thin-film resistors to shunt
the junctions to reduce the hysteresis. A better solution would be to tailor the tunnel barrier material so that
1
Fig. 1. Shown is a sketch of a normalinsulatornormal, NIN, tunneling system. N represents a normal, or nonsuperconducting metal film. I (nominally an insulator) represents a thin, dielectric layer through which quantum-mechanical
tunneling can occur. The normalmetal films are typically approximately 10 nm to 100 nm thick and the dielectric layer is
approximately 1 nm thick. Barrier heights, , are typically in the range of 0.1 eV to 3 eV. At low voltages I R 1 (V + V 3 )
with 1 V 2 .
it was somewhere between a metal and insulator. More exactly, this suggests producing a material close to the
metal-insulator transition. This subject is discussed at further length in connection with research in producing
high-temperature superconductor junctions.
MetalInsulatorMetal Tunneling
A basic tunneling system consists of a metalinsulatormetal (MIM) structure as depicted in Fig. 1. Generally
the system comprises thin metal films 10 nm to 100 nm in thickness, separated by a thin dielectric layer
1 nm in thickness. This dielectric can either be a native oxide created by the oxidation of the base metal
electrode (a good example of which is aluminum) or a deposited artificial barrier, a good example of which is
a semiconductor such as silicon. Tunneling in native and artificial barriers has been reviewed as a separate
topic (9).
If the metals are not in the superconducting state, we have normalinsulatornormal (NIN) tunneling
(10). The currentvoltage characteristics of NIN junctions have been extensively studied in the context of
quantum-mechanical electron tunneling through a potential barrier. The height of the barrier, , governs the
rate of change of current with applied voltage. Barrier heights typically range from 0.1 eV to 3 eV. Simmons
11 has shown that for applied voltages V
where 1 V 2 .
Fig. 2. Shown is a sketch of a superconductorinsulatornormal, SIN, tunneling system. N represents a normal, or

nonsuperconducting metal film and S represents a superconducting film. I (nominally an insulator) represents a thin, dielectric layer through which quantum-mechanical tunneling can occur. The normal-metal films are typically approximately
10 nm to 100 nm thick and the dielectric layer is approximately 1 nm thick. The currentvoltage trace shown is for zero
temperature (T = 0).
At a given voltage, the current varies with barrier thickness, d and barrier height, as
where d is in nanometers and is in volts (12). Tunneling in more complex although thematically similar
semiconductor systems is discussed by Sze 13.
If we now introduce a superconductor (Fig. 2) as one of the metal elements of the tunnel junction, to
produce a superconductorinsulatornormal (SIN) system, the currentvoltage characteristic becomes highly
modified. Most notable is the introduction of a region in which, at zero temperature, no current flows until a
voltage /e is reached, where is the so called energy gap of the superconducting film. Energy gaps associated
with conventional superconductors such as Pb, Pb alloys, Nb, NbTi, and NbSn are in the few meV (1 meV =
10 3 eV) range, while energy gaps associated with the high-temperature superconductor materials, such as
YBaCuO, are in the 30 meV range.
If we introduce a second superconductor to make an superconductorinsulatorsuperconductor (SIS)
system, something unique occurs (Fig. 3). In this case, two branches develop in the system. If we start at zero
current and increase the current slightly, no voltage will develop across the junction until the critical current,
Ic , is reached. The critical current is related to the energy gaps of the superconductors by Ic = /4(1 + 2 )/eR,
at zero temperature. For example, for Pb and Nb, = 1.37 and 1.53 meV, respectively.
If the critical current is exceeded, a junction driven by a current source will jump (at constant current) over
to the quasi-particle curve. Further increasing the current moves one to higher voltages on the quasi-particle
curve and reducing the current explores the rest of this curve. The zero-voltage state is not recovered until the
current is reduced to zero. If a finite impedance source drives the junction, this switching will occur along the
load-line of the system. This remarkable situation is due the fact that coherent tunneling of Cooper pairs of
electrons (paired electrons in the superconducting state) can occur, producing a zero-voltage superconducting
tunneling state in the system. When the critical current is exceeded, these pairs are broken into quasi particles
with properties akin to electronhole pairs created across a semiconductor energy gap (13).
The current can be described by the Josephson equations (14,15,16)
where is the phase difference between the superconductors. Therefore, increasing the applied current from
zero is equivalent to introducing a quantum-mechanical phase difference between the superconducting elements of the junction.
Beside SIS systems, properly configured SNS systems can also exhibit Josephson effects, as long as the
normal-metal channel between the superconducting materials is of the appropriate geometry. This means that
the normal metal is either a microscopic weak-link connection, point contact, or a thin-film metal constriction
as opposed to a dielectric material (4,17).
Alternating Current Josephson Effect

A nonzero dc voltage, V = constant, across a Josephson junction implies that
which means that
Therefore, a finite voltage across a Josephson junction gives rise to a radio frequency (RF) current of
where 2e/h = 483.598 THz V 1 . The presence of an alternating current associated with the appearance of a dc
voltage across the junction is the ac Josephson effect. This means that at finite voltage, the current comprises a
dc component with a superimposed ac modulation at the Josephson frequency. A currentvoltage characteristic
thus represents the time-averaged voltage across the junction as a function of applied current (when the system
is driven by a finite-impedance source). The junction thus acts like an RF frequency-to-voltage converter.
Josephson junctions are in fact actually used as high-frequency sources. For a given junction, roughly 10
nW of power can be produced, although most this is dissipated in the junction resistance. The linewidth of the
radiation is given by
where 0 = h/2e = 2.068 10 15 Wb is the magnetic flux quantum. For T 1 K and R 1 , this means that
f 1 MHz. The linewidth can be reduced by coupling to a high-Q cavity, although with the concomitant loss
of overall operational bandwidth.
Fig. 3. Shown is a sketch of a superconductorinsulatorsuperconductor, SIS, tunneling system. S represents a superconducting film. I (nominally an insulator) represents a thin, dielectric layer through which quantum-mechanical tunneling can
occur. The normalmetal films are typically approximately 10 nm to 100 nm thick and the dielectric layer is approximately
1 nm thick. Increasing the current applied across the device produces no voltage until the critical current, Ic is reached, at
which point increasing or decreasing current drives the system along the nonzero-voltage, quasiparticle currentvoltage
characteristic. The load-line of a system driven by a nonzero impedance source is also indicated. Applying a magnetic field
to a sufficiently small junction (see text) will produce a periodic depression of the critical current. is the magnetic flux
entering the junction. = HW(d + 1 + 2 ), where is the penetration depth and 0 = h/2e = 2.068 10 15 Wb.
To achieve useful output power levels, arrays of junctions are produced to create tunable millimeter wave
sourcestypically as low-noise local oscillators for radio astronomy mixer applications. Junction arrays can
produce 2 W to 6 W of RF power in the 340 GHz to 440 GHz band. To achieve this, all junctions in the
array must be phase locked (18).
The ac Josephson effect can be manifested as Shapiro steps (Fig. 3) by coupling microwave radiation into
the junction. This produces a series of steps of equal width V = (h/2e) where f is the frequency of the applied
microwaves. An important application that exploits this phenomenon, and also uses series arrays of (here
Nb-based) junctions, is the standard volt. Here, a junction array is driven by a microwave source to produce in
excess of 1 V dc on the output terminals of the device (19,20). Only fundamental constants and the frequency of
an external oscillator, which can be established with high accuracy, determine the output voltage. The National
Institute of Standards and Technology (NIST) has developed Josephson arrays for this purpose and they now
serve as the primary US standard volt.
Recent work in this particular area has also included Nb/AuPd/Nb (SNS-type) junctions. These systems
use a normalmetal alloy instead of a dielectric barrier, through which Josephson coupling can also occur to
produce a junction that is inherently resistively shunted (21). Another refinement is to use Ti (22) instead of
AuPd. Titanium is potentially more desirable, because its resistivity (at 4.2 K) is higher than AuPd and it is
compatible with whole-wafer processing techniques.
Experimental work based on a stacks of Josephson junctions (23) has also been explored for potential
submillimeter oscillator applications with, for example, NbCN/MgO/NbCN (24), NbCN/NgO/NbCN (25), and
Nb/Al/AlOx/Nb (26,27) systems. Stacks of Josephson junctions may also be useful as inductive elements in
resistive or rapid single-flux-quantum (RSFQ) digital circuitry. One technical challenge here is producing
junctions with uniform Josephson critical currents, Ic (28).
Microwave irradiation has also been explored on a more purely experimental basis with high-temperature
superconductors. These materials can be viewed as a stack of superconducting CuO planes, with an interplanar
quantum-mechanical coupling that can vary from one material to another. For example, the layer-to-layer
coupling in BaSrCaCuO is far less than in YBaCuO. In fact, in the former case the coupling is similar to the
Josephson coupling occurring in an SIS tunnel junction. Thus BaSrCaCuO is thought capable of mimicking
the behavior of a stack of individual Josephson junctions. With this picture in mind, researchers have applied
microwave radiation to BaSrCaCuO mesas. Microwave steps were in fact observed, but with a voltage spacing
greater than that expected for Shapiro steps (29). Mesa-type TlBaCaCuO devices have also been produced (30)
with microwave applications in mind. Again, however, these systems remain experimental in nature.
Magnetic-Field Effects
The current distribution in a Josephson junction will be uniform if the width, W, of the junction (see Fig. 3) is
small compared with the Josephson penetration depth, J , with
where J c = Ic /A with A the junction area, and t = 1 + 2 + d. The quantities are the penetration depths of
the superconductors composing the junction. This is the scale over which an applied external magnetic field
will penetrate into a superconductor. Here d again is the thickness of the barrier (1,2,3,4,5,6,31).
In the case where we apply a static magnetic field to the junction along the direction of the plane of the
barrier, we will suppress the Josephson current. If we meet the criterion that the junction is small compared
Fig. 4. Shown is the currentvoltage characteristic of a nominally SIS junction created by using a scanning tunneling
microscope tip to peel up a portion of the surface of a high-temperature superconducting BiSrCaCO (2212) film and hold it
in proximity to the underlying material. Although not an ideal characteristic, this demonstrates the ability to achieve (and
provides a means to study in detail) tunneling in the high-temperature superconductor materials. From Miyakawa et al.,
Ref. 32.
with the Josephson penetration depth, this suppression of the critical current will take the form
where Ic0 is the zero-field critical current. For Pb, Nb, NbSn, 50 nm to 100 nm; for NbN, 300 nm; and
for YBaCuO 30 nm and 200 nm. In the last case, parallel and perpendicular refer to the directions
along and perpendicular to the CuO planes.
If a junction is not small by this definition, the current distribution will not be uniform and the suppression
of the critical current will not follow the simple sinx/x behavior noted above. Indeed, the geometry of the junction
can be controlled to produce a critical-current behavior parametrically tailored to specific applications, as in
the case with the use of junctions as elements for digital applications. In addition, the spatial nonuniformity of
the barriers of individual junctions can be diagnosed through a deconvolution of critical current versus applied
magnetic field characteristics (3).
As far as applications are concerned, this general phenomenology implies that a large penetration depth is
generally undesirable. A material such as NbN, which has a larger energy gap than Nb (and from this standpoint
represents a superior material) suffers from this limitation. However, this can be in part engineered around
by creating hybrid NbN/Nb layer pairs in which a balance of the higher critical temperature of NbN (14 K as
opposed to 9.25 K for Nb) and lower penetration depth of Nb is reached. Another important example of the use
of hybrid materials systems also involves Nb as Nb/Al bilayers; these are used to produce high-quality tunnel
junctions as is discussed later (see section entitled Digital Applications.)
Finally we note that SIS junctions have been made with high-temperature superconductor materials as
well. As shown in Fig. 4, such devices have been created with the use of a scanning tunneling microscope tip
which is crashed into the surface of a BiSrCaCuO crystal (32). As the tip is pulled back up, a junction is created
between superconducting material remaining on the tip and the underlying crystal. While hardly a practical
device, with less-than-perfect currentvoltage characteristics, this nonetheless illustrates the potential for
creating devices from high-temperature superconductor material and has provided a valuable vehicle for
fundamental studies of the system.
Fig. 5. The resistively shunted junction (RSJ) model for tunneling in SIS systems. This diagram illustrates that the
degree of hysteresis in a junction with a given resistance and capacitance is governed by the magnitude of the parameter
= (2e/h)Ic R2 C. The parameter can be viewed as setting the degree of damping in the circuit.
The Resistively Shunted Junction (RSJ) Model

The application of Josephson junctions generally requires control of the degree of junction hysteresis. One
way of achieving this is to shunt the junction with an external resistor. This usually means the use of a thin
film of Au or AuCu alloy deposited across the SIS junction to produce a shunt in the range of 1 . In some
cases, junctions can be produced with internal microscopic resistive links in the barrier layer to achieve the
same effect. More sophisticated approaches, involving tuning the properties of tunnel barriers to achieve a
conductivity near the metalinsulator transition, have also been proposed and are discussed further in HighTemperature Superconductor Ramp Junctions.
Figure 5 shows the electrical equivalent circuit of a shunted Josephson junction with a total resistance
R biased by a current source (shunted with an external resistor to produce a resistance far lower than the
original tunneling resistance of the junction). Represented in this manner there are three current paths: the
Josephson current, the ohmic current, and the displacement current from the junction capacitance. The total
current (for a system driven by a current source) is thus
Along with the Josephson relation
this can be rewritten as
where c = (2e/)Ic R represents the upper operational frequency of the junction and p = (2eIc /C)1/2 is the
plasma or lower-bound propagation frequency of the system. It is customary to define the StewartMcCumber
parameter
The dependence of the current-voltage characteristics of a Josephson junction on is also depicted in

Fig. 5. Small and large values of represent the high- and low-damping limits, respectively. The most desirable
operating regime for many applications is 1. As noted by Likharev 4, a useful way of parameterizing is as
where again J c = Ic /A, A is the junction area (W2 in Fig. 3), r is the dielectric constant of the junction barrier
material, 0 = 8.85 10 12 F/m, and d is the barrier thickness. Improved performance at high frequencies
generally implies smaller areas and a concomitant increase in J c to achieve 1 (33). Note that in principle
is an intrinsic property of the superconductors. Its actual value, however, can be lower than the indicated
theoretical result due to gap suppression at film surfaces and other effects.
Superconducting Quantum Interference Devices

If we place two Josephson junctions in a loop (Fig. 6) we form a SQUID, in this case a so-called dc SQUID.
To produce a device intended for the measurement of flux, the junctions are shunted, as discussed previously,
to suppress junction hysteresis. The flux associated with an applied magnetic field generates a circulating
current in the device, which suppresses the Josephson current. An applied flux thus modulates the voltage
across the device with a period equal to the magnetic flux quantum 0 . The result of this is a device that is very
sensitive to magnetic flux. SQUIDs are sensitive to flux to magnitudes applied 0 = h/2e = 2.068 10 15
Wb. Practical SQUIDs are capable of measuring fields to the 10 fT magnetic field range, or 10 34 JHz 1
at 105 Hz (34,35), making them by far the most sensitive device in existence for measuring magnetic fields
and flux. Field versions of SQUIDs can usually achieve an order of magnitude more sensitivity than flux gate
magnetometers.
SQUIDs have a variety of applications, including commercial biomedical SQUID arrays that noninvasively
monitor and map heart and brain function. Other uses include nondestructive evaluation (NDE) (e.g., crack
detection in metals), oceanic anomaly detection, gravity-wave antennas, earthquake monitoring, magnetic
monopole detection, dark-matter searches, and other fundamental scientific investigations (6,36,37).
A key factor limiting the performance of SQUIDs is 1/f noise. This stems from shallow trapping of both
fluxoids in the thin-film elements of the SQUID and tunneling electrons in the barriers of the Josephson
junctions (shot noise). The observed 1/f noise typically has a knee in the 1 Hz to 10 Hz range with an associated
spectral flux noise density 10 10 2 0 /Hz 1 at 0.1 Hz. Optimizing SQUID performance is especially important
for biomedical applications since operation down to the 0.1 Hz range is necessary (38).
The flux noise energy of a SQUID can be written as (39)
10
Fig. 6. Shown is a sketch of a dc SQUID (superconducting quantum interference device) comprising a loop with two
Josephson junctions. Shunt resistances are used to control the damping parameter () of the junctions. The currentvoltage
characteristics and voltageflux characteristics are shown for an applied flux .
which generally suggests the need for devices with the smallest inductance and capacitance, operating at the
lowest temperatures. The use of low-temperature superconductors (Pb, Pb-alloys, Nb, and so on) generally
means operation at 4.2 K, (the boiling point of liquid helium at 1 atm) and temperatures less than 80 K for the
high-temperature superconductor materials. A method successfully used to reduce inductance is the washer
loop (39). Because of the Meissner effect associated with superconductivity, flux transformers can be used
to collect and couple flux from a larger, external superconducting loop and present it to the SQUID using a
multiturn thin-film coil.
High-temperature superconductors have also been relatively successfully explored for SQUID applications. An approach to creating Josephson junctions in high-temperature superconductor materials is to simply
make a step in a substrate to create a step-edge junction. Material grown across the edge acquires a defect that
creates a weakly linked Josephson junction, with characteristics similar to a resistively shunted SIS system.
It was realized that the actual angle of the step or ramp was important in consistently creating junctions
with desired characteristics (40,41,42). Along the same lines, junctions made with silicon-on-sapphire (SOS)
substrates have demonstrated RSJ characteristics (43,44) and were used to make the first SQUIDs operating
at 77 K using step-edge techniques (45).
A different approach to the use of high-temperature superconductor materials is the use of bicrystal substrates. Here, two single crystals (for example, of SrTiO3 ) are fused together at a predetermined angle to create
an off-axis interface. Junctions are made by depositing a thin film of high-temperature superconductor material
across this disruptive interface. Bicrystal work in particular has consistently shown RSJ-type currentvoltage
characteristics and critical currents that are reproducibly correlated to the angular displacement of the
11
Fig. 7. SQUID structure from IBM. Most of the structure seen in the figure is the superconducting loop containing a flux
dam, visible as the triangular restriction on the left, to reduce device noise by limiting circulating currents in the flux loop
lines. The Josephson junctions incorporated into the structure are step-edge devices. The overall size of the SQUID is 1
1 cm. From Sun et al., Ref. 53.
substrate crystal lattices. These bicrystal junctions have been successfully incorporated into YBaCuO (46,47,48)
and BaSrCaCuO SQUIDs (49,50). Bicrystal junctions of MgO (51) have also been discussed for use in SQUIDs
targeted for nondestructive evaluation systems. Noise studies (52) suggest that bicrystal and step-edge techniques generally produce devices of comparable quality. However, although the techniques are suitable for
few-device applications such as SQUIDs, they are not appropriate for medium- or large-scale integration.
The practical implementation of high-temperature SQUID technology is shown in Fig. 7. The work is
from Kochs group at IBM (53). Here junctions are formatted into a large loop containing flux dams, one of
which is shown to the left as a constriction in the loop. The constriction creates a fuselike link to help eliminate
noise-generating current loops.
Beside the double-junction dc SQUID, there is also the single-junction RF SQUID, depicted in Fig. 8. With
this device, flux quantization in the ring, along with the corresponding variation of the Josephson current with
flux, produce a variation in the inductance of the SQUID loop. This change in loop inductance is coupled to and
thus shifts the resonant frequency of an external, RF-driven tank circuit. Because of its relative simplicity and
the requirement for only a single, shunted Josephson junction, this scheme was initially and has continued to
be the device of choice for many commercial applications, notably for nondestructive evaluation and SQUID
voltmeters.
We note finally that, irrespective of the type of SQUID employed, applications such as biomedical diagnostics typically require operation within the confines of a carefully shielded magnetic enclosure. However, for
12
Fig. 8. Shown the schematic of an RF SQUID. This type of SQUID uses a single (shunted) Josephson junction in a loop.
Applied external flux () acts to alter the impedance of the loop, which is inductively coupled to and shifts the frequency
of an RF driven tank circuit.
such applications such as nondestructive evaluation field work, SQUID gradiometers can be employed that
are more immune from background magnetic fields and can be optimized for unshielded performance. This
implies a compromise between field sensitivity and voltage modulation characteristics. For example, SQUIDs
for nondestructive evaluation (54) are cited as having a flux noise at 1 Hz of 134 0 Hz 1/2 and 60 0 Hz 1/2
at 10 Hz, qualifying them as prototype commercial systems (55).
In addition to the measurement of flux, SQUIDs can also be configured to perform other functions.
This includes the measurement of ultrasmall voltages. SQUID voltmeters are commercially available for
measurements in the 1 pV rangelimited by noise to a range of 10 10 V/Hz 1/2 at 100 . Radio frequency
SQUIDs have also been employed as low-noise amplifiers to 100 MHz.
Mixing and Detection

The zero-voltage branch of a Josephson junction (Fig. 3) can respond at frequencies up to f (4/h) (e/h)Ic R
1 THz for low-temperature superconductors and 10 THz for high-temperature materials. As a result
high-frequency Josephson mixing has been extensively studied. Unfortunately, heterodyne mixing using the
zero-voltage branch of the Josephson tunneling characteristic is seriously degraded by the appearance of excess
noise due to noise down conversion and the dynamic impedance related to the ac Josephson effect (56).
However, the extreme nonlinearity of the quasi-particle portion of the currentvoltage characteristic, as
the tunneling current turns on for voltages just above the sum-gap voltage, has been successfully exploited
for mixing (56,57,58) in the same manner as high-frequency diodes are used. For superconducting mixing,
the Josephson current (zero-voltage curve) becomes a nuisance and can be suppressed by applying a small
magnetic field. Such devices are generally referred to as SIS mixers, with maximum operating frequencies the
same as noted previously.
SIS mixers are typically employed when uncompromising low-noise performance is required (such as with
radio astronomy) and where very low (1 W) local-oscillator power is required. These mixers are typically used
in the 40 GHz to 1 THz region, with single sideband (SSB) noise temperatures at 100 GHz of 4 K. Overall
13
noise temperatures for 40 GHz to 1 THz are typically within an order of magnitude of the quantum noise
limit of hf /kB . This currently exceeds the performance of high-electron-mobility transistor (HEMT) devices.
One critical element in mixing is junction quality. The magnitude of the tunneling current for voltages
below the turn-on voltage for quasi-particle tunneling, that is, voltages below the sum-gap voltage, is a crital
factor in mixer noise performance. Leakage conduction in this regime needs to be as small as possible and is
limited in theory only by thermal excitations, but in practice can be dominated by imperfections in the tunnel
barrier itself. For this reason, junctions based on oxidized tin and other soft metals were initially employed
because of their extremely low so-called subgap conductance due to the high quality of the Sn-oxide barrier.
However metals such as Sn are not physically robust with respect to thermal cycling (from their 1 K operating
temperature to room temperature).
Mixing requires a very high quality dielectric barrier integrated into superconductive elements with
the highest possible energy gap (and hence critical temperature) to provide the highest operating frequency.
Matching requirements also mean that junction resistance be in the 20 to 100 range and that 1 < 2RCf <
10. Practical compromises have resulted in the use of hybrid systems such as Nb-based electrodes, aluminumoxide barriers, and Pb-alloy counterelectrodes to form NbAl2 O3 Pb-alloy junctions. This is a good union
between the thermal stability of Nb and the high-quality dielectric properties of Al2 O3 . NbAl2 O3 Nb junctions
have also been adopted for mixing applications. The upper operational limit for Nb-based junctions is 1.3 THz.
Junctions based on NbN (with a critical temperature of 14 K compared with the 9.25 K transition temperature
of niobium) have also been made, which have an operational mixing limit of 2.5 THz. Early NbN work
explored both NbNMgONbN (59,60), and NbNAlNNbN systems (61,62,63). Recent NbN work has produced
devices with double sideband receiver noise temperatures of 1450 K at 600 GHz and 2800 K at 950 GHz (64).
Mixing experiments have also been conducted with high-temperature superconductor materials. For
example, bicrystal silicon-based junctions have been produced for this purpose (65). Bicrystal results on silicon
and sapphire (66) appear to be comparable to those achieved with SrTiO3 in terms of the overall nature of their
currentvoltage characteristics, but without the disadvantage of the large dielectric effects of SrTiO3 . Both
Shapiro steps (67) and harmonic mixing of a 1.6 THz signal have been observed (68) in such systems. However,
high-temperature superconductor tunnel systems that have demonstrably fulfilled the stringent requirements
for quasi-particle mixing applications have yet to be produced. These applications will require (non-RSJ-like)
high-quality, thin-film, SIS type junctions with low subgap conductance.
Digital Applications
Researchers with an interest in digital applications such as ultrahigh-speed analog-to-digital conversion (ADC)
have been attracted by the fundamental properties of Josephson devices. Josephson junction are fast, with 1
ps switching times; have low switching power, P I2 c R 0.1 W; and can be matched to the impedance of
modern microstrip technology (69).
The most extensive effort to use Josephson devices to create a prototype superconducting computer was
the much-discussed program at IBM. Although the program was not successful in achieving its ultimate goal,
a great deal was learned and much progress was made (70). Lead-alloy technology that produced junctions
with reproducible characteristics and that could be thermally cycled was used. The logic elements were based
on latching circuitry with underdamped junctions (p c ) such that the binary state was defined by the
voltage state of a Josephson junction. For example, a junction can be switched from a zero-voltage state to
a nonzero voltage state along the load-line of the device. As discussed by 69, this type of approach has fast
set times (1 ps) but slow reset times (10 ns). Other difficulties with this arrangement include problems in
achieving uniform device clocking across an entire logic array and high latching power (1 W) per gate. More
recent work with single-flux-quantum (SFQ) logic (referring to logic based on the presence of single fluxiods in
SQUID-based logic elements) has switched to Nb-based junctions (71,72,73,74) in a variety of formats.
14
A new logic type, which is fundamentally different from latching-type logic, has also been introduced.
Called RSFQ (resistive/rapid single flux quantum) logic (69,75), it is a hybrid logic family, wherein junctions
are configured to have both logic and (dynamic) memory functions and logic operations are performed on pulses
originating from traditional SFQ devices. This approach addresses to some extent a major problem that was
encountered in the IBM experimt: matching SFQ logic and memory elements. The approach also recovers
much of the fundamentally fast switching times and low switching power of Josephson devices (the latter since
junctions are open for only a small part of a clock period). Switching powers of 10 7 W/gate (10 18 J/bit)
implies that packing densities of 107 gates/cm2 are possible at speeds to 500 GHz with 1 m linewidths and
at greater speeds for narrower lines (4).
In present practice, much of the effort is not in computing per se but high-speed digital processing,
especially ADC [and agile programmable voltage standards (76,77)] with RSFQ as a favored digital logic
format (78,79). For example, 1024-bit shift registers have been operated at up to 20 GHz (80).
The current application of Josephson technology is still largely based on niobium. Niobium technology is
mature and now appears to be the standard for digital Josephson applications. Niobium based junctions are
physically robust and amenable to large-scale integration as a result of the important innovation of replacing
the native oxide that grows on niobium, which generally forms a poor quality barrier with a relatively large
dielectric constant (140 fF/m2 ) (81,82). The idea is that the niobium base electrode is capped in situ with a
thin aluminum film (1 nm to 8 nm thick), which is oxidized to completion to form a surface-layer junction and
in the process prevents the growth of NbOx (9,83,84). This scheme successfully combines the relatively high
critical temperature, T c , of the (robust) Nb underlayer (9.25 K) with the unsurpassed qualities of Al2 O3 , which
is an excellent barrier material with a relatively low dielectric constant (60 fF/m2 ) (84) compared with Nb
oxide. A large-scale process for creating many such junctions with a simple anodization process was developed
by 85,86). Likharev has reviewed the complexities of junction fabrication for digital and other applications (4).
A good example of contemporary large-scale Josephson technology is the work at HYPRES Inc., which
produces large-scale integrated Nb-based Josephson tunnel junction circuitry using RSFQ logic. A recently
manufactured large-scale integration ADC is shown in Fig. 9. Target applications for such systems would
include high-resolution ADCs for radar and time-to-digital converters (TDCs) to measure the timing of events
in high-energy and nuclear physics experiments. These systems have demonstrated flash ADC with 6-bit
resolution, in the 1 GHz to 10 GHz operating range (87).
Another potential application for the fast switching speed of Josephson junctions is cross-bar and related
switching matrices for switching between processors and memory (88,89). This is important to high-speed
telecommunications and computation applications.
We finally note that other approaches based on quantum-limited-logic have been proposed and may be
competitive in some areas. One of these is single-electron logic (SEL), which is based on charging effects in
ultrasmall capacitance tunnel junctions (and not the Josephson effect). SEL is operationally similar to RSFQ
logic, but the former is based on the presence or absence of single electrons rather than magnetic fluxoids
(loosely speaking, bits based on single electrons with charge e as opposed to single fluxoids with flux 0 ) (4,69).
High-Temperature Superconductor Systems

The appearance of high-temperature superconductor materials has spurred a broad-based effort to produce
Josephson junctions for digital and other applications. Some of these approaches were discussed previously
in connection with SQUIDs. Beside this work, other approaches have been taken to produce a successful
junction format. Certainly all the high-speed applications that are envisioned for niobium-based Josephson
technology such as flash ADC, wideband ADCs, transient-event digitizers, and crowbar switches could be envisioned for high-temperature superconductor materials as well. Barring thermodynamic considerations, operation at higher temperatures would generally be an advantage. The biggest hurdle now for high-temperature
15
Fig. 9. Modern analog/digital converter from HYPRES, Inc. using niobium technology. The chip is a 6-bit flash ADC with
a 32-word shift resister memory and operates at 16 GS/s.
superconductor digital applications is the uniformity of the Josephson critical current density (Josephson critical current per unit junction area) for junctions across a single chip and chip-to-chip critical current uniformity.
A maximum variation of a few percentage points in critical current density is probably necessary for large-scale
applications (90), and this has yet to be achieved. At present, achievable spreads in critical current density are
more in the vicinity of (or perhaps somewhat less than) 10 percent. This single consideration is a dominant
factor governing which basic device format will be successful for high-temperature superconducting materials,
a number of which have been and are currently being explored in an attempt to find a suitable candidate to
meet this criterion.
We again note that high-temperature superconductors can be viewed as quasi-two-dimensional
materialsa stack of Josephson coupled two-dimensional superconducting layers. This basic physical structure, combined with a propensity for naturally occurring grain boundarieswhich also act as Josephson
weak linkshas led to the investigation of intrinsic Josephson effects (91,92,93,94,95,96,97,98,99). Some of
the devices based on naturally occurring, intergranular weak links have exhibited good critical-current and
normal-state characteristics. However, the use of naturally occurring weak links to create junctions per se has
typically lead to poor reproducibility. Focused ion beam (FIB) techniques have also been used in an attempt to
artificially induce defects at given locations by inducing substrate damage (100).
A successful approach to creating SNS type junctions in preselected locations is focused electron beam
(FEB) writing. Here, an electron beam is directed at a location on a superconducting thin film, creating a narrow, damaged line of material. The material then acts as a normal metal to create an HTS/N/HTS (HTS,
high-temperature superconductor) device (101,102,103,104,105,106). The technique can produce RSJ-like
16
Fig. 10. Sketch of the ramp junction geometry used with high-temperature superconductor materials, shown here with
YBaCuO films.
microwave behavior (107,108). An example is the work at Cambridge (104), where a computer-controlled
electron beam (350 KeV at 400 pA) is swept across narrow lines of YBaCuO. This process has produced junctions with an SNS character (17) with long-term room-temperature stability. One drawback of the large-scale
use of FEB is its slow materials processing speed, since each device must be individually electron-beam written.
High-Temperature Superconductor Ramp Junctions. The idea of creating a step in a substrate
has also been extended to make SNS and SIS type structures where the normal and insulating materials are
separately introduced films as opposed to defect-modified high-temperature superconductor material. Originally, SNS junctions used normal metals such as gold; however, such devices appear to have been dominated by
interface resistance. More recently, both cobalt-doped YBaCuO and gallium-doped PrBaCaO have been used
as generic barrier materials in YBCO/barrier/YBCO systems, selected because they are physically compatible
with YBaCuO and their conductance properties can be tuned with doping level. Cobalt-doped YBaCuO has a
relatively low resistance, as opposed to gallium-doped PrBaCuO, and thus imposes a somewhat limited device
operating temperature range because the material becomes superconducting below some temperature (109),
although relatively narrow critical current density spreads (12 percent variation) have been observed with
the material (110,111). Such junctions have also been employed in SQUIDs (112).
Gallium-doped PrBaCuO tunnel barriers tend to produce relatively high resistance barriers and must be
operated below 77 K. Nevertheless, the material can produce junctions with values of Ic R in a technologically
useful range (1 mV) and with independently adjustable critical current density and resistance (113), which
is important for engineering considerations. In one view of tunneling in PrBaCuO, supercurrent (zero-voltage)
conduction occurs via direct tunneling through the barrier whereas quasi-particle (nonzero-voltage) conduction
occurs via resonant tunneling channels within the barrier (114). The fundamentals of this issue were also
addressed, with device applications in mind, by work with amorphous silicon barriers (115).
HTS/I/HTS tunnel junctions using PrBaCuO barriers in a ramp format have been produced through a
variety of techniques (116,117). A sketch of the standard ramp junction format used with high-temperature
superconductor materials is shown in Fig. 10. All tend to produce devices with RSJ-like characteristics [with Ic R
1 mV at 4.2 K, which scales with PrBaCuO doping (118)]. The combination of step-edge substrates and doped
barriers has been promising overall, and these devices have been modeled by microscopic theory (17), although
the full applicability of standard proximity-effect theory with regard to high-temperature superconductor
materials remains an open question (119).
To date, small circuits have been put together with these junctions, which are designed with tolerance to
the large spreads that exist at present in junction critical current (120,121,122,123,124). Clearly, again, the
challenge is to improve on the reproducibility of critical currents in this or any other device configuration that
may present itself.
Related to this work are engineered interface YBaCuO/I/YBaCuOtype junctions. Here, the barriers are
produced by relatively subtle damage to the base YBaCuO film and then YBaCuO is simply deposited on top.
The suggestion is that the barrier comprises a thin layer of YBaCuO with oxygen disorder or deficiency perhaps
a few nanometers thick (125). Other groups have been experimenting with this and related techniques (126).
In summary, a variety of promising approaches have been explored with high-temperature superconductor materials to produce junctions for SQUID and digital applications. The primary goal is to produce junctions
17
that are parametrically reproducible, and the secondary goal is to produce a thin-film barrier compatible with
the high-temperature superconductor materials with transport properties close to the metalinsulator transition. The latter would allow the junction to be self-shunting and thus allow for a broader use of the devices. This
balancing act is challenging in light of the problems of material compatibility with high-temperature superconductor systems and unanswered questions regarding the physics of barriers (especially high-temperature
superconductor-compatible materials) near the metalinsulator transition.
High-Temperature Superconductor Applications:Technical Considerations. A great deal of effort has gone into the application of high-temperature superconductor materials for SQUIDs and SFQ/RSFQ
logic circuitry. Such circuits need nonhysteretic junctions with Ic 50 A to 500 A and R 1 for 10 < T
< 77 K, which is achieved with low-temperature superconductors at 4.2 K by shunting. In the case of hightemperature superconductor materials, compatible barrier materialssuch as doped PrBaCuO or YBaCuO
typically have far lower resistivities than barriers such as Al2 O3 . Therefore, as noted, these barrier materials
have been examined with a view toward obtaining high-temperature superconductorbased junctions in the
parametrically desirable region where shunting would be unnecessary (109,127,128).
In Likharevs description of SFQ circuits (129) it was shown that for niobium junctions with Ic R 0.2
mV, operation can be in the 100 GHz range. For SFQ to work, one needs series or parallel loops of junctions
with inductance L such that (1/2)0 < LIc < (3/2)0 . This limits the acceptable range of Ic for junctions. A
lumped-parameter analysis (130) indicates that at 10 K the minimum Ic for Ic R 0.1 mV to 1.0 mV is 0.2
m and for high temperatures, 60 K to 70 K, Ic will need to be 1.2 mA. The analysis also indicates that to
avoid latching and ensure sufficiently fast signal propagation, the condition on junction resistance will be that
Rmax 4 for r = 40 with 1 m lines and 2 for r = 100, where r is the dielectric constant of the material
adjacent to the signal lines.
This analysis and the results of other work (131,132) suggest that devices using PrBaCuO-based barriers
can achieve critical currents, Ic , in the range of 105 A/cm2 and still be nonhysteretic. The ultimate conclusion
is that that PrBaCuO barriers can give junctions parametrically compatible with SFQ applications.
Three-Terminal Devices
One overriding characteristic of a Josephson junction is that it is a two-terminal device, more like a forwardbiased diode with a hysteretic current-voltage characteristic than a transistor, which has a gate or base terminal. However, there have been a number of proposed and experimentally examined three-terminal superconducting devices. These include hybrid superconductorsemiconductor devices, such as junction and field-effect
transistors, and nonequilibrium superconducting devices, such as stacked-junctions and injection-controlled
weak links (133). At present none of these devices has found its way into standard use, because they generally have not exhibited above unity gain, although in theory this appears to be possible (134). One inherent
problem is the high carrier density of conventional superconducting materials, making it difficult to modulate
device transconductance. However, high-temperature superconductor materials, which have far lower carrier
densities and perhaps other parametric advantages for certain device configurations (135), provide somewhat
more promise for three-terminal devices.
Early work on Josephson FETs per se has included InAlAs/InGaAs HEMT-type configurations, where
a gate controls the magnitude of the Josephson current (136). The overall prospects for Josephson FETs
(137), and experimental results with HEMPT devices (138) have also been presented. Generally it appears
ultimately possible to create a device with voltage gain, driving loads of 100 . FET-type structures have
also been created in YBaCuO/barrier/YBaCuO SIS-type systems, where a base layer is used to apply a strong
electric field across the barrier region using an external gate, causing a change in surface carrier density and a
corresponding modulation of the Josephson current. Examples of this work include YBaCuO/SrTiO3 /YBaCuO
and related structures (139,140,141,142,143,144,145,146) and Au/SrTiO3 /YBaCuO (NIS) structures (147,148).
18
Fig. 11. The figure shows the structure of a proposed magnetic/superconductor memory element. Shown are parallel (a)
and antiparallel (b) configurations of magnetization in the M layers. Figure from Sangjun et al., Ref. 170.
In general, FET structures have to date not shown anything but relatively weak effects, although work
on a variety of promising systems continues (149). The work is motivated in part because the devices have
the potential to serve as a Josephson-to-semiconductor logic interface. The general problem to be addressed is
that Josephson logic output voltages are 3 mV (for low-temperature superconductors) while a CMOS device
operates at 1 V. Some work has been specifically directed at this issue (150,151), including interface electronics
combining 4 K Josephson drivers and 77 K semiconducting HEMT amplifiers with a predicted speed of 1 GHz
to 3 GHz (152).
A variety of other schemes have also been advanced for both low- and high-temperature systems. These fall
under the generic categories of electric-field controlled devices (153,154,155,156,157), quasi-particle injection
devices (158,159), flux-flow transistors (160,161), and dielectric-base transistors (66,162,163,164,165).
Although these ideas continued to be pursued, an achievable device format with useful gain has yet to
emerge unambiguously. Certainly, three-terminal high-temperature superconductor devices will require higher
quality junctions and for FETs, dielectric gates compatible with high-temperature superconductor materials
and better geometries to provide improved coupling of electric fields into junctions. Thorough discussions of
three-terminal Josephseon devices are available in the literature (155,167,168,169).
Related Superconducting Devices

While not superconducting tunneling devices, several systems have recently been developed that could interface
with superconducting devices and logic. One example, illustrated in Fig. 11, is a memory element based on
superconductormagnetic multilayers (170).
19
There has also recently been considerable interest in the physics and device implications of tunnel junctions containing magnetic elements (171,172,173,174,175), which may also be compatible with and complementary to Josephson circuitry.
Acknowledgments
I wish to thank Alan Kleinsasser for many useful comments on the text. I also wish to acknowledge insightful
discussions with M. R. Beasley, D. A. Moore, J. M. Rowell, and J. Talvacchio and comments by Alan Kadin. I
also gratefully acknowledge support during the preparation of this work by the Department of Energy, Division
of Materials Sciences, through grant DE-FG02-88ER45373.
BIBLIOGRAPHY
1. T. Van Duzer C. W. Turner Principles of Superconductive Devices and Circuits, New York: Elsevier North Holland,
1981.
2. A. Barone G. Paterno Principles and Applications of the Josephson Effect, New York: Wiley, 1982.
3. A. Barone G. Patterson Physics and Application of the Josephson Effect, New York: Wiley, 1982.
4. K. K. Likharev Dynamics of Josephson Junctions and Circuits, New York: Gordon and Breach, 1986.
5. H. Weinstock M. Nisenoff Superconducting Electronics, Berlin: Springer-Verlag, 1989.
6. J. C. Gallop SQUIDs, the Josephson Effects and Superconducting Electronics, New York: Adam Hilger, 1990.
7. S. T. Ruggiero D. A. Rudman Superconducting Devices, New York: Academic Press, 1990.
8. T. P. Orlando K. A. Delin Foundations of Applied Superconductivity, Reading, MA: Addison-Wesley, 1991.
9. S. T. Ruggiero Artificial tunnel barriers, in S. T. Ruggiero, D. A. Rudman (eds.), Superconducting Devices, New York:
Academic Press, 1990, pp. 373390.
10. I. Giaever Metal-insulator-metal tunneling, in E. Burstein and S. Lundqvist (eds.), Tunneling Phenomena in Solids,
New York: Plenum, 1969, pp. 1930.
11. J. G. Simmons Low-voltage current-voltage relationships of tunnel junctions, J. Appl. Phys., 34: 238, 1963.
12. J. G. Simmons Image force in metal-oxide-metal tunnel junctions, in E. Burnstein and S. Lundqvist (eds.), Tunneling
Phenomena in Solids, New York: Plenum, 1969, pp. 135148.
13. S. M. Sze Physics of Semiconductor Devices, New York: Wiley, 1981.
14. B. D. Josephson Possible new effects in superconducting tunneling, Phys. Lett., 1: 251, 1962.
15. B. D. Josephson The discovery of tunneling supercurrents, Science, 184: 527, 1974.
16. P. W. Anderson J. M. Rowell Probable observation of the Josephson superconducting tunneling effect, Phys. Rev. Lett.,
10: 230, 1963.
17. K. A. Delin W. A. Kleinsasser Stationary properties of high-critical-temperature proximity effect Josephson junctions,
Supercond. Sci. Technol., 9: 227, 1996.
18. J. Lukens Josephson arrays as high frequency sources, in S. T. Ruggiero and D. A. Rudman (eds.), Superconducting
Devices, New York: Academic Press, 1990, pp. 135167.
19. S. P. Benz Superconductor-normal-superconductor junctions for programmable voltage standards, Appl. Phys. Lett.,
67: 2714, 1995.
20. C. A. Hamilton C. J. Burroughs S. P. Benz Josephson voltage standarda review, IEEE Trans. Appl. Supercon., 7:
3756, 1997.
21. S. P. Benz, et al. Stable 1 volt programmable voltage standard, Appl. Phys. Lett., 71: 1866, 1997.
22. L. Fritsch, et al. Superconductornormal metalsuperconductor Josephson junctions with Ti interlayer, Appl. Phys.
Lett., 73: 1583, 1998.
23. I. P. Nevirkovets Fabrication, dc characteristics of vertically stacked SIS-type structures for use as low-temperature
detectors, Supercond. Sci. Technol., 8: 575, 1995.
24. A. V. Ustinov H. Kohlstedt C. Heiden Coherent flux-flow in vertically stacked long Josephson tunnel junctions, IEEE
Trans. Appl. Superconduct., 5: 2743, 1995.
20
25. A. Shoji T. Kikuchi H. Yamamori Uniformity of critical currents of vertically stacked Josephson junctions, in Extended
Abstr. 5th Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 157, 1997.
26. M. G. Blamire, et al. Characteristics of vertically-stacked planar tunnel junction structures, IEEE Trans. Magn.,
MAG-25: 1135, 1989.
27. I. P. Nevirkovets L. P. Stryzhko A. V. Poladich Photon-assisted tunneling in stacked tunnel structures by millimeter
wave irradiation, Physica B, 194196: 2395, 1994.
28. S. Lomatch E. D. Rippert J. B. Ketterson Multilayer Josephson junction flux quantum devices, IEEE Trans. Appl.
Supercond., 5: 3147, 1995.
29. M. Scheuerman J. T. Chen J. J. Chang Interaction between microwaves and a single vortex in a long Josephson
tunnel junction, J. Appl. Phys., 54: 3286, 1983.
30. S. Yoshikawa, et al. Intrinsic Josephson junctions ofTl2 Ba2 Ca CuOx thin films for high frequency devices, Physica C,
293: 44, 1997.
31. M. Tinkham Introduction to Superconductivity, 2nd ed., New York: McGraw-Hill, 1996.
32. N. Miyakawa, et al. Strong dependence of the superconducting gap on oxygen doping from tunneling measurements
on Bi2 Sr2 Cau2 O7 , Phys. Rev. Lett., 80: 157, 1998.
33. A. W. Kleinsasser R. A. Buhrman High-quality submicron niobium tunnel junctions with reactive-ion-beam oxidation,
Appl. Phys. Lett., 37: 841, 1980.
34. D. J. van Harlingen R. H. Koch J. Clarke Dc SQUID near the quantum noise limit, Physica (Utrech), B108: 1083,
1981.
35. D. J. van Harlingen R. H. Koch J. Clarke Superconducting quantum interference device with very low magnetic flux
noise energy, Appl. Phys. Lett., 41: 197, 1982.
36. Y. Zhang, et al. Applications of high-temperature SQUIDs, Appl. Supercond., 3: 367, 1995.
37. F. Ludwig, et al. Multilayer magnetometers based on high-Tc SQUIDs, Appl. Supercond., 3: 383, 1995.
38. G. L. Romani The use of SQUIDs in the study of biomagnetic fields, in H.-D. Hahlbohm and H. Lubbig (eds.), SQUID
85. Berlin: de Gruyer, 1985.
39. J. Clarke SQUIDs: Principles, noise and applications, in S. T. Ruggiero and D. A. Rudman (eds.), Superconducting
40. J. Gao, et al. Controlled preparation of all high-Tc SNS-type edge junctions and dc SQUIDs, Physica C, 171: 126,
1990.
41. C. L. Jia, et al. Microstructure of epitaxial YBa2 Cu3 O7 films on step-edge SrTiO, Physica C, 175: 545, 1991.
42. R. B. Laibowitz, et al. All high Tc edge junctions and SQUIDs, Appl. Phys. Lett., 56: 686, 1990.
43. Y. A. Boikov, et al. Biepitaxial Josephson junctions with high critical current density based on YBa2 /Cu3 /O7 films
on silicon on sapphire, J. Appl. Phys., 77: 1654, 1995.
44. B. N. Hunt, et al. High-Tc edge-geometry SNS weak link on silicon-on-sapphire substrates, Physica C, 230: 141, 1994.
45. M. J. Burns, et al. Demonstration of YBa2 Cu3 O7 and complementary metal-oxide-semiconductor device fabrication
on the same sapphire substrate, Appl. Phys. Lett., 63: 1282, 1993.
46. D. Koelle, et al. High performance dc SQUID magnetometers with single layer YBa2 /Cu3 /O7 x flux transformers,
Appl. Phys. Lett., 63: 3630, 1993.
47. R. Cantor, et al. Low-noise, single-layer YBa2 /Cu3 /O7 x dc SQUID magnetometers at 77 K, IEEE Trans. Appl.
Supercond., 5: 2927, 1995.
48. F. Dillnn V. N. Giyantsev M. Siegel Performance of YBa2 /Cu3 /O7 direct current SQUIDs with high-resistance, Appl.
Phys. Lett., 69: 1948, 1996.
49. T. Amrein, et al. Thin film Bi2 Sr2 CaCu2 O8+x Josephson junctions and direct current superconducting quantum
interference devices on (001) SrTiO3 bicrystals, Appl. Phys. Lett., 63: 1978, 1993.
50. T. Takami, et al. Dc superconducting quantum interference devices with BiSrCaCuO bicrystal grain boundary
junctions at 77 K, Jpn. J. Appl. Phys., 35: L391, 1996.
51. C. Carr, et al. Electromagnetic nondestructive evaluation: Moving HTS SQUIDs, including field nulling and dual
frequency measurements, IEEE Trans. Appl. Supercond., 7: 3275, 1997.
52. L. Hao J. C. Macfarlane C. M. Pegrum Excess noise in YBa2 Cu3 O7 thin film grain boundary Josephson junctions and
devices, Supercond. Sci. Technol., 9: 678, 1996.
53. J. Z. Sun, et al. Improved process for high-Tc superconducting step-edge junctions, Appl. Phys. Lett., 63: 1561,
1993.
21
54. A. Cochran, et al. Experimental results in non-destructive evaluation with HTS SQUIDs, in Proc. EUCAS 95,
Institute Physics Conf., 2: 1511, 1995.
55. L. N. Morgan, et al. Electromagnetic nondestructive evaluation with simple HTS SQUIDs: Measurements and
modeling, IEEE Trans. Appl. Supercond., 5: 3127, 1995.
56. Q. Hu P. L. Richards Quasiparticle mixers and detectors, in S. T. Ruggiero and D. A. Rudman (eds.), Superconducting
57. J. R. Tucker M. J. Feldman Quantum detection at millimeter wavelengths, Rev. Mod. Phys., 57: 1055, 1985.
58. P. L. Richards Q. Hu Superconducting components for infrared and millimeter-wave receivers, Proc. IEEE, 77: 1233,
1989.
59. J. A. Stern, et al. NbN/MgO/NbN SIS tunnel junctions for sub mm wave mixers, IEEE Trans. Magn., 25: 1054, 1989.
60. W. R. McGrath, et al. Performance of NbN superconductive tunnel junctions as SIS mixers at 205 GHz, IEEE Trans.
Magn., 27: 2650, 1991.
61. W. Zhen A. Kawakami Y. Uzawa NbN/AlN/NbN tunnel junctions with high current density up to 54 kA/cm2 , Appl.
Phys. Lett., 70: 114, 1997.
62. Y. Uzawa, et al. Submillimeter wave responses in NbN/AlN/NbN tunnel junctions, Appl. Phys. Lett., 66: 1992, 1995.
63. Y. Uzawa W. Zhen A. Kawakami Quasi-optical submillimeter-wave mixeres with NbN/AlN/NbN tunnel junctions,
Appl. Phys. Lett., 69: 2435, 1996.
64. P. Dielman, et al. Shot noise in NbN SIS junctions suitable for THz radiation detection, in Appl. Superconductivity,
Proc. of EUCAS 1997 3rd. Eur. Conf. Appl. Superconductivity, 1: 421, 1997.
65. J. Chen, et al. YBa2 Cu3 O7 angle grain boundary junction on Si bicrystal substrate, Jp. J. Appl. Phys., 30: 1964,
1991
66. P. Seidel S. Linzen F. Schmidl High-Tc thin films and Josephson junctions on silicon and sapphire substrates, in
Extended Abstr. 5th Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 178, 1997.
67. P. A. Rosenthal E. N. Grossman Terahertz Shapiro steps in high temperature SNS Josephson junctions, IEEE Trans.
Microw. Theory Tech., 42: 547, 1994.
68. K. Nakajima et al. THz response of YBCO grain boundary junctions, in Extended Abstr. 5th Int. Workshop High-Temp.
Supercond. Electron Devices, Tokyo, FED-157: 185, 1997.
69. K. K. Likharev V. K. Semenow A. B. Zorin New possibilities for superconductor devices, in S. T. Ruggiero and D. A.
Rudman (eds.), Superconducting Devices, New York: Academic Press, 1990, pp. 149.
70. Collection of articles, IBM J. Res. Develop., 24, 1980.
71. S. Kotani et al. Feasibility of an ultra-high-speed Josephson multiplier, IEEE J. Solid-State Circuits., SC-22: 98,
1987.
72. H. Hayakawa Computing, in S. T. Ruggiero, D. A. Rudman (eds.), Superconducting Devices, New York: Academic
Press, 1990, pp. 101134.
73. T. van Duzer G. Lee Digital signal processing, in S. T. Ruggiero and D. A. Rudman (eds.), Superconducting Devices,
New York: Academic Press, 1990, pp. 197225.
74. R. S. Withers Wideband analog signal processing, in S. T. Ruggiero and D. A. Rudman (eds.), Superconducting
75. K. K. Likharev O. A. Mukhanov V. K. Semenov Resistive single flux quantum logic for the Josephson-junction digital
technology, in H.-D. Hahlbohm and H. Lubbig (eds.) SQUID 85, Berlin: de Gruyer, 1985, pp. 11031108.
76. S. P. Benz C. J. Burroughs C. A. Hamilton Operating margins for a pulse-driven programmable voltage standard,
77. S. P. Benz C. A. Hamilton A pulse-driven programmable Josephson voltage standard, Appl. Phys. Lett., 68: 3171,
1996.
78. K. K. Likharev V. K. Semenov RSFQ logic/memory family: A new Josephson-junction technology for sub-teraherzclock-frequency digital systems, IEEE Trans. Appl. Supercond., 1: 3, 1991.
79. K. K. Likharev Ultrafast superconducting electronics: RSFQ technology roadmap, Proc. Low Temp. Phys. Conf. LT-21,
1996.
80. O. A. Mukhanov RSFQ 1024-bit shift register for acquisition memory, IEEE Trans. Appl. Supercond., 3: 3102, 1993.
81. J. H. Magerlein Specific capacitance of Josephson tunnel junctions, IEEE Trans. Magn., 17: 286, 1981.
82. R. F. Broom et al. Niobium oxide-barrier tunnel junction, IEEE Trans. Elect. Dev., ED-27: 1998, 1980.
83. E. L. Wolf, et al. Proximity electron tunneling spectroscopy I. Experiments on Nb, J. Low Temp. Phys., 40: 19, 1980.
22
84. M. Gurvitch M. A. Washington H. A. Huggins High quality refractory tunnel junctions using thin aluminum layers,
Appl. Phys. Lett., 42: 472, 1983.
85. H. Kroger L. N. Smith D. W. Jillie Selective niobium anodization process for fabricating Josephson tunnel junctions,
Appl. Phys. Lett., 39: 280, 1981.
86. D. Jillie et al. All-refractor, Josephson logic circuits, IEEE J. Sol. State Circuits, SC-18: 173, 1983.
87. P. Bradley A 6-bit Josephson flash A/D converter with GHz input bandwidth, IEEE Trans. Appl. Supercond., 3: 2550,
1993.
88. T. Sterling P. Messina P. H. Smith Enabling Technologies for Petaflops Computing, Cambridge, MA, MIT Press, 1996.
89. N. B. Dubash P.-F. Yuh V. V. Brozenets SFQ data communication switch, IEEE Trans. Appl. Supercond., 7: 2681,
1997.
90. D. L. Miller J. X. Przybysz J. H. Kang Margins and yields of SFQ circuits in HTS materials, IEEE Trans. Appl.
Supercond., 3: 2728, 1993.
91. Y. Ishimaru et al. Observation of boundary Josephson junction with d-wave pairing characteristics, Jpn. J. Appl.
Phys., 34: L1532, 1995.
92. R. Kleiner, et al. Intrinsic Josephson effects in high-Tc superconductors, Phys. Rev., B49: 1327, 1994.
93. R. Kleiner, et al. Dynamic behavior of Josephson-coupled layered structures, Phys. Rev., B50: 3942, 1994.
94. A. Irie G. Oya Flux flow behavior in the intrinsic Josephson junctions in Bi2 Sr2 CaCu2 Oy single crystals, Physica C,
235: 3277, 1994.
95. Y. I. Latyshev J. E. Nevelskaya P. Monceau Dimensional crossover for intrinsic dc Josephson effect in Bi2 Sr2 CaCu2 O8
2212 single crystal whiskers, Phys. Rev. Lett., 77: 932, 1996.
96. A. Yurgens, et al. Strong temperature dependence of the c-axis gap parameter of Bi2 Sr2 CaCu2 O8+ intrinsic Josephson
junctions, Phys. Rev., B53: R8887, 1996.
97. A. Irie M. Sakakibara G. Oya Growth and tunneling properties of (Bi,Pb)2 Sr2 CaCu2 Oy single crystals, IEICE Trans.
Electron, E77-C: 1191, 1994.
98. A. Irie G. Oya Microwave response of intrinsic Josephson junctions in BSCCO single crystals, IEEE Trans. Appl.
Supercond., 5: 3267, 1995.
99. A. Irie M. Iwama G. Oya Novel microwave-induced steps of intrinsic Josephson junctions in mesa-shaped BSCCO
single crystals, Supercond. Sci. Technol., 9: A14, 1996.
100. C. Neumann et al. Fabrication of high Ic Rn YBCO-Josephson-junctions on MgO-substrates using a focused-ion-beam
system, Physica, C210: 138, 1993.
101. A. J. Pauza et al. High-Tc Josephson junctions by electron beam irradiation, IEEE Trans. Appl. Supercond., 3: 2405,
1993.
102. S. K. Tolpygo, et al. Normal-state properties and Josephson effects in HTS weak links produced by electron beam,
103. B. A. Davidson et al. Superconductor-normal-superconductor behavior of Josephson junctions scribed in
Y1 Ba2 Cu3 O7 by a high-brightness electron source, Appl. Phys. Lett., 68: 3811, 1996.
104. D. F. Moore et al. Asymmetric YBaCuO Interferometers and SQUIDs made with focused electronbeam irradiation
junctions, IEEE Trans. Appl. Supercond., 7: 2494, 1997.
105. W. E. Booij et al. Electrodynamics of closely coupled YBCO junctions, IEEE Trans. Appl. Supercond., 7: 3025,
1997.
106. A. J. Pauza et al. Electron-beam damaged high-temperature superconductor Josephson junctions, J. Appl. Phys., 82:
5612, 1997.
107. S.-J. Kim et al. Successful fabrication of bicrystal Si substrates for YBa2 bCu3 O7 y Josephson junctions, Cryogenics,
35: 901, 1995.
108. S.-J. Kim et al. Microwave induced steps of YBCO weak link damaged by focused electron beam irradiation, in
Extended Abstr. 5th Int. Workshop on High-Temp. Supercond. Electron Devices, Tokyo, FED-157, 182, 1997.
109. L. Antognazza et al. Proximity effect in YBa2 Cu3 O7 /YBa2 Cu1 x Cox )3 O7 /YBa2 Cu3 O7 junctions: From the clean
limit to the dirty limit with pair breaking, Phys. Rev., B51: 8560, 1995.
110. W. H. Mallison et al. A multilayer YBa2 Cu3 Ox Josephson junction process for digital circuit applications, Appl. Phys.
Lett., 68: 3808, 1996.
111. H. Sato et al. Improvement in parameter spreads of YBaCuO/PrBaCuO/YBaCuO trilayer junctions, Jpn. J. Appl.
Phys., 35: L1411, 1996.
23
112. Q. X. Jia, et al. Directly coupled direct current superconducting quantum interference devices based on ramp-edge
Ag: YBCO/PBCO/Ag: YBCO junctions, Appl. Phys. Lett., 72: 3068, 1998.
113. M. A. J. Verhoeven High-T c superconducting ramp-type junctions., PhD thesis, University of Twente, The Netherlands, 1996.
114. M. A. J. Verhoeven, et al. Ramp-type junction parameter control by Ga doping of PrBa2 Cu3 O7 barriers, Appl. Phys.
Lett., 69: 848, 1996.
115. S. J. Bending M. R. Beasley Transport processes via localized states in thin a-Si tunnel barriers, Phys. Rev. Lett., 55:
324, 1985.
116. M. A. J. Verhoeven G. J. Gerritsma H. Rogalla Ramp-type junctions with very thin PBCO barriers, in Proc. EUCAS
95 Conf., Edinburg, 1995, p. 1395.
117. J. B. Barner A. W. Kleinsasser High-T c superconductornormalsuperconductor junctions with polyimide-passivated
ambient-temperature edge formation, IEEE Trans. Appl. Supercond., 7: 2502, 1997.
118. H. Myoren et al. Terahertz response of high-T c ramp-type junctions on MgO, in Extended Abstr. 5th Int. Workshop
High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 188, 1997.
119. E. Demler et al. Proximity effect and Josephson coupling in the SO(5) theory of high-T c superconductivity, Phys. Rev.
Lett., 80: 2917, 1998.
120. M. A. J. Verhoeven et al. Ramp type HTS Josephson junctions with PrBaCuGaO barriers, IEEE Trans. Appl.
Supercond., 5: 2095, 1995.
121. A. E. Lee J. A. Luine C. L. Pettiette-Hall Co-doped-YBCO SNS junctions for superconductive integrated circuits,
122. M. G. Forrester et al. A single flux quantum shift register operating at 65 K, IEEE Trans. Appl. Supercond., 5: 3401,
1995.
123. V. K. Kaplunenko Voltage divider based on submicron slits in a high Tc superconducting film and two bicrystal grain
boundaries, Appl. Phys. Lett., 67: 282, 1995.
124. S. J. Berkowitz Int. Workshop on Superconductivity, High-temperature superconducting circuits using SNS long
junctions transistors, High Temperature Supercond. Electron. Fund. Appl., Int. Superconductivity Technol. Centre,
Tokyo Japan, 1996, p. 40.
125. B. H. Moeckly, et al. Interface-engineering high-Tc Josephson junctions, Appl. Supercond, 6: 317, 1998.
126. J. Talvacchio, private communication.
127. K. Char HTS SNS Josephson junctions: Interfaces and mechanisms, Physica C, 282287: 419, 1997.
128. L. Antognazza K. Char T. H. Geballe High-T c edge junctions with YPrBaCuCoO barrier layers near the metal
insulator transition, Appl. Phys. Lett., 70: 3152, 1997.
129. S. V. Polonsky et al. New RSGQ circuits (Josephson junction digital devices), IEEE Trans. Appl. Supercond., 3: 2566,
1993.
130. J. Yoshida et al. High-T c Josephson junction technology for digital applications, in Extended Abstr. 5th Int. Workshop
131. J. Yoshida T. Nagano Tunneling and hopping conduction via localized states in thin PBCO barriers, Phys. Rev., B55:
11860, 1997.
132. T. Nagano, et al. Thin film growth for fabrication of YBaCuO/PrBaCuCoO/YBaCuO ramp-edge junctions, in Extended
Abstr. 5th Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo FED-157: 233, 1997.
133. A. W. Kleinsasser W. J. Gallagher Three-terminal devices, in S. T. Ruggiero and D. A. Rudman (eds.), Superconducting
134. K. K. Likharev et al. Josephson junction with lateral injection as a vortex transistor, IEEE Trans. Magn., 15: 420,
1979.
135. W. E. Booij, et al. Supercurrents in closely spaced high T c Josephson junctions, in Proc. ATS-Workshop 96, Twente,
The Netherlands, 1996, p. 6.
136. T. Akazaki et al. A Josephson field effect transistor using an InAs-inserted-channel In0.52 Al0.48 /As/In0.53 /Ga0.47 As
inverted modulation-doped structure, Appl. Phys. Lett., 68: 418, 1996.
137. A. W. Kleinsasser T. N. Jackson Prospects for proximity effect superconducting FETs, IEEE Trans. Magn., MAG-25:
1274, 1989.
138. H. Takayanagi T. Akazaki Superconducting three-terminal devices using HEMT structures, in Extended Abstr. 5th
Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 153, 1997.
24
139. J. Mannhart, et al. Electric field effect on superconducting YBaCuO films. Z. Phys., B83: 307, 1991.
140. X. X. X. Xi, et al. Electric field effects in high T c superconducting ultrathin YBaCuO films, Appl. Phys. Lett., 59: 3470,
1992.
141. K. Matsui, et al. Analysis on electric field effect in Al/SrTiO3 /YBa2 Cu3 Oy structure, Jpn. J. Appl. Phys., 31: L1342,
1992.
142. K. Joosse, et al. Multilayers for high-Tc superconducting electric field effect devices, J. Alloys Comp., 195: 723, 1993.
143. T. Nakamura et al. Study on a superconducting channel transistor, Sumitomo Elect. Tech. Rev., No. 38, 35, 1994.
144. R. Schneider R. Auer Temporal response of a high-Tc superconducting field effect transistor, Appl. Phys. Lett., 67:
2075, 1995.
145. T. Nakamura H. Tokuda M. Iiyama Electric field effects of SrTiO/YBaCuO multilayers grown by all-MBE, IEEE
146. T. Fujii, et al. Metalinsulatorsuperconductor field-effect-transistor using SrTiO/YBaCuO heteroepitaxial films,
Jpn. J. Appl. Phys., 31: L612, 1992.
147. T. Saito, et al. Fabrication of a High-T c superconducting field effect transistor by ion beam sputtering. IEEE Trans.
148. T. Goto, et al. Properties of high-T c field effect devices with ion beam sputtered SrTiO Film, in Extended Abstr. 5th
Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 159, 1997.
149. T. Nakamura H. Inada M. Iiyama Superconducting field effect transistor of YBCO/SrTiO3 /YBCO tri-layer structures,
in Extended Abstr. 5th Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 123, 1997.
150. U. Ghoshal, et al. CMOS amplifier designs for Josephson-CMOS interface circuits, IEEE Trans. Appl. Supercond., 5:
2640, 1995.
151. D. Crupta et al. Optimization of hybrid JJ/CMOS memory operating temperatures, IEEE Trans. Appl. Supercond.,
7: 3307, 1997.
152. K. Hikosaka, et al. Josephson-semiconductor high-speed interface circuit using 77 K analog HEMT amplifiers, in
Extended Abstr. 5th Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 55, 1997.
153. J. Mannhart High-Tc transistors, Supercond. Sci. Technol., 9: 49, 1996.
154. J. Chen et al. Possible three-terminal device with YBCO angle grain boundary, IEEE Trans. Appl. Supercond., 1:
102, 1991.
155. J. Mannhart et al. Large electric field effects in YBa2 Cu3 O7 films containing weak links, Appl. Phys. Lett., 62: 630,
1993.
156. Z. G. Ivanov Field effect transistor based on a bi-crystal grain boundary Josephson junction, IEEE Trans. Appl.
Supercond., 3: 2925, 1993.
157. B. Mayer J. Mannhart H. Hilgenkamp Electric field controllable Josephson junctions of high quality in high-Tc
superconductors, Appl. Phys. Lett., 68: 3031, 1996.
158. I. Iguchi N. Nukui K. Lee Dynamic Cooper-pair breaking by tunnel injection of quasiparticles into a high-Tc
YBa2 Cu3 O7 superconductor, Phys. Rev., B50: 457, 1994.
159. E. Kume H. Hiroyuki I. Iguchi Quasi-particle injection effect of YBCO thin films using double-injector junctions having two-holes, in Extended Abstr. 5th Int. Workshop High-Temp. Supercond. Electron Devices, Tokyo, FED-157: 167,
1997.
160. D. J. Frank M. J. Brady A. Davidson A new superconducting-base transistor, IEEE Trans. Magn., MAG-21: 721,
1985.
161. T. Yamada, et al. (Ba, Rb)BiO3 base and YBa2 Cu3 Ox flux flow transistors, in Extended Abstr. 5th Int. Workshop
162. H. Tamura A. Yoshida S. Hasuo Transistor action based on field-effect controlled current injection into an
insulator/SrTiO3 interface, Appl. Phys. Lett., 59: 298, 1991.
163. A. Yoshida, et al. Dielectric-base transistor using YBa2 Cu3 O7 x /NdGaO3 /SrTiO3 heterostructures, J. Appl. Phys., 71:
5284, 1992.
164. T. Hato et al. Improved emitter-base junction with In2 O3 in dielectric-base transistor Jpn J. Appl. Phys., 34: 6379,
1995.
165. C. Yoshida, et al. Electric field effect in LaTiO3 /SrTiO3 heterostructure, Jpn. J. Appl. Phys., 35: 5691, 1996.
166. T. Hato, et al. Dielectric-base transistor with doped channel, Appl. Phys. Lett., 70: 2900, 1997.
25
167. A. W. Kleinsasser W. J. Gallagher Three-terminal devices, in S. T. Ruggiero and D. A. Rudman (eds.), Superconducting
Devices, 372. New York: Academic Press, 1990, pp. 325372.
168. A. W. Kleinsasser H. Weinstock R. W. Ralston (eds.) Superconducting Devices, Dordrecht: Kluwer Academic, 1993, p.
249.
169. J. Mannhart Changes in the superconducting properties of high-Tc cuprates produced by applied electric fields, Mod.
Phys. Lett., B6: 555, 1992.
170. O. Sangjun D. Youm M. R. Beasley A superconductive magnetoresistive memory element using controlled exchange
interaction, Appl. Phys. Lett., 71: 2376, 1997.
171. J. S. Moodera J. Nowak J. M. van de Veerdonk Interference magnetism and spin wave scattering in ferromagneticinsulator-ferromagnet tunnel junctions, Phys. Rev. Lett., 80: 2941, 1998.
172. P. K. Wong J. E. Evetts M. G. Blamire High conductance small area magnetoresistive tunnel junctions, Appl. Phys.
Lett., 73: 384, 1998.
173. S. Sankar A. E. Berkowitz D. J. Smith Spin-dependent tunneling in discontinuous Co-SiO2 magnetic tunnel junctions,
Appl. Phys. Lett., 73: 535, 1998.
174. J. Z. Sun, et al. Temperature and bias dependence of magnetoresistance in doped manganite thin film trilayer
junctions, Appl. Phys. Lett., 73: 1008, 1998.
175. K. Inomata Y. Saito Spin-dependent tunneling through layered ferromagnetic nanoparticles, Appl. Phys. Lett., 73:
1143, 1998.
S. T. RUGGIERO
University of Notre Dame
108

Superconductors differ from normal metals in some truly remarkable ways. In a normal metal, conduction electrons behave as individual particles that move thorough the material,
interacting with each other and with the array of ions that
1000
HC
800
B (mT)
form the background lattice. When a voltage is applied, the

electrons carry current, scatter randomly, show electrical resistance, and obey Ohms law. Electrons in a Pb wire at 8 K,
for example, will be scattered from impurity atoms and dissipate energy. This electron-scattering process can be modeled
by treating the electrons as single particles subject to the
Coulomb and exchange forces of the neighboring electrons.
The individual-particle picture works well, and one can discuss one electron without concern for the events happening to
electrons that are many atomic spacings away.
In a superconducting metal, this individual-particle picture does not hold. Rather, the motion of one electron is
highly correlated with other electrons in the metal over very
large distances. Again using Pb as an example, when a sample is cooled below the superconducting transition temperature Tc of 7.25 K (1), the electrical resistance drops by more
than a factor of 1015 in a temperature interval of a few millikelvins. In the course of this transition the motion of the electrons transforms from a single-particle picture to a highly correlated picture. These correlations extend over macroscopic
distances and the electrical resistance vanishes. During the
superconducting transition, the wave function of the electrons
develops a certain rigidity, and the electrons tend to move as
a giant molecule rather than as individual particles.
This rigidity is reflected in a special feature of superconductivity called phase locking. Electrons, like all elementary
particles, have both a particle and a wave character and the
wave properties are described by a wave function Aei
with both an amplitude A and a phase . In the course of the
superconducting transition, the phase of any one superfluid
electron becomes tightly locked to the phase of all the neighboring superfluid electrons. This rigidity and phase locking of
the wave function leads to most of the remarkable properties
of superconductors. It leads to the suppression of scattering,
to zero electrical resistance, to the expulsion of magnetic flux
from the interior of a superconductor, and to all of the marvelous aspects of quantum interference effects that are so crucial
to microelectronic devices. It is important to understand how
this rigidity and phase locking come about.
There are two very broad classes of superconductors that
are distinguished from one another by the way magnetic flux
is distributed inside the material. In one class, called type I,
the magnetic flux is totally excluded from the interior of a
bulk sample, and the magnetic induction B is zero everywhere except for a thin layer that is a few hundred nanometers thick near the surface. This thin layer is called the penetration depth , and the B 0 state is called the Meissner
state. In this thin layer, surface currents flow to cancel the
applied field and give B 0 in the interior. It is useful to
recall here that B in the interior is the sum of the applied
field, 0H, plus the field due to circulating currents in the
material, 0M, where H is the magnetic field and M is the
volume magnetization. When the external magnetic field
reaches a certain critical magnitude called the thermodynamic critical field Hc, the superconducting state collapses
and the material reverts to the normal state at higher fields.
This is illustrated by the dashed curve in Fig. 1.
In the second class of materials called type II, as you increase the magnetic field from zero, the flux is totally excluded at first, just as for type I material. Then, at some characteristic field called Hc1, the magnetic flux begins to enter
the material in the form of tiny vortices of magnetic flux,
109
HC2
600
Type ll
400
HC1
200
0
200
400
600
0 H(mT)
Type l
800
1000
Figure 1. B vs 0H plot to illustrate the defining characteristics of

type I and type II superconductors.
called flux quanta 0. Each quantum carries 2.07 1015 Wb

of magnetic flux. With further increases in H, the quantized
vortices flood into the material until an upper critical field,
Hc2, is reached where the sample again reverts to the normal
state. These lower and upper critical fields are illustrated on
the diamond points of Fig. 1.
The purpose of this article is to review the microscopic origins of these two types of magnetic behavior for superconductors. We will discuss the features that distinguish type I from
type II materials, and describe the way in which type I materials can be transformed into type II materials by alloying.
Discussion will include both the classical superconductors and
the cuprate superconductors that were discovered in 1986 (2)
to have extraordinarily high Tc. After taking a brief look at a
few typical examples of type I and type II behavior, we will
review some features of the superconducting wave function
that leads to the remarkable properties of superconductors.
Then we will return to a more detailed discussion of more
subtle aspects of type I and type II behavior. Various practical
methods to measure the characteristic fields of Hc, Hc1, and
Hc2 will be explored. Throughout, there will be an attempt to
provide useful equations for computation, even though the
derivation of the relations is beyond the scope of this presentation. The goal here is to discuss the behavior of these two
classes of superconductors using simple physical pictures
wherever possible.
TYPICAL EXAMPLES
Superconductivity is a very common phenomenon. Approximately half the elements in the periodic table and a large
number of intermetallic compounds show the effect. Often, it
is necessary to go to rather low temperatures and in some
cases it is necessary to apply high pressure to transform the
material into a superconductor. Although there are exceptions, pure elemental metals tend to be type I superconductors and alloys tend to be type II superconductors.
Low-Tc Type I Superconductors
In many classical superconductors, such as high-purity Pb,
Hg, and In (1,3), the magnetization curves, M versus H, exhibit a Meissner (B 0) behavior up to some critical field at
which the magnetic flux suddenly collapses into the sample
110
100
Type I superconductors
80
0 M(mT)
T = 1.28 K
60
Pb
40
Hg
20
In
0
20
40
60
0 H(mT)
80
100
Figure 2. Magnetization curves for Pb, Hg, and In at 1.28 K showing

typical values of Hc.
as shown in Fig. 2 for 1.28 K data. Generally speaking, very

pure metals that have s-band or p-band conduction electrons
at the Fermi surface will show this so-called type I behavior.
Very pure metals that have a relatively high Fermi velocity
vF and a relatively low Tc tend to be type I superconductors.
A more detailed look at the magnetization curves reveals
that the collapse of flux into a bulk sample occurs over a finite
magnetic field interval. When the flux begins to enter the
sample, the slope of the line is governed by demagnetizing
effects (4), that is, effects associated with the shape of the
sample. As shown in Fig. 3, a magnetic field applied to a superconductor in the Meissner state has a larger field at the
equator of the ellipse than the applied field Ha. For this elliptical sample, flux begins to enter the sample at Ha (1
)Hc and the transition is complete at Hc. is called the demagnetizing factor. Because the free-energy difference between the superconducting and normal state is directly related to Hc, it is called the thermodynamic critical field curve.
Intermediate State
Between (1 )Hc and Hc, the material breaks up into an
intermediate state in which both Meissner regions and normal regions coexist in the sample. When the screening currents that are flowing in the penetration depth region near
the surface reach a critical value, normal regions will nucleate at the surface and propagate into the interior of the sam-
100
ple. Usually this happens when the kinetic energy of the electrons in the circulating screening current are a significant
fraction of characteristic superconducting energy, kBTc. Here,
kB is the Boltzmann constant. The detailed shape of the Meissner regions in the intermediate state depends on structure
of the material and the shape of the sample. Normal regions
nucleate at the surface and often propagate into the interior
in the form of lamellar regions as shown in Fig. 4. For a long
slender sample with the magnetic field parallel to the long
axis of the sample, is close to zero and the region of the
intermediate state is narrow. This behavior is illustrated by
the data in Fig. 2. For a sphere, is , and the magnetization
curve is illustrated in Fig. 3. For a flat plate with the magnetic field perpendicular to the plane of the plate, is close to
1, and the sample enters the intermediate state at fields far
below Hc. Shoenberg, p. 103 (5), gives a detailed discussion of
the intermediate state for spheres and Tinkham, p. 25 (6),
gives a detailed discussion for flat slabs.
Low-Tc Type II Superconductors and the Vortex State
Magnetization curves for three different transition metals are
shown in Fig. 5. Very pure Ta is a type I material (7) with an
abrupt collapse of the superconducting state similar to Pb.
The other two very pure transition metals, Nb and V, show a
much broader transition to the normal state and are called
type II superconductors (8,9). When very pure metals show
type II behavior, they are called intrinsic type II superconduc-
Demagnetization factor
= 0.33 = 0.10 = 0.01
80
40
20
0
Transition metals
60
60
0 M(mT)
0 M(mT)
Figure 4. A typical magnetic flux distribution in the intermediate

state of a type I superconductor as lamella of normal-state regions
(shaded area) collapse into the interior. This is for a flat plate with
H perpendicular to the plate.
20
40
60
0 H(mT)
80
Ta 2.2 K
Nb 7.7 K
40
V 4.2 K
20
100
Figure 3. Change in the shape of the magnetization curves for a

given material as the demagnetizing factor increases. The inset
shows an elliptical sample in the Meissner state with the field at the
equator larger than the applied field.
20
40
60
0 H(mT)
80
Figure 5. Magnetization curves for three transition metals to illustrate both type I (Ta) and type II (Nb and V) behavior.
Changes with increasing
60
Type l
0 M(mT)
tors. The interval between Hc1 and Hc2 is called the vortex
state because in this magnetic field interval, the sample fills
with vortices, each carrying one quantum of flux, 0. In all of
these intrinsic type II materials, the conduction electrons at
the Fermi surface are mostly d band in character, so the
Fermi velocity is relatively low. In addition, Tc is relatively
high. The magnetization data in three samples shown here
are highly reversible, and therefore the first flux entry is very
close to the field at which a vortex is thermodynamically stable in the material, Hc1. For the data shown in Fig. 5, a temperature is chosen so that Hc is in the vicinity of 50 to 60 mT.
This permits an easy comparison of the shapes of the curves.
As the temperature is decreased, the ratio of Hc2 /Hc typically
rises by about 50%, and the slope of the magnetization at
Hc2 always decreases. Among the transition metals, very pure
Ta (9) is an exception in that it is type I. A very small amount
of impurity, however, will transform it to a type II material
very similar to Nb and V.
111
40
Partial
first order
20
Second order
20
40
60
80
0 H(mT)
100
120
Figure 6. Transformation from type I to type II behavior as l is gradually increased. For just greater than 0.707, there is an attractive
interaction between vortices and a first-order transition at Hc1. This
illustrates the change in shape of magnetization curves as impurities
are added.
Low-Tc Alloys
(1)
Small- materials are type I, and large- materials are called

type II. The transition from type I to type II behavior occurs
at 1/ 2 0.707. In a type I superconductor, Hc is directly
related to the free-energy difference between the superconducting and normal state by the area under the magnetization curve,
Hc
Gn Gs =
0
0 MdH = 0 H 2c /2
(2)
In type II materials, Hc retains this definition. The ratio

from Eq. (1) also is related to the characteristic critical fields
on the magnetization curve via
= Hc2 / 2Hc
(3)
The range of the correlations among the electrons and hence
in a superconductor can be reduced by shortening the normal-state mean free path of the electrons, l. Hence, one
should expect that alloying would reduce both l and and

would cause a transition from type I to type II behavior if
remains roughly constant. This was shown to be true, for example, by Kumpf (10), who demonstrated that pure Pb with
0.4 could be converted to a type II material with 0.84
with the addition of 2.0 at. % Tl. For this Pb system, it requires a resistivity of n 2 cm to change Pb from a type
I to a type II superconductor. To convert pure Ta from type I
to type II, it requires alloying until n 0.5 cm.
The sketch in Fig. 6 shows the changes that occur in the
magnetization curves as is gradually increased for a material similar to Ta. For substantially less than 0.707, the
magnetization curves are typical type I with a complete firstorder transition to the normal state at Hc. A first-order transition has a latent heat, but a second-order transition has no
latent heat. For comparable to 1/ 2, there is a partial firstorder transition at Hc1 with a tail in the magnetization extending out to Hc2 as shown by the dashed line in Fig. 6. For
substantially larger than 2, the phase transition is second
order at Hc1. As shown by Auer and Ullmaier (7), the range of
showing the type II behavior with a first-order transition
at Hc1 depends on temperature. Figure 7 is a sketch showing
roughly the expected behavior. The horizontal line on Fig. 7
is 1/ 2.
Value
To understand the difference between type I and type II superconductors on a microscopic scale, it is essential to know
that there is a characteristic distance, called the coherence
distance, , over which the superconducting wave function can
change. In a superconductor, the basic charge-carrying unit
in the system is a highly correlated pair of electrons called
the Cooper pair. The minimum distance in which the density
of superconducting electrons, ns, can change from the value in
the bulk superconductor to zero in a normal metal is roughly
the size of these Cooper pairs. The coherence distance, or the
size of the Cooper pairs, is a very important property of a
superconductor and typically varies from about 1 m in Al to
as small as 2 nm in the cuprate superconductors. In the transition metals such as Nb and V 30 nm and in the conventional type I materials such as Sn, Pb, Hg, and In 500 nm.
The critical factor governing the shape of the magnetization curves of superconductors is the ratio of the magnetic
field penetration depth to the coherence distance, / . This
ratio is sufficiently important that it is given a symbol of its
own,
1.2
Type ll second-order transition
0.8
Type ll
first - order transition
= 0.707
0.4
00
Type l first-order transition
0.2
0.4
0.6
T/TC
0.8
1.2
Figure 7. The transformation from type I to type II behavior on a

vs. T plot. This qualitatively shows the boundaries of these three different types of phase transitions.
112
Because l is closely connected to the normal-state resistivity n, can be related to n and the value for the pure
metal, 0, by the useful relation
= 0 + 7.5 105 1/2 n
(4)
where is the electronic specific constant and n is the resistivity. If you use units where is in erg/cm3 K2 and n is in
cm, then the constant is 7.5 103. For the extreme dirty
or short mean-free-path limit, Hake (11) showed that Ti16
at. %Mo samples with 100 cm can be nearly reversible with a value of 66.
Among the high-purity s-p band metals, ranges from 0.01
for Al, to 0.15 for Sn, to 0.4 for Pb. Among the high-purity
transition-metal superconductors, ranges from 0.36 for Ta,
to 0.78 for Nb, to 0.90 for V. Nitrogen impurities are particularly good to show the transition from type I to type II behavior because they go into the lattice of Ta as statistically distributed interstitial atoms. Hence, N decreases the electronic
mean free path without forming clusters that would substantially increase the pinning of vortices and irreversibility effects. As shown by Auer and Ullmaier (7), Ta is transformed
from a type I to a type II superconductor when the sample
residual resistivity 0 585 cm and attains a value of
about 1.5 when 0 2000 cm.
High-Tc Materials
The high-temperature superconductors (HTS) such as
La1.85Sr0.15CuO4, La(214), are qualitatively different from all of
the metals that had been studied before. They are different
because the normal state is created by doping an insulator.
They also are different because they tend to be rather anisotropic with the charge carriers moving most easily in the a-b
planes of the CuO2 sheets. The parent cuprate for La(214)
with no Sr doping, La2Cu1O4 is not a superconductor and is
not a metal. Rather, it is an antiferromagnetic insulator. If,
however, part of the trivalent La ions are replaced by divalent
Sr ions, holes are created in the copper oxide planes and the
material becomes an anisotropic normal metal and a superconductor with Tc of about 42 K. In these oxide conductors,
the carrier mobility along the copper oxide planes is much
higher than it is perpendicular to the copper oxide planes, so
both the normal conductivity and the superconductivity are
quite anisotropic.
In the superconducting state, the magnetization data for
the cuprate superconductors show type II behavior with very
high values, commonly about 100. Hence, Hc2 is over
100 times larger than Hc. For many of these cuprates, the
penetration depth is about 200 nm, and the coherence distance along the copper oxide planes is about 2 nm. Along the
c axis, the coherence distance is even smaller at 0.5 nm.
PHYSICAL PICTURES FOR TYPE I AND TYPE II PHENOMENA
Superconductivity is rather special in the field of condensedmatter physics because it is a manifestation of quantum mechanics on a macroscopic scale. If one induces a supercurrent
to flow in a superconducting ring, the circulating charge carriers obey the Bohr-Sommerfeld quantization condition even if
the diameter of the ring is hundreds of micrometers. In a single atom, such as the hydrogen atom, the angular momentum
of the circulating electron is quantized because the wave function of the electron must be single valued going around the
atom by 2. This quantization of angular momentum in the
hydrogen atom is then reflected in a quantization of the magnetic moment in units of Bohr magnetons. The same kind of
quantization occurs for superconducting electrons circulating
in a large ring. Because the electrons are phase locked, the
wave function must be single valued and the Bohr-Sommerfeld quantization condition, pdq n, is obeyed. Here, p is
the momentum, dq is the path element, and n is an integer.
This creates quantized circulating currents, and the resulting
flux also is quantized in units of 0 h/2e.
Even though quantum mechanics is fundamental to understanding superconductivity, there are some simple pictures
that enable the beginner to visualize where the electrons or
holes are and how they interact. The goal of this section is to
present some of the vocabulary and ideas of the Bardeen-Cooper-Schrieffer (BCS) (12) theory in a way that one can picture
the basic behavior of type I and type II superconductors.
Idea 1: All Superconductors Look Alike
There is a great deal of similarity in the physical properties
of superconductors. If the thermodynamic critical field is plotted as a function of temperature or the superconducting energy gap in the excitation spectrum, , as a function of temperature, the curves have the same functional form and the
values scale with the transition temperature. The ratio of
Hc(T 0)/Tc is always about 10 mT/K, and the ratio of
(T 0)/kBTc 1.8 for the low Tc superconductors. Because
superconductors are so similar, there is a reasonable expectation that the details of the metal are not terribly important
in the basic explanation of superconductivity. A rather general and simple theory may explain the effect.
Idea 2: BCS Theory
In the development of a theory of superconductivity, BCS constructed the superconducting ground state by taking special
combinations of normal-state wave functions. At the superconducting transition, the space dependence of the wave functions do not change, but rather the occupation probability of
a given state changes. To create the superconducting wave
function, the electrons are allowed to exchange phonons, thus
scattering around the Fermi surface and coherently mixing
the normal-state wave functions. Instead of the random occupation of states at the Fermi surface that occurs in a normal
metal, the system can gain energy via phonon exchange if the
electronic states are occupied in pairs. This pairing increases
the amount of phonon exchange that can occur and each phonon exchange lowers the energy a bit. To define terms more
explicitly, pair occupation means that if the state with momentum k is occupied, then the state with momentum k
also is occupied. Similarly, if the state with k is empty, then
the state with k also is empty. The pairs are chosen to
have equal and opposite momentum because, by symmetry,
this choice gives the maximum number of final states for the
electron-phonon scattering and maximizes the phonon exchange. As the metal undergoes the superconducting transition, it is the probability of occupation that changes. It
changes from a random occupation to a pair occupation.
In BCS theory, three essential variables are used. First,
the normal-state density of states at the Fermi surface, N(0),
is a measure of the number of electrons participating. The

number of electrons that can take advantage of phonon exchange is determined by N(0). Second, the Debye energy D
is a measure of the range of phonons available for the electron-phonon exchange process. Third, the strength of the electron-phonon interaction, V, is a measure of the amount of energy the electrons gain in a phonon exchange. Within the
theory, the superconducting transition temperature Tc will be
calculated to be
kB Tc ~D e1/N (0)V
(5)
This provides a connection between the transition temperature and the three variables in the theory. If, in addition, one
works out the minimum energy to create an excitation out of
the superfluid ground state by disrupting one pair of electrons, at T 0, this turns out to be
0 1.8kB Tc
(6)
Disrupting a pair simply means preventing that pair of states

from participating in the coherent phonon exchange. This
minimum energy to create an excitation in a superconductor,
, is often called the order parameter or the energy gap, and
typically it is about 1 meV for a Tc of 7 K. The value of
decreases from the T 0 value, 0, as the temperature rises
and goes to zero at Tc. Our goal here is to provide a physical
picture to go with these equations.
A first step in the development of the BCS theory was the

Cooper-pair problem (13). Cooper showed that there is a fundamental instability in an electron gas if one introduces an
attractive interaction that scatters electrons around the
Fermi surface. In this problem, you start with one pair of electrons outside a Fermi sea. Without the electronphonon interaction, the energies available to the electrons are the usual
free-electron energies above the Fermi energy EF, as shown
by the dotted lines in Fig. 8. The solid horizontal lines are
filled states below the Fermi sea. When the electronphonon
interaction is introduced, the new energies for the perturbed
EF
system, W, are represented by the heavy solid lines. Here

they are plotted as a function of the strength of the interaction, V. Note that if V is positive or repulsive, all the energies
are pushed up a bit above the unperturbed dotted line values.
Note also that if V is negative or attractive, all of the energies
are pushed down a bit except the level just above EF. This
level is pushed far below EF. The splitoff of this state far below EF means that a free-electron gas with random occupation
of states is unstable to the formation of pairs when an attractive interaction is introduced. The energetically most favored
combination of electrons for this phonon exchange occurs
when the pair has zero center-of-mass momentum (electrons
have equal and opposite momentum), hence, giving the lowest
energy. Experiments have shown that the pairs of electrons
also are known to have opposite spin. Presumably this antiparallel spin arrangement arises from a quantum-mechanical
effect called exchange forces. Electrons with opposite spins
are closer together than electrons with parallel spins so they
can take better advantage of the electronphonon interaction.
Hence, pairs with opposite spin and momentum have the
lower energy. If some other type of interaction were to cause
pairing, then a parallel spin arrangement is possible.
A special feature of the Cooper pair problem is that the
normal-state wave functions that are being mixed by the electronphonon interaction all add in phase at some point in
real space to form a wave packet. This packet is the so-called
Cooper pair and by the summation of occupied states, it has
a real-space extent of the coherence distance,
= ~vF / 0 = 0.18~vF/kB Tc
Idea 3: The Cooper-Pair Problem
Figure 8. Energy-level diagram for the Cooper-pair problem. Here

W is the new energy of the pair state and V is the strength of the
perturbing interaction. Note that for large and negative V, one state
falls far below the Fermi energy.
113
(7)
Equation (7) shows that the coherence distance, or the size of

the Cooper pairs, is directly related to the ratio of vF /Tc. Because the s-p band metals tend to have relatively high vF and
low Tc, they would be expected to have large coherence distances and hence tend to be type I superconductors. The basic
charge-carrying unit in the circulating current around a superconducting vortex in a type II material in the vortex state
is the Cooper pair.
Idea 4: BCS Theory and Pair-Pair Correlations
The Cooper-pair problem is only part of the picture because
it describes just one pair of electrons outside a Fermi sea. The
total problem must deal with all of the electrons and must
deal with their correlated motion. If typical numbers are put
into Eq. (7), Al has 1600 nm, Sn has 230 nm, Pb has
83 nm, and Nb has 40 nm. Given the normal density
of electrons in a metal, there are thousands to millions of
pairs occupying the space of any single pair. Hence there is a
great deal of pair-pair overlap. These highly interacting and
overlapping Cooper pairs provide a physical picture for the
origin of phase locking over macroscopic distances.
The high-temperature superconductors are a very special
case because the coherence distance is so short. Typically,
2 nm so the overlap of Cooper pairs is much less in this
class of materials and the phase locking is less strong. For
HTS, the number of pairs overlapping any one pair is measured in dozens rather than thousands to millions. This leads
to a less rigid vortex lattice and a greater susceptibility to
flux creep.
The BCS problem is similar in some ways to a variational
calculation in an undergraduate quantum-mechanics course.
114
One guesses a form for the wave function with the pair occupation number as an adjustable parameter. Next one writes
an expression for the free energy of the system in terms of
this trial wave function and minimizes the energy as a function of pair occupation. In both the Cooper and BCS problems,
a paired trial wave function is selected in order to maximize
the final states available for electron-phonon scattering. One
result of the theory is that it is energetically favorable to mix
normal-state wave functions within an energy band that is
wide near EF. All of the pairs share the same states so there
is an energy advantage for correlated motion of many pairs to
make most efficient use of the states available for electronphonon exchange. Pair-pair correlations are critical because
they provide the fundamental mechanism to propagate phase
coherence over long distances.
To create an excitation from the superconducting ground
state, one of the pair states is simply disrupted. If a pair state
is broken, say by injecting an electron into one of the two
states of a pair in the metal, then other pairs cannot use that
channel for phonon exchange. That pair state is removed from
the coherent phonon exchange for all of the other electrons.
This raises the energy of all the electrons in the ground state.
In computing the ground-state energy and the excitation energies, one has to give up the single-particle picture and go to
a highly correlated many-paired electron picture in which the
disruption of one pair changes the energy of all the pairs in
the ground state.
Idea 5: Phase Locking and Rigidity of the Wave Function
A central feature of superconductivity is that the electrons
phase lock into a many-electron ground-state wave function
that has a substantial amount of rigidity. If the system is
disturbed, it responds as a giant unit rather than responding
as individual particles. In the Cooper problem, a pair of electrons is formed by mixing normal-state wave functions so that
they all add in phase at some point in space. In the BCS problem, pair-pair correlations play a central role because all of the
pairs are using the same states for electron-phonon scattering. This leads to a highly correlated ground-state superfluid
that extends over macroscopic distances. The ground state is
somewhat like a giant macromolecule in that the superfluid
is phase locked and responds as a rigid unit. A common analog in chemistry would be a benzene molecule in which the
electrons in this benzene ring respond as a rigid unit to small
stimuli. In a superconducting Pb wire 1 m long, the electrons
at one end are phase locked to the electrons at the other end.
Pairs, of course, can be disrupted by many processes: by
thermal (kBT) excitations, by electromagnetic absorption, or
by electron injection. In all of these cases the disrupted electrons or excitations behave just like normal-state electrons.
The ground-state electrons can be thought of as a superfluid
with density ns, with a condensation energy or pairing energy
per pair. The excitations out of the ground state can be
thought of as a normal fluid with density nn. In this framework the sum of these probabilities must add to one, ns
nn 1. This picture is quite analogous to the two-fluid model
originally proposed by London (14). As the temperature rises,
the incoherent phonon scattering becomes larger and overwhelms the coherent phonon exchange and the material reverts to the normal state.
Zero Electrical Resistance

Electrical resistance in a material such as Pb disappears below the superconducting transition temperature because a
significant fraction of the electrons transform into a coherent,
phase-locked, superfluid ground state having considerable rigidity in the wave function. In a normal metal, such as Pb
above Tc, the electrons are relatively uncorrelated and behave
as individual particles. Electrical resistance arises because individual electrons scatter from impurity atoms, phonons, or
some other excitation in the metal as random or thermal
events. In a superconductor, electrons are locked together in
a giant rigid wave function and individual scattering is essentially eliminated. The reason that the transition is so sharp
is that there is such overkill in the pair-pair correlation. With
millions of pairs overlapping the region of any one pair, one
can get substantial pair-pair correlation even if only 0.1% of
the electrons are in the superfluid state. This can occur within
a temperature interval of a millikelvin.
Meissner Flux Exclusion in Type I Materials
In thinking about the ability of a type I superconductor to
push magnetic flux out of the interior to give B 0, it is
important to remember that the free-energy difference per
atom to push the flux out is very small. The value of Gn Gs
given by Eq. (2) can be evaluated to be about 108 eV per
atom. This is tiny compared with most electronic processes,
which are about 1 eV. Hence, it does not take much energy to
exclude the flux.
There are several competing energies in this problem. If a
magnetic field penetrates the interior of a sample where the
paired superfluid density is high, then there is a tendency for
the spins to align with the field and thus break pairs. This
would raise the free energy. This rise in free energy would be
prevented if circulating currents are created at the surface of
the sample to cancel the magnetic field to give B 0. There
is, of course, a kinetic energy cost in creating these circulating
supercurrents. For small and H Hc1, it is energetically
more favorable for the material to develop a supercurrent at
the surface and provide a B 0 condition in the interior than
it is to lose the condensation energy associated with breaking
pairs as the spins align with the applied magnetic field. If,
however, is greater than 1/ 2 and H Hc1, it is energetically favorable for quantized vortices to enter the material.
The time dependence of the response of the electrons to an
applied magnetic field is governed by the plasma frequency.
With the application of an external field, the superconducting
electrons respond as a coherent plasma, and the screening
length for magnetic fields is given by
2 = m /0 ns e2
(8)
where m* is an effective mass and e is the charge on the electron. For very pure metals, 50 nm and for the HTS materials, 170 nm. Because is governed by the superfluid
density, it does not vary from superconductor to superconductor as strongly as , which is governed by the ratio of
vF /kBTc.
GinzburgLandau Equations
Very early in the development of the theory of superconductivity, Ginzburg and Landau (15) developed a very general
and yet very powerful formulation of the problem. They assumed that there is an order parameter that behaves much
like a wave function, and they further assumed that this wave
function is related to the local density of superconducting
electrons by 2 ns. The free energy is then written as the
sum of a kinetic energy term, a potential energy term, and a
magnetic term:

2

1 ~
h2
eA
+ ||2 + ||4 +
Gs Gn =

2m
i
2
8
(9)
Here, the potential energy term is a power-series expansion

in 2 and h2 /8 is the magnetic energy term. Minimizing this
free energy with respect to the order parameter gives the famous GinzburgLandau (GL) equation:
+ ||2 +
1
2m
~
i
2
eA
=0
(10)
Excellent discussions of solutions of these equations are

given by Fetter and Hohenberg (16), Tinkham (6), and de
Gennes (17).
A central factor determining whether a material is a type
I or type II superconductor is the boundary energy between a
superconducting region and a normal region. If a superconductornormal-metal boundary has a positive surface energy,
then the system will minimize the boundary area and a Meissner solution will be expected. If the superconductornormalmetal boundary has a negative surface energy, then it will be
expected that the material will be unstable against a breakup
into small domains in order to maximize the amount of superconductornormal-metal boundary area. In the original paper
(15), it was shown by numerical calculations that for Eq. (10),
the onset of negative surface energy occurs when the ratio
of two characteristic lengths, the penetration depth and the
coherence distance, is equal to 0.707 or
1/ 2
115
If there are no surface barriers to flux entry, vortices will nucleate and move into the interior above Hc1. As the applied
field increases, the vortices crowd closer together. At high
magnetic field, when the cores of the vortices begin to overlap,
the sample goes normal. This occurs at
Hc2 =
0
= 2Hc
2
2
(14)
Combining Eq. (13) with Eq. (14) gives a convenient rule of

thumb that Hc1Hc2 Hc2 ln .
Large- Case: Repulsive Interaction Between Vortices
For the case of 1/ 2, the GinsburgLandau equations
have relatively simple solutions and there are rather good
physical pictures to describe the behavior of vortices. A reasonably good model to describe the vortex is a normal core of
radius with circulating currents sufficient to give one quantum, 0, of flux. The superconducting order parameter is
zero in the center of the core and rises toward the bulk value
with a characteristic length . To estimate the free energy per
unit length of the vortex, gv, the free energy per unit volume,
0Hc2 /2, can be multiplied by the area of the core, 2. A more
accurate calculation (6) gives
gv (0 H 2c /2)(4 2 ) ln
(15)
The magnetic field around the vortex is given by (6)

h(r) =
0
K (r/)
22 0
(16)
where K0 is the zeroth-order Hankel function of imaginary

argument. The force on a quantized vortex in the presence of
a current density J is (6)
f p = J0
(17)
(11)
where a0 is the lattice spacing of the triangular lattice. More

detailed calculation shows that vortices are first stable in a
superconductor at
where J can be either the current density caused by the circulating current of other vortices or a transport current density
that is externally applied. In this large- regime, the force
between vortices is repulsive and the Abrikosov theory (18)
shows that the vortices will arrange themselves in a triangular array. As the field rises above Hc1, the vortices flow into
the interior under the influence of the magnetic pressure of
the applied field.
The high-Tc cuprate superconductors normally have a coherence distance in the a-b plane on the order of ab 2 nm
and a penetration depth on the order of 200 nm so the
values are in the range of 100. In this extreme type II
limit, the size of the Cooper pairs is very small compared to
the vortex size and several simplifying assumptions can be
made in the development of models to describe the magnetization curves.
The Hao and Clem model (19) for the magnetization curves
of high-Tc superconductors is typical of approaches that can
be used. They developed a variational method in which the
trial wave function is written as
Hc
Hc1 = ln
2
2
+ 02
This criterion then specifies whether a superconductor is type

I or type II.
Vortex State
Building on the GL result, Abrikosov (18) predicted in 1957
that flux entering a type II superconductor is in the form of
quantized vortices. A quantized vortex with 0 of flux is the
minimum unit of flux that can enter. He showed that if
1/ 2, then a regular array of flux tubes or vortices would
form in the interior of the material. The stable state is a triangular array so the area of a flux tube would be
3 2

a = 0
2 0
B
(12)
(13)
(18)
116
1
Superfluid density
Y (124)
0.8
2
0.6
0.4
0.2
0
0.2
0.4
0.6
h(H/Hc2)
0.8
Figure 9. Superfluid density as a function of magnetic field for

Y(123). These data show that the superfluid density falls linearly
with magnetic field and approaches zero as H goes to Hc2.
where is the distance from the core of the vortex, and both
v and are adjustable variables in the trial wave function.
They start with the free energy including the core energy and
minimize with respect to the trial variables of v and . The
magnetization curves are found to scale to a universal function on a 0M/ 2Hc vs H/ 2Hc plot. Both Y1Ba2Cu3O7 and
Y1Ba2Cu4O8 are found to obey the Hao-Clem model very well
(19). In addition, it was found that the superfluid density averaged over one vortex unit cell falls linearly as H/Hc2 for
fields greater than 0.3Hc2 over the entire range where there
is thermodynamic reversibility and measurements can be
made (20). Figure 9 shows that ns 2 is linear in H/Hc2 for
this high-Tc Y(123) material.
Surface Superconductivity
The surface of the superconductor modifies the potential for
the superfluid electrons, and superconductivity will persist to
fields above Hc2 in a narrow layer near the surface. With the
applied field parallel to the flat surface (17), Ginzburg-Landau theory with plane surface boundary conditions predicts
that a superconducting layer about a coherence distance thick
will be present up to Hc3 1.7Hc2. Experiment verifies that
this basic idea is correct. For the case of pure Nb, the measured ratio of Hc3 /Hc2 for pure Nb varies from 1.78 at low temperatures to 1.70 at T/Tc 0.9 (21). At higher temperatures,
Hc3 /Hc2 approaches 1.0 at Tc. Altering of the surface condition,
say, by a normal metal coating will suppress or destroy this
surface state.
Small- Case: Attractive Interaction Between Vortices
For the case in which is comparable to , there can be overlap of the vortex cores and there can be an attractive interaction between vortices. This effect was established experimentally by Essmann and Trauble (22) with experiments in which
they decorated the surface of Nb with Fe spheres about 4 nm
in diameter. To perform these experiments, typically an array
of vortices is trapped in a coin-shaped Nb sample by applying
a magnetic field above Hc1 and then turning the field off. A
smoke of Fe was then created by evaporating Fe metal in
an atmosphere of a few Torr of He gas. The tiny Fe particles
that are created follow the flux lines to the point on the surface where the core of the vortex emerges. Once the Fe particles touch the Nb, they stick very strongly. To image the vortex lattice, the Fe is stripped off the surface by a graphite
replication technique and viewed in a transmission electron
microscope. For the initial flux entry in the first-order transition of Hc1, illustrated by the dashed curve of Fig. 6, it is found
that there is a two-phase region. There are clusters of a few
hundred vortices all on a triangular lattice and separated by
about 200 nm. Between these clusters there are Meissner-like
or vortex-free regions. In this two-phase region, the spacing
of the vortices is independent of magnetic field. As the magnetic field increases, the sample fills with vortices having 200
nm spacing. The magnetization curve in this region is linear
with a slope governed by the demagnetizing factor. An abrupt
change in the slope of the magnetization curve occurs when
the sample is just filled with vortices at the lattice spacing
governed by the attractive interaction between vortices. This
characteristic field is denoted by B0. At higher magnetic fields,
the vortices are pushed closer together and the magnetization
curve is similar to the repulsive force case for high- materials. Auer and Ullmaier (7) performed a very systematic study
of these same effects, as in Ta is systematically increased
from type I to type II behavior by alloying with N.
To analyze quantitatively the transition from type I to type
II behavior by alloying, it is useful to focus on the connection
between the coherence distance and the normal-state electronic mean free path l. With small additions of impurity, is
given by
1/ 1/0 + 1/
(19)
where 0 is the intrinsic or clean limit of . At higher impurity

concentrations, l becomes comparable to 0 and the diffusion
limit is more appropriate. In this regime
= 0
(20)
Experiments to Determine Hc1 and Hc2
Many factors can lead to errors in determining both Hc1 and
Hc2. The lowest field for which a vortex is thermodynamically
stable in a superconductor is defined to be Hc1. It is a difficult
quantity to measure because thermodynamic equilibrium is
not achieved easily at fields close to Hc1. The most common
error arises from the presence of surface barriers to flux entry
into the sample. For a cylindrical superconductor with a radius of curvature much larger than both and (17, p. 79),
the applied field must be larger than Hc1 before a vortex will
nucleate. As the field increases from zero, Meissner screening
currents flow in the superconductor within a distance of the
surface. When a vortex starts to nucleate at the surface and
move into the interior of the sample, an image vortex develops
to pull the vortex back toward the surface. The competition
between the image force pulling the vortex toward the surface
and the Meissner currents pushing the vortex into the interior creates a surface barrier to flux entry. Often the fluxentry field is comparable to Hc. To overcome this effect, the
surface needs to be rough on the scale of the penetration
depth or the surface needs to be coated in some way to suppress the surface barrier to zero.
The highest field for which a vortex is thermodynamically
stable in a superconductor is defined to be Hc2. For a sample
that obeys the Ginzburg-Landau theory, there is a sharp
change in slope at the second-order phase transition of a reversible magnetization curve that identifies Hc2. This is the
most reliable measurement of Hc2. Measurements of the elec-
SURFACE ACOUSTIC WAVE APPLICATIONS
trical resistivity is less reliable because it depends on the motion or depinning of vortices and often is not a measure of
the point of thermodynamic stability of a vortex. For a clean
classical superconductor such as Nb, the electrical resistance
goes to zero at very nearly the same field that the magnetization goes linearly to zero, and both resistivity and magnetization methods can be used to determine Hc2. For high-Tc materials, the situation is more complicated. Because is so small,
fluctuation effects smear out the normal-metal to superconductor transition so that is is typically 3 K wide. Both the
magnetization curves and the electrical resistivity are greatly
rounded even for a very perfect sample. For this case, it is
better to move to lower fields at which fluctuations are negligible and use a fit of the M vs H data to the HaoClem theory
to determine Hc2.
BIBLIOGRAPHY
1. D. L. Decker, D. E. Mapother, and R. W. Shaw, Critical field measurements on superconducting lead isotopes, Phys. Rev., 112:
18881897, 1958.
2. G. Bednorz and K. A. Muller, Possible high Tc superconductivity
in the Ba-La-Cu system, Z. Phys. B, 64: 189197, 1986.
3. D. K. Finnemore and D. E. Mapother, Superconducting properties of tin, indium, and mercury below 1 K, Phys. Rev., 140:
A507A518, 1965.
4. R. B. Goldfarb, Demagnetizing factors. In J. Evetts (ed.), Concise
Encyclopedia of Magnetic and Superconducting Materials, Oxford:
Pergamon, 1992.
5. D. Shoenberg, Superconductivity, Cambridge: Cambridge University Press, 1960.
6. M. Tinkham, Introduction to Superconductivity, 2nd ed., New
York: McGraw-Hill, 1996.
7. J. Auer and H. Ullmaier, Magnetic behavior of type II superconductors with small Ginsburg-Landau parameter, Phys. Rev., 7:
136145, 1973.
8. D. K. Finnemore, T. F. Stromberg, and C. A. Swenson, Superconducting properties of high purity Nb, Phys. Rev., 149: 231243,
1966.
9. J. J. Wollan et al., Phase transition at Hc1 for superconducting
Nb and V, Phys. Rev., 10: 18741880, 1974.
10. U. Kumpf, Magnetization curves for type II superconductors with
small Ginsburg-Landau parameter, Phys. Status Solidi, 44: 829
843, 1971.
11. R. R. Hake, Mixed state paramagnetism in high field type II superconductors, Phys. Rev. Lett., 15: 865868, 1965.
12. J. Bardeen, L. N. Cooper, and J. R. Schrieffler, Theory of superconductivity, Phys. Rev., 108: 11751204, 1957.
13. L. N. Cooper, Bound electron pairs in a degenerate Fermi gas,
Phys. Rev., 104: 11891190, 1956.
14. F. London, Superfluids, New York: Wiley, 1950, Vol. 1.
15. V. L. Ginzburg and L. D. Landau, Regarding a theory of superconductivity, Zh. Expt. Teor. Fiz., 20: 10641072, 1950; 20: 1064
1077, 1950.
16. A. L. Fetter and P. C. Hohenberg, Theory of type II superconductors. In R. D. Parks (ed.), Superconductivity, Marcel Dekker: New
York, 1969.
17. P. G. de Gennes, Superconductivity of Metals and Alloys, New
York: W. A. Benjamin, 1966.
18. A. A. Abrikosov, Magnetic properties of superconductors, Zh.
Expt. Teor. Fiz., 32: 14421450, 1957 (Sov. Phys. JETP, 5: 1174
1182, 1957).
117
19. Z. Hao and J. R. Clem, Limitations of the London model for the
reversible magnetization of type II superconductors, Phys. Rev.
Lett., 67: 23712373, 1991.
20. J. Sok et al., Reversible magnetization, critical fields, and vortex
structure in grain aligned YBa2Cu4O8, Phys. Rev., 51: 6035
6040, 1992.
21. J. E. Ostenson, J. R. Hopkins, and D. K. Finnemore, Surface superconductivity in Nb and V, Physica, 55: 502506, 1971.
22. U. Essmann and H. Trauble, Direct observation of individual flux
lines in type II superconductors, Phys. Lett., A24: 526527, 1967.
D. K. FINNEMORE
Iowa State University
SUPERLATTICES, MAGNETIC. See MAGNETIC EPITAXIAL

LAYERS.
SUPPLY AND DEMAND SIDE MANAGEMENT. See

LOAD MANAGEMENT.

03 Applied Superconductivity

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

03 Applied Superconductivity

Загружено:

Авторское право:

Доступные форматы

BIOLOGICAL EFFECTS OF ELECTROMAGNETIC

FIELDS WITH EMPHASIS ON HEALTH AND

External eld strength, measured in volts/m (electric

Internal eld, Ei (i.e., the electric eld induced within

Induced current density in A/m2 . The current density

where is the tissue density (kg/m3 ).

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

(i.e., the induced electric eld is proportional to the time

magnetically induced electric eld E is given by

where f is the frequency (Hz).

where o is the permittivity of space (8.85 1012 F/m) (all

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

Effects (50/60 Hz)

stimulation of motor cortex in brain and

Adapted from References 15 and 16.

Approximate energy penetration depth, m

Threshold for perception (10 s exposure, 0.024 m2 exposure area),W/m2

Table 4. Exposure Standards for 60-Hz Electric and Magnetic Fields

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

Figure 2. Whole-body average SAR in a prolate spheroidal model

4. A buildup of charge exists on the surface of the body, as

where f is the frequency in Hz and

is the loss tangent of the material. The SAR as a function

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

At frequencies above about 10 GHz, the energy penetration

Thermal Effects. Thermal effects are produced by Joule

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

currents can produce serious burns, which might be more

magnetic eld-permanent dipole or magnetic eld-induced

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

Low Frequency Fields

Bone Repair Stimulated by Pulsed Magnetic Fields

Cardiac Excitation and Fibrillation

Naturally occurring (endogenous) electric elds exist in the

Nerve Stimulation by Pulsed Magnetic Fields

Perception and Pain

The thresholds for perception of microwave energy have

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

Other Thermal Effects

Cataract is an established hazard of microwave enBurns

ergy, which has been reported occasionally in humans

Consequences of Long-Term Exposures

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

Powerline Fields. Most epidemiology studies related to

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

tal electrocution claims the lives of about 1000 individuals

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

Biological Effects of Electromagnetic Fields with Emphasis on Health and Safety

59. Kheifets, L.; Shimkhada, R. Childhood Leukemia and EMF:

CORPORATE MODELING WITH DSS AND ERP

Model Management Module

DECISION SUPPORT SYSTEMS (DSS)

vantage of using a DSS is that it is capable of integrating

Corporate Modeling with Dss and Erp Systems

Figure 2. Framework of MRP II.

duction of ERP systems in the early 1990s stemmed from

available. ERP systems are not purely software systems;

Corporate Modeling with Dss and Erp Systems

Corporate Modeling with Dss and Erp Systems

Supply Chain M anagement

Figure 3. Applications of the ERP system (source: Reference 18).

demand network design (20). The presented solution is

Corporate Modeling with Dss and Erp Systems

tic models, discriminate analysis, and data envelopment