IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-9, NO.

1, FEBRUARY 1974

Characterization of Surface Channel CCD l[mage Arrays at Low Light Levels

MARV1 N H, WHITE,
SENIOR MEMBER, IEEE,

DONALD INGHAM

R, LAMPE, A, MACK,

MEMBER,

IEEE,

FRANIKLYN

C. BLAHA,

MEMBER,

IEEE,

AND

MEMBER,

IEEE

AbsfracfThe characterization of surface channel charge-coupled device (CCD) lime imagers with front-surface imaging, interline transfer, and 2-phase stepped oxide, silicon-gate CCD registers is presented in this paper. The analysis, design, and evaluation of 1 X 64 CCD line arrays are described in terms of their performance at low light levels. The signal-to-noise (S/N) is formulated in terms of charge at the collection diode. A dynamic range of 80 dB and a noise equivalent signal (NES), where S/N = 1, of 135 electrons is achieved with a picture element time of 20 @ and an integration time of 1.32 ms in the absence of a fat zero. A unique CMOS readout
circuit, element which time uses correlated double sampling within a picture

electrometer long distances

amplifier. signal within

Thus, the

with

the CCD and

principle

photon-generated

charge may be transported silicon amplified

over at low

input noise charge levels, Although the clock and video signal levels are noninteracting within the CCD imager, there is an interaction at the collection diode. We have developed a method of signal processing called correlated dwble sampling [5], [6] to remove the switching transients, l/f eliminate surface-state the Nyquist capacitance noise associated With with this the techsigreset switch/node combination, and suppress

removes the Nyquist noise of the reset switch, eltimates switching transients, and suppresses lowfrequency noise to provide low-noise analog signal processing of the video signals. Thk paper describes the responsivit y, resolution, spectral, and noise measurements on silicon-gate CCD sensors and CCD inter~me shift-registers. The influence of transfer inefficiency and electrical fat-zero insertion on resolution and noise is described at low light levels.
window,

noise contributions, the intrinsic current. within the whereas, from shift

nique we have realized nal (NES) shot the the analog video signal

noise equivalent In the CCD the array in non-CCD switch [8].

of the CCD imager which is set by the thermal imager by an arrays with Thus, from dual rea in signal shift is processed register,

noise of the leakage CCD

is transferred by the closure a digital we require

sensor to the video [7],

I. 1NTRODUCTION HARGE-COUPLED invention

c

preamplifier (CCDS), analog and high since their [2], lines with delay investigation a CCD both quirement

of an address register analog on yield register,

devices low dispersive, propagation delay such

gate pulse from imager

[1] signal

and experimental

characteristics and this

have promised low-noise

the sensor and the shift places limitations within II. CCD

with

dynamic

due to the presence

range. The CCD a monolithic silicon-gate monolithic CCD

analog

line may be coupled

of nonuniformities

the array.
LINE ARRAYS

photosensor, or diffused preamplifier

as a {semitransparent and a low-noise level imagGeneral Fig. low light

photodiode,

to provide

Considerations 1 illustrates the line transfer array functional requires block dia-

ing. The transfer inefficiency e, in practical surface-channel CCD structures, is determined by surface-state trapping [3] which limits dispersion the free-charge [4] transfer process line. (N 10SS and introduces The CCD = number in image into the analog delay

gram, The interline pulse I#)T,at the start photocharge responding

approach

a transfer the stored to a cor-

of a line time, to transfer sensor locations CCD transfers

analog delay resolution

line should or modulation

have a NPE <0.1 transfer

from the individual

of bits, P = number

of phases) to prevent

bit in the parallel-to-serial clock system

shift register. the charge

function carriers to

In our case, a 2-phase

(MTF) degradation. The CCD imager uses the flow of minority transfer diode. CCD in the the video signal to a low capacitance The collection diode is typically imager, whereas, the readout Z-V addressed image arrays

along the. shift register to a CMOS readout circuit. The shift register is a 2-phase stepped-oxide geometry with a surface field and potential profile as shown in Fig 2. Fig. 3 is an enlargement of the CCD line array after the definition of the shift register with an aluminum interconnection, Thd sensors are constructed polycrystalline silicon-gate {100) diffusion (n-type, diffusion elements and inhibits with transparent, with on between conan n+ three adThe ductive, stopper jacent electrodes substrates)

collection

0.25 pF in the line capacitance is at least an or-

der of magnitude larger in value. The low capacitance provides a reduction in noise and an increase in voltage swing at the gate electrode of an on-chip MOS

sides. This

reduces the interaction blooming

in the array.

Manuscript The authors Westinghouse

received May 8, 1973; revised July are with the Advanced Technolo ti. Electric Corporation, Baltimore,

23,

1973.

H&tirles

sensor bias voltage is adjusted such that the maximum collected charge at any sensor location cannot overflow the storage well of the shift register. The fourth side of

IEEE JOURNAL 0!3 SOLID-STATE CIRCUITS, FEBRUARY 1974

SENSOR BIAS

CCD SILICON

GATE

SENSOR

ARRAY

@JM MUX TRANSFER PULSE f#JT /////// +1 q {= PARALLEL CCD SHIFT TRANSFER GATE ///// TO SERIAL REGISTER READOUT

PULSE

2 PHASE SIGATE

@JR RESET PULSE

OUTPUT

Fig.

1.

CCD

line

array

functional

block

diagram

illustrating

interline

transfer.

0 I

246 1

1
I 0-\ 1 \,

8 I

10 f rn) *

12 I

14 (

16 1

0 D_
18 20
I

22
I

26 I

1
-2 t -4

D!stmce(.

8 -6 H 0

Zr- -- -4------,-!L E5 (M(n) = 84 Vlcm

q
\ \

\ \

-8 IE I I

-lo t : : u -12 I 1 -14 -16 : I t I -16 I -,;

ii 1 -----}-_ --T
ND = lX1015

I +, (+

I L I I

l-------
Cm-3

I ~ t \

.%s . (jX ,olo,.m-2 a

-22 -24 #

##

t---

+ . .. ~
I

Iil
I

Fig,

2.

Two-phase stepped oxide CCD unit Longitudinal electric field relative

cell with to electrode

curves of periodic surface potential cross section at top of figure.

and

the

CCD

sensor is stoppered potential with

by holding

the transfer

fier. The voltage ter is also

waveform

on the gate of the electrome4. Below the waveform

gate at a positive

respect to the substrate

dkplayed

in Fig.

during the integration period to form an n+ accumulation layer. The sensors are defined by and aluminum light shield which also serves to cover the CCIl shift-register and CMOS readout circuits. The CCD line array is constructed of 128 CCD sensors with a 2P offset in the along(P = 15 ~m, Az = 22 ~m, Ay = 18 ~m). track direction The dimension Ax is in the across-track rection and Ay is in the along-track direction. Fig. 4 illustrates sists of a multiplex biased collection the CMOS gate readout circuit which con(i.e., (i.e., muxgate MOS +J1), a reverseamplielectron-scan dimechanical-scan

four distinct timing intervals are labeled for discussion. These four timing intervals comprise a pizel (i.e., picture element) time method called and form correlated the basis of a signal double sampling. processing The node ca-

pacitance at the collecting diode is 0.25 pF and is not influenced by the parasitic n+/p- diode of the reset switch which is reverse-biased to prevent discharge of the collecting node by the reset feedthrough pedestal when the reset switch provides rests over ground cuit from about is turned off. The aluminum light shield since it a cir0.03 pF of the node capacitance output

1.3 to 1.5 pm of deposited pickup.

Si02 and forms collection

diode, n-channel

MOS reset switch electrometer

plane to shield the sensitive

reset gate 4R), and p-channel

wmTE

et al.:

ccD

IMAGE ARR.4ys

Fig.

3.

(ICD

interline-transfer

line

array

with

CMOS

readout

circuit

(prior

aluminum

light

shield).

MUX swl-rc +M (=1 4!.4. % RESET SWITCH 1 -VR L< c H

= COLLECTING DIOOE

OUTPUT

i g m=

VDT

-v ELECTROMETER AMPLIFIER

:1

.l

I
DIODE

RESET

READ RESET

MUX SIGNAL

REAO SIGNAL

44
COLLECTING

cm~<

II

E~
Fig. 4. On-chip correlated doublesampling (CDS) readout separate sequential circuit steps. with

MUX ON

MUX OFF

gate

voltage

waveform

comprised

of

III.

CORRELATED DOUBLE SAMPLING ANALOG SIGNAL PROCESSOR

to be detected the random output

and amplified.

In addition

to this problem, give rise to

rise and fall

times

of the clocks

In the use of a CCD is charge vantageous construction. Iimitation action integration, to provide

sensor array and sampling on-chip signal

the natural techniques processing

frequency components which fall in the passband of the video preamplifier, The minimum detectable signal or noise equivalent set by the system signal (NES) after preamplification was noise in these image arrays. It seemed

are adand rea major charge

In the introduction to Z-Y addressed pulses image with

we mentioned arrays the video signal

was the inter-

as though all of the effort was placed on sensor development with little emphasis on the problem of signal detection and, video processing. With the advent of CCD

of the clock

IEEE JOURNAL OF SOLID-STATE CIRCUITS, FEBRUARY 1974 ~ +V R, w F

MUX

-1=

SWITCH

COLLECTING DIODE ( /

R2 NOOE N COMPOSITE VIDEO

w= cd RESET SWITCH -4 #R + -vDT VG *

in

1/

_ + BUFFER AMPLIFIER I+

SAMPLE

C2

@ ~c PREAMPLIFIER = CLAMP

-v ~

I
v

I
of a CDS processor

I--J-i=

-Vc
I
GAIN=G

ELECTROMETER AMPLIFIER

CMOS READOUT CIRCUIT

Fig. 5.

Schematic

diagram

with

critical

capacitances,

noise

sources,

and signal

nodes.

imagers, until

the integrity detection

of the signal occurs at of this point

charge is maintained a common cannot readout collection be emphasized and clock intercircuit. All of collec-

signal

diode, The importance too strongly action the pixel is localized information

since the problem

of signal

at a common

passes through

a common

tion diode, integrating capacitance, and electrometer amplifier. This is true for the line and area CCD imagers. We also mentioned in the introduction the low output capacitance (C = 0.25 pF) at the collection diode. For a well designed 1O-MHZ in, which video amplifier, may as used with a vidicon, amplifier lection the shunt capacitance is shown Thus, be 20-30 pF, For a to the preC/g~ of the at

.
I 1
0
I

I
1

noise current diode

at the input the small

in Fig. 5, the equivalent is C iJg~. increases

noise charge at the colcharge ratio

~
--i
0 ~
The in timing Fig. 6. is in 4) the of diagram At the

1
____ ____ _____ ____ ----__

- -_

electrometer

the signal-to-noise

the collection diode for such noise currents as shot noise, preamplifier noise, surface-state noise, and supply noise. Let us examine the four distinct timing ployed in the readout circuit of Fig. 4.
1)

intervals reset switch

emis C is

Reset:

The

n-channel voltage

MOSFET V~ with

turned

on and the voltage

V~ across the capacitor

Fig. 6. Timing diagram for CCD signal processor. The four steps correspond to explanation in the

line array with CDS analog are separated in sequence to text.

reset to the reference

a noise uncertainty
for start to the the of last bit signal the processing reset (i.e., to interval is illustrated the pixel +2 and ~~

V., This noise voltage may be introduced through inadequate filtering of the reference supply voltage and the Nyquist noise contribution of the reset switch, [9], [10] where the latter is given by V. = (iiT/C) ~ or in terms of noise charge Q. = (k!fC) ~. The full Nyquist voltage appears across C when the electrical time constant formed by the series resistance of the reset switch and integration capacitance C is much less than the time the reset switch is on. The p-channel electrometer is connected to an operational fier in the CCD signal amplifier processor which circuit is the preamplishown in Fig. 5.

charge Fig. and

transit the

well adj scent

see phase the

electrical diode.

muxgate

collection

.%)

turned

Read Reset: After the n-channel reset switch is off, the voltage present on the gate of the electrom-

eter consists of a feedthrough pedestal AVR and a noise voltage V.. With the reset switch off, the gate voltage is holding on a high impedance point with a time constant switch 1 of seconds. In the read reset interval, the clamp

WHITE et al.:

ccD

IMAGE ARRAYS

is turned on and Cl is charged to a voltage the voltage on the gate of the electrometer. turned clamped off and one side of the capacitor, or dc restored to a reference

indicative of Switch 1 is node N, is V,, while voltage in the

Fig.

voltage

L\\
IT(UI)12
TIT

1 -COS u,

(-+

CORRELATE

OOUBLESAMPLING

PREAMPLIFIER ROLL-OFF

the other side of the capacitor ous sample of the gate voltage. across the clamp ferential the on-chip gate switch from or incremental

represents this moment caused diode

the instantaneon is the difa change With

The instantaneous by

charge

or collection

volta~e.

2Tr/T

37r/T

47riT

57r/T

61r/T

7r IT

6T/T

clamp switch turned off, the measured reset level is holding on the high impedance node N formed by the clamp capacitor amplifier. 3) Muz the high) to the (minority moment, transferred pixel and Signal: charge At the start is raised diode. to The of the mux signal interval storage the well pixel of pixel (+2 goes charge charge transfer collection Cl, and the noninverting input of the buffer

7.

Filter

characteristic
T =

of

CDS T/2.

analog

processor

with

since it is correlated a removal

within

a pixel

time,

and this means

in the

of a noise charge

+,,1 goes low carriers)

collection

(1) for a 0.25 pF capacitor. also suppressed signal processor The transfer components intervals, of may which function, the be written The l/f surface-state noise is

discharges

the voltage

V@ as shown charge pedes-

in the signal there

waveform

of Fig. 4. If charge,

we assume, for the

is no pixel

then the only

to the gate electrode

is the feedthrough

by the filter

characteristic

of the analog

tal AVfl, = V,. Cm,/C, where V., is the mux voltage swing and C., the feedthrough capacitance from the muxgate to the collection when diode. The charge is removed, is turned off as shown is introduced, collection +,. may however, 4. A may which diode. the muxgate for in Fig.

is shown as follows. which as

(1 e-)
~,uo

acts on any time-varying clamp and sample

signal

between

&Tyquist noise of Qa2 = 2KTC~ be minimized muxgate natively, does not overlap with the

C,. <: 0.01 pF for the case where the Alterat muxgate be held

m) =
the end of the clamp

To ~ +

(2)

an overlapping diode

where To is the signal gain and r is the delay time between pulse ancl the end of the sampreple pulse, amplifier. The zeros The press and ~,. is the bandwidth Fig. 7 illustrates of , = T/2, features zero jitter, where 2Nr/ T noise clock generation Thus, of the front-end T filter (N (O =
is

a fixed d-c potential to the collection

the clock ~s transferring

charge cur-

(see Fig. 4). In t$e absence of any

optical pixel charge we would collect the leakage rent from the sensor and the shift register wells. .4) Read the running Signal: output

a plot of the filter of this

characterisdouble

tic for a value important of l/j double

the clock period. are the = O, 1, 2, ). from power

After the mux signal is turned off, voltage on node N is the time differclamped increment read reset Thus, reset level and the introduced (clamp) by the leakage and read

IT(OJ12 at u = and pulse low-frequency

ence between

the previously

at the origin

O) serves to supzero at even also supprocessing the

hame reset level plus signal closure of the mux switch of the reset level signal includes related which (sample) Nyquist within between intervaJs).

arising double

(i.e., there is negligible

suppliesj harmonics MOS

etc., and the noise

of the fundamental amplifier.

frequency the signal

the reset noise, which The signal increment, register leakage current of the sample and passed to the outis a sequence of pixel

presses surface-state electrometer

in the p-channel of the sensor

noist and V~ power swpply noisej is cor-

a pixel time window. is amplified video stream

does not degrade but enhances the qualities element by removing the Nyquist l/f noise while amplifying the these features, we have automatic to increase the dynamic clamped by the reference filter out the clock

consists of sensor and shift processor

noise and filtering

added to photocharge, put of the signal switch 2. The output

by the closure

signal. In addition to dark level subtraction

element responses free from reset noise and proportional to the minority carrier signal increment introduced by closure of the mux switch. The correlated double sampling moves switching transients similar que [11 ] which storing quent subtraction (clamping) the actual from (CDS ) method reto an earlier techniin lieu of incre-

range since the video signal is voltage V,. There is no need to and higher order har-

fundamental

monics; and the video output is already in a format for image display or further data processing (see Fig. 8). IV. Analytical
RESPONSIVITY AND NoISE CoivsrzmrAT1oNs

used a gated charge integrator

diode reset level for subse-

the reset level plus signal

Formulation of what we mean by

ment to give the signal increment without reset noise. The Nyquist noise of the reset switch has been removed

We can begin by an examination responsivity. We will

define this quantity

to be R in pA/

6
+VB

IEEE JOURNAL OF SOLID-STATE CIRCUITS, FEBRUARY 1974

Mux Switch % ~ Y Collecting Diode RB

Video
1

c@ +RA
Reset Switch + VDT

AID Converter

Statistical Computer Analysis

c T~
Electrometer

rT
+C(Clamp) +S(%mple)

-vR

Fig.

8.

Test

apparatus

for

measurement

of

responsivity

and

noise

with

typical

output

waveforms,

mW,/m2

or current

output

per input

irradiance,l

which to

amplifier processor in output AE. factor signal The

feedback following voltage responsivity

resistance,

G the gain

of the signal density by a the

makes the responsivity we attempt an ideal 6000 K). to formulate blackbody

a ffinction

of sensor area. Suppose by reference T, (e.g., TS =

the preamplifier,

and AV the change exposure

the responsivity

for a change in input

source of temperature

of each sensor is degraded of transfers collection with it reaches the

Thus, we can write RLH,(X) dk = individual H,(k) dh sensor responsivity (3)

(1 E)~ where 2A~ is the number undergoes current before output, the integral irradiance in the numerator is obtained

i,
R=

diode.

In practice, represents supplied

of (3), which a test source The irradiance over the entire

,X / ~k.

whose spectral

profile is known. ,. by the test source, whale effective

and R, = spectral where H, is the specified reference nm), quantum temperature spectral area efficiency. band of the A the responsivity = he/A

spectral response of the sensor (i.e., 200 nm to 1200 nm for silicon), is converted to an effective irradiance from a

m(NA
& and Al the = 800 the effective

6000

blackbody

source

in

the

400 nm

to

800 nm

window. The measurement A/D in the variance (4) converter, array In practice, procedure provides output involves from of A/D the use of a 10-b of 1024 b. which an accuracy

source irradiance, (e.g., Al = sensor, of and ~(A) (3) interval

400 nm, &

the signal

each sensor element at each irradiance bits. The mean and The corresponds

Integration

for the specified yields

is sampled

1024 times in A/D

source and wavelength A

level and recorded mean represents to the uncertainty ples. With these

in terms

R = 0.186v

[ 18 Pm X 22 pm
experimental

are calculated

bits for each signal.

PA mW/m3

the signal while the variance

where ~ has been assumed to be constant length interval. The

over the wavemay be

or noise as determined over 1024 samquantities we can determine linearity, cirse-

responsivity

referred to the gate capacitance C of the electrometer in Fig. 8 and expressed in terms of quantities in Fig. 5, R(1 e)~ = CA V g. R,GAE AQ = AE at collection diode (5)

streaking, responsivity, noise, etc. The measurement cuit is illustrated in Fig. 8 which shows the timing quence (discussed in Section III) at the

preamplifier

output and video output signal. The noise is converted from rms A/D bits to an equivalent input exposure dencalled the noise equivalent sity (microj oules/meter2), signal (NES), by multiplying the rms A/D bits by the reciprocal slope of the transfer curve at the particular irradiance level (exposure density). The transfer curve is essentially a plot of signal A/D bits versus input exposure density. The reciprocal slope of the transfer is called the quantizing interval and is given as curve

= responsivity

measured

where C is the node capacitance at the gate of the electrometer, g~ the electrometer transconductance, Rr the pre-

1R

may

easily

be

converted

to

electrons/micro

jouleJmeterZ.

WHITE

et

at. : CCD IMACE ARRAYS

7
TABLE I RIMDOUT CIRCUIT AND
PROCESSOR

(6) Thus, if we have Z3,~, bits variance becomes (~J/m2). by the signal above (7) procedure, power etc. noise at a specified irradi-

EXPERIMENTAL

PARAMETERS

OF CCD

ANALOG gm = 250 pmho R~=75 iI A=18pm X22pm 2N = 5 to 133

SIGNAL

ance level, then the NES NES The NES, which

G = 2.57 AE 290 pJ/m~ ~ = 1.32 ms Vfi = 12V

AV ;

= 3.56 V ~ cii_$ pF

= Q,B,~,

is measured terms: from analog

consists of 4 principal 1) System, noise supplies, 2) pulse

processor, vibrations, This

removes

the Nyqui.st

noise of the reset switch; include this term ma-y be written as

however,

jitter, by

mechanical

to be general we will of signal-to-noise

in the formulation

Chip noise from is determined

the CCD

sensor array. design

which

geometrical

and fabricain arriva back-

tion processes for the chip. 3) Radiation ing signal ground with shot noise from photons limited performance. noise from NES the uncertainty is given as associated Q1. size of the quantizing interval the fluctuation represents and which

s =
N

signal

charge

noise charge

R(l e)2~E
= (kT(C If + C.,) + = q[~m 1 in + l~s(l (5)-(7), ~)z~]r) then

(9) the chip NES

4) Qucmti.zation the finite The measured

we set the S/N

or total

becomes
NES

c,,,

@T(c

C,t)

!ilLR

~I,s(l

e)~]r)

R(l E)2N (lo)

where the radiation density which can heavy voltage and varies logic filtering the linearly

shot noise term involves quantization with time noise [12]. The

the exposure an error noise Measurements Table The I lists the experimental with a 1
x

assumes

system

measurements CCD calculated

of paramline array. from these

be reduced

painstaking debugging of the examination of grounding, waveforms, on supplies, careful layout

to

through

eters associated experimental parameters

64 element

responsivity

of

the analog

[see (5) ] is R(exp) 0.065 PA = mw,m,-

signal processor, low nc,ise preamplifier, biasing resistor li?~ for a given RF noise contribution. The quantization mined lution by the dynamic or accuracy

and high source reduce Nyquist interval is deterand the resoWe can deterand see the effect at different noise NES for the CCD this CCD relative fringes gate CCD pm) and substitution

()
with

406e~J/mz

(11)

range requirements

of (11) into q(effective)

(4) yields = 0.35 quantum displays electrodes efficiency (12) for

in the measurements,

mine the system noise by measurement, of radiation exposure The in (8). chip We shot :noise through NES can is determined formulate densities.

measurements

sensor, The effective be taken profile response The relative with

sensor must spectral

caution

since the

by the remaining the chip

interference over a lk A of the 12 re(~j =

analytically

due to the 2k A polysilicon sensor compared

at the collection diode in Fig. 4. There noise on the CCID chip: 1) Nyquist itance 2) Thermal shift where 3) Surface noise at the reset switch Qf12 := lcT(J. shot noise associated with time. with

are 3 sources of and output capac-

oxidation. photodiode

spectral a diffused quantum

responses p+/p-/n efficiency

are shown in Fig, 9. The photodiode a constant

sponse illustrates the sensor and 6) ~]T, of between equivalent clear that the transfer vantage

q = 0.65 for an it is ad-

400 nm and 800 nm and the responsivity area size is 0,124 pA/ (mW/m*). a photodiode over its CCD sensor would sensor counterpart. Thus, Both

register

Q.z =

QIILR +

11,s (1

here 7 =

line readout

have a definite

state noise associated

sensors

charge to and from Si-SiOz interface states within the sensor, shift register and electrometer amplifier Q: where
frequency,

are overcoated with approximately 3 pm of SiO~. The response of the CCD sensor array to input irradiance (responsivity) may be determined in another manner through the radiation the total (NES) 2 plotted mz). We notice the straight the CCD sities above 50&J/m2, The shot noise. Fig, 10 illustrates versus exposure density E (@/ line obtained limited quantizing 0.30 pJ/m2 which indicates denfor these and the reat exposure interval . bit,

== kTC, , surface state capacitance

C~t is an effective

determined processor,

by the number of transfers, the clock the effective bandwidth of the signal etc. that the correlated double sampling

becomes background was Q1 =

We have mentioned

measurements

8
06,

IEEE JOURNAL OF sOLID-STATE CIRCUITS, FEBRUARY 1974

(O74)2

09 .

V= 065
m

05 / 04 / DATA CORRESPOND TO NESTFOR MIDPOINTS OF EXPOSURE DENSITY LEVELS.

08

:= A

(0.62)2

$ ~ ~~ x % y
0 % .

07

0.6 ; 05
/

m 03 Zj + g 02 - (O42)2 $2 ~ (

* / A

, 0.45)2 RAOIATIONSHOTt NOISE= ~

\: \ , ;
\ \\ \ \ I

01

y\

H > ij cd .

04

(1) CCD 1=035

1/
o

50

100

150

200

E l#Jln12) EXPOSURE DENSITY

03

1 (3) P+IN X, = 2Pm/

Fig. 10. Effect of equivalent signal 0.30 p.J/mZ ,bit.

radiation ( NES)

shot noise on with a quantizing

the total interval,

noiseQI n

0.2

0,1

val

measured was 0.40 pJ/m2 of 0.30 ~J. This total which is

(160e-) for a quantizing interNES is less than the Nyquist

noise contribution
0. 300 400 500 600 700
A (rim)

800

900

1000

Iloo

NES (Nyquist)
0.70 ~J (2kTC)1/2 = ~ (284e-) R where the factor of 2 is used to illustrate readings of clamp

Fig. 9. Relative spectral responses of (1) a CCD sensor with a 0.2 .um silicon gate over a 0.1 ~m SiO, and of two diffused photodiodes; (2) a p+/p-/n diode 12 ~m deep; and (3) a p+/n diode 2 pm deep.

(14)

the uncorrelated and sample, The is referred for this to the

case of 2 independent sponsivity 10 [see (8)] qAE R =; A(NES)2 obtained is 1.6 X 10-9 X 140 X 10-6 -0.065pA = [(0.72)2 (0.42)2] X 10-2 = mW/m2(13) in agreement with theresponsivity obtained through (5) (1 x CCD by from the slope of the curve in Fig. The system input the CCD ments ments sensor line array particular time dynamic array NES

NES values shown in Fig 11 are for the basic sensor chip. = 0.15 ,PJ/m2 which amplifier. is about 75 dB of the electrometer range with As Fig. 11 illustrates, particular measure=

a chip NES leakage

= 260e7; however, indicate uniform current in Fig.

on lower were

devices under

a chip NES irradiance

135e- and a dynamic performed with the 7 = 1.32 ms with the device IL,,S was current 8 V)

range of 80 dB, All of the measure(line noise For the limitation leakage measured as the shot on the chip.

a: calculated in (11). T%e dynamic range of a CCD 64 element) adray is shown in Fig. had a high quantizing current clamp current, linearly between buildup

11. This interval

associated example, current current leakage V,$ =

Q1 = 0.80 ~J/m2 as measured intervals N-b (see Fig.

11 had a leakage (the sensor leakage with the register sensor bias voltage contribution. (240e-) (15)

bit and a leakage the offset sumed to 4). The leakage

of 7.0 pA/well and sample to a first down the

ILR = 7.0 pA in the register negligible because compared of the low

approximation, CCD

is asshift

and the shot noise NES = q~~r)

register such that the last well in the register carries approximately 2N times the leakage charge as the first well. This leakage current is affected severely by the choice of which clock voltages which are 8 and grounded). A reduction of the clock will reduce the leakage current the resolution will be impaired 26 V (substrate voltage amplitudes however, charge

NES(shot) accounts

= 0,591 % m

for most of the noise. In the low leakage chip NES = [0.33 pJ/m current (135e-) ] noise to the leakage of 1.5 pA/well

devices the measured can be attributed or a NES =

substantially, due to inadequate

[0.275 pJ/m2

(112e-) ]. The remaining

transfer. Fig, 11 illustrates the variance or noise measured for the correlated and uncorrelated sampling case. To obtain the uncorrelated noise we reversed the clamp and sample sequence such that they were not performed in the same pixel time window. This is proof of the correlated double sampling technique in action. Measurements had been taken lower leakage current on another CCD line array with a (1.5 pA/well) and the total NES

is attributed to the surface-state noise and in the low leakage device this is a NES = [0.18 pJ/m2 (74e-) ]. The noise measurements do not appear to change across the 1 x 64 element line array and the values quoted are representative of the average noise along the array. One reason for the noise to remain surface-state amplifier. constant across the array is from of is that the major the electrometer noise contribution

There

is also a suppression

WHITE

et

al.:

CCD IMAGE

ARRAYs

----t ,
I

It, 0 ?-

_._+___
.aAe) N

(suomala

w.161s,aqumu

10
,ooo~ ~ BLANKING INTERVAL EXPOSURE DENSITY E- IgOZ(pJ/m2)

IEEE JOURNAL

OF SOLID-STATE

CIRCUITS,

FEBRUARY

1974

NEST CORRESPOND TO MIDPOINT OF EXPOSURE DENSITY LEVELS N EST

0,74 1A/JIn2)

0,62

0,52

-l -. 3 . ~ ~ : w . a

0,43 - .__, _ .. ___

O 42 _ _

10 _ _ ___

--R -0065

UNIFORM IRRADIANCE FAT ZERO pA/mV//In2

(NO FAT ZERO)

(406,/ /IJ/m2) MEASUREMENTS ON DEVICE NO, 27 ,( EXPOSURE TIME= LINE TIME) = I,32MSEC BLANKING TIME =4o #SEC NESCHIP = 0,33 # J/m2

NE% = NES2CHIP + NSyST2 + N*RAD + 0t2 SYST 0 15p Jlm2 T

NRAD = O 64 /JJ/rn2 01 0 ~= 5 10 0087 15 20 25 30 35 40

45

50

55

60

65

RELATIVE ELEMENT POSITION FROM COLLECTING OIODE

Fig. 12. Low light level response uniformity and noise performance of a 1 x 64 element CCD Iihe array with and without a 20 percent fat zero. (CDS analog signal processing at a clock frequency j. n 50 kHz.)

the surface-state tion state capacitance

noise within adjacent may be written

C,t
w

the array wells as

by the correlaThe surface

performance formance or held still lines. In the

and noise sources array. for approximately absen~e of a fat dark) varies

which

influence

the perbetween response high for

of noise between

[13].

of the CCD

The line array

is (blanked

~j$l,,A
(states/cm*.
of x the 1010 MOS cm-z

2 pixel times zero, the video across the array:

(16) eV)
elecev-l

(response

minus

where N,t is the interface state density and A the area of the gate electrode
trometer. near the Typical band edge values at 50 of kHz N,t and H 4

the first pixels (i.e., pixels nearest collection diode) after the blanking interval and low for the last pixels. Inefficient during transfer the blanking of excess interval diode charge, which accumulates potential wells only in the larger

A = 6.2 x

I&G cm2 of

yield

a C.t = 0.04 pF and a noise contribution = (lcTC ,)1 ~ = 0.20$

nearest the collecting

(see Fig. 3), contributes

NES (surface state)

(80e-)

(17)

to the high response. Inefficient transfer of photocharge along the CCD register causes the video output from the last pixels to be diminished first pixels by residual in the next charge which video line. If is we added to the

which is close to the observed value. Fig. 12 illustrates the responsivity and noise performance of the 1 x 64 element CCD line array discussed in the preceding paragraphs. Table II summarizes the noise

inject an electrical fat zero (2o percent full into the shift register, the output is uniform lines time in Fig. window these (12] ] because produced conditions of improved nonuniform ency. An electrical fat zero injected a high, was

CCD well) [see dashed efficipixel well

transfer

over the entire full CCD [4.o

noise at the ~J/m2

z N, I is an effective transfer function of the surface states.

surface-state density signal processor and

modified by the distribution

the of

output. under

The output

noise for a 10 percent approximately

WHITE

etCd.:CDIMACJE ARRAYS C
TABLE II
OF ~OISE PERFORMANCE OF CCD IMAGING ARRAY (No EVZCTRICAL FAT ZBRO)

11

SUMMARY

R =
Total

406e-/pJ/mz NES (measured) at 166 ~, = ~~ () x system noise) = 112 = 0,74$

~z

Calculated

radiation

shot

noise

~,64@
m2

Signal

processor noise

noise

(measured lIZ = (().64)2 --

015+J . mz

Quantizing NES (chip)

(!Jz iZ ()

0.087

~,

((0.74,)2

(0.15)2

(().087 )2)2

o,33@ mz. o,34@

m2

(kTC,,;

+qI~~r)2(

theoretical)

NES

(shot)=

~~ (k!/c,,)/ = r

= =

0.275:, 020@
mz

NES

(surface)

Fig. 13. Low light level imaging with and without fat zero. (Collection diode at photo-bottom for a.t photo-top for middle pictures.)

a 20 percent end pictures,

(1600e-)] about

and consistedof

irregularburst injections

noise. When was narrowed to noise for noise. fur-

electrical fat zero the transfer inefficiency c d 1 x 10-4 from low frequencies to 2 MHz, which is sufficient to operate injection variation straight both line and area CCD and there is less than in NES line with imagers. We have made with with * a fat-zero in from range * 2 5 percent variation measurements on the response uniformity array entire zero)

the time window 1 percent a 20 percent [3.0 pJ/m2 The method

for electrical

clf the pixel window,

the output

full CCD

well was reduced to approximately no evidence a truly of irregular requires fat zero injection

& 5 percent The

(1200 e-) ] with of electrical

response across a 1 X 64 element linearity over the fat

across the array. an electrical

deviation dynamic is within

ther investigation process.

to produce

low noise injection

(> 60 dB percent. Imaging in

V. Low
Transfer Inejkiency

L:IGHT

LEVEL

IMAGING

Evaluation the CCD line arrays we employed image motion in the on all at

at Low Imagers

Light

Levels

In order to evaluate a rotating while the line array

Surface-Channel The modulation the CCD

CCD

drum to provide direction. levels

the along track

was scanned

electronically

transfer

function from

(MTF)

[4] is deterof to well

across track the drum irradiance

A transparency by a calibrated to

was mounted source with source

mined by the charge transfer to transfer charge

inefficiency

e. The ability

and irradiated

one potential

referenced

a blackbody

another is described by the transfer inefficiency E, which in a practical CCD is limited by surface-state trapping [3] lcTN, , = ~ where N,t is the well interface N,ig the ln f/to) state signal density charge f. (typically in the =

6000 K in the 400 nm to 800 nm band. Fig. 13 illustrates the low light level performance of the CCD 1 x 64 element which line array with 3 ms exposure time (0.75 ms line Our minirnurn zero is 0.97 signal, if noise,3 is readout time) is about and a highlight with irradiance current. fat of 18.7 mW/m2,

120 pA of signal

(18)
= CCD 10 zero

detectable mW/m2

signal

a 20 percent

or about 6 pA. The minimum the fat zero with or about 0.4 pA. Notice the electrical the elements furthest

detectable negligible

we could introduce 0.08 mW/m2 if we remove *White House

1 X 100/cm2. eV), shift register surface-state kHz). A small

the loss in resolution from the collecting

(charges/cmz)

~and ),0the characteristic (typically 1 charge or fat

fat zero. In the scene of the

escape frequency amount

of background

is required to fill the interface reduce the transfer inefficiency. measurements, the center of variable transfer with the various inefficiency

states permanently and Our transfer inefficiency profile elements, slit placed indicate line array. in a

diode are at the top of the picture. It is apparent that the long number of empty CCD wells (as denoted by the black tree area), affects the resolution severely since these wells do not have a background radiation fat zero to provide low transfer inefficiencies.
electrical the simple injection of a fat zero [14] shot noise associated with

a 110ym optical sensor

across the CCD

The transfer efficiency improves as leakage accumulates to fill the surface states along the shift register. With an

~ A method for low-noise, can give a IVES less than the fat zero.

12

IEEE JOURN.4L OF SOLID-STATE CIRCUITS, FEBRUARY 1974

Since a CCD requirements interline

imager

uses an analog

shift

register With the

we the

fringes silicon which design diffused system if digital

caused gate CCD

by

this

silicon

gate

thickness efficiency

and

the

must have low leakage register of low leakage we transfer method

elements in addition sensor elements. can separate

to the causes

underlying

SiOz. The

effective

quantum

of the

sensor in the range from 400 nm to 800 with a photodiode should array with the ~ = 0.65, to the a should Qr, use of Consideration imaging be given ratio

nm was ~ = 0.35, compared is constant. of the CCD photodiode

of leakage in an imager by simply turning on and off the transfer gate. In general, we encounter more leakage in the shift register than the sensor elements. transfer register. will begin imaging approach If to free-running for a period pictures One advantage register or the leakage of the interline can be operated throughout current processor. the high imagers. VI. We called nates, have developed
CONCLUSIONS

for the sensor. The g~/C collection and the quantizing is used, should

is that the shift to distribute

be increased

at the output

diode to overcome interval be decreased in

noise limitations; signal processing with dynamic

the

the register saturate that

is blanked the analog with

held stationary spots The quality

of time, then the local clarksignal in Fig. is possible 13 illustrate CCD

accordance

range and data rate limitations.

illustrated

AGKNOWLEDGMEXT

The authors wish to express their appreciation to R. M. McLouski for the diffusions and oxidations, P. R. Reid for the polysilicon depositions, C. J. Taylor would J. Grossman for for the chemevapand ical processing, They the photolithography, D. H. McCann discussions, (Manager), M. N.

a method sampling

of

signal

processing removes the cadiode, elimiand in-

correlated transients the Nyquist

do?lble

which collection

and D. S. Herman orations. for his valuable Giuliano

for the special aluminum/silicon also like to thank and technical to W. S. Corak Laboratory, assistance

switching pacitance

at the output provides

noise of the reset switch-output dc restoration range, and suppresses The analog

combination,

to express appreciation State Systems

creases dynamic

I/f noise contributions. which uses this technique frequencies imum imager from ured the intrinsic detectable with this

surface state and signal processor, with clock We have meas-

and FL C. Gallagher Technology

(Supervisors),

of the Solid

and to G. Strull,

has been operated

800 Hz to over 3 MHz. noise-equivalent-signal input exposure The density primary with and technique. in the CCD inefficiency

Manager of the Advanced their encouragement and accorded to C. Lesniewski of the manuscript.

Technology Laboratories, for support. A special thanks is ancl S. Viscomi for preparation

ONESI or minof the limitation thermal shift CCD to leak-

REFERENCES
[11

the N13S is the shot noise associated age current The surface generated channel sensor c ~ imager

register. geo-

requires

a 20 percent

fat zero for transfer metrical electrical channel method indicate ever, [14] with

10-4 to obtain

resolution. The excess noise associated with the fat zero limits the sensitivity of the surface C(3D ancl attention must be directed toward a of low noise electrical a noise charge enable limited in range injection. For an electrical howcircuit

fat zero introduced a redesign should an NES

by a gated diode, our measurements of 1000 to 1200 electrons; electrical channel leakage injection CCDS For

S~st. Tech. J., vol. 49,, p. 587, 1970. M. F. Tompsett, and G. E. Smith, Experi[21 G. F. Amelio, mental ve~~fication of the charge coupled device concept, Bell S~st. Tech. J., vol. 49, p. 593:1970. The quantitative effects of interface states [31 M. F. Tompsett, on the performance of CCIYS, presented at the Int. Electron Devices Meeting, Washington, D. C., Oct. 1971; also Tech. Papers Abstracts, p. 70, and IEEE Trans. Electron Devices, vol. ED-20, pp. 4555, Jan. 1)73. [41 W. B. Joyce and W. J. Bertram. Linearized dispersion relation and Greens function for discrete charge tra~sfer devices ~~i~, incomplete transfer, Bell Syst. Tech. J., vol. 50, p, 1741, [51 M. H. White, D. R, Lampe, F. C. Blaha, and I. A. Mack, [Charge coupled device (CCD) imaging at low light leveis, presented at the IEEE Int. Electron Devices Meeting, Washington, D. C.j 1972. Characterization of charge coupled device line and area~array imaging at low light levels? presented at the IEEE Int. Solid-State Circuit Conf., Phdadelphia, Pa., 1973. R. H. Dyck, and G. P. Weckler, (A new self-scanned photodiode array, Solid-State Technol., p. 37, July 1971. M. Ashikawa, hT. Ko~ke, T, Kamiyama, and S. Kubo, A new spl~e noise elimination technique for photosensitive arrays, presented at the IEEI!j Int. Solid-State Circuits Conf., Philadelphia, Pa., 1973. J. E. Carries and W. F. Kosonocky, Noise sources in chargecoupled devices, RCA Rev., vol. 33, p. 327, 1972. D. F. Barbe, (Noise and distortion considerations in chargecoupled devices. lllect~on, Lett., vol. 8, p. 207, 1972, J. D. Plummer and J. D. Meindl, MOS electronics for a portable reading aid for the blind, IEEE ~. Solid-State Circuits, vol. SC-7, pp. 111-120, Apr. 1972. J. A. Betts, Signal Processing, Modulation and Noise. New York: Elsevier, 1971. K. K, Thornber and M. F. Tompsett, Spectral density of noise generated in charge transfer devices, IEEE Trans. .Electron Devices, vol. ED-20, p. 456, Apr. 1973. S. P. Emmons and D. D. Buss, Techniques for introducing a low noise fat zero in CCDS, presented at the Device Res. Conf., Boulder, Colo., June 1973.

W. S. Boyle and G. onductor devices, Bell

E.

Smith,

[Charge-couplecj

semic-

of the input surface

to operate our simple

[61

by thermal circuit with

shot noise and fat zero, we

[71 [81

a dynamic

excess of 80 dB.

gated diode injection

a 20 percent

have achieved greater than 60 dB dynamic range with less than * 2 percent deviation from linearity over this range and less than ~ 3 percent variation in responsivity dark across out and NES current the across a 1 x variation We fat was have 64 element less than made line * array. 3.4 The percent withexposure at an inte-

[91 [101 [111

array. a noise

measurements uniform

an electrical

zero and under charge

to indicate gration primarily

of 135 electrons

time of 1.32 ms. The noise was associated with the thermal leakage current in the CCD current on the better

[121 [131

shift register. The thermal leakage arrays was about 50 nA/cm2. Spectral response dicate a nonuniform

and responsivity measurements inspectral response due to interference

[141

WmTE

et al.:

ccD IMAGE ARRAys

13 ment on pyroelectric detectors, advanced explosives and weapons detectors, and more recently on charge-coupled devices (CCDS) for imaging and both analog and digital signal processing. He is a Fellow Engineer at the Westinghouse Advanced Technology Labs.

Technology

Labs.

Donald R. Lanipe (S57M68) was born in Baltimore, Md.~ on January 6, 1940. He received the B.E.S. and Ph.D. degrees in electrical engineering from the Johns Hopkins University, Baltimore, Md., in 1960 and 1966, respectively, and the M. S.E.E. degree from the California Institute of Technology, Pasadena, in 1961. Since 1961 he has been employed by the Westinghouse Electric Corporation, Defense and Space Center, Baltimore, Md. Until

Ingham Montreal.

A.

Mack (fl65-M67) was born in P. Q.. . Canada. on June 1.1944. He .

received the B .S.E.E. degree from the University of California, Berkeley, in 1967, and the M. S.E.E. degree from San Jose State College, San Jose, Calif., in 1969. Since 1968 he has been employed by the Westinghouse Defense and Electronic Systems Center, Baltimore, Md,, in the design and characterization of integrated circuits. In
particular, he has worked on CMOS selfscanned photodiode arrays and more recently, CCD self-scanned CCD sensor imaging. He is an Engineer at the Westinghouse Advanced Technology Labs.

1967 he was engaged in the study of electron paramagnetic spectra primarily and relaxation at microwave frequencies. He was involved
with microwave integrated active duty. Subsequently, circuit during two years of U.S. Army he has performed research and develop-