Вы находитесь на странице: 1из 13

See discussions, stats, and author profiles for this publication at: https://www.researchgate.


nanoMOS 2.5: A two-dimensional simulator for quantum transport in double-

gate MOSFETs

Article  in  IEEE Transactions on Electron Devices · October 2003

DOI: 10.1109/TED.2003.816524 · Source: IEEE Xplore

241 612

5 authors, including:

Zhibin Ren Rachana Venugopal

IBM CMR Institute of Technology


Sebastien Goasguen M.s. Lundstrom

Clemson University Purdue University


Some of the authors of this publication are also working on these related projects:

Scientific Software Development for nanoHUB View project

All content following this page was uploaded by M.s. Lundstrom on 21 May 2013.

The user has requested enhancement of the downloaded file.


nanoMOS 2.5: A Two-Dimensional Simulator for

Quantum Transport in Double-Gate MOSFETs
Zhibin Ren, Ramesh Venugopal, Sebastien Goasguen, Member, IEEE, Supriyo Datta, Fellow, IEEE, and
Mark S. Lundstrom, Fellow, IEEE

Abstract—A program to numerically simulate quantum trans- phase [17]. The method has been developed into a practical tool
port in double gate metal oxide semiconductor field effect transis- for simulating one-dimensional resonant tunneling diodes [14],
tors (MOSFETs) is described. The program uses a Green’s func-
and more recently, it has been extended to two-dimensional
tion approach and a simple treatment of scattering based on the
idea of so-called Büttiker probes. The double gate device geometry bulk MOSFETs [13], [31]. It is also now widely used to explore
permits an efficient mode space approach that dramatically lowers conduction at the mesoscopic [6] and molecular scales [32]. In
the computational burden and permits use as a design tool. Also comparison to phenomenological quantum approaches [24],
implemented for comparison are a ballistic solution of the Boltz- [29], the Green’s function method offers the advantage of rigor,
mann transport equation and the drift-diffusion approaches. The
program is described and some examples of the use of nanoMOS but the cost can be a heavy computational burden. The simple
for 10 nm double gate MOSFETs are presented. geometry of the fully-depleted SOI MOSFET (single or double
gate) reduces this computational burden. It also permits the
Index Terms—Boltzmann transport equation, Büttiker probes,
double gate, drift-diffusion, MOSFETs, nanoMOS, quantum use of a mode space approach that, while not suitable for bulk
transport, scattering. MOSFETs, dramatically lowers the computational time for
ultra-thin-body SOI MOSFETs [35].
For ballistic transport, the NEGF formalism is equivalent to
solving the Schrödinger equation, but the NEGF formalism is

A S METAL oxide semiconductor field effect transistor

(MOSFET) channel lengths rapidly shrink below 100 nm,
current research focuses on understanding device physics and
readily extendible to two- and three-dimensions or to the use of
molecular orbitals [22]. It also provides a clear prescription for
treating large contacts attached to an intrinsic device [28]. The
ultimate scaling limits as well as the practical issues that need NEGF formalism also provides a sound conceptual basis upon
to be addressed to achieve channel lengths at the 10 nm scale which additional physics (such as scattering) can be included
[1]. Computational studies can help address these issues, but as needed. See [9], [23] for a tutorial introduction to the NEGF
the methods commonly used in computer-aided design tools method and [7] for an example of its application to conduction
do not include (or do so only phenomenologically) important in molecules.
quantum mechanical effects. What is needed is a full quantum This paper describes nanoMOS2.5, a simulation program that
transport model—one that describes quantum phenomena uses the NEGF approach to simulate fully-depleted, nanoscale,
(especially confinement and tunneling) as well as phase ran- SOI MOSFETs. It begins in Section II with a description of
domizing scattering. Computational efficiency is necessary to the methods focusing on the mode space approach. Section III
permit use for engineering design and to explore numerous
discusses some questions relating to the numerical implementa-
design options. It would be useful for such a program to be
tion. In Section IV, we describe a phenomenological treatment
based on an approach that can be extended to nonconventional
of scattering, and in Section V some results are presented.
devices, such as carbon nanotube FETs [2], [10], [20] and
other molecular transistors. In this paper, we describe such a
computer simulation tool, nanoMOS2.5. II. APPROACH
The program, nanoMOS, uses the nonequilibrium Green’s In this section, we describe the approach used in nanoMOS
function (NEGF) approach, which provides a rigorous de- from a wavefunction perspective, because ultra-thin body DG
scription of quantum transport and interactions that randomize MOSFETs can be treated by a one–dimensional (1-D) approach,
and for 1-D quantum transport with the simple scattering
Manuscript received August 29, 2002; revised June 11, 2003. This work model currently implemented in nanoMOS, the wavefunction
was supported by the Semiconductor Research Corporation under Contract
NJ-99-724 and an ARO Defense University Research Initiatives in Nanotech-
and NEGF approaches are equivalent. We use a wavefunction
nology (DURINT) Grant. The review of this paper was arranged by Editor H. description because it is more familiar to readers of this journal,
Sakaki. but in the Appendix, we translate the key equations into the
Z. Ren is with IBM, Yorktown Heights, NY 10598 USA (e-mail: zhibinr@
us.ibm.com). language of the NEGF formalism. Readers are referred to a
S. Goasguen, S. Datta, and M. S. Lundstrom are with Purdue University, West tutorial introduction [9] or to standard references [15], [18] for
Lafayette, IN 47907-1285 USA. a more extensive discussion of the NEGF approach.
R. Venugopal is with Texas Instruments, Dallas, TX 75235 USA (e-mail:
venugopal@ti.com). Fig. 1 sketches the double gate MOSFET geometry that we
Digital Object Identifier 10.1109/TED.2003.816524 assume; we seek a solution for the intrinsic device only. Because
0018-9383/03$17.00 © 2003 IEEE

Fig. 1. Double gate MOSFET structure examined in this work. An oxide

thickness of 1 nm, and a body thickness of 3 nm have been used to highlight
quantum effects within this device geometry. The power supply is 0.4 V and
the gate work function is adjusted to obtain an I (10 A/m) consistent
with the ITRS specifications for a high performance transistor. The S/D doping
is 2 10 cm and the transistor is assumed to be wide (Y-dimension is
Fig. 2. Sketch of the generic subband energy versus position along the channel
treated as infinite).
(x). Also shown are the semi-infinite source-drain contacts (bounded by open
rectangles) and the boundary conditions for injection of a unit amplitude from
we assume that the device is wide in the -direction, the wave- the source end. The nodes within the device are numbered 1 to N and the active
device extends from x = 0 (source) to x = L (drain).
function can be written as ,
where is obtained from
where is the expansion coefficient of with re-
spect to the mode-space eigenvector, .
(1) Equation (4) is the decoupled mode space transformation
of (1); the key assumption is that the shape of the mode does
where is the longitudinal energy. One could directly dis- not change rapidly along . Although the potential, ,
cretize (1) in real space, which is often necessary, but it imposes varies from source to drain, the largest variation is with position,
a heavy computational burden [35]. For thin body, SOI MOS- . For a thin body, the shape of the mode (i.e., its -dependence)
FETs, however, quantum confinement in the -direction intro- changes slowly along the channel. A careful comparison of
duces subbands, and for a thin body, only a few subbands are the decoupled approach with a direct, real-space discretization
occupied. Accordingly, we expand the wavefunction in an or- shows that the decoupled approach produces results that are
thonormal basis as [35] essentially identical to the exact solution for body thicknesses
of up to at least 5 nm [35]. There are, however, conditions for
(2) which the decoupled approach fails [35]. One example is the
transition from the flared out contact to the ultra-thin channel
where are the eigenfunctions (modes) associated (which is not treated in nanoMOS), and another is the bulk
with confinement in the -direction as defined in Fig. 1. These MOSFET.
eigenfunctions and the associated eigenenergies are obtained Equation (4) is a 1-D wave equation that greatly reduces the
by solving a one-dimensional wave equation in the -direction size of the two-dimensional (2-D) problem. In the ballistic limit,
within each vertical slice of the device along the dimension each mode (subband) can be treated separately. Each transverse
mode is also independent, because we assumed there is no po-
tential variation in the -direction. In this 1-D equation, instead
of the bottom of the conduction band, , we have the
(3) bottom of the relevant subband, , which is determined
The energy, , represents the bottom of subband, , which by a solution to (3) in the confinement direction. (Accordingly,
varies with position, , along the channel. The envelope wave- when we plot energy band diagrams, we will plot rather
functions are assumed to be zero at the oxide/Si interfaces if than the bottom of the conduction band, ). We also see
electron penetration into the oxide regions is neglected (other- from (4) that the relevant energy for the 1-D problem is not the
wise, the zero boundary is extended to the gate contact/oxide total energy, , but the longitudinal energy, .
interfaces). Because the decoupled mode space solution reduces the 2-D
By using the orthonormal basis of (2), we can transform (1) problem to a set of 1-D problems, one for each subband and
to a mode-space basis. By retaining only a few occupied modes, transverse energy, the solution procedure is much like that of a
the computational burden can be significantly reduced com- true 1-D problem. Fig. 2 is a sketch of the subband energy versus
pared to a direct discretization in two-dimensional real space position for one subband. Semi-infinite contacts are attached to
[8]. The geometry of the double gate MOSFET offers yet an- the device at the source and drain ends. Because the potential
other simplification. If we assume that the shape of the confined in the contacts is assumed to be uniform, the solutions in the
mode does not change along , i.e., then (2) semi-infinite contacts are plane waves. If a unit amplitude wave
becomes [26], [35] is injected from the left (source) contact, then some portion re-
flects from the device and some transmits across and exits the
perfectly absorbing right (drain) contact

(4) (5a)

and where is the local density of states due to injection

(5b) from the drain, which is computed from the wavefunction with
boundary conditions analogous to (5a) and (5b). Note that
where and are the reflection and transmission coefficients in deriving (9), an equilibrium Fermi–Dirac distribution was
for source injection into mode and is the length of the active assumed to prevail within the S/D contacts only. No assumption
device. was made on the shape of the distribution function within the
By solving (4) subject to the boundary conditions, (5a) and active device. In the ballistic case, the source and drain injected
(5b), we find the wavefunction due to the injection of a unit states are independent, and can be filled by the appropriate
amplitude wave from the source. The corresponding electron Fermi functions as shown in (9). Equation (9) clearly indicates
density versus position for confined mode, , at each , is that under nonequilibrium conditions (the source Fermi level
obtained by summing the contributions from each transverse is different from the drain Fermi level), the net 2-D electron
mode distribution within the device, which is a combination of source
and drain injected streams, cannot be characterized by a single
(6) Fermi level. Its shape is not equilibrium Fermi Dirac.
To obtain the current due to source injection into mode , at
where refers to the -component of the wavevector of an each we evaluate
electron with total energy, , in the source contact and the sub- (10)
script, 1, refers to injection from the first (left or source) contact.
The probability that the state at energy, , in the source contact
is occupied is give by the source Fermi function because we as- where is the current transmission coefficient from contact
sume that scattering maintains thermal equilibrium in the con- 1 to contact 2 for electrons in subband, . The net current due
tacts. to mode is obtained by summing (10) over all the positive
The 2-D charge density for mode at site is obtained by states and by subtracting the corresponding drain injected
summing (6) over all of the positive states (negative states component. On converting the sums over the ’s to integrals
are not injected into the device from the source). Since the width in (10), and by performing the integral over transverse energy
in the transverse direction, , is assumed to be large, we can analytically we find
convert the sum over transverse states to an integral over the
transverse energy. We can also convert the sum over the injected (11a)
states to an integral over longitudinal energy. The integral
over transverse energy can be done analytically and the final where
result expressed as [26]

(7a) (11b)
and is the Fermi function which is analytically integrated over
where we let the top of the band approach infinity and transverse energy (spin degeneracy is included in ). The total
current is obtained by summing the contributions from each sub-
F band and valley.
Finally, the current transmission coefficient for mode,
(7b) is

with (12)

(8) which, from (5a) can be expressed in terms of the computed

wavefunction as
being the local density of states due to injection from the source (13)
(spin degeneracy is included in the square root factor). In (7b)
F is the Fermi–Dirac integral of order [3] and F is
the Fermi function integrated over the transverse energy. To find III. NUMERICAL SOLUTION
the total electron density within the device, we sum the contribu- A. Solving for the Wavefunction
tions from each subband and valley and include the contribution
due to injection from the drain To evaluate the expression for the electron density and current
in the previous section, the wavefunction within the device must
be known. In the confinement direction, (3) is a standard eigen-
value problem solved by finite differences. In the longitudinal
direction, we discretize (4) on a finite difference grid imposing
the boundary conditions, (5a) and (5b), to find

where is the discretized Hamiltonian operator where is the three-dimensional (3-D) effective density-of-
states, and the function, , is the Fermi–Dirac integral of
order . We regard (22) as a mathematical change of variables
with an on-site energy of to a quantity, , whose physical significance under off-equi-
librium conditions is unclear, but unimportant. When (22) is
(16) inserted into (21), a nonlinear Poisson equation results. After
computing from the quantum transport model, a cor-
for a finite difference grid with node spacing, . The energy, responding is computed from (22). This “quasi-Fermi
, in (15) is the effective potential energy for electrons in level” is used in a nonlinear Poisson equation that is obtained by
mode . The “self-energy” matrices, and account inserting (22) in (21) and solved by Newton’s method. The ad-
for the open boundary conditions [(5a) and (5b)] and are vantage of this approach is that it builds a negative feedback into
the iterative process. If the potential increases (conduction band
decreases) during the Poission solution, the subsequent trans-
(17b) port solution will increase as carriers flow to regions of
The vector, is a source term accounting for injection lower energy. This coupling is built into the Poisson equation
from the left contact (source). It has only one nonzero compo- when (22) is used. The approach has proven effective in pre-
nent, the first vious quantum and semiclassical transport simulations [5], [34]
and proved similarly effective here.

(18) C. Boundary Conditions

Boundary conditions must be specified for both the transport
For injection from the drain, we use boundary conditions anal- equation and for Poisson’s equation. For Poisson’s equation,
ogous to (5a) and (5b), and the corresponding vector has a the boundary conditions at the gate electrodes (Dirichlet) and
nonzero component in position, . at the oxide/air interfaces (Neumann) are standard. Boundary
The solution to (14) gives the value of the wavefunction, , conditions at the interface between the S/D extensions and the
at each of the finite difference nodes. The formal solution is flared out contacts are also needed. A more complete solution
(19) would include the flared out region to explore the resistive drops
that may occurs at the wide/narrow transition. In this work, we
where seek simple, upper limit boundary conditions that define our
(20) “ideal” contacts.
External to the intrinsic device being simulated, we assume
is the retarded Green’s function in a discrete basis. Because large source and drain contacts where scattering maintains
the matrix in (14) is tridiagonal, it can be efficiently solved by thermal equilibrium. We solve the wave equation assuming
Gaussian elimination; the carrier density and current are then a unit amplitude injected wave, then weight by the Fermi
evaluated from the computed wavefunctions. Alternatively, as function of the appropriate contact. The Fermi levels in the
shown in the Appendix, we can express all of the results in expressions presented in Section II refer to the Fermi levels
terms of the retarded Green’s function. For this simple, in the equilibrium source and drain contacts that are external
1-D problem, the Green’s function approach may appear to be to the region being simulated. The self-energy matrices are
a complicated way to solve a simple problem. The advantage of expressed in terms of the wavevector, , in the contact from
the NEGF formalism become apparent when extensions to two which the electrons are injected. For a uniform contact with
and three-dimensions are contemplated, when an atomic basis simple bands
is essential, or when a rigorous treatment of scattering is neces-
sary. (23a)

B. Poisson’s Equation but the discrete grid modifies this relation to

The program, nanoMOS, computes a self-consistent solution (23b)
to a quantum transport equation and Poisson’s equation
where is the grid spacing. The two expressions are nearly
(21) equal when . Care must be taken to ensure that the
maximum energy used is well below the top of the band so that
Equation (7) gives the electron density per unit area within the
(23b) approximates (23a). It should also be recognized that the
device; to convert it to a density per unit volume, it is distributed
maximum energy that can be safely used depends on the grid
in the -direction according to the computed eigenfunction in
spacing, .
the confinement direction. Direct use of (21) leads to slow con-
For ballistic transport, the boundary conditions on Poisson’s
vergence. Instead, we solve a nonlinear Poisson equation. The
equation must be carefully specified. Fig. 3 is a sketch of a
electron density evaluated from the wavefunction can be related
generic subband profile from the source to the drain. Under low
to a quasi-Fermi level by
gate bias and high drain bias, the source-to-channel barrier is
(22) high. Therefore most electrons injected from the source (empty

Fig. 3. Illustration of why floating boundary conditions are assumed for the
potential at the contacts. The empty circles on the source side, below the dotted
line (source to channel barrier) represent source injected electrons reflected by
the barrier. The drain injected electrons are represented by filled circles.

circles in Fig. 3) reflect from the source to channel potential bar-

rier, and both positive and negative velocity states ( states) are
occupied in the source extension as a result of source injec-
tion. Under high gate and drain bias, however, fewer electrons
are reflected from the barrier, and the electron density in the
source extension decreases (there is very little contribution
to the overall electron concentration at the source end due to
drain side injection as the drain voltage is high). If the poten-
tial at the source contact is fixed, space-charge neutrality (the
2-D donor doping concentration equals the 2-D electron density
obtained by integrating along ), which is a result of self-con- Fig. 4. Summary of the ballistic solution scheme.
sistent electrostatics, cannot be maintained. (Note that the loca-
tion is the Fermi level is fixed by the large, thermal equilibrium 2) For each position, , along the channel, (3) is solved to
source reservoir.) Therefore, in order to maintain space-charge find the eigenfunctions and eigenenergies versus position.
neutrality in the source/drain (S/D) extensions, the electrostatic A grid spacing of in the -direction is typical.
potential floats relative to the fixed Fermi potentials in the S/D 3) A grid in longitudinal energy is defined (with a typical
contacts. We allow for this to happen by specifying grid spacing of 0.5 meV) and (4) is solved to find the
wavefunction due to injection of a unit amplitude wave at
(24) longitudinal energy, , from both the source and drain
contacts. The electron density for each mode is evalu-
ated from (7b). This step is repeated for each longitudinal
at the idealized source and drain contacts. This boundary con-
energy in the grid and for each occupied confined mode
dition causes the potential at the source end to float down, thus
(as set by the user). The total carrier density is obtained
allowing more carriers to enter into the device from the source
by summing the contributions from each longitudinal en-
reservoir. This allows macroscopic space-charge neutrality to be
ergy, confined mode, and valley as in (9).
maintained (Fig. 8) at the source end irrespective of the biasing
4) The nonlinear Poisson equation is solved to update the
condition. The floating boundary condition eliminates the need
electrostatic potential. The maximum change in potential
to resolve the potential at the point where the thin body couples
is compared to the convergence criteria, and the process
to the large S/D. It represents an ideal upper limit contact. In
continues until convergence is achieved. The minimum
practice, the transition from the flared out S/D to the ultra-thin
change in potential that can be achieved is essentially the
intrinsic device could introduce quantum mechanical parasitic
spacing in the energy grid.
resistances that are not treated within nanoMOS. The validity
5) After convergence is achieved, the transmission coeffi-
of this approach is supported by the results that will be shown
cient and current at each energy are evaluated from (13).
in Section V.
The contributions from each energy are summed to find
the total current from (11). The process is repeated for
D. Solution Procedure each mode, and the results summed to compute the total
The solution procedure is summarized in Fig. 4; it consists of current.
the following steps. Although we have described the solution process from a wave-
1) An initial guess for the 2-D electrostatic potential is de- function perspective, because this approach may be more fa-
fined. (We use a drift-diffusion simulation as our initial miliar to readers of this journal, the actual equations solved in
guess.) nanoMOS are expressed in the Green’s function formalism as

described in the Appendix. The Green’s function formalism is

a generalization of the concepts outlined this far, which can be
naturally extended to model devices in two or three dimensions,
including the effects of scattering.


The NEGF formalism provides a prescription for including
scattering, but scattering greatly increases the computational
burden because it couples the longitudinal and transverse en-
ergies. Instead of treating transverse modes as independent and
integrating over them analytically, we need a grid in transverse
Fig. 5. Illustration of the concept of Büttiker probes. The probe self-energy is
energy too. And instead of treating each longitudinal energy in- adjusted to tune scattering rates and the probe Fermi-levels to obtain zero net
dependently, they must be coupled. Scattering has been treated current at each scatterer.
rigorously in NEGF simulations of MOSFETs [30], [36], but
the resulting computational burden limits the use of such simu- them into the device. Following the prescription for filling states
lations. The advantage of the NEGF approach lies in its ability from the two real contacts, (9), we have
to treat quantum transport using a atomic level Hamiltonian, if
necessary, but we also need a simple way to capture the main
effects of scattering.
We treat scattering in a way that is analogous to the well-
known relaxation time approximation to the collision operator
for the Boltzmann equation
The spectral density, , gives the local density-of-
which describes carriers out-scattering from a state at a rate,
states everywhere along the channel due to injection from the
, and in-scattering from a thermal equilibrium distribution.
probe located at . We have assumed a single scatterer at
(Under nonequilibrium conditions, the magnitude of the in-scat-
with a Fermi level, , but in practice, probes are placed at
tering rate is adjusted to preserve current continuity.) An analo-
each node and each one has a different Fermi level. The electron
gous treatment for the wave equation can be implemented with
density at due to injection from all reservoirs (the source and
an idea due to Büttiker [4]. As shown in Fig. 5, we conceptu-
drain contacts as well as the floating probes used to introduce
ally attach a floating contact to each node. Carriers are removed
scattering) is
from the device and injected into the floating contact where they
are thermalized and re-injected into the device. (29)
Since each scattering center is phenomenologically treated
as an additional contact, nanoMOS models the effect of out-
where the sum runs over all the reservoirs, each with its own
scattering by adding an additional potential to the Schrödinger
Fermi level. The source and drain Fermi levels are fixed by the
equation [9]
applied source-drain bias; the Fermi level of each probe is de-
termined by requiring that current is conserved. The current in
(26) subband, , at reservoir, , is given by an expression analogous
to (11b)
Note that the scattering potential is similar to the “self-energies”
[(17a) and (17b)] used to account for the open boundary condi-
tions [(5a) and (5b)], and that the coupling energy between the
device and the scatterer, , can be smoothly varied to mimic (30)
specific low-field mobilities as described in [36]. Our scattering where the sum over is over the remaining reservoirs. The trans-
model is phenomenological and analogous to the relaxation time mission coefficient between any two reservoirs is computed in a
approximation. It captures the cumulative effects of all types manner analogous to (12). The source and drain contacts carry
of scattering, including phonon scattering. When scattering is current, but the Büttiker probes do not, so for Büttiker probe,
present, the retarded Green’s function in (20) becomes
(27) (31)

Equation (26) does a reasonable job of describing scattering Equation (31) gives a set of constraints on the Fermi levels
in MOSFETs if the scattering strength is calibrated to an ap- of each probe. This set of constraining equations is solved it-
propriate mobility. The diagonal self-energy results because the eratively for the Fermi level of the probe [36]. Note that the
scattering potential is assumed to be a -function at each node. resulting position dependent Fermi level behaves in much the
We have described the process of out-scattering electrons to same way as the quasi-Fermi level in conventional semicon-
a probe where they are thermalized. The next step is to re-inject ductor theory and that the re-injection process (in-scattering)

Fig. 6. Conduction and first two subband minima, as a function of position in Fig. 7. Total 3-D electron density, n(x; z ), in the on-state. The thin silicon
the on-state (V = V = 0:4 V). Note that the conduction band is a function body is volume inverted, and the electron density goes to zero at the top and
bottom oxide/silicon interface (0, 3 nm). Quantum effects due to confinement
of both x and z , while the subband minima are functions of x alone. The lower
subband minima is for the unprimed valleys with two fold degeneracy, while the are accurately captured by nanoMOS.
higher minima is for the primed valleys with four fold degeneracy.

from the probes to the device mixes different subband popula-

tions, thus capturing the effects of inter subband and inter valley

Several studies of nanoscale MOSFET device physics and de-
sign issues that make use of nanoMOS simulations have already
been published [25]–[27], [19]. Our purpose in this section is
to present some simulation results that illustrate the capabili-
ties of the program. The simulated device (recall Fig. 1) is an
idealized structure for which we do not treat the flared-out S/D
contacts that would be present in an actual device. The n-type
source and drain regions are doped at cm , the channel
is intrinsic, and the S/D junctions are abrupt. No gate-to-S/D Fig. 8. Areal electron density (cm ) along the channel is plotted in the
off-state (V = 0; V = 0:4 V) from both the quantum (solid line) and the
overlap is assumed. The oxide thickness is 1.0 nm for both top classical (dashed line) ballistic transport models at room temperature. Quantum
and bottom gates, and the silicon film thickness is 3.0 nm. The mechanical tunneling through the source-to channel barrier results in a higher
gate workfunction was set to 4.22 eV in order to produce an channel charge density in case of the quantum model. Charge neutrality is
achieved in the S/D regions in both cases.
off-current of A/ m, consistent with the 2016 node of
the International Technology Roadmap for Semiconductors [1].
The power supply is V and the gate length (equals quantum mechanical tunneling into the source-channel barrier
the channel length as the S/D junctions are abrupt) is 10 nm. increases the carrier density in the channel. A careful examina-
Fig. 6 is a plot of the first two subband minima versus po- tion of the figure also reveals that the quantum mechanical car-
sition in the on-state. Also shown is a plot of the conduction rier density just outside the channel is slightly reduced, an effect
band minimum versus position. Note that the conduction band first observed by Willander [37]. Fig. 9 compares the
varies in two dimensions (in the -direction along the channel versus characteristics from ballistic quantum and classical
from the source to the drain as well in the -direction across nanoMOS simulations. As expected, quantum mechanical tun-
the thickness of the silicon body). The subband minima, how- neling increases the off-current, but note that it also decreases
ever, are determined by solving the Schrödinger equation in the the on-current. In the on-state, self-consistent gate electrostatics
direction, and they vary only in the -direction. This uncou- tries to maintain a charge density of , at
pled mode space approach is what allows us to treat transport in the top of the source-to-channel barrier irrespective of the so-
each subband with a 1-D Green’s function approach. lution scheme (quantum or classical) used to simulate the de-
Fig. 7, a plot of the electron density within the device under vice. In quantum simulations, some of this charge is due to
on-state conditions, shows the quantum confinement of carriers tunneling electrons (with channel directed energies below the
in the -direction. The profile varies approximately as source-to-channel barrier). Since these electrons are evanescent,
in the S/D regions, which indicates that most electrons reside they carry less current than their thermionic counterparts. There-
in the first subband (primed and unprimed). Classical Boltz- fore, the on-current from classical simulations is higher than
mann and NEGF simulations are compared in Fig. 8, which that from quantum simulations as shown in Fig. 9. Classically,
shows the integrated electron density (per cm ) versus posi- if one views the on-current as a product of a 2-D electron den-
tion in the off-state ( V). As expected, sity multiplied by a velocity, it looks like the electron velocity

space approach for the -direction, the effect of quantum

confinement on the threshold voltage is included in both
simulations. Fig. 10(a) shows that even for a 10 nm channel
length, MOSFETs are expected to behave classically. Quantum
mechanics increases the threshold voltage and decreases the
on-current at a given threshold voltage, but no quantum oscil-
lations are observed. In Fig. 10(b), we compare the common
source characteristics with and without scattering. The Büttiker
probe strength has been adjusted to obtain a low field mobility
of 50 cm /V-s in the heavily doped source drain regions. In
the channel, a doping dependent mobility model is used to
adjust the probe self-energy. It is clear from Fig. 10(b) that
the on-current is significantly reduced due to scattering in the
Fig. 9. I versus V characteristics for the model device from both the
quantum (solid line) and classical (dashed line) ballistic transport models. The heavily doped S/D extensions. This reduction in the on-current
ballistic off-current is higher from the quantum model due to source-to-channel once scattering is turned on is primarily due to
1) source parasitic resistance which degrades the effective
gate to source voltage;
2) change in the shape of the distribution (toward equi-
librium Fermi–Dirac) due to reflections, which causes
a reduction in the injection velocity at the top of the
source-to-channel barrier.
The presence of a parasitic drain resistance does not degrade the
on-current as severely as the source parasitic resistance because
it only affects the drain to source voltage and not the gate to
source voltage.
The conduction bandstructure for silicon is composed of
three sets of ellipsoids. The first set of ellipsoids with their
longitudinal effective mass oriented along the confinement di-
rection, gives rise to the so-called unprimed series of subbands
(a) and the two other sets of ellipsoids that have their transverse
effective masses oriented along the confinement direction
give rise to the primed series of subbands. These subbands,
with their anisotropic effective masses, are independently
treated within nanoMOS-2.5. In the ballistic limit, there is no
mixing between electrons from different subbands, but once
scattering is turned on, electrons populations from different
subbands mix as a result of in-scattering from Büttiker probes.
The ballistic, on-state source to drain transmission coefficient
versus energy is plotted in Fig. 11. The locations of the first
unprimed (dashed line) and the first primed (dash-dot line)
subband minima are also indicated in Fig. 11. Note that the
net transmission coefficient (Fig. 11) includes contributions
from all subbands (both primed and unprimed), at each energy.
Therefore, the net transmission coefficient begins to increase
Fig. 10. (a) I versus V characteristics for the model device from both below the first subband minimum (due to source-to-channel
the quantum (solid line) and classical (dashed line) ballistic transport models.
The ballistic on-current is lower from the quantum model as compared to the tunneling) then smoothly approaches unity above the first
classical model. (b) The I versus V characteristics for the model device subband minimum, and increases once again at higher energies
from both the quantum ballistic (dashed line) and quantum dissipative (solid increases. At high energies, the net transmission coefficient
line) transport models.
approaches three, because we chose to include one subband
from each conduction band valley for our simulations (valley
(which is derived from the current and the 2-D electron density) degeneracy is not accounted for in Fig. 11).
is lower for quantum simulations when compared to classical It is instructive to look at the floating boundary conditions
simulations, in the on-state. more closely. To explain the floating boundary condition, we
The classical and quantum mechanical ballistic common modify our idealized 3 nm body DG MOSFET structure by ap-
source characteristics are compared in Fig. 10(a). Note that the pending heavily doped ( cm ) regions to the left
term classical or quantum refers to the treatment of transport and right ends of the device. The profile of the first subband from
along the channel (the -direction). Because we use a mode the source to the drain, for the modified device structure under

Fig. 11. Ballistic source-to-drain transmission is plotted as a function of
longitudinal energy (E ) in the on-state. The top of the source-to-channel
barrier for both the unprimed (dashed) and the primed (dashed with dots)
subbands is also indicated. Note that the transmission is nonzero for longi-
tudinal energies below the top of the subband barrier (tunneling) and increases
smoothly to unity. The primed subband transmission goes to two instead of one
due to contributions from two sets of valleys.

Fig. 13. (a) Energy resolved Local Density of States (LDOS) is plotted along
the channel in the on-state from the quantum ballistic transport model. Light
areas imply a low density of states, while dark regions indicate a high density of
states. Coherent oscillations in the LDOS are the result of reflections from the
barrier. Non-zero LDOS in the forbidden region (below the subband energy),
(a) leads to tunneling through the source-to-channel barrier. The LDOS from the
primed (blue line) band is higher than the unprimed (red line) due to degeneracy.
(b) The energy resolved LDOS is plotted along the channel in the on-state from
the quantum diffusive transport model. Coherent oscillations in the LDOS are
washed out as a result of scattering. A small, potential drop in the source and
drain regions is also discernible.

barrier gives rise to strong reflections, which maintain a near

equilibrium distribution in the regions even when a large
bias is applied to the drain and gate. Therefore, a fixed poten-
tial boundary condition based on charge neutrality can be used
when solving Poisson’s equation for this modified device. The
regions maintain a near equilibrium distribution even under
On examining the subband profile [Fig. 12(a)] of the modified
(b) device at high gate bias, it is clear that the subband is unchanged
Fig. 12. (a) Subband profile, E (x), for the device in Fig. 1 with an n in the heavily doped regions, but floats to a lower value in
source extension added under low (solid line) and high (dashed line) gate bias the source region of the intrinsic device. This observation
(V = 0 and 0:6 V) at a drain bias of 0.6 V. The potential (subband) floats can be explained by examining Fig. 3. At low gate biases, both
to a lower energy in the n region, but remains unchanged in the n region.
(b) Subband profile, E (x), for the device in Fig. 1 with the n source the positive and the negative halves of the distribution in the
extension and fixed boundary conditions (dashed line) and without the n source are predominantly filled by the source Fermi level. When
extension and floating boundary conditions (solid line) at a high gate voltage. the gate bias is increased to higher and higher values, the number
of source injected electrons reflected off the source-to-channel
a high drain bias and for a high and low gate bias is ( barrier is reduced because the source-to-channel barrier height
and 0 V, V) is shown in Fig. 12(a). The presence of decreases. Although nearly one half of the distribution is un-
the heavily doped regions, creates a large potential barrier occupied at the source, 2-D electrostatics requires that charge
for electrons injected from the source and drain contacts. This neutrality has to be maintained (integrated doping equals the

drain bias, however, confinement is lost near the drain and nu-
merous modes much be treated. This makes a real-space dis-
cretization more suitable [13], [31]. To treat the flared our con-
tact in a FD UTB SOI MOSFET, mode coupling must also be
included. The transition from the thick contact to the thin body
could introduce resistance. For this problem one could use a
coupled mode space approach [8] or a real-space discretization.
The treatment of scattering in nanoMOS is a simple, phe-
nomenological one. A well-defined prescription for including
scattering exists, but the computational burden rapidly gets out
of hand if one resolves scattering between transverse modes
[36]. When using a phenomenological approach, one must
be careful to mimic the correct physics. Our use of Büttiker
probes that relax the energy of scattered carriers mimics the
Fig. 14. Energy resolved electron density is plotted along the channel in the
on-state from the quantum ballistic transport model. In the heavily doped S/D relaxation of longitudinal energy that occurs in MOSFETs
regions, multiple subbands are occupied. However much of the current in the [19]. The strength of the NEGF approach lies in its ability to
channel is from the first subband. The ballistic source injected charge that treat quantum confinement, reflection, and tunneling and our
propagates to the drain without any energy relaxation is clearly observed.
phenomenological approach provides nanoMOS with a way
to include the first order effects of scattering. Monte Carlo
2-D hole density) at every point within the source. To achieve simulations can include a much more detailed treatment of
charge neutrality, the electron density residing in the positive scattering, but it is necessary to include quantum effects phe-
velocity states nearly doubles between the low and high gate nomenologically [33], [29]. The two approaches complement
bias conditions. To accomodate this increased charge the sub- each other; the NEGF approach is preferable when quantum
band (potential) floats to a lower energy as seen in Fig. 12(a) transport is the first order issue and Monte Carlo simulation
(dotted line). when scattering is the key issue.
Carriers injected from the heavily doped region are pre- The nanoMOS program was written in a scripting language
dominantly backscattered by the built in barrier, so the poten- (Matlab) to permit rapid development and debugging. Since
tial in this region is unchanged with gate bias. Now, if we re- much of the computation occurs in compiled matrix routines,
move the regions but use the floating boundary condition the performance penalty is slight. For larger problems, tool-
to terminate the regions, we observe the potential behavior boxes to parallelize the scripts are being developed [11]. When
plotted in Fig. 12(b) (solid line). Fig. 12(b), demonstrates that the time comes for a production CAD program, nanoMOS may
the subband under bias for the device with floating boundaries have to be re-written in C or Fortran, but the scripting language
( - - ) is identical to that of the device with fixed boundaries approach greatly facilitates program development and provides
( - -- - ) within the region of interest ( - - ). adequate performance for use as a research tool.
This indicates that the floating boundary condition does capture
the effect of coupling a ballistic device to a scattering contact.
The energy-resolved local density of states (LDOS) under
ballistic on-state conditions is plotted in Fig. 13(a). The states The program, nanoMOS 2.5, simulates quantum transport
below the barrier are due to tunneling, and the strong influence in fully-depleted, ultra-thin-body SOI MOSFETs with an effi-
of quantum mechanical reflections is apparent. The LDOS for ciency that permits its use on a workstation. We have described
the higher (primed) subband is larger (darker in the grayscale the methods and approaches used in the program as well as the
plot) because of the four-fold degeneracy of the conduction band simplifying approximations that deliver its computational effi-
valleys. Fig. 13(b) shows the same plot in the presence of scat- ciency. The NEGF approach used in nanoMOS provides a solid
tering, which is seen to reduce the contrast of the interference base for simulating electronic devices at the nanoscale which
pattern. Finally, Fig. 14 shows the energy resolved ballistic elec- can be extended to increase the fidelity of the physics (e.g., im-
tron density which results in filling up the LDOS according to proving the bandstructure and the treatment of scattering) or to
the Fermi levels in the source and drain contacts. Carrier tun- treat much different devices such as carbon nanotube transis-
neling below the source barrier is observed as well as oscilla- tors or molecular conductors. The nanoMOS program is a step
tions in the source and drain, which are due to reflections from toward a new generation of simulation tools that will allow de-
the barrier. vice engineers to explore new classes of electronic devices. The
program is available for use through the WWW or for access to
VI. DISCUSSION its source code [21].
Although the nanoMOS program has proven to be a useful
tool for exploring device physics and design, it is necessarily APPENDIX
based on a number of simplifying assumptions. The mode space GREEN’S FUNCTION FORMULATION
approach is particularly efficient for fully-depleted (FD), ultra- Although we described nanoMOS from a wave function per-
thin-body (UTB), SOI MOSFETs, for which only a few, uncou- spective in the text, nanoMOS is actually based on a Green’s
pled modes need to be treated. In a bulk MOSFET under high function formulation of the same problem. The nonequilibrium

Green’s function formalism offers advantages when more com- The velocities at the injection and exit points are
plicated geometries or basis functions are considered and when
scattering needs to be included in a more rigorous manner. In
this Appendix we translate the wavefunction approach of the
text into the NEGF formalism. Several references provide a (A8a)
fuller discussion of the method [9].
To simulate a MOSFET, we need to compute the electron
density and the current. In the wavefunction picture, (9), we
need at each node, where is determined by the (A8b)
Greens function according to (19). To get , consider
the generalization Using these results, we find

(A1) (A9)

where Since

.. (A2)
we have
is an vector giving the value of the wavefunction at each
node, and the superscript, , denotes the Hermetian transpose. which can be expressed as
The matrix product in (A2) is

.. .. (A3) To summarize, after computing the Green’s function, we find
. . the transmission coefficient from (A12), which is then used with
(11b) to find the current. Note that (A12) can be generalized to
By defining an matrix find the transmission between any two probes as

.. (A4) which is useful when Büttiker probes are used to include scat-
. . tering.

[where is the self-energy describing the connection to the
[1] International Technology Roadmap for Semiconductors (ITRS) [On-
source contact as given by (17a)] we can use (A1) to generalize line]. Available: http://public.itrs.net
the spectral function to [2] A. Bachtold, P. Hadley, and C. Dekker, “Logic circuits with carbon nan-
otube transistors,” Science, vol. 294, p. 1317, 2001.
(A5) [3] J. S. Blakemore, “Approximations for Fermi–Dirac integrals, especially
the functions F ( ) to describe electron density in a semiconductor,”
Solid-State Electron., vol. 25, p. 1067, 1982.
which is an matrix. Note that the term, , [4] M. Büttiker, “Four-terminal phase coherent conductance,” Phys. Rev.
in (8) is . Instead of the electron density as in (9), Lett., vol. 57, p. 1761, 1986.
we find the density matrix due to source injection as [5] M. Cahay, M. McLennan, S. Datta, and M. S. Lundstrom, “Importance
of space-charge effects in resonant tunneling devices,” Appl. Phys. Lett.,
vol. 50, p. 612, Mar. 9, 1987.
(A6) [6] S. Datta, Electronic Transport in Mesoscopic Systems. Cambridge,
MA: Cambridge Univ. Press, 1997.
[7] P. Damle, A. W. Ghosh, and S. Datta, “First principles analysis of molec-
Analogous expressions exist for injection from the drain, so the ular conduction using quantum chemistry software,” Chem. Phys., vol.
total density matrix is . In summary, from the 281, p. 171, 2002.
[8] P. Damle, A. W. Ghosh, and S. Datta, “Nanoscale device modeling,”
Green’s function, (20), we find the density matrix whose diag- Molec. Nanoelectron., 2002.
onal elements give the electron density at each node. [9] S. Datta, “Nanoscale device modeling: The Green’s function method,”
To find the current, (11b) still applies, but we need to express Superlatt. Microstruct., vol. 28, p. 253, 2000.
[10] V. Derycke, R. Martel, J. Appenzeller, and Ph. Avouris, “Carbon nan-
the transmission coefficient, , in terms of the Green’s otube inter- and intramolecular logic gates,” Nano Lett., vol. 1, p. 453,
function. Assuming injection from the source as in (5), we find 2001.
the incident and transmitted currents as [11] Standard (2003). [Online]. Available: http://www.ece.purdue.edu/celab
[12] A. Asenov, A. R. Brown, and J. R. Watling, “The use of quantum poten-
tials for confinement in semiconductor devices,” in Proc. 5th Int. Conf.
(A7a) Modeling Simulation Microsyst., San Juan, PR, Apr., 21–25 2002, pp.
and 490–493.
[13] D. Jovanovic and R. Venugopal, Proc. 7th Int. Workshop Computat.
(A7b) Electron.. Glasgow, U.K., 2000.

[14] Standard (2003). [Online]. Available: http://www-hpc.jpl.nasa.gov/ Ramesh Venugopal was born in Chennai, India. received the Ph.D. degree in
PEP/gekco/nemo/nemo.html electrical and computer engineering from Purdue University, West Lafayette,
[15] L. V. Keldysh, “Diagram technique for nonequilibrium processes,” Sov. IN, in 2003.
Phys. JETP, vol. 20, p. 1018, 1965. He is now with Texas Instruments, Dallas, TX and his research interests in-
[16] J. Knoch, B. Lengeler, and J. Appenzeller, “Quantum simulations of clude device physics, design, and simulation.
an ultrashort channel single-gated n-MOSFET on SOI,” IEEE Trans.
Electron Devices, vol. 49, p. 1212, July 2002.
[17] R. Lake and S. Datta, “Nonequilibrium Green’s function method applied
to double barrier resonant-tunneling diodes,” Phys. Rev B, vol. 45, p.
6670, 1992. Sebastien Goasguen (S’99–M’01) was born in
[18] D. C. Langreth, “Linear and nonlinear electron transport in solids,” in Rennes, France, on March 22, 1974. He received
NATO Advanced Study Instruction Series B. New York: Plenum, 1976, the B.S. degree in electrical engineering from
vol. 17, p. 3. the Polytechnic Institute of Toulouse, France, in
[19] M. S. Lundstrom and Z. Ren, “Essential physics of nanoscale MOS- 1997, the M.S. degree (with honors) in electronics
FETs,” IEEE Trans. Electron Devices, vol. 49, p. 133, Jan. 2002. research, from King’s College of London, London,
[20] P. L. McEuen, M. S. Fuhrer, and H. Park, “Single-walled carbon nan- U.K., in 1998, and the Ph.D. degree in electrical
otube electronics,” IEEE Trans. Nanotechnol., vol. 1, p. 78, Mar. 2002. engineering from Arizona State University, Tucson,
[21] Nanotechnology Simulation Hub (2003). [Online]. Available: in 2001.
http://www.nanohub.purdue.edu His main area of interest was global modeling of
[22] M. Nekovee, B. Geurts, H. M. J. Bootsand, and M. F. H. Schuurmans, microwave active circuits including the use of neural
“Failure of extended-moment-equation approaches to describe ballistic networks and wavelet based numerical methods. In September 2001, he joined
transport in submicrometer structures,” Phys. Rev. B, vol. 45, p. 6643, Purdue University, West Lafayette, IN, as a Post-Doctoral Research Associate
1992. and became a Visiting Professor in August 2002. Since September 2002, he has
[23] M. Paulsson, F. Zahid, and S. Datta, “Resistance of a molecule,” Tech. acted as Technical Director for the Network for Computational Nanotechnology,
Rep., www.arxiv.org/abs/cond-mat/0 208 183, 2002. in charge of high-performance computing solution to nanotechnology compu-
[24] C. S. Rafferty, B. Biegel, Z. Yu, M. G. Acona, J. Bude, and R. W. Dutton, tational challenges and the implementation of a cyber-infrastructure.
“Multidimensional quantum effects simulation using a density gradient
model and script level programming technique,” in Proc. SISPAD’98,
1998, p. 137.
[25] Z. Ren, R. Venugopal, S. Datta, M. S. Lundstrom, D. Jovanovic, and J.
G. Fossum, “The ballistic nanotransistor: A simulation study,” in IEDM Supriyo Datta (F’96) was born on February 2, 1954.
Tech. Dig., 2000, p. 715. He received the B.Tech. degree from the Indian Insti-
[26] Z. Ren, “Nanoscale MOSFETs: Physics, simulation, and design,” Ph.D. tute of Technology, Kharagpur, in 1975 and the Ph.D.
dissertation, Purdue Univ., West Lafayette, IN, Dec. 2001. degree from the Univeristy of Illinois, Urbana-Cham-
[27] Z. Ren, R. Venugopal, S. Datta, and M. S. Lundstrom, “Examination of paign, in 1979.
design and manufacturing issues in a 10 nm double gate MOSFET using In 1981, he joined Purdue University, West
nonequilibrium Green’s function simulation,” in IEDM Tech. Dig., Dec. Lafayette, IN, where he is currently the Thomas
3–5, 2001, p. 107. Duncan Distinguished Professor in the School of
[28] J. H. Rhew and M. S. Lundstrom, “Benchmarking macroscopic transport Electrical and Computer Engineering. He is the
models for nanotransistor TCAD,” J. Computat. Electron., 2002. author of Surface Acoustic Wave Devices (Engle-
[29] L. Shifren, A. Akis, and D. K. Ferry, “Correspondence between quantum wood Cliffs, NJ: Prentice-Hall, 1986), Quantum
and classical motion: Comparing bohmian mechanics with a smoothed Phenomena (Reading, MA: Addison-Wesley, 1989), and Electronic Transport
effective potential,” Phys. Lett. A, vol. 274, p. 75, 2000. in Mesoscopic Systems (Cambridge U.K.: Cambridge, 1995). His current
[30] A. Svizhenko, M. Anantram, and T. Govindan, “The role of scattering in research interests are centered around the physics of nanostructures and
nanotransistors,” IEEE Trans. Electron Devices, vol. 50, pp. 1459–1466, includes molecular electronics, nanoscale device physics, spin electronics and
June 2003. mesoscopic superconductivity.
[31] A. Svizhenko, M. Anantram, T. Govindan, B. Biegel, and R. Venu- Dr. Datta received the NSF Presidential Young Investigator Award and the
gopal, “Nano-transistor modeling: Two dimensional Green’s function IEEE Centennial Key to the Future Award in 1984, the Frederick Emmons
method,” J. Appl. Phys., vol. 91, p. 2343, 2002. Terman Award from the ASEE in 1994, and shared the SRC Technical
[32] W. Tian, S. Datta, S. Hong, R. Riefenberger, J. I. Henderson, and C. P. Excellence Award, 2001 and the IEEE Cledo Brunetti Award, 2002. He is a
Kubiak, “Conductance spectra of molecular wires,” J. Chem. Phys., vol. Fellow of the American Physical Society (APS) and the Institute of Physics
109, p. 2874, 1998. (IOP).
[33] H. Tsuchiya and U. Ravaioli, “Particle Monte-Carlo simulation of
quantum phenomena in semiconductor nanostructures,” J. Appl. Phys.,
vol. 89, p. 4023, 2001.
[34] F. Venturi, R. K. Smith, E. C. Sangiorgi, M. R. Pinto, and B. Ricco,
“A general purpose device simulator coupling Poisson and Monte Carlo Mark S. Lundstrom (F’94) received the B.E.E. and
transport with applications to deep submicron MOSFETs,” IEEE Trans. M.S.E.E. degrees from the University of Minnesota,
Computer-Aided Design, vol. 8, p. 360, Apr. 1989. Minneapolis, in 1973 and 1974, respectively, and
[35] R. Venugopal, Z. Ren, S. Datta, M. S. Lundstrom, and D. Jovanovic, the Ph.D. degree from Purdue University, West
“Simulating quantum transport in nanoscale MOSFETs: Real versus Lafayette, IN, in 1980.
mode space approaches,” J. Appl. Phys., vol. 92, p. 3730, 2002. He is the Scifres Distinguished Professor of
[36] R. Venugopal, M. Paulsson, S. Goasguen, S. Datta, and M. Lundstrom, Electrical and Computer Engineering at Purdue
“A simple quantum mechanical treatment of scattering in nanoscale tran- University, West Lafayette, where he also directs the
sistors,” J. Appl. Phys., vol. 93, p. 5613, 2003. NSF Network for Computational Nanotechnology.
[37] Y. Fu, M. Karlsteen, M. Willander, N. Collaeert, and K. De Meyer, Before attending Purdue, he worked at Hewlett-
“Quantum transport and I-V characteristics of quantum sized field-ef- Packard Corporation, Loveland, CO, on integrated
fect-transistors,” Superlatt. Microstruct., vol. 24, no. 2, p. 111, 1998. circuit process development and manufacturing. His current research interests
center on the physics of semiconductor devices, especially nanoscale transis-
tors. His previous work includes studies of heterostructure devices, solar cells,
heterojunction bipolar transistors, and semiconductor lasers. During the course
of his Purdue career, he has served as Director of the Optoelectronics Research
Zhibin Ren was born in China. He received the Ph.D. degree in electrical en- Center and Assistant Dean of the Schools of Engineering.
gineering from Purdue University, West Lafayette, IN, in 2001. Dr. Lundstrom received several awards for teaching and research, most re-
He is currently with IBM Corporation, Yorktown Heights, NY. His research cently the 2002 IEEE Cledo Brunetti Award and the 2002 Semiconductor Re-
interests are primarily centered on device physics, modeling, and experimental search Corporation Technical Achievement Award for work on nanoscale elec-
characterization of MOSFETs. tronics. He is a Fellow of the American Physical Society.

View publication stats