Universit di Pisa

S. Scaluafeiii, u. Cuiatola, u. Iannaccone, !"#$%& ()*+&"), )- &.$ /)*&012,, 3#2,45)#& 67+2&"), 2,8
9)"44),:(%.#;8",<$# 67+2&"), -)# =2,)4%2*$ >?(@634 , IEEE Tiansactions on Election Bevices, !", 11,
pp.29u1-29u9 (2uu7).

Direct Solution of the Boltzmann
Transport Equation and Poisson
Schrdinger Equation for Nanoscale
MOSFET
#$%&'() #+',-'&%../
Bialog Semiconuuctoi umbB, Kiichheim untei Teck-Nabein
0/,1%.$) 23.'$),'
NXP Semiconuuctois, Leuven
0/34%55% 6'(('++)(%
Bipaitimento ui Ingegneiia uell'Infoimazione: Elettionica, Infoimatica, Telecomunicazioni,
0niveisit ui Pisa


IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 54, NO. 11, NOVEMBER 2007 2901
Direct Solution of the Boltzmann Transport
Equation and PoissonSchrdinger Equation
for Nanoscale MOSFETs
Stefano Scaldaferri, Gilberto Curatola, and Giuseppe Iannaccone
AbstractWe propose an efcient and fast algorithm to solve
the coupled PoissonSchrdinger and Boltzmann transport equa-
tions (BTE) in two dimensions. The BTE is solved in the relaxation
time approximation within each subband obtained from the direct
solution of the Schrdinger equation. The proposed approach,
considering a subband-based transport formalism, allows to fully
explore the entire range from drift-diffusion to ballistic regime in
nanoscale eld-effect transistors. Quantum effects are also fully
taken into account by the direct solution of the Schrdinger
equation. The model is implemented in the NanoTCAD2D device
simulator and used to study the device performance of a 25-nm
channel-length MOSFET. The inuence of scattering on the elec-
tron distribution function and on device characteristics is analyzed
in detail.
Index TermsBallistic transport, Boltzmann transport
equation, modeling, nanoscale MOSFET device, Schrdinger
equation.
I. INTRODUCTION
T
HE CONTINUOUS scaling of device dimensions to meet
the requirements in terms of silicon area occupation and
device performance that is set by the Semiconductor Industry
has led to a rapid increase of the complexity of CMOS tech-
nology and, consequently, of the TCAD tools needed to obtain
reliable simulations.
Monte Carlo techniques for the solution of the Boltzmann
transport equation (BTE) offer a high degree of accuracy and
allow to properly describe the essential physics of nanoscale
devices even in strong off-equilibrium conditions [1][3]. Un-
fortunately, the huge computational resources required are
among the reasons why the vast majority of routinely per-
formed simulations are still based on simplied approaches,
like the drift-diffusion (DD) or hydrodynamic (HD)-type
models [4], [5].
Manuscript received September 19, 2006; revised July 24, 2007. This work
was supported by the European Commissions Information Society Technolo-
gies Programme under PULLNANO Project Contract IST-026828. The review
of this paper was arranged by Editor S. Datta.
S. Scaldaferri is with Dialog Semiconductor GmbH, 73230 Kirchheim unter
Teck-Nabern, Germany.
G. Curatola is with NXP Semiconductors, 3001 Leuven, Belgium
(e-mail: gilberto.curatola@nxp.com).
G. Iannaccone is with the Dipartimento di Ingegneria dellInformazione:
Elettronica, Informatica, Telecomunicazioni, Universit di Pisa, 56122 Pisa,
Italy (e-mail: g.iannaccone@iet.unipi.it).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TED.2007.906927
However, all the simplied models (DD and HD) rely on
strong approximations on the carrier distribution function,
which is necessary to make the problem analytically man-
ageable. Such approximations may invalidate the results of
simulations in the case of ultrascaled devices and/or strong off-
equilibrium conditions; moreover, the BTE [6][8] is strictly
semiclassical and does not allow to take into account quantum
effects which become predominant as the dimensions of the
device enter the sub-100-nm regime.
Even if, recently, some Monte Carlo tools [9][11] have been
modied in order to include the quantum effects by means
of approximate models [15][18] or by direct coupling the
Schrdinger equation to the Boltzmann equation, the com-
puting time is still excessive for industrial application, and
therefore, such approaches are conned to the computational
electronics community.
We will show in the following that in the case of conned
systems, instead, where the dimensionality is reduced due to
the quantum connement, it is possible to directly solve the
BTE (i.e., without a statistical approach) and to couple it to
the Schrdinger equation without large penalty in the compu-
tational efciency of the algorithm.
Previous attempts in this direction have been presented in the
literature by Goldsman et al. [12][14]. We want to stress here
that our main goal is to develop a 2-D simulator for nanoscale
MOSFETs that satises the following requirements.
1) It self-consistently solves the coupled Poisson
SchrdingerBoltzmann equation in two dimensions
without a priori assumptions on the carrier distribution
functions.
2) Computing time is maintained comparable to the com-
mercial DD or HD tools.
3) It accounts for mass anisotropy (the Schrdinger equation
is solved three times for electrons and twice for holes).
4) It can be applied to a wide variety of semiconductor
materials, like Si, Ge, AlGaAs, SiGe, and strained-Si.
5) It includes a fully ballistic model for electrons and
holes.
6) Differently from other approximate quantum models
proposed in the literature, our approach allows to have
full information on the subband structure. Therefore, it
permits a detailed evaluation of the different scattering
rates in each subband and among different subbands
given the knowledge of the envelope wavefunctions at
each position along the channel.
0018-9383/$25.00 2007 IEEE
2902 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 54, NO. 11, NOVEMBER 2007
Fig. 1. Schematic representation of an nMOSFET: x is the direction of
quantum connement, and y is the direction of electron transport. We suppose
that the device characteristics are uniform along the z-direction (not shown).
As far as scattering is concerned, we present here the results
of simulations performed in the simplied assumption of neg-
ligible intersubband scattering which gives us the freedom of
treating electron transport in each subband independently. Ad-
ditional results obtained with a more detailed scattering model
will be presented elsewhere [19]. Here, we are mainly interested
in demonstrating the feasibility of the proposed approach.
This paper is organized as follows. In Section II-A, we
present the algorithm for the solution of the BTE, assuming a
known subband prole and in the relaxation time approxima-
tion (RTA). For simplicity, we also assume a scattering time
independent of energy, but the algorithm can be easily general-
ized. The full self-consistent PoissonSchrdingerBoltzmann
approach is then described in detail in Section II-B.
In Section III, we highlight the numerical issues connected
with the discretized form of the BTE.
Finally, in Section IV, we present the results of simulations
performed on a benchmark nanoscale transistor, i.e., the well-
tempered MOSFET with an effective channel length of 25 nm
proposed by Antoniadis et al. [20]. Electrical characteristics
are presented for a whole range of operating regimes ranging
from DD to fully ballistic. Furthermore, details on the electron
distribution function under strong off-equilibrium conditions
are given.
A metric that allows to quantify the contribution of scattered
electrons on the total current is then dened and evaluated
for different scattering conditions. Conclusion nally ends this
paper.
II. MODEL
Let us consider the nMOSFET shown in Fig. 1, where x
represents the direction of quantum connement, and y repre-
sents the direction of electron transport. Electrons moving from
source to drain must overcome an energy barrier whose height
is modulated via the voltage applied to the gate electrode. As a
consequence of gate-induced vertical eld, carriers are conned
at the oxide/silicon interface in condition of strong inversion.
This causes energy quantization in the transverse direction.
The position (in energy) of the available states obtained
from the solution of the Schrdinger equation varies along the
current ow direction due to the varying conning potential
along the channel. Available states are therefore grouped in
Fig. 2. Subband prole for an applied external voltage V
ds
. Fermi levels at
the boundaries are imposed by the source and drain contacts.
different subbands. A generic subband prole along the current
ow direction is shown in Fig. 2, where E
SB
(y) represents the
subband edge at coordinate y.
The Fermi levels at the two boundaries are set by the source
and drain contacts E
FS
and E
FD
, respectively. Indeed, we
assume that left- and right-moving electrons are injected in the
channel with the chemical potential of the emitting reservoir.
Considering the steady-state Boltzmann equation, for each
subband, we assume that the kinetic term along x is imposed
by energy quantization, and therefore, the distribution function
in each subband only depends on the y coordinate and of the
wave vector along directions y and z.
Under these assumptions, the steady-state Boltzmann equa-
tion can be written as
v
y
f(y, k)
y

q
y

f(y, k)
k
y
= (f) (1)
where k = k
y

y + k
z

z represents the electron wave vector in

the transport plane, q is the electron charge, is the reduced
Planck constant, v
y
is the y-component of the electrons group
velocity,
y
is the y-component of electric eld, and f is
the electron distribution function in the phase space in the
considered subband.
The electric eld in the transport direction can easily be
evaluated given the knowledge of the subband prole and reads

y
=
1
q
dE
SB
(y)
dy
. (2)
The operator (f) at the right-hand side of (1) represents the
collision term, which can be expressed as
(f) =

S(k

; k)f(y, k

) [1 f(y, k)] +

S(k; k

)f(y, k) [1 f(y, k

)] (3)
where S(k

; k) represents the scattering rate from state k

to
state k.
SCALDAFERRI et al.: DIRECT SOLUTION OF THE BTE AND THE PS EQUATION FOR NANOSCALE MOSFETs 2903
A. One-Subband Transport
Under the RTA hypothesis, the scattering term can be simpli-
ed in the form
(f) =
f
eq
[E
fn
(y), k] f(y, k)
(k)
. (4)
The term f
eq
[E
fn
(y), k] represents an equilibrium distribution
function for the 2-D electron gas, i.e., a FermiDirac distri-
bution. E
fn
(y) represents the quasi-Fermi level at coordinate
y, and (k) is the relaxation time which represents the time
required for the carrier population to reach an equilibrium
distribution (thus, described by a FermiDirac distribution
function) as a consequence of the scattering processes.
The equilibrium distribution f
eq
can conventionally be
expressed as
f
eq
[E
fn
(y), k] =
_
_
1 + exp
E
SB
(y)+

2
k
2
y
2m
y
+

2
k
2
z
2m
z
E
fn
(y)
k
B
T
_
_
1
(5)
where m
y
and m
z
are the electron effective masses in
y- and z-directions, k
B
is the Boltzmann constant, and T is the
temperature.
The RTA approximation must, however, ensure charge con-
servation, as expressed by the following relation (in the present
form, the following relation strictly holds when the intersub-
band scattering is neglected):
_
k
1
2
2
f(y, k)dk =
_
k
1
2
2
f
eq
_
E
fn(y)
, k
_
dk. (6)
Previous simulations performed on the same device have shown
indeed that in strong inversion conditions, almost 95% of the
total carrier concentration resides in the rst subband [21]. This
result justies the approximation of negligible intersubband
scattering and gives us strong condence in the simulation
results.
The left side of (6) represents the effective carrier concen-
tration n(y) associated to the considered subband, while the
right side represents the equivalent concentration at equilibrium
which is a known nonlinear function of the electron quasi-Fermi
level and is given by
n
eq
(y)=
g
e
k
B
T

m
y
m
z

2
ln
_
1 + exp
E
fn
(y) E
SB
(y)
k
B
T
_
(7)
where g
e
represents the degeneracy of each group of equivalent
conduction band minima of silicon (in our case, g
e
= 2).
In particular, the charge-conservation property gives us an
important relation between the equilibrium and quasi-Fermi
levels out of equilibrium. Inverting (6) with the use of (7) allows
us to express the quasi-Fermi level E
fn
(y) in the considered
subband as
E
fn
(y) = g (n(y)) (8)
where g is a known nonlinear function.
In the rest of this paper, we will assume for simplicity
a relaxation time constant with energy. The extension to an
energy-dependent relaxation time is, however, straightforward.
Under this hypothesis, (1) can be rewritten as follows:
v
y
f(y, k)
y

q
y

f(y, k)
k
y
=
f
eq
(E
fn
(y), k) f(y, k)

.
(9)
If we dene
F(y, k
y
)
_
k
z
1
2
2
f(y, k)dk
z
F
eq
(E
fn
(y), k
y
)
_
k
z
1
2
2
f
eq
(E
fn
(y), k) dk
z
(10)
we can integrate both terms of (9) in k
z
and obtain
v
y
F(y, k
y
)
y

q
y

F(y, k
y
)
k
y
=
F
eq
(E
fn
(y), k
y
)F(y, k
y
)

.
(11)
The term F(y, k
y
) represents the derivative of the electron
distribution function with respect to the y-component of the
wave vector. Indeed, by denition
_
F(y, k
y
)dk
y
= n(y). (12)
The computation of F
eq
(E
fn
(y), k
y
) from the quasi-Fermi
levels requires instead the evaluation of Fermi integral on the
order of (1/2)
F
eq
(E
fn
(y), k
y
) =

2m
z
k
B
T
2
2

1
2
_
E
k
B
T
_
(13)
where
E E
fn
(y)
_
E
SB
(y) +

2
k
2
y
2m
y
_
F

1
2
(x) =

_
y=0
y

1
2
1 + exp(y x)
dy. (14)
With the previous assumptions, the original problem has been
reduced to a nonlinear system composed by (11) and (8) whose
unknowns are the distribution functions F(y, k
y
) and the quasi-
Fermi levels E
fn
(y). Furthermore, the complexity of the prob-
lem has been reduced with respect to the original case since the
discretization domain is now a 2-D space whose components
are the electron transport direction y, in real space, and the
y-component of the wave vector, k
y
, in k space.
A box integration method [22] is used to discretize (11) and
(8) onto nonuniform rectangular grids. If we indicate the num-
ber of internal grid points with N, the result of discretization is
a nonlinear algebraic system
A F =
F
eq

(15)
2904 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 54, NO. 11, NOVEMBER 2007
Fig. 3. Flow chart for the solution of BTE for a given subband prole.
where A is a pentadiagonal N N matrix, F is the N-element
vector of the discretized distribution function, F
eq
is the
N-element vector of the discretized equilibrium distribution
function (further details on the discretization procedure are
given in the next section).
The system is solved using a standard xed point method, as
sketched in the ow diagram of Fig. 3. At the generic kth step,
the linear algebraic system obtained from the discretization of
(11), assuming a known value for F
eq
(E
fn
(y), k
y
), is solved
using a successive-overrelaxation iterative method [23].
The right-hand side of the system is then updated according
to the following procedure.
1) Carrier concentration is evaluated using the calculated
distribution function.
2) The quasi-Fermi levels are obtained by solving (8).
3) The new guess of F
eq
is calculated by the relation (13).
The algorithm is considered converged when the norm of
the difference between F
eq
values at two consecutive steps
falls below an externally set threshold. At this point, electron
distribution functions and its moments (carrier concentration
and current) can easily be obtained.
B. Self-Consistent Multisubband Approach
In the previous section, we have supposed to start from a
known subband prole which is not dependent on the carrier
concentration obtained from the solution of BTE. A variation
in the carrier density translates, via the Poisson equation, into
a variation in the electrostatic potential and, therefore, in the
subband prole obtained from the solution of the Schrdinger
equation.
To reach self-consistency, the previously described method
is inserted into our 2-D PoissonSchrdinger (PS) solver [24],
[25], as schematically described in the ow diagram of Fig. 4.
At each step, the potential prole obtained from the PS solver
is used to evaluate the new guess of the distribution function
which is then used to estimate the carrier density. This term
is inserted afterward into the Poisson equation, and the self-
consistent procedure continues until convergence is reached.
III. NUMERICAL ISSUES
In this section, we will focus on the numerical issues arising
from the discretization of the BTE. First, we specify boundary
conditions on the integration domain, and then, we identify an
appropriate discretization scheme.
A. Boundary Conditions
The integration domain of (11) is the yk
y
plane. In the real-
space dimension, the boundary conditions are set by source and
drain contacts, while in the momentum space, we can properly
x an interval [k
y
min
k
y
max
] outside of which the electron
distribution function can be considered negligible. Hence, we
can impose Dirichlet boundary conditions on the momentum
space assuming the distribution function to be zero for k
y
=
k
y
max
and k
y
= k
y
min
.
The imposition of boundary conditions in the real space
requires instead more subtle argumentations: BTE has indeed
only the rst derivative in space, but we would specify boundary
conditions at both contacts. This seeming inconsistency can be
overcome assuming the hypothesis of black body contacts.
In particular, distribution-function values are specied at each
contact only for emitted electrons. An equilibrium distribution
function with Fermi level E
FS
(E
FD
) is therefore xed at
the source (drain) electrode for states characterized by k
y
>
0 (k
y
< 0).
B. Discretization
In order to discretize (11), both dimensions of the integration
area can be divided into nonuniform steps. If we indicate the
number of points along y and k
y
axes with N
y
and N
k
, we
can identify the coordinates of real space and momentum-
space dimensions with y[i] and k
y
[j], respectively, where i =
1, . . . , N
y
, and j = 1, . . . , N
k
.
Let (i, j) be the grid point of coordinates (y
i
, k
yj
) and F
i,j
be the electron distribution function at that point. According to
a box integration approach, the device domain is divided into as
many nonoverlapping cells as internal grid points.
Integrating (11) on each cell leads to the following nite
difference equation:
v
y
j
[F
i+
1
2
,j
F
i
1
2
,j
]k
j

q
y
j

[F
i,j+
1
2
F
i,j
1
2
]y
i
=
F
eq
i,j
F
i,j

k
j
y
i
(16)
where subscripts that are applied to physical quantities denote
coordinates of the domain space, while k
j
and y
i
represent
the dimensions of the elementary cell associated with the (i, j)
grid point. Equation (16) is a balance of uxes in the control
SCALDAFERRI et al.: DIRECT SOLUTION OF THE BTE AND THE PS EQUATION FOR NANOSCALE MOSFETs 2905
Fig. 4. Flow chart of the self-consistent solution of the BTE and the PS equation.
Fig. 5. Contour plot of the doping prole of the 25-nm MOSFET considered
in the simulations (concentrations are in cubic meters).
volume: The rst two terms at the left-hand side are diffusion
and drift uxes, respectively, while the right-hand side is the
scattering ux. An upwinding technique is used to obtain
a stable discretization and to avoid the unphysical negative
values for F.
IV. SIMULATION RESULTS
The method described in Section II has been used to study
the electrical characteristics of a benchmark device, i.e., the
well-tempered MOSFET proposed by Antoniadis et al. [20].
In spite of the reduced dimensions, short channel effects (SCEs)
in the device are maintained at a reasonable level, owing to the
ideal doping prole considered. In particular, Fig. 5 shows
the net doping concentration in the transistor: Source and drain
proles are Gaussian, while a superhalo doping is implanted in
the substrate in order to reduce SCEs. The oxide thickness is
1.5 nm, and the effective channel length, which is measured
at the points where minority carrier concentration in the chan-
nel falls below 2 10
19
cm
3
, is approximately 25 nm. The
polysilicon gate has a donor concentration of 5 10
20
cm
3
to
reduce polysilicon depletion effects.
Transport regime of carriers in the channel can be selected
by specifying the scattering time in the input of the BTE solver.
In the present simulations, we choose a value in the range
[10
14
s 10
11
s], which allows us to explore the entire DD
to ballistic transport regime.
For the BTE solution, we have adopted a grid with N
y
= 951
points in the real-space dimension and N
k
= 1000 points in the
momentum dimension.
Fig. 6. Convergence of self-consistent algorithm iterations for V
gs
= 1 V,
V
ds
= 1 V, and = 10
13
s.
In order to improve the stability of the algorithm described
in Section II, an underrelaxation technique has been used to
update the subband proles at each different iteration step.
By indicating the subband proles at two successive iteration
steps obtained from the PS solver with E
k1
i
and E
k
i
, the new
subband prole E
new
i
for the BTE solver is obtained as
E
new
i
= E
k1
i
+
_
E
k
i
E
k1
i
_
. (17)
The stability of the implemented algorithm and its conver-
gence properties is rst analyzed in Figs. 6 and 7. In particular,
Fig. 6 shows the residual versus the number of iterations for a
bias condition (V
gs
= 1 V, and V
ds
= 1 V) and for a scattering
time ( = 10
13
s).
As is clearly visible, the convergence rate strongly depends
on the bias conditions applied to the structure. Fig. 7 shows the
iteration number as a function of the voltage applied to the gate
electrode V
gs
for V
ds
= 1 V and a scattering time = 10
13
s.
In accordance with the residual behavior, the time required to
reach convergence increases with an increasing bias.
The effect of carrier scattering on the electrical character-
istics of the device is analyzed in the following for a xed
drainsource voltage V
ds
= 1 V and a gate bias sweeping
from V
gs
= 0.3 V to V
gs
= 1.3 V. Fig. 8 shows the transfer
2906 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 54, NO. 11, NOVEMBER 2007
Fig. 7. Iterations required for convergence versus V
gs
for V
ds
= 1 V and =
10
13
s.
Fig. 8. Transfer characteristics for V
ds
= 1 V and scattering times =
10
11
s, 10
12
s, 10
13
s, and 10
14
s.
characteristics in semilogarithmic scale for scattering times
= 10
11
s, 10
12
s, 10
13
s, and 10
14
s.
The threshold voltage V
T
for the considered structure is
about 0.65 V. This high value is the consequence of the very
high halo doping concentration required to suppress the SCEs
arising from the protruding of the source and drain depletion
regions into the channel. It is worth noting that in a real device,
the ad hoc doping prole considered would induce a severe
current-drive capability reduction as well as strong band-to-
band tunneling effects at the junctions.
Subthreshold slope is close to 100 mV/decade, and as ex-
pected, it is barely affected by the scattering time selected.
Conversely, the current drive is strongly dependent on the scat-
tering conditions. As shown in Fig. 8, indeed, for a high scatter-
ing time set as input ( = 10
12
s), the current approaches the
theoretical ballistic limit, while a strong current degradation can
be observed as the scattering time is progressively decreased
down to = 10
14
s.
Fig. 9 shows the distribution-function color maps related to
the rst subband for high bias condition (V
gs
= 1.3 V, and
V
ds
= 1 V). The four cases analyzed roughly correspond to a
transition from a full DD regime ( = 10
14
s) to a full ballistic
case ( = 10
11
s).
In the BB case, as theoretically predicted [26][28], we can
clearly observe the ballistic peak in the drain region and a
hemi-FermiDirac distribution-function prole at the top of
the barrier due to the suppression of electrons coming from
the drain contact. Indeed, in the case of high source-to-drain
bias, carriers injected from the drain reservoir do not have
enough energy to overcome the energy barrier set by V
DS
and
cannot reach the source contact. The distribution function at the
source contact is therefore represented by a hemi-FermiDirac
distribution.
At the other extreme, going toward a DD regime ( =
10
14
s), collision randomize carriers motion, and the distribu-
tion function has a quasi-equilibrium behavior at each channel
position.
It is worth noting that while the terminal current is not
strongly dependent on in the range [10
11
10
13
s], the
carrier distribution shows instead a strong dependence on the
scattering rate, particularly in the polysource implant super-
position region. Conversely, for = 10
14
s, a change in the
carrier distribution directly translates into a reduced terminal
current.
This behavior can be explained with the help of Fig. 10,
where the rst subband prole along the channel is shown for
each scattering time considered. The top of the barrier is located
in the source-contact proximity, while accumulation regions for
y in the range [30 nm15 nm] and [15 nm30 nm] are the
effects of the superposition of polygate with source and drain
extensions, respectively.
The maximum of the barrier is practically coincident with
the source contact. For > 10
13
s, almost all source emitted
electrons cross the barrier, and the barrier peak is barely af-
fected by a change in . Any scattering event occurring after
the maximum point is then revealed by the presence of a quasi-
symmetric electron fraction but does not affect the current at
the terminals.
For = 10
14
s, instead, scattering directly inuences the
barrier height which modies the electron fraction overcoming
the barrier and, consequently, the terminal current.
A. Ballistic Fraction
As observed in the previous section, a ballistic transport
regime can be identied by a peak in the electron distribution
functions in the drain region which gradually vanishes when
approaching a collision-dominated DD regime. This property
can be used to dene a synthetic parameter indicating the
transport regime for a given structure.
Let us refer to Fig. 11 where the electron distribution function
F(y
d
, v
y
) at the drain contact (y
d
= 47.5 nm) is represented for
bias conditions (V
ds
= 1 V, and V
gs
= 1.3 V) and a scattering
time ( = 10
13
s).
We can identify a quasi-symmetric component of the distrib-
ution function F
s
(y
d
, v
y
) dened as follows:
F
s
(y
d
, v
y
) =
_
F(y, v
y
), for < v
y
< v
y
min
0, otherwise
(18)
and a ballistic component F
b
(y
d
, v
y
)
F
b
(y
d
, v
y
) =
_
0, for < v
y
< v
y
min
F(y, v
y
), otherwise
(19)
SCALDAFERRI et al.: DIRECT SOLUTION OF THE BTE AND THE PS EQUATION FOR NANOSCALE MOSFETs 2907
Fig. 9. Electron distribution function in the rst subband as a function of four different scattering times for V
gs
= 1.3 V and V
ds
= 1 V. For an operating
regime ranging from fully (top left) ballistic to (bottom right) DD, it is clearly visible that the ballistic peak at the drain side progressively disappears due to the
randomizing effect of collisions.
Fig. 10. Energy prole of the rst subband as a function of position along the
channel and the scattering time. In particular, different scattering times have
been considered: = 10
11
s, 10
12
s, 10
13
s, and 10
14
s.
where v
y
min
, as shown in Fig. 11, ideally represents the
separation point between the ballistic and quasi-symmetric
components. From (18) and (19), the following relation natu-
rally follows:
F(y
d
, v
y
) = F
b
(y
d
, v
y
) + F
s
(y
d
, v
y
). (20)
In a similar way, we can identify a drift-diffusive component
of the current at the terminals J
d
dened as follows:
J
d
=
_
k
y
F
s
(y
d
, v
y
)dv
y
. (21)
If we indicate the total current at the terminals with J
T
, then
the term B
d
= 1 J
d
/J
T
can represent a synthetic parameter
indicating transport regime, which we will call ballistic fraction
(1 B
d
) represents instead the diffusive fraction).
Fig. 12 shows the diffusive and ballistic fractions r plotted
versus the scattering time for V
ds
= 1 V, and V
gs
= 1.3 V.
Fig. 11. Electron distribution function at the drain contact for V
ds
= 1 V,
V
gs
= 1.3 V, and = 10
13
s. The ballistic and drift-diffusive components
are highlighted.
Fig. 12. Drift-diffusive rate versus scattering time for V
ds
= 1 V and
V
gs
= 1.3 V.
As we expected, R
d
reaches a saturation point close to unity
for low scattering-time values (drift-diffusive regime), while it
exhibits a linear decreasing prole for the increasing values of
scattering time.
2908 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 54, NO. 11, NOVEMBER 2007
V. CONCLUSION
We have presented an algorithm for the self-consistent so-
lution of the BTE and the PS equation in two dimensions,
which accurately captures the essential physics of ultrascaled
eld-effect transistors. The proposed approach, based on a
subband-transport formalism, allows to account for the quan-
tum effects due to the gate-induced vertical conning eld.
Furthermore, the direct solution of the BTE in each subband
that is obtained from the direct solution of the Schrdinger
equation coupled to a PS solver allows to fully explore the
entire DD to ballistic (BB) regime, which is a key capability
for modern device simulators.
The proposed method has been then used to study a bench-
mark device [20] with a 25-nm channel length. In particular, the
carrier distribution function and the terminal currents have been
carefully analyzed as a function of the scattering events in the
device. The main features of the DD and BB transports have
been demonstrated as well as the capability of modeling the
intermediate operating regime. Finally, a synthetic parameter,
indicating the transport regime and the ability of scattering
events to alter the terminal current, has been dened and
evaluated as a function of the relaxation time.
REFERENCES
[1] M. V. Fischetti and S. E. Laux, Monte Carlo analysis of electron
transport in small semiconductor devices including band-structure and
space-charge effects, Phys. Rev. B, Condens. Matter, vol. 38, no. 14,
pp. 97219745, Nov. 1988.
[2] C. Jacoboni and L. Reggiani, The Monte Carlo method for the so-
lution of charge transport in semiconductors with applications to co-
valent materials, Rev. Mod. Phys., vol. 55, no. 3, pp. 645705,
Jul. 1983.
[3] U. Ravaioli and D. K. Ferry, MODFET ensemble Monte Carlo model
including the quasi-two-dimensional electron gas, IEEE Trans. Electron
Devices, vol. ED-33, no. 5, pp. 677681, May 1986.
[4] R. Stratton, Diffusion of hot and cold electrons in semiconductor
barriers, Phys. Rev., vol. 126, no. 6, pp. 20022014, Jun. 1962.
[5] K. Bltekjr, Transport equations for electrons in two-valley semicon-
ductors, IEEE Trans. Electron Devices, vol. ED-17, no. 1, pp. 3847,
Jan. 1970.
[6] D. Ventura, A. Gnudi, G. Baccarani, and F. Odeh, Two-dimensional
MOSFET simulation by means of a multidimensional spherical harmonics
expansion of the Boltzmann transport equation, Solid State Electron.,
vol. 36, no. 4, pp. 575581, Apr. 1993.
[7] K. Banoo and M. S. Lundstrom, Direct solution of the Boltzmann trans-
port equation in nanoscale Si devices, in Proc. SISPAD, 2000, pp. 5053.
[8] T. Grasser, R. Kosik, C. Jungemann, H. Kosina, B. Meinerzhagen, and
S. Selberherr, in Proc. IWCE-10, 2004, pp. 3637.
[9] B. Winstead and U. Ravaioli, A coupled Schrdinger/Monte Carlo tech-
nique for quantum-corrected device simulation, in Proc. Device Res.
Conf., Jun. 2527, 2001, pp. 169170.
[10] H. Tsuchiya, B. Winstead, and U. Ravaioli, Quantum potential ap-
proaches for nano-scale device simulation, in Proc. IWCE, Glasgow,
U.K., pp. 118119.
[11] L. Lucci, P. Palestri, D. Esseni, and L. Selmi, Multi-subband
Monte Carlo modeling of nano-MOSFETs with strong vertical quantiza-
tion and electron gas degeneration, in IEDM Tech. Dig., Dec. 5, 2005,
pp. 617620.
[12] W. Liang, N. Goldsman, I. Mayergoyz, and P. Oldiges, 2-D MOSFET
modeling including surface effects and impact ionization by self-
consistent solution of the Boltzmann, Poisson, and hole-continuity
equations, IEEE Electron Devices, vol. 44, no. 2, pp. 257267,
Feb. 1997.
[13] C. K. Huang and N. Goldsman, Non-equilibrium modeling of tunneling
gate currents in nanoscale MOSFETs, Solid State Electron., vol. 47,
no. 4, pp. 713720, Apr. 2003.
[14] C. K. Huang and N. Goldsman, 2-D self-consistent modeling of
MOSFET quantum effects by solving the Schrdinger and Boltzmann
system of equation, in Proc. Simul. Semicond. Process. Devices, 2001,
pp. 148151.
[15] M. J. VanDort, P. H. Woerlee, and A. J. Walker, A simple model
for quantisation effects in heavily-doped silicon MOSFETs at inver-
sions condition, Solid State Electron., vol. 37, no. 3, pp. 411414,
Mar. 1994.
[16] W. Hnsch, T. Vogelsang, R. Kirchner, and M. Orlowski, Carrier trans-
port near the Si/SiO
2
interface of a MOSFET, Solid State Electron.,
vol. 32, no. 10, pp. 839849, Oct. 1989.
[17] M. G. Ancona and H. F. Tiersten, Macroscopic physics of the silicon
inversion layer, Phys. Rev. B, Condens. Matter, vol. 35, no. 15, pp. 7959
7965, May 1987.
[18] A. Wettstein, A. Schenk, and W. Fichtner, Quantum device-simulation
with the density-gradient model on unstructured grids, IEEE Trans.
Electron Devices, vol. 48, no. 2, pp. 279284, Feb. 2001.
[19] G. Iannaccone et al. unpublished.
[20] D. Antoniadis, I. Djomehri, K. Lackson and S. Miller, [Online]. Available:
http://www-mtl.mit.edu/well
[21] G. Fiori and G. Iannaccone, Modeling of ballistic nanoscale
metaloxidesemiconductor eld effect transistors, Appl. Phys. Lett.,
vol. 81, no. 19, pp. 36723674, Nov. 2002.
[22] S. Selberherr, Analysis and Simulation of Semiconductor Devices. New
York: Springer-Verlag, 1984, pp. 175180.
[23] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numer-
ical Recipes in CThe Art of Scientic Computing. Cambridge, U.K.:
Cambridge University Press, 1992, pp. 866870.
[24] G. Curatola, G. Doornbos, Y. V. Ponomarev, J. J. P. Loo, and
G. Iannaccone, Detailed modeling of sub-100-nm MOSFETs based on
Schrdinger DD per subband and experiments and evaluation of the
performance gap to ballistic transport, IEEE Trans. Electron Devices,
vol. 52, no. 8, pp. 18511858, Aug. 2005.
[25] G. Curatola and G. Iannaccone, NANOTCAD2D: Two-dimensional code
for the simulation of nanoelectronic devices and structures, Comput.
Mater. Sci., vol. 28, no. 2, pp. 342352, 2003.
[26] A. Rahman, J. Guo, S. Datta, and M. S. Lundstrom, Theory of ballistic
nanotransistors, IEEE Trans. Electron Devices, vol. 50, no. 9, pp. 1853
1864, Sep. 2003.
[27] Z. Ren, R. Venugopal, S. Datta, M. S. Lundstrom, D. Jovanovic, and
J. Fossum, The ballistic nanotransistor: A simulation study, in IEDM
Tech. Dig., Dec. 1013, 2000, pp. 715718.
[28] G. Timp, J. Bude et al. The ballistic nanotransistor, in IEDM Tech. Dig.,
1999, pp. 5558.
Stefano Scaldaferri was born in Lagonegro, Italy,
on November 30, 1978. He received the M.Sc. degree
in electrical engineering from the University of Pisa,
Pisa, Italy, in 2003, with a thesis on modeling of
electronic transport in nanoscale eld-effect tran-
sistors, and the Master of Engineering degree in
embedded systems design from the ALaRI Institute
of Lugano, Lugano, Switzerland, in 2004, with a
thesis on application-driven optimization of VLIW
architectures.
From September 2004 to September 2006, he was
with the Inneon Technologies, Munich, Germany, working as an IC Design
Engineer for memory products. Since September 2007, he has been with the
Dialog Semiconductor GmbH, Kirchheim unter Teck-Nabern, Germany, as an
IC Designer for audio and power management applications.
SCALDAFERRI et al.: DIRECT SOLUTION OF THE BTE AND THE PS EQUATION FOR NANOSCALE MOSFETs 2909
Gilberto Curatola was born in Italy, on May 5,
1975. He received the M.Sc. degree in electrical
engineering and the Ph.D. degree for a thesis on the
subject of quantum effects and transport in nanoscale
eld-effect transistors from the University of Pisa,
Pisa, Italy, in 2000 and 2005, respectively.
Since January 2005, he has been with the
NXP Semiconductors, Leuven, Belgium, working
on device simulations for CMOS applications. His
research interest includes characterization and sim-
ulation of deep-submicrometer silicon devices with
emphasis on quantum effects and mobility.
Giuseppe Iannaccone received the M.S. and Ph.D.
degrees in electrical engineering from the Universit
di Pisa, Pisa, Italy, in 1992 and 1996, respectively.
In 1996, he took a permanent position as
a Researcher with the Italian National Research
Council, and in the same year, he obtained a
faculty position with the Dipartimento di Ingegneria
dellInformazione: Elettronica, Informatica, Teleco-
municazioni, University of Pisa, rst as an Assis-
tant Professor and, then, since January 2001, as an
Associate Professor of electronics. He has authored
or coauthored more than 110 papers published in peer-reviewed journals and
more than 70 papers in proceedings of international conferences. His research
interests include transport and noise in nanoelectronic and mesoscopic devices,
development of device modeling and TCAD tools, and the design of extremely
low-power circuits and systems for radio-frequency identication and ambient
intelligence scenarios.
Dr. Iannaccone has coordinated a few European and National Projects
involving multiple partners and has acted as Principal Investigator in several
research projects funded by public agencies at the European and National
levels, and by private organizations. He is in the technical committee of few
international conferences and serves as a referee for the leading journals in the
elds of condensed matter physics, device electronics, and circuit design.