Professional Documents
Culture Documents
6, JUNE 2007
Abstract—We have developed analytical physically based mod- Compact and accurate models of the threshold voltage, which
els for the threshold voltage [including the drain-induced barrier are the DIBL and the subthreshold swing for DG MOSFETs,
lowering (DIBL) effect] and the subthreshold swing of undoped are needed in order to facilitate and extend the use of these
symmetrical double-gate (DG) MOSFETs. The models are derived
from an analytical solution of the 2-D Poisson equation in which devices in integrated circuits. However, a fully 2-D analyti-
the electron concentration was included. The models for DIBL, cal threshold voltage and subthreshold swing models for DG
subthreshold swing, and threshold voltage roll-off have been veri- MOSFETs, and particularly for undoped devices, are still miss-
fied by comparison with 2-D numerical simulations for different ing. Chen et al. [8] developed a 2-D model for the threshold
values of channel length, channel thickness, and drain–source
voltage; very good agreement with the numerical simulations has
voltage roll-off of DG undoped devices, but that model did not
been observed. include DIBL effects. Kranti et al. [9], Chen et al. in [10],
and Suzuki et al. [11], [12] presented models that accounted
Index Terms—Device modeling, double gate (DG), downscaling,
drain induced barrier lowering (DIBL), MOSFET, subthreshold for the DIBL effect, but the devices considered were doped,
swing, threshold voltage. and the effect of the mobile charge density was neglected. It
has to be remarked that undoped DG MOSFETs show better
I. INTRODUCTION performances than doped ones because of their higher mobility.
In undoped devices, the effect of the mobile charge density
Fig. 1. Symmetrical DG MOSFET considered in this paper. (a) 3-D device structure. (b) Cross section.
of a symmetric undoped DG MOSFET including the effect of The channel is undoped (∼ = 1016 cm−3 ), the n+ source and
the mobile charge density. The dependences of channel length, drain are highly doped. All calculations have been done at room
thickness, and drain–source voltage are accounted for. temperature.
We have used an appropriate definition of the threshold The channel electrostatics is governed by the Poisson equa-
voltage for these devices. In bulk MOSFET, it is usually defined tion. If the device is undoped
as the gate voltage at which the surface potential is equal to
q
two times the Fermi potential [16]. Nevertheless, this definition ∇2 φ(x, y) = n (1)
εsi
is not adequate for DG MOSFETs (particularly for undoped
devices), where the inversion and the accumulation take place in where φ is the electrostatic potential referenced to the Fermi
the whole film. The threshold voltage can be instead defined as level in the source [3], [4], the electron density is given as
the gate voltage at which the minimum sheet density of carriers
Qinv reaches a value QTH that can be identified as the onset of n = ni e(φ−φF )/VT (2)
the turn-on condition [8], [17].
We observed a very good agreement with 2-D numerical where ni is intrinsic electron density in silicon, VT is the
simulations of the threshold voltage and the subthreshold swing thermal voltage, and φF is the nonequilibrium quasi-Fermi
for different values of channel lengths and thickness and from level referenced to the Fermi level in the source, satisfying the
low to high drain–source voltage values. The structure of this following boundary conditions:
paper is the following: Section II addresses the derivation of the
φF (0, y) = 0 (3)
potential model from an analytical solution of the 2-D Poisson’s
equation. In Section III, we derive an expression of the location φF (L, y) = VDS (4)
of the minimum potential. In Section IV, the development of the
threshold voltage model, using the value of the minimum poten- where VDS is the drain voltage. The boundary conditions for φ
tial, is presented, and its dependence with the silicon thickness, are given as
channel length, and drain–source voltage is compared with
∂φ(x, y)
2-D numerical simulation results (using DESSIS–integral of Cox (VGS − φMS − φ(x, y = t0 )) = − εSi (5)
squared error). Conclusions are summarized in Section V. ∂y y=t0
φ(0, y) = Vbi (6)
Fig. 1 shows the cross section of the symmetrical where VGS is the gate voltage, φMS is the gate work func-
DG-MOSFET considered in this paper. We have assumed a tion referenced to intrinsic silicon, Vbi is the built-in voltage
DG-MOSFET with Si–SiO2 interface parallel to (100) plane. given as approximately 0.6 V, and t0 is half the Si thickness
1404 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 54, NO. 6, JUNE 2007
t0 = tSi /2. Assuming drift-diffusion transport, the drain– φ1 (x, y) is the solution of the residual 2-D Poisson’s equation:
current can be written as
q
∇2 φ1 (x, y) = ni eφ0 (y)/VT eφ1 (x,y)/VT − 1 . (17)
t0 εsi
dφF
ID = q · µ · W n(x, y) · dy
dx Assuming that φ1 /VT is small, (17) can be reduced to be a
−t0
Laplace equation. This is a reasonable approximation in well-
t0 behaved devices, with not too strong short-channel effects. This
dφF
= q · µ · ni · W e[φ(x,y)−φF ]/VT dy (8) is equivalent to use the superposition of a 1-D solution of the
dx
−t0 Poisson’s equation assuming a 1-D distribution of the mobile
or charge and a 2-D solution of Laplace’s equation. Therefore,
φ1 (x, y) can be considered the solution of
Vds
e−φF /VT dφF ∂2 ∂2
ID = q · ni · W µ L0 (9) φ1 (x, y) + 2 φ1 (x, y) = 0. (18)
∂x ∂y
dx
0
t0 Using (11), the boundary conditions of φ1 (x, y) can be
eφ(x,y)/VT dy
−t0
written as
where W is the device width, µ is the electron low field φ1 (0, y) = Vbi − φ0 (y) (19)
mobility. In the subthreshold regime, the quasi-Fermi potential φ1 (L, y) = VDS + Vbi − φ0 (y) (20)
retains its value at the source end in most of the channel [8];
in fact, this region with a constant value of φF is the only and
region which significantly contributes to the integral in the
εox ∂φ1 (x, y)
numerator on the right-hand side of (9) [18]. Therefore, for · [0 − φ1 (x, y = to )] = −εsi · . (21)
practical purposes, in the 2-D Poisson’s equation (1), we can tox ∂y y=to
use the expression of the electron density with φF = 0 (value
The solution for the 1-D potential term φ0 (y) is given by [3]:
of the quasi-Fermi potential at the source)
2
Bn
n = ni eφ(x,y)/VT . (10) φ0 (y) = VT · ln sec2 (Bn · y) . (22)
2·δ
In order to find an analytical solution, we apply the superposi-
The solution of (18) with the boundary conditions (19)–(21) is
tion principle, and we write the potential φ (x, y) as the sum of
(for details about the procedure, see Appendix)
two terms: φo (y), which is the solution of the 1-D Poisson’s
equation in the direction perpendicular to the channel, and
x−L
φ1 (x, y) = C0 · eλ t0 + C1 · e−λ t0 · cos(λ · y)
x
φ1 (x, y), which is the solution of the residual 2-D differential (23)
equation:
where
φ(x, y) = φ0 (y) + φ1 (x, y). (11)
q
δ= ni · t20 (24)
Therefore, φ0 (y) is the solution of εsi · Vt
q and
∇2 φ0 (y) = n (12)
εsi
C0 = S1 · VDS + Vbi · 1 − e−L to
λ
− S2 · φso (25)
where
C1 = S1 · Vbi · 1 − e−L to − VDS · e−L to − S2 · φso .
λ λ
and
4 · λ · cos λ2 · 1 − e−L to
λ
εox ∂φ0
· [VGS − φms − φ0 (y = t0 )] = −εsi · (15)
tox ∂y y=to S2 = (28)
[2 · λ + sin(2 · λ)] · 1 − e−2L to
λ
where εox · to
Cr = (29)
tsi tox · εsi
t0 = (16)
2 2 · λ tan(λ) = Cr . (30)
EL HAMID et al.: TWO-DIMENSIONAL THRESHOLD VOLTAGE AND SUBTHRESHOLD SWING MOSFETs 1405
where
∂φ1 (x, y)
= 0. (33)
∂x xmin
−L tλ cos( λ
2) −L tλ
L t0 V bi · 1 − e 0 + V DS − λ · sin(λ) 1 − e 0 · φ so
xmin = − · ln (34)
2 2·λ −L tλ −L tλ cos( λ ) −L λ
Vbi · 1 − e 0 − VDS · e 0 − λ · sin(λ) 2
1 − e t0 · φso
1406 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 54, NO. 6, JUNE 2007
Fig. 3. Threshold voltage roll-off versus channel length for different channel Fig. 4. DIBL coefficient versus channel length for different channel thickness.
thickness. The discrete circles and diamonds lines are accounting for numerical
results by [22] VDS = 1 V.
The threshold voltage value at low drain–source voltage is,
and from (36)
λ L λ 1 QTH
· e− 2 t0 · cosh xmin −
L λ
Sgs = 2 · S2 · cos . VTH0 = φms + · VT ln − Sdso (42)
2 2 t0 1 − Sgso 2ni · to
(38)
In devices where the channel is long with respect to the channel where
thickness, Sgs is close to zero. Therefore, the long channel
threshold voltage can be written as L λ
1 + cos(λ) sinh 2 t0
e−2L t0 ·
λ
Sgso = (43)
1 + sin(2 · λ)/2 · λ
QTH sinh L tλ0
VTH = φms + VT ln − Sds . (39)
2ni · to
and
The threshold voltage derived in (36) tends to the theoretical
long channel value (39) as the channel length increases. For
cos λ2 /λ sinh L2 tλ0
long enough channels, Sds tends to zero, and the threshold Sdso = · · Vbi . (44)
voltage expression reduces to the one reported by Chen et al. [8] 1 + sin (2 · λ) /2 · λ sinh L tλ0
QTH
VTH = φms + VT ln . (40) The DIBL coefficient can therefore be written as:
2ni · to
Equations (40) and (36) constitute the threshold voltage QTH 1 1
compact models for both short and long channel devices. The DIBL = VT ln −
2ni · to 1 − Sgso 1 − Sgs
DG-MOSFET threshold voltage roll-off can be written as the
difference between the values of threshold voltage calculated Sdso Sds
− − /[Vds,high − Vds,low ].
using (40) and (36) 1 − Sgso 1 − Sgso
(45)
QTH 1
∆VTH = VT ln 1− − Sds . (41)
2ni · to 1 − Sgs
As shown in Fig. 4, very good agreement has been obtained
with 2-D numerical simulations using DESSIS-ISE for channel
The threshold voltage roll-off shows a good agreement with the
lengths down to 30 nm. This confirms that the approximations
numerical results published by Frank and Dennard [22], which
done do not significantly hamper the accuracy of the model.
includes the DIBL effect, as shown in Fig. 3.
The decrease of the threshold voltage with Vds is due to
the DIBL effect. The DIBL has to be determined from the
V. SUBTHRESHOLD SWING MODEL
difference between the threshold voltage at high drain–source
voltage value (e.g., 1 V) and the threshold voltage value at low To find an expression for the subthreshold swing, following
drain–source voltage (0.1 V), using (36) (see Fig. 4). [10], we assume that the subthreshold drain current ID is
EL HAMID et al.: TWO-DIMENSIONAL THRESHOLD VOLTAGE AND SUBTHRESHOLD SWING MOSFETs 1407
A PPENDIX
We use variable separations to find the solution of (18), i.e.,
[15] X. P. Liang and Y. Taur, “A 2-D analytical solution for SCEs in DG Jaume Roig Guitart was born in Roses, Catalonia,
MOSFETs,” IEEE Trans. Electron Devices, vol. 51, no. 9, pp. 1385–1391, Spain, in 1976. He received the B.S. degree in
Sep. 2004. physics and the Ph.D. degree in microelectronics en-
[16] S. M. Sze, Physics of Semiconductor Devices, 2nd ed. New York: Wiley, gineering from Universitat Autònoma de Barcelona,
1981. Barcelona, Spain, in 1999 and 2004, respectively.
[17] Y. Ma, Z. Li, L. Liu, L. Tian, and Z. Yu, “Effective density-of-states From 2000 to 2005, he was with the Centre Na-
approach to QM correction in MOS structure,” Solid State Electron., cional de Microelectronica (CNM-IMB), Barcelona,
vol. 44, no. 3, pp. 401–407, Mar. 2000. Spain, where he was involved in high-voltage smart
[18] T. A. Fjeldly and M. Shur, “Threshold voltage model and subthreshold power devices in SOI and bulk technologies. He
regime of operation of short-channel MOSFET’s,” IEEE Trans. Electron was with the Intégration de Systèmes de Gestion de
Devices, vol. 40, no. 1, pp. 137–145, Jan. 1993. l’Energie (ISGE) team in the Laboratoire d’Analyse
[19] L. Ge and J. G. Fossum, “Analytical modeling of quantization and volume et d’Architecture des Systèmes (LAAS)/Centre National de la Recherche
inversion in thin Si-film DG MOSFETs,” IEEE Trans. Electron Devices, Scientifique (CNRS), Toulouse, France, in 2005. His current research activity
vol. 49, no. 2, pp. 287–294, Feb. 2002. deals with semiconductor power device physics. He is author and coauthor of
[20] G. Chindalore, S. A. Hareland, S. Jallepalli, A. F. Tasch, Jr., C. M. Maziar, more than 40 scientific articles in international journals and conferences.
V. K. F. Chia, and S. Smith, “Experimental determination of thresh-
old voltage shifts due to quantum mechanical effects in MOS electron
and hole inversion layers,” IEEE Electron Device Lett., vol. 18, no. 5,
pp. 206–208, May 1997.
[21] A. Kumar, J. Kedzierski, and S. E. Laux, “Quantum-based simulation Benjamin Iñíguez (M’96–SM’03) received the B.S.,
analysis of scaling in ultrathin body device structures,” IEEE Trans. M.S., and Ph.D. degrees in physics from the Uni-
Electron Devices, vol. 52, no. 4, pp. 614–617, Apr. 2005. versity of the Balearic Islands (UIB), Palma, Spain,
[22] D. J. Frank and R. H. Dennard, “Device scaling limits of Si MOSFETs and in 1989, 1992, and 1996, respectively. His doctoral
their application dependencies,” Proc. IEEE, vol. 89, no. 3, pp. 259–288, research focused on the development of computer-
Mar. 2001. aided design models for short-channel bulk-Si and
SOI MOSFETs.
From 1997 to 1998, he was working as a Post-
doctoral Research Scientist at the Electrical, Com-
Hamdy Abd El Hamid was born in 1970. He puter and Systems Engineering (ECSE) Department,
received the B.S. degree in electrical engineer- Rensselaer Polytechnic Institute (RPI), Troy, NY,
ing from Higher Technological Institute, Tenth of where he studied advanced devices such as short-channel a-Si and poly-Si thin-
Ramadan, Egypt, in 1994, and the M.S. degree in film transistors, GaN HFETs, and heterodimensional MESFETs. From 1998
electronic engineering from Ain Shams University, to 2001, he was a Research Scientist (Postdoctoral Marie-Curie Grant Holder)
Cairo, Egypt, in 2000. Since 2003, he has been in the Microelectronics Laboratory, Université catholique de Louvain (UCL),
working toward the Ph.D. degree at the Departament Louvain-la-Neuve, Belgium, working on the characterization and modeling of
d’ Enginyeria Electrònica, Elèctrica i Automàtica thin-film and ultrathin-film SOI MOSFETs from dc to RF conditions. Since
(DEEEA), Universitat Rovira i Virgili (URV), 2001, he has been with the Departament d’ Enginyeria Electrònica, Elèctrica i
Tarragona, Spain. Automàtica (DEEEA), Universitat Rovira i Virgili (URV), Tarragona, Spain,
He was a Graduate Visiting Student in the as Titular Professor. His current research interests are characterization and
Electrical Engineering Department, University of Liverpool, Liverpool, U.K., modeling of advanced electron devices, particularly nanoscale multiple-gate
from April to June 2005, and in the Microelectronics Laboratory, Université MOSFETs and organic and polymer TFTs.
catholique de Louvain (UCL), Louvain-la-Neuve, Belgium, from April to June Dr. Iñíguez was awarded the Distinction of the Catalan Government for
2006 and in February 2007. His research focuses on the physics and modeling the Promotion of University Research in 2004. In 2007 he obtained the IET
of nanoscale MOSFETs. He will obtain his Ph.D. in May 2007. Circuits, Devices and Systems Premium.