Anda di halaman 1dari 6

Journal of the Korean Physical Society, Vol. 47, November 2005, pp.


Quantum-Mechanical Modeling and Simulation of a Novel Nano-Scale FET:

Center-Channel (CC) Double-Gate (DG) MOSFET
Kidong Kim and Taeyoung Won
Department of Electrical Engineering, School of Engineering, Inha University, Incheon 402-751
In this paper, a novel device structure (Si1x Gex /Si/Si1x Gex heterostructure), which is named
as center-channel (CC) double-gate (DG) MOSFET, is proposed. The device performance of the
proposed FET structure was investigated with our two-dimensional quantum-mechanical simulator
which is based upon the self-consistent solution of Poisson-Schr
odinger equations and the current
continuity equation. The CC operation of CC-NMOS is confirmed by considering the electron
density distribution and the band lineups as well as the lowest-energy wave function. The currentvoltage characteristics including the transconductance (Gm ) of CC-MOSFET are carefully compared
with those of the conventional DG-NMOS in order to observe the distinct feature of the proposed
FET structure. The simulation revealed that the proposed FET exhibits enhancement of about
1.6x for current drive and 60 % for Gm . Finally, short-channel effects were considered for CC
and DG MOSFET, both of which demonstrate excellent sub-threshold behaviors and open the
possibility of device scaling down to sub-20 nm.
PACS numbers: 85.30.De
Keywords: Quantum-mechanical simulation, Center-channel operation, Double-gate MOSFET, Shortchannel effects, Coupled Poisson-Schr
odinger equations

Recent approaches for nano-scale semiconductor devices place a lot of emphasis on materials and structures [13]. The enhancement of carrier mobility in the
conduction channel needs high-mobility materials such
as Si1x Gex and strained-Si. However, a conventional
MOSFET with strained-Si channel has some disadvantages such as a limit of scaling, high-electric field in the
channel, and corresponding degradation of mobility. For
the solution of these problems, we propose and simulate a high-performance center-channel (CC) double-gate
(DG) structure incorporating the materials Si1x Gex
and strained-Si. To fulfill the numerical simulation of
nano-scale structures such as DG MOSFET, we need to
obtain a self-consistent solution of the coupled PoissonSchr
odinger equations. The CC-NMOS device exhibits
enhancement of current drive and switching speed caused
by a decrease of surface roughness scattering and mobility enhancement of the strained-Si channel. In this
paper, we report our self-consistent QM approach for
the analysis of CC-NMOS. The electrical performance of
CC-NMOS was carefully investigated in terms of channel
length (Lg ) (10 80 nm) and gate oxide thickness (Tox )
(2 nm). Device optimizations were performed in order to suppress the short-channel effects (SCE) through
some critical parameters such as subthreshold swing,
Tel: +82-32-860-8686; Fax: +82-32-875-7436

threshold-voltage (Vt ) roll-off and drain-induced barrier

lowering (DIBL). At the same time, the differences between the CC-NMOS and DG-NMOS simulation results,
and especially enhancement of current drive, were examined by comparing the band lineups, lowest-energy wave
function, and electron density.


The two-dimensional (2-D) QM model for describing
the effect of strain on the band structure of Si and bound
states is based on the self-consistent solution of coupled
Poisson-Schrodinger equations [4]:
[(x, y)(x, y)] = (x, y),
(x, y) = q[n(x, y) + p(x, y) + ND
(x, y) NA (x, y)], (2)


n (x, y) + V (x, y)n (x, y) = En n (x, y),
where is the dielectric constant, the electrostatic potential, the total charge density, n and p the electron
and hole concentrations, ND
and NA the ionized donor
and acceptor concentrations, n (x, y) the wave function
of nth eigenstates, ~ Plancks constant divided by 2,
m (x,y) the effective mass, En the energy of nth eigenstates, and V the potential energy, which is given by
V (x, y) = Ec (x, y) q(x, y). Here, Ec (x, y) is the


Quantum-Mechanical Modeling and Simulation Kidong Kim and Taeyoung Won

band offset in the conduction band. The densities of

ionized donors and acceptors are expressed as
ND (x, y)
1 + gD e(EF x (x,y)ED )/kB T
NA (x, y)
NA (x, y) =
1 + gA e(EA EF y (x,y))/kB T

(x, y) =


where ND (x, y) and NA (x, y) are the donor and acceptor

concentrations, gD and gA the degeneracy factors of the
energy levels, EF n (x, y) and EF p (x, y) the quasi-Fermi
levels, ED and EA the ionization energies of donor and
acceptor, kB the Boltzmann constant, and T the temperature.
We used the mixed Dirichlet and von Neumann boundary conditions for the solution of the Schr
odinger equation, which solve the Schr
odinger equation with Dirichlet and Von Neumann boundary conditions, respectively,
and normalize the states to 1/2. The mixed Dirichlet
and von Neumann boundary conditions are given by the
following equation:
| (z) |2 dz = 1/2.
This means that we obtain a constant function by
summing cosine functions from Dirichlet boundary conditions and sine functions from von Neumann boundary
conditions with normalization to 1/2.
By solving the Schr
odinger equation, we obtain the
quantized states which are occupied with the local quasiFermi levels. The 2-D quantum electron density is found
by using [5]
n(x, y) =

1 p
E F Ej
2m kB T
| f (x, y) |2 F1/2
kB T
where Ej and j (x, y) are the energy and the wave function of the j th eigenstate, EF (x, y) the Fermi level, and
Fk the Fermi-Dirac integrals of order k. These integrals
are defined as follows:
uk du
Fk () =
, k 1,
(k + 1) 0 1 + eu
and also have the following property:
Fk () = Fk1 (), k 1.


We obtain the semi-classical current solution from the

1st moment of the Boltzmann equation by adopting a
simple drift-diffusion model for the electron current:
Jn (x, y) = n (x, y)n(x, y)EF n (x, y).


The continuity equation for current density is given by

Jn (x, y) = 0.



For a model of carrier mobility, we employed a physicsbased and semi-empirical model which takes various scattering mechanisms into account by using the Mathiessen
rule [6]:



where b is the carrier mobility in the bulk, ac the carrier mobility from the surface acoustic phonon scattering,
and sr the carrier mobility from surface roughness.
The bulk mobility, b , uses the empirical model expressed by Masetti et al. [7], which depends on the impurity concentration and temperature:
max (T ) 0


1 + (NA /Cr )
1 + (Cs /NA )

max (T ) = max
b (NA , T ) = 0 +

where NA is the local acceptor concentration, and init

= 2.42, the initial estimate for .
The electron mobility for the acoustic phonon scattering and nondegenerate surface, ac , is expressed by [7]
ac (E , T ) =
+ 1/3

where E is the electric field perpendicular to the direction of current flow. Further, B and C are fitting parameters with initial values of Binit = 3.1 108 cm/s and
Cinit = 3.0 107 (V/cm)2/3 Kcm/s based on physical
quantities such as deformation potential, mass density of
silicon, and effective thickness of inversion layer [8].
At very high perpendicular electric field, surface
roughness scattering sr significantly affects the inversion layer mobility. This is described as follows [9]:
sr (E ) =

2 ,


where is a constant depending on the oxide growth

condition, and the initial value of is = 6 1014 V/s.
We consider the decrease of surface roughness scattering of CC-NMOS by specifying the sr of Eq. (12) and
include the velocity-overshoot effect of strained-Si and
dependence on Ge concentration [10].
For the above carrier-mobility model, we arranged the
optimized results for the model fitting parameters in Table 1 for electrons and holes.
In order to obtain self-consistent QM solutions, we employed an iterative procedure [11]. We start out by calculating the electric potential. Next, the program calculates a charge distribution by using an iteration scheme.
Thereafter, the program determines self-consistent solutions of Poisson-Schrodinger and current equations [11].
The Newton method has been employed, called the conjugate gradient method, with a constraint that should
satisfy error criteria as outlined in [12]. The discretization of main equations is performed through a boxintegration method to apply the finite-difference method.


Journal of the Korean Physical Society, Vol. 47, November 2005

Fig. 2. Schematic diagram of CC-NMOS considered in this


Fig. 1. Flow diagram of the quantum-mechanical simulation.

Fig. 3. Comparison of band lineups of CC and DG-NMOS
for Lg = 30 nm, Vd = 0.1 V, and Vg = 1.5 V.

The flow diagram of the quantum-mechanical simulation

is demonstrated in Figure 1.
Figure 2 schematically shows the CC-NMOS considered in this work. Two metal gates with gate length
Lg and work function M = 4.1 eV are located symmetrically on both sides of Si1x Gex /Si/Si1x Gex heterostructure. The source and drain regions are modeled
as ohmic contacts with a doping of 2 1020 /cm3 , and
the channel region under the gate is undoped. The Si,

Table 1. Mobility model parameters.

cm2 /Vs
cm2 /Vs
1 43.4
cm /Vs
9.68 10
2.23 1017
3.43 10
6.10 10



4.75 10
9.93 10
1.74105 N0.125
8.84105 N0.125
NA in cm3

5.82 1014
2.05 1014
From Ref. 7

Si1x Gex and oxide (Tox ) layer thickness is 3 nm, 1 nm,

and 2 nm, respectively, wherein Lg varies from 10 nm to
80 nm.


In this work, we confirmed the CC operation, the enhancement of drive current and Gm , and short-channel
effects (SCE) of the proposed CC-NMOS structure.
First, we confirmed the CC operation of CC-NMOS
through the band lineups, lowest-energy wave function,
and electron density. Figure 3 shows a comparison of
conduction-band lineups of CC and DG-NMOS with Lg
= 30 nm, Vd = 0.1 V, and Vg = 1.5 V. From this
result, we understand the difference of band offsets in
CC and DG-NMOS, the formation of a quantum well of
CC-NMOS from Si1x Gex /Si/Si1x Gex heterostructure,
and corresponding CC operation. The lowest-energy
wave function of CC-NMOS is compared with that of
DG-NMOS in Figure 4 at the condition of Lg = 30 nm,
Vd = 0.1 V, and Vg = 1.5 V. Contrary to DG-NMOS,
the wave function of CC-NMOS is confined to the center of the channel, meaning CC operation. Figure 5

Quantum-Mechanical Modeling and Simulation Kidong Kim and Taeyoung Won

Fig. 4. Comparison of lowest energy wave function of CC

and DG-NMOS for Lg = 30 nm, Vd = 0.1 V.


Fig. 6. Comparison of electron density of CC and DGNMOS for Lg = 30 nm, Vd = 0.1 V, and Vg = 1.0 V.

Fig. 7. Id -Vg characteristics of CC and DG-NMOS for Lg

= 30 nm, Vd = 0.1 V.

Fig. 5. Quantum electron density for Lg = 30 nm, Vd =

0.1 V, and Vg = 1.0 V. (a) DG-NMOS, (b) CC-NMOS.
Fig. 8. Gm characteristics of CC and DG-NMOS for Lg =
30 nm, Vd = 0.1 V.

shows the quantum electron density for Lg = 30 nm,

Vd = 0.1 V, and Vg = 1.0 V: (a) DG-NMOS, (b) CCNMOS. The electron density of DG-NMOS shows peak
values at both sides of the channel. On the other hand,
the electrons of CC-NMOS are confined at the center of
Si1x Gex /Si/Si1x Gex heterostructure, definitely show-

ing CC operation. The cross-section quantum electron

densities as seen for a cut through the Si channel of CC
and DG-NMOS are compared in Figure 6. These results
visibly show the different shapes of the channels in CC


Fig. 9. Id -Vg characteristics for CC-NMOS with Lg of 10

80 nm, Vd = 0.1 V.

Journal of the Korean Physical Society, Vol. 47, November 2005

Fig. 11. Vt roll-off and DIBL of CC-NMOS in terms of

Lg .

acteristics are demonstrated in Figure 11. From these results, we confirmed that both CC and DG-NMOS show
excellent subthreshold behaviors and successfully suppress the SCE.


Fig. 10. Subthreshold swing of CC and DG-NMOS in

terms of Lg at Vd = 0.1 V.

and DG-NMOS.
Figure 7 demonstrates a comparison of current-voltage
(Id -Vg ) characteristics of CC and DG-NMOS at the condition of Lg = 30 nm and Vd = 0.1 V. This simulation result shows enhancement of about 1.6x for current drive.
The transconductances (Gm ) of CC and DG-NMOS are
compared in Figure 8. The Gm of CC-NMOS increases
by about 60 % compared with that of DG-NMOS, showing Gm = 1428.3 S/m for CC-NMOS, and Gm = 891.3
S/m for DG-NMOS. This is due to enhancement of carrier mobility caused by the strained-Si and a decrease of
surface roughness scattering.
Finally, we estimated the SCE of CC and DG-NMOS
by performing a simulation with respect to Lg . Figure 9
shows the Id -Vg curves for CC-NMOS, produced under
the condition of Lg varying from 10 nm to 80 nm at Vd =
0.1 V. As Lg decreases, the short-channel behavior visibly comes in to sight. Therefore, we analyzed the SCE in
terms of subthreshold swing, threshold voltage (Vt ) rolloff, and drain-induced barrier lowering (DIBL). Figure
10 shows the subthreshold-swing of CC and DG-NMOS
in terms of Lg at Vd = 0.1 V. Vt roll-off and DIBL char-

In this paper, we report our two-dimensional numerical modeling and simulation results for center-channel
(CC) double-gate (DG) MOSFET in comparison with
those for conventional DG structure. CC operation of
CC-NMOS is confirmed by band lineups, lowest energy
wave function, and electron density. The simulation results also reveal that current drive and transconductance
are remarkably enhanced and short-channel effects are
appreciably suppressed. Our simulation results imply
that CC-NMOS structure is a promising candidate for
implementing sub-20 nm MOSFETs.

This work was supported by the Research Program of
Inha University.

[1] T. Krishnamohan, C. Jungemann and K. C. Saraswat,
in SISPAD Tech. Dig. (Munich, Sep., 2004), p. 191.
[2] A. Svizhenko, M. P. Anantram, T. R. Govindan and B.
Biegel, J. Appl. Phys. 91, 2343 (2002).
[3] S. H. Jin, Y. J. Park and H. S. Min, J. Korean Phys.
Soc. 44, 87 (2004).
[4] K. Kim, O. Kwon, J. Seo and T. Won, J. Korean Phys.
Soc. 45, 1384 (2004).
[5] A. Trellakis and A. T. Galick, J. Appl. Phys. 81, 7880

Quantum-Mechanical Modeling and Simulation Kidong Kim and Taeyoung Won

[6] I. H. Cho, B. G. Park, J. D. Lee and J. H. Lee, J. Korean
Phys. Soc. 42, 233 (2003).
[7] G. Masetti, M. Severi and S. Solmi, IEEE Trans. Electron Devices. 30, 764 (1983).
[8] C. T. Sah, T. H. Ning and L. L. Tschopp, Surface Science. 32, 561 (1972).
[9] C. Lombardi, S. Manzini, A. Saporito and M. Vanzi,
IEEE Trans. Computer-Aided Design, 7, 1164 (1988).


[10] F. M. Bufler and W. Fichtner, IEEE Trans. Electron

Devices 50, 278 (2003).
[11] M. Sabathil, S. Hackenbuchner, J. A. Majewski, G. Zandler and P. Vogl, J. Computat. Electron 1, 81 (2002).
[12] S. E. Laux, A. Kumar and M. V. Fischetti, in IEDM
Tech. Dig. (San Francisco, Dec., 2002), p. 715.