Semi-Classical Transport Theory. Outline: What is Computational Electronics? Semi-Classical...
-
Upload
stephen-greer -
Category
Documents
-
view
228 -
download
3
Transcript of Semi-Classical Transport Theory. Outline: What is Computational Electronics? Semi-Classical...
Semi-Classical Transport Theory
Outline:
What is Computational Electronics?
Semi-Classical Transport Theory Drift-Diffusion Simulations Hydrodynamic Simulations Particle-Based Device Simulations
Inclusion of Tunneling and Size-Quantization Effects in Semi-Classical Simulators Tunneling Effect: WKB Approximation and Transfer Matrix Approach Quantum-Mechanical Size Quantization Effect
Drift-Diffusion and Hydrodynamics: Quantum Correction and Quantum Moment Methods
Particle-Based Device Simulations: Effective Potential Approach
Quantum Transport Direct Solution of the Schrodinger Equation (Usuki Method) and Theoretical
Basis of the Green’s Functions Approach (NEGF) NEGF: Recursive Green’s Function Technique and CBR Approach Atomistic Simulations – The Future
Prologue
Semi-Classical Transport Theory
It is based on direct or approximate solution of the Boltzmann Transport Equation for the semi-classical distribution function f(r,k,t)
which gives one the probability of finding a particle in region (r,r+dr) and (k,k+dk) at time t
Moments of the distribution function give us information about: Particle Density Current Density Energy Density
3
11 1
8 k k kk
Fkk
k r k k k k k k k
f VE f f d f f f f
t
D. K. Ferry, Semiconductors, MacMillian, 1990.
Semi-Classical Transport Approaches
1.Drift-Diffusion Method2.Hydrodynamic Method3.Direct Solution of the Boltzmann Transport
Equation via: Particle-Based Approaches – Monte Carlo
method Spherical Harmonics Numerical Solution of the Boltzmann-Poisson
Problem
C. Jacoboni, P. Lugli: "The Monte Carlo Method for Semiconductor Device Simulation“, in series "Computational Microelectronics", series editor: S. Selberherr; Springer, 1989, ISBN: 3-211-82110-4.
1. Drift-Diffusion Approach
Constitutive Equations
Poisson
Continuity Equations
Current Density Equations
1
1
J
J
n n
p p
nU
t q
pU
t q
D AV p n N N
( ) ( )
( ) ( )
n n n
p p p
dnJ qn x E x qD
dxdn
J qp x E x qDdx
S. Selberherr: "Analysis and Simulation of Semiconductor Devices“, Springer, 1984.
Numerical Solution Details
Linearization of the Poisson equation Scharfetter-Gummel Discretization of the
Continuity equation De Mari scaling of variables Discretization of the equations
Finite Difference – easier to implement but requires more node points, difficult to deal with curved interfaces
Finite Elements – standard, smaller number of node points, resolves curved surfaces
Finite Volume
Linearized Poisson Equation φ→φ + δ where δ= φnew - φold
Finite difference discretization: Potential varies linearly between mesh points Electric field is constant between mesh points
Linearization → Diagonally-dominant coefficient matrix A is obtained
2/ / / /
2
2/ / / /
2
/ /
/
/
old old old oldT T T T
old old old oldT T T T
old oldT T
newV V V V V V V Vi i
i
newV V V V V V V Vnewi i
iT
V V V V oldi
T
new old
en end Ve e C n e e
dx
en end Ve e V e e C n
dx V
ene e V
V
V V
Scharfetter-Gummel Discretization of the Continuity Equation
Electron and hole densities n and p vary exponentially between mesh points → relaxes the requirement of using very small mesh sizes
The exponential dependence of n and p upon the potential is buried in the Bernoulli functions
1/ 2 1 1/ 2 1 1/ 2 1 1/ 2 11 12 2 2 2
1/ 2 1 1/ 2 1 1/ 2 112 2 2
n n n ni i i i i i i i i i i i
i i i iT T T T
n n ni i i i i i i i i
iT T T
D V V D V V D V V D V VB n B B n B n U
V V V V
D V V D V V D V VB p B B
V V V
1/ 2 112
ni i i
i i iT
D V Vp B p U
V
( )1x
xB x
e
Bernoulli function:
Scaling due to de MariVariable Scaling Variable Formula
Space Intrinsic Debye length (N=ni)
Extrinsic Debye length (N=Nmax)
2Bk T
Lq N
Potential Thermal voltage * Bk T
Vq
Carrier concentration Intrinsic concentration
Maximum doping concentration
N=ni
N=Nmax
Diffusion coefficient Practical unit
Maximum diffusion coefficient
2
1cm
Ds
D = Dmax
Mobility *
DM
V
Generation-Recombination 2
DNR
L
Time 2LT
D
Numerical Solution Details
Governing Equations ICS/BCS
DiscretizationSystem of Algebraic Equations
Equation (Matrix) Solver
ApproximateSolution
Continuous Solutions
Finite-Difference
Finite-Volume
Finite-Element
Spectral
Boundary Element
Hybrid
Discrete Nodal Values
Tridiagonal
SOR
Gauss-Seidel
Krylov
Multigrid
φi (x,y,z,t)
p (x,y,z,t)
n (x,y,z,t)
D. Vasileska, EEE533 Semiconductor Device and Process Simulation Lecture Notes,Arizona State University, Tempe, AZ.
Numerical Solution Details
Poisson solvers: Direct
Gaussian Eliminatioln LU decomposition
Iterative Mesh Relaxation Methods
• Jacobi, Gauss-Seidel, Successive over-Relaxation Advanced Iterative Solvers
• ILU, Stone’s strongly implicit method, Conjugate gradient methods and Multigrid methods
G. Speyer, D. Vasileska and S. M. Goodnick, "Efficient Poisson solver for semiconductor device modeling using the multi-grid preconditioned BICGSTAB method", Journal of Computational Electronics, Vol. 1, pp. 359-363 (2002).
Complexity of linear solvers
2D 3D
Sparse Cholesky: O(n1.5 ) O(n2 )
CG, exact arithmetic:
O(n2 ) O(n2 )
CG, no precond: O(n1.5 ) O(n1.33 )
CG, modified IC: O(n1.25 ) O(n1.17 )
CG, support trees: O(n1.20 ) -> O(n1+ ) O(n1.75 ) -> O(n1.31 )
Multigrid: O(n) O(n)
n1/2 n1/3
Time to solve model problem (Poisson’s equation) on regular mesh
LU Decomposition• If LU decomposition exists, then for a tri-diagonal matrix A,
resulting from the finite-difference discretization of the 1D Poisson equation, one can write
where
Then, the solution is found by forward and back substitution:
n
n
nnn
nnn c
cc
abcab
cabca
1
22
11
3
2
111
222
11
...
......
1......
...1
1
.........
nkcab
a kkkkk
kk ,...,3,2,,, 1
111
1,2,...,1,,
,,...,3,2,,
1
111
nixcg
xg
x
nigfgfg
i
iiii
n
nn
iiii
Numerical Solution Details of the Coupled Equation Set
Solution Procedures: Gummel’s Approach – when the constitutive
equations are weekly coupled Newton’s Method – when the constitutive
equations are strongly coupled Gummel/Newton – more efficient approach
D. Vasileska and S. M. Goodnick, Computational Electronics, Morgan & Claypool,2006.D. Vasileska, S. M. Goodnick and G. Klimeck, Computational Electronics: From Semiclassical to Quantum Transport Modeling, Taylor & Francis, in press.
Schematic Description of the Gummel’s Approach
initial guessof the solution
solvePoisson’s eq.
Solve electron eq.Solve hole eq.
nconverged?
converged?n
y
y
initial guessof the solution
solvePoisson’s eq.
Solve electron eq.Solve hole eq.
nconverged?
converged?n
y
y
initial guessof the solution
Solve Poisson’s eq.Electron eq.
Hole eq.
Updategeneration rate
nconverged?
converged?n
y
y
initial guessof the solution
Solve Poisson’s eq.Electron eq.
Hole eq.
Updategeneration rate
nconverged?
converged?n
y
y
Original Gummel’s scheme Modified Gummel’s scheme
Constraints on the MESH Size and TIME Step
The time step and the mesh size may correlate to each other in connection with the numerical stability:
The time step t must be related to the plasma frequency
The Mesh size is related to the Debye length
*
2
m
ne
sp
max
s TD
VL
qN
When is the Drift-Diffusion Model Valid? Large devices where the velocity of the carriers is
smaller than the saturation velocity
The validity of the method can be extended for velocity saturated devices as well by introduction of electric field dependent mobility and diffusion coefficient:
Dn(E) and μn(E)
0 0.8 1.6 2.42.4
x 10-7
0
0.5
1
1.5
2
2.5x 10
5
x (m)
Velo
cit
y (
m/s
)
80 nm FD SOI nMOSFET
0 0.9 1.8 2.72.7
x 10-7
0
0.5
1
1.5
2
2.5x 10
590 nm FD SOI nMOSFET
x (m)0 1 2 33
x 10-7
0
0.5
1
1.5
2
2.5x 10
5100 nm FD SOI nMOSFET
x (m)
Tgate=300KTgate=400KTgate=600K
0 1.2 2.4 3.63.6
x 10-7
0
0.5
1
1.5
2
2.5x 10
5
x (m)
Velo
cit
y (
m/s
)
120 nm FD SOI nMOSFET
0 1.4 2.8 4.24.2
x 10-7
0
0.5
1
1.5
2
2.5x 10
5
x (m)
140 nm FD SOI nMOSFET
0 1.8 3.6 5.45.4
x 10-7
0
0.5
1
1.5
2
2.5x 10
5
x (m)
180 nm FD SOI nMOSFET
Tgate=300KTgate=400KTgate=600K
0 2.5 5 7.5
x 10-8
0
0.5
1
1.5
2
2.5x 10
5
x (m)
Velo
cit
y (
m/s
)
25 nm FD SOI nMOSFET
0 0.45 0.9 1.351.35
x 10-7
0
0.5
1
1.5
2
2.5x 10
5
x (m)
45 nm FD SOI nMOSFET
0 0.6 1.2 1.81.8
x 10-7
0
0.5
1
1.5
2
2.5x 10
5
x(m)
60 nm FD SOI nMOSFET
Tgate=300KTgate=400KTgate=600K
sourcechannel
drain
25 nm 140 nm100 nm
What Contributes to The Mobility?
Scattering Mechanisms
Defect Scattering Carrier-Carrier Scattering Lattice Scattering
CrystalDefects
Impurity Alloy
Neutral Ionized
Intravalley Intervalley
Acoustic OpticalAcoustic Optical
Nonpolar PolarDeformationpotential
Piezo-electric
Scattering Mechanisms
Defect Scattering Carrier-Carrier Scattering Lattice Scattering
CrystalDefects
Impurity Alloy
Neutral Ionized
Intravalley Intervalley
Acoustic OpticalAcoustic OpticalAcoustic Optical
Nonpolar PolarNonpolar PolarDeformationpotential
Piezo-electric
D. Vasileska and S. M. Goodnick, "Computational Electronics", Materials Science and Engineering, Reports: A Review Journal, Vol. R38, No. 5, pp. 181-236 (2002)
Mobility Modeling
Mobility modeling can be separated in three parts: Low-field mobility characterization for bulk
or inversion layers High-field mobility characterization to
account for velocity saturation effect Smooth interpolation between the low-field
and high-field regions
Silvaco ATLAS Manual.
(A) Low-Field Models for Bulk Materials
Phonon scattering:
- Simple power-law dependence of the temperature
- Sah et al. model: acoustic + optical and intervalley phonons combined via Mathiessen’s rule
Ionized impurity scattering:
- Conwell-Weiskopf model
- Brooks-Herring model
(A) Low-Field Models for Bulk Materials (cont’d)Combined phonon and ionized impurity scattering:
- Dorkel and Leturg model:
temperature-dependent phonon scattering +ionized impurity scattering + carrier-carrier interactions
- Caughey and Thomas model:
temperature independent phonon scattering + ionized impurity scattering
(A) Low-Field Models for Bulk Materials (cont’d)
- Sharfetter-Gummel model:
phonon scattering + ionized impurity scattering (parameterized expression – does not use the Mathiessen’s rule)
- Arora model:
similar to Caughey and Thomas, but with temperature dependent phonon scattering
(A) Low-Field Models for Bulk Materials (cont’d)Carrier-carrier scattering
- modified Dorkel and Leturg model
Neutral impurity scattering:
- Li and Thurber model:
mobility component due to neutral impurity scattering is combined with the mobility due to lattice, ionized impurity and carrier-carrier scattering via the Mathiessen’s rule
(B) Field-Dependent Mobility
The field-dependent mobility model provides smooth transition between low-field and high-field behavior
vsat is modeled as a temperature-dependent quantity:
/1
0
0
1
)(
satvE
E = 1 for electrons = 2 for holes
cm/s
600exp8.01
104.2)(
7
Lsat T
Tv
(C) Inversion Layer Mobility Models
CVT model: combines acoustic phonon, non-polar optical
phonon and surface-roughness scattering (as an inverse square dependence of the perpendicular electric field) via Mathiessen’s rule
Yamaguchi model: low-field part combines lattice, ionized impurity
and surface-roughness scattering there is also a parametric dependence on the
in-plane field (high-field component)
(C) Inversion Layer Mobility Models (cont’d)Shirahata model:
uses Klaassen’s low-field mobility model takes into account screening effects into the
inversion layer has improved perpendicular field dependence for
thin gate oxides
Tasch model: the best model for modeling the mobility in MOS
inversion layers; uses universal mobility behavior
Generation-Recombination MechanismsClassification
Twoparticle
One step(Direct)
Two-step(indirect)
Energy-level
consideration
• Photogeneration• Radiative recombination• Direct thermal generation• Direct thermal recomb.
• Shockley-Read-Hall (SRH) generation-recombination
• Surface generation-recombination
• Shockley-Read-Hall (SRH) generation-recombination
• Surface generation-recombination
Threeparticle
Impactionization
Auger
• Electron emission• Hole emission• Electron capture• Hole capture
Pure generation process
Hydrodynamic Modeling
In small devices there exists non-stationary transport and carriers are moving through the device with velocity larger than the saturation velocity In Si devices non-stationary transport occurs
because of the different order of magnitude of the carrier momentum and energy relaxation times
In GaAs devices velocity overshoot occurs due to intervalley transfer
T. Grasser (ed.): "Advanced Device Modeling and Simulation“, World Scientific Publishing Co., 2003, ISBN: 9-812-38607-6 M.M. Lundstrom, Fundamentals of Carrier Transport, 1990.
Velocity Overshoot in Silicon
-5x106
0
5x106
1x107
1.5x107
2x107
2.5x107
0 0.5 1 1.5 2 2.5 3 3.5 4
1 kV/cm5 kV/cm10 kV/cm50 kV/cm
time [ps]
Drift
velo
city
[c
m/s
]
0
0.05
0.1
0.15
0.2
0.25
0 0.5 1 1.5 2 2.5 3 3.5 4
1 kV/cm5 kV/cm10 kV/cm50 kV/cm
Ene
rgy
[eV
]
time [ps]
Scattering mechanisms:
• Acoustic deformation potential scattering• Zero-order intervalley scattering (f and g-
phonons)• First-order intervalley scattering (f and g-
phonons)
g
f
kz
kx
ky
g
f
g
ff
kz
kx
ky
X. He, MS Thesis, ASU, 2000.
How is the Velocity Overshoot Accounted For?
In Hydrodynamic/Energy balance modeling the velocity overshoot effect is accounted for through the addition of an energy conservation equation in addition to: Particle Conservation (Continuity Equation) Momentum (mass) Conservation Equation
Hydrodynamic Model due to Blotakjer
Constitutive Equations: Poisson +• More convenient set of balance equations is in terms of n, vdand w:
colld
d
Bdd
coll
d
dddd
colld
tw
e
vmw
kn
nw
tw
tme
vnmnwnm
mmt
tn
ntn
)(
2
*
32
)(
*
*21
*32
)*(*
)(
2
2
vE
vv
vE
vvv
v
Closure
• To have a closed set of equations, one either:(a) ignores the heat flux altogether(b) uses a simple recipe for the calculation of the heat flux:
)(*2
5,
2
wvm
nTkTn B q
• Substituting T with the density of the carrier energy, the momentum and energy balance equations become:
Momentum Relaxation Rate• The momentum rate is determined by a steady-state MC
calculation in a bulk semiconductor under a uniform bias electric field, for which:
dp
dpcoll
dd
vmeE
w
wme
tme
t
*)(
0)(**
v
EvEv
K. Tomizawa, Numerical Simulation Of Submicron Semiconductor Devices.
Energy Relaxation Rate• The emsemble energy relaxation rate is also determined by a
steady-state MC calculation in a bulk semiconductor under a uniform bias electric field, for which:
0
0
)(
0)(
ww
ew
wwetw
etw
dw
wdcoll
d
vE
vEvE
K. Tomizawa, Numerical Simulation Of Submicron Semiconductor Devices.
Validity of the Hydrodynamic Model
Source Drain
Gate oxide
BOX
tox
tsi
tBOX
LS Lgate LD
feature 14 nm 25 nm 90 nm
Tox 1 nm 1.2 nm 1.5 nm
VDD 1V 1.2 V 1.4 V
Overshoot EB/HD
233% / 224% 139% / 126% 31% /21%
Overshoot EB/DD with series resistance
153%/96% 108%/67% 39%/26%
0 0.2 0.4 0.6 0.8 1 1.2 1.40
0.5
1
1.5
2
2.5
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
DDEBHDDD SREB SRHD SR
Silvaco ATLAS simulations performed by Prof. Vasileska.
90 nm device
SR = series resistance
Validity of the Hydrodynamic Model
Source Drain
Gate oxide
BOX
tox
tsi
tBOX
LS Lgate LD
feature 14 nm 25 nm 90 nm
Tox 1 nm 1.2 nm 1.5 nm
VDD 1V 1.2 V 1.4 V
Overshoot EB/HD
233% / 224% 139% / 126% 31% /21%
Overshoot EB/DD with series resistance
153%/96% 108%/67% 39%/26%
0 0.2 0.4 0.6 0.8 1 1.2 1.40
0.5
1
1.5
2
2.5
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
DDEBHDDD SREB SRHD SR
Silvaco ATLAS simulations performed by Prof. Vasileska.
25 nm device
0 0.2 0.4 0.6 0.8 1 1.20
1
2
3
4
5
6
7
Drain Voltage [V]D
rain
Cu
rre
nt [
mA
/um
]
DDHDEBDD SREB SRHD SR
SR = series resistance
0 0.2 0.4 0.6 0.8 1 1.2 1.40
0.5
1
1.5
2
2.5
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
DDEBHDDD SREB SRHD SR
Validity of the Hydrodynamic Model
Source Drain
Gate oxide
BOX
tox
tsi
tBOX
LS Lgate LD
feature 14 nm 25 nm 90 nm
Tox 1 nm 1.2 nm 1.5 nm
VDD 1V 1.2 V 1.4 V
Overshoot EB/HD
233% / 224% 139% / 126% 31% /21%
Overshoot EB/DD with series resistance
153%/96% 108%/67% 39%/26%
0 0.2 0.4 0.6 0.8 1 1.20
1
2
3
4
5
6
7
Drain Voltage [V]D
rain
Cu
rre
nt [
mA
/um
]
DDHDEBDD SREB SRHD SR
Silvaco ATLAS simulations performed by Prof. Vasileska.
0 0.2 0.4 0.6 0.8 10
2
4
6
8
10
12
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
DDEBHDDD SREB SRHD SR
14 nm device
SR = series resistance
Failure of the Hydrodynamic Model
0 0.2 0.4 0.6 0.8 10
2
4
6
8
10
12
14
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1020 cm-3
1019 cm-3
0.1 ps
0.3 ps
0.2 ps
Silvaco ATLAS simulations performed by Prof. Vasileska.
90 nm
25 nm
14 nm
0 0.2 0.4 0.6 0.8 1 1.20
1
2
3
4
5
6
7
8
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1020 cm-3
1019 cm-3
0.1 ps
0.2 ps
0.3 ps
0 0.2 0.4 0.6 0.8 1 1.2 1.40
0.5
1
1.5
2
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1019 cm-3
1020 cm-3
0.1 ps
0.2 ps0.3 ps
Failure of the Hydrodynamic Model
0 0.2 0.4 0.6 0.8 1 1.2 1.40
0.5
1
1.5
2
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1019 cm-3
1020 cm-3
0.1 ps
0.2 ps0.3 ps
Silvaco ATLAS simulations performed by Prof. Vasileska.
90 nm
25 nm
14 nm
0 0.2 0.4 0.6 0.8 10
2
4
6
8
10
12
14
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1020 cm-3
1019 cm-3
0.1 ps
0.3 ps
0.2 ps
0 0.2 0.4 0.6 0.8 1 1.20
1
2
3
4
5
6
7
8
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1020 cm-3
1019 cm-3
0.1 ps
0.2 ps
0.3 ps
Failure of the Hydrodynamic Model
0 0.2 0.4 0.6 0.8 1 1.20
1
2
3
4
5
6
7
8
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1020 cm-3
1019 cm-3
0.1 ps
0.2 ps
0.3 ps
0 0.2 0.4 0.6 0.8 1 1.2 1.40
0.5
1
1.5
2
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1019 cm-3
1020 cm-3
0.1 ps
0.2 ps0.3 ps
Silvaco ATLAS simulations performed by Prof. Vasileska.
90 nm
25 nm
14 nm
0 0.2 0.4 0.6 0.8 10
2
4
6
8
10
12
14
Drain Voltage [V]
Dra
in C
urr
en
t [m
A/u
m]
1020 cm-3
1019 cm-3
0.1 ps
0.3 ps
0.2 ps
Summary
Drift-Diffusion model is good for large MOSFET devices, BJTs, Solar Cells and/or high frequency/high power devices that operate in the velocity saturation regime
Hydrodynamic model must be used with caution when modeling devices in which velocity overshoot, which is a signature of non-stationary transport, exists in the device
Proper choice of the energy relaxation times is a problem in hydrodynamic modeling