quantum corrections for monte carlo simulationmcc.illinois.edu/summerschool/2002/umberto...
TRANSCRIPT
Quantum Corrections forMonte Carlo Simulation
Brian Winstead and Umberto RavaioliBeckman Institute
University of Illinois at Urbana-Champaign
2002 School on Computational Material Science May 21-31, 2002
Outline
• Quantum corrections for quantization effects– Effective potential– Wigner-based– Schrödinger-based
• Quantum corrections for tunneling• Extending Schrödinger-based correction to device
simulation
2002 School on Computational Material Science May 21-31, 2002
20 30 40 50 60 70
0
5
10
15
20
25
30
35
Lateral direction (nm)
Dep
th d
irect
ion
(nm
)
Source
Channel
Drain
Monte Carlo Snapshot of a MOSFET
2002 School on Computational Material Science May 21-31, 2002
Motivation for Quantum Corrections
• Full-quantum transport is often impractical• Goal is to extend the validity of semi-classical
Monte Carlo to the 10-nm regime• Quantum corrections can extend the validity of
Monte Carlo in a practical way– Mixed quantum/classical effects are treated in a
unified fashion– Little extra computational overhead is added in
both 2D and 3D
2002 School on Computational Material Science May 21-31, 2002
Role of Quantum Corrections
• Monte Carlo particles correctly represent themotion of wave packets centroids in the crystal
• Monte Carlo does not account for interferenceeffects due to rapidly varying applied fields orheterojunctions
• Quantum corrections can capture non-coherentinterference effects
• Coherent transport effects are small for puresilicon devices above 10-nm regime
2002 School on Computational Material Science May 21-31, 2002
Corrections in Monte Carlo
Quantum effects can be approximated in MonteCarlo by correcting the classical potential
Tunneling
SizeQuantization
UU + Uc
2002 School on Computational Material Science May 21-31, 2002
Snapshot from MOS Capacitor
0 5 10 15−0.15
−0.1
−0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Depth (nm)
Ene
rgy
(eV
) potentialparticles
E1 = 0 eV
Tox
= 3 nmV
g = 1.0 V
NA = 2 x 1017 cm−3
2002 School on Computational Material Science May 21-31, 2002
Effective Potential
• Feynman developed the effective potential in the1960s and applied it to quantum corrections instatistical mechanics
• Particles feel nearby potential due to quantumfluctuations around classical path of least action
• Veff is a non-local function of the nearby potential
mkTadxexVxV a
xx
eff12
,')()(2
2)'( 2
2 ==−
−
∫
2002 School on Computational Material Science May 21-31, 2002
Properties of Effective Potential
• Simple to implement and to calculate• Not sensitive to noise from Monte Carlo• Works best for smooth, symmetric potentials• a can be treated as a fitting parameter describing
the “size” of the particle• Detailed solution near large heterojunctions is
typically incorrect and cannot be fit
2002 School on Computational Material Science May 21-31, 2002
0 0.5 1 1.5 2 2.5 3
1012
1013
She
et c
harg
e cm
−2
Gate bias (V)
NA = 2 x 1017 cm−3
ND
= 2 x 1017 cm−3
Tox
= 30 Å
Schrödinger−PoissonSchrödinger−correctedWigner−based correctedEffective potential corrected
Effective potential for MOS
• Effective size a can be tuned to match sheetcharge density
2002 School on Computational Material Science May 21-31, 2002
0 1 2 3 4 5 610
16
1017
1018
1019
1020
Con
cent
ratio
n (c
m−
3 )
Depth below interface (nm)
Tox
= 30 Å N
A = 2 x 1017 cm−3
Vg = 0.25 V
0.5
1.0
2.0
3.0
Schrödinger−PoissonEffective potential corrected MC
Effective potential for MOS
2002 School on Computational Material Science May 21-31, 2002
...192024 5
5
5
54
3
3
3
32
+∂∂
∂∂
+∂∂
∂∂
−
∂∂
∂∂
+∂∂
∂∂
−=∂∂
pf
xH
pf
xH
pf
xH
xf
pH
tf
ww
www
Wigner Formulation
• The Wigner method was developed in the 1930sfor quantum correction to statistical mechanics
• Wigner formulation of quantum mechanicsformally separates the quantum and classicalcontributions to the equation of motion
QuantumContribution
ClassicalBoltzmann equation
2002 School on Computational Material Science May 21-31, 2002
• We start from the general Wigner function representation ofquantum transport
where
is the density matrix.
• The Wigner function is the quantum equivalent of thedistribution function in the semi-classical Boltzmannequation.
Quantum corrections for Monte Carlo simulation
( ) ( )31, 2, 2j p y
wf r p dy e x y x yr◊= + -Ú
( ), 'x xr
2002 School on Computational Material Science May 21-31, 2002
• The quantum transport equation of the Wigner function hasthe form (parabolic bands, ballistic)
• Direct solution of the Wigner transport equation is still aconsiderable numerical challenge.
• We are interested in determining a truncated expansion ofthe quantum equation, that resembles the standardBoltzmann equation, so that the standard Monte Carlotechnique can be applied with minor modifications.
Quantum corrections for Monte Carlo simulation
3
* 21
( , ) ( , , )
( , ) sin( )2 2
ww w
f pf d s dP K s P f r p P t
t m
s sK s P r U r s PU
p
∂+ ◊ — = +
∂
= + - - ◊
Ê ˆÁ ˜Ë ¯
È Ê ˆ Ê ˆ ˘Í ˙Ë ¯ Ë ¯Î ˚
ÚÚ
2002 School on Computational Material Science May 21-31, 2002
• The complete Wigner transport equation, inclusive of collisionterms, can be reformulated to resemble Boltzmann equation
• At first order, we truncate considering only α = 1.
Quantum corrections for Monte Carlo simulation
( )( )
( )
r r k
12 1
r k1
1v
1
4 2 1 ! C
ff U f
t
fU f
t
aa
aa a
+•+
=
∂+ ◊ — - — ◊ —
∂
- ∂+ — ◊ — =
∂+
Ê ˆÁ ˜Ë ¯Â
2002 School on Computational Material Science May 21-31, 2002
• The truncated equation has a form resembling Boltzmannequation
where Fqc contains the quantum correction to the forces.
• With the modified forces, the particles move as if under theinfluence of a classical potential, but following equivalentquantum trajectories.
• The quantum correction essentially modifies the potential energyfelt by the particles.
Quantum corrections for Monte Carlo simulation
r k1v qc
C
f ff F ft t
∂ ∂Ê ˆ+ ◊— - ◊— = Á ˜Ë ¯∂ ∂
2002 School on Computational Material Science May 21-31, 2002
• The corrected forces can be evaluated by making assumptions ondistribution function and bandstructure. For
Quantum corrections for Monte Carlo simulation
( ) ( )
( ) ( )
2 22 22 22 2
2 22 22 22 2
2
1 3 324
1 3 324
with ;
qcx x x x y y y yx
qcy y y y x x x xy
x yx B
U UF U k k k kx x y
U UF U k k k ky y x
m k T
g g g g
g g g g
g g
Ê ˆÏ ¸È ˘ È ˘Ô ÔÁ ˜Í ˙ Í ˙Ì ˝Á ˜Í ˙ Í ˙Ô ÔÁ ˜Î ˚ Î ˚Ó ˛Ë ¯
Ê ˆÏ ¸È ˘ È ˘Ô ÔÁ ˜Í ˙ Í ˙Ì ˝Á ˜Í ˙ Í ˙Ô ÔÁ ˜Î ˚ Î ˚Ó ˛Ë ¯
∂ ∂ ∂=- - - - + - -∂ ∂ ∂
∂ ∂ ∂=- - - - + - -∂ ∂ ∂
= =2
y Bm k T
k k1exp (r) fB
f E U Ek T -
Ï ¸È ˘= - + -Ì ˝Î ˚Ó ˛
( )
( )
22
k k, ,
momentum centroid
2, ,
i i
ii x y z
x y z
k kE
mk k k k
-=
-=
= =
Â
Displaced Maxwellian Parabolic bands
2002 School on Computational Material Science May 21-31, 2002
• The forces obtained have still some practical problems in regionslike sharp interfaces, where quantum effects are prominent.
• To obtain a smooth potential, we can use approximate relationsobtained by integrating the displaced Maxwellian distribution withthe momentum.
• We obtain these alternative expressions for the second orderderivatives of the potential
Quantum corrections for Monte Carlo simulation
( ) ( )2 22 2
2 2 2 2ln ln;B B
n nU Uk T k Tx x y y
∂ ∂∂ ∂- -∂ ∂ ∂ ∂
2002 School on Computational Material Science May 21-31, 2002
• The “smooth” version of the quantum corrected forces is
• This formulation has explicit momentum dependence andimproves upon previous results in the literature where themomentum terms were evaluated with the thermal energy
Quantum corrections for Monte Carlo simulation
( ) ( )
( ) ( )
2 2222 22 2
2 22 22 22 2
ln ln )3 3
24
ln ln3 3
24
qc Bx x x x x y y y y
qc By y y y y x x x x
k T n nF U k k k k
x x y
k T n nF U k k k k
y y x
g g g g
g g g g
∂ ∂ ∂= - - - - + - -
∂ ∂ ∂
∂ ∂ ∂= - - - - + - -
∂ ∂ ∂
Ê ˆÏ ¸Ô ÔÈ ˘È ˘Ì ˝Á ˜Í ˙Î ˚ Î ˚Ë ¯Ô ÔÓ ˛Ê ˆÏ ¸Ô ÔÈ ˘ È ˘Ì ˝Á ˜Í ˙ Î ˚Î ˚Ë ¯Ô ÔÓ ˛
( )22
2 2i i B
i
k k k Tm-
=
2002 School on Computational Material Science May 21-31, 2002
• Test of the quantum corrections in Monte Carlo:Single GaAs/AlGaAs/GaAs barrier with a fixed potential (1)
Quantum corrections for Monte Carlo simulation
2002 School on Computational Material Science May 21-31, 2002
• Single GaAs/AlGaAs/GaAs barrier with a fixed potential (2)
Quantum corrections for Monte Carlo simulation
2002 School on Computational Material Science May 21-31, 2002
• Single GaAs/AlGaAs/GaAs barrier with a fixed potential (3)
Quantum corrections for Monte Carlo simulation
2002 School on Computational Material Science May 21-31, 2002
• Single GaAs/AlGaAs/GaAs barrier with a fixed potential (4)
Quantum corrections for Monte Carlo simulation
2002 School on Computational Material Science May 21-31, 2002
Wigner Correction
• Approximate quantum series– Truncate to the first non-classical term– Assume displaced maxwellian– Parabolic bands
• Leads to a momentum-dependent correction
• Possible to avoid last two approximations bymaking use of Monte Carlo full band
2002 School on Computational Material Science May 21-31, 2002
• Averaging out the momentum-dependence
• Feynman showed this can also be derived as anapproximation to the effective potential
• Further approximation leads to the densitygradient method in drift-diffusion
Simplified Wigner Correction
)ln(12
2*
2
nm
VV ∇−→
2002 School on Computational Material Science May 21-31, 2002
Properties of Wigner Correction
• Requires long run times due to Monte Carlonoise in ∇2ln(n) and restricts grid spacing
• Works well in drift-diffusion where noise isnot an issue
• Unlike effective potential, ∇2ln(n) is local
• For MOS, a single fitting parameter was foundto adjust the correction at the oxide interface
• Requires no fitting in the silicon region
2002 School on Computational Material Science May 21-31, 2002
0 2 4 6 8 10
1017
1018
1019
1020
Ele
ctro
n co
ncen
trat
ion
(cm
−3 )
Depth below interface (nm)
Schroedinger−PoissonMonte Carlo
Tox
= 3nmN
A = 2 x 1017 cm −3
Vg = 0.25 V
0.5
2.0
1.0
3.0
Wigner-corrected MOS Inversion
Wigner correction is accurate across range ofbiases—empirical nox = 1e15 cm-3 to fit interface
2002 School on Computational Material Science May 21-31, 2002
Wigner-corrected MOS Accumulation
nox = 1e15 cm-3
0 2 4 6 8 1010
17
1018
1019
1020
Ele
ctro
n co
ncen
trat
ion
(cm
−3 )
Depth below interface (nm)
Schroedinger−PoissonMonte Carlo
Tox
= 3 nmN
D = 2 x 1017 cm−3
Vg = 0.25 V
0.5
1.0
2.0
3.0
2002 School on Computational Material Science May 21-31, 2002
Schrödinger-Based Correction
• Treats quantum effects in the directionperpendicular to transport
• Accurate• No fitting parameters• Not sensitive to noise in the Monte Carlo
concentration estimator• Efficient, additional computation time is small
2002 School on Computational Material Science May 21-31, 2002
Applying Schrödinger Correction
• Schrödinger equation is solved along 1D slices ofthe 2-D domain
• Self-consistent Monte Carlo potential is the inputto Schrödinger and quantum density, nq, is output
• Concentration is linked to the correction with aBoltzmann dependence
{ } tqcp kTzVzVq ezn /)()()( +−∝
2002 School on Computational Material Science May 21-31, 2002
Consistent with Non-equilibrium Transport
• Schrödinger energy levels/wavefunctions are filledon a Boltzmann distribution
• Within each slice, the correction forces the shapeof the quantum density onto Monte Carlo
• No Fermi level is required• Relative concentration between the slices is
determined naturally by Monte Carlo transport
2002 School on Computational Material Science May 21-31, 2002
Degenerate Statistics in MOSDegeneracy can be important in highly invertedMOS structures
0 0.5 1 1.5 2 2.5 3−4
−3
−2
−1
0
1
2
3
4
5
6
Sub
band
Ene
rgy
(eV
/kT
)
Gate potential (V)
Tox
= 30 Å N
A = 2 x 1017 cm−3
Fermi energySubband 0Subbands 1,2Subband 3
2002 School on Computational Material Science May 21-31, 2002
0 0.5 1 1.5 2
1019
1020
Vg = 3.0 V
Tox
= 30 Å N
A = 2 x 1017 cm−3
Con
cent
ratio
n (c
m−
3 )
Depth below interface (nm)
Quantum−corrected MCDegenerate statisticsNon−degenerate statistics
Correction Satisfies DegeneracyVqc(z) is only a function of the relative nq(z) andthe first three subbands have a similar shape
2002 School on Computational Material Science May 21-31, 2002
Quantum-corrected Simulation Flow
Potential
Energy levels
Quantum density
Quantum correction
Move Particles inCorrected Potential
ρε −=∇⋅∇ pV
)( qcpr VVqdtkd
+∇=
ψψψ EVdzd
mdzd
p =+
− *
2 12
∑ −∝E
kTEq
tezEzn /2),()( ψ
( ) opqt VzVznkT +−− )()(log
2002 School on Computational Material Science May 21-31, 2002
Schrödinger-corrected MOS Inversion
0 1 2 3 4 5 610
16
1017
1018
1019
1020
Con
cent
ratio
n (c
m−
3 )
Depth below interface (nm)
Tox
= 30 Å N
A = 2 x 1017 cm−3
Vg = 0.25 V
0.5
1.0
2.0
3.0
Schrödinger−PoissonMonte Carlo
2002 School on Computational Material Science May 21-31, 2002
Schrödinger-corrected MOS Accumulation
0 1 2 3 4 5 610
17
1018
1019
1020
Con
cent
ratio
n (c
m−
3 )
Depth below interface (nm)
Tox
= 30 Å N
D = 2 x 1017 cm−3
Vg = 0.25 V
0.5
1.0
2.0
3.0
Schrödinger−PoissonMonte Carlo
2002 School on Computational Material Science May 21-31, 2002
Schrödinger-corrected Double-gate
−20 −15 −10 −5 0 5 10 15 2010
17
1018
1019
1020
Con
cent
ratio
n (c
m−
3 )
Depth direction (nm)
Vgs
= 0.6 V T
ox = 20 Å
NA = 1 x 1015 cm−3
Tsi
= 5 nm 10 nm 20 nm 40 nm
Schrödinger−PoissonMonte Carlo
2002 School on Computational Material Science May 21-31, 2002
Schrödinger vs Other Corrections for MOS
• Typical behavior of different corrections for MOS
0 0.5 1 1.5 2 2.5 3 3.5 4−0.2
0
0.2
0.4
0.6
0.8
1
Depth below interface (nm)
Pot
entia
l (V
) Vg = 1.0 V
Tox
= 30 Å N
A = 2 x 1017 cm−3
Schrödinger−Poisson potential Ideal correctionSchrödinger correctionWigner correction Effective potential, a = 5 Å
, nox
= 1e15 cm−3
2002 School on Computational Material Science May 21-31, 2002
Schrödinger vs Other Corrections for MOS
0 0.5 1 1.5 2 2.5 3 3.5 410
17
1018
1019
Depth below interface (nm)
Con
cent
ratio
n (c
m−
3 )
Vg = 1.0 V
Tox
= 30 Å N
A = 2 x 1017 cm−3
Schrödinger−Poisson Schrödinger−based correction Wigner correction Effective potential, a = 5 Å
, nox
= 1e15 cm−3
2002 School on Computational Material Science May 21-31, 2002
Adding Tunneling to Monte Carlo
• Calculation of transfer matrix tunneling probabilityfor each Monte Carlo particle is accurate
• Implementation of transfer matrix is cumbersomeand can be inefficient
• Effective potential is simple and gives reasonableresults for current through a small barrier
• Useful in current project to extend MOCA to SiGeor could be applied to source/channel tunneling
2002 School on Computational Material Science May 21-31, 2002
Effective Potential for Tunneling
103
104
105
106
100
101
102
103
104
105
106
107
J (A
/cm
2 )
Field (V/cm)
T = 300 Ka = 14 Å
φb = 0.1 eV
φb = 0.2 eV
φb = 0.4 eV
Transfer MatrixEffective Potential
2002 School on Computational Material Science May 21-31, 2002
Extending Schrödinger Correction to Devices
• Heating occurs in the direction ⊥ to transport• Cannot make use of an electron “temperature”
because it is not well-defined for non-equilibrium• Define a “transverse” temperature Tt to describe
the variation of the concentration with potential inthe ⊥ direction
• Tt is validated if a single temperature at each pointalong the transport path accurately describes thevariation
2002 School on Computational Material Science May 21-31, 2002
0 0.5 1 1.5 210
18
1019
1020
Con
cent
ratio
n (c
m−
3 )
Depth below interface (nm)
Source (drain is similar)
Beginning of channel
Middle of channel
End of channel
ConcentrationTemperature derived concentration
Validating Transverse Temperature
Tt(z) accurately describes potential→concentrationfor highly non-equilibrium transport in 25nm FET
2002 School on Computational Material Science May 21-31, 2002
Typical Transverse Temperature
• Transverse temperature for a 25-nm MOSFET insaturation bias
30 35 40 45 50 55 60 6510
2
103
104
Distance along channel (nm)
Tem
pera
ture
(K
)
Source Channel Drain
Average Energy/kBTransverse Temperature
2002 School on Computational Material Science May 21-31, 2002
Effect of Heating in the ⊥ Direction
0 0.5 1 1.5 2 2.5 3
1018
1019
1020
Con
cent
ratio
n (c
m−
3 )
Depth below interface (nm)
Cut at source/channel transition
Cut at channel/drain transition
Transverse temperatureNo transverse temperatureNo quantum correction
2002 School on Computational Material Science May 21-31, 2002
104
105
106
102
103
Mob
ility
(cm
2 /Vs)
Effective Normal Field (V/cm)
Cooper and NelsonTakagi et. al.Monte Carlo
Surface Scattering
Surface scattering model of Yamakawa, usingroughness parameters obtained from experiment
2002 School on Computational Material Science May 21-31, 2002
25 nm MOSFET – concentration
5e+16
5e+17
5e+18
1e+19
2e+19
4e+19
8e+19
2e+20
20 30 40 50 60 70
0
1
2
3
4
5
6
Lateral direction (nm)
Dep
th d
irect
ion
(nm
)
Leff = 25 nmTox = 15 ÅVg = 1 VVd = 1 V
2002 School on Computational Material Science May 21-31, 2002
25 nm MOSFET – corrected potential
−1
−0.9
−0.8
−0.7
−0.6
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
20 30 40 50 60 70
0
1
2
3
4
5
6
7
8
9
10
Lateral direction (nm)
Dep
th d
irect
ion
(nm
)
Leff = 25 nmTox = 15 ÅVg = 1 VVd = 1 V
2002 School on Computational Material Science May 21-31, 2002
Properties along transport path
20 30 40 50 60 700
0.5
1
1.5
2
2.5
3x 10
7
0
0.12
0.23
0.35
0.47
0.58
0.7A
verage energy (eV)
Ave
rage
vel
ocity
(cm
/s)
Distance along channel (nm)
S−MCMC only
2002 School on Computational Material Science May 21-31, 2002
Conclusions• Quantum corrections can be used to extend the validity of
Monte Carlo device simulation to the 10-nm regime• Wigner-based corrections
– accurate, and momentum-dependent is interesting– somewhat impractical due to noise in Monte Carlo
• Effective potential– simple and fast– accurate for small heterojunctions
• Schrödinger-based correction– accurate and efficient with no fitting parameters– best choice for size quantization effects