a benchmark study of the wigner monte-carlo method

27
A benchmark study of the Wigner Monte-Carlo method J.M. Sellier * , M. Nedjalkov ** , I.Dimov * , S. Selberherr ** * Bulgarian Academy of Sciences, IICT, Sofia ** TU Wien, Inst. Microelec., Vienna LSSC 2013 [email protected]

Upload: others

Post on 08-Nov-2021

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A benchmark study of the Wigner Monte-Carlo method

A benchmark study of the Wigner

Monte-Carlo

method

J.M. Sellier *, M. Nedjalkov **, I.Dimov *, S. Selberherr **

* Bulgarian Academy of Sciences, IICT, Sofia** TU Wien, Inst. Microelec., Vienna

LSSC 2013

[email protected]

Page 2: A benchmark study of the Wigner Monte-Carlo method

Topics

• Motivations

• The Wigner equation and its semi-discrete form

• A Monte Carlo approach based on particles sign

• The annihilation technique

• Wigner vs. Schroedinger benchmark test

Page 3: A benchmark study of the Wigner Monte-Carlo method

Motivations

Page 4: A benchmark study of the Wigner Monte-Carlo method
Page 5: A benchmark study of the Wigner Monte-Carlo method

Quantum Effects – First set of

challenges• When device dimensions are reduced, quantum effects start

to appear.

• Particles can tunnel through barriers, energies are discretized.

• The behavior of an electron is more similar to a wave than to• The behavior of an electron is more similar to a wave than toa particle, the dynamics is described by Schroedingerequation.

• Boltzmann equation does not predict quantum effects.

• A new approach must be taken.

Page 6: A benchmark study of the Wigner Monte-Carlo method

Quantum effects – Second set of

challenges

• Scattering effects are relevant even in nanodevices

• Full-Quantum approach

Time-dependent phenomena• Time-dependent phenomena

• Self-Consistent Calculations

• A simulator that possibly runs on a common machine in reasonable run-times

Page 7: A benchmark study of the Wigner Monte-Carlo method

The Wigner Equation and its

Semi-Discrete FormSemi-Discrete Form

Page 8: A benchmark study of the Wigner Monte-Carlo method

The Wigner equation

• The Wigner equation reads

• where

( ) ( ) ( ) ( )tkrQftkrfkt

tkrfWWrk

W ,,,,1,, rrrrr

h

rr

rr =∇∇+∂

∂ ε

• where

( ) ( ) ( )',',', krfkkrVkdkrQf WWW

rrrrrrrr −= ∫

( ) ( ) ∫

−−

+= ⋅−

2

'

2

''

2

1, ' r

rVr

rVerdi

krV rkidW

rr

rrr

h

rr rr

π

Page 9: A benchmark study of the Wigner Monte-Carlo method

Wigner equation in semi-discrete form

• Taking into account the fact that a semiconductor device

has limited dimensions, it is possible to re-formulate the

Wigner equation in a semi-discrete form.

( ) ( ) ( ) ( )tnMrfnrVtMrfkMmt

tMrfWWWr

W ,,,,,,,

*−=∇∆+

∂∂

∑+∞ rrrrh

r

r

• The phase space is discretized w.r.t. the pseudo-wave

vector coordinates.

( ) ( ) ( )tnMrfnrVtMrfkMmt W

nWWr ,,,,,

*−=∇∆+

∂ ∑−∞=

( ) ( ) ( )( )∫ −−+= ⋅∆−2/

0

211,

Lskm

W srVsrVesdLi

nrVrrrrr

h

r rr

Page 10: A benchmark study of the Wigner Monte-Carlo method

Comment

• The Wigner equation is a difficult task in a

finite difference framework.

• The distribution is known to be rapidly varying • The distribution is known to be rapidly varying

and the diffusion term cannot be calculated

correctly.

( )tkrfWr ,,rr

r∇

Page 11: A benchmark study of the Wigner Monte-Carlo method

A Monte-Carlo Approach Based on

Particles SignParticles Sign

Page 12: A benchmark study of the Wigner Monte-Carlo method

Wigner equation in integral form

• The semi-discrete Wigner equation can be re-

formulated in terms of an integro-differential

equation.

( ) ( )( ) ( )( )−=∫− ,0,, 0 mxfetmxf i

dyyx

W

trr

• where

( ) ( )( )

( )( ) ( ) ( )( ) ( ) ( )( ) ( )∫ ∑∞ ∞+

−∞=

−−−∫Γ

=−

0'

''''',,'',',''

,0,,

' xtxxttemmxtmtxfdt

mxfetmxf

Dm

dyyx

W

iW

t

trrrrr

r

θδθγ

( ) ( ) ( )∑∑+∞

−∞=

−+∞

−∞=

+ ==m

Wm

W mxVmxVx ,,rrrγ

( )( ) ( )( ) ( )( ) ( )( ) ','','','',,' mmWW txmmtxVmmtxVmmtx δγ rrrr +−−−=Γ ++

( ) ( )''*

ttm

kmxtx −∆−=

rhr

Page 13: A benchmark study of the Wigner Monte-Carlo method

Mean value of a function

• Finally, using the fact that the adjoint equation of the

integro-differential equation is a Fredholm integral

equation of second type, one can show that:

( ) ( ) ( ) ( )∑ ∑∫∫+∞ +∞∞

=−= W AtmxAtmxfxddtA ,,, τδτ rrr

• where (for example)

( ) ( ) ( ) ( )∑ ∑∫∫−∞= =

=−=m i

iW AtmxAtmxfxddtA0

0,,, τδτ

( ) ( ) ( )( ) ( )( )∑∫+∞

−∞=

−∫='

0',',' 0

mi

dyyx

ii mxAemxfxdAi ττ

τγrr

( ) ( ) ( )( ) ( )

( ) ( )( ) ( )( ) ( )τδ

θτ

γ

γ

−∫Γ⋅

⋅∫=

−∞+

−∞=

+∞

−∞=

−∞

∑∫

∑∫∫

ttmtxAemmxdt

xemxfdxdtA

t

t

t

i

dyyx

mt

mD

dyyx

iii

,,',,

,'

11'

'101

'1

'

0r

Page 14: A benchmark study of the Wigner Monte-Carlo method

Physical interpretation of the terms

• A particle starts at xi with momentum m’∆k at time 0. The exponent gives the probability that the particle remains on the trajectory, provided that the scattering rate isϒ.

• Consider as a particle generation rate.

The Wigner potential gives rise to the creation of two particles, one positive and one negative, and the sign carries the quantum information.

( ) ( ) ( )∑∑+∞

−∞=

−+∞

−∞=

+ ==m

Wm

W mxVmxVx ,,rrrγ

Page 15: A benchmark study of the Wigner Monte-Carlo method

Similarities in the 2 approaches

• Boltzmann • Wigner

( ) ( )∑+∞

−∞=

+=m

W mxVx ,rrγ

dtdttkWtkWdttPt

]'))'((exp[))(()(0∫−=

rr

∫= )',(')( kkSdkkWrr

dtdttxtxdttPt

]'))'((exp[))(()(0∫−= rr γγ

Page 16: A benchmark study of the Wigner Monte-Carlo method

The Annihilation Technique

Page 17: A benchmark study of the Wigner Monte-Carlo method
Page 18: A benchmark study of the Wigner Monte-Carlo method

Some comments

• The only unknown of the problem is the quasi

distribution function

• Particles with opposite signs in the same • Particles with opposite signs in the same

phase-space cell do not contribute to the

calculation of the unknown

• Created particles can annihilate (i.e. removed)

Page 19: A benchmark study of the Wigner Monte-Carlo method
Page 20: A benchmark study of the Wigner Monte-Carlo method
Page 21: A benchmark study of the Wigner Monte-Carlo method

Wigner vs. Schroedinger

Benchmark TestBenchmark Test

Page 22: A benchmark study of the Wigner Monte-Carlo method

The Schroedinger Equation

and its Discretization

Page 23: A benchmark study of the Wigner Monte-Carlo method

The Schroedinger Equation

and its Discretization

Page 24: A benchmark study of the Wigner Monte-Carlo method
Page 25: A benchmark study of the Wigner Monte-Carlo method
Page 26: A benchmark study of the Wigner Monte-Carlo method
Page 27: A benchmark study of the Wigner Monte-Carlo method