Efficient Numerical Modeling of Random Rough Surface ... · 15 Limitations of Monte-Carlo...
Transcript of Efficient Numerical Modeling of Random Rough Surface ... · 15 Limitations of Monte-Carlo...
Efficient Numerical Modeling of Random Rough Surface Effects
in Interconnect Internal Impedance Extraction
CHEN CHEN QuanQuan & WONG & WONG NgaiNgai
Department of Electrical & Electronic EngineeringDepartment of Electrical & Electronic Engineering
The University of Hong KongThe University of Hong Kong
2
OutlineOutlineBackground
Modeling (Effective Parameters)
Computation (Modified SIE Method)
Results
Conclusion
3
Surface Roughness in InterconnectsSurface Roughness in Interconnects
4
Sources of Surface RoughnessSources of Surface RoughnessUnintentional sourcesUnintentional sources–– Technology limitationsTechnology limitations–– Process variationsProcess variations
Intentional sourcesIntentional sources–– Electronic deposition Electronic deposition –– Chemical etchingChemical etching–– AnnealingAnnealing
Enhance the cohesion between metal and Enhance the cohesion between metal and dielectricdielectric
5
Impact of Surface Roughness on Internal Impedance
Impact of Surface Roughness on Internal Impedance
Current under smooth surface Current under rough surface
More resistive lossLonger
current pathLargercurrent loop
Higher internal inductance
Higher resistance
Interaction between rough surface and currentInteraction between rough surface and current
6
High Frequency EffectsHigh Frequency Effects•• Rough surface effects is insignificant in low Rough surface effects is insignificant in low frequencies frequencies (large skin depth, small roughness)(large skin depth, small roughness)
•• It becomes significant in high frequencies It becomes significant in high frequencies (comparable skin depth and roughness)(comparable skin depth and roughness)
Skin depth = 6.5microns when f = 0.1 GHz (From Intel)
7
ModelingModeling
Model the impact of random rough surface on interconnect internal
impedance
8
Effective Resistivity Effective Resistivity ρρee & Effective Permeability & Effective Permeability μμee
Capture the increase of resistance and internal Capture the increase of resistance and internal inductance caused by surface roughnessinductance caused by surface roughness
Current Extraction Tools
Layout Information
Rough Surface Module
Effective Resistivity & PermeabilityRough Surface
RLC Circuit
Effective ParametersEffective Parameters
9
Analytical FormulationAnalytical FormulationFor effective resistivity
–– h h –– RMS height; RMS height; δδ-- skin depthskin depth
–– Widely used in practical design, BUTWidely used in practical design, BUT
–– Inaccurate Inaccurate (only h is considered)(only h is considered)
For effective permeability–Unavailable
2121 tan 1 .4e
hρ ρπ δ
−⎡ ⎤⎛ ⎞= +⎢ ⎥⎜ ⎟
⎝ ⎠⎢ ⎥⎣ ⎦
10
Numerical Formulation of ρeNumerical Formulation of ρe
Power loss equivalence Ps = Pr
{ }0
20
Re d ( )2 rSe zU z
HP
H l lρδδρ == ∫
Smooth surface power loss2
0
2s
H lP
ρδ
=
Rough surface power loss
( ){ }*0 Re d
2r S
HP zU zρ= ∫
V
11
Numerical Formulation of μeNumerical Formulation of μe
{ }2 *
0 Im d ( )2r S
HW zU zμδ= ∫
{ }200
Im d ( )2 rSe
WH
zUlH
zl
μδμδ
== ∫
20
2s
H lW
μδ=
Smooth surface magnetic energy
Rough surface magnetic energy
Magnetic energy equivalence Ws = Wr
V
12
Governing EquationGoverning Equation2
0 01 ( ) ( , ')d ( , ') ( ) d 12S cp
y z G z zzG z z U z H H zz n
∂ ∂⎛ ⎞= + + ⎜ ⎟∂ ∂⎝ ⎠∫ ∫
2( ) ( )( ) 1ˆ
y z H zU zz n
∂ ∂⎛ ⎞= + ⎜ ⎟∂ ∂⎝ ⎠
0( )H r H r S= ∈Boundary condition
(1)0 1( , ') ( ' )G z z H k z z= −
Surface unknown
Green’s function
13
Modeling of Random Rough SurfaceModeling of Random Rough Surface
Described by Described by stationary stationary stochastic processstochastic process with with Probability Density Probability Density FunctionFunction and and Correlation FunctionCorrelation Function
2
1 2
1P ( ) exp( )22zyhhπ
= −
η --- correlation length
21 2
1 2 2( , ) exp( )g
z zC z z
η−
= −
Most rough surfaces in reality are randomMost rough surfaces in reality are random
14
Conventional Statistical Solver-- Monte-Carlo Method
Conventional Statistical Solver-- Monte-Carlo Method
Input Layout information
Random Rough Surface Generator
Direct Integral Equation Solver
Computation Output Mean of the solution
S1 S2 S3 Sn-1 SnA large number of
surface realizations
R1 R2 R3 Rn-1 Rn
Corresponding solutions
15
Limitations of Monte-Carlo methodLimitations of Monte-Carlo methodProbabilistic natureProbabilistic nature–– Mean value is also a random variableMean value is also a random variable
Slow convergenceSlow convergence–– More than 2500 More than 2500
runs to converge runs to converge within 1%within 1%
0 500 1000 1500 2000 2500 3000 3500 40000
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Number of Monte−Carlo runs
Co
nve
rgen
ce o
f m
ean
of ρ
e (%
)
16
ComputationComputationEfficient Stochastic Integral Equation
(SIE) method
17
Stochastic Integral EquationStochastic Integral Equation (SIE)(SIE) methodmethod–– Use mean value as unknownUse mean value as unknown–– OneOne--pass solution pass solution –– Deterministic natureDeterministic nature
Two stepsTwo steps– Zeroth-order approximation
(Uncorrelatedness assumption )– Second-order correction
18
Zeroth-Order ApproximationZeroth-Order ApproximationDirectly applying ensemble average on both sides
2
0 01 ( ) ( , ')d ( , ') ( ) d 12S cp
y z G z zz G z z U z H H zz n
∂ ∂⎛ ⎞= + + ⎜ ⎟∂ ∂⎝ ⎠∫ ∫
1( ) ( ( )) ( )f x P f x f x+∞
−∞= ∫Ensemble average
2
0 01 ( ) ( , ')d ( , ') ( ) d 12S cp
y z G z zz G z z U z H H zz n
∂ ∂⎛ ⎞= + + ⎜ ⎟∂ ∂⎝ ⎠∫ ∫
Assuming the Green’s function G and the surface unknown U are statistically independent (Uncorrelatedness Assumption)
New unknown
19
Inaccurate zeroth-order approximation
Zeroth-order ApproximationZeroth-order ApproximationMatrix equation format
AU L=
(0) 1U A L−=Cause: uncorrelatedness assumption
( , ') ( ) ( , ') ( )G z z U z G z z U z≠
(0) (0)TV L U=
2( , ) d d ( , ) ( , ; , )ik i k i k i k i k i kA G z z y y P y y G y y z z= = ∫∫
Deterministic quantities
( , ')G z z
( )U z
20
Primitive Second-Order CorrectionPrimitive Second-Order CorrectionImproveImprove the accuracy of the mean Vthe accuracy of the mean V
(0) (2)V V V= +Second-order
correction term(2) ( )TV trace A D−=
(0) (0)( ) ( ) ( ) ( )Tvec D A A A A U U= − ⊗ − ⊗
Zeroth-order approximation term
Limit: Also timeLimit: Also time--consumingconsuming
2 2N NF
×
21
Computational BottleneckComputational BottleneckHigh dimensional infinite integration in High dimensional infinite integration in FF
iknm ik mn ik mnF A A A A= −
4d d d d ( , , , ) ( , ) ( , )ik mn i k m n i k m n i k m nA A y y y y P y y y y G y y G y y= ∫∫ ∫∫Standard technique: Gauss Hermite quadrature– Complexity grows exponentially with the integral
dimension (Curse of dimensionality)
4-D infinite integration
( )1 dimension O N− →( )44 dimension O N− →
22
5 10 15 20 25 307
7.5
8
8.5
9
9.5
10
10.5
11
11.5x 10
−5
Number of quadrature points
|<A
ijAm
n>|
Gauss HermiteAccurate Value
More than 30quadrature points for 1D integration
(81000 points for 4D)
No. of Points for Gauss Hermite QuadratureNo. of Points for Gauss Hermite Quadrature
23
Improved Formulation of SIE (2D case)Improved Formulation of SIE (2D case)Translation invariance of GreenTranslation invariance of Green’’s functions function
2( , ) d d P ( , ) ( , )) (2Dik i k i k i k i kA G y y y y y y G y y= = ∫∫
( , ) ( , ) (0, )i k i i d dG y y G y y y G y= + =
1dˆd P ( ) ( ) (1( ) )ˆ Dd d dik d y y G yA G y= = ∫
ˆ( , ) ( )i j dG y y G y=
PP1d 1d –– Probability density function of yProbability density function of ydd
ThusThus
24
Partial Probability Density FunctionPartial Probability Density Function
2 2
2 2 22 2
2 ( ) ( )1P ( , ) exp2 (1 )2 1
i i i d i di d
y cy y y y yy yh ch cπ
⎛ ⎞− + + += −⎜ ⎟−− ⎝ ⎠
2 2
2 2 22 2
21( , ) exp2 (1 )2 1
i i k ki k
y cy y yP y yh ch cπ
⎛ ⎞− += −⎜ ⎟−− ⎝ ⎠
( ) ( )
1d 2
2
2
P ( ) d P ( , )
1 exp4 12 1
d i i i d
d
y y y y y
yh ch cπ
= +
⎛ ⎞−= ⎜ ⎟⎜ ⎟−⎝ ⎠
∫
-
25
Improved Formulation of SIE (4D case)Improved Formulation of SIE (4D case)
4d d d d ( , , , ) ( , ) ( , )ik mn i k m n i k m n i k m nA A y y y y P y y y y G y y G y y= ∫∫ ∫∫
1 1 1 2 1 22dˆ ˆd d P ( , ) ( ) ( )d d d d d dik mn y y y yA A G y G y= ∫∫
1 2d k i d n my y y y y y= − = −LetLet
1 2 1 2 1 22d 4P ( , ) d d P ( , , , )d d d d i i d m m dy y y y y y y y y y= + +∫WhereWhere
26
ResultsResults
27
Numerical VS Analytical Formulation(Different Correlation Length)
Numerical VS Analytical Formulation(Different Correlation Length)
28
SIE VS MIE (Mean ρe Ratio)SIE VS MIE (Mean ρe Ratio)
Gaussian surface (σ= 1μm)
29
SIE VS MIE (Mean μe Ratio)SIE VS MIE (Mean μe Ratio)
Gaussian surface (σ= 1μm)
30
CPU Time Comparison (Unit: second)
CPU Time Comparison (Unit: second)
MethodMethodh /h /δδ= 1= 1 h /h /δδ= 2= 2
MIEMIE
(1500run)(1500run)7160.37160.3 14484.714484.7
SIESIE
(Original)(Original)6367.36367.3 6544.76544.7
SIESIE
(Modified)(Modified)198.7198.7 200.3200.3
(Gaussian rough surface (Gaussian rough surface ------ σσ=1=1μμm, m, ηη=1 =1 μμm)m)
32X
31
Conclusion
An efficient numerical approach for modeling the impact of surface
roughness on interconnect internal impedance
Conclusion
An efficient numerical approach for modeling the impact of surface
roughness on interconnect internal impedance
32
Numerical effective parameters–Model the rough surface effects on internal
impedance –Take all statistical information into account
Modified SIE method–One-pass solution for mean values–Halve infinite integral dimension by partial
PDF formulation
ConclusionConclusion
33