EE382V Fall 2006

VLSI Physical Design Automation

Lecture 5. Circuit Partitioning (III)

Spectral and Flow

Prof. David Pan

dpan@ece.utexas.edu
Office: ACES 5.434

10/18/08 1
Recap of what we have learned
❁KL Algorithm
❁FM algorithm
❁ Variation and Extension
❁Multilevel partition (hMetis)
❁Simulated Annealing

2
 Partitioning Algorithms

❁Two elegant partition algorithms

❁ although not the fastest
❁Learn how to formulate the problem!
❁=> Key to VLSI CAD
(1) Spectral based partitioning algorithms
(2) Max-flow based partition algorithm

3
Spectral Based Partitioning Algorithms

a c a b c d a b c d
a 0 1 0 3 a 4 0 0 0
3 3
1 b 1 0 0 4 b 0 5 0 0
A = D =
c 0 0 0 3 c 0 0 3 0
b 4 d d 0 0 0 10
d 3 4 3 0

D: degree matrix; A: adjacency matrix; D-A: Laplacian matrix

Eigenvectors of D-A form the Laplacian spectrum of G

4
Eigenvalues and Eigenvectors

A x Ax
 x1   a11 x1 + a12 x2 +  + a1n xn 
 a11 a12  a1n    
    =  
... 
 an1 an 2  ann  xn   an1 x1 + an 2 x2 +  + ann xn 


If Ax=λx
then λ is an eigenvalue of A
x is an eignevector of A w.r.t. λ
(note that Kx is also a eigenvector, for any constant K).

5
 A Basic Property

 a11  a1n  x1 
  
xTAx = (x1 , , xn )    
a  
 n1  ann  xn 
 x1 
n n
 
=  ∑xi ai1,  ∑xi ain,   
 i =1 i =1  
 xn 
= ∑x x a i j ij
i, j

6
 Basic Idea of Laplacian Spectrum Based Graph
Partitioning
❁Given a bisection ( X, X’ ), define a partitioning vector

−1 i ∈X
= =
x ( x1 , x2 , , xn ) s.t. xi 
1 i ∈ X'

clearly, x ⊥ 1, x ≠ 0 ( 1 = (1, 1, …, 1 ), 0 =(0, 0, …, 0))

❁For a partition vector x: ∑ (x

( i , j )∈ E
i − xj )2= 4 * C(X, X’)

❁Let S = {x | x ⊥ 1 and x ≠ 0}. Finding best partition vector x

such that the total edge cut C(X,X’) is minimum is relaxed to finding
x in S such that ∑ ( xi − xj )2 is minimum
( i , j )∈ E

❁Linear placement interpretation: ….

minimize total squared wirelength x1 x2 x3 xn

7
 Property of Laplacian Matrix

(1) x Qx =
T
∑Q x x ij i j

= xT Dx − xT Ax
= ∑d x i i
2
− ∑Aij xi xj
….
= ∑d x 2
i i
− ∑2 a x x ij i j
( i , j )∈E x1 x2 x3 xn
= ∑( x i
− xj )2 squared wirelength
( i , j ) ∈E

= 4 * C(X, X’) If x is a partition vector

T
Therefore, we want to minimize
x Qx

8
 Property of Laplacian Matrix (Cont’d)

( 2 ) Q is symmetric and semi-definite, i.e.

(i) x Qx =
T
∑Q x x
ij i j
≥0
(ii) all eigenvalues of Q are ≥0
( 3 ) The smallest eigenvalue of Q is 0
corresponding eigenvector of Q is x0= (1, 1, …, 1 )
(not interesting, all modules overlap and x0∉S )
( 4 ) According to Courant-Fischer minimax principle:
the 2nd smallest eigenvalue satisfies:
xT Qx
λ = min 2
x in | x |
S

9
Results on Spectral Based Graph Partitioning

❁ Min bisection cost c(x, x’) ≥ n⋅λ/4

❁ Min ratio-cut cost c(x,x’)/|x|⋅|x’| ≥ λ/n
❁ The second smallest eigenvalue gives the
best linear placement
❁ Compute the best bisection or ratio-cut
based on the second smallest
eigenvector

10
Computing the Eigenvector

❁Only need one eigenvector

(the second smallest one)

❁Q is symmetric and sparse

❁Use block Lanczos Algorithm

11
Interpreting the Eigenvector

❁Linear Ordering of modules

❁Try all splitting points, choose the best one
using bisection of ratio-cut metric

23 10 22 34 6 21 8

23 10 22 34 6 21 8

12
Experiments on Special Graphs

❁GBUI(2n, d, b) [Bui-Chaudhurl-Leighton 1987]

2n nodes
n n d-regular
min-bisection≈b

GBUI Results
Cluster 2
Value of Xi

Module i

Cluster 1
13
Some Applications of Laplacian Spectrum

❁Placement and floorplan

[Hall 1970]
[Otten 1982]
[Frankle-Karp 1986]
[Tsay-Kuh 1986]

❁Bisection lower bound and computation

[Donath-Hoffman 1973]
[Barnes 1982]
[Boppana 1987]

❁Ratio-cut lower bound and computation

[Hagen-Kahng 1991]
[Cong-Hagen-Kahng 1992]

14