www.jntuworld.

com

Floorplanning
Inputs to the oorplanning problem:

D L

A set of blocks, xed or exible. Pin locations of xed blocks. A netlist.

Objectives: Minimize area, reduce wirelength for (critical) nets, maximize routability, determine shapes of exible blocks
7

w w
6 1

2 3
1

An optimal floorplan, in terms of area

A nonoptimal floorplan
1

www.jntuworld.com

N J
4 2

. w
5

u t jn
4

U T
7 5

r o w

. ld

m o c
3

www.jntuworld.com

Floorplan Design

D L

x Modules: Area: A=xy

w w

N J
3 1 5 e 2

. w
b

U T
c

u t jn
c

r o w

Aspect ratio: r <= y/x <= s

. ld
a

m o c
y 2 b 3 6

Rotation: Module connectivity

d 5

f
2

www.jntuworld.com

www.jntuworld.com

Floorplanning: Terminology
Rectangular dissection: Subdivision of a given rectangle by a nite # of horizontal and vertical line segments into a nite # of non-overlapping rectangles.

D L

1 2

4 6

w w
5 7

H
2 1 H
V

N J
H
3 2
V

. w
2

U T
V 6 7 4 5

Skewed slicing tree: One in which no node and its right child are the same.
V H
1 V
6 7 V

u t jn
5 7
6

Slicing tree: A binary tree, where each internal node represents a vertical cut line or horizontal cut line, and each leaf a basic rectangle.

Slicing structure: a rectangular dissection that can be obtained by repetitively subdividing rectangles horizontally or vertically.

r o w

. ld

m o c

H
H
3

Nonslicing floorplan

Slicing floorplan

A slicing tree (skewed) Another slicing tree (nonskewed)

3

4 5

www.jntuworld.com

www.jntuworld.com

Floorplan Design by Simulated Annealing

Related work

D L

Wong & Liu, A new algorithm for oorplan design, DAC86. Consider slicing oorplans.

Wong & Liu, Floorplan design for rectangular and L-shaped modules, ICCAD87. Also consider L-shaped modules.

Ingredients: solution space, neighborhood structure, cost function, annealing schedule?

N J

Wong, Leong, Liu, Simulated Annealing for VLSI Design, pp. 3171, Kluwer academic Publishers, 1988.

w w

. w

u t jn

U T

r o w

. ld

m o c

www.jntuworld.com

www.jntuworld.com

Solution Representation
An expression E = e1 e2 . . . e2n1 , where ei {1, 2, . . . , n, H, V }, 1 i 2n 1, is a Polish expression of length 2n 1 i 1. every operand j , 1 j n, appears exactly once in E ; 2. (the balloting property) for every subexpression Ei = e1 . . . ei, 1 i 2n 1, #operands > #operators.

D L

1 6 H 3 5 V 2 H V 7 4 H V # of operands = 4 ....... = 7 # of operators = 2 ....... = 5

Polish expression Postorder traversal. ijH : rectangle i on bottom of j ; ijV : rectangle i on the left of j .

w w

4
H

6 1 2 3
1

2 7 5

6 E = 16H2V75VH34HV

E = 16+2*75*+34+* Postorder traversal of a tree!

5

www.jntuworld.com

N J
H

. w

u t jn

U T
V

r o w

. ld
3 4

m o c

www.jntuworld.com

Solution Representation (contd)

D L

V
1
H 2 3

3 1 2

V
4 1

nonskewed!
Nonskewed cases

w w

....... HH ........

Question: How to eliminate ambiguous representation?

6

www.jntuworld.com

N J

n j . w

U T

w tu
H H

E = 123H4VV

or

. ld

m o c
H 2 3

E = 123HV4V

skewed!
V V

....... VV ........

www.jntuworld.com

Normalized Polish Expression

A Polish expression E = e1 e2 . . . e2n1 is called normalized i E has no consecutive operators of the same type (H or V ).

D L

Given a normalized Polish expression, we can construct a unique rectangular slicing structure.

7 6

5 2

w w
1

E = 16H2V75VH34HV
A normalized Polish expression

N J

. w3

u t jn

U T
H 2 7 6

r o w
V

. ld
V V 5

m o c
H 3 4

www.jntuworld.com

www.jntuworld.com

Neighborhood Structure
Chain: HV HV H . . . or V HV HV . . .

D L

1 6 H 3 5 V 2 H V 7 4 H V

3 types of moves:

M 2 (Chain Invert): Complement some chain (V = H, H = V ). M 3 (Operator/Operand Swap): Swap two adjacent operand and operator.

N J

M 1 (Operand Swap): Swap two adjacent operands.

w w

. w

U T

Adjacent: 1 and 6 are adjacent operands; 2 and 7 are adjacent operands; 5 and V are adjacent operand and operator.

u t jn

r o w
chain

. ld

m o c

www.jntuworld.com

www.jntuworld.com

Eects of Perturbation

D L

3 2 M2 1

4 3 1 2

M1

M 1 and M 2 moves are OK.

Check M 3 moves: Assume that the M3 move swaps the operand ei with the operator ei+1 , 1 i k 1. Then, the swap will not violate the balloting property i 2Ni+1 < i.
Nk : # of operators in the Polish expression E = e1 e2 . . . ek , 1 k 2n 1.

N J

Check the M 3 moves! Reject illegal M 3 moves.

w w

n j . w

U T

Question: The balloting property holds during the moves?

w tu

12V4H3V

12V3H4V

or

12H3H4V

. ld
4

m o c
2 M3 1

4 3

12H34HV

www.jntuworld.com

www.jntuworld.com

Cost Function
= A + W .
A: area of the smallest rectangle

D L

R O
r o w
3 2 M2 1

W : overall wiring length

: user-specied parameter
4 3 1 2 M1 1 3

. ld
4

m o c
2 1 4 3

cij : # of connections between blocks i and j .

dij : center-to-center distance between basic rectangles i and j .

N J
10

W =

ij cij dij .

w w

. w

u t jn
4 2

U T

M3

A: 12H34HV

www.jntuworld.com

www.jntuworld.com

Cost Evaluation: Shape Curves

Shape curves correspond to dierent kinds of constraints where the shaded areas are feasible regions.

D L

wi hi hi wi

y = six

hi

wi

xi >= a, yi >= b

w w

N J

Bounding area

j . w
y

y = six

U T
y = six
y

u t n
y = ri x
x

hi

W
corner points

wi

r o w

feasible region

. ld
x
x

m o c
i

1 y = six y = r ix y = r x

y = ri x

hi

1 x y=s
i

wi

wi hi

xi >= a, yi >= b or xi >= b, yi >= a

xi >= a, yi >= b xi yi >= A

xi >= a, yi >= b, xi yi >= A or xi >= b, yi >= a, xi yi >= A

(a) rigid, fixed orientation

(b) rigid, free (c) flexible, fixed (d) flexible, free orientation orientation orientation
11

www.jntuworld.com

www.jntuworld.com

Area Computation

D L

{ (5,5) (9,4) }

2
H
{ (3,5) (6,,4) }

{ (2,5) (3,4) } H

2 3

2 1

{ (6,2) (3,3) }

V
1

O
5

2
3

4 { (1,2) (2,1) } { (2,2) }

{ (1,3) (3,1) } { (2,3) (3,2) }

u1 v

u2

w w
u1 v

u2

u2 w

u1+u2

2
3

N J
V

. w

v+w
u1

u t jn

max{u1, u2}

U T
V
H
4

{ (2,3) (3,2) } { (2,2) }

r o w
6

V { (3,2) }

. ld

m o c
5 3 6 4
H

max{v, w}

Wiring cost?
12

www.jntuworld.com

www.jntuworld.com

Incremental Computation of Cost Function

Each move leads to only a minor modication of the Polish expression. At most two paths of the slicing tree need to be updated for each move.
H
V

D L

O
V

W
1

M1

N J

E = 12H34V56VHV

w w

. w

U T
3

u t jn
6

r o w
H

. ld
V
2

m o c
H
V
4
5

V 6

E = 12H35V46VHV

13

www.jntuworld.com

www.jntuworld.com

Incremental Computation of Cost Function

D L

(contd)
H

H
V

O
6

M2
1

H
2

2
3
4

w w
1 2

N J
V
5
H

. w
H
4

U T
V
1

E = 12H34V56VHV

u t jn

r o w

. ld
V

m o c
V
H

5
4

E = 12H34V56HVH

M3

H
V 6

E = 12H34V56VHV

E = 123H4V56VHV
14

www.jntuworld.com

www.jntuworld.com

Annealing Schedule
Initial solution: 12V 3V . . . nV .

D L

R
Ti = riT0 , i # of accepted moves < 5%, run out of time.

= 1, 2, 3, . . .; r = 0.85.

At each temperature, try kn moves (k = 510). Terminate the annealing process if temperature is low enough, or

w w

N J
15

. w

u t jn

www.jntuworld.com

U T

r o w

. ld

m o c
n

www.jntuworld.com

Algorithm: Simulated Annealing Floorplanning(P, , r, k) 1 begin 2 E 12V 3V 4V . . . nV ; /* initial solution */ 3 Best E ; T0 ln(avg ; M M T uphill 0; N = kn; P) 4 repeat 5 M T uphill reject 0; 6 repeat 7 SelectMove(M ); 8 Case M of 9 M1 : Select two adjacent operands ei and ej ; N E Swap(E, ei , ej ); 10 M2 : Select a nonzero length chain C ; N E Complement(E, C ); 11 M3 : done F ALSE ; 12 while not (done) do 13 Select two adjacent operand ei and operator ei+1 ; 14 if (ei1 = ei+1 ) and (2Ni+1 < i) then done T RU E ; 15 N E Swap(E, ei , ei+1 ); 16 M T M T + 1; cost cost(N E ) cost(E ); cost 17 if (cost 0) or (Random < e T ) 18 then 19 if (cost > 0) then uphill uphill + 1; 20 E N E; 21 if cost(E ) < cost(best) then best E ; 22 else reject reject + 1; 23 until (uphill > N ) or (M T > 2N ); 24 T = rT ; /* reduce temperature */ 25 until ( reject > 0.95) or (T < ) or OutOf T ime; MT 26 end

D L

w w

N J

. w

u t jn

U T

r o w

. ld

m o c

16

www.jntuworld.com

www.jntuworld.com

Floorplanning by Mathematical Programming

Sutanthavibul, Shragowitz, and Rosen, An analytical approach to oorplan design and optimization, 27th DAC, 1990. Notation:

D L

wi, hi: width and height of module Mi.

(xi, yi): coordinate of the lower left corner of module Mi. ai wi/hi bi: aspect ratio wi/hi of module Mi. (Note: We dened aspect ratio as hi/wi before.)

Linear constraints? Objective function?

N J

Goal: Find a mixed integer linear programming (ILP) formulation for the oorplan design.

w w

. w

u t jn

U T

r o w

. ld

m o c

17

www.jntuworld.com

www.jntuworld.com

wi hi

D L
Mi

Area = hi * wi Aspect ratio = wi / hi

R
u t jn

(xi, yi)

w w

www.jntuworld.com

N J

. w

U T

r o w

. ld

m o c

www.jntuworld.com

Nonoverlap Constraints
Two modules Mi and Mj are nonoverlap, if at least one of the following linear constraints is satised (cases encoded by pij and qij ): pij qij Mi to the left of Mj : xi + w i xj 0 0 Mi below Mj : yi + h i y j 0 1 Mi to the right of Mj : xi wj xj 1 0 Mi above Mj : yi h j yj 1 1

D L

w w
hi

wi

wj hj

wi

(xj, yj) (xi, yi) xi + wi <= xj

(xj, yj) (xi, yi) xi + wi > xj

18

www.jntuworld.com

N J

. w

xi + w i yi + h i xi w j yi h j

xj + W (pij + qij ) yj + H (1 + pij qij ) xj W (1 pij + qij ) yj H (2 pij qij )

U T

Introduce two 0, 1 variables pij and qij to denote that one of the above inequalities is enforced; e.g., pij = 0, qij = 1 yi + hi yj is satised.

u t jn

Let W, H be upper bounds on the oorplan width and height, respectively.

r o w

. ld

m o c

www.jntuworld.com

Cost Function & Constraints

Minimize Area = xy , nonlinear! (x, y : width and height of the resulting oorplan) How to x?

D L

Fix the width W and minimize the height y !

1. no two modules overlap (i, j : 1 i < j n); 2. each module is enclosed within a rectangle of width W and height H (xi + wi W, yi + hi H, 1 i n); 4. pij , qij {0, 1}. wi, hi are known.

N J
19

3. xi 0, yi 0, 1 i n;

w w

. w

u t jn

www.jntuworld.com

U T

Four types of constraints:

r o w

. ld

m o c

www.jntuworld.com

Mixed ILP for Floorplanning

Mixed ILP for the oorplanning problem with rigid, xed modules. min y

D L

subject to

Size of the mixed ILP: for n modules,

# continuous variables: O(n); # integer variables: O(n2 ); # linear constraints: O(n2 ). Unacceptably huge program for a large n! (How to cope with it?) Popular LP software: LINDO, lp solve, etc. 20

w w

www.jntuworld.com

N J

. w

U T

xi + wi W, yi + hi y, xi + wi xj + W (pij + qij ), yi + hi yj + H (1 + pij qij ), xi wj xj W (1 pij + qij ), yi hj yj H (2 pij qij ), xi , y i 0 , pij , qij {0, 1},

u t jn

r o w

. ld

1in 1in 1i<jn 1i<jn 1i<jn 1i<jn 1in 1i<jn

m o c

(1) (2) (3) (4) (5) (6) (7) (8)

www.jntuworld.com

Mixed ILP for Floorplanning (contd)

Mixed ILP for the oorplanning problem: rigid, freely oriented modules. min y

D L

For each module i with free orientation, associate a 0-1 variable ri : ri = 0: 0 rotation for module i. ri = 1: 90 rotation for module i. M = max{W, H }. 21

www.jntuworld.com

N J

subject to xi + ri hi + (1 ri )wi W, yi + ri wi + (1 ri )hi y, xi + ri hi + (1 ri )wi xj + M (pij + qij ), yi + ri wi (1 ri )hi yj + M (1 + pij qij ), xi rj hj + (1 rj )wj xj M (1 pij + qij ), yi rj wj (1 rj )hj yj M (2 pij qij ), xi , yi 0, pij , qij {0, 1},

w w

. w

u t jn

U T

r o w

. ld

1in 1in 1i<jn 1i<jn 1i<jn 1i<jn 1in 1i<jn

m o c

(9) (10) (11) (12) (13) (14) (15) (16)

www.jntuworld.com

Flexible Modules
Assumptions: wi , hi are unknown; area lower bound: Ai . Module size constraints: wi hi Ai ; ai Hence, wmin =
wi hi

D L

bi .
Ai , bi

Ai ai , wmax =

Ai bi , hmin =

hmax =

wi hi Ai nonlinear! How to x?

Can apply a rst-order approximation of the equation: a line passing through (wmin , hmax ) and (wmax , hmin ).

Substitute i wi + ci for hi to form linear constraints (xi , yi , wi are unknown; i , j , ci , cj can be computed as above).

w w

N J

j . w

h i = i w i + c i hmax hmin i = wmin wmax ci = hmax i wmin

U T

u t n

r o w

. ld

Ai . ai

m o c

/ y = mx + c / / slope / / c = y0 mx0 /

22

www.jntuworld.com

www.jntuworld.com

h Ai = wi * hi

D L

h max

hi =

i wi

+ ci

h min

wmin

wmax

w w

www.jntuworld.com

N J

. w

u t jn

U T

r o w

. ld

m o c

www.jntuworld.com

Reducing the Size of the Mixed ILP

Time complexity of a mixed ILP: exponential! Recall the large size of the mixed ILP: # variables, # constraints: O(n2 ). How to x it?

D L

How to select next subgroup of modules? linear ordering based on connectivity.

w w

N J
23

. w

www.jntuworld.com

U T

How to minimize the # of required variables?

u t jn

Questions:

Key: Solve a partial problem at each step (successive augmentation)

r o w

. ld

m o c

www.jntuworld.com

Next group of modules

D L

w w

www.jntuworld.com

N J

. w

u t jn
W

U T

r o w

Partial floorplan

. ld

m o c

www.jntuworld.com

Reducing the Size of the Mixed ILP (contd)

Size of each successive mixed ILP depends on (1) # of modules in the next group; (2) size of the partially constructed oorplan.

D L

R
(a) C3 C2 C1 (c)

Keys to deal with (2)

Minimize the problem size of the partial oorplan.

Replace the already placed modules by a set of covering rectangles.

Dead space

w w

C4

R4

Horizontal cut edges

(d)

N J
R5 R3 R2 R1

. w

u t jn

U T
(b)

# rectangles is usually much smaller than # placed modules.

r o w

. ld

m o c

24

www.jntuworld.com