Circuit placement on arbitrarily shaped regions using the self-organization principle

1
Circuit placement on arbitrarily shaped regions
using the self-organization principle

• Paper by Sung-Soo Kim,
• Chong-Min Kyuing,
• Korea, 1992

Leonid Tolstoy UMASS 2004
2
Overview

• Self-organization assisted placement
  algorithm(SOAP), for circuit placement in
  arbitrarily shaped regions

• two-dimensional (2-D) arbitrary shaped regions
• nonplanar surfaces of three-dimensional objects
• three-dimensional volumes

Based on the self-organizing Kohonen algorithm
for Neural Networks
3
Advantages

• General algorithm for circuit placement in
  arbitrary shaped region, even 3D, and gives
  optimal solution.
• Earlier methods of placement in rectilinear
  regions, where result heavily depends on
  arbitrary partition of the entire region into a
  number of rectangular subregions, often give
  suboptimal solutions.

4
Self-Organizing Assisted Placement

• Placement consists in finding the positions of
  modules such that closely related modules are
  placed near one another, and total
  interconnection length is minimized.

• Relationship with Neural Network (NN)- Number
  of input nodes corresponds to dimensionality (2D,
  3D) of the region- Circuit modules correspond
  to output NN nodes- Variable weighs wij from
  input nodes (si) to output nodes (rj) correspond
  to i-th coordinates of module mj (x and y)

5
General Self-Organizing Assisted Placement Alg-m.

• 1 step. Initialization t0all modules are
  clustered around the center of the region.

2 step. Present a new pointX(t)x1(t),x2(t)
3 step. Select closest to X module j
4 step. Update position of module j, and its
neighbors within the distance s(i) from module j.
5 step. Increment t.
6 step. If s(t)0 stop,else go to step 2.
6
Neighborhood of a module

• Neighborhood of a module j at time t is a set of
  modules lying within distance s(t) from module j,
  where s(t) is a time- decreasing function, which
  determines the size of the neighborhood.

Position update
Neighborhood at T1
Neighborhood at T2
Neighborhood at T3
7
Global Placement of Macro cells using SOAP

• In the basic SOAP each module is considered as a
  point, which leads to overlaps in placement.
• Modified algorithm considers size and shape of
  modules.

• Basic SOAP is used to obtain the initial
  placement in the initial region (about 25 of
  the entire region).
• Then gradually increase size of the modules,
  preserving current placement configuration.
• Use reorientation of modules to eliminate
  overlaps.

8
Global Placement for Macro cells, using modified
SOAP
1 step. Initial placement using basic SOAP
(using 25 of area, and modules represented by
point).
2 step. Gradual expansion of module size.
3 step. Modules rotation and orientation for min
wiring length.
5 step. If module size is actual- stop,else go
to step 2.
9
3-D circuit placement

• Occurs when control schemes have to be placed on
  the interior/exterior surface of the mechanical
  object,or when it is necessary to place circuits
  modules on the surface of the 3D objects to
  minimize total volume or wiring distances.
• Placement on multiple layers - high packing
  density and high performance.

SOAP Algorithm modifications -3 input nodes
(correspond to x,y,z coordinates) -sort cells
for each surface, project on it.
10
3D circuit placement results
Placement on the 3D surface and on multiple
layers
11
Conclusions

• This algorithm can work with arbitrary shaped
  regions, non-planar surfaces, and 3D volumes.
• SOAP algorithm doesnt requires region
  partitioning (essential in earlier works).
• This algorithm is quite competitive with other
  algorithms even in rectangular regions.

12
?