Scribd
Upload
5 views

Lab 3

This document discusses several existing techniques for circuit partitioning and floorplanning, including LIFO bucket management, lookahead gain schemes, clustering and unclustering approaches, and spectral bipartitioning methods. It introduces the EIG algorithm, which uses eigenvectors of the Laplacian matrix to embed graph vertices in a k-dimensional space for multi-way partitioning. Finally, it describes algorithms such as MELO that perform vector partitioning based on eigenvector mappings to obtain high-quality partitioning solutions.

Uploaded by

Rubick
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
5 views

Lab 3

This document discusses several existing techniques for circuit partitioning and floorplanning, including LIFO bucket management, lookahead gain schemes, clustering and unclustering approaches, and spectral bipartitioning methods. It introduces the EIG algorithm, which uses eigenvectors of the Laplacian matrix to embed graph vertices in a k-dimensional space for multi-way partitioning. Finally, it describes algorithms such as MELO that perform vector partitioning based on eigenvector mappings to obtain high-quality partitioning solutions.

Uploaded by

Rubick
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
You are on page 1/ 1

than first-in-first-out (FIFO) queues or random management as in FM.

They
also showed that the LIFO selection scheme results in improvement over ran-
dom schemes for KFM [Sanchis, 1989]. Under the LIFO scheme, most recently
visited modules are placed near the beginning of the buckets, implicitly caus-
ing the neighborhoods or clusters of modules to be moved together. This has a
similar “cluster removal” effect as the CLIP scheme [Dutt and Deng, 1996b].
The authors of [Hauck and Borriello, 1997] examined many of the exist-
ing techniques for FM and presented a methodology for determining the best
mix of these approaches. These techniques include various clustering and un-
clustering schemes, initial partitioning creation, LIFO bucket management
[Hagen et al., 1997], lookahead gain scheme [Krishnamurthy, 1984], and net
partitioning. The result is a novel bipartitioning algorithm that includes both
new and existing techniques.
EIG Algorithm
Spectral bipartitioning for ratio-cut minimization has been extended to multi-
way partitioning by [Chan et al., 1994]. Their approach involves finding the
k-smallest eigenvalue/eigenvector pairs of the Laplacian of the circuit graph.
The eigenvectors provide an embedding of the graph’s n vertices into a k-
dimensional subspace. They also proposed a clustering heuristic to reduce the
size of the problem before the eigenvalue/eigenvector computation.
The authors of [Riess et al., 1994] derived minimal ratio-cut partitioning
solutions from the placement result obtained by the Gordian-L algorithm [Sigl
et al., 1991a]. Gordian-L is an analytical placer, where the original quadratic
placement formulation is modified to optimize linear objective under linear
constraint. The 1-dimensional placement obtained by Gordian-L determines
an ordering among the cells. Partitioning solutions are generated based on this
ordering and evaluated in terms of ratio cut.
Spectral partitioning methods use the eigenvectors of the Laplacian matrix
of the netlist graph to obtain partitioning solutions. Given d eigenvectors, the
authors of [Alpert and Yao, 1995] map each vertex in the netlist graph to a
vector in d-dimensional space such that these vectors constitute an instance
of the vector partitioning problem. When all the eigenvectors are used, they
showed that graph partitioning exactly reduces to vector partitioning. Based on
this result, the authors presented an algorithm named MELO that is based on
a simple vertex ordering scheme that can be used to yield high-quality 2-way
and multi-way partitioning solutions.
The authors of [Alpert and Kahng, 1995a] presented a spectral partition-
ing algorithm that exploits both the geometric embedding and netlist topology
information. The geometric embedding is done by the computation of d eigen-
vectors of the Laplacian matrix. This embedding is then partitioned based on
the topological information from the netlist. They begin with a d-dimensional

You might also like