Efficient network flow based ratio-cut netlist hypergraph partitioning

  • Authors:

  • Sachin B. Patkar;Himanshu Sharma;H. Narayanan

  • Affiliations:

  • Department of Mathematics, Indian Institute of Technology, Bombay, India;Department of Electrical Engineering, Indian Institute of Technology, Bombay, India;Department of Electrical Engineering, Indian Institute of Technology, Bombay, India

  • Venue:
  • ICECS'03 Proceedings of the 2nd WSEAS International Conference on Electronics, Control and Signal Processing
  • Year:
  • 2003

Quantified Score

Hi-index 0.01

Visualization

Abstract

We present a new efficient heuristic that finds good ratio-cut bipartitions of hypergraphs that model large VLSI netlists. Ratio cut measure is given by the ratio of the number of nets cut between two blocks and the product of the cardinality (size) of each block. Hypergraphs model VLSI netlists in a more natural fashion than do graphs. Our new heuristic may be considered a hybrid between the approaches of [10] and [12]. It makes use of maxflow-mincut algorithm as a subroutine that is invoked only a (small) constant number of times. Comparisons with the state-of-art partitioners such as hmetis [6](hmetis was run several times to yield min-cuts for diffently balanced bipartitions) show that the heuristic successfully finds better ratio-cut bipartitions in most of the cases. Although maxflow-mincut algorithms are theoretically slower than multilevel algorithm of hmetis the actual running times of our heuristic is found to be of the same order. The experiments have been conducted on ISPD98 benchmarks of sizes upto 200K cells.