Stable Multiway Circuit Partitioning for ECO

  • Authors:
  • Yongseok Cheon;Seokjin Lee;Martin D. F. Wong

  • Affiliations:
  • The University of Texas at Austin;The University of Texas at Austin;University of Illinois at Urbana-Champaign

  • Venue:
  • Proceedings of the 2003 IEEE/ACM international conference on Computer-aided design
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

We propose a new stable multiway partitioning algorithm,where stability is defined as an additional quality of a partitioningsolution. The stability of a partitioning algorithm isan important criterion for a partitioning based placement toachieve timing closure through the repetition of the placementprocedure [Invited talk: Important Placement Considerations for Modern VLSI Chips]. Given a previous partitioning result P*on an original netlist hypergraph H* and a partially modifiednetlist hypergraph H, a new cost function with similarityfactor is defined to produce a new partition P on Hwhich is similar to the original partition P*. The proposedalgorithm is the first approach that quantifies the degree ofsimilarity of a current partition to the original partition usingsimilarity cost. Our goal is to build a new partition ina relatively short run time, whose cut quality is not muchdegraded from that of the original partition P* while it preservesas much of the previous groupings in P* as possible.The proposed partitioner is especially beneficial to engineeringchange order (ECO) applications, where partial modificationsof a netlist are handled by the incremental methodologyin a design iteration cycle. Our approach helps ECOplacers maximize the incremental capability since the portionsto be re-placed are minimized. Experimental resultsshow that the proposed algorithm achieves a high qualitypartition comparable to a state-of-the-art multilevel partitionerhMetis [Multilevel k-way hypergraph partitioning], while many portions of the groupings in the previous partition are preserved in the current partition.The tradeoff between similarity and cut quality with respectto a varying similarity coefficient is also shown.