An improved hierarchical placement technique using clustering and region refinement

Abstract

An effective hierarchical placement technique has been developed based on improved clustering and region refinement algorithms. The region refinement algorithm has been developed for an efficient and effective iterative improvement of row-based cell placement. For efficiency, the placement refinement is performed using multiple levels of hierarchy and "closely connected" cells are clustered. The physical layout area containing cells is hierarchically partitioned into bins. In the region refinement algorithm, each bin becomes the active bin by turns and the placement is optimized by moving cells between the active bin and its neighbor bins. To obtain optimal placement results, partitioning and region refinement algorithms are performed several times at each hierarchical level. Several standard cell benchmark circuits from MCNC are placed using the suggested placement technique and about 8% better results have been obtained when compared with TW7.0.