Parallel standard cell placement on a cluster of workstations

Abstract

In this paper we report experiences on a parallel implementation of a standard cell placement algorithm on a cluster of Myrinet connected PCs. The implementation is based on a recently developed placement tool (Feng Shui) that extends recursive bisection placement to incorporate global aspects of the design using an efficient optimization called iterative deletion. Contrary to previous attempts at parallelizing placement algorithms, initial experimental results show significant performance improvement with small reduction in the placement quality. Furthermore, the reduction in the placement quality does not increase with the number of processors.

Parallelizing post-placement timing optimization

“…To that end, partitioners find balanced partitions of tasks and assign each partition to one of multiple processors. There have been investigations of parallel databases for CAD [9] and parallel algorithms for various CAD tasks, such as placement [2,7,12], routing [6,8] and logic simulation [4,10]. In post-placement timing optimization, however, tasks are heavily connected and less amenable to efficient parallelization, due to the excessive need for communication and/or synchronization.…”