A Timing Optimization Method Based on Clock Skew Scheduling and Partitioning in a Parallel Computing Environment

This paper describes the implementation of a heuristic method to perform non-zero clock skew scheduling of digital VLSI circuits in a parallel computing environment. In the proposed method, circuit partitions that have low number of timing paths between partitions are formed. Clock skew scheduling is applied independently to each partition-sequentially or in parallel on a computing cluster-and results are iteratively merged. The scalability of the proposed method is superior compared to conventional non-zero clock skew scheduling techniques due to the reduction of analyzed circuit sizes (partition sizes) at each iteration step and the potential to parallelize the analyses of these partitions. It is demonstrated that after only the first iteration step of the proposed method, feasible clock schedules for 65% of the ISCAS'89 benchmark circuits are computed. For these circuits, average speedups of 2.1X and 2.6X are observed for sequential and parallel application of clock skew scheduling to partitions, respectively.