Multithreaded systolic computation: A novel approach to performance enhancement of systolic arrays

In this paper we propose a synergy of processing on parallel processor arrays (systolic or SIMD) and multithreading named multithreaded systolic computation. Multithreaded systolic computation enables simultaneous execution of independent algorithm data sets, or even different algorithms, on systolic array level. This approach results in higher throughput and improved utilization of programmable systolic arrays, when processing elements are designed with pipelined functional units. The real benefit of multithreaded systolic computation is that it increases the throughput of systolic arrays without changing systolic algorithm.