An empirical study of on-chip parallelism

Abstract

A methodology is presented for empirically determining the amount of parallelism on a CMOS VLSI chip. Six chips are measured, and the effect of input choice and circuit size is studied. The unexpectedly low parallelism measured here suggests that certain strategies for parallel simulators may be doomed, and earlier efforts to extrapolate parallelism from small circuits to large circuits may have been overly optimistic.<>