Accurate high level datapath power estimation

Abstract

The cubic switching table is a new data structure for estimating datapath switching at a high level. It is constructed during behavioral simulation, and is used to estimate the switching for any particular datapath schedule and binding. Time to extract the estimate from the table is independent of the original simulation size. For n operations in the RTL description, it takes O(n/sup 3/) time to perform the extraction. We show that an exact switching table would require exponential size, but experimental results show that the cubic table is accurate, with typical error under 5%.