Quick generation of temporal power waveforms for RT-level hard macros

Abstract

Power characterization of complex macros is essential to enable accurate RT-level power estimation. Existing characterization procedures focus on the average value of power. In this paper, we take a fresh look at this problem. We propose a fast, yet accurate technique to determine a time-dependent power model, i.e., a temporal power waveform, which is able to fully characterize the power behavior of a hard macro in response to a realistic input stream consisting of typical usage patterns. Our approach is simulation-based, and resorts to a mix of high-level, fast cycle-based simulation with low-level, slow accurate simulation of the long set of input patterns. Results are extremely satisfactory, since the average error between the power waveforms generated by our tool and the exact ones is always within 1%, while the reduction in the execution time ranges between one and two orders of magnitude.