An Improved Power Macro-Model for Arithmetic Datapath Components

BIB
Domenik Helms, Eike Schmidt, Arne Schulz, Ansgar Stammermann, Wolfgang Nebel
PATMOS
We propose an improved power macro-model for arithmetic datapath components, which is based on spatio-temporal correlations of two consecutive input vectors and the output vector. Based on the enhanced Hamming-distance model [3], we introduce an additional spatial distance for the input vector and the Hamming-distance of the output vector to improve model accuracy significantly. Experimental results show that the models standard deviation is reduced by 3% for small components and up to 23% for complex components. Because of its fast and accurate power prediction, this model can be used for fast high-level power analysis.
9 / 2002
inproceedings
Proceedings of PATMOS 02, Seville, Spain
PRO-DASP
Power Reduction for Digital Audio Signal Processing (DFG Project of the University of Oldenburg)
POET
Power Optimization of Embedded systems