Power consumption estimation of data-processing instructions

Abstract

Nowadays embedded and autonomous systems are widely used. The performance requirements that were important previously are substituted with the requirements of the minimal power consumption. Optimisation of power consumption can be provided in different ways. The most flexible approach is optimisation of the software. It is proposed that experimental data must be generalised with the help of neural networks in order to estimate the power consumption during software executing. Methodology of development of training set for the estimation of power consumption of instructions will be represented. Data-processing instructions of ARM7TDMI will be used as reference.