An Accurate Instruction-Level Energy Estimation Model and Tool for Embedded Systems

Abstract

Estimating the energy consumption of applications is a key aspect in optimizing embedded systems energy consumption. This paper proposes a simple yet accurate instruction-level energy estimation model for embedded systems. As a case study, the model parameters were determined for a commonly used ARM7TDMI-based microcontroller. The total energy includes the energy consumption of the processor core, Flash memory, memory controller, and SRAM. The model parameters are instructions opcode, number of shift operations, register bank bit flips, instructions weight and their Hamming distance, and different types of memory accesses. Also, the effect of pipeline stalls have been considered. In order to validate the proposed model, a physical hardware platform equipped with energy measurement capabilities was developed. We have conducted experiments on several embedded applications from MiBench benchmark suite and the results show less than 6% error in the energy consumption estimation. We have also developed an energy profiler tool for the systems that use ARM7TDMI processors by embedding the model parameters in an instruction-level profiler from the SimpleScalar toolset which provides valuable information and guidelines for software energy optimization.