Green-Pow: Power Modeling and Estimation of Software Applications for Embedded Processors

This paper presents an early-phase high-level power model for software applications running on embedded processors. Several studies have been conducted on the performance and power estimation of embedded processors. However, the high demand for customized low-power embedded processors, particularly when implementing Artificial Intelligence in the Internet of Things, requires a fast, yet accurate power estimation. This paper proposes a high-level model to estimate power consumption and execution time at the instruction level for software applications running on the ARM Cortex-M3 implemented on an FPGA. The development of this model is based on real experimental power measurements from basic and machine-learning benchmarks. Finally, the accuracy of the proposed model is validated experimentally, achieving a relative error of 3.5 % in power estimation and 6.36 % in execution time estimation. This work demonstrates the potential for fast and accurate power estimation in embedded systems development.