Modeling Software Driven Power Consumption

Abstract

Power management issues are a major factor that threatens to curtail the progress of Moore's law in the 21st century (Borkar and Malone, 2004). Techniques to calculate and optimize the power consumption of hardware components are well known and understood, yet there exists no formal technique for modeling the effect of running software on the power dissipation for an arbitrary system. System designers have been forced to over design systems based on worst case estimates, or make design changes based on empirical measurements after the hardware and software is complete. Since there exists no formal method for modeling how software affects power dissipation, there is no way to optimize the software to minimize power consumption. This paper demonstrates a formal approach towards establishing a power consumption model for any processor running on an arbitrary target. Power consumption models are developed and tested for three DSPs with radically different internal architecture through the use of modeling, measurement, and statistical techniques