Tool for Automated Instruction Set Characterization for Software Power Estimation

The complexity and functionality of mobile digital devices is continuously growing. This results in a higher energy consumption of such devices. To counteract this trend, it is mandatory to accomplish software power optimizations based on accurate power consumption models characterized for the processor. This paper presents an environment for automated instruction set characterization based on physical power measurements. The generic design of this characterization system enables an easy portability to other architectures. For an accurate current measurement, a high-performance sampling technique has been established, which can be either clock or energy driven. The performance of those techniques is analyzed, and the advantages over the conventional solution of a series resistor are discussed. During the characterization of different processor platforms, it could be shown that the characterization effort can be reduced from three man-months to two man-weeks.