Parameter-Aware Energy Models for Embedded-System Peripherals

Energy models support monitoring and prediction of energy use, which is essential for the development and usage of transiently powered systems. However, model generation is a time-consuming and repetitive task. Also, available energy modeling solutions typically assume hardware configurations to be constant, although configuration changes can significantly impact hardware behaviour. Here we present a work-in-progress algorithm for the automatic generation of configuration-aware energy models for system peripherals. We determine the influence of configurable hardware parameters on model attributes and generate functions to describe it. We also propose a new transition energy model to improve energy accounting accuracy without additional overhead. Initial tests show promising results with mean absolute model error less than 1.5 % for various hardware configurations.