Estimating the Worst-Case Energy Consumption of Embedded Software

The evolution of battery technology is not being able to keep up with the increasing performance demand of mobile embedded systems. Therefore, battery life has become an important design constraint. As battery-operated embedded devices are deployed in mission critical systems, designers should ensure that the energy constraints are satisfied in addition to the timing constraints - the battery should not drain out before a critical task completes execution. Giving these guarantees requires the knowledge of the worst-case execution time and energy of a task. Significant progress has been made in estimating the worst-case execution time through static analysis. In contrast, existing energy estimation techniques use average-case execution profile of a program and as such cannot guarantee the satisfiability of energy constraints. In this paper, we present a static analysis technique to estimate the worst-case energy consumption of a task on complex micro-architectures. Estimating a bound on energy is non-trivial as it is unsafe to assume any direct correlation with the bound on execution time. Experimental evaluation with a number of benchmark programs indicates the accuracy of our worst-case energy consumption estimates.