Incorporating temperature-leakage interdependency into dynamic voltage scaling for real-time systems

Energy efficiency is critical for many application specific real-time systems. Dynamic voltage scaling (DVS) is one of the most effective and well-studied techniques. In this paper, we study the interdependency of temperature and leakage and how it influences DVS. We derive an analytic temperature-leakage model, which has an average error of 0.5°K from the accurate numerical result. This temperature-leakage model enables us to perform temperature aware DVS for total energy minimization without using on-chip temperature sensors. We find that the most energy efficient way to complete a single task is, unlike the existing approaches that use high voltage to save leakage, to scale voltage down to the lowest level without missing the task's deadline. Based on this new finding, we propose an online DVS algorithm to schedule multiple tasks on real-time system. Simulation results show that our algorithm can achieve total energy saving over a state-of-the-art leakage aware DVS approach by as high as 14% and more than 9% on average.