ALT-DVS: Dynamic Voltage Scaling with Awareness of Leakage and Temperature for Real-Time Systems

Abstract

The increasing leakage power and chip temperature in microprocessors are adversely affecting the performance and power consumption in real-time embedded systems. However, the traditional dynamic voltage scaling (DVS) technique, a popular and effective low-power design method, does not include the leakage and temperature impacts. In this work, we consider these two impacts when designing DVS-enabled real-time systems. We study the following problems: how to achieve the maximum computation and how to achieve a given amount of computation with the minimum total energy with the leakage and temperature variance available. The enhanced DVS technique ALT-DVS we proposed leverages the on-chip temperature sensor and the established system-level thermal model to determine the appropriate operating voltage and frequency for a given task, such that each task can be completed before its deadline while the system does not overheat. Our simulation results show that ALT-DVS not only saves more than 20% total energy over pure DVS technique, but also reduces the number of tasks that cannot be completed by DVS due to system overheat.