A Novel Software Solution for Localized Thermal Problems

Abstract. In this paper, we propose a temperature-aware DFS (Dynamic Frequency Scaling) technique using the performance counters that is already embedded in the commercial microprocessors. By using performance counters and simple regression analysis, we can predict the localized temperature and efficiently schedule the tasks considering the temperature. The proposed technique is especially beneficial to potential localized thermal problems that are inevitable due to limited number of costly CMOS thermal sensors. When localized thermal problems that were not detected by thermal sensors are found after fabrication, the thermal problems can be avoided by the proposed software solution without re-fabrication costs. The evaluation results show that the proposed technique is comparable to the DFS technique using CMOS thermal sensors.