Thermal Sensor Allocation for SoCs Based on Temperature Gradients

Jun Yong Shin,  Fadi Kurdahi,  Nikil Dutt
University of California Irvine


Recently, numerous techniques have been proposed so that the temperature distribution of a chip can be managed dynamically during its operation, and these dynamic thermal management (DTM) schemes rely on on-chip thermal sensors in order to get the accurate temperature information. The challenging question is how to allocate a proper number of sensors on a die in order to get the accurate thermal information at runtime. In this paper, we propose a novel approach for thermal sensor allocation and thermal profile reconstruction over an entire die area using image processing and computer vision techniques. Test results show that the RMSE over an entire die area can be as low as 1.19℃ when the temperature difference between the maximum and the minimum on a die is 9.29℃, and the averaged absolute error at the hottest spot on a die is 0.18℃ in case we use six thermal sensors.