Abstract
Avoiding thermal hotspots and large temperature gradients are the ultimate design goals for the new generation of systems on chip. Monitoring the system's behavior allows Dynamic Thermal Management (DTM) techniques to prevent critical temperatures, and therefore enhance design reliability, reduce the energy consumption, and increase the chip lifetime. This paper addresses the self-awareness for the chip temperature monitoring. A cost-effective method is presented for the ondie thermal sensor placement in order to provide estimations at run-time of the overall chip thermal behavior. The proposed algorithms systematically choose the best trade-off between accuracy and overhead. The surface of the chip is decomposed into several thermally homogeneous regions. A case-study experiment shows the accuracy of our method compared to the related works.