In this paper, we develop an accurate dynamic thermal management DTM approach considering the interdependency of temperature and leakage. By modeling the leakage-thermal interdependence as a quadratic polynomial, we develop accurate analytical equations that capture the thermal transient. We also identify all the situations in which thermal runaway would occur, which should be avoided by DTM.We then present a discrete dynamic programming algorithm that performs thermal-aware task and speed scheduling using the model we derived.Compared to a linear leakage-thermal model, owing to our more accurate model, our scheme resulted in 18.2% better performance while maintaining the temperature below constraint.
