Optimal Multi-Processor SoC Thermal Simulation via Adaptive Differential Equation Solvers
Thermal management is a critical challenge in the design of high performance multi-processor system-on-chips (MPSoCs). Therefore, accurate and fast thermal modeling tools are necessary for efficiently analyzing the thermal profiles of MPSoCs. This paper advances state-of-the-art MPSoC thermal modeling approaches in several directions. Our first contribution is a novel matrix statespace compatible representation of MPSoC thermal behavior. This representation can be used to choose the “best fit” solver among various ordinary differential equation (ODE) solvers according to the required accuracy and simulation speed. Then, we exploit this representation to develop an adaptive thermal simulation infrastructure that provides the shortest simulation time for the desired thermal modeling accuracy and the given MPSoC floorplan. The experimental results, which are based on a commercial 8-core MPSoC, show that our thermal simulation method achieves both higher thermal estimation accuracy (6x better) and faster simulation time (up to 70%) when compared to state-of-the-art MPSoC thermal simulators.