Thermal modelling for temperature control in MPSoC´s using Timed Continuous Petri Nets

This paper proposes a thermal modelling methodology for Multiprocessor Systems on a Chip (MPSoC) by using Timed Continuous Petri Nets (TCPN). The derived model includes the heat generation, temperature and power consumption. The proposed methodology is related to finite differences technique where the solution´s precision is given by the number of Petri net places. The main advantages of the proposed methodology are that the model is built incrementally, no further calibration stages are needed to tune the model, the derived model is as precise as finite element or differences methods and based on this model, an approximate algebraic thermal model can be easily derived to efficiently predict the MPSoC temperature. Hence the proposed thermal model can be exploited by a Model Predictive Controller (MPC) in thermal aware schedulers or other kind of schedulers.