A thermal stress-aware algorithm for power and temperature management of MPSoCs

In this work, we propose a thermal stress-aware algorithm for the management of the power and temperature in MPSoCs. The algorithm, which uses a heuristic approach, controls the power consumption, maximum temperature, thermal cycles, and temporal/spatial thermal gradients of MPSoCs. At the top level, the decision on turning the cores on and off is made based on the constraints of peak temperature, maximum spatial thermal gradient, and power consumption. At the next tier, the optimal frequencies (and supply voltages) of the ON cores, formulated in a convex optimization problem, are determined again based on satisfying the constraints of the maximum total power consumption, peak temperature, thermal cycles, and also temporal thermal gradient. The technique may be applied to both the heterogeneous and homogenous MPSoCs. The efficacy of the proposed approach in reducing the thermal cycles as well as temporal thermal gradient is evaluated by comparing its results with a similar previous power and temperature management approach. The evaluation which is performed on 8-core processors under Splash2 benchmarks, demonstrates the ability of the suggested technique in limiting a considerable reduction in the thermal stress parameters.