Regression Based Run-Time Thermal Estimation for Multi-core Processors
Praveen Kumar, M. G. Mini · 2020 International Conference on Smart Electronics and Communication (ICOSEC) · 2020
The increasing power density in deep sub-micron devices, computationally efficient run-time thermal estimation techniques are important to implement real-timereal-time thermal management policies. This work proposes a polynomial regression-based methodology for the run time estimation of dynamic power and steady-statesteady-state temperature of the various logical units in a multicore processor. The impact of the workload characteristics which affect the dynamic power and steady-statesteady-state temperature of the processing elements of a core are investigated to develop the polynomial regression model. Experimental results show that the proposed regression-based power and thermal estimation methodology is computationally efficient with an estimation Root Mean Square Error (RMSE) of 0.010 for power estimation and 0.0166 for steady-statesteady-state temperature estimation compared to the benchmarking schemes.