KNN Regression based Machine Learning Model for the Mitigation of Aging due to Thermal Cycling in Multi Core Processors

Processor life-time reliability is an important concern in today’s high end embedded devices that are being designed with high performance multi-core processors. These processing cores are manufactured with high level of integration for meeting the high functionality per unit area demands of the applications. The lifetime reliability of such densely integrated processor cores is significantly affected by thermal stress as temperature cycles can cause failure of the cores. The damages that are caused by the cycles in the temperature accumulates over time. These thermal cycles are highly dependent on the characteristics of the workload that are being executed by the processing cores. This work proposes a K-Nearest Neighbor (KNN) based machine learning approach for the estimation of the temperature of the processor cores due to thermal cycling by analyzing the workload characteristics. This work also proposes a scheduling scheme to improve the lifetime reliability of the processing cores by appropriately allocating the workload to the cores in a multicore system. Using the experimental results, we show that the proposed methodology is accurate and suitable to implement in real-time applications.