vioft2nntf2t|tblJournal|Abstract_paper|0xf4ffe5322c00000043c8000001001000
The increasing computing requirements in modern embedded systems demand high performance processing cores with highly down-scaled devices. System reliability due to ageing is a major concern in such processing cores. Run-time adaptations with ageing-aware schedulers are getting more attention to increase the lifetime reliability of the processing cores. Accurate and efficient thermal estimation of processing cores based on the characteristics of workloads is important to implement run-time adaptation schemes. In this work we have developed K-Nearest Neighbor (KNN) and Linear Regression (LR) machine learning models for the estimation of thermal profiles of the major power consuming logical units of a multi-core processor. Prediction performance of KNN and LR machine learning models in the design of an Electron Migration (EM)-ageing aware scheduler for improving the lifetime of multi-core processors is evaluated. The ageing-aware scheduler takes inputs from the trained models for the estimation of ageing effect and perform scheduling based on performance and reliability requirement. Experimental results show that predictive performance is better for KNN model compared to the LR model.