Reliability Enhancement of Multi-Core Processors using Machine-Learning Techniques

As a result of technology scaling, power density of multi-core chips increases and leads to temperature hotspots which accelerate device aging and chip failure. Moreover tremendous efforts to reduce power consumption by employing low-power techniques decreases the reliability of new design generation. In this work, we first discuss the state-of-the-art methods for predicting workload dynamics and we compare their performance. We, then, introduce a prediction method based on Support Vector Regression (SVR), which accurately predicts the workload behavior several steps ahead. To evaluate the effectiveness of our approach, we use UltraSPARC T1 processor along with Sun Solaris operating system. We incorporate OS and architectural level sensors and knobs and our preliminary results show our predictive method achieve higher accuracy.