A Machine Learning Enabled Long-Term Performance Evaluation Framework for NoCs

Jie Hou, Qi Han, Martin Radetzki

Abstract: The rapidly increasing transistor density enables the evolution of many-core on-chip systems. Networks-on-Chips (NoCs) are the preferred communication infrastructure for such systems. Technology scaling increases the susceptibility to failures in the NoC's components. However, such a NoC can still operate at the cost of performance degradation. Therefore, it is not sufficient to analyze the performance and reliability of a NoC separately. In this paper, we propose a machine learning enabled performability evaluation framework to treat both aspects together. It applies Markov reward models. In addition, it leverages machine learning techniques to obtain different performance metrics under consideration of faulty routers and various simulation parameters quickly, which is a challenging task in an analytical manner. Moreover, we use a mesh-based NoC to demonstrate our methodology. Long-term performances of mesh 8x8 under XY and fault-tolerant negative-first routing algorithms are evaluated.