Deep Reinforcement Learning Enabled Self-Configurable Networks-on-Chip for High-Performance and Energy-Efficient Computing Systems

Network-on-Chips (NoC) has been the superior interconnect fabric for multi/many-core on-chip systems because of its scalability and parallelism. On-chip network resources can be dynamically configured to improve energy efficiency performance NoC. However, large complex design space in heterogeneous NoC architectures becomes difficult explore within a reasonable time optimal trade-offs performance. Furthermore, reactive resource management is not effective preventing problems, such as thermal hotspots, from happening adaptive systems. Therefore, we propose machine learning techniques provide proactive solutions an instant NoC-based computing We present deep reinforcement technique configure voltage/frequency levels routers links both high while meeting global budget constraint. Distributed agents proposed, where agent configures router associated intelligently based on system utilization application demands. Additionally, neural networks are used approximate actions distributed agents. Simulations results 256-core 16-core under real applications synthetic traffic show that proposed self-configurable approach improves energy-delay product (EDP) by 30-40% compared traditional non-machine-learning solution.