Fault-Tolerant Routing Algorithm for Mesh based NoC using Reinforcement Learning

Abstract

Network-on-Chip (NoC) is an efficient and scalable on-chip interconnection framework, designed to address communication challenges in multiprocessor system-on-chips (MPSoCs) based architectures. The major backbone of the communication infrastructure of NoC are links and routers. In the nano-scale era, NoCs are susceptible to faults that may degrade the system performance at a higher level. In NoC, faults can occur at any abstract level of the communication infrastructure. Hence, there is a need for fault-tolerant strategies that address reliability while improving system performance. In this paper, a fault-tolerant routing algorithm is proposed to address link and router faults in the mesh-based NoC design. Unlike traditional approaches, a reinforcement learning (RL) based fault-tolerant routing algorithm has been proposed to address the link and router faults in NoC. The simulations are performed to examine the efficiency of the proposed routing algorithm by varying the number of links and router faults and by scaling the network size. The simulation results show that the proposed routing algorithm found an optimal path in the likelihood of multiple links and router faults, resulting in a seamless delivery of data packets in the mesh network.