A methodology to compute long-term fault resilience of NoCs under fault-tolerant routing algorithms

Networks-on-Chips (NoCs) are the preferred communication interconnect for many-core systems. Technology scaling increases the susceptibility to failures in the NoC’s components. Fault-tolerant routing algorithms are applied to maintain reliable operation in the presence of such failures. Fault resilience is a measure of a NoC’s ability to tolerate faults. In this paper, we investigate how fault resilience develops over a long period of time during which faults appear and disappear. We apply Markov reward models to investigate long-term fault-resilience of mesh-based NoCs under three different routing algorithms. Fault-tolerant XY routing brings the best long-term fault resilience even though it is less adaptive than fault-tolerant negative-first routing. We further investigate how long-term fault resilience develops with scaling towards larger NoCs.