Addressing transient routing errors in fault-tolerant Networks-on-Chips
Amir Charif, Nacer-Eddine Zergainoh, M. Nicolaidis · 2016
NoCs (Networks-on-Chips) are being viewed as the paradigm of choice for on-chip communication in modern SoCs (Systems-on-chips). Unfortunately, continuous technology downscaling is rendering NoC components increasingly susceptible to failure, to a point where it is no longer an option to design such systems without accounting for reliability issues. In this work, we concern ourselves with faults affecting one of the most important logical units in NoC routers, namely the Route Computation Unit. To prevent deadlocks and packet loss, which may result from misrouting, we propose a solution to detect and correct route computation errors. With compatibility with non-minimal fault-tolerant adaptive routing algorithms in mind, two detection methods are proposed. Lazy detection only detects the faults that may result in an error immediately following route computation, and leaves it to the next hops to detect and correct other errors. Strict detection detects all fatal errors before the affected packets leave the faulty router, making it possible to correct all errors using a single rerouting mechanism. Finally, we propose a novel method to safely correct routing errors and deliver all packets to their destination without resorting to retransmission.