Handling of SETs on NoC Links by Exploitation of Inherent Redundancy in Circular Input Buffers

The miniaturization of nanometer technologies beyond the sub-micron domain has jeopardized the reliability of on-chip network links, making them more susceptible to Single Event Transients (SETs) during system's run-time. Using retransmission approaches has been proposed in the literature for handling SETs in Network-on-Chip (NoC) links. However, those approaches either suffer from significant latency overhead or impose additional retransmission buffers, which consume more area. This paper proposes the ReUSE mechanism as a transient fault mitigation mechanism for Network-on-Chip links. The mechanism takes advantage of the inherent redundancy in the input buffers of NoC routers and reuses these for SET mitigation on the NoC links. By using a parity checker for fault detection, the approach can cover all SETs during run-time and return to normal operation in maximum one clock cycle after the disappearance of the fault. The experimental results show that the proposed mechanism prevents network-wide failure even in harsh environments with up to 80 million random faults on links per second. The ReUSE mechanism imposes 18% area overhead and 7.8% critical path delay overhead to the baseline NoC router.