Scalable Bio-inspired Fault Detection, Isolation and Recovery in NoCs

Abstract

High density silicon architectures, including Networkson-Chip (NoC), require an increasing level of fault tolerance with adaptivity to faulty conditions post deployment. An unobtrusive bio-inspired approach is proposed to facilitate both online and real-time detection and isolation of faulty NoC channels. The proposed approach is based on a novel Fault Detection Unit (FDU) which uses the combination of synapse inspired excitatory and inhibitory responses as a method of fault detection. The FDU integrates with the NoC router to inform the routing algorithms when faults are detected, enabling re-routing as a recovery method. The key contribution is results for online detection, isolation and recovery in a NoC architecture for a range of test conditions. The low area/power overheads and scalability advantages over similar online techniques are also presented.