High Performance and Connection-retaining Fault-Tolerant Routing Algorithm for NoCbased many core Systems

Abstract

Communication prerequisites in multi-core systems are highly convoked by egressing the network-on-chip (NoC) architecture. NoC should cover communication reliability issues that are essential factor in communication between multiple cores embedded on a chip. Mostly, fault tolerance routing algorithms are based on rerouting data packets around the faults for reliable communication in presence of faults. But this rerouting packets might bring on non-minimal path causing increase in latency and congestion around the faults that further affect the performance. Proposed algorithm is able to tolerate both faulty links and routers by utilizing one and two virtual channels along X and Y dimensions. To bypass the faults, it always furnishes a shortest route for packets as long as the route exists. Packets are routed to shortest path even if faulty router is located on the route directly between current and destination router by utilizing the non-faulty links connected to the faulty router. This technique optimally reduces the congestion due to rerouting packets around the faults. Moreover the network congestion is balanced by adaptively selection of output channel whenever current and destination routers are located at distance of greater than two hop counts along both dimensions. Experimental results, in presence of six faults for 6x6 mesh network, manifest capability of up to 99.3% reliability for a system under functionality of proposed technique.