A load balancing method for improving fault tolerance in mesh based networks on chip

Abstract

This paper presents an adaptive routing algorithm for 2D mesh network-on-chips (NoCs). The algorithm is based on DyXY routing algorithm. Our proposed routing algorithm is a very low cost fault-tolerant routing method to tolerate at least one faulty link in mesh-based on chip networks. It is a distributed, adaptive, congestion-aware and deadlock-free routing algorithm where only two virtual channels are used for adaptiveness and fault-tolerance. This is obtained by using two congestion flags between each two nodes which demonstrate the existence of congestion in a row or column. The same flags may be used to alarm a link failure in a row or column. The network performance, fault-tolerance capability and hardware overhead are computed through appropriate simulations. The experimental results show that the overall reliability and throughput of a NoCs is significantly elevated against at least one link failure with only a small latency overhead.