Distributed adaptive routing for spidergon NoC

Abstract

Spidergon is a popular NoC (Network-On-Chip) developed to realize cost effective multi-processor SoC (MPSoC) using a fixed optimized topology [1]. Increasing diversity of applications, quality of service requirements and deterministic routing schemes inhibit the performance by creating congestion bottlenecks. This paper presents an adaptive routing algorithm that exploits the path diversity of spidergon NoC and selects the optimal path on the basis of congestion level (CL). CL depicts current traffic conditions. Proposed scheme is compared with native deterministic routing schemes of spidergon NoC i.e. aFirst and aLast. Experimental results demonstrate that our algorithm distributes traffic evenly while reducing hot-spots resulting in a considerable performance improvement.