Regional Selection Mechanism for Traffic-balanced Adaptive Routing Algorithms in Mesh-based NoC Architectures

摘要

With deep submicron scaling and integrating multi and many cores on a single chip, Network-on-Chip (NoC) has been proposed as a scalable and cost-effective solution for inter-core communication. Data exchange between cores does not follow a uniform traffic pattern, which frequently leads to congestion in central routers of a mesh-based NoC. Congestion limits the performance of NoC and results in larger packets transmission latency, higher thermal, and more power consumption. This paper proposes a traffic-balanced selection mechanism to distribute traffic load over the network and mitigate the blocking of packets. The proposed scheme based on congestion information of neighbors routers estimates congestion of all possible minimal paths for a packet, then conduct packet through a non-congested path. It uses congestion information of a stair-shaped region in routing decision. The stair-shaped region provides adequate and up-to-date congestion information by taking into account local and non-local traffic conditions. Experimental results show the proposed method improves both saturation throughput and average latency of 9.91%-17.45% and 30.48%-64.23%, respectively, compared to similar approaches for all considered traffic patterns.