A Traffic Intensive Virtual Channels Allocation Scheme in Network-on-Chip

The Network-on-Chip (NoC) phenomenon has emerged as a highly reliable and efficient communication system that improves the scalability of modern chips with lesser communication overhead. Virtual Channels (VC)-based flow control strategy plays a vital role to make NoC an efficient communication system. While transitioning from source to destination IP cores, allocation of the packets (flits) to the next router’s available VCs is done as per an allocation strategy. An ideal VC allocation determines an effective usage of resources, which typically results in an increasing number of packets received at the destination(s) and can minimize the per-hop delay as a performance enhancement. However, recognizing the influence of diverse traffic and load configurations over multiple source-destination paths on NoCs’ performance is crucial to the realization of the many-core architectures. This paper introduces an artificial neural networks (ANN)-based VC allocation technique that prioritizes the most appropriate VCs for allocation and provides all-round performance improvisation. Here, adjustable weights (priorities) are set to each VC in each port of every router, to distribute traversing flits among accessible VCs. The proposed scheme is compared with the state-of-the-art First-Come-First-Serve (FCFS)-based VC allocation under two distinct routing strategies. Conducted experiments under various diverse traffic scenarios for the proposed scheme demonstrate an upsurge of up to 5.75% in the number of packets received at destinations, while also incurring up to 24.4% lesser network delay as compared to the FCFS-based VC allocation scheme.