Hybrid Wireless Network-on-Chip Architectures with Load-Balanced Congestion-Aware Routing Algorithm

Network-on-Chip (NoC) faces growing challenges in maintaining low-latency, energy-efficient communication as core counts in System-on-Chip architectures continue to rise. Wireless Network-on-Chip (WiNoC) has emerged as a promising solution by integrating wireless links with traditional wired mesh networks, enabling single-hop long range communication. However, with limited wireless bandwidth and non-uniform traffic patterns, WiNoC systems are prone to wireless congestion and inefficient resource utilisation. This paper proposes a load-balanced congestion-aware routing algorithm for WiNoC (LCRAW) that addresses these issues. The algorithm changes packet distribution between lower-wired paths and upper-layer wireless paths dynamically. The wired mesh network is partitioned into small subnets, and source-to-destination distance guides optimal path selection. By monitoring network load conditions, LCRAW adaptively reduces congestion risk in both layers and improves overall throughput. Simulation results using synthetic traffic patterns on the Noxim cycle-accurate simulator show that LCRAW, compared to XY and load-balanced congestion-aware (LTCA) routing, improves throughput by up to 14% and 9%, reduces latency by 19% and 12%, and decreases energy consumption by 13% and 10%, respectively. For application traces, LCRAW improves throughput by up to 25% and 14%, reduces latency by 16% and 9%, and decreases energy consumption by 51% and 35%, compared to XY and LTCA, respectively.