EHPRA: Enhanced Hybrid Probabilistic Routing Algorithm with Smart Detour for Fault-and Congestion-Aware ۲D Mesh NoC

Network-on-Chip(NoC) routing in many-core systems must satisfy two conflicting requirements: low-latency packet delivery and resilience against congestion-induced failures in heavily used links. This paper presents EHPRA (Enhanced Hybrid Probabilistic Routing Algorithm), a lightweight extension of HPRA for ۲D mesh NoCs that introduces a Smart Detour mechanism to improve routing quality under hotlink and hotspot traffic conditions. Unlike random detour approaches, EHPRA evaluates multiple intermediate candidates and selects the least congested path using a destination-aware congestion score plus a hop penalty. The algorithm preserves HPRA’s probabilistic adaptation concept while improving path selection quality through local link-usage history. A simulation framework on a ۱۶×۱۶ mesh compares EHPRA against HPRA, Minimal-Oblivious, Random Detour, and XY routing over one million trials per scenario. Results show that EHPRA achieves the lowest fault counts in both hotspot (۱۵۰,۳۲۰) and hotlink (۲۱۳,۳۷۵) tests. Compared with HPRA, EHPRA reduces faults by ۱۰.۲۹% in hotspot traffic and ۲۷.۲۰% in hotlink traffic. Although EHPRA may increase average hops versus strictly minimal algorithms, it provides the best reliability-latency trade-off, achieving the minimum weighted overall cost across all compared methods.

کلیدواژه ها:

Network-on-Chip ، ۲D Mesh ، Adaptive Routing ، Fault Tolerance ، Congestion Awareness ، Smart Detour ، HPRA ، EHPRA