Integrating Cognitive Models into Network-on-Chip Architectures for Resilient and Adaptive Autonomous Systems

As autonomous systems grow in complexity, the need for intelligent, resilient on-chip communication infrastructures becomes critical. Traditional Network-on-Chip (NoC) architectures, while effective for parallelism and throughput, struggle to adapt to the dynamic demands of robotics, edge AI, and real-time processing. This paper proposes a cognitive NoC architecture embedding lightweight reinforcement learning agents within each router, transforming interconnects into distributed, self-adaptive systems. Agents perceive local states, learn optimal routing strategies, and adapt dynamically to faults and traffic variations. Simulation results demonstrate substantial improvements in latency, throughput, fault recovery, and convergence, with modest hardware overhead. This work advances the cognitive hardware paradigm by establishing a decentralized intelligence framework for scalable, resilient computing in next-generation AI platforms.