Hierarchical Agent Monitoring NoCs: A Design Methodology with Scalability and Variability

Adopting the bio-inspired system architecture, a hierarchical agent monitoring NoC design method is proposed. Based on circuit conditions traced at the run-time, system settings including resource utilization and power supplies are configured hierarchically and adaptively at each architecture level by corresponding agents. This methodology provides systematic design approaches to VLSI circuits under the influence of unpredictable variations and stringent power constraints. Each level of agents autonomously monitor their local components while being supervised by their higher level counterparts. In this manner, the overall performance and efficiency can be maximized for systems of any scale. The feasibility and innovation of this methodology are explained elaborately with low-power optimization and fault-tolerance services, and architectural alternative analysis on the regular NoC platform is presented. An FFT study case demonstrates the functional effectiveness and flexibility of agent-monitored adaptive system configuration in case of component failures.