Novel Agent-Based Management for Fault-Tolerance in Network-on-Chip

We introduce a novel agent-based reconfiguring concept for futures network-on-chip (NoC) systems. The necessary properties to increase architecture level fault tolerance are introduced. The system control is modeled as multi-level agent hierarchy that is able to increase application fault-tolerance and performance with autonomous reactions of agents. The agent technology adds a system level intelligence level to the traditional NoC system design. The architecture and functions of this system are described on conceptual level. Communication and reconfiguring data flows are presented as study cases. Principles of reconfiguration of a NoC on faulty environment are demonstrated and simulated. Probability of reconfiguration success is measured with different latency requirements and amount of redundancy by Monte Carlo simulations. The effect of network topology in reconfiguration of a faulty mesh was also under research in the simulations.