An odd-even scheme to prevent a packet from being corrupted and dropped in fault tolerant NoCs

Packet corruption, misrouting, and dropping have become an extra burden on network performances due to stuck-at and open faults on network-on-chip (NoC) interconnects. Existing works for testing interconnect faults have addressed either shorts and/or stuck-ats with the assumption that the opens do not exist on interconnects. A new distributed test scheme that addresses coexistent stuck-at and open faults on NoC interconnects is proposed. The scheme is governed by a set of odd/even router and cores and takes account of testing of a subset of interconnects in turn. Results achieve 100% fault coverage in terms of packets received and dropped, and test coverage in terms of link-wires tested. Results also show evaluation of different performance metrics affected by the faulty links in a NoC.