Efficiently using data splitting and retransmission to tolerate faults in networks-on-chip interconnects

In this paper, a fault tolerance technique is proposed for the protection of NoC interconnects. The technique is based on data splitting and retransmission, providing for high NoC-based SoC reliability. For error detection, the proposed approach splits the communication data into two parity encoded halves. For error correction, only the erroneous half is doubled and retransmitted. Compared to the use of a conventional Hamming Code, we show that, for medium to large NoCs, the approach proposed in this work is less silicon costly, leads to the design of faster fault-free routers and, for the majority of faults, presents no performance degradation at system level.