Extended Multi-Region Hamming-based ECC for Mitigating Clustered Faults in 3D Memories
摘要
Three-dimensional stacked memory architectures provide high bandwidth and density while exhibiting increased susceptibility to radiation-induced single-bit upsets and clustered multi-bit upsets. Traditional Hamming-based error-correcting codes offer narrow fault protection. This work proposes an extended multi-region Hamming-based ECC that incorporates redundant bits for each row, column, and diagonal, enhancing resilience to multi-bit errors in both 2D and 3D memories. The approach is evaluated by using exhaustive fault-injection analysis. To assess the ECC’s performance, 30,000 randomly generated fault cases were evaluated by injecting 1 to 12 errors, and were compared to previous works. The proposed ECC consistently outperforms existing works, achieving 90% correction capability up to eight errors. In the context of 3D memories, two data configurations were evaluated: 8 × 8 and 6 × 8, each analysed under four distinct physical bit-placement layouts. For each configuration, 100,000 incident-centred multi-die fault injections were performed across three error intensity profiles. On average, the 8 × 8 data configuration achieves correction rates exceeding 90%. Notably, the 6 × 8 data configuration, despite employing 64 fewer redundant bits, attains comparable correction performance of 90% and, in several cases, outperforms the 8 × 8 configuration. The results suggest that combining the extended multi-region Hamming ECC with strategic bit mapping enables reliable and redundancy-efficient fault tolerance in 3D memory systems.