Data Block Partitioning Methods to Mitigate Stuck-At Faults in Limited Endurance Memories

Deep scaling in conjunction with increased process variation has resulted in increasingly faulty memories. Emerging memories, particularly phase-change and resistive memories, can experience stuck-at faults due to limited endurance. Partition and flip (PAF) schemes partition data into blocks and invert these blocks as needed to ensure data that is written matches the stuck-at cells. In this paper, we propose two novel correction schemes that substantially enhance the fault-tolerance capabilities of existing PAF techniques. First, dynamic partitioning increases the number of possible configurations with equivalent auxiliary bits. At high fixed error rates, the increase in configurations results in improved write error rates for flip-N-write and Aegis partitioning by 7%-72% and 5-53x, respectively. Our second novel partitioning method, relaxed partitioning, dramatically and effectively increases the partitioning search space by specifying minimally overlapping configurations. Through Monte Carlo simulations, data-aware dynamic partitioning tolerates 25% and 27% more faults over its lifetime than Aegis with 36 and 43 auxiliary bits per 512-bit data block, respectively, while relaxed partitioning achieves an extra 15% and 24% additional improvement while requiring two fewer overhead bits per data block.