Answers are grounded in this paper and cite their sources.
Resistive nonvolatile memories (NVMs) promise significant performance improvement over existing NVM candidates. However, fabrication nonidealities and parasitics on the access path cause cell location-dependent variations in the total resistance received at the read circuitry. Write characteristics delivered to each cell, as well as the optimal write conditions for each cell, are also location-dependent. In this article, we propose a 2-D calibration scheme to address these variations. The proposed scheme joins row and column calibrations to create a correction grid at each crosspoint on the array and effectively cancels many spatial patterns. The enabling circuit and algorithmic modifications are described. We assess the 2-D calibration scheme in a 28-nm256× 256memory array, and show reduction in variability across multiple gradient patterns compared to conventional calibration methods. For the same calibration granularity, 2-D calibration achieves between 41% and 99% improvement depending on the amount of calibration bits. For the same amount of total calibration bits, the 2-D calibration scheme reduces the variability between 39% and 99%.
No takes yet. Share an insight, caveat, or question.
Lee et al. (2020) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: