기계 학습을 이용한 셀 문턱 전압 분포 기반의 Soft Decision 센싱 레벨 예측

Abstract

Along with the rapid growth of the NAND flash memory market, the phenomenon of stacking 3D NAND flash memory is also steadily increasing. As the number of stacks increases, NAND flash memory will inherit different cell threshold voltage distributions for different physical characteristics. Furthermore, this phenomenon intensifies as the deterioration of data neglect is added. Considering the threshold voltage of these various cells, it becomes difficult to derive the sensing level during the operation of embedded memory products. In this paper, we propose a Sensing Level (SL) prediction method for making Soft Decision (SD) using machine learning. The proposed method experimentally confirmed the possibility of constructing a model that reflects the threshold voltage distributions of various cells. The prediction accuracy of the model confirmed an excellent performance of 94 to 99%, improving 36 to 52%p compared to that of the probability-based prediction method.