Machine Learning-Based Prediction of Threshold Voltage Distribution Due to Lateral Migration in 3-D NAND Flash Memory

Abstract

In this study, we propose a deep neural network (DNN)-based simulator and method to predict and analyze the changes in threshold voltage (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {t}}$ </tex-math></inline-formula>) distribution due to the lateral migration (LM) in 3-D NAND flash memory. We focus on the most dominant failure mechanism in long-term retention, LM, by considering influencing sources such as cell variation (CV), trap variation (TV), and random pattern (RP) simultaneously. As a key component of our method, we introduce the retention slope, which represents how quickly data in a cell degrades over time. By training the DNN model on this, we can predict the degradation rate of each cell under various conditions. After confirming the model’s accuracy through test evaluation, the DNN-based simulator performs a Monte Carlo simulation of random strings to predict <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {t}}$ </tex-math></inline-formula> distribution at multiple retention time (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${t}_{R}$ </tex-math></inline-formula>) points. Detailed analysis of the results shows the impacts of each source and provides a more comprehensive understanding of LM than previous studies. Our approach can offer valuable insights and guidelines for improving retention characteristics of 3-D NAND flash memory.