Machine Learning-Based Investigation of Temperature Effects on Lateral Migration in 3-D nand Flash Memory

In this study, we investigate the temperature effects of lateral migration (LM) in 3-Dnandflash memory using a machine learning (ML) approach. The LM is not only the dominant failure mechanism in long-term retention but also significantly degrades the lifetime estimation at room temperature (RT) due to its temperature dependence. Therefore, we model deep neural networks (DNNs) at$25~^{\circ }$C,$55~^{\circ }$C,$85~^{\circ }$C, and$115~^{\circ }$C, and embed them into a Monte Carlo framework to predict and analyze the effects that are difficult to capture through measurements. In this process, cell variation (CV) and trap variation (TV) arising from limitations of the fabrication process, as well as pattern dependence (PD) from program data, are considered. To cover various conditions of retention time (${t}_{\text {R}}\text {)}$points and bit errors, the stretched exponential function (SEF) was employed, and the variations of its parameters with temperature are analyzed. A key finding of this study is that the extracted activation energy (${E}_{\text {a}}\text {)}$is nonconstant, depending on the error threshold and variation in CV, TV, and PD. The results from our methodology provide deeper insights into LM and also offer design guidelines for more accurate lifetime estimation in 3-Dnandflash memory.