A machine learning framework for predictive electron density modelling to enhance 3D NAND flash memory performance

Abstract

Data storage in electronic devices has been revolutionised by 3D NAND flash memory. However, polycrystalline silicon and grain boundaries offer issues that greatly affect memory performance in terms of string current and Program-Erase Threshold Voltage window (Vt –Window). Scientists need to learn more about how grain size, channel thickness, and trap density affect electron behaviour to improve the efficiency of memory chips. Regression models are used in this work to forecast fluctuations in electron density along the channel in 3D NAND string devices. The dataset, which was derived using TCAD simulations, has a sizable number of samples that show the electron density as a function of channel length. We assess their performance using R2 scores and RMSE values using regression models such as Linear Regression, Random Forest, K-Neighbour Regressor, Decision Tree, Gradient Boosting, XGBRegressor, CatBoosting Regressor, and AdaBoost Regressor. By improving our knowledge of how electrons behave in transistor channels, this work contributes to the optimisation of 3D NAND flash memory.

Keywords

• 3D NAND
• Flash memory
• Grain size
• Channel thickness
• Charge trap
• Machine learning