Machine Learning-Driven Modeling for Fast and Accurate Z-Interference Estimation in 3D NAND Scaling

This study presents a CatBoost-based regression model to predict Z-interference (Z-ITF) in scaled 3D NAND Flash. Unlike traditional TCAD or curve fitting, the model uses initial/final attack and victim cell states, improving adaptability to future scaling. It accurately predicts Vth shifts across various WL pitches, showing a 2-3× Z-ITF increase as ON pitch shrinks from 63 nm to 50 nm. The model helps identify critical scaling limits for improved reliability and error correction.