Prediction of Random Telegraph Noise-Induced Threshold Voltage Shift and Its Scaling Dependency Using Machine Learning

Abstract

Random telegraph noise (RTN) shifts the threshold voltage (Vt) of 3D NAND flash memory cells, making it a key factor of the device malfunction. The aim of this study is to predict the distribution of RTN induced ${\mathrm { V}}_{\mathrm { t}}$ shift in 3D NAND flash memory. Artificial neural network (ANN)-based machine learning (ML) is used for this prediction. With 2000 samples, ANN is trained and tested to predict the ${\mathrm { V}}_{\mathrm { t}}$ shift of random cells with high reliability. Furthermore, ANN is applied to predict the tendency of RTN-induced ${\mathrm { V}}_{\mathrm { t}}$ shift in scaled 3D NAND. Compared to prior works which has required far more measurements or simulations, the predictions are shown to shorten the time spent to obtain the distribution. Based on these predictions, the dependency of the decay constant on cell variation is investigated, which is a most critical parameter in analyzing the RTN distribution. This indicates that it is possible to apply ANN-based ML to predict various characteristics of 3D NAND flash memory in a much shorter time and to develop numerical models of related parameters.

Keywords

• 3-DNAND flash memory
• random telegraph noise (RTN)
• polycrystalline Si (poly Si) channel
• macaroni MOSFETs
• artificial neural network (ANN)
• machine learning (ML)