Artificial neural network models for metal-ferroelectric-insulator-semiconductor ferroelectric tunnel junction memristor

Abstract

Metal-Ferroelectric-Insulator-Semiconductor (MFIS) structure ferroelectric tunnel junction (FTJ) memristor becomes one of the most promising candidates for next-generation memories. Two numerical simulation methods have been developed and analyzed previously to calculate the tunneling current in the MFIS-FTJ, one by using the Wentzel–Kramers–Brillouin (WKB) method with the band profile obtained from Poisson's equation, the other by self-consistently solving Poisson's equation and the drift-diffusion transport equations with a tunneling-induced carrier generation rate. However, numerical methods can be computationally expensive, especially for device design with various parameters. In this work, an artificial neural network (ANN) model that can predict the device performance is proposed, the model can reduce computational cost while maintaining good accuracy. In addition, to further investigate the applicable conditions of the two simulation methods mentioned above, we also develop an ANN model to predict the relative differences between the two methods' results under different conditions, and the prediction results show good agreement with numerical results.