Prediction of Electrical Characteristics of Fin Field-effect Transistor Devices Based on Simulation Using Deep Learning Method

Nowadays, silicon-based fin field-effect transistor (FinFET) devices have become the top choice for integrated circuit designers because they can handle the majority of application scenarios. In this article, the authors propose applying deep learning models to predict the electrical characteristics of devices using their structural parameters, aiming to solve the problems of complexity, time consumption, and convergence difficulty in traditional simulations. The authors first determine the electrical characteristics of simulations on FinFET devices using technology computer-aided design (TCAD). Different deep learning models were constructed in this study to predict various electrical parameters and characteristics of integrated circuit devices based on different datasets and prediction tasks, and high levels of accuracy were achieved. For instance, the average normalized mean square error of the predicted electrical parameters of FinFET devices based on TCAD simulation was less than 1.4 x 10-5, while the average relative errors of the predicted DC and AC characteristics of FinFET devices based on Berkeley short-channel insulated gate FET model (BSIM) simulation were 7.12 x 10-3 and 4.8 x 10-3, respectively. These results demonstrate that the proposed deep learning models can effectively predict the electrical parameters and characteristics of integrated circuit devices, providing strong support for device design and optimization.