Knowledge-based neural network SPICE modeling for MOSFETs and its application on 2D material field-effect transistors

Abstract

As the traditional scaling of silicon metal-oxide-semiconductor field-effect transistors (MOSFETs) reaches its physical limit, research efforts on novel semiconductor devices are increasingly desired. To enable the joint optimization of early-stage circuit design and process of novel devices, the rapid creation of an accurate compact model of these devices with the capability to cover process variations is required. In this work, a knowledge-based neural network (KNN) modeling method is proposed. This method separates the geometrical variables from the other input variables of the device, where the geometrical variables are modeled with physics-based analytical equations, while the remaining part is modeled by an artificial neural network. The KNN model takes advantage of the automated numerical fitting capability of the neural network and the geometrical scalability from device physics. The created KNN model is first validated with silicon MOSFET data from the industry standard BSIM6 and shows more than 20% accuracy improvement as compared with the traditional neural network model. Furthermore, MoS₂ field-effect transistors and circuits, such as ring oscillators, standard cells, and logic functional circuits, are experimentally fabricated for model verification. The results show that the KNN model is capable of predicting the electrical characteristics of devices beyond the measurement geometry and facilitates the accurate simulations of statistical circuits with respect to experimental data. This work paves the way for future circuit designs and simulations of novel semiconductor devices.