A Physical Insight Based Neural Network Model for Small Signal Behavior of MOSFET in Terahertz Applications

In this paper, a physical insight based neural network small signal MOSFET model is proposed, which can accurately and efficiently describe the high-frequency behavior of MOSFET. By incorporating prior physical knowledge in the preprocessing stage of neural networks, the model demonstrates the enhancement of precision and reliability and is capable of describing non-quasi-static effects as well as other parasitic effects and second-order effects precisely. The proposed model is validated with the small signal data obtained by TCAD simulation up to 1THz, and the model can predict high-frequency y-parameters with 3σ error less than 3%.