A Scalable Knowledge-Based Neural Network Model for GaN HEMTs With Accurate Trapping and Self-Heating Effects Characterization

In this article, a scalable knowledge-based neural network (KBNN) large-signal model of gallium nitride (GaN) high-electron-mobility transistors (HEMTs) with accurate trapping and self-heating effects characterization is developed. An improved empirical drain current model is proposed and added to the neural network as prior knowledge, thereby establishing the drain current model, including self-heating effect. A new empirical equation is proposed to model the buffer-related trapping effect more accurately. Taking Angelov capacitance models as prior knowledge, the KBNN capacitance models are completed. Moreover, the scaling characteristics of the proposed KBNN model are studied. The developed model has been fully verified by different sizes of GaN HEMTs. Good agreement between the model simulation results and the measurement data, including current–voltage ( $I$ – $V$ ), $S$ -parameters, power characteristics, and load-pull data, confirms the effectiveness of the proposed model. The proposed scalable KBNN model is fast and accurate and would be useful for accurate large-signal modeling of large gate periphery GaN HEMTs for high-power radio frequency (RF) applications.