Accurate Modeling of GaN HEMTs and MMICs for Cryogenic Electronics Applications Utilizing Artificial Neural Network

The prosperity of quantum computing has boosted the development of cryogenic electronics. The wide bandgap gallium nitride (GaN)-based devices show the potential to develop such cryogenic systems due to their decent characteristics in harsh environments. However, robust and reliable device and circuit models for GaN devices are still absent and urgently in demands. Herein, we report an artificial neural network (ANN)-based compact model for both device- and circuit-level characteristics of GaN high-electron-mobility transistors (HEMTs) within a wide temperature range from 300 to 4.2 K. Based on the built ANN architecture, the dc and RF device models are produced, which can quickly capture and predict the relationship between device performance and multiple variables, including ambient temperatures, operating frequencies, biased voltages, and device sizes, with high accuracy over 99%. Furthermore, the ANN models are implemented in the circuit design of a GaN monolithic microwave integrated circuit (MMIC), showing a mean relative error (MRE) value of less than 4% under both 300 and 4.2 K compared with measured results. These results verified the robustness and generalization abilities of the proposed ANN models in the tape-out production of GaN-based devices and circuits for future quantum and cryogenic electronics applications.