Artificial Neural Network-Based (ANN) Physical Model for GaN HEMTs Based on Terminal Charge

An artificial neural network (ANN)-based physical model is proposed for accurately predicting the static characteristics of GaN High Electron Mobility Transistor (HEMT) devices, and two distinct modeling approaches are thoroughly investigated. The first approach leverages the MIT Virtual Source GaN-HEMT (MVSG) framework, where terminal charge data (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{q}_{\text{is}}$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{q}_{\text{id}}$</tex>) is initially generated via Silvaco TCAD simulations. This data is subsequently used to train the neural network, focusing on the source and drain terminal charges. The second approach directly models the drain currentvoltage (I-V) relationship using Silvaco simulations to create a training dataset for the ANN. In the final MVSG-based model, the neural network achieves a Mean Squared Error (MSE) of 0.0700 and a Mean Absolute Error (MAE) of 0.0837 for the source charge <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{q}_{\text{is}}$</tex>, and an MSE of 0.0173 and MAE of 0.0367 for the drain charge <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{q}_{\text{id}}$</tex>. When applied to the MVSG theoretical formulation, the predicted drain current yields an MSE of 0.0172 and an MAE of 0.0823. Additionally, the I-V model demonstrates an MSE of 0.0894 and an MAE of 0.0152, illustrating the superior performance of the ANN-based approach in terms of both accuracy and computational efficiency. These results highlight the effectiveness and potential of ANN models in advancing GaN HEMT device modeling.