Modeling of GaN HEMT I-V Characteristics Using Artificial Intelligence: Optimization Study

The drain-source current versus the drain-source voltage characteristics (Ids-Vds) of the Gallium Nitride (GaN) HEMT are modeled by applying the Artificial Neural Networks (ANN) approach. This modeling method needs a great amount of training data, which means that we had to measure many Ids-Vds characteristics for different gate source voltages (V gs) values that is not of well practice. Furthermore, an arbitrary choice of the samples used in the training process can increase the model error. An accurate method for optimizing the number and the values of V gs values, utilized in the training step, and for optimizing the ANN neurons number is carried out. We have used the Genetic Algorithm (GA) and the Levenberg-Marquard Algorithm (LMA) to optimize the ANN weights and biases. A weak model error is achieved with the LMA optimizer by applying the mean square error as a fitness function. Then an optimization algorithm is developed in the purpose to get optimal values and numbers of V gs and to get optimal numbers of ANN neurons for ANN training. The optimization methodology is validated by using simulated Ids-Vds curves.