Comparative Analysis of TCAD augmented ML Algorithms in modeling of AlGaN/GaN HEMTs

Abstract

In this study, a computer-aided design (TCAD) supported machine learning framework is built to predict the intrinsic parameters of GaN HEMT, such as VTH (Threshold Voltage) and gm (Transconductance). TCAD was used to generate the training data set constituting the ID-VGS characteristics of the GaN HEMT. This is achieved by changing multiple input parameters (e.g. the Al mole fraction (x), gate metal work function, AlGaN barrier thickness and gate length). We deployed numerous ML algorithms and an ANN (artificial neural network) to predict the VTH and gm of GaN HEMT. We compared the performance of these ML algorithms and found that the boosting and ensemble algorithms provide better results in terms of accuracy. We showed that Random Forest and Gradient Boost were most effective in predicting VTH with an R² value of 0.99 each, and for gm prediction, Gradient Boost was most effective with an R² of 0.92.