A Neural Network-based Manufacturing Variability Modeling of GaN HEMTs

A new technique to accurately model the manufacturing variability of GaN HEMT using a neural network(NN) is presented in this paper. Compact model parameters are automatically generated through Principal component analysis (PCA) parameters from variations in I-V data. Together with the bias conditions, the compact model parameters are used to train a neural network. The NN-based compact model captures the I-V behavior of 115 GaN HEMT with excellent accuracy. The trained neural network is converted to a standard Verilog-A file that can be imported to a circuit simulator. The NN-based compact model is further evaluated in terms of complexity and simulation speed. The presented technique shows great potential in developing a fast, flexible, and accurate NN-based compact model that can be applied to any device technology.