Deep Learning-Based ASM-HEMT High Frequency Parameter Extraction

A fast and accurate deep learning (DL) based ASM-HEMT high frequency (HF) model parameter extraction is presented for the first time. The parameter extraction starts with creating a nominal model by extracting ASM-HEMT I–V parameters. The nominal model is used for Monte Carlo simulation of preselected ASM-HEMT HF parameters to generate 90K training data, with a total of 796 million S-parameter data points from a frequency sweep of 14 different bias conditions. The DL model is then trained to instantly predict ASM-HEMT HF parameters from the S-parameter data. The results show that the proposed approach can provide accurate model results, obtaining an error lesser than 10%. The presented approach shows a fast and accurate means for HF parameter extraction with an accuracy typically achieved in manual parameter extraction.