Radiation effects modeling of InP-based HEMT based on neural networks

Abstract

This paper has proposed a novel radiation effects modeling methodology based on neural networks for InP-based high-electron-mobility transistors (HEMTs). 2 MeV proton radiation has been performed with dose of 1 × 1012 H+/cm2, 5 × 1012 H+/cm2, 1 × 1013 H+/cm2, 5 × 1013 H+/cm2, 1 × 1014 H+/cm2. The radiation neural network models were comparatively constructed based on Feedforward Neural Network (FNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM). Results indicate that the LSTM network outperforms the FNN and RNN networks in the modeling for both drain-source current (IDS) and S-parameters, which demonstrates superior prediction accuracy with smaller fitting error. The proposed modeling approach offers an accurate characterization for the radiation effects of InP-based HEMT devices, without the need to consider the complex degradation process associated with radiation, thus providing practical guidelines for the space applications of such devices.