Modeling of Psuedomorphic High Electron Mobility Transistor Using Artificial Neural Network

Abstract

AbstractModeling electronic devices has become a numerous work needed to keep up the development of the applications of the communication systems. Since the physical modeling of electronic devices is complex and time consuming. Recent research is using machine learning in modeling electronic devices, especially Artificial Neural Network (ANN) to reduce complexity and time consumption. In this paper, an enhanced model using ANN is proposed to model the scattering (S-) parameters in frequency range from 0.5 to 18 GHZ of GaAs Pseudomorphic High-Electron-Mobility-Transistor (pHEMTs). The proposed model uses four parameters which are the drain-source voltage, the drain–source current, the channel width, and the operating frequency as input parameters for the proposed ANN model to produce s-parameters for the pHEMT. Excellent agreements between the manufacturer’s datasheet (ATF-34143) and model-calculated data are achieved. The ANN outputs are represented in the form of amplitude-phase and achieved higher performance than existing models. In modeling S11 the best training performance reached is 5.642 *10–16 at epoch 392, in modeling at S21 the best performance reached is 2.869*10–21 at epoch 60, in modeling S12 the best performance reached is 4.3601*10–20 at epoch 20 and in modeling S22 the best performance reached is 5.7795*10–23 at epoch 144.KeywordsArtificial neural networks-parametersPseudomorphic high-electron-mobility-transistor