Bias dependent scalable noise models of MESFETs/HEMTs based on neural networks

Abstract

A bias-dependent scalable microwave MESFET/HEMT noise model is proposed in this paper. It is based on a multilayer perceptron neural network that produces noise parameters at its outputs for device gate width, biases and frequency presented at its inputs. In that way determination of the noise parameters is enabled for various values of gate width and for all operating points over a wide frequency range. Once the network is trained its structure remains unchanged. After the network training, the noise parameters determination is done without additional optimizations and without need for the measured data that are required for the network training only.