An Automatic Parameter Extraction Method Based on Autoencoder for PIN Diode Model

This paper proposes a machine learning accelerated automatic parameter extraction method for non-linear models. The generalization ability of artificial neural network (ANN) for nonlinear systems enables the trained networks to accurately capture the nonlinear characteristics between the input and output of device models. This method only requires a random combination of parameters and the corresponding results as the training set without predefining parameter distributions. The parameter extraction of the test data takes only a few seconds, avoiding the iterative process when facing new data. A PIN macro model is used for demonstrating the efficiency and accuracy of this method, which shows an MSE of 0.77% for DC part with 800 training groups and an MSE of 0.56% for transition part with 600 training groups.