Fast Frequency-Domain Analysis for Parametric Electromagnetic Models Using Deep Learning

The design of microwave devices relies on the modeling and simulation of mathematical models of such devices. Usually, the mathematical models are very large, causing long simulation times and making the design process very slow. In this work, we propose a deep learning method for fast frequency-domain simulation of parametric models of microwave devices. We focus on the fast evaluation of the parameter-dependent transfer function via approximating the transfer function with a neural network. After being properly trained using limited data of the transfer function, the trained neural network can evaluate the transfer function at any parameter sample with acceptable accuracy. The proposed deep learning method is tested on a PEEC model of a microwave device and shows its efficiency in fast predicting the transfer function.