Neural Network Modeling of Antennas on Package for 5G Applications

The design and optimization of microwave devices require rigorous electromagnetic (EM) analysis requiring excessive computational resources. To mitigate this, neural network based machine learning methods can be used to generate surrogate models for EM performance analysis. In this study, we constructed a spectral transposed convolutional neural network model for a microstrip patch antenna printed on a package. The model takes geometrical parameters and material properties of the antenna as inputs and generates S-11 parameter within the band of 23-33 GHz as output. The results showed a very good correlation between simulations and predictions. This verifies that NN based models can be used for EM analysis of antennas, particularly at the earlier stages of a design process.