Modeling Power-Bus Structures with Physics-Informed Neural Networks

This paper presents electromagnetic modeling of power-bus structures with physics-informed neural networks. A deep neural network (DNN) is used as a solution surrogate to the Helmholtz equation describing a physical law of voltage distributions in the parallel-plate structures. Its equation is directly embedded in the loss function of the DNN to be trained. The developed method can be regarded to be data-free and mesh-free. It is applied to the analysis of input impedance of a circular power-bus structure with dielectric and conductor losses. The predicted result is verified in comparison with the theoretical one.