Input Skip-connections Feedforward Neural Network for Parametric Electromagnetic Modeling of Passive Components

As an effective parametric modeling technique, the artificial neural network-based model provides accurate and rapid predictions of electromagnetic (EM) response behaviors for devices. This paper proposes an input skip-connections feedforward neural network (ISFNN) for parametric modeling of passive components. The ISFNN facilitates feature combination by introducing multiple skip-connections that incorporate design features into intermediate hidden layers. Solely through the network architecture design, the ISFNN enables effectively learning the complex nonlinear relationships between geometrical param-eters and EM responses. A three-pole H-plane filter example is employed to verify the effectiveness of the model. Compared to other modeling methods, the ISFNN achieves the smallest testing errors. This model achieves better modeling performance with fewer training samples, thereby reducing the time required for model development.