Scalable and Rapid Design of Pixelated Passive Networks via S-Parameter Matrix Synthesis and Cascaded Deep Learning

Designing large-scale pixelated passive networks is often hindered by high simulation costs and data scarcity. A scalable design methodology is proposed to overcome simulation costs and data scarcity by utilizing S-parameter matrix synthesis and cascading network architectures. Training samples are rapidly generated via matrix operations, minimizing reliance on full-wave simulations. Design complexity is managed by cascading smaller models to represent larger topologies, reducing dataset dependency. The approach is demonstrated through simple and extended bandpass filters, confirming that high-performance designs are achieved with significantly reduced turnaround times.