Wave Propagation Approaches for Design Optimization of Asymmetric Double-Layer Gold Gratings: Logic- and Learning-Based Perspectives

This study investigates logic- and learning-based wave propagation approaches for optimizing asymmetric double-layer gold grating structures to maximize transmission and polarization performance. The logic-based wave method (LgWM) utilizes the multiple-layer model and transfer matrix calculations to identify optimal design parameters for specific wavelength targets, offering an efficient yet limited approach for single-wavelength optimization. In contrast, the learning-based wave method (LnWM) based on the explainable neural network (XNN) is developed to predict TM transmission spectra across a broad range of parameter combinations and wavelengths. By leveraging XNN frameworks, this LnWM addresses challenges of simulation domain constraints and parameter sparsity, enabling dense predictions across all design combinations. Results demonstrate that the LgWM achieves higher extinction ratios but is constrained in parameter flexibility, while the LnWM supports customizable objective functions and outperforms in multi-wavelength optimization. Specifically, the LnWM identifies a broadband optimal design with 2.5 times higher extinction ratios across 3–5 µm wavelengths compared to the LgWM. The complementary characteristics of these wave propagation approaches provide versatile tools for optimizing asymmetric grating structures, balancing efficiency and flexibility to meet diverse design objectives. The source code and trained models are made publicly available at https://github.com/AVIP-laboratory/Asymmetric-Au-gratings.