Machine learning-based model inference for spectral response of photonic crystals

Photonic Crystals (PhCs) are materials with a periodic arrangement of dielectric or metallic components that can manipulate the flow of photons. Conventional techniques for computing the spectrum response of these artificial structures are time-consuming, laborious, and susceptible to human errors. This paper presents a novel approach incorporating Machine Learning (ML) in forming PhC-based periodic structures. The structures are designed from the Two-Dimensional (2D) PhCs having air holes in a dielectric material. These crystalline structures work on the Guided Mode Resonance (GMR) principle and find their use in numerous applications, including optical filters in Near Infrared Range (NIR). However, the conventional methods for analyzing the output spectra of PhCs need to be revised due to the complexities in designing compound optical filters. Therefore, an automated process employing ML is required. As a result, in our work, we form 2D crystalline structures using the Finite-Difference Time Domain (FDTD) method and predict various spectral responses of PhCs via ML-based linear regression. The proposed mathematical models provide efficient results and considerably less simulation efforts and time compared to the traditional manual methods.