Computational screening of photonic crystals using AI-driven methods
Annotatsiya
Photonic crystals (PCs) are crucial in optoelectronics, serving as electromagnetic materials with dielectric structures for applications such as mirrors and waveguides. They control light using photons, inducing resonance phenomena and acting as gratings to generate intense electric fields that boost nonlinear optical interactions. Designing artificial PCs at the nanoscale presents challenges for traditional high-throughput trials like evolutionary algorithms and neural networks. PCs influence light propagation via Bragg scattering, which allows for innovative applications like slowing light and creating high-Q lasers based on cavity geometry. These artificial structures manipulate light through periodic arrangements that align with the wavelength, offering a novel method of light control distinct from conventional optics. The unique optical properties of PCs diverge from traditional materials, prompting extensive research in optical electronics. The interest in automating photonic device design has led to developments in inverse photonics, with PCs providing a distinctive basis for this approach. An AI-driven, data-centric method screens designs for topological photonic crystal devices influenced by the spin-Hall effect. A dataset of 6,919 configurations is divided into ordinary and topological insulators, with surrogate models predicting high-level finite-difference time-domain calculations rapidly. This methodology accelerates exploration of complex design spaces while avoiding AI-related issues, producing a varied collection of topological crystal phases, six of which were fabricated and confirmed experimentally. The principles recovered facilitate model-aided inverse design of valley-Hall splitters based on desired functionality.