Inverse design of topological valley-locked slow light rainbow trapper based on physics-aware latent diffusion model.

Valley photonic crystals (VPCs), leveraging topologically protected edge states, provide robust mechanisms for light wave manipulation and propagation. The design of VPCs primarily relies on the configuration and arrangement of the unit cell structures. In conventional structural design, the trial-and-error approach relying on prior structural templates and full-wave simulations leads to significant inefficiencies. In recent years, various inverse design algorithms have been widely adopted for structural generation. However, these methods often fail to meet the demands of multi-objective structural generation under limited physical constraints. We propose a physics-aware latent diffusion model (PALDM). This generative framework enables efficient generation of VPC unit cell structures by embedding physics-aware constraints into the latent diffusion process. Using PALDM, we designed ten different unit cell structures with parametric gradient variations. The topological slow light waveguide composed of these structures achieved slow light rainbow trapping in the topological bandgap, validating the capability of PALDM to generate VPCs with tailored frequency response. The physics-aware approach thus offers a pathway for advanced topological photonic device design.