Photonic modes prediction via multi-modal diffusion model

Abstract

Abstract The concept of photonic modes is the cornerstone in optics and photonics, which can describe the propagation of the light. The Maxwell’s equations play the role in calculating the mode field based on the structure information, while this process needs a great deal of computations, especially in the handle with a three-dimensional model. To overcome this obstacle, we introduce the multi-modal diffusion model to predict the photonic modes in one certain structure. The Contrastive Language–Image Pre-training (CLIP) model is used to build the connections between photonic structures and the corresponding modes. Then we exemplify Stable Diffusion (SD) model to realize the function of optical fields generation from structure information. Our work introduces multi-modal deep learning to construct complex mapping between structural information and optical field as high-dimensional vectors, and generates optical field images based on this mapping.