Near-field distribution prediction and inverse design of spherical hyperbolic metamaterial cavity using multimodal models

Abstract

The near-field information of optical structures is vital for understanding the behavior of materials at minuscule scales and innovating new technologies. However, the acquisition of near-field information involves substantial data, and traditional numerical simulations are constrained by computational resources and time. To solve this problem, we propose a method that integrates the multimodal model Stable Diffusion with the deep subwavelength spherical hyperbolic metamaterial cavity, to establish the mapping relationship between the electric field distribution and the cavity parameters. After training, the model can accurately and quickly predict the electric field map at preset structural information, and it is highly consistent with simulated results. Additionally, we use the Contrastive Language-Image Pretraining (CLIP) algorithm to retrieve the structural information based on the given electric field distribution, for the inverse design of the metamaterial cavity. The results demonstrate that our method provides a fast and accurate way to obtain near-field information and also can accelerate the on-demand design of optical metamaterials and other structures, with promising implications for real-world applications.