Rapid inverse design of large-scale freeform meta-optics with the neighborhood-attention transformer

Abstract

Metasurfaces are progressively reshaping traditional optical paradigms and pushing the boundaries in complex applications where compact designs are essential. However, the design of metasurfaces demands substantial computational resources to numerically solve Maxwell's equations—particularly for large-scale photonic systems. Conventional forward design using electromagnetic solvers is based on specific approximations that may not effectively address complex problems. In contrast, existing inverse design methods are a stepwise process that is often time-consuming. Here, we overcome these challenges by presenting MetaE-former architecture, a Neighborhood Attention Transformer-based transfer learning framework that enables customized surrogate solver development through fine-tuning of pre-trained neural networks with only thousands of data, facilitating highly efficient task-adaptable inverse design of metasurfaces. Moreover, this method achieves global solutions for hundreds of nanostructures simultaneously, providing up to a 250,000-fold speedup during the optimization stage compared with solving for individual meta-atoms based on the FDTD method. As examples, we demonstrate a binarized high-numerical-aperture (~ 1.31) metalens and several optimized structured-light meta-generators. Our method significantly improves the beam shaping adaptability with metasurfaces and paves the way for quick design of large-scale metadevices with high accuracy.