Deep learning enabled moiré chiral metasurfaces

Moiré metasurfaces stacked with twist angles exhibit remarkable optical chirality. However, the quasi-periodic nature of moiré superlattices introduces substantial computational challenges across a wide range of twist configurations. These challenges arise from the simultaneous presence of extremely large supercells at small twist angles, the high dimensionality of the design parameter space, and the strongly nonlinear relationship between structural geometry and optical response. The combined effect of these factors significantly complicates both the forward modeling and inverse design of moiré-based systems. To overcome these limitations, an innovative deep learning-based design strategy is proposed to accelerate the prediction of spectra and the inverse design of chiroptical responses in moiré chiral metamaterials. The results demonstrate that the proposed model can efficiently and accurately perform both forward prediction and inverse design. Moreover, it enables the customization of metasurfaces through digital fitting, significantly reducing the time and resources required compared to traditional design approaches. Furthermore, the study explores the application of moiré chiral metasurfaces in the differentiation of biomolecular enantiomers, showcasing their advanced capabilities for enantiomer-specific biological analysis. These findings offer valuable insights into the fundamental design principles of twisted stacked aperture-based metasurfaces and lay the groundwork for developing novel chiral metamaterials with exceptional performance.