Machine learning-assisted characterization of optical forces near gradient metasurfaces

Gradient metasurfaces provide a rich platform to control and manipulate optical forces, particularly appealing to design customized optical traps for nanoscale objects. A challenging step in the optimization and design of such structures is the forward simulation, as the metasurface geometry can be highly spatially variant while unit-cell simulations are insufficient to model the local surface-particle interactions. Here, we present a deep-learning-based approach to accurately and efficiently model near-field optical forces in complex nanostructure configurations. While traditional full-wave simulations are computationally expensive and not easily scalable, our deep learning model can capture a wide range of local interactions, ideal for optical trap designs. Our model predicts optical forces both parallel and normal to the surface at different distances above the surface. After an initial training stage, this approach significantly reduces computation time while maintaining high accuracy, making it a valuable tool for designing optical nanotweezers.



This material is based upon work supported by the National Science Foundation under Grant No. 2138869.