Arbitrary Multifunctional Vortex Beam Designed by Deep Neural Network

As topological charge constitutes an infinite-dimensional Hilbert space, vortex beam has numerous applications in optical communications and other fields where signal capacity is a vital requirement. Multifunctional vortex beams, showing up to different controllable responses subjected to separate combinations of polarization states, have significantly exhibited improved capacity of signal transport. Relying on prior physical knowledge, complex requirement brings tremendous challenge to the design of multifunctional vortex beams. Here, a deep-learning-based platform for designing metasurfaces is proposed, which can intelligently generate predesigned multifunctional vortex beams. Employing the proposed strategy, the demonstrations of bifunctional and trifunctional vortex beams are consistent with the design targets. Three samples are fabricated and measured by a Michelson interferometer. Clear observed interference patterns revealed the topological nature of the generated vortex beams, unambiguously justifying the design platform. This intelligent design strategy, which may inspire new ideas in other scientific fields, lays a solid foundation for the high-performance application of multifunctional vortex beams. This work fully exploits the potential of vortex beams for large-scale dense data communication and quantum optics with high quantum numbers, which may further promote the development of the integrated photonic chip.