Deep learning enabled topological design of exceptional points for multi-optical-parameter control

Abstract

Abstract Metasurfaces are 2D artificial nanostructures that exhibit fascinating optical phenomena and flexible capabilities. Multi-optical-parameter metasurfaces have advantages over single-function or single-dimensional metasurfaces, especially in practical applications like holography, sub-diffraction imaging, and vectorial fields. However, achieving multi-optical-parameter control is challenging due to a lack of design strategy, limited manipulation channels, and signal-to-noise ratio problems. Exceptional points (EPs) possess inherent polarization decoupling properties and allow for amplitude and wavelength modulation, opening up research prospects for multi-optical-parameter electromagnetic field modulation and developing compact integrated devices. Leveraging deep learning, we observe topological charge conservation and utilize the topologically protected optical parameter distribution around scattered EPs. Based on these, we introduce amplitude-phase multiplexing and wavelength division multiplexing devices. Our work allows rapid and precise discovery of EPs topology, offers a powerful tool for digging related physics, and provides a paradigm for multi-optical parametric manipulation with high performance and less crosstalk, which is critical for imaging, encryption, and information storage applications.