Vectorial-Holography metasurface empowered by Orthogonality-Simplified Machine learning.

[Display omitted] • We propose the orthogonality-simplified machine learning framework to design vectorial-holography metasurface. • A structure with orthogonality is designed to independently control x- and y- polarized waves. • Resistors are embedded in meta -atoms to tailor amplitude. • The design of machine learning architecture is simplified based on the physical principle of orthogonality. • The vectorial-holography metasurface is designed by customizing the polarization. Metasurfaces can provide unprecedented degree of freedom in manipulating electromagnetic waves and have been introduced to holography. Aiming to explore the full capability for information presentation, vectorial metasurface holography is sprung up, which exhibits mesmerizing capability in carrying information compared with scalar counterpart. However, the more flexible modulation means the more complex design dimension, which hinders the development of vectorial metasurface holography. In this work, we propose an orthogonal I-shaped structure embedded with resistors to synthesize arbitrary polarization with less crosstalk. Benefiting from the independent control of phase and amplitude on orthogonal base, the orthogonality-simplified machine learning framework is employed to assist vectorial metasurface holography design. As a proof-of-concept, a vectorial metasurface holography carrying multi-polarization information was designed, simulated and measured. All the results exhibit a high degree of consistency, which fully demonstrates the effectiveness of our design. Encouragingly, our method paves a new route to simplify machine learning framework based on physical significance, which can be handily extended to more functional structures. [ABSTRACT FROM AUTHOR]