Efficient Gradient-Based Metasurface Optimization toward the Limits of Wavelength-Polarization Multiplexing.

Area of Science:

• Metasurfaces
• Optical Engineering
• Computational Photonics

Background:

• Metasurfaces enhance capacity using polarization and wavelength multiplexing.
• Simultaneously achieving limits for both techniques and computational efficiency remains challenging.
• Existing methods struggle with computational demands for complex tasks.

Purpose of the Study:

• To develop a computationally efficient method for simultaneous polarization and wavelength multiplexing in metasurfaces.
• To experimentally validate a gradient-based optimization algorithm using deep neural networks (DNNs).
• To demonstrate high-capacity multiplexing for advanced optical applications.

Main Methods:

• Developed a gradient-based optimization algorithm powered by deep neural networks (DNNs).
• Implemented a single-layer metasurface for nine multiplexed channels (3 wavelengths × 3 polarizations).
• Utilized the DNN for efficient inverse design of metasurface elements.

Main Results:

• Achieved simultaneous polarization and wavelength multiplexing limits with high computational efficiency.
• Demonstrated nine multiplexed channels on a single-layer metasurface.
• Reached 96% classification accuracy in simulations and 91.5% in experiments for a 36-class image recognition task.

Conclusions:

• The DNN-based method enables high-capacity multiplexing in metasurfaces efficiently.
• This approach overcomes computational limitations of previous multiplexing techniques.
• Sets a new benchmark for gradient-based inverse design in advanced optical elements.