Reconfigurable origami hologram based on deep neural networks

Area of Science:

• Metasurfaces and Nanophotonics
• Holography and Optical Engineering
• Materials Science and Engineering

Background:

• Reconfigurable and multifunctional metasurfaces are crucial for diverse applications but often suffer from structural complexity and bulk due to active components.
• Existing metasurface designs face challenges in achieving both reconfigurability and simplicity.
• Passive metasurfaces offer a potential solution to reduce complexity and volume.

Purpose of the Study:

• To propose and demonstrate a reconfigurable passive hologram utilizing an origami structure.
• To achieve dynamic control over holographic patterns by transforming between planar and 3D configurations.
• To explore the application of deep neural networks for phase control in reconfigurable metasurfaces.

Main Methods:

• Development of a passive metasurface based on an origami structure capable of mechanical deformation.
• Utilizing deep neural networks trained on an NVIDIA Tesla k80 GPU to determine phase distributions for holographic generation.
• Demonstrating hologram generation of 'Z' and 'L' shapes under right-hand circular polarization (RHCP) in planar and zigzag states.

Main Results:

• Successful generation of distinct holograms ('Z' and 'L') in planar and 3D zigzag configurations (35° slant angles).
• Mechanical stretching and compression enable seamless transformation between 2D and 3D metasurface states.
• Deep neural network training achieved in 11.88 seconds over 100 epochs for phase optimization.

Conclusions:

• The proposed origami-based reconfigurable passive metasurface offers a flexible and simple approach to dynamic holography.
• This method overcomes the limitations of complex active components in traditional reconfigurable metasurfaces.
• The technology shows significant potential for advancements in optical imaging and data processing applications.