Machine learning-assisted design of polarization-controlled dynamically switchable full-color metasurfaces

Area of Science:

• Photonics and Metamaterials
• Optical Engineering
• Materials Science

Background:

• Dynamic color tuning is crucial for displays, steganography, and encryption.
• Existing color-switching methods often require external stimuli, complicating device structure and limiting applications.
• Need for stimuli-independent dynamic color tuning in advanced display technologies.

Purpose of the Study:

• To propose and demonstrate polarization-controlled hybrid metal-dielectric metasurfaces for dynamic color tuning.
• To realize full-color display and tunable structural colors by altering incident light polarization.
• To develop a neural network for efficient metasurface design and color prediction.

Main Methods:

• Fabrication of hybrid metal-dielectric metasurfaces.
• Utilizing the polarization angle of incident light to control and tune colors dynamically.
• Training a bidirectional neural network for color prediction and inverse design of geometric parameters.

Main Results:

• Achieved full-color display and dynamic color tuning solely by adjusting the polarization angle.
• Demonstrated high accuracy in color prediction (93.18%) and inverse parameter design (92.37%) using the neural network.
• The proposed method avoids material property changes and external stimuli for color switching.

Conclusions:

• Presents a simplified approach for dynamic structural color tuning using polarization-controlled metasurfaces.
• Accelerates the design process for full-color metasurfaces, reducing reliance on empirical design.
• Offers valuable insights for developing novel color filters and advancing photonics research.