Back-propagation optimization and multi-valued artificial neural networks for highly vivid structural color filter...

Area of Science:

• Optics and Photonics
• Materials Science
• Artificial Intelligence

Background:

• Traditional metasurface design methods face limitations in efficiency and performance.
• Existing approaches can suffer from extrapolation issues and local minima during optimization.
• The need for high-performance color filters with extended color gamuts is growing.

Purpose of the Study:

• To introduce a novel, data-driven deep learning technique for designing color filter metasurfaces.
• To achieve highly efficient metasurface designs that surpass existing configurations.
• To enable the creation of color filters with vivid colors beyond the standard sRGB gamut.

Main Methods:

• Utilized a data-driven approach employing deep learning for inverse design.
• Combined Multi-Valued Artificial Neural Networks (MV-ANNs) with back-propagation optimization.
• Trained and evaluated designs using a dataset of 585 distinct metasurface geometries.

Main Results:

• Identified highly efficient metasurface color filter designs outperforming the entire dataset.
• Overcame limitations of previous methods, avoiding poor performance from extrapolation and local minima.
• Successfully generated reliable and efficient metasurface configurations producing exceptionally vivid colors.

Conclusions:

• The developed deep learning technique offers a powerful and reliable method for metasurface design.
• This approach enables the creation of color filters with superior color reproduction capabilities.
• The technique is extensible for designing various pixellated metasurface configurations with diverse functionalities.