Deep neural network enabled active metasurface embedded design.

Area of Science:

• Photonics and optical engineering
• Materials science
• Artificial intelligence in scientific design

Background:

• Active metasurfaces offer tunable optical properties but their design is computationally intensive.
• Traditional design methods struggle with complex, objective-driven optimization for active photonic devices.

Purpose of the Study:

• To introduce a novel deep learning framework for the forward modeling and inverse design of photonic devices with active metasurfaces.
• To demonstrate a significant reduction in computational overhead for metasurface design.
• To facilitate accurate, objective-driven design of complex photonic functionalities.

Main Methods:

• Combining neural network-based metasurface design with scattering matrix optimization.
• Utilizing a phase-change material (Ge2Sb2Se4Te - GSST) integrated into a metasurface atop a silicon heater.
• Employing distributed Bragg reflectors (DBRs) for device fabrication.

Main Results:

• Successfully designed a continuously tunable bandpass filter in the mid-wave infrared spectrum.
• Achieved a narrow passband (∼10 nm), high quality factors (Q-factors ∼ 10^2), and substantial out-of-band rejection (optical density ≥ 3).
• Demonstrated simplification of computational overhead and accurate objective-driven design.

Conclusions:

• The proposed deep learning approach offers an efficient and accurate method for designing active photonic devices.
• This generalized methodology is applicable to a wide range of photonic devices incorporating active metasurfaces.
• The technique significantly advances the capabilities for creating tunable and reconfigurable optical components.