Photonic-dispersion neural networks for inverse scattering problems

Area of Science:

• Optics and Photonics
• Computational Physics
• Materials Science

Background:

• Inverse scattering problems are crucial for inferring material properties from optical responses.
• Challenges include large parameter spaces and experimental noise, hindering accurate reconstruction.
• Reconstructing grating profiles is a key inverse scattering problem with significant practical applications.

Purpose of the Study:

• To propose a robust strategy for reconstructing grating profiles using neural networks and photonic dispersions.
• To overcome challenges posed by large parameter ranges and experimental noise in inverse scattering.
• To achieve high-precision, rapid reconstruction of grating profiles.

Main Methods:

• Utilized forward-mapping and inverse-mapping neural networks for grating profile reconstruction.
• Developed a parameters-to-point forward-mapping neural network for generating analytical photonic dispersions.
• Implemented a Fourier-optics-based angle-resolved imaging spectroscopy for single-shot dispersion measurement.

Main Results:

• Successfully reconstructed grating profiles with geometric features spanning hundreds of nanometers with nanometric sensitivity.
• Achieved reconstruction in several seconds, demonstrating high efficiency.
• Forward-mapping algorithm showed excellent linear correlation (R² > 0.982) with atomic force microscopy measurements, even with experimental noise.

Conclusions:

• The proposed strategy offers a robust and efficient solution for inverse scattering problems, specifically grating profile reconstruction.
• Neural networks combined with photonic dispersions provide high accuracy and sensitivity.
• The experimental setup enables rapid and informative data acquisition for real-time analysis.