Grating waveguides by machine learning for augmented reality.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Computational Science

Background:

• Grating waveguides are crucial components in optical systems, including augmented reality.
• Traditional numerical simulation methods, like finite-element analysis, are computationally intensive and time-consuming.
• Optimizing grating waveguide structures requires exploring numerous design parameters.

Purpose of the Study:

• To propose a novel machine-learning-based method for the design of grating waveguides.
• To significantly reduce computation time compared to existing finite-element-based methods.
• To achieve optimal design of high-efficiency grating waveguide structures using artificial intelligence.

Main Methods:

• Exploited structural parameters including grating slanted angle, depth, duty cycle, coating ratio, and interlayer thickness.
• Utilized a multi-layer perceptron algorithm implemented in the Keras framework.
• Trained the model on a dataset of 3000-14,000 samples.

Main Results:

• Achieved training accuracy with a coefficient of determination > 99.9% and average absolute percentage error of 0.5%-2%.
• A hybrid structure grating demonstrated a diffraction efficiency of 94.21% and uniformity of 93.99%.
• The hybrid structure grating exhibited superior performance in tolerance analysis.

Conclusions:

• The proposed artificial intelligence waveguide method enables optimal design of high-efficiency grating waveguide structures.
• This AI-driven approach offers significant computational advantages over traditional methods.
• Provides theoretical guidance and technical reference for AI-based optical design.