Tandem neural network-assisted inverse design of highly efficient diffractive slanted waveguide grating

Area of Science:

• Optics and Photonics
• Computer Science
• Materials Science

Background:

• Virtual reality (VR) near-to-eye displays increasingly utilize diffractive optical elements.
• Slanted waveguide gratings are crucial for efficient light coupling in these displays.
• Current design methods for gratings can be complex and time-consuming.

Purpose of the Study:

• To develop an automated and efficient method for designing slanted waveguide gratings.
• To leverage deep learning for inverse design of optical components.
• To optimize grating performance for multi-wavelengths and various incident angles.

Main Methods:

• Proposed a tandem neural network (TNN) combining a generative flow-based invertible neural network and a fully connected neural network.
• Employed deep learning-driven inverse design for grating optimization.
• Optimized coupling efficiencies for red, green, and blue beams across 0°-15° incident angles.

Main Results:

• Achieved peak transmittance near 100%, average transmittance of 92%, and illuminance uniformity of 98%.
• Successfully deduced grating structural parameters inversely within milliseconds to seconds.
• Demonstrated the TNN's capability for multi-wavelength optimization.

Conclusions:

• The proposed TNN offers an efficient and rapid inverse design solution for slanted waveguide gratings.
• This deep learning approach can significantly expedite the development of advanced VR near-eye display components.
• The methodology provides a paradigm for designing diverse types of waveguide gratings.