Adjoint method-based Fourier neural operator surrogate solver for wavefront shaping in tunable metasurfaces

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Materials Science

Background:

• Traditional wavefront optimization methods like Gerchberg-Saxton and adjoint optimization are computationally intensive due to iterative simulations.
• Metasurfaces offer precise control over light but require efficient optimization techniques for tunable elements.

Purpose of the Study:

• To introduce a Fourier neural operator (FNO)-based surrogate solver for efficient wavefront optimization in tunable metasurface controls.
• To overcome the computational burden of existing iterative simulation-based methods.

Main Methods:

• Developed an FNO-based surrogate solver to estimate gradients for tunable meta-atoms without direct numerical simulations.
• Validated the solver's accuracy by comparing its performance against the adjoint optimization method.

Main Results:

• The FNO surrogate solver achieved a residual of 0.02 compared to the normalized figure of merit from the adjoint method.
• Inference time was 887.5 times faster than conventional simulation-based optimization methods.
• Demonstrated highly accurate gradient estimations with respect to meta-atom changes.

Conclusions:

• The FNO-based surrogate solver offers a computationally efficient and fast alternative for wavefront shaping.
• This advancement enables ultra-fast optical wavefront shaping, reconfigurable intelligent metasurfaces, and improved biomedical imaging.