Diffractive Optical Networks
All-optical information processing with deep learning-designed diffractive surfaces
Diffractive optical networks are deep learning-designed structured surfaces that process information as light passes through them — performing computation at the speed of light, with no external power beyond the illumination itself.
My work in this area includes universal polarization transformations using diffractive polarization transformers (Advanced Materials, 2023), nonlinear encoding in diffractive information processing using linear optical materials (Light: Science & Applications, 2024), massively parallel universal approximation of nonlinear functions (eLight, 2025), and unidirectional focusing of light using structured diffractive surfaces (Advanced Optical Materials, 2025).