All-optical information-processing capacity of diffractive surfaces

Abstract

Optical computing: the potential of layered diffractive surfaces Layers of materials that diffract light with variable spacing between them can be adjusted or “trained” to perform information-processing tasks using light alone. Diffraction is the alteration of the propagation of light waves by structural features of the materials they encounter. Aydogan Ozcan and colleagues at the University of California, Los Angeles, USA, performed an analysis of optical neural networks composed of spatially engineered diffractive surfaces. They explored the power of multilayered networks to perform optical processing tasks, including image recognition and classification. They also determined mathematical rules describing the performance limits of the networks in relation to the number of diffractive surfaces they contained. Their work is relevant to various diffractive surfaces, including metasurfaces patterned with features smaller than the wavelength of light, and plasmonic materials governed by the coherent behavior of surface electrons.