Holographic-disk-based optical neural network

Abstract

Holographic photothermoplastic disks with two-dimensional Fourier hologram registration obtained by infrared laser heating of the recording layer is described. These disks have high information recording density (~10(5) bits/mm(2)), high quality of the registered Fourier holograms (eta = 5-7%, contrast ratio 70:1), the possibility of information rerecording, and invariance to shift in thin phase holograms. A holographic-disk-based scheme of an optical neural network with outer-product implementation is suggested. The main advantages of this approach are discussed, and the results of such a neural network modeling are described.