Cascaded-grating holography for artificial neural networks

A holographic technique for implementing optical neural networks is described in which each connection weight is distributed among a series of cascaded angularly and spatially multiplexed photorefractive gratings. This greatly reduces cross talk caused by the conical Bragg degeneracy associated with a single grating, reduces distortions caused by beam coupling, and improves the utilization of pixels in the input spatial light modulator. Experimental examples of high-quality holographic storage and optical implementations of the perceptron and backpropagation neural networks are also discussed.