Noise issues in holographic photorefractive interconnections: application to neural networks

We investigated the influence of the imperfections of some optical components in a reconfigurable holographic interconnection system. These interconnections are recorded in a photorefractive crystal by interfering two light beams modulated by two spatial light modulators, We experimentally show that such set-ups are limited to small interconnect capacities by the finite contrast of the spatial light modulators. We proposed two techniques to improve the performances of such interconnection systems. We used this set-up to implement the first layer of a two layer neural network. With this system, we successfully recognised 101 chinese ideograms. (C) 1997 Elsevier Science B.V.