ELECTRO-HOLOGRAPHIC ARTIFICIAL NEURAL NETWORKS

A new approach for implementing artificial neural networks (ANNs), bas ed on a novel concept in volume holographic storage - electroholograph y (EH) - is described. EH is based on the use of paraelectric photoref ractive crystals in which the photorefractive mechanism is created in the paraelectric phase. Thus the quadratic electrooptic effect is used to transform the information carrying space charge fields into birefr ingence gratings, and therefore the diffraction efficiency is governed by an external electric field. This principle enables the constructio n of a page oriented architecture in which each neuron is implemented as a local electronic circuit, governing the diffraction efficiency of a small volume hologram representing the axon of the neuron. The EH a pproach is expected to enable the construction of networks with 10(5)- 10(6) neurons, each with 10(3)-10(4) synapses.