Photorefractive crystal helps classify data

Using optically interconnected electronics, researchers in France have constructed a type of self-organizing neural network called a Kohonen map. The network is constructed around a holographic refractive crystal. The team from the Laboratoire Charles Fabry (a joint venture of the Institute D'Optique, Orsay, where it is housed, and the Centre National de la Research Scientifique) showed that the system could correctly classify 100 items, each with more than 100 features. Though the system suffers from problems often associated with sensitive photorefractive materials (such as hologram erasure, low diffraction efficiency, and material nonuniformity), it demonstrates that optoelectronic approaches can produce interesting and high-capacity results. Even in this demonstrator, more than 106 weights are updated per second.