Abstract

The photorefractive effect is a nonlinear optical phenomenon exhibited in a variety of electrooptic crystal whereby index gratings can be written holographically in real time and also be stored for extended periods of time. Because of this unique capability, photorefractive crystals have been used to implement the adaptive interconnections required for neural network systems. In such optical realizations of neural networks, complex interconnection patterns are stored in the form of volume holographic gratings in a photorefractive medium which can be accessed in parallel to achieve large data processing rates. Moreover, the dynamic nature of photorefractive holograms allow the implementation of learning algorithms whereby the contents of the stored interconnections are incrementally modified to optimize certain performance criteria.

© 1991 Optical Society of America