Abstract

Holographic memories offer the advantages of high storage density, distributed storage, and fast parallel access. These characteristics can be important for parallel opto-electronic computers such as neural networks. The memory can be used to hold a training set which is repetitively displayed during learning. In addition, if partitioning is used to solve problems requiring more neurons than the actual number of processors, the memory can also be used to hold connection weight and threshold information. In this paper, we present results from a photorefractive memory using incremental recording scheduling and binary orthogonal phase codes image addressing. We show how this approach allows modification of the content and diffraction efficiency of the stored images, and how multiple images can be combined by complex amplitude addition and subtraction during reconstruction.

© 1991 Optical Society of America