Abstract

The rapid growth of optical disk storage techniques provides an advantage for optics in large capacity information storage and processing. Here we propose an optical disk-based neural network architecture for high speed and large capacity associative processing. The interpattern association (IPA) neural network model is used to generate the tristate interconnection weight matrices (IWMs). Each IWM is stored in two blocks (i.e., positive and negative matrices) on the disk, and read out in parallel by separate pulse laser beams. The two readout beams are encoded in perpendicular polarizations and then directed to a parallel optical matrix-vector processor, which consists of a lenslet array, an input SLM, and two photodetector arrays. Parallel buffers and a postprocessing circuit would be used to alleviate the electronic bottleneck to some extent.

© 1989 Optical Society of America