Optical associative memory based on a neural net model has been implemented for 12x9 neurons, by using a holographic interconnection method. To achieve accurate positioning of each hologram element in the hologram array, a stepping motor-driven positioner for holographic recording is fabricated. Also a camera thresholding circuit has been made for thresholding operation needed in the neural net model.
The memory system stores five vectors stably. It is shown that the memory system could correct approximately up to 20 bits of errors regardless of the position of error bits and up to 40 bits when the correlation between stored vectors is small enough.
