Abstract
We evaluate Inner-product and outer-product algorithms for associative memory. We show that the page-serial, bit-parallel inner-product method has lower storage requirements and system complexity than do other algorithms, and it achieves minimum error recall. Storage capacity can be traded off for fault tolerance. The high capacity (1 Gb) and high throughputs (1.5 Gb/s) obtainable with the motionless parallel-readout optical disk make it suitable to implement an associative memory capable of fast retrieval (25 ms). The associative memory system consists of a detector array, an Si/PLZT exclusive-OR gate-array, and a variable-threshold detector (VTD) with fast local-decision circuitry. The intensity at the VTD is a measure of the mismatch between the query image and the stored image that is being read. Two modes of operation are possible. If the threshold is preset at the VTD, all disk images close to the query are retrieved in one rotation. Alternatively, all Hamming distances are calculated during the first rotation of the disk, and the closest match is retrieved during the next rotation.
© 1990 Optical Society of America
PDF ArticleMore Like This
P. Marchand, A. Krishnamoorthy, P. Ambs, S. Esener, and S. H. Lee
TuK2 OSA Annual Meeting (FIO) 1990
Demetri Psaltis, Alan A. Yamamura, Mark A. Neifeld, and Seiji Kobayashi
ME3 Optical Computing (IP) 1989
Mark A. Neifeld and Demetri Psaltis
WA4 Optical Computing (IP) 1991