This project concerns the development and practical apparatus implementation of modern methods on information processing, suitable for industrial applications.
Content-Addressed Memory (CAM) is one of the most promising applications of completely parallel Neuron Networks (NN) architecture. It is especially adept at fast pattern recognition and related problems, where many hypotheses are pursued in parallel.
Conventional NN attempt to achieve good performance via dense interconnections of simple neurons. (The latter being a threshold element over the weighted sum of its inputs). The basic limitation of the present models of CAM is their low capacity, which is roughly given by the number of interconnections per neuron.
The only way out to reduce the number of interconnections with the increase of the network's storage capacity is the use of more complex neurons, which take into account various cross-correlations of its inputs. The capacity of such networks is known to be much higher, than that of the conventional ones.
We propose to use the waveguide holographic technique to implement high-order cross correlations. The multisegment diode lasers may be used both as light sources and as nonlinear elements. Such a technology permits the integrated implementation of NN, suitable for mass production.
The final goal of the present program is the development of CAM with high storage capacity and diluted interconnections. This willopen jbbe way for silicon CAM implementations, which is now strongly restricted by the limited fan-out of electronic gates.
In LPI there exists a laboratory of ultra-fast optoelectronics, which have an experience in picosecond laser diode technology, integrated optics and holography. To reach the project's goals the experienced specialists, familiar with neurocomputing are needed. VNIITF has a number of weapons scientists, who match this requirement. This project provides them an opportunity to redirect their talents to peacefull activities.