Design of an optical content-addressable parallel processor for expert systems

Abstract

The slow execution speed of current rule-based systems (RBS's) has restricted their application areas. To improve the speed of RBS's, researchers have proposed various electronic multiprocessor systems as well as optical systems. However, the electronic systems still suffer in performance from the large amount of required time-consuming pattern-matching and comparison operations at the core of RBS's. And optical systems do not fully exploit the available parallelism in RBS's. We propose an optical content-addressable parallel processor for expert systems. The processor executes the three basic RBS operations, match, select, and act, in a highly parallel fashion. Additionally, it extracts and exploits all possible parallelism in a RBS. Distinctive features of the proposed system include the following: (1) two-dimensional representation of data (knowledge) and control information to exploit the parallelism of optics in the three RBS units; (2) capability of processing general-domain knowledge expressed in terms of variables, numbers, symbols, and comparison operators such as greater than and less than; (3) the parallel optical match unit, which performs the two-dimensional optical pattern matching and comparison operations; (4) a novel conflict-resolution algorithm to resolve conflicts in a single step within the optical select unit. The three units and the general-knowledge representation scheme are designed to make the optical content-addressable parallel processor for expert systems suitable for any high-speed general-purpose RBS.