Design and simulation of digital optical computing systems for artificial intelligence.

Abstract

Rule-based systems (RBSs) are one of the problem solving methodologies in artificial intelligence. Although RBSs have a vast potential in many application areas the slow execution speed of current RBSs has prohibited them from the full exploitation of their vast potential. In this dissertation, to improve the speed of RBSs, we explore the use of optics for the fast and parallel RBS architectures. First, we propose an electro-optical rule-based system (EORBS). Using two-dimensional knowledge representation and a monotonic reasoning scheme, EORBS provides highly efficient implementation of the basic operations needed in rule-based systems, namely, matching, selection, and rule firing. The execution speed of the proposed system is theoretically estimated and is shown to be two orders of magnitude faster than the current electronic systems. Although EORBS shows the best performance in execution speed compared to other RBSs, the monotonic reasoning scheme restricts the application domains of EORBS. In order to overcome this limitation on the application domain in EORBS, a general purpose RBS, called an Optical Content-Addressable Parallel Processor for Expert Systems (OCAPP-ES) is proposed. Using a general knowledge representation scheme and a parallel conflict resolution scheme, OCAPP-ES executes the three basic RBS operations on general knowledge (including variables, symbols, and numbers) in a highly parallel fashion. The performance of OCAPP-ES is theoretically estimated and is shown to be an order of magnitude slower than that of EORBS. However, the performance of OCAPP-ES is still an order of magnitude faster than any other RBS. Furthermore, OCAPP-ES is designed to support the general knowledge representation scheme so that it can be a high speed general purpose RBS. To verify the proposed architectures, we developed a modeling and simulation methodology for digital optical computing systems. The methodology predicts maximum performance of a given optical computing architecture and evaluates its feasibility. As an application example, we apply this methodology to evaluate the feasibility and performance of OCAPP which is the optical match unit of OCAPP-ES. The proposed methodology is intended to reduce optical computing systems' design time as well as the design risk associated with building a prototype system.