On Supporting Associative Access and Processing over Dynamic Knowledge Bases

Abstract

Dynamic knowledge bases are a fact of life in many artificial intelligence applications. Using current techniques, however, it is not always possible to provide the desired level of associative access to them whilst meeting real-time, or even near-real-time, performance criteria. This paper argues the case for a hardware associative storage system that uses symbolic pattern matching as its access mechanism. A working prototype of such a system, designed as a co-processor for a workstation host, is then described. The coprocessor is based on an array of custom designed VLSI 'smart memory' chips. These combine storage and search/processing logic on the same die. Parallelism is exploited both on chip and between chips to yield a high system performance. The paper concludes with some examples of how this hardware can be used to support real applications.