Abstract

The use of associative memories—storage devices that allow data retrieval based on contents—has often been suggested to speed up the performance of many applications. Until recently, using such content-addressable memories (CAMs) was unfeasible due to their high hardware cost. However, the advent of VLSI has made the class of fully-parallel associative memory cost-effective for implementation. This paper briefly overviews design of several fully parallel associative memories proposed in the literature, concentrating on the design of fully-parallel θ-search CAMs.Existing market realities require that product development be fast and predictable. As a result, design flexibility and automation are becoming increasingly important design features. Using the various CAM designs reviewed, the paper collects the features of these designs into a general, modular CAM organization and describes its major components. The modular CAM organization can be used to design application specific CAMs of varying degrees of functionality. Design and space complexity of a sample associative memory suitable for relational database operations is studied. Finally, the application of genetic algorithms as a means to developing automated design tools for fabrication of modular VLSI design chips is discussed.Given a library of CAM modules, the desired functionality and a set of speed and area constraints, this optimization technique produces a suitable CAM design. The proposed technique has been implemented and its performance measure is briefly addressed.