A neural inspired associative memory

Associative memories can be modeled theoretically from neural networks using the feature extractor and a MAXNET. However, from the practical standpoint, the feature extractor has the inconvenience of needing complex operations and/or analog weight values, while the MAXNET is difficult to implement due to the high interconnectivity and analog weight values. The proposed solution is to use an encoding scheme which performs the Hamming distance calculation on encoded vectors, equivalent to a real-valued distance calculation. A memory compression technique is then applied to reduce the number of bits. For the MAXNET, synchronous digital circuit techniques are used. The resulting associative memory processors are simple, modular, and not too costly in terms of logic. The proposed content addressable memory matches data according to that which produces a minimum distance between the searched key and each of the stored record´s fields. Each field´s relevance in the search operation is controlled using a mask