Comparison of an efficient direct storage BAM model with traditional neural bidirectional memories

This paper proposes an efficient and improved model of a direct storage bidirectional memory, DBAM, which directly stores the X and Y associated sets of M bipolar binary vectors, requires O(N) or about 15% of interconnections of weight strength plusmn1, and is computationally very efficient as compared to other outer-product type BAM models that require O(N²) complex interconnections with weight strength ranging between plusmnM. It is simple, robust in structure, VLSI realizable, modular and expandable, and the addition or deletion of a pair of vectors does not require changes in the strength of interconnections of the entire memory matrix. Retrieval constraints and orthogonality issues and restrictions on the length, in bits, and number of vectors to be stored are discussed. The analysis of signal to noise ratio, storage capacity, and performance of the proposed model has been carried out. Simulation results show that it has log_eN time´s higher storage capacity, superior performance, faster convergence and retrieval time.