A highstorage capacity content-addressable memory and its learning algorithm

Abstrud-Hopfield’s neural networks show retrieval and speed capabili-ties that make them good candidates for content-addressable memories (CAM’s) in problems such as pattern recognition and optimization. This paper presents a new implementation of a VLSI fully interconnected neural network with only two binary memory points per synapse (the connection weights are restricted to three different values: + 1,O and- 1). The small area of single synaptic cells (about lo4 pm’) allows the implementation of neural networks with more thut 500 neurons. Because of the poor storage capability of Hebb’s learning rule, especially in VLSI neural networks where the range of the synapse weights is limited by the number of memory points contained in each connection, a new algorithm is proposed for programming a Hopfield neural network as a high-storage capacity CAM. The results of the VLSI circuit programmed with this new algorithm are very promising for pattern recognition applications. I