Multiple fiber coupler associative memory with bit significance retrieval

Summary form only given, as follows. A multiple-optical-fiber-coupler associative memory based on the Hopfield model is proposed and demonstrated experimentally. It realizes 25/sup 2/ global interconnections. The coupling ratio from fiber to fiber represents a synaptic weight of the connection between units. A remarkable feature of the optical-fiber neural networks is that the global connection between two-dimensional units can be easily achieved, resulting in instantaneous and automatic multiplication of the weight matrix and input vector. By introducing bit significance whereby the input vector is weighted bit by bit, the retrieval capability of the stored vector is much improved even in the presence of the deviation of the fiber coupling ratios from their ideal values in the multiple-fiber network.<>