A multilayered superconducting neural network implementation

We present the results of numerical simulations of a novel neural networking implementation utilizing multilayered Josephson junction (or series array) based synaptic circuits with local memory. These synaptic circuits utilize single flux quanta for both neural information and synaptic weight programming, and we present a simple circuit that can implement Hebbian learning at a completely local level, with global control over the rates of both learning and forgetting in synapses.