Neural network synapse device using single-electron tunnel junctions

We propose a novel single-electron synapse device using parallel-coupled ultra-small tunnel junction arrays. The switching function and the ohmic characteristic of the coupled arrays are exploited and their robustness to the background-charge fluctuation effect is examined. Based on the synaptic function of the coupled arrays, we propose two types of the Hopfield neural networks which can operate with minimal errors: one directly corresponds to a CMOS neural network architecture and operates with binary-weight, analog-signal. The other one is an ordinary Hopfield network having binary-signal, analog-weight.