Hybrid RSFQ-QFP Superconducting Neuron
M. Ameen Jardine, Coenrad J. Fourie
Abstract
Neuromorphic computing is the implementation of neural networks in hardware, in a manner mimicking the human brain. We present a superconducting synaptic circuit for use in such a system. The proposed system makes use of quantum flux parametron architecture to attain both excitatory and inhibitory weightings, and rapid single flux quantum architecture to adjust the magnitude of the synapse weighting. We show that the proposed synaptic circuit is capable of being tuned to eight different weightings, and these adjustments can all be done on the fly. We also demonstrate the synapse's capabilities in full neuron architecture using simple perceptron logic. The result is successfully training the system to emulate simple Boolean logic gates, from scratch.