Neuromorphic Computing with Josephson Junction Neurons

Josephson Junctions are superconducting circuit components whose behavior can be described by a second-order, non-linear differential equation. This makes them an ideal tool for exploring and modeling complicated systems, such as neurons. This abstract will give background for studying fluxon dynamics (the behavior of a quantized amount of magnetic flux) in Josephson Junctions arrays and the possibility of demonstrating learning in a neural circuit. When cooled below TC, current loops in the array can cause fluxons to become trapped between junctions in the array. At a certain current, ISW, or thermal energy level, a fluxon will begin to move around the array, and a voltage is detectable. ISW, however, can vary significantly. It is strongly suspected that this variation is caused by production uncertainty in the size of the junctions, akin to a particle moving over hills of different sizes. Macroscopic quantum tunneling is also a suspect for these variations. The demonstration of learning involves splitting artificial neuron spikes down two different "axons" compose of more Josephson Junctions and observing the difference in arrival time of these spikes to a "learning gate" composed of an inductor and a SQUID. What we are able to observe here is called spike-timing dependent plasticity (if the spikes are close, the coupling strength is increased and vice versa). In simulations, unsupervised learning and pattern recognition have been successfully demonstrated.