High-speed low-power neuromorphic systems based on magnetic Josephson junctions
Michael L. Schneider, Christine A. Donnelly, Stephen E. Russek
Josephson junctions and single flux quantum (SFQ) circuits form a natural neuromorphic technology with SFQ pulses and superconducting transmission lines simulating action potentials and axons, respectively. Josephson junctions consist of superconducting electrodes with a nanoscale barrier that modulates the coupling of the complex superconducting order parameter across the junction. When the order parameter undergoes a 2π phase jump, the junction emits a voltage pulse with an integrated amplitude of a flux quantum ϕ_0=h/(2e^2 )=2.068 x 〖10〗^(-15) Vs. The coupling across a junction can be controlled and modulated by incorporating nanoscale magnetic structure in the barrier. The magnetic state of embedded nanoclusters can be changed by applying small currents or applied field pulses, thereby enabling both unsupervised and supervised learning. The advantage of this magnetic superconducting technology is that it combines natural spiking behavior and plasticity in a single nanoscale device and is orders of magnitude faster and lower energy than room temperature technologies. Maximum operating frequencies are above 100 GHz, while spiking and training energies are ~10-21 J, 10-18 J, respectively. This technology can operate close to the thermal limit, which at 4 K, is considerably lower than in a human brain that operates at 310 K. The transition from deterministic to stochastic behavior can be studied with small temperature modifications. Here we present a tutorial on the spiking behavior of Josephson junctions, the use of nanoscale magnetic structure to modulate the coupling across the junction, the design and operation of magnetic Josephson junctions, device models and simulation of magnetic Josephson junction neuromorphic circuits, and potential neuromorphic architectures based on hybrid superconducting/ magnetic technology.