Superconducting Elements for Artificial Neural Network

Reducing power consumption is a crucial task in the development of novel supercomputers with non-von Neumann architecture, based on Artificial Neural Networks (ANNs). The work presents design and characteristics of adiabatic superconducting neuron based on Magnetic Josephson Junction (MJJ), allow reduction in energy loss during the operation of that artificial neural network down to attojoule levels. Other important element of ANN - synapse is proposed based on layered superconductor-ferromagnet layered nanostructures. This allows for a significant reduction in energy dissipation during the passage of a single pulse. Theoretical and experimental analysis of electronic properties of Nb/Co multilayers with different thicknesses of two ferromagnetic layers Co and several stacking periods demonstrated that magnetization switching of the Co layers leads to modulations of superconductivity in the superlattice with corresponding tunable kinetic inductances (TKIs) of the structure which can serve as artificial synapse - tunable superconducting component of ANN.