High-Tc Josephson Neurons And Synapses: Towards Ultrafast And Energy Efficient Superconducting Neuromorphic Computing (JOSEPHINE)

Basic data for this project

Description

We aim at realizing a novel class of high-temperature Josephson junctions (JJs) that behave as artificial neurons and synapses. These JJs will enable a new neuromorphic computing paradigm, in which neural networks are much faster, more energy efficient and compact than with non-superconducting approaches, and possess novel capabilities (combined sensitivity to light, magnetic and electric fields). Via these rupture ingredients, JOSEPHINE will dramatically enhance the impact of neuromorphics on its broad range of projected applications: from artificial intelligence (where it would allow supercomputer-level processors at a fraction of the environmental cost) to the control of autonomous vehicles, the Internet of Things, and novel medical applications. That constitutes the long-term vision for the science we propose. To reach that goal, we will use different strategies to realize high-Tc Josephson junctions whose weak-links are active and can be changed "in operando" by external stimuli. Those strategies include "weak links" modified by a nanoscale redox reaction, by the motion of domain walls in a ferromagnet, or by locally doping a graphene or a 2D semiconductor. Once realized, these JJs will be implemented and tested in neural networks to demonstrate their performance and their transformative effect on neuromorphics. The proposed strategy exploits recent breakthrough results of the partners (physical effects that will be implemented) and synergizes their complementary expertise via a multidisciplinary approach that marries traditionally distant disciplines: neural network engineering, superconducting electronics, and various facets of solid-state physics (superconductivity, magnetism, Dirac materials, and electrochemistry).