Josephson junction neurons for neuromorphic computing
Abstract
Compared to silicon circuits, superconducting circuits made from Josephson junctions operate with vastly less power and at higher speeds. Although much work has been done on conventional digital processors using Josephson... [ view full abstract ]
Compared to silicon circuits, superconducting circuits made from Josephson junctions operate with vastly less power and at higher speeds. Although much work has been done on conventional digital processors using Josephson junctions, comparatively little has been done with neuromorphic processors. Here we describe a Josephson junction approach to neuromorphic circuits. We configure an analog neuron model using two junctions so that one activates and the other resets the firing. The resulting circuit provides biologically realistic action potentials, threshold input response and post-firing refractory period. Our proof-of-principle experiment couples two such mutually coupled neurons and demonstrates many components of the neural toolbox: spiking neurons with delay- and amplitude-adjustable synapses, and real-time state detection. We measure their relative firing which is seen to be asynchronous or to synchronize in one of two states: symmetric (in-phase) or anti-symmetric (anti-phase). The state can be toggled back and forth by changing the strength and delay of the synapse in a phase-flip bifurcation. The synchronization state is measured in real time and the bifurcation diagrams match circuit simulations. We propose a second circuit to demonstrate Hebbian learning and we speculate about future processors.
Authors
-
Kenneth Segall
(Colgate University)
-
Daniel Schult
(Colgate University)
Topic Areas
Topics: Neuromorphic, or “brain inspired”, computing , Topics: Nonlinear Dynamical Systems and Edge of Chaos , Topics: Superconducting or cryogenic computing
Session
PS-1 » Poster Session (19:00 - Monday, 17th October, Ballroom Foyer)
Paper
RC_Segall.pdf
Presentation Files
The presenter has not uploaded any presentation files.