An active dendritic tree can mitigate fan-in limitations in superconducting neurons

Abstract

Superconducting electronic circuits have much to offer with regard to neuromorphic hardware. Superconducting quantum interference devices (SQUIDs) can serve as an active element to perform the thresholding operation of a neuron's soma. However, a SQUID has a response function that is periodic in the applied signal. We show theoretically that if one restricts the total input to a SQUID to maintain a monotonically increasing response, a large fraction of synapses must be active to drive a neuron to threshold. We then demonstrate that an active dendritic tree (also based on SQUIDs) can significantly reduce the fraction of synapses that must be active to drive the neuron to threshold. In this context, the inclusion of a dendritic tree provides dual benefits of enhancing computational abilities of each neuron and allowing the neuron to spike with sparse input activity.

Reconfigurable Neuromorphic Computing: Materials, Devices, and Integration

“…Digital superconducting systems, as a promising post‐CMOS concept, employing cryoelectronic device technologies, have shown significant improvements in neuromorphic computing for both energy‐efficient and speed‐up purposes. [ 201–207 ]…”

“…Digital superconducting systems, as a promising post-CMOS concept, employing cryoelectronic device technologies, have shown significant improvements in neuromorphic computing for both energy-efficient and speed-up purposes. [201][202][203][204][205][206][207] Josephson junctions (JJs) [59,61,62,75,[208][209][210] and superconducting nanowires [211][212][213][214] are the two principal forces that directly access neuromorphic computing at the device level, with particular implementations often exactly dual with each other. [77] For instance, the ion channel dynamics of LIF neurons can be efficiently mimicked by two cascading JJs, and the emulation of neuronic relaxation oscillation can be realized by a nanowire resistor incorporation as well.…”

SuperMind: a survey of the potential of superconducting electronics for neuromorphic computing

“…With this form of optoelectronic neuron, the maximum neuronal firing rate must be reduced due to the SPD recovery time, while fast JJ processing is leveraged for computations performed within each neuron. One proposed architecture uses dendritic trees with dendrites very similar to typical JJ neurons [33,36]. Sophisticated neurons thus envisioned are closely akin to the elaborate neural processors comprising human cortical neurons [37,38], wherein dendritic computations appreciably augment neuronal functionality [39].…”

Superconducting Bio-Inspired Au-Nanowire-Based Neurons

“…The shape of this spike can be quite close to the one produced in neurophysiological processes [ 22 ]. It was shown that an artificial neuron can be implemented using only two Josephson junctions [ 22 , 23 ] (see illustration in Figure 1 ). This is an order of magnitude less than the number of transistors in CMOS counterparts [ 16 ].…”

“…While the superconducting neural networks of various kinds are rapidly developed currently [ 22 , 23 , 24 , 28 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 ], their complexity is severely limited by the low integration density of superconducting circuits [ 41 , 42 ]. One of the main reasons for this is a comparatively large area (an order of a micron to few tenths of a micron squared) of commonly used superconductor-insulator-superconductor tunnel Josephson junction [ 32 ].…”