Abstract
Superconducting circuits that operate by propagation of small voltage or current pulses, corresponding to propagation of single flux or charge quantum, are naturally suited for implementing spiking neuron circuits. Quantum phase-slip junctions (QPSJs) are 1-D superconducting nanowires that have been identified as exact duals to Josephson junctions, based on charge-flux duality in Maxwell’s equations. In this paper, a superconducting quantized-charge circuit element, formed using quantum phase-slip junctions, is investigated for use in high-speed, low-energy superconducting spiking neuron circuits. By means of a SPICE model developed for QPSJs, operation of this superconducting circuit to produce and transport quantized charge pulses, in the form of current pulses, is demonstrated. The resulting quantized-charge-based operation emulates spiking neuron circuits for brain-inspired neuromorphic applications. Additionally, to further demonstrate the operation of QPSJ-based neuron circuits, a QPSJ-based integrate and fire neuron circuit is introduced, along with simulation results using WRSPICE. Estimates for operating speed and power dissipation are provided and compared to Josephson junction and CMOS-based spiking neuron circuits. Current challenges are also briefly mentioned.
Highlights
Solid-state neuromorphic hardware has recently experienced remarkable improvement due to advanced, highlyscaled CMOS technology and emerging memory devices.1–8 These non-von Neumann architectures show great promise for solving complex problems, such as object recognition and decision-making, faster and more efficiently than conventional von Neumann architectures.9,10 a bottleneck arises when scaling CMOS technology to smaller nodes, as its energy efficiency is several orders of magnitude lower than the human brain
With regard to practical implementation of Quantum phase-slip junctions (QPSJs)-based circuits, existing challenges include determination of optimal materials that exhibit more ideal QPS behaviors, as well as charge offset and fluctuations, which have been previously observed as challenges with single-electron devices
We introduce a superconducting island circuit comprised of two QPSJs, to emulate a neuron circuit with tunable threshold, firing frequency, and firing mode
Summary
Solid-state neuromorphic hardware has recently experienced remarkable improvement due to advanced, highlyscaled CMOS technology and emerging memory devices. These non-von Neumann architectures show great promise for solving complex problems, such as object recognition and decision-making, faster and more efficiently than conventional von Neumann architectures. a bottleneck arises when scaling CMOS technology to smaller nodes, as its energy efficiency (energy per operation) is several orders of magnitude lower than the human brain. With regard to practical implementation of QPSJ-based circuits, existing challenges include determination of optimal materials that exhibit more ideal QPS behaviors, as well as charge offset and fluctuations, which have been previously observed as challenges with single-electron devices.. With regard to practical implementation of QPSJ-based circuits, existing challenges include determination of optimal materials that exhibit more ideal QPS behaviors, as well as charge offset and fluctuations, which have been previously observed as challenges with single-electron devices.36,37 These challenges can be compared to the stray magnetic flux challenges associated with more conventional JJ-based superconducting electronics, where these challenges have been successfully overcome through suitable designs and operational techniques.. We explore several ideas for using QPSJs in superconducting spiking neuron circuits and demonstrate that this technology can emulate certain neuron functions with operation comparable to the (very low) power dissipation and high-speed of JJ-based superconducting neuromorphic circuits. The simulation results in WRSPICE demonstrate that these circuits are able to mimic cortical neuron spiking behaviors and circuits can be designed to function like an integrate and fire neuron
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
