-
Self-training superconducting neuromorphic circuits using reinforcement learning rules
Authors:
M. L. Schneider,
E. M. Jué,
M. R. Pufall,
K. Segall,
C. W. Anderson
Abstract:
Reinforcement learning algorithms are used in a wide range of applications, from gaming and robotics to autonomous vehicles. In this paper we describe a set of reinforcement learning-based local weight update rules and their implementation in superconducting hardware. Using SPICE circuit simulations, we implement a small-scale neural network with a learning time of order one nanosecond. This netwo…
▽ More
Reinforcement learning algorithms are used in a wide range of applications, from gaming and robotics to autonomous vehicles. In this paper we describe a set of reinforcement learning-based local weight update rules and their implementation in superconducting hardware. Using SPICE circuit simulations, we implement a small-scale neural network with a learning time of order one nanosecond. This network can be trained to learn new functions simply by changing the target output for a given set of inputs, without the need for any external adjustments to the network. In this implementation the weights are adjusted based on the current state of the overall network response and locally stored information about the previous action. This removes the need to program explicit weight values in these networks, which is one of the primary challenges that analog hardware implementations of neural networks face. The adjustment of weights is based on a global reinforcement signal that obviates the need for circuitry to back-propagate errors.
△ Less
Submitted 29 April, 2024;
originally announced April 2024.
-
Fan-out and Fan-in properties of superconducting neuromorphic circuits
Authors:
M. L. Schneider,
K. Segall
Abstract:
Neuromorphic computing has the potential to further the success of software-based artificial neural networks (ANNs) by designing hardware from a different perspective. Current research in neuromorphic hardware targets dramatic improvements to ANN performance by increasing energy efficiency, speed of operation, and even seeks to extend the utility of ANNs by natively adding functionality such as sp…
▽ More
Neuromorphic computing has the potential to further the success of software-based artificial neural networks (ANNs) by designing hardware from a different perspective. Current research in neuromorphic hardware targets dramatic improvements to ANN performance by increasing energy efficiency, speed of operation, and even seeks to extend the utility of ANNs by natively adding functionality such as spiking operation. One promising neuromorphic hardware platform is based on superconductive electronics, which has the potential to incorporate all of these advantages at the device level in addition to offering the potential of near lossless communications both within the neuromorphic circuits as well as between disparate superconductive chips. Here we explore one of the fundamental brain-inspired architecture components, the fan-in and fan-out as realized in superconductive circuits based on Josephson junctions. From our calculations and WRSPICE simulations we find that the fan-out should be limited only by junction count and circuit size limitations, and we demonstrate results in simulation at a level of 1-to-10,000, similar to that of the human brain. We find that fan-in has more limitations, but a fan-in level on the order of a few 100-to-1 should be achievable based on current technology. We discuss our findings and the critical parameters that set the limits on fan-in and fan-out in the context of superconductive neuromorphic circuits.
△ Less
Submitted 14 August, 2020;
originally announced August 2020.
-
A superconducting nanowire spiking element for neural networks
Authors:
Emily Toomey,
Ken Segall,
Matteo Castellani,
Marco Colangelo,
Nancy Lynch,
Karl K. Berggren
Abstract:
As the limits of traditional von Neumann computing come into view, the brain's ability to communicate vast quantities of information using low-power spikes has become an increasing source of inspiration for alternative architectures. Key to the success of these largescale neural networks is a power-efficient spiking element that is scalable and easily interfaced with traditional control electronic…
▽ More
As the limits of traditional von Neumann computing come into view, the brain's ability to communicate vast quantities of information using low-power spikes has become an increasing source of inspiration for alternative architectures. Key to the success of these largescale neural networks is a power-efficient spiking element that is scalable and easily interfaced with traditional control electronics. In this work, we present a spiking element fabricated from superconducting nanowires that has pulse energies on the order of ~10 aJ. We demonstrate that the device reproduces essential characteristics of biological neurons, such as a refractory period and a firing threshold. Through simulations using experimentally measured device parameters, we show how nanowire-based networks may be used for inference in image recognition, and that the probabilistic nature of nanowire switching may be exploited for modeling biological processes and for applications that rely on stochasticity.
△ Less
Submitted 29 July, 2020;
originally announced July 2020.
-
A Power Efficient Artificial Neuron Using Superconducting Nanowires
Authors:
Emily Toomey,
Ken Segall,
Karl K. Berggren
Abstract:
With the rising societal demand for more information-processing capacity with lower power consumption, alternative architectures inspired by the parallelism and robustness of the human brain have recently emerged as possible solutions. In particular, spiking neural networks (SNNs) offer a bio-realistic approach, relying on pulses analogous to action potentials as units of information. While softwa…
▽ More
With the rising societal demand for more information-processing capacity with lower power consumption, alternative architectures inspired by the parallelism and robustness of the human brain have recently emerged as possible solutions. In particular, spiking neural networks (SNNs) offer a bio-realistic approach, relying on pulses analogous to action potentials as units of information. While software encoded networks provide flexibility and precision, they are often computationally expensive. As a result, hardware SNNs based on the spiking dynamics of a device or circuit represent an increasingly appealing direction. Here, we propose to use superconducting nanowires as a platform for the development of an artificial neuron. Building on an architecture first proposed for Josephson junctions, we rely on the intrinsic nonlinearity of two coupled nanowires to generate spiking behavior, and use electrothermal circuit simulations to demonstrate that the nanowire neuron reproduces multiple characteristics of biological neurons. Furthermore, by harnessing the nonlinearity of the superconducting nanowire's inductance, we develop a design for a variable inductive synapse capable of both excitatory and inhibitory control. We demonstrate that this synapse design supports direct fanout, a feature that has been difficult to achieve in other superconducting architectures, and that the nanowire neuron's nominal energy performance is competitive with that of current technologies.
△ Less
Submitted 29 June, 2019;
originally announced July 2019.
-
Synchronization dynamics on the picosecond timescale in coupled Josephson junction neurons
Authors:
Ken Segall,
Matthew LeGro,
Steven Kaplan,
Oleksiy Svitelskiy,
Shreeya Khadka,
Patrick Crotty,
Daniel Schult
Abstract:
Conventional digital computation is rapidly approaching physical limits for speed and energy dissipation. Here we fabricate and test a simple neuromorphic circuit that models neuronal somas, axons and synapses with superconducting Josephson junctions. The circuit models two mutually coupled excitatory neurons. In some regions of parameter space the neurons are desynchronized. In others, the Joseph…
▽ More
Conventional digital computation is rapidly approaching physical limits for speed and energy dissipation. Here we fabricate and test a simple neuromorphic circuit that models neuronal somas, axons and synapses with superconducting Josephson junctions. The circuit models two mutually coupled excitatory neurons. In some regions of parameter space the neurons are desynchronized. In others, the Josephson neurons synchronize in one of two states, in-phase or anti-phase. An experimental alteration of the delay and strength of the connecting synapses can toggle the system back and forth in a phase-flip bifurcation. Firing synchronization states are calculated >70,000 times faster than conventional digital approaches. With their speed and low energy dissipation (10-17 Joules/spike), this set of proof-of- concept experiments establishes Josephson junction neurons as a viable approach for improvements in neuronal computation as well as applications in neuromorphic computing.
△ Less
Submitted 8 February, 2017; v1 submitted 16 August, 2016;
originally announced August 2016.
-
Breather decay into a vortex/anti-vortex pair in a Josephson Ladder
Authors:
K. Segall,
P. Williams,
O. Svitelskiy,
D. Edwards,
N. Zhelev,
G. Brummer,
J. J. Mazo
Abstract:
We present experimental evidence for a new behavior which involves discrete breathers and vortices in a Josephson Ladder. Breathers can be visualized as the creation and subsequent annihilation of vortex/anti-vortex pairs. An externally applied magnetic field breaks the vortex/anti-vortex symmetry and causes the breather to split apart. The motion of the vortex or anti-vortex creates multi-site br…
▽ More
We present experimental evidence for a new behavior which involves discrete breathers and vortices in a Josephson Ladder. Breathers can be visualized as the creation and subsequent annihilation of vortex/anti-vortex pairs. An externally applied magnetic field breaks the vortex/anti-vortex symmetry and causes the breather to split apart. The motion of the vortex or anti-vortex creates multi-site breathers, which are always to one side or the other of the original breather depending on the sign of the applied field. This asymmetry in applied field is experimentally observed.
△ Less
Submitted 18 March, 2014;
originally announced March 2014.
-
Numerical simulation of thermal noise in Josephson circuits
Authors:
Kenneth Segall,
Dan Schult,
Ushnish Ray,
Toshiro Ohsumi
Abstract:
We present a method to numerically add thermal noise to the equations of motion for a circuit of Josephson junctions. A new noise term, which we call "linearly interpolated Gaussian noise," replaces the usual white noise process. It consists of random noise values spaced at a chosen time interval and linearly interpolated in-between. This method can be used with variable time step solvers, allowin…
▽ More
We present a method to numerically add thermal noise to the equations of motion for a circuit of Josephson junctions. A new noise term, which we call "linearly interpolated Gaussian noise," replaces the usual white noise process. It consists of random noise values spaced at a chosen time interval and linearly interpolated in-between. This method can be used with variable time step solvers, allowing more precise control over the error while ensuring that fast dynamics are not missed by the solver. We derive the spectral density of such a noise term and compare it to a white noise process. Then we demonstrate the technique by computing the switching dynamics of a circuit of two Josephson junctions and comparing the results to the traditional method.
△ Less
Submitted 18 October, 2016; v1 submitted 2 October, 2011;
originally announced October 2011.
-
Josephson junction simulation of neurons
Authors:
Patrick Crotty,
Daniel Schult,
Ken Segall
Abstract:
With the goal of understanding the intricate behavior and dynamics of collections of neurons, we present superconducting circuits containing Josephson junctions that model biologically realistic neurons. These "Josephson junction neurons" reproduce many characteristic behaviors of biological neurons such as action potentials, refractory periods, and firing thresholds. They can be coupled togethe…
▽ More
With the goal of understanding the intricate behavior and dynamics of collections of neurons, we present superconducting circuits containing Josephson junctions that model biologically realistic neurons. These "Josephson junction neurons" reproduce many characteristic behaviors of biological neurons such as action potentials, refractory periods, and firing thresholds. They can be coupled together in ways that mimic electrical and chemical synapses. Using existing fabrication technologies, large interconnected networks of Josephson junction neurons would operate fully in parallel. They would be orders of magnitude faster than both traditional computer simulations and biological neural networks. Josephson junction neurons provide a new tool for exploring long-term large-scale dynamics for networks of neurons.
△ Less
Submitted 23 February, 2010; v1 submitted 15 February, 2010;
originally announced February 2010.
-
Switching Current Measurements in Josephson Rings
Authors:
K. Segall,
A. Dioguardi,
N. Fernandes,
J. J. Mazo
Abstract:
We present switching current measurements in niobium-aluminum oxide-niobium underdamped Josephson rings. Underdamped Josephson rings display hysteretic current-voltage curves which can be characterized by their switching current, the value of applied current at which the junctions switch to the energy-gap voltage. The value of the switching current is strongly affected by temperature and by the…
▽ More
We present switching current measurements in niobium-aluminum oxide-niobium underdamped Josephson rings. Underdamped Josephson rings display hysteretic current-voltage curves which can be characterized by their switching current, the value of applied current at which the junctions switch to the energy-gap voltage. The value of the switching current is strongly affected by temperature and by the presence of fluxons in the system. We observe a very small voltage (0.2 nV) across the ring prior to switching, indicating a low-velocity fluxon diffusion state before the jump to a full running state. In analogy with previous work on single junctions, we analyze the switching current data with the process of thermal activation over a dissipation barrier, where the system switches from a low velocity state to a high velocity state. We find that our data agrees qualitatively with this description, further supporting the observation of fluxon diffusion.
△ Less
Submitted 18 July, 2008;
originally announced July 2008.
-
Thermal depinning of fluxons in discrete Josephson rings
Authors:
J. J. Mazo,
F. Naranjo,
K. Segall
Abstract:
We study the thermal depinning of single fluxons in rings made of Josephson junctions. Due to thermal fluctuations a fluxon can be excited from its energy minima and move through the array, causing a voltage across each junction. We find that for the initial depinning, the fluxon behaves as a single particle and follows a Kramers-type escape law. However, under some conditions this single partic…
▽ More
We study the thermal depinning of single fluxons in rings made of Josephson junctions. Due to thermal fluctuations a fluxon can be excited from its energy minima and move through the array, causing a voltage across each junction. We find that for the initial depinning, the fluxon behaves as a single particle and follows a Kramers-type escape law. However, under some conditions this single particle description breaks down. At low values of the discreteness parameter and low values of the dam**, the depinning rate is larger than the single particle result would suggest. In addition, for some values of the parameters the fluxon can undergo low-voltage diffusion before switching to the high-voltage whirling mode. This type of diffusion is similar to phase diffusion in a single junction, but occurs without frequency-dependent dam**. We study the switching to the whirling state as well.
△ Less
Submitted 30 June, 2008;
originally announced June 2008.
-
Multiple junction biasing of superconducting tunnel junction detectors
Authors:
K. Segall,
J. J. Mazo,
T. P. Orlando
Abstract:
We describe a new biasing scheme for single photon detectors based on superconducting tunnel junctions. It replaces a single detector junction with a circuit of three junctions and achieves biasing of a detector junction at subgap currents without the use of an external magnetic field. The biasing occurs through the nonlinear interaction of the three junctions, which we demonstrate through numer…
▽ More
We describe a new biasing scheme for single photon detectors based on superconducting tunnel junctions. It replaces a single detector junction with a circuit of three junctions and achieves biasing of a detector junction at subgap currents without the use of an external magnetic field. The biasing occurs through the nonlinear interaction of the three junctions, which we demonstrate through numerical simulation. This nonlinear state is numerically stable against external fluctuations and is compatible with high fidelity electrical readout of the photon-induced current. The elimination of the external magnetic field potentially increases the capability of these types of photon detectors and eases constraints involved in the fabrication of large detector arrays.
△ Less
Submitted 4 October, 2004;
originally announced October 2004.
-
Dynamics and Energy Distribution of Non-Equilibrium Quasiparticles in Superconducting Tunnel Junctions
Authors:
K. Segall,
C. Wilson,
L. Li,
L. Frunzio,
S. Friedrich,
M. C. Gaidis,
D. E. Prober
Abstract:
We present a full theoretical and experimental study of the dynamics and energy distribution of non-equilibrium quasiparticles in superconducting tunnel junctions (STJs). STJs are often used for single-photon spectrometers, where the numbers of quasiparticles excited by a photon provide a measure of the photon energy. The magnitude and fluctuations of the signal current in STJ detectors are in l…
▽ More
We present a full theoretical and experimental study of the dynamics and energy distribution of non-equilibrium quasiparticles in superconducting tunnel junctions (STJs). STJs are often used for single-photon spectrometers, where the numbers of quasiparticles excited by a photon provide a measure of the photon energy. The magnitude and fluctuations of the signal current in STJ detectors are in large part determined by the quasiparticle dynamics and energy distribution during the detection process. We use this as motivation to study the transport and energy distribution of non-equilibrium quasiparticles excited by x-ray photons in a lateral, imaging junction configuration. We present a full numerical model for the tunneling current of the major physical processes which determine the signal. We find that a diffusion framework models the quasiparticle dynamics well and that excited quasiparticles do not equilibrate to the lattice temperature during the timescales for tunneling. We extract physical timescales from the measured data, make comparisons with existing theories, and comment on implications for superconducting mesoscopic systems and single-photon detectors.
△ Less
Submitted 31 May, 2004; v1 submitted 2 February, 2004;
originally announced February 2004.
-
DC measurements of macroscopic quantum levels in a superconducting qubit structure with a time-ordered meter
Authors:
D. S. Crankshaw,
K. Segall,
D. Nakada,
T. P. Orlando,
L. S. Levitov,
S. Lloyd,
S. O. Valenzuela,
N. Markovic,
M. Tinkham,
K. K. Berggren
Abstract:
DC measurements are made in a superconducting, persistent current qubit structure with a time-ordered meter. The persistent-current qubit has a double-well potential, with the two minima corresponding to magnetization states of opposite sign. Macroscopic resonant tunneling between the two wells is observed at values of energy bias that correspond to the positions of the calculated quantum levels…
▽ More
DC measurements are made in a superconducting, persistent current qubit structure with a time-ordered meter. The persistent-current qubit has a double-well potential, with the two minima corresponding to magnetization states of opposite sign. Macroscopic resonant tunneling between the two wells is observed at values of energy bias that correspond to the positions of the calculated quantum levels. The magnetometer, a Superconducting Quantum Interference Device (SQUID), detects the state of the qubit in a time-ordered fashion, measuring one state before the other. This results in a different meter output depending on the initial state, providing different signatures of the energy levels for each tunneling direction. From these measurements, the intrawell relaxation time is found to be about 50 microseconds.
△ Less
Submitted 20 November, 2003; v1 submitted 19 November, 2003;
originally announced November 2003.
-
Impact of time-ordered measurements of the two states in a niobium superconducting qubit structure
Authors:
K. Segall,
D. Crankshaw,
D. Nakada,
T. P. Orlando,
L. S. Levitov,
S. Lloyd,
N. Markovic,
S. O. Valenzuela,
M. Tinkham,
K. K. Berggren
Abstract:
Measurements of thermal activation are made in a superconducting, niobium Persistent-Current (PC) qubit structure, which has two stable classical states of equal and opposite circulating current. The magnetization signal is read out by ram** the bias current of a DC SQUID. This ram** causes time-ordered measurements of the two states, where measurement of one state occurs before the other. T…
▽ More
Measurements of thermal activation are made in a superconducting, niobium Persistent-Current (PC) qubit structure, which has two stable classical states of equal and opposite circulating current. The magnetization signal is read out by ram** the bias current of a DC SQUID. This ram** causes time-ordered measurements of the two states, where measurement of one state occurs before the other. This time-ordering results in an effective measurement time, which can be used to probe the thermal activation rate between the two states. Fitting the magnetization signal as a function of temperature and ramp time allows one to estimate a quality factor of 10^6 for our devices, a value favorable for the observation of long quantum coherence times at lower temperatures.
△ Less
Submitted 26 February, 2003;
originally announced February 2003.
