Search

Neuromorphic computing is a brain-inspired approach to hardware and algorithm design that efficiently realizes artificial neural networks. Neuromorphic designers apply the principles of biointelligence discovered by neuroscientists to design efficient computational systems, often for applications with size, weight and power constraints. With this research field at a critical juncture, it is crucial to chart the course for the development of future large-scale neuromorphic systems. We describe approaches for creating scalable neuromorphic architectures and identify key features. We discuss potential applications that can benefit from scaling and the main challenges that need to be addressed. Furthermore, we examine a comprehensive ecosystem necessary to sustain growth and the new opportunities that lie ahead when scaling neuromorphic systems. Our work distils ideas from several computing sub-fields, providing guidance to researchers and practitioners of neuromorphic computing who aim to push the frontier forward.Approaches for the development of future at-scale neuromorphic systems based on principles of biointelligence are described, along with potential applications of scalable neuromorphic architectures and the challenges that need to be overcome. [ABSTRACT FROM AUTHOR]

We present multiplexed gradient descent (MGD), a gradient descent framework designed to easily train analog or digital neural networks in hardware. MGD utilizes zero-order optimization techniques for online training of hardware neural networks. We demonstrate its ability to train neural networks on modern machine learning datasets, including CIFAR-10 and Fashion-MNIST, and compare its performance to backpropagation. Assuming realistic timescales and hardware parameters, our results indicate that these optimization techniques can train a network on emerging hardware platforms orders of magnitude faster than the wall-clock time of training via backpropagation on a standard GPU, even in the presence of imperfect weight updates or device-to-device variations in the hardware. We additionally describe how it can be applied to existing hardware as part of chip-in-the-loop training or integrated directly at the hardware level. Crucially, because the MGD framework is model-free it can be applied to nearly any hardware platform with tunable parameters, and its gradient descent process can be optimized to compensate for specific hardware limitations, such as slow parameter-update speeds or limited input bandwidth. [ABSTRACT FROM AUTHOR]

Superconducting optoelectronic hardware has been proposed for large-scale neural computing. In this work, we expand upon the circuit and network designs previously introduced. We investigate circuits using superconducting single-photon detectors and Josephson junctions to perform signal reception, synaptic weighting, and integration. Designs are presented for synapses and neurons that perform integration of rate-coded signals as well as detect coincidence events for temporal coding. A neuron with a single integration loop can receive input from thousands of synaptic connections, and many such loops can be employed for dendritic processing. We show that a synaptic weight can be modified via a superconducting flux-storage loop inductively coupled to the current bias of the synapse. Synapses with hundreds of stable states are designed. Spike-timing-dependent plasticity can be implemented using two photons to strengthen and two photons to weaken the synaptic weight via Hebbian-type learning rules. In addition to the synaptic receiver and plasticity circuits, we describe an amplifier chain that converts the current pulse generated when a neuron reaches threshold to a voltage pulse sufficient to produce light from a semiconductor diode. This light is the signal used to communicate between neurons in the network. We analyze the performance of the elements in the amplifier chain to calculate the energy consumption per photon created. The speed of the amplification sequence allows neuronal firing up to at least 20 MHz, independent of connectivity. We consider these neurons in network configurations to investigate near-term technological potential and long-term physical limitations. By modeling the physical size of superconducting optoelectronic neurons, we calculate the area of these networks. A system with 8100 neurons and 330 430 total synapses will fit on a 1 × 1 cm 2 die. Systems of millions of neurons with hundreds of millions of synapses will fit on a 300 mm wafer. For multiwafer assemblies, communication at light speed enables a neuronal pool the size of a large data center (10 5 m 2) comprised of trillions of neurons with coherent oscillations at 1 MHz. [ABSTRACT FROM AUTHOR]

Emerging neuromorphic hardware promises to solve certain problems faster and with higher energy efficiency than traditional computing by using physical processes that take place at the device level as the computational primitives in neural networks. While initial results in photonic neuromorphic hardware are very promising, such hardware requires programming or "training" that is often power-hungry and time-consuming. In this article, we examine the online learning paradigm, where the machinery for training is built deeply into the hardware itself. We argue that some form of online learning will be necessary if photonic neuromorphic hardware is to achieve its true potential. [ABSTRACT FROM AUTHOR]

We present multiplexed gradient descent (MGD), a gradient descent framework designed to easily train analog or digital neural networks in hardware. MGD utilizes zero-order optimization techniques for online training of hardware neural networks. We demonstrate its ability to train neural networks on modern machine learning datasets, including CIFAR-10 and Fashion-MNIST, and compare its performance to backpropagation. Assuming realistic timescales and hardware parameters, our results indicate that these optimization techniques can train a network on emerging hardware platforms orders of magnitude faster than the wall-clock time of training via backpropagation on a standard GPU, even in the presence of imperfect weight updates or device-to-device variations in the hardware. We additionally describe how it can be applied to existing hardware as part of chip-in-the-loop training or integrated directly at the hardware level. Crucially, because the MGD framework is model-free it can be applied to nearly any hardware platform with tunable parameters, and its gradient descent process can be optimized to compensate for specific hardware limitations, such as slow parameter-update speeds or limited input bandwidth. [ABSTRACT FROM AUTHOR]

Optical communication achieves high fanout and short delay advantageous for information integration in neural systems. Superconducting detectors enable signaling with single photons for maximal energy efficiency. We present designs of superconducting optoelectronic neurons based on superconducting single-photon detectors, Josephson junctions, semiconductor light sources, and multi-planar dielectric waveguides. These circuits achieve complex synaptic and neuronal functions with high energy efficiency, leveraging the strengths of light for communication and superconducting electronics for computation. The neurons send few-photon signals to synaptic connections. These signals communicate neuronal firing events as well as update synaptic weights. Spike-timing-dependent plasticity is implemented with a single photon triggering each step of the process. Microscale light-emitting diodes and waveguide networks enable connectivity from a neuron to thousands of synaptic connections, and the use of light for communication enables synchronization of neurons across an area limited only by the distance light can travel within the period of a network oscillation. Experimentally, each of the requisite circuit elements has been demonstrated; yet, a hardware platform combining them all has not been attempted. Compared to digital logic or quantum computing, device tolerances are relaxed. For this neural application, optical sources providing incoherent pulses with 10 000 photons produced with an efficiency of 10 − 3 operating at 20 MHz at 4.2 K are sufficient to enable a massively scalable neural computing platform with connectivity comparable to the brain and thirty thousand times higher speed. [ABSTRACT FROM AUTHOR]

Computer-aided design (CAD) has become a critical element in the creation of nanopatterned structures and devices. In particular, with the increased adoption of easy-to-learn programming languages like python, there has been a significant rise in the amount of lithographic geometries generated through scripting and programming. However, there are currently unaddressed gaps in usability for open-source CAD tools—especially those in the GDSII design space—that prevent wider adoption by scientists and students who might otherwise benefit from scripted design. For example, constructing relations between adjacent geometries is often much more difficult than necessary—spacing a resonator structure a few micrometers from a readout structure often requires manually coding the placement arithmetic. While inconveniences like these can be overcome by writing custom functions, they are often significant barriers to entry for new users or those less familiar with programming. To help streamline the design process and reduce barrier to entry for scripting designs, we have developed PHIDL, an open-source GDSII-based CAD tool for python. [ABSTRACT FROM AUTHOR]

Equilibrium propagation is a learning framework that marks a step forward in the search for a biologically-plausible implementation of deep learning, and could be implemented efficiently in neuromorphic hardware. Previous applications of this framework to layered networks encountered a vanishing gradient problem that has not yet been solved in a simple, biologically-plausible way. In this paper, we demonstrate that the vanishing gradient problem can be mitigated by replacing some of a layered network's connections with random layer-skipping connections in a manner inspired by small-world networks. This approach would be convenient to implement in neuromorphic hardware, and is biologically-plausible. [ABSTRACT FROM AUTHOR]

We demonstrate adiabatically tapered fibers terminating in sub-micron tips that are clad with a higher-index material for coupling to an on-chip waveguide. This cladding enables coupling to a high-index waveguide without losing light to the buried oxide. A technique to clad the tip of the tapered fiber with a higher-index polymer is introduced. Conventional tapered waveguides and forked tapered waveguide structures are investigated for coupling from the clad fiber to the on-chip waveguide. We find the forked waveguide facilitates alignment and packaging, while the conventional taper leads to higher bandwidth. The insertion loss from a fiber through a forked coupler to a sub-micron silicon nitride waveguide is 1.1 dB and the 3 dB bandwidth is 90 nm. The coupling loss in the packaged device is 1.3 dB. With a fiber coupled to a conventional tapered waveguide, the loss is 1.4 dB with a 3 dB bandwidth extending beyond the range of the measurement apparatus, estimated to exceed 250 nm. [ABSTRACT FROM AUTHOR]

We demonstrate cryogenic, all on-chip, single-photon-level photonic integrated circuits on a III-V platform with waveguide-coupled quantum-well sources and tungsten silicide superconducting nanowire single-photon detectors. We have measured the dark count rates below 10−3 counts/s and have reduced the cross talk to an adjacent waveguide by 30 dB. [ABSTRACT FROM AUTHOR]

We design, fabricate, and characterize integrated photonic routing manifolds with 10 inputs and 100 outputs using two vertically integrated planes of silicon nitride waveguides. We analyze manifolds via top-view camera imaging. This measurement technique allows the rapid acquisition of hundreds of precise transmission measurements. We demonstrate manifolds with uniform and Gaussian power distribution patterns with mean power output errors (averaged over 10 sets of 10 inputs) of 0.7 and 0.9 dB, respectively, establishing this as a viable architecture for precision light distribution on-chip. We also assess the performance of the passive photonic elements comprising the system via self-referenced test structures, including high-dynamic-range beam taps, waveguide cutback structures, and waveguide crossing arrays. [ABSTRACT FROM AUTHOR]

We propose and experimentally demonstrate a photonic routing architecture that can efficiently utilize the space of multi-plane (3D) photonic integration. A wafer with three planes of amorphous silicon waveguides was fabricated and characterized, demonstrating <3 ×10-4 dB loss per out-of-plane waveguide crossing, 0.05±0.02 dB per interplane coupler, and microring resonators on three planes with a quality factors up to 8.2 × 104. We also explore a phase velocity mapping strategy to mitigate the cross talk between co-propagating waveguides on different planes. These results expand the utility of 3D photonic integration for applications such as optical interconnects, neuromorphic computing and optical phased arrays. [ABSTRACT FROM AUTHOR]

We demonstrate cryogenic, electrically injected, waveguide-coupled Si light-emitting diodes (LEDs) operating at 1.22 μm. The active region of the LED consists of W centers implanted in the intrinsic region of a p-i-n diode. The LEDs are integrated on waveguides with superconducting nanowire single-photon detectors (SNSPDs). We demonstrate the scalability of this platform with an LED coupled to eleven SNSPDs in a single integrated photonic device. [ABSTRACT FROM AUTHOR]

Engineering the electromagnetic environment of a nanometre-scale light emitter by use of a photonic cavity can significantly enhance its spontaneous emission rate, through cavity quantum electrodynamics in the Purcell regime. This effect can greatly reduce the lasing threshold of the emitter, providing a low-threshold laser system with small footprint, low power consumption and ultrafast modulation. An ultralow-threshold nanoscale laser has been successfully developed by embedding quantum dots into a photonic crystal cavity (PCC). However, several challenges impede the practical application of this architecture, including the random positions and compositional fluctuations of the dots, extreme difficulty in current injection, and lack of compatibility with electronic circuits. Here we report a new lasing strategy: an atomically thin crystalline semiconductor-that is, a tungsten diselenide monolayer-is non-destructively and deterministically introduced as a gain medium at the surface of a pre-fabricated PCC. A continuous-wave nanolaser operating in the visible regime is thereby achieved with an optical pumping threshold as low as 27 nanowatts at 130 kelvin, similar to the value achieved in quantum-dot PCC lasers. The key to the lasing action lies in the monolayer nature of the gain medium, which confines direct-gap excitons to within one nanometre of the PCC surface. The surface-gain geometry gives unprecedented accessibility and hence the ability to tailor gain properties via external controls such as electrostatic gating and current injection, enabling electrically pumped operation. Our scheme is scalable and compatible with integrated photonics for on-chip optical communication technologies. [ABSTRACT FROM AUTHOR]

Fast, high efficiency and low error single-photon sources are required for the implementation of a number of quantum information processing applications. The fastest triggered single-photon sources to date have been demonstrated using epitaxially grown semiconductor quantum dots (QDs), which can be conveniently integrated with optical microcavities. Recent advances in QD technology, including demonstrations of high temperature and telecommunications wavelength single-photon emission, have made QD single-photon sources more practical. Here we discuss the applications of single-photon sources and their various requirements, before reviewing the progress made on a QD platform in meeting these requirements. [ABSTRACT FROM AUTHOR]

n investigation of the dielectric properties of polyamine buffer prepared human chromosomes is presented in this paper. Chromosomes prepared in this buffer are only a few micrometers in size and shaped roughly like spherical discs. Dielectrophoresis was therefore chosen as the method of manipulation combined with a custom designed microfluidic system containing the required electrodes for dielectrophoresis experiments. Our results show that although this system is presently not able to distinguish between the different chromosomes, it can provide average data for the dielectric properties of human chromosomes in polyamine buffer. These can then be used to optimize system designs for further characterization and even sorting. The experimental data from the dielectrophoretic manipulation were combined with theoretical calculations to extract a range of values for the permittivity and conductivity of human polyamine buffer prepared chromosomes. [ABSTRACT FROM AUTHOR]

We propose and experimentally demonstrate a photonic crystal nanocavity with multiple resonances that can be tuned nearly independently. The design is composed of two orthogonal intersecting nanobeam cavities. Experimentally, we measure cavity quality factors of 6600 and 1000 for resonances separated by 382 nm; we measure a maximum separation between resonances of 506 nm. These structures are promising for enhancing efficiency in nonlinear optical processes such as sum/difference frequency and stimulated Raman scattering. [ABSTRACT FROM AUTHOR]

We demonstrate enhanced second harmonic generation in a gallium phosphide photonic crystal waveguide with a measured external conversion efficiency of 5 × 10-7/W. Our results are promising for frequency conversion of on-chip integrated emitters having broad spectra or large inhomogeneous broadening, as well as for frequency conversion of ultrashort pulses. [ABSTRACT FROM AUTHOR]

We demonstrate a quantum dot single photon source at 900 nm triggered at 100 MHz by a continuous wave telecommunications wavelength laser followed by an electro-optic modulator. The quantum dot is excited by on-chip-generated second harmonic radiation, resonantly enhanced by a GaAs photonic crystal cavity surrounding the InAs quantum dot. Our result suggests a path toward the realization of telecommunications wavelength-compatible quantum dot single photon sources with speeds exceeding 1 GHz. [ABSTRACT FROM AUTHOR]

We demonstrate second harmonic generation at telecommunications wavelengths in photonic crystal cavities in (111)-oriented GaAs. We fabricate 30 photonic crystal structures in both (111)- and (100)-oriented GaAs and observe an increase in generated second harmonic power in the (111) orientation, with the mean power increased by a factor of 3, although there is a large scatter in the measured values. We discuss possible reasons for this increase, in particular, the reduced two photon absorption for transverse electric modes in (111) orientation, as well as a potential increase due to improved mode overlap. [ABSTRACT FROM AUTHOR]

Indistinguishable single photons are necessary for quantum information processing applications. Resonant or quasiresonant excitation of single quantum dots provides greater single photon indistinguishability than incoherent pumping, but is also more challenging experimentally. Here, we demonstrate high signal to noise quasiresonant excitation of InP/InGaP quantum dots. The excitation is provided via second harmonic generated from a telecommunications wavelength laser resonant with the fundamental mode of a photonic crystal cavity, fabricated at twice the quantum dot transition wavelength. The second harmonic is generated using the χ(2) nonlinearity of the InGaP material matrix. [ABSTRACT FROM AUTHOR]