Synapse Device Research Articles

Multi-Stimuli-Responsive Synapse Based on Vertical van der Waals Heterostructures.

Brain-inspired intelligent systems demand diverse neuromorphic devices beyond simple functionalities. Merging biomimetic sensing with weight-updating capabilities in artificial synaptic devices represents one of the key research focuses. Here, we report a multiresponsive synapse device that integrates synaptic and optical-sensing functions. The device adopts vertically stacked graphene/h-BN/WSe2 heterostructures, including an ultrahigh-mobility readout layer, a weight-control layer, and a dual-stimuli-responsive layer. The unique structure endows synapse devices with excellent synaptic plasticity, short response time (3 μs), and excellent optical responsivity (105 A/W). To demonstrate the application in neuromorphic computing, handwritten digit recognition was simulated based on an unsupervised spiking neural network (SNN) with a precision of 90.89%, well comparable with the state-of-the-art results. Furthermore, multiterminal neuromorphic devices are demonstrated to mimic dendritic integration and photoswitching logic. Different from other synaptic devices, the research work validates multifunctional integration in synaptic devices, supporting the potential fusion of sensing and self-learning in neuromorphic networks.

A Vis‐SWIR Photonic Synapse with Low Power Consumption Based on WSe2/In2Se3 Ferroelectric Heterostructure

AbstractNeuromorphic visual sensory and memory systems that can sense, learn and memorize optical information have great potential in many areas such as image recognition and autonomous driving. However, most current artificial neuromorphic vision technology is suffering from large power consumptions (>100 pJ per switching), high circuitry complexity, and difficulty in miniaturization due to the physical separation of the optic sensing, processing, and memory units. Here, a photonic neuromorphic device based on WSe2/In2Se3 van der Waals (vdW) heterostructure is developed to meet the requirements of high‐performance photonic synapse devices. Owing to the optically controlled ferroelectric polarization switching, the device shows light‐tunable synaptic functions. For the first time, the light‐stimulated synaptic behavior extends from visible light to short‐wavelength infrared region (up to 1800 nm). Taking advantage of the ultra‐low dark current of the heterojunction, the power consumption of the photonic synapse can be as low as 258 fJ per switching under a bias of −0.1 V, comparable to biological synapses. These findings demonstrate a novel photonic synapse, deepen the understanding of light‐controlled polarization switching in semiconducting ferroelectric materials, and open up new application fields in multifunctional electronic and photonic devices based on 2D ferroelectric vdW heterostructures.

In-depth understanding of physical mechanism of the gradual switching in AlOxNy-based RRAM as memory and synapse device

This paper reported the Pt/AlOxNy/Ta structure with oxygen scavenging metal Ta as a top electrode in order to achieve excellent data storage and the artificial synaptic function. As a series resistance layer and a thermal enhanced layer, the TaOx interface layer between the resistance switching (RS) layer and the top electrode (TE) can greatly affect the RS characteristics. These effects can be attributed to properties of the interface layer such as thermal and electrical conductivities, suggesting that properties of the interface layer should be carefully and systematically considered if excellent RS characteristics are desirable. Based on the electrical testing and current conduction mechanism analyses, the major factors affecting the bipolar gradual RS behavior can be concluded as the property of the interface layer between the RS layer and TE, the Joule heating caused by current, and the distribution of the electric field on the interface layer.

Humidity‐Enabled Organic Artificial Synaptic Devices with Ultrahigh Moisture Resistivity

AbstractOrganic memristors with homogeneous resistive switching behaviors can emulate the functionalities of biological synapses, making them promising candidates for use in brain‐inspired neuromorphic computing. Among various environmental parameters, humidity affecting device operational stability is a critical issue that organic memristors face in practical applications. A high humidity resistant organic memristor is demonstrated here using a small molecule material F16CuPc. Unencapsulated devices show excellent moisture stability, maintaining reliable memristive switching behavior after 1 year of exposure to air at 60% high relative humidity and remaining operational when immersed in water for 96 h. A proton conduction mechanism is discovered responsible for the resistive switching effect. The continuously adjustable conductance in F16CuPc‐based memristive devices enables the mimicking of biological synapse functions. These results open up a path to a humidity tolerant neuromorphic computing system based on high humidity resistant artificial synapse devices.

Novel training method for metal-oxide memristive synapse device to overcome trade-off between linearity and dynamic range

Synapse devices are essential for the hardware implementation of neuromorphic computing systems. However, it is difficult to realize ideal synapse devices because of issues such as nonlinear conductance change (linearity) and a small number of conductance states (dynamic range). In this study, the correlation between the linearity and dynamic range was investigated. Consequently, we found a trade-off relationship between the linearity and dynamic range and proposed a novel training method to overcome this trade-off.

Multiwavelength Optoelectronic Synapse with 2D Materials for Mixed-Color Pattern Recognition.

Neuromorphic visual systems emulating biological retina functionalities have enormous potential for in-sensor computing, with prospects of making artificial intelligence ubiquitous. Conventionally, visual information is captured by an image sensor, stored by memory units, and eventually processed by the machine learning algorithm. Here, we present an optoelectronic synapse device with multifunctional integration of all the processes required for real time object identification. Ultraviolet-visible wavelength-sensitive MoS2 FET channel with infrared sensitive PtTe2/Si gate electrode enables the device to sense, store, and process optical data for a wide range of the electromagnetic spectrum, while maintaining a low dark current. The device exhibits optical stimulation-controlled short-term and long-term potentiation, electrically driven long-term depression, synaptic weight update for multiple wavelengths of light ranging from 300 nm in ultraviolet to 2 μm in infrared. An artificial neural network developed using the extracted weight update parameters of the device can be trained to identify both single wavelength and mixed wavelength patterns. This work demonstrates a device that could potentially be used for realizing a multiwavelength neuromorphic visual system for pattern recognition and object identification.

Robust and Low-Power-Consumption Black Phosphorus-Graphene Artificial Synaptic Devices.

Two-dimensional (2D) black phosphorus (BP) materials, as the most promising building blocks for the development of artificial synapses, have attracted more and more attention. However, the instability of exfoliated 2D BP structures still remains a problem in the development of artificial synapse devices. In this study, the robust and low-power-consumption artificial-synaptic-based BP was successfully manufactured. The synapse devices have high stability in the air atmosphere and do not show obvious degradation over 3 months. The obtained devices not only implement the main function of synapses but also perform the function of dendritic neural synapses and simple logical operations, revealing their very strong learning behavior. The high mobility of 2D BP as well as the coupled effect and quantum confinement effect of the graphene oxide quantum dot (GOQD) can greatly boost the performance of BP-based synapse devices, such as low power consumption (62 pW) and high sensitivity (ultrasmall stimuli at an amplitude of -20 mV). Moreover, benefiting from the GOQD and the interaction between BP and graphene, the main dominated mechanism of the BP-graphene synapse device can be the capture and release of electrons by the 2D BP and GOQD instead of the conductive filament.

On-chip learning of a domain-wall-synapse-crossbar-array-based convolutional neural network

Abstract Domain-wall-synapse-based crossbar arrays have been shown to be very efficient, in terms of speed and energy consumption, while implementing fully connected neural network algorithms for simple data-classification tasks, both in inference and on-chip-learning modes. But for more complex and realistic data-classification tasks, convolutional neural networks (CNN) need to be trained through such crossbar arrays. In this paper, we carry out device–circuit–system co-design and co-simulation of on-chip learning of a CNN using a domain-wall-synapse-based crossbar array. For this purpose, we use a combination of micromagnetic-physics-based synapse-device modeling, SPICE simulation of a crossbar-array circuit using such synapse devices, and system-level-coding using a high-level language. In our design, each synaptic weight of the convolutional kernel is considered to be of 15 bits; one domain-wall-synapse crossbar array is dedicated to the five least significant bits (LSBs), and two crossbar arrays are dedicated to the other bits. The crossbar arrays accelerate the matrix vector multiplication operation involved in the forward computation of the CNN. The synaptic weights of the LSB crossbar are updated after forward computation on every training sample, while the weights of the other crossbars are updated after forward computation on ten samples, to achieve on-chip learning. We report high classification-accuracy numbers for different machine-learning data sets using our method. We also carry out a study of how the classification accuracy of our designed CNN is affected by device-to-device variations, cycle-to-cycle variations, bit precision of the synaptic weights, and the frequency of weight updates.

Ferrimagnetic Synapse Devices for Fast and Energy-Efficient On-Chip Learning on Crossbar-Array-Based Neural Networks (A Device-Circuit-System Costudy)

We propose a ferrimagnetic domain-wall synapse device as an alternative to its ferromagnetic counterpart for faster and more energy-efficient on-chip learning on a crossbar-array-based analog-hardware neural network. We first use micromagnetics to model domain-wall motion in a Co-Gd-bilayer-based ferrimagnetic device. We show that at the angular-momentum-compensation temperature, the ferrimagnetic-domain-wall velocity in this device is about 2–2.5 times ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$22.5\times $ </tex-math></inline-formula> ) higher than the ferromagnetic-domain-wall velocity in a CoFe/Pt device at room temperature, for the same value of current density. This velocity ratio is consistent with recent experimental findings. Next, we incorporate the results from our micromagnetic model in SPICE circuit simulation of a crossbar-array-based fully connected neural network (FCNN) using a Verilog-A model for each synapse in the crossbar array. We show that as a consequence of faster domain-wall motion in the ferrimagnetic device compared to the ferromagnetic device, for the same duration of synaptic-weight-modulating current pulses, total energy consumption in the synapses for on-chip learning is five times ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$5\times $ </tex-math></inline-formula> ) lower in the ferrimagnetic-synapse-based FCNN compared to the ferromagnetic-synapse-based FCNN. Similarly, for the same amount of energy consumed for learning, time taken for on-chip learning is five times ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$5\times $ </tex-math></inline-formula> ) lower for the former compared to the latter.

Neuromorphic chip integrated with a large-scale integration circuit and amorphous-metal-oxide semiconductor thin-film synapse devices

Artificial intelligences are promising in future societies, and neural networks are typical technologies with the advantages such as self-organization, self-learning, parallel distributed computing, and fault tolerance, but their size and power consumption are large. Neuromorphic systems are biomimetic systems from the hardware level, with the same advantages as living brains, especially compact size, low power, and robust operation, but some well-known ones are non-optimized systems, so the above benefits are only partially gained, for example, machine learning is processed elsewhere to download fixed parameters. To solve these problems, we are researching neuromorphic systems from various viewpoints. In this study, a neuromorphic chip integrated with a large-scale integration circuit (LSI) and amorphous-metal-oxide semiconductor (AOS) thin-film synapse devices has been developed. The neuron elements are digital circuit, which are made in an LSI, and the synapse devices are analog devices, which are made of the AOS thin film and directly integrated on the LSI. This is the world's first hybrid chip where neuron elements and synapse devices of different functional semiconductors are integrated, and local autonomous learning is utilized, which becomes possible because the AOS thin film can be deposited without heat treatment and there is no damage to the underneath layer, and has all advantages of neuromorphic systems.

HfZrOx-based capacitive synapses with highly linear and symmetric multilevel characteristics for neuromorphic computing

A hardware based artificial neural network (ANN), which holds the potential to alleviate the computation load and energy of a digital computer, has propelled the development of memory devices that can resemble the synapse. Memcapacitors, especially based on ferroelectric materials, with theoretically no static power, nondestructive readout, and multiple polarization states, are expected to have good energy efficiency and endurance as emerging artificial synapses. However, conventional ferroelectric devices are characterized with extremely high remnant polarization, which requires high energy for polarization state updating and always leads to low linearity and symmetry in updating properties. Here, we show a memcapacitive synapse based on an Au/HfZrOx (HZO)/Au ferroelectric memcapacitor with moderate remnant polarization that can offer unexceptionable updating properties for building an ANN. The memcapacitor demonstrates more than 64 weight states with an ultralow weight updating energy of ≤3.0 fJ/μm2. Both potentiation and depression synaptic characteristics show an ultralow non-linearity of &amp;lt;10−2. Based on these properties, a two-layer restricted Boltzmann machine is built based on this memcapacitive synapse, and it can be trained to reconstruct incomplete images. The reconstructed images show reduced Euclidean distance to originals in comparison with that of the incomplete images. Furthermore, the memcapacitive synapse is also tested by a handwritten digits recognition task based on a simple perceptron, and the pattern recognition accuracy is as high as 93.4%. These results indicate that the HZO-based capacitive synapse devices have great potential for future high-efficiency neuromorphic systems.

In-depth analysis on electrical parameters of floating gate IGZO synaptic transistor affecting pattern recognition accuracy

The reliable conductance modulation of synaptic devices is key when implementing high-performance neuromorphic systems. Herein, we propose a floating gate indium gallium zinc oxide (IGZO) synaptic device with an aluminum trapping layer to investigate the correlation between its diverse electrical parameters and pattern recognition accuracy. Basic synaptic properties such as excitatory postsynaptic current, paired pulse facilitation, long/short term memory, and long-term potentiation/depression are demonstrated in the IGZO synaptic transistor. The effects of pulse tuning conditions associated with the pulse voltage magnitude, interval, duration, and cycling number of the applied pulses on the conductance update are systematically investigated. It is discovered that both the nonlinearity of the conductance update and cycle-to-cycle variation should be critically considered using an artificial neural network simulator to ensure the high pattern recognition accuracy of Modified National Institute of Standards and Technology (MNIST) handwritten digit images. The highest recognition rate of the MNIST handwritten dataset is 94.06% for the most optimized pulse condition. Finally, a systematic study regarding the synaptic parameters must be performed to optimize the developed synapse device.

In-depth analysis on electrical parameters of floating gate IGZO synaptic transistor affecting pattern recognition accuracy.

The reliable conductance modulation of synaptic devices is key when implementing high-performance neuromorphic systems. Herein, we propose a floating gate IGZO synaptic device with an aluminum trapping layer to investigate the correlation between its diverse electrical parameters and pattern recognition accuracy. Basic synaptic properties such as excitatory postsynaptic current, paired pulse facilitation, long/short term memory, and long-term potentiation/depression are demonstrated in the IGZO synaptic transistor. The effects of pulse tuning conditions associated with the pulse voltage magnitude, interval, duration, and cycling number of the applied pulses on the conductance update are systematically investigated. It is discovered that both the nonlinearity of the conductance update and cycle-to-cycle variation should be critically considered using an artificial neural network simulator to ensure the high pattern recognition accuracy of Modified National Institute of Standards and Technology (MNIST) handwritten digit images. The highest recognition rate of the MNIST handwritten dataset is 94.06% for the most optimized pulse condition. Finally, a systematic study regarding the synaptic parameters must be performed to optimize the developed synapse device.

Light-stimulated artificial photonic synapses based on solution-processed In-Sn-Zn-O transistors for neuromorphic applications

Artificial photonic synapse devices (PSDs) hold great promise for the realization of next-generation artificial vision systems and processing units through a synergistic combination of brain-inspired neuromorphic computing and high levels of parallelism. Here, we demonstrate artificial PSDs based on solution-processed In-Sn-Zn-O (indium-tin-zinc oxide, ITZO) thin films capable of mimicking various neuromorphic functions. In particular, a transistor structure was adopted for PSDs to enable a facile control of the photo-response characteristics by gate biasing. With optimized gate bias condition, enhanced electrical conductance modulation was possible which can improve the energy efficiency of PSDs. In addition, we investigated the dependency of photo-response characteristics on light pulse waveforms to find out the correlation between various pulse parameters and the photo-current generation. Based on these findings, we demonstrated the emulation of associative learning which is one of the important cognitive functions of the brain. Moreover, to verify the translation of optically derived synaptic behaviors of ITZO PSDs into artificial neuromorphic computing, pattern recognition of handwritten digit patterns was demonstrated showing an accuracy up to 90.3%.

Forming‐Free Pt/Ti/AlOx/CeOx/Pt Multilayer Memristors with Multistate and Synaptic Characteristics

Due to their unique electrical performance and simple structure, memristors exhibit excellent application prospects for future information technology. In this work, we fabricated Pt/Ti/AlOx/CeOx/Pt memristors demonstrating electroforming‐free bipolar resistive switching behavior with low operating voltage (−1 to 1 V), stable endurance, and retention. Space‐charge limited conduction (SCLC) as well as the formation and rupture of conductive filaments are responsible for the resistive switching behavior. Increasing the magnitude of the RESET voltage could generate multistate resistive switching. We studied the synaptic characteristics of the device by obtaining multilevel conductance states and investigating the relationship between the device resistance, pulse amplitude, pulse width, and pulse number. By applying programmed pre and postsynaptic spiking pulses, spike‐timing dependent plasticity was observed. This study shows that the device is suitable for multivalue storage and can be used as an electronic synapse device in artificial neural networks.

Titanium oxide-based optoelectronic synapses with visual memory synergistically adjusted by internal emotions and ambient illumination.

Brain-inspired neuromorphic computing has become one of the critical technologies to overcome the bottleneck of von Neumann architecture. It is a vital step to construct a brain-like neuromorphic computing system at the hardware level by utilizing artificial synaptic devices. Compared with electronic synaptic devices, optoelectronic synaptic devices have the advantages of low power consumption, low crosstalk, and high bandwidth. Artificial optoelectronic synapses, analogous to retinal structure, can directly respond to and process light signal information to mimic the neuromorphic visual system. As high-level nerve impulses, both generated and regulated, emotions affect the strength and persistence of memory. Ambient illumination can provide visual perception to distinguish the size, color, and other characteristics of objects as well as affect the nonvisual functions of individuals, such as emotional states, thereby affecting learning and memory function. Herein, an artificial optoelectronic synapse composed of ITO/TiO2−x/p-Si was proposed. A variety of biologically dependent synaptic plasticity relating to learning and memory, including short-term synaptic plasticity, long-term synaptic plasticity, and learning-forgetting-relearning multifunctional advanced synaptic activity, was successfully simulated. A 3 × 3 artificial optoelectronic synapse array based on 9 devices was constructed to mimic the functions of visual learning and memory affected by internal emotion and ambient illumination. The proposed artificial optoelectronic synapse will exhibit great potential in visual and image information perception and memory.

Surface acoustic wave controlled skyrmion-based synapse devices

Magnetic skyrmions, which are particle-like spin structures, are promising information carriers for neuromorphic computing devices due to their topological stability and nanoscale size. In this work, we propose controlling magnetic skyrmions by electric-field-excited surface acoustic waves in neuromorphic computing device structures. Our micromagnetic simulations show that the number of created skyrmions, which emulates the synaptic weight parameter, increases monotonically with increases in the amplitude of the surface acoustic waves. Additionally, the efficiency of skyrmion creation is investigated systemically with a wide range of magnetic parameters, and the optimal values are presented accordingly. Finally, the functionalities of short-term plasticity and long-term potentiation are demonstrated via skyrmion excitation by a sequence of surface acoustic waves with different intervals. The application of surface acoustic waves in skyrmionic neuromorphic computing devices paves a novel approach to low-power computing systems.

Alternative negative weight for simpler hardware implementation of synapse device based neuromorphic system

Lately, there has been a rapid increase in the use of software-based deep learning neural networks (S-DNN) for the analysis of unstructured data consumption. For implementation of the S-DNN, synapse-device-based hardware DNN (H-DNN) has been proposed as an alternative to typical Von-Neumann structural computing systems. In the H-DNN, various numerical values such as the synaptic weight, activation function, and etc., have to be realized through electrical device or circuit. Among them, the synaptic weight that should have both positive and negative numerical values needs to be implemented in a simpler way. Because the synaptic weight has been expressed by conductance value of the synapse device, it always has a positive value. Therefore, typically, a pair of synapse devices is required to realize the negative weight values, which leads to additional hardware resources such as more devices, higher power consumption, larger area, and increased circuit complexity. Herein, we propose an alternative simpler method to realize the negative weight (named weight shifter) and its hardware implementation. To demonstrate the weight shifter, we investigated its theoretical, numerical, and circuit-related aspects, following which the H-DNN circuit was successfully implemented on a printed circuit board.

Photoelectroactive artificial synapse and its application to biosignal pattern recognition

In recent years, optoelectronic artificial synapses have garnered a great deal of research attention owing to their multifunctionality to process optical input signals or to update their weights optically. However, for most optoelectronic synapses, the use of optical stimuli is restricted to an excitatory spike pulse, which majorly limits their application to hardware neural networks. Here, we report a unique weight-update operation in a photoelectroactive synapse; the synaptic weight can be both potentiated and depressed using “optical spikes.” This unique bidirectional operation originates from the ionization and neutralization of inherent defects in hexagonal-boron nitride by co-stimuli consisting of optical and electrical spikes. The proposed synapse device exhibits (i) outstanding analog memory characteristics, such as high accessibility (cycle-to-cycle variation of <1%) and long retention (>21 days), and (ii) excellent synaptic dynamics, such as a high dynamic range (>384) and modest asymmetricity (<3.9). Such remarkable characteristics enable a maximum accuracy of 96.1% to be achieved during the training and inference simulation for human electrocardiogram patterns.

Flexible artificial synapse with relearning function based on ion gel-graphene FET

Due to its high-speed parallel processing, the human brain is superior to the von Neumann system computer in processing images, speech and perception. This paper proposes a flexible artificial synaptic device fabricated on graphene field effect transistors with ion gel. The channel with 100 µm is fabricated on a flexible substrate, and the weights of nerve synapses are adjusted by ion movement in the ion gel. The artificial synapse successfully mimics synaptic functions under low drain voltage (3 mV), such as excitatory postsynaptic current, inhibition postsynaptic current, paired-pulse facilitation and long-term plasticity with microampere current, and artificial synapse has a function similar to human forgetting and relearning. In order to integrate artificial synapse device with other circuits, the equivalent circuit of the artificial synapse simulates the plasticity of the synapse in this paper. These provide a new basis for the simulation and production of flexible artificial synapse, and the excellent long-term plasticity characteristics have huge application potential in memory functions.