Synapse Device Research Articles

On-chip adaptive matching learning with charge-trap synapse device and ReLU activation circuit

For the hardware implementation of artificial intelligence, neuromorphic systems have major advantages in terms of their energy consumption and massively parallel operation compared to conventional computing systems. For general-purpose neuromorphic systems, the on-chip learning of large-scale deep neural networks (DNN) is an essential function. However, compared to a backpropagation algorithm of DNN, an on-chip learning technology, which can be efficiently implemented in hardware without accuracy degradation, has not yet been developed. Consequently, off-chip learning-based neuromorphic systems that perform only inference operations are a promising approach to the first step in the commercialization of neuromorphic systems. To address the limitation of off-chip learning that cannot cope with real-time errors, we proposed on-chip adaptive matching learning (AML). By adding a spare single-layer neural network where an on-chip AML was carried out in parallel to the main neural network, it was possible to implement an adaptive neuromorphic system that can correct errors during real-time applications. For hardware implementation, we proposed a synapse device, synapse array, and neuron circuit. Finally, we conducted a system-level simulation of the adaptive neuromorphic system to verify the feasibility of the proposed on-chip AML.

Neuro‐Receptor Mediated Synapse Device Based on Crumpled MXene Ti3C2Tx Nanosheets

AbstractArtificial synapse devices can simulate the neuro‐transmission in a completely electronic way, but the neuro‐biochemical responses are still a challenge for them. Here, a novel three‐terminal (3T) neuro‐receptor‐mediated (acetylcholine receptor (AChR) as a proof‐of‐concept) synapse device (NR‐S) based on the solution–MXene interface is presented. It is demonstrated that the synaptic plasticity behavior triggered by neuro‐transmitter (ACh) and the pathogenic autoantibody (AChR‐ab) induced neuronal damage that can be detected and recorded. The improved sensitivities, including the linear responses to ACh in an extremely wide range (1 am to 1 µm) and ultra‐low (1 am) limit of detection, are obtained using crumpled MXene. Furthermore, the ability of the proposed NR‐S to determine the tiny neuronal injury caused by only 10 ng mL−1 AChR‐ab is conceptually proven. Collectively, the novel 3T NR‐S has good application prospects in the field of the neuromorphic chip for not only realizing the bionic simulation of the chemically modulated or injured neuro‐transmission but also offering an efficient experimental platform for neuro‐biochemistry studies.

An Optogenetics-Inspired Flexible van der Waals Optoelectronic Synapse and its Application to a Convolutional Neural Network.

Optogenetics refers to a technique that uses light to modulate neuronal activity with a high spatiotemporal resolution, which enables the manipulation of learning and memory functions in the human brain. This strategy of controlling neuronal activity using light can be applied for the development of intelligent systems, including neuromorphic and in-memory computing systems. Herein, a flexible van der Waals (vdW) optoelectronic synapse is reported, which is a core component of optogenetics-inspired intelligent systems. This synapse is fabricated on 2D vdW layered rhenium disulfide (ReS2 ) that features an inherent photosensitive memory nature derived from the persistent photoconductivity (PPC) effect, successfully mimicking the dynamics of biological synapses. Based on first-principles calculations, the PPC effect is identified to originate from sulfur vacancies in ReS2 that have an inherent tendency to form shallow defect states near the conduction band edges and under optical excitation lead to large lattice relaxation. Finally, the feasibility of applying the synapses in optogenetics-inspired intelligent systems is demonstrated via training and inference tasks for the CIFAR-10 dataset using a convolutional neural network composed of vdW optoelectronic synapse devices.

Retina‐Inspired Structurally Tunable Synaptic Perovskite Nanocones

AbstractArtificial photonic synapses with morphologically controlled photoreception, allowing for area‐dependent tunable light reception as well as information storage and learning, have potential for application in emerging photo‐interactive neuro‐computing technologies. Herein, an artificially intelligent (AI) photonic synapse with area‐density‐tunable perovskite nano‐cone arrays templated in a self‐assembled block copolymer (BCP) is presented, which is based on a field effect transistor with a floating gate of photoreceptive perovskite crystal arrays preferentially synthesized in a micro‐phase‐segregated BCP film. These arrays are capable of electric charge (de)trapping and photo‐excited charge generation, and they exhibit versatile synaptic functions of the nervous system, including paired‐pulse facilitation and long‐term potentiation, with excellent reliability. The area‐density variable perovskite floating gate developed by off‐centered spin coating process allows for emulating the human retina with a position‐dependent spatial distribution of cones. 60 × 12 arrays of the developed synapse devices exhibit position‐dependent dual functions of receptor and synapse. They are AI and exhibit a pattern recognition accuracy up to ≈90% when examined using the Modified National Institute of Standards and Technology handwritten digit pattern recognition test.

Solid-state electrolyte gated synaptic transistor based on SrFeO2.5 film channel

In the past few years, synaptic device has become a hot topic in material science because of the urgent demand of neuromorphic computing. Ionic liquids gated three-terminal synaptic transistors have drawn extensive attention due to their outstanding energy efficiency, linearity, and symmetry. However, utilization of liquid electrolyte is not appropriate for the requirement of the practical application. Herein, a novel three-terminal synaptic transistor is designed and constructed using a solid-state electrolyte gate and a brownmillerite SrFeO2.5 thin film channel. Topotactic phase transformation can be induced between insulative brownmillerite SrFeO2.5 and conductive perovskite SrFeO3−δ. Nonvolatile conductance switching of the SrFeOx film can be realized by inserting and extracting of oxygen ions with electrolyte gating. The essential synaptic learning functions including excitatory postsynaptic current, and short-term/long-term plasticity, are successfully mimicked in the three-terminal synaptic transistor. This synapse device achieves a large scale adjusted range and multistable plasticity, demonstrating a new way to achieve key component of upcoming neuromorphic circuitry.

Polarity Reversal Effect of a Memristor From the Circuit Point of View and Insights Into the Memristor Fuse

As the memristor device is asymmetrical in nature, it is not a bilateral element like the resistor in terms of circuit functionality. Thus, it causes hindrance in some memristor-based applications such as in cellular nonlinear network neighborhood connections and in some application areas where its orientation is essentially expected to act as a bilateral circuit element reliable for bidirectional communication, for example, in signal and image processing or in electrical synapse devices. We introduce a memristor-based network for each purpose where we replace the conventional series resistances by memristors. The memristor asymmetry is described from the circuit point of view allowing us to observe its interaction within the network. Moreover, a memristor fuse is proposed in order to achieve the memristive effect with symmetry, which is formed basically by connecting two memristors antiserially. We, therefore, analyze the memristor fuse from its basic principle along with the theoretical analysis and then observe the response from the circuit point of view.

Artificial Synapses: A Reliable All‐2D Materials Artificial Synapse for High Energy‐Efficient Neuromorphic Computing (Adv. Funct. Mater. 27/2021)

In article number 2011083, Congli He, Guangyu Zhang, and co-workers report an all-2D materials two-terminal floating-gate memory as an artificial synapse device for high energy-efficient neuromorphic computing. It exhibits linear and symmetric weight update behaviors with high reliability and tunability. A large number of states up to 3000, high switching speed of 40 ns, and low energy consumption of 18 fJ for a single pulse event are realized, demonstrating great potential for high-speed and low-power neuromorphic computing applications.

Phase field study on the performance of artificial synapse device based on the motion of domain wall in ferroelectric thin films

Artificial neural networks have gained intensive attention in recent years because of their potential in effectively reducing energy consumption and improving computation performance. Ferroelectric materials are considered to be promising candidates for artificial synapses because of their multiple and nonvolatile polarization states under external stimuli. Despite artificial ferroelectric synapses with multilevel states, long retention and fast switching speed have been reported, and some key fundamental issues, e.g., the influence of domain wall configuration and evolution on the performance of synapse behaviors, also remain unclear. In this work, we study the performance of artificial synapses based on the motion of 180° ferroelectric domain walls of stripe domain and cylinder domain in ferroelectric thin films via a dynamical phase field model. The results demonstrate that artificial synapses based on the stripe domain exhibit high linearity and symmetry in weight update under a weak electric field, compared with the cylinder domain. Based on such artificial synapses, the accuracy of an artificial neural network for the Modified National Institute of Standards and Technology handwritten digit recognition is over 92%. This work provides a domain-wall-based strategy to improve the weight updating linearity and symmetry of artificial synapse devices and the recognition accuracy of artificial neural networks.

Enhancement of Synaptic Characteristics Achieved by the Optimization of Proton-Electron Coupling Effect in a Solid-State Electrolyte-Gated Transistor.

Presently, the 3-terminal artificial synapse device has been in focus for neuromorphic computing systems owing to its excellent weight controllability. Here, an artificial synapse device based on the 3-terminal solid-state electrolyte-gated transistor is proposed to achieve outstanding synaptic characteristics with a human-like mechanism at low power. Novel synaptic characteristics are accomplished by precisely tuning the threshold voltage using the proton-electron coupling effect, which is caused by proton migration inside the electrolyte. However, these synaptic characteristics are degraded because traps at the interface of channel/electrolyte disturb the proton-electron coupling effect. To minimize degradation, the oxygen plasma treatment is performed to reduce interface traps. As a result, symmetric weight updates and outstanding synaptic characteristics are achieved. Furthermore, high repeatability and long-term plasticity are observed at low operating power, which is essential for artificial synapses. Therefore, this study shows the progress of artificial synapses and proposes a promising method, a low-power neuromorphic system, to achieve high accuracy.

Amorphous-Metal-Oxide-Semiconductor Thin-Film Planar-Type Spike-Timing- Dependent-Plasticity Synapse Device

An amorphous-metal-oxide-semiconductor (AOS) thin-film planar-type spike-timing-dependent-plasticity (STDP) synapse device has been developed. The AOS is a non-rare material and can be easily deposited, and therefore it is inexpensive. On the other hand, the STDP is promising as a learning principle for neuromorphic systems. In this study, first, an AOS thin-film planar-type STDP synapse device is actually fabricated. Next, spiking pulses are applied to the synapse device as pre-signals and post-signals. Finally, STDP characteristics, namely, long-term-depression (LTD) and long-term-potentiation (LTP), are observed, which seem to be due to charge injection into the underlayer insulator film.

A domain wall-magnetic tunnel junction artificial synapse with notched geometry for accurate and efficient training of deep neural networks

Inspired by the parallelism and efficiency of the brain, several candidates for artificial synapse devices have been developed for neuromorphic computing, yet a nonlinear and asymmetric synaptic response curve precludes their use for backpropagation, the foundation of modern supervised learning. Spintronic devices—which benefit from high endurance, low power consumption, low latency, and CMOS compatibility—are a promising technology for memory, and domain-wall magnetic tunnel junction (DW-MTJ) devices have been shown to implement synaptic functions such as long-term potentiation and spike-timing dependent plasticity. In this work, we propose a notched DW-MTJ synapse as a candidate for supervised learning. Using micromagnetic simulations at room temperature, we show that notched synapses ensure the non-volatility of the synaptic weight and allow for highly linear, symmetric, and reproducible weight updates using either spin transfer torque (STT) or spin–orbit torque (SOT) mechanisms of DW propagation. We use lookup tables constructed from micromagnetics simulations to model the training of neural networks built with DW-MTJ synapses on both the MNIST and Fashion-MNIST image classification tasks. Accounting for thermal noise and realistic process variations, the DW-MTJ devices achieve classification accuracy close to ideal floating-point updates using both STT and SOT devices at room temperature and at 400 K. Our work establishes the basis for a magnetic artificial synapse that can eventually lead to hardware neural networks with fully spintronic matrix operations implementing machine learning.

Neuromorphic Devices: Artificial Neuron and Synapse Devices Based on 2D Materials (Small 20/2021)

In article number 2100640, Stephen J. Pearton, Gwan-Hyoung Lee, Jihyun Kim, and co-workers review the recent progress on 2D-material-based neuromorphic devices. Artificial neuromorphic devices, including synapse and neuron devices, have been classified according to their working principles. The unique properties of 2D materials make it possible to emulate biological neural components, enabling to realize the neuromorphic devices that emulate advanced neural functions with high energy efficiency and excellent performance.

Alloyed High-k-Based Resistive Switching Memory in Contact Hole Structures

Resistive random-access memory (RRAM) devices are noticeable next generation memory devices. However, only few studies have been conducted regarding RRAM devices made of alloy. In this paper, we investigate the resistive switching behaviors of an Au/Ti/HfTiOx/p-Si memory device. The bipolar switching is characterized depending on compliance current under DC sweep mode. Good retention in the low-resistance state and high-resistance state is attained for nonvolatile memory and long-term memory in a synapse device. For practical switching operation, the pulse transient characteristics are studied for set and reset processes. Moreover, a synaptic weight change is achieved by a moderate pulse input for the potentiation and depression characteristics of the synaptic device. We reveal that the high-resistance state and low-resistance state are dominated by Schottky emissions.

Artificial Neuron and Synapse Devices Based on 2D Materials.

Neuromorphic systems, which emulate neural functionalities of a human brain, are considered to be an attractive next-generation computing approach, with advantages of high energy efficiency and fast computing speed. After these neuromorphic systems are proposed, it is demonstrated that artificial synapses and neurons can mimic neural functions of biological synapses and neurons. However, since the neuromorphic functionalities are highly related to the surface properties of materials, bulk material-based neuromorphic devices suffer from uncontrollable defects at surfaces and strong scattering caused by dangling bonds. Therefore, 2D materials which have dangling-bond-free surfaces and excellent crystallinity have emerged as promising candidates for neuromorphic computing hardware. First, the fundamental synaptic behavior is reviewed, such as synaptic plasticity and learning rule, and requirements of artificial synapses to emulate biological synapses. In addition, an overview of recent advances on 2D materials-based synaptic devices is summarized by categorizing these into various working principles of artificial synapses. Second, the compulsory behavior and requirements of artificial neurons such as the all-or-nothing law and refractory periods to simulate a spike neural network are described, and the implementation of 2D materials-based artificial neurons to date is reviewed. Finally, future challenges and outlooks of 2D materials-based neuromorphic devices are discussed.

Effectiveness of preventive measures for increased abrasion of hard dental tissues and its complications in young people involved in athletic gymnastics

BACKGROUND: Currently, there is an increased interest of young people in fitness, strength sports, as well as a desire to be in excellent physical shape. However, the influence of amateur sports loads using large weights on the condition of the dental system of young people has been little studied.
 AIM: The article analyzes the effectiveness of combined treatment, using an occlusive splint, of dental disorders in young people engaged in athletic gymnastics. The results of the study showed that the proposed method, which prevents the progressive auto-destruction of hard dental tissues in young men, improves the condition of periodontal tissues, primarily the gums, as well as the functional parameters of the temporomandibular joint dysfunction and masticatory muscles.
 MATERIALS AND METHODS: All the examined patients were divided into two groups. The observation group included 62 young men who regularly engage in athletic gymnastics, of which 31 people used individual occlusal tires during training (patent No. 188470 of April 15, 2019) and 31 people who continued sports training without tires.
 Research methods, tools used, equipment. In the course of the study, inspection, probing, and evaluation were performed according to the criteria of the KPU and KPUp indices, the Green-Vermillion hygiene Index (1964), and increased erasability was evaluated using the Smith-Knight index (1984), also used the enamel resistance test according to the method of V.R. Okushko, L.I. Kosareva, and I.K. Lutskaya (1984), the papillary-marginal alveolar indices in the Parma modification (1960) and the complex periodontal index according to P.A. Leus (1988), and the morpho-functional diagnosis of the temporomandibular joint using the Lira-100 device (NMBT Group, Russia).
 Methods of research-surface electromyography on the Synapse device (Neurotech, Russia), odontometry (caliper ADA Mechanic 150, China).
 RESULTS: The study showed that the activities of young men in gyms with large weights contribute to the development of a number of dental disorders.
 СONCLUSION: The inclusion of an occlusive splint used by young people during sports activities in the complex of therapeutic and preventive measures contributes to the positive dynamics of the state of dental status.

Brain-inspired ferroelectric Si nanowire synaptic device

We herein demonstrate a brain-inspired synaptic device using a poly(vinylidene fluoride) and trifluoroethylene (PVDF-TrFE)/silicon nanowire (Si NW) based ferroelectric field effect transistor (FeFET). The PVDF-TrFE/Si NW FeFET structure achieves reliable synaptic plasticity such as symmetrical potentiation and depression, thanks to the reversible dynamics of the PVDF-TrFE permanent dipole moment. The calculated asymmetric ratio of potentiation and depression is as low as 0.41 at the optimized bias condition, indicating a symmetrical synaptic plasticity behavior. Pattern recognition accuracy based on the actual synaptic plasticity data of the synaptic device can be estimated via the CrossSim simulation software. Our simulation result reveals a high pattern recognition accuracy of 85.1%, showing a potential feasibility for neuromorphic systems. Furthermore, the inverter-in-synapse transistor consisting of the Si NW FeFET synapse and resistor connected in series is able to provide energy-efficient logic circuits. A total noise margin [(NMH + NML)/VDD] of 41.6% is achieved, and the power consumption [Ps = VDD(ID,L + ID,H)/2] of the logic-in-synapse transistor is evaluated to be 0.6 µW per logic gate. This study would shed light on the way toward a brain-inspired neuromorphic computing system based on the FeFET synapse device.

Polyvinylalcohol (PVA)-Assisted Exfoliation of ReS2 Nanosheets and the Use of ReS2-PVA Composites for Transparent Memristive Photosynapse Devices.

Two-dimensional (2D) transition metal dichalcogenides (TMDs) have attracted significant attention for their outstanding optoelectrical properties. Unlike most TMDs with layer-dependent photoresponsivity, rhenium disulfide (ReS2) shows excellent thickness-independent photoresponsivity. Herein, we show a surfactant-free polyvinyl alcohol (PVA)-assisted exfoliation method for 2D-TMDs in aqueous solution and a transparent photosensitive memristor synapse device based on ReS2 nanosheets composited with PVA. ReS2 nanosheets are obtained via PVA-assisted exfoliation. After exfoliation, the ReS2-PVA dispersion solution is spin-coated on a substrate and dried to form a nanocomposite film without additional processing. Transparent memristors are then fabricated on plastic or glass substrates to demonstrate the applicability of the ReS2-PVA film. The devices show "write once, read many" memory behavior with a high ON/OFF current ratio (1.0 × 104 at 0.5 V) during electrical operation. In the high resistive state, synaptic functions with long-term memory behavior are successfully mimicked by applying photonic stimuli to the transparent ReS2-PVA memristors. The excitatory postsynaptic current stimulated by the photosignal is gradually reduced by electric stimuli. The proposed PVA-assisted exfoliation method is cost-effective, environmentally friendly, and applicable to various TMD nanomaterials. Furthermore, the ReS2-PVA nanocomposite film obtained via a simple solution-based process demonstrates excellent photosynaptic behavior.

Excellent Pattern Recognition Accuracy of Neural Networks Using Hybrid Synapses and Complementary Training

To overcome the performance degradation in hardware neural networks (NNs) with non-ideal synapse devices, we proposed a novel neuromorphic architecture with both TiO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</sub> -based interfacial RRAM and CBRAM-based filamentary RRAM for highly accurate NN training and long-term inference reliability. We used a threshold-triggered training scheme, in which interfacial and filamentary RRAMs were programmed in a complementary fashion. This took advantage of the long retention time of the filamentary RRAM and the high-resolution, symmetric weight update in the interfacial RRAM. Additional evaluation of device parameters, such as linearity, precision, variation, and retention time, was conducted. An excellent pattern recognition accuracy of ~97% was achieved during training with the MNIST dataset. Thus, reliable inference accuracy after training was maintained using the filamentary RRAM.

Integrated neuromorphic computing networks by artificial spin synapses and spin neurons

One long-standing goal in the emerging neuromorphic field is to create a reliable neural network hardware implementation that has low energy consumption, while providing massively parallel computation. Although diverse oxide-based devices have made significant progress as artificial synaptic and neuronal components, these devices still need further optimization regarding linearity, symmetry, and stability. Here, we present a proof-of-concept experiment for integrated neuromorphic computing networks by utilizing spintronics-based synapse (spin-S) and neuron (spin-N) devices, along with linear and symmetric weight responses for spin-S using a stripe domain and activation functions for spin-N. An integrated neural network of electrically connected spin-S and spin-N successfully proves the integration function for a simple pattern classification task. We simulate a spin-N network using the extracted device characteristics and demonstrate a high classification accuracy (over 93%) for the spin-S and spin-N optimization without the assistance of additional software or circuits required in previous reports. These experimental studies provide a new path toward establishing more compact and efficient neural network systems with optimized multifunctional spintronic devices.

Improved switching and synapse characteristics using PEALD SiO2 thin film in Cu/SiO2/ZrO2/Pt device

As the demand for data increases rapidly, neuromorphic computing, which can process parallel data and consume low power, is attracting to mimic the human nervous system. Resistive switching memory is considered as a viable element to realize neuromorphic computing due to high power and area efficiency. Although many previous studies have emphasized the linearity of analog switching, which is essential for realizing an ideal artificial synapse, it is still limited to control the diffusion of oxygen vacancies and suppress abrupt switching. Switching mechanism depends on the growth and dissolution of oxygen-deficient filaments, which could cause random variation. In this study, to further improve linear conductance change characteristics, we investigate the role of inserting a diffusion limiting layer in the synapse device operated by the migration of metal cations. Inserting thin SiO2 was prepared by plasma enhanced atomic layer deposition at the Cu/ZrO2 interface. This effectively mitigates the random diffusion of Cu ions and can perform multi-level RESET switching characteristics, which significantly improve the linearity of conductance with the number of pulses. In addition, various synaptic characteristics of potentiation and depression, short-term memory to long-term memory transition, pulse-paired facilitation, and post-tetanic potentiation, spike timing dependent plasticity were successfully emulated.