Advanced Optoelectronic Devices for Neuromorphic Analog Based on Low-Dimensional Semiconductors

Neuromorphic systems can parallelize the perception and computation of information, making it possible to break through the von Neumann bottleneck. Neuromorphic engineering has been developed over a long period of time based on Hebbian learning rules. The optoelectronic neuromorphic analog device combines the advantages of electricity and optics, and can simulate the biological visual system, which has a very strong development potential. Low-dimensional materials play a very important role in the field of optoelectronic neuromorphic devices due to their flexible bandgap tuning mechanism and strong light-matter coupling efficiency. This review introduces the basic synaptic plasticity of neuromorphic devices. According to the different number of terminals, two-terminal neuromorphic memristors, three-terminal neuromorphic transistors and artificial visual system are introduced from the aspects of the action mechanism and device structure. Finally, the development prospect of optoelectronic neuromorphic analog devices based on low-dimensional materials is prospected.     

Optoelectronic Perovskite Synapses for Neuromorphic Computing

Simulating the human brain for neuromorphic computing has attractive prospects in the field of artificial intelligence. Optoelectronic synapses have been considered to be important cornerstones of neuromorphic computing due to their ability to process optoelectronic input signals intelligently. In this work, optoelectronic synapses based on all‐inorganic perovskite nanoplates are fabricated, and the electronic and photonic synaptic plasticity is investigated. Versatile synaptic functions of the nervous system, including paired‐pulse facilitation, short‐term plasticity, long‐term plasticity, transition from short‐ to long‐term memory, and learning‐experience behavior, are successfully emulated. Furthermore, the synapses exhibit a unique memory backtracking function that can extract historical optoelectronic information. This work could be conducive to the development of artificial intelligence and inspire more research on optoelectronic synapses.     

Broadband Optoelectronic Synapse Enables Compact Monolithic Neuromorphic Machine Vision for Information Processing

Traditional machine vision is suffering from redundant sensing data, bulky structures, and high energy consumption. Biological-inspired neuromorphic systems are promising for compact and energy-efficient machine vision. Multifunctional optoelectronics enabling multispectrum sensitivity for broadband image sensing, feature extraction, and neuromorphic computing are vital for machine visions. Here, an optoelectronic synapse is designed that enables image sensing, convolutional processing, and computing. Multiple synaptic plasticity triggered by photons can implement photonic computing and information transmission. Convolutional processing is realized by ultralow energy kernel generators fully controlled by photons. Meanwhile, the device shows the ability of conductance modulations under electronic stimulations that implement neuromorphic computing. For the first time, this two-terminal broadband optoelectronic synapse enables front-end retinomorphic image sensing, convolutional processing, and back-end neuromorphic computing. The integrated photonic information encryption, convolutional image preprocessing, and neuromorphic computing capabilities are promising for compact monolithic neuromorphic machine vision systems.     

HfO2-Based Memristor as an Artificial Synapse for Neuromorphic Computing with Tri-Layer HfO2/BiFeO3/HfO2 Design

Neuromorphic devices are among the most emerging electronic components to realize artificial neural systems and replace traditional complementary metal–oxide semiconductor devices in recent times. In this work, tri-layer HfO₂/BiFeO₃(BFO)/HfO₂ memristors are designed by inserting traditional ferroelectric BFO layers measuring ≈4 nm after thickness optimization. The novel designed memristor shows excellent resistive switching (RS) performance such as a storage window of 10⁴ and multi-level storage ability. Remarkably, essential synaptic functions can be successfully realized on the basis of the linearity of conductance modulation. The pattern recognition simulation based on neuromorphic network is conducted with 91.2% high recognition accuracy. To explore the RS performance enhancement and artificial synaptic behaviors, conductive filaments (CFs) composed of Hafnium (Hf) single crystal with a hexaganal lattice structure are observed using high-resolution transmission electron microscopy. It is reasonable to believe that the sufficient oxygen vacancies in the inserting BFO thin film play a crucial role in adjusting the morphology and growth of Hf CFs, which leads to the promising synaptic and enhanced RS behavior, thus demonstrating the potential of this memristor for use in neuromorphic computing.     

Bidirectionally Photoresponsive Optoelectronic Transistors with Dual Photogates for All-Optical-Configured Neuromorphic Vision

Bio-inspired neuromorphic vision sensors, integrating optical sensing, and processing functions have attracted significant attention for developing future low-power and high-efficiency imaging systems. However, the compulsory electrical signal modulation to achieve inhibitory behaviors in most reported neuromorphic vision sensors results in additional hardware and computational latency. Herein, bidirectional photoresponsive optoelectronic synapses based on In₂O₃/Al₂O₃/Y6 phototransistors are achieved, realizing all-optical-configured synaptic weight updates enabled by dual photogates. The inhibitory and excitatory photoresponses originate from the photogating effects provided by trapped photogenerated electrons in Al₂O₃ under near-infrared light and the ionized oxygen vacancies in In₂O₃ under ultra-violet light, respectively. The bidirectional phototransistor illustrates outstanding optoelectronic synaptic characteristics with low nonlinearity and asymmetry, demonstrating high efficiencies in both preprocessing and postprocessing tasks, such as noise reduction, contrast enhancement, and pattern recognition. The proposed dual-photogate optoelectronic synapses provide effective strategies to construct high-efficiency neuromorphic vision sensors and in-sensor computing systems.     

All-Optically Controlled Retinomorphic Memristor for Image Processing and Stabilization

Image stabilization is a crucial field in machine vision, aiming to eliminate image blurring or distortion caused by the camera or object jitter. However, traditional image stabilization techniques often suffer from the drawbacks of requiring complex equipment or extensive computing resources, resulting in inefficiencies. In contrast, the human retina performs a highly efficient all-in-one system, encompassing the detection and processing of light stimuli. In this study, an all-optically controlled retinomorphic memristor based on the Cs_xFA_yMA_1-x-yPb(I_zBr_1-z)₃ is proposed, which integrates perception, storage, and processing functions. This memristor exhibits significant advantages in image stabilization. It is capable of positively and negatively modulating its conductance using specific intensities (11.8 and 0.9 mW cm⁻², respectively) of red light (630 nm). To demonstrate the effectiveness of the proposed approach, handwritten digit recognition simulations are conducted. The application of specific light stimuli effectively highlights the characteristics of blurred images. The processed images are then fed into a conductance-mapped neural network for rapid recognition. Remarkably, the recognition rates of the processed images reach 83.5% after 19 000 iterations, surpassing the performance of blurred images (only 56.2% after 19 000 iterations). These results highlight the immense potential of retinomorphic memristors as the hardware foundation for next-generation image stabilization systems.     

Bio-Inspired Artificial Perceptual Devices for Neuromorphic Computing and Gesture Recognition

Artificial perception technologies capable of sensing and feeling mechanical stimuli like human skins are critical enablers for electronic skins (E-Skins) needed to achieve artificial intelligence. However, most of the reported electronic skin systems lack the capability to process and interpret the sensor data. Herein, a new design of artificial perceptual system integrating ZnO-based synaptic devices with Pt/carbon nanofibers-based strain sensors for stimuli detection and information processing is presented. Benefiting from the controllable ion migration after indium doping, the device can emulate various essential functions, such as short-term/long-term plasticity, paired-pulse facilitation, excitatory post-synaptic current, and synaptic plasticity depending on the number, frequency, amplitude, and width of the applied pulses. The Pt/carbon nanofibers-based strain sensors can detect subtle human motion and convert mechanical stimuli into electrical signals, which are further processed by the ZnO devices. By attaching the integrated devices to finger joints, it is demonstrated that they can recognize handwriting and gestures with a high accuracy. This work offers new insights in designing artificial synapses and sensors to process and recognize information for neuromorphic computing and artificial intelligence applications.     

Set/Reset Bilaterally Controllable Resistance Switching Ga-doped Ge2Sb2Te5 Long-Term Electronic Synapses for Neuromorphic Computing

Long-term plasticity of bio-synapses modulates the stable synaptic transmission that is quite related to the encoding of information and its emulation using electronic hardware is one of important targets for neuromorphic computing. Ge₂Sb₂Te₅ (GST) based phase change random access memory (PCRAM) has become a strong candidate for complementary-metal-oxide-semiconductor (CMOS) compatible integrated long-term electronic synapses to cope with the high-efficient and low power consumption data processing tasks for neuromorphic computing. However, the performance of PCRAM electronic synapses is still quite limited due to the challenges in linear and continuous conductance regulation, which originates from the fast and uncontrollable resistance switching characteristic of conventional PCRAM for the data storage application. Here an in-depth study is reported on the impact of gallium (Ga) doping on GST (GaGST) structural properties and on the corresponding 0.13 µm CMOS technology fabricated PCRAM integrated devices with a mushroom structure. The Ga doping effectively retarded the crystallization process of GST by augmenting the local disorder of GeTe_4-nGe_n tetrahedron, which subsequently leads to the Set/Reset bilaterally controllable resistance switching of corresponding PCRAM devices. The optimized 6.5%GaGST electronic synapses demonstrate gradual resistance switching characteristics and a good multilevel retention feature and eventually exhibit outstanding long-term synaptic plasticity like potentiation/depression and spiking time dependent plasticity in four forms. Such long-term electronic synapses are applied to handwritten digits recognition (96.22%) and CIFAR-10 image categorization (93.6%) and attain very high accuracy for both tasks. These results provide an effective method to achieve high performance PCRAM electronic synapses and highlight the great potential of GaGST PCRAM as a component for future high-performance neuromorphic computing.     

Compensated Ferrimagnet Based Artificial Synapse and Neuron for Ultrafast Neuromorphic Computing

Spintronic devices are considered a possible solution for the hardware implementation of artificial synapses and neurons, as a result of their non-volatility, high scalability, complementary metal-oxide-semiconductor transistor compatibility, and low power consumption. As compared to ferromagnets, ferrimagnet-based spintronics exhibits equivalently fascinating properties that have been witnessed in ultrafast spin dynamics, together with efficient electrical or optical manipulation. Their applications in neuromorphic computing, however, have still not been revealed, which motivates the present experimental study. Here, by using compensated ferrimagnets containing Co_0.80Gd_0.20 with perpendicular magnetic anisotropy, it is demonstrated that the behavior of spin-orbit torque switching in compensated ferrimagnets could be used to mimic biological synapses and neurons. In particular, by using the anomalous Hall effect and magneto-optical Kerr effect imaging measurements, the ultrafast stimulation of artificial synapses and neurons is illustrated, with a time scale down to 10 ns. Using experimentally derived device parameters, a three-layer fully connected neural network for handwritten digits recognition is further simulated, based on which, an accuracy of more than 93% could be achieved. The results identify compensated ferrimagnets as an intriguing candidate for the ultrafast neuromorphic spintronics.     

Electrospun Nanofiber-Based Synaptic Transistor with Tunable Plasticity for Neuromorphic Computing

Biological synapses are the operational connection of the neurons for signal transmission in neuromorphic networks and hardware implementation combined with electrospun 1D nanofibers have realized its functionality for complicated computing tasks in basic three-terminal field-effect transistors with gate-controlled channel conductance. However, it still lacks the fundamental understanding that how the technological parameters influence the signal intensity of the information processing in the neural systems for the nanofiber-based synaptic transistors. Here, by tuning the electrospinning parameters and introducing the channel surface doping, an electrospun ZnO nanofiber-based transistor with tunable plasticity is presented to emulate the changing synaptic functions. The underlying mechanism of influence of carrier concentration and mobility on the device's electrical and synaptic performance is revealed as well. Short-term plasticity behaviors including paired-pulse facilitation, spike duration-dependent plasticity, and dynamic filtering are tuned in this fiber-based device. Furthermore, Perovskite-doped devices with ultralow energy consumption down to ≈0.2554 fJ and their handwritten recognition application show the great potential of synaptic transistors based on a 1D nanostructure active layer for building next-generation neuromorphic networks.     

Emerging MXene-Based Memristors for In-Memory,Neuromorphic Computing,and Logic Operation

Confronted by the difficulties of the von Neumann bottleneck and memory wall, traditional computing systems are gradually inadequate for satisfying the demands of future data-intensive computing applications. Recently, memristors have emerged as promising candidates for advanced in-memory and neuromorphic computing, which pave one way for breaking through the dilemma of current computing architecture. Till now, varieties of functional materials have been developed for constructing high-performance memristors. Herein, the review focuses on the emerging 2D MXene materials-based memristors. First, the mainstream synthetic strategies and characterization methods of MXenes are introduced. Second, the different types of MXene-based memristive materials and their widely adopted switching mechanisms are overviewed. Third, the recent progress of MXene-based memristors for data storage, artificial synapses, neuromorphic computing, and logic circuits is comprehensively summarized. Finally, the challenges, development trends, and perspectives are discussed, aiming to provide guidelines for the preparation of novel MXene-based memristors and more engaging information technology applications.     

CuInP2S6-Based Electronic/Optoelectronic Synapse for Artificial Visual System Application

With the development of information processing, various neuromorphic synaptic devices have been proposed, including novel devices mimicking multiple sensory systems in the biosome, in which vision is a vital source of information. Due to the pressing issue of high energy consumption and the ever-increasing complexity of practical application scenarios, there is an urgent need to investigate optoelectronic synapses with simple structure but multifunctional capabilities, thereby broadening their application scope. Here, remarkable performances in both electrical and optical operation modes are achieved in multilayer graphene/CuInP₂S₆/Au electronic/optoelectronic device. By modulating the electrical and optical pulses, both short-term and long-term memory can be emulated in the same device, while visual perception, processing, and memorizing functions are demonstrated in this single cell with relatively low energy consumption. In addition, light adaptive behavior can also be simulated through optical–electrical cooperative modulation in the device, further providing a novel and promising strategy for future applications in artificial visual systems.     

Carbon Nanotube Optoelectronic Synapse Transistor Arrays with Ultra-Low Power Consumption for Stretchable Neuromorphic Vision Systems

High-performance stretchable optoelectronic synaptic transistor arrays are key units for constructing and mimicking simulated neuromorphic vision systems. In this study, ultra-low power consumption and low-operation-voltage stretchable all-carbon optoelectronic synaptic thin film transistors (TFTs) using sorted semiconducting single-walled carbon nanotubes (sc-SWCNTs) modified with CdSe/ZnS quantum dots as active layers on ionic liquid-based composite elastomer substrates are first reported. The resulting stretchable TFT devices show enhancement-mode characteristics with excellent electrical properties (such as the record on/off ratios up to 10⁵, negligible hysteresis, and small subthreshold swing), excellent mechanical tensile properties (such as the only 12.4% and 6.4% degradations of the carrier mobility after 20% vertical and horizontal strain stretching), and optoelectronic synaptic plasticity (for the recognition of Morse codes) with ultra-low power consumptions (15.38 aJ) at the operating voltage from −1 to 0.2 V. At the same time, the designed nonvolatile conductance of the stretchable SWCNT optoelectronic synapse thin film transistors (SSOSTFTs) stimulated by UV light and the bending angle are first used to simulate stretchable neuromorphic vision systems (including the functions of the crystalline lens and optic cone cells as bionic eyes) for detecting the atmospheric environment with a record accuracy of 95.1% as a bionic eye.     

Photoferroelectric Perovskite Synapses for Neuromorphic Computing

Halide perovskite is an emerging material with excellent optoelectronic properties, and also widely used in neuromorphic devices. Recently, halide perovskite has been redefined as exhibiting extraordinary multifunction, e.g., photoferroelectricity. Herein, this work employs a composite material consisting of halide perovskite and organic ferroelectric material to develop a new photoferroelectric synapse, and the photoferroelectricity and some synaptic plasticity are investigated. By the corresponding test analysis, it is demonstrated that photoelectricity and ferroelectricity can reinforce each other in this photoferroelectric composite material. Versatile synaptic behaviors of the nervous system, including paired-pulse facilitation/paired-pulse depression, post-tetanic potentiation /post-tetanic depression, and spiking-rate-dependent plasticity, are successfully simulated. Particularly, the classical conditioning in Pavlov's dog experiment can be replicated in the photoferroelectric synapse to realize the learning function of the brain, including memory loss and recovery. This work could be conducive to the application of multifunctional perovskite materials in synapse devices and neuromorphic computing.     

Artificial Synapses Based on Multiterminal Memtransistors for Neuromorphic Application

Neuromorphic computing, which emulates the biological neural systems could overcome the high‐power consumption issue of conventional von‐Neumann computing. State‐of‐the‐art artificial synapses made of two‐terminal memristors, however, show variability in filament formation and limited capacity due to their inherent single presynaptic input design. Here, a memtransistor‐based arti?cial synapse is realized by integrating a memristor and selector transistor into a multiterminal device using monolayer polycrys‐talline‐MoS₂ grown by a scalable chemical vapor deposition (CVD) process. Notably, the memtransistor offers both drain‐ and gate‐tunable nonvolatile memory functions, which efficiently emulates the long‐term potentiation/depression, spike‐amplitude, and spike‐timing‐dependent plasticity of biological synapses. Moreover, the gate tunability function that is not achievable in two‐terminal memristors, enables significant bipolar resistive states switching up to four orders‐of‐magnitude and high cycling endurance. First‐principles calculations reveal a new resistive switching mechanism driven by the diffusion of double sulfur vacancy perpendicular to the MoS₂ grain boundary, leading to a conducting switching path without the need for a filament forming process. The seamless integration of multiterminal memtransistors may offer another degree‐of‐freedom to tune the synaptic plasticity by a third gate terminal for enabling complex neuromorphic learning.     

Redox Memristors with Volatile Threshold Switching Behavior for Neuromorphic Computing

The spiking neural network(SNN), closely inspired by the human brain, is one of the most powerful platforms to enable highly efficient, low cost, and robust neuromorphic computations in hardware using traditional or emerging electron devices within an integrated system. In the hardware implementation, the building of artificial spiking neurons is fundamental for constructing the whole system. However, with the slowing down of Moore’s Law,the traditional complementary metal-oxide-semiconductor(CM...     

Neuromorphic Synapses for Artificial Dendrites

We describe neuromorphic, variable-weight synapses onartificial dendrites that facilitate experimentation with correlativeadaptation rules. Attention is focused on those aspects of biologicalsynaptic function that could affect a neuromorphic network'scomputational power and adaptive capability. These include sublinearsummation, quantal synaptic noise, and independent adaptationof adjacent synapses. The diffusive nature of artificial dendritesadds considerable flexibility to the design of fast synapsesby allowing conductances to be enabled for short or for variablelengths of time. We present two complementary synapse designs:the shared conductance array and the self-timed synapse. Bothsynapse circuits behave as conductances to mimic biological synapsesand thus enable sublinear summation. The former achieves weightvariation by selecting different conductances from an on-chiparray, and the latter by modulating the length of time a constantconductance remains activated. Both work with our interchip communicationsystem, virtual wires. For the present purpose of comparing variousadaptation mechanisms in software, synaptic weights are storedoff chip. This simplifies the addition of quantal weight noiseand allows connections from different sources to the same dendriticcompartment to have independent weights.     

All-Optically Controlled Artificial Synapse Based on Full Oxides for Low-Power Visible Neural Network Computing

Artificial synapse devices are dedicated to overcoming the von Neumann bottleneck. Adopting light signals in visual information processing and computing is vital for developing next-generation artificial neuromorphic systems. A strategy to construct all-optically controlled artificial synaptic devices based on full oxides with amorphous ZnAlSnO/SnO heterojunction in a two-terminal planar configuration is proposed. All synaptic behaviors are operated in the visible optical pathway, with excitatory synapse under red (635 nm) light and inhibitory synapse under green (532 nm) and blue (405 nm) lights. Based on the different inhibitory effects, two modes of long-term depression (LTD) and RESET processes can be implemented through green and blue lights, respectively. The energy consumption of an event can be as low as 0.75 pJ. A three-layer perceptron model is designed to classify 28 × 28-pixel handwritten digital images and performed supervised learning using a backpropagation algorithm, demonstrating the bio-visually inspired neuromorphic computing with a training accuracy of 92.74%. The all-optically controlled artificial synapses with write/erasure behaviors in visible RGB region and rational microelectronic process, as presented in this work, are essential in developing future artificial neuromorphic systems and highlight the huge potential of next-generation computer systems.     

Novel Electronics for Flexible and Neuromorphic Computing

Emerging classes of flexible electronic systems that can be attached to a wide range of surfaces from wearable clothes to internal organs have driven significant advances in communication protocols (e.g., Internet of Things, augmented reality) and clinical research, shifting today's personal computing paradigm. The field of “system on plastic” is on the verge of an innovative breakthrough toward a hypercognitive society by being fused with current neuromorphic applications in the spotlight, which can offer intelligent services such as personalized feedback therapy and autonomous driving. The novel concept of electronics for flexible and neuromorphic computing requires an important research leap in micro‐/nanoelectronics on plastics, system‐level integration techniques (interconnection and packaging), and synaptic devices. Here, representative advances and developments in the area of flexible and neuromorphic technologies are reviewed with regard to device configurations, materials, fabrication processes, and their potential research fields.