For the conventional von Neumann based vision systems, the sensing, memory, and processing units are separated. Shuttling of redundant data between separated image sensing, memory, and processing units causes a high latency and energy consumption. To break these limitations, the next-generation neuromorphic visual systems, which integrate light information sensing, memory, and processing, can reduce the data transfer, thus improving their time and energy efficiencies. As the basis of the hardware-implementing of neuromorphic visual systems, optoelectronic artificial synapse devices have been extensively investigated in recent years. By integrating the functions of synaptic devices and light-sensing elements, the optoelectronic artificial synapse devices pave the way for constructing new neuromorphic vision systems with low latency, high energy efficiency and good reliability. Many materials are widely utilized for optoelectronic artificial synapse devices, and operation mechanisms of the present optoelectronic artificial synapse devices mainly include the ionization and dissociation of oxygen vacancy, the trapping/detrapping of photogenerated carriers, the light-induced phase change, and the interaction between light and ferroelectric materials. In this short review, the recent progresses in optoelectronic artificial synapse devices are introduced from the perspectives of their operation mechanisms. Besides, advantages and challenges of the devices are analyzed from the view of operation mechanisms. Finally, the advanced prospect and research aspect of optoelectronic artificial synapse devices are outlined for the application.

In recent years, inspired by the unique operation mode of the human brain, emulation of the perception and computing functions of synapses and neurons by artificial neuromorphic devices has attracted more and more attention. So far, many researches have been reported about neuromorphic transistors (NMT), but most devices are fabricated on rigid substrates. The flexible neuromorphic transistors can not only realize signal transmission and training learning at the same time, but also carry out nonlinear spatio-temporal integration and cooperative regulation of multiple signals. It can also closely fit the soft human skin and withstand the high physiological strain of organs and tissues. More importantly, flexible neuromorphic transistors have unique advantages and application potential in detecting low amplitude signals at physiologically relevant time scales in biological environments due to their designable flexibility and excellent biocompatibility. Flexible neuromorphic transistors have been widely used in electronic skin, artificial vision system, intelligent wearable system, and other fields. At present, it is one of the most important tasks to develop low-power consumption, high-density integrated flexible neuromorphic transistors. In this paper, the research progress of NMT based on different flexible substrates is reviewed. In addition, the bright application prospect of flexible neuromorphic transistors is prospected. This review provides a reference for the development and application of flexible neuromorphic transistors in the future.

Two-dimensional transition metal dichalcogenides are appealing materials for the preparation of nanoelectronic devices, and the development of memristors for information storage and neuromorphic computing using such materials is of particular interest. However, memristor arrays based on two-dimensional transition metal dichalcogenides are rarely reported due to low yield and high device-to-device variability. Herein, the 2D MoTe₂ film was prepared by the chemical vapor deposition method. Then the memristive devices based on 2D MoTe₂ film were fabricated through the polymethyl methacrylate transfer method and the lift-off process. The as-prepared MoTe₂ devices perform stable bipolar resistive switching, including superior retention characteristics (>500 s), fast switching (~60 ns for SET and ~280 ns for RESET), and excellent endurance (>2000 cycles). More importantly, the MoTe₂ devices exhibit high yield (96%), low cycle-to-cycle variability (6.6% for SET and 5.2% for RESET), and low device-to-device variability (19.9% for SET and 15.6% for RESET). In addition, a 3×3 memristor array with 1R scheme is successfully demonstrated based on 2D MoTe₂ film. And, high recognition accuracy (91.3%) is realized by simulation in the artificial neural network with the MoTe₂ devices working as synapses. It is found that the formation/rupture of metallic filaments is the dominating switching mechanism based on the investigations of the electron transport characteristics of high and low resistance states in the present MoTe₂ devices. This work demonstrates that large-scale two-dimensional transition metal dichalcogenides film is of great potential for future applications in neuromorphic computing.

The analog channel conductance modulation of electrolyte-gated transistors (EGTs) is a desirable property for the emulation of synaptic weight modulation and thus gives them great potential in neuromorphic computing systems. In this work, an all-solid-state electrochemical EGT was introduced with a low channel conductance (~120 nS) using amorphous Nb₂O₅ and Li-doped SiO₂ (Li_xSiO₂) as the channel and gate electrolyte materials, respectively. By adjusting the applied gate voltage pulse parameters, the reversable and nonvolatile modulation of channel conductance were achieved, which was ascribed to reversible intercalation/deintercalation of Li⁺ ions into/from the Nb₂O₅ lattice. Essential functionalities of synapses, such as the short-term plasticity (STP), long-term plasticity (LTP), and transformation from STP to LTP, were simulated successfully by conductive channel modulation of the EGTs. Based on these characteristics, a simple associative learning circuit was designed by parallel a resistor between the gate and the source terminals. The Pavlovian dog classical conditioning behavior was simulated based on associative learning circuit, where the resistor represented the unconditioned synapse and shared the gate voltage with EGT according to the proportion of its resistance, and the resistance between gate and source for negative feedback regulation of synaptic weights. These results demonstrate the potential of EGT for artificial synaptic devices and provide an insight into hardware implementation of neuromorphic computing systems.

Memristor, fusing the functions of storage and computing within a single device, is one of the core electronic components to solve the bottleneck of von Neumann architecture. With the unique volatile/non-volatile resistive switching characteristic, memristor can simulate the function of synapses/neurons in brain well. In addition, due to the compatibility with traditional complementary metal-oxide-semiconductor (CMOS) processes, metal-oxide-based memristors have received a lot of attention. In recent years, many kinds of metal-oxide memristors based on single dielectric layer have been proposed. However, there are still some problems such as the instability of switching voltage, fluctuation of high/low resistance state and poor endurance of memristive device. Thus, the researchers have successfully optimized the device performance by introducing the double dielectric layer into the metal-oxide memristors. In this article, we introduce the advantages of double dielectric layers-based metal-oxide memristors, and discuss their mechanism and design of double dielectric layers-based metal-oxide memristors. Eventually, we introduce their potential applications in neuromorphic computing. This review provides some enlightenment on how to design high-performance metal-oxide memristor based on double dielectric layers.

Indium-gallium-zinc-oxide (IGZO)-based electric-double-layer (EDL) transistors have great applications for neuromorphic perception and computing systems because of their low processing temperature, high homogeneity, and plentiful ionic dynamics. However, IGZO-based EDL transistors have problems of high leakage current (>10 nA), high energy consumption and abnormal current spikes, which are the main obstacles to the development of neuromorphic computing systems based on such devices. In this work, a novel IGZO neuromorphic transistor with Al₂O₃/chitosan stacked gate dielectric was proposed. Compared with the monolayer chitosan gate dielectric transistor, the device with Al₂O₃/chitosan layer showed low subthreshold swing of 78.3 mV/decade, a low gate leakage current of 1.3 nA (reduced by about 98%), a large hysteresis window of 3.73 V (increased by about 3.4 times), a low excitable postsynaptic current of 0.86 nA (decreased by about 97%) and an energy consumption of 1.7 pJ for a spike event (0.5 V, 20 ms). Additionally, the emulation of spiking synaptic function and the synergistically modulation of the channel current were also realized, and the abnormal current spike caused by high leakage in synaptic plasticity simulation was also effectively avoided. The results suggest that the inserting of high-k dielectric layer can effectively improve the leakage current, energy consumption and performance of neuromorphic devices, which has substantial value for future ultra-low energy consumption neuromorphic perception and computing systems.

The exploration of flexible electronic devices with information processing functions of biological neurons is of great significance for the development of intelligent wearable technologies. Due to lack of inherent mechanical flexibility, conventional threshold-switching memristor based on rigid materials that can implement the computing functions of biological neurons is difficult to fulfill the requirements for potential applications in the future. In this work, an intrinsically stretchable threshold-switching memristor was prepared by using silver nanowire-polyurethane composite as the dielectric layer and liquid metal as the electrodes, respectively. Under application of a sweeping voltage, the device exhibited reliable threshold switching characteristics, which was switched from the high resistance state (HRS) to the low resistance state (LRS) during device programming and spontaneously relaxed to the HRS upon voltage application. Further analysis shows that the underlying mechanism can be attributed to the dynamic formation and rupture of discontinuous silver conductive filaments formed between silver nanowires. In the pulse programming mode, memristor device is able to emulate the integration and firing characteristics of biological neurons, suggesting its great potential as an artificial neuron. Moreover, the pulse amplitude and pulse interval modulated neuronal spiking behaviors are successfully replicated using such devices. Under 20% tensile strain, the threshold-switching memristor shows negligible changes in the operating parameters during device switching and neuronal function implementations, suggesting its excellent mechanical flexibility and stability. This work provides important guidelines for the development of high-performance stretchable artificial neuronal devices and next-generation intelligent wearable systems.

As the basic and essential unit of neuromorphic computing system, artificial synaptic devices exhibit great potential in accelerating the high-performance parallel computation, artificial intelligence, and adaptive learning. Among them, electrolyte-gated synaptic transistors (EGSTs) have received increasing attention as the next generation neuromorphic devices owing to its controllable channel conductance. The devices exhibit the abilities of simulating the short-term plasticity (STP) and long-term plasticity (LTP) of the neural synapses. However, most of EGSTs exhibit short persistence for LTP and their channel conductance is difficult to be adjusted due to the rapid self-discharge of the electric double layer. In this work, the EGSTs based on water-induced In₂O₃ as the channel and chitosan as gate electrolyte were constructed and the O₂ plasma treatments were performed. The formation of traps on the channel surface is caused by the O₂ plasma treatments, which leads to capturing hydrogen ions at interface of the electrolyte/channel layer, and the device performance exhibits an enlarged hysteresis window, so as to regulate LTP of EGSTs. Biological synaptic functions, including excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), STP, and LTP, were mimicked by electrochemical doping and electrostatic coupling effects. Meanwhile, based on the experimentally verified potentiation/depression characteristics of the EGSTs, a three-layer artificial neural network is applied for handwritten digit recognition, and simulation tests can obtain high recognition accuracy of 94.7%. These results reveal that surface plasma treatment is one of the key technologies to affect the device performance, which has great potential in regulating synaptic function of EGSTs.

Mimicking of brain perceptual processing mode is of great importance for the design of bionic intelligent perceptual system. On the meantime, adopting functional materials with biocompatibility and biodegradability to construct environment-friendly neuromorphic devices is also an important aspect for synaptic electronics. Here, gelatin/carboxylated chitosan (GEL/C-CS) composite electrolyte film was adopted as gate dielectrics in oxide neuromorphic transistors. Synaptic plasticities, including excitory post synaptic current and paired pulse facilitation, were mimicked on the oxide neuromorphic transistor under different humidities. A quantitative processing method for tactile recognition of objects was proposed based on the spike number dependent synaptic plasticity. An artificial neural network was built in further. Recognition accuracy of MNIST handwritten digits is above 90%. Data from above evaluation show that the proposed GEL/C-CS gated neuromorphic device has a promising application potential in the design of bionic intelligent perceptual systems and brain inspired neuromorphic systems.

Brain-inspired neuromorphic computing refers to simulation of the structure and functionality of the human brain via the integration of electronic or photonic devices. Artificial synapses are the most abundant computation element in the brain-inspired system. Memristors are considered to be ideal devices for artificial synapse applications because of their high scalability and low power consumption. Based on Ohm’s law and Kirchhoff’s law, memristor crossbar arrays can perform parallel multiply-accumulate operations in situ, leading to analogue computing with greatly improved speed and energy efficiency. Oxides are most widely used in memristors due to the ease of fabrication and high compatibility with CMOS processes. This work reviews the research progress of oxide memristors for brain-inspired computing, mainly focusing on their resistance switching mechanisms, device structures and performances. These devices fall into three categories: electrical memristors, memristors controlled via both electrical and optical stimuli, and all-optically controlled memristors. The working mechanisms of electrical memristors are commonly related to microstructure change and Joule heat that are detrimental to device stability. The device performance can be improved by optimizing device structure and material composition. Tuning the device conductance with optical signals can avoid microstructure change and Joule heat as well as reducing energy consumption, thus making it possible to address the stability problem. In addition, optically controlled memristors can directly response to external light stimulus enabling integrated sensing-computing-memoring within single devices, which are expected to be used for developing next-generation vision sensors. Hence, the realization of all-optically controlled memristors opens a new window for research and applications of memristors.

Multistage pain perception is of great significance for surviving the outside harmful stimuli for organisms. In this work, using a sodium alginate biopolymer electrolyte as neurotransmitter layer, a 5×5 array of junctionless transistoris successfully fabricated for pain perception. The device operates well at low voltage (2 V) with a large current on-off ratio (>10⁴) and on-state current (>10 μA). This coplanar-gate array can not only emulate the important functions of synapses, such as excitatory postsynaptic current, paired-pulse facilitation, and dynamic filtering, but also successfully mimic pain-perception and sensitization abilities of the artificial nociceptor network. Furthermore, this work also successfully emulates the multistage spatio-temporal sensitization in the nociceptor network. Construction of this kind of network system provides a new way for the application of the next-generation neuromorphic brain-like system.

As a reversible, non-volatile, and resistive state mutation information storage and processing device, the resistive switching (RS) memory is expected to solve the inherent physical limitations of the traditional memory and von Neumann bottleneck, and has received widespread attention. Taking advantage of rapid carrier migration characteristics and excellent photoelectric conversion performance, halide perovskite optoelectronic RS memory devices present excellent resistive switching performance. In recent years, researches on storage and computing applications of the halide perovskite RS memory developed unprecedentedly; whereas, the working mechanisms of halide perovskite RS memory still remain unclear. This review analyzes the working mechanism of halide perovskite RS memory, compares the regulation characteristics of conduction filaments (CFs) and energy level matching (ELM), summarizes the constraints of various mechanisms, reveals the repeated formation and dissolution of CFs under light illumination and electric field, as well as Schottky barrier between the perovskite transfer layer and other layer, dominates the On/Off ratio, threshold (Set/Reset) voltage and performance stability of halide perovskite optoelectronic RS memory, and prospects the applications of halide perovskite RS memory in artificial intelligence bionic synapses, in-memory computing, and machine vision.

A memristor with analogue resistive switching (RS) memory behaviors could provide enough conductance states for high-efficiency neuromorphic computing because this type RS memory feature can avoid conductance clamping, steeply change, and computing invalidation. Simulating the behavior of biological synapses under stimulus pulse can better reveal the bionic characteristic mechanism of electronic devices and provide support for high performance neuromorphic computation. Synaptic paired-pulse facilitation (PPF) is an important characteristic of biological synapses, reflecting the facilitation and adaptation process under external stimuli, which is crucial to reveal the working mechanism of neurons. A memristor with the structure of the Ag/FeO_x/ITO was prepared by RF magnetron sputtering, which was designed by energy band engineering for the PPF demonstration. Experimental measurement of the electric properties illustrates that the developed memristor displays an excellent asymptotic nonlinear resistance switching behaviors, which is so called analogue RS memory behavior. Importantly, this developed memristor presents this analogue RS memory behavior during 3000 I-V sweepings, provides dissociable 16 conductance states that could be well maintained for 10⁴ s, illustrating that these available conductance states are nonvolatile. Based on the energy band structure and oxygen vacancy (V_O) defects, a physical mechanism, which involved trap sites softly filled by the injection electron, electron tunneling between the potential barrier built by the contact of Ag/FeO_x and FeO_x/ITO, and the V_O migration that accompanied a volatile feature to some extent, is proposed to comprehend the observed analogue RS memory behaviors. According to this mechanism, a typical PPF feature is obtained after modulating the voltage pulse width and amplitude. The observed analogue RS memory behaviors and PPF behaviors show a promising potential and advantage in neuromorphic computing.