The rapid development of information technologies such as the cloud computing, and artificial intelligence (AI) has accelerated humanity's progression toward an intelligent society^[1−3]. However, this progress has also led to an exponential increase in global data volumes, placing greater demands on the speed and efficiency of information processing. In addition, the existing computing systems face significant challenges, including both the breakdown of Moore’s Law and the von Neumann bottleneck, making it difficult to meet the computational requirements of an intelligent society^[4−9]. To address the severe mismatch between supply and demand, it is imperative to develop new devices and computing paradigms to drive the transformation of computing technology. Inspired by the human brain, neuromorphic computing technology, characterized by in-memory, parallel, and analog computing, offers a promising solution for significantly enhancing the data processing capabilities while reducing energy consumption^[10−12]. It presents a feasible approach for efficiently handling the vast amounts of data generated in an intelligent society.

Neuromorphic computing technology requires novel fundamental devices distinct from traditional CMOS transistors to fully harness its potential, enabling high throughput, energy efficiency, and area-efficient information processing^[13−16]. Electrolytes are ionically conductive but electronically insulating materials, providing a substantial capacitance compared to the conventional dielectrics of transistors. This high capacitance ensures effective channel modulation at low operating voltages, allowing them for the detection and amplification of weak signals^{[17, 18]}. Electrolyte-gated transistors (EGTs) represent a new class of devices that align with the characteristics of neuromorphic computing^{[19, 20]}. EGTs offer advantages such as decoupled read and write operations, rich dynamic properties, excellent channel modulation capabilities, and ultra-low power consumption, making them well-suited as foundational devices for neuromorphic computing. Neuromorphic computing systems built on such devices are expected to significantly improve the computational efficiency, with promising applications in the internet of things (IoT) and edge computing, where the energy-efficient computation is critical^[21−24].

Currently, most electrolyte-gated neuromorphic transistors are driven by electrical signals^[25]. However, compared to electrical signals, light offers several advantages, including ultra-high speed, broad bandwidth, and low crosstalk^[26−28]. As a result, the optoelectronic neuromorphic devices exhibit unique advantages in achieving the ultra-low power consumption and ultra-high-speed neuromorphic computing^[29−32]. Such devices typically require both light signal detection and neuron-like functionality. They combine photoreception and signal processing capabilities, holding a great potential for building intelligent perceptional systems^{[33, 34]}. In this review, the recent progresses of electrolyte-gated optoelectronic transistors are reviewed for neuromorphic applications. We primarily examine the operating mechanisms of EGTs, categorized the electrolytes, and introduced their cutting-edge applications. Firstly, the information processing units of biological neural systems are presented, and followed by a detailed introduction of EGTs and two different operational modes. Subsequently, as shown in Fig. 1, the main optoelectronic synapses and neuromorphic applications are discussed, focusing on the synaptic plasticity, spatiotemporal integration functions, and different artificial perception systems (vision, pain, and tactile sensation). Finally, the current challenges and future outlooks of neuromorphic EGTs are given in the last section of this review.

Through the intricate interconnections of tens of thousands of neurons and synapses, the brain's computational model exhibits greater flexibility and sophistication^{[35, 36]}. Fig. 2(a) presents a schematic diagram of a chemical synapse, comprising three distinct components: the presynaptic terminal, synaptic cleft, and postsynaptic terminal, respectively. The transmission and processing of information within chemical synapses are complex processes, which is typically initiated by the onset of an action potential^{[37, 38]}. The action potential induces the opening of voltage-gated calcium channels. Upon activation of voltage-gated calcium channels, the electrical signal activity in the presynaptic neuron is transduced into the release of neurotransmitters, subsequently altering the postsynaptic current or potential (Fig. 2(b)). The release of excitatory neurotransmitters, such as glutamate, induces excitatory postsynaptic currents/potentials (EPSC/EPSP), whereas the release of inhibitory neurotransmitters, such as GABA, results in inhibitory postsynaptic currents/potentials (IPSC/IPSP). The summation of EPSCs and IPSCs within the postsynaptic neuron determines whether an action potential will be generated. Therefore, the generation of EPSCs and IPSCs is pivotal for information processing, learning, and memory in the human brain^[39−41].

Synaptic behaviors can be divided into short-term and long-term synaptic plasticity, based on the complexity of synaptic weight connections within neural networks (Fig. 2(c)). The previously discussed excitatory and inhibitory postsynaptic potentials rapidly fluctuate in response to externally applied stimulus signals. Upon removal of the external stimulus, EPSCs and IPSCs rapidly dissipate, typically persisting for only 1 to 100 ms. This short-term response characteristic in biological synapses bears a strong resemblance to short-term memory in human brain models. Additional synaptic behaviors have also been reported, as depicted in Fig. 2(d), including paired-pulse facilitation (PPF), spike-timing-dependent plasticity (STDP), and spike-rate-dependent plasticity (SRDP), respectively. STDP and SRDP, both forms of long-term plasticity, entail extended response times and more intricate connections^[45]. These mechanisms are widely regarded as fundamental to the learning and memory processes of human brain.

In neuromorphic systems, various electronic components, including comparators and reset circuits, consume a substantial portion of the circuit area^{[46, 47]}. However, the human brain, comprising approximately 10¹¹ neurons, remains the ultimate model that the neuromorphic hardwares aspire to emulate. Recent years have witnessed the development of various artificial neuromorphic devices, including memristors, phase-change memory, magnetic tunnel junctions, ferroelectric field-effect transistors, and electric-double-layer (EDL) thin-film transistors^[48−50]. These devices are capable of generating spiking signals independently of complex circuit assistance. Owing to their diverse operating principles, each artificial neuron device adheres to its own operational scheme, contingent on whether its input and output are voltage or current. Given that biological neurons receive ions from synapses and generate membrane potentials in the form of spikes, artificial neuron devices are optimally suited to mimic this operational mode, specifically by receiving current input (I_in) and emitting voltage output (V_out).

As previously discussed, substantial advancements have been achieved in the development of artificial neural synapses and neuromorphic systems. Furthermore, electrolyte-gated synaptic transistors, leveraging ion-conductive electrolytes, have garnered significant attention from the research community. Ion-conductive electrolytes facilitate EDL modulation, allowing electrolyte-gated transistors to well emulate the dynamic functions of neural network system. The EDL effect would induce significant changes of channel conductance, which manifests as relaxation characteristics^[51]. Given that these ion behaviors can be readily employed for expanding into a multi-gate configuration, the millisecond-scale time frame closely aligns with the memory and forgetting processes of synapses, holding high promise in neuromorphic electronics.

In the 1950s, Bell Labs reported the electrolyte materials could effectively modulate the surface potential of semiconductors, thereby validating the concept of using electrolyte materials in transistors^[52]. The principal motivation for employing electrolyte materials as the gate dielectric layerlies in their large capacitance (>1 μF/cm²) from electrolytes. This capacitance is approximately 10 times greater than that of high-k dielectrics (e.g., Ta₂O₅) for a film thickness of around 100 nm and 5 times greater than that of ultrathin dielectrics (e.g., self-assembled monolayers)^[53]. The primary advantage of such large capacitance is that it enables transistors to operate at lower voltages while delivering substantially higher drive currents. The influence of capacitance on a transistor's operating voltage and drain current can be elucidated using the standard equation for drain current (I_D) within the linear region^[54]:

1ID=(W/L)μC(VG−VD)VD,

where W is the channel width, L is the channel length, and μ is the carrier mobility, respectively. From the above equation, it can be observed that a large capacitance would enhance the I_D, revealing that the electrolyte capacitance can improve the gate/channel coupling significantly. These advantages have led to the EGTs garnering widespread attention for different fields, including the flexible electronics, printed electronics, biochemical sensing, and neuromorphic computing. Fig. 3 illustrates two common device structures of EGTs^[55]. The gate can be expanded into multiple gates, thereby forming a multi-input device architecture. The electrolyte materials of EGTs include polymer electrolytes, ionic liquids, ion gels, and inorganic solid electrolytes, respectively. The ions of electrolyte would respond to an electric field with a migration behavior, which is very beneficial for emulating the ion information processing in biological neural system.

Electrolyte-gated neuromorphic photoelectric transistors utilized electrolytes to modulate the electrical properties and photoelectric response of the device, thereby mimicking the signal transmission and synaptic functions of biological neurons. These devices controlled the conductivity and current response in the semiconductor channel through the movement of ions in the electrolyte. When a gate voltage (V_g) was applied, ions within the electrolyte layer migrated toward the interface of the semiconductor channel. A positive voltage caused cations to move to the surface of the channel, while a negative voltage led anions to approach the surface. These ions formed an ionic layer at the interface of the semiconductor channel, resulting in a change in local potential, which in turn altered the conductivity of the channel. This phenomenon was akin to modifying the "doping" level of the channel, thereby regulating the output current of the device. Additionally, illumination of the channel by a light source excited and separated photo-generated carriers (electrons and holes). The quantity of photo-generated carriers was proportional to the intensity of the light. Since the ions in the electrolyte had already modulated the surface potential of the channel, the transport path of the photo-generated carriers was altered, consequently changing the photocurrent. This allowed the photoelectric transistor to modulate and amplify the light signal. The ion dynamics in EGTs exhibits the considerable complexity. The operational mechanisms can be classified into electrostatic control and electrochemical doping, respectively.

In the EDL model, it can be simplified into two compact layers with opposite charges, as illustrated in Fig. 4(a). Subsequently, Gouy−Chapman model was developed by introducing a diffuse layer which accounts for the thermal and electrical motion of ions in EDL due to ion aggregation^[56]. Stern further advanced this model by incorporating an inner region (called as the Stern layer)^{[57, 58]}. The Stern model employs the inner Helmholtz plane (IHP) and the outer Helmholtz plane (OHP) to delineate two distinct types of ion accumulation on the electrolyte surface. This model, known as the Gouy−Chapman−Stern model, has been extensively employed for the applications of EGTs.

When the semiconductor channel of EGTs permits the ion penetration from electrolyte, the ions would move into semiconductor channel, which is called as electrochemical process (i.e. doping). In 2014, the phenomenon of electrochemical doping was first experimentally validated in a transistor that utilized ionic liquids as the gate dielectric material and samarium nickel oxide (SmNiO₃) as the semiconductor channel. This device demonstrated substantial advantages over traditional transistor devices, particularly in the linear variation of conductivity. Typical pairing of electrolyte and channel material includes the perylenetetracarboxylic dianhydride (PTCDA)/MoS₂ (Fig. 4(b))^[59], modified lithium perchlorate (LiClO₄)/MoO₃ (Fig. 4(c))^[60], respectively.

The high capacitance characteristic of EDL allows the EGTs to demonstrate exceptional electrical performance. To date, various electrolytes have been developed, including the ionic liquids, ion gels, polymer electrolytes, and other inorganic solid electrolytes.

Ionic liquids constitute a class of molten salts consisting exclusively of cations and anions, characterized by melting points below 100 °C, such as [EMIM-TFSI] and [BMIM]PF₆ (Fig. 5(a))^{[61, 62]}. Gajar et al. first employed ionic liquids to construct an EGT in 1992^[63]. This breakthrough demonstrated that ionic liquids could supplant traditional SiO₂ dielectrics, thereby enhancing the long-field modulation capability of transistors.

As illustrated in Fig. 5(b), an aromatic ring molecular nano-transistor has been reported with a graphene-molecule−graphene single-molecule junction^[64]. The incorporation of an ionic liquid effectively mitigates charge transport challenges in single-molecule transistors. This ionic liquid can precisely modulate the alignment between the frontier orbitals of molecular and the Fermi level of graphene, thereby tuning well the charge transport of junction. Moreover, bipolar charge transport was realized in these electrochemically inactive molecular systems. The work lays a crucial foundation for the realization of controllable single-molecule electronic devices and paves the way for further exploration of quantum transport and novel physical phenomena in low-temperature single molecules.

Artificial intelligence has achieved groundbreaking advancements in numerous fields, including image recognition, speech recognition, and natural language processing. Zhu et al. explored a neuromorphic application utilizing electrolyte-gated In₂O₃ material through the study of flexible synaptic transistors (Fig. 5(c))^[65]. The flexible transistors were fabricated using solution-processing techniques. These flexible transistors exhibit low power consumption, high performance, and exceptional bendability. Through ion migration and relaxation dynamics in the electrolyte, they successfully simulated various critical synaptic behaviors, including excitatory postsynaptic currents, paired-pulse facilitation, high-pass filtering characteristics, and long-term memory performance. This research offers novel insights and methodologies for advancing the fields of neuromorphology and artificial intelligence.

Moreover, ion-liquid-gated field-effect transistors (FETs) utilizing transition metal dichalcogenides (TMDs) serve as an advanced platform for probing the physical phenomena of band filling and charge carrier accumulation within these systems (Fig. 5(d))^[66]. Through quantitative analysis of both longitudinal and transverse conductivities, the intricate process of carrier accumulation in MoSe₂ and WSe₂ monolayers was meticulously examined. The precise conditions under which the chemical potential penetrates distinct valleys in the monolayer band structure—specifically, the K and Q valleys in the conduction band and the spin-split K valleys in the valence band—were identified, revealing significant non-monotonic behavior in both longitudinal conductivity and Hall slope.

As depicted in Fig. 5(e), the MoS₂-based electric-double-layer transistor (EDLT) regulated by DEME-TFSI ionic liquid exhibited the distinct bipolar transfer characteristics under ion gate modulation, conclusively demonstrating that the ion gate doping effect is robust enough to induce a PN-type transition in MoS₂. By precisely configuring the voltages applied to the drain and ion gate, an upward electric field is established between the channel near the drain and the gate, prompting anions in the ionic liquid to accumulate on the channel material's surface, thereby injecting holes^[67]. Conversely, a downward electric field forms between the channel near the source and the gate, leading to the accumulation of cations on the channel material's surface, thereby injecting electrons and forming an in-plane homojunction PN junction. While ionic liquids provide substantial benefits in enhancing carrier mobility and reducing subthreshold swing in transistors, they also present several challenges, including the potential for redox reactions with electrodes and slow ion migration. Furthermore, given that ionic liquids are inherently in liquid form, they pose significant challenges for device integration, thus imposing numerous limitations on their practical applications. Consequently, researchers have shifted towards employing solid-state ion gels as alternatives to ionic liquids, which not only retains the potent control effects of double-layer capacitance but also significantly enhances the ease of device integration.

Ion gels are solid mixtures with ionic conductivity, functioning as gel polymer electrolytes composed of salts and organic polymers^[68]. These materials typically consist of lithium salts, such as Li₂N, LiClO₄, LiPF₆, and LiAsF₆, combined with polymer matrices like PEO, PVDF, and PMMA. One of the primary advantages of PEO is its high solvation capacity, enabling it to form complexes with various alkali salts. The (−CH₂−CH₂O−)_n units in the polymer backbone facilitate cation migration, further enhancing its conductivity.

Compared to ionic liquids, one significant advantage of ion gels was their ease of integration into devices. Traditional methods for integrating ion gels with devices included the drop-casting method and the spin-coating method. The drop-casting method involved directly depositing the electrolyte onto the device, followed by heating to dry the solvent, thus forming a thin electrolyte film. While this approach was straightforward, the resulting thickness was difficult to control. Compared to drop-casting, spin-coating not only achieved a more uniform film but also allowed for precise control of film thickness. Zan et al. employed both methods to prepare PEO/LiClO₄ films^[69]. As shown in Fig. 6(a), the electrolyte layer produced via spin-coating was visibly more uniform and exhibited better film-forming properties. This study further demonstrated the relationship between electrolyte film thickness and spin-coating speed, as well as it between the specific capacitance and frequency, as illustrated in Fig. 6(b). Notably, the PEO not only could be integrated using the above methods but also had the unique capability of being combined with electron-beam lithography (EBL). This allowed for the patterning of ion gels, enabling more precise device integration.

Xu et al. introduced a PEO/CsClO₄ ion gel into a 2H-MoTe₂ transistor structure, as illustrated in Fig. 6(c). Fig. 6(d) presented the transfer characteristic curves of the ion gate both before and after the introducing of PEO/CsClO₄^[70]. Wu et al. utilized the PEO/LiClO₄ ion gel to form a p-type region, achieving an in-plane homojunction PN device, as shown in Figs. 6(e)−6(g)^[71]. These findings provided critical experimental evidence for the design and optimization of 2D electronic devices based on the ion gel as gate dielectrics.

Polymer electrolytes are a class of electrolytes by dissolving inorganic salts into a polymer matrix^[72]. This requires specific interactions between the polymer chains and the salts to ensure that the salts dissolve uniformly within the polymer. The most representative type of polymer electrolyte is the poly (ethylene oxide) (PEO)/AClO₄ system, where A represents the cations such as Li or K. As shown in Fig. 7(a), the Frisbie’s group utilized PEO/LiClO₄ polymer electrolyte as the gate dielectric and poly(3-hexylthiophene) (P₃HT) as the channel layer, respectively^[73]. Due to such the atomic-level flatness and good chemical compatibility at electrolyte/channel interface, the field-effect mobility was significantly enhanced.

Bio-polymer electrolyte is another kind of electrolyte materials based on biological polymers. These materials exhibit excellent biocompatibility and biodegradability, making them suitable for the applications of bioelectronics and biosensors. Li et al. successfully achieved a simple ionic-electronic junctionless oxide transistor array with pain perception capabilities by utilizing the coplanar proton-coupling bio-polymer electrolyte of sodium alginate, as illustrated in Fig. 7(b)^[74]. They further developed an artificial nociceptor network, allowing it to mimic the transmission and modulation of pain signals, as depicted in Fig. 7(c). Liu et al. demonstrated the biodegradable synaptic oxide transistors^[75]. By using a chitosan as the bio-polymer gate dielectric, they successfully fabricated a biodegradable synaptic transistor on a graphene-coated polyester film, as shown in Fig. 7(d). This transistor exhibited excellent performance, including the low operating voltage, high field-effect mobility, high on/off ratio, and low subthreshold swing. Different synaptic functions such as EPSC, PPF, and synaptic filtering were successfully emulated, as illustrated in Fig. 7(e). Moreover, the device could dissolve in water within a short period, demonstrating potential for applications in biodegradable electronics. As shown in Fig. 7(f), Gou et al. constructed a novel artificial synapse using the coupling of a biopolymer electrolyte and a SnO₂ nanowire transistor^[76]. This device can adjust the synaptic behavior by multiport regulation of synaptic inputs, and in addition, a neural network was constructed. The design and functional simulation of artificial synapse provides important technical support for the development of novel neuromorphic computing systems.

Inorganic electrolytes provide superior chemical stability while maintaining comparable conductivity^{[77, 78]}. Inorganic materials, including porous SiO₂, Al₂O₃, silicates, zeolites, and copolymer sodium alginate, have been utilized as gate electrolytes in EGTs^[79]. As shown in Fig. 8(a), an oxide transistor utilizing yttria-stabilized hafnia (YSH) solid-state electrolyte is presented. It exhibited the enhanced retention characteristics for artificial synapse applications^[80]. Channel conductance was modulated by precisely tuning the proton-electron coupling strength through the input signal, as illustrated in Fig. 8(b).

To tackle the critical challenges of reliability and variability of 2D semiconductor device, Park et al. proposed a robust 2D artificial synaptic transistor with a solid-state lithium silicate electrolyte film as the gate dielectric, as shown in Fig. 8(c)^[81]. The device exhibited good low cycle variation during long-term potentiation and depression, resulting in the statistical simulation of discrete states, as shown in Fig. 8(d)). Furthermore, as illustrated in Fig. 8(e), a SrFeO_2.5 transistor with solid-state electrolyte dielectric can control the insertion and extraction of oxygen ions, achieving a non-volatile conductivity switching^[82]. This device can provide a novel approach for critical components in future neuromorphic circuits.

In neuromorphic EGTs, the channel corresponds to the postsynaptic terminal, while the channel conductivity represents the synaptic weight. The volatile and non-volatile conductance can be used to emulate short-term plasticity and long-term plasticity of synapses, respectively. Fundamental synaptic behaviors, such as EPSC/IPSC, PPF/PPD, high-pass/low-pass filtering, STDP/SRDP, and higher-order synaptic hyperplasticity, can be all realized in EGTs^[83−85].

As shown in Fig. 9(a), Gkoupidenis et al. developed a neuromorphic organic electrochemical transistor (OECT) based on PEDOT as channel and KCl aqueous electrolyte as gate dielectric, respectively^[86]. The PEDOT channel layer was reversibly doped back to its initial high-conductivity state as shown in Figs. 9(b) and 9(c). Both STP and STD of plasticities could be simultaneously achieved in this synaptic device. STDP plays a critical role in learning and memory, determining whether synaptic weight undergoes LTP or LTD, and thus regulating the strength of synaptic connections.

As shown in Figs. 9(d) and 9(e), STDP behavior can be achieved in chitosan-gated ITO EGTs. STM can transition into LTM. This behavior can be realized in synaptic ITO EGTs proposed by Yu et al.^[87]. As illustrated in Fig. 9(f), when the pulse interval gradually decreased, the EPSC significantly increased due to the facilitation effect, resulting in a typical STM behavior. As shown in Fig. 9(g), the EPSC responseclosely resembled the process of transitioning from STM to LTM. Synaptic hyperplasticity behavior has been extensively reported in two-terminal memristors, whereas the work is rather limited for transistor devices. As shown in Figs. 9(h)−9(j), John et al. successfully demonstrated the regulation of STDP behavior as well as LTP and LTD through hyperplasticity in 2D MoS₂ EGTs^[88]. Unlike memristor devices, this device employed a multi-gate ion-coupled structure to achieve multiple operating modes, including the electrical, ionic, and optical signals. The synaptic weight could be maintained within a dynamic range, thereby enabling hyperplasticity to regulate synaptic plasticity.

The brain's processing of information is closely related to both temporal and spatial dimensions. In neural networks, neurons can non-linearly integrate input signals that reach their dendrites, which have complex structures and functions^[89−93]. Dendritic integration is a crucial information-processing function in neural networks, including the integration of events that occur at different times (temporal integration) and the integration of events that occur simultaneously at different regions of the dendrites (spatial integration). Most neural information processing is not limited to purely spatial or temporal integration. For example, directional selectivity depends on the spatiotemporal integration of action potentials generated in the sensory layer.

Orientation selectivity is a common phenomenon in the primary visual cortex. Gkoupidenis et al. demonstrated this function using an array of PEDOT OECTs with a 3 × 3 configuration of coplanar gold electrodes (Fig. 10(a))^[94]. In this coplanar multi-gate structure, the closer the distance between gate and drain is, the stronger the channel current becomes (Fig. 10(b)). This spatial inhomogeneity could be exploited to achieve orientation selectivity. Fig. 10(c) shows a polar plot of the EPSCs for various spatial orientation input pulses, closely resembling the orientation tuning curves observed in the primary visual cortex. Inspired by this, Wan et al. developed a neuromorphic system consisting of a photodetector and synaptic EGTs to simulate edge detection, as shown in Fig. 10(d)^[96]. This integration effect could switch from sublinear to superlinear integration, providing a versatile approach to neuromorphic computing.

Accurate sound localization in biological nervous systems plays a crucial role in information exchange, foraging, and predator avoidance. The brain achieves sound localization by detecting the time difference (Δt) when sound signals reach each ear. He et al. developed an artificial neural network based on chitosan-gated EGTs to emulate the spatiotemporal sound localization function, as shown in Fig. 10(e)^[98]. This result demonstrates that an artificial neural network built from EGT array can mimic the sound localization function of biological brain. Pavlovian conditioning is a type of associative learning process. This coordinated learning process can be emulated by using the highly interconnected lateral multi-terminal EGTs. Fu et al. developed a device based on solid-state ion gel that could achieve classical conditioning using just two gates, as shown in Fig. 10(f)^[99]. By emulating the neural network of the human brain, a device with remarkable flexibility and electrical stability was developed. Significant synaptic functions, including STDP and enhancement/inhibition, were successfully replicated, and the Pavlov learning rule was simulated by a multi-gated device. The findings of this research demonstrate that this adaptable neuromorphic device offers an approach to implementing flexible neuromorphic systems, such as electronic skin, human-computer interfaces, and soft robots.

The human body possesses five primary senses, including the vision, touch, hearing, smell, and taste, that furnish the brain with a wealth of information. Inspired from human sensory system, the artificial perception systems endow the next-generation photoelectronic devices with unprecedented vitality^{[100, 101]}. The EGTs have the significant advantages in sensing and actuation due to their strong resemblance to neural cells, indicating that this kind of devices is a good candidate for the emerging neuromorphic electronics.

The human eye constitutes a sophisticated visual apparatus through capturing light stimuli across a specific wavelength spectrum from the external environment. Artificial visual perception systems are capable of realizing the neuromorphic functions such as recognition, learning, and memory, respectively^{[102, 103]}. These are particularly coveted in the development of next-generation robots and humanoid sensory platforms.

As illustrated in Fig. 11(a), Jin et al. introduced an innovative optoelectronic In₂O₃ transistor array for emulating the artificial visual adaptability^[104]. This work tackles the challenge of adaptability for light intensities within artificial visual systems. By modulating gate bias and illumination conditions, the transistor array effectively facilitates visual information processing and adaptive responses (Fig. 11(b)). Fig. 11(c) depicts the continuous variation of Δ_IN corresponding to the "X" and "Y" coordinates within the device array. The In₂O₃ transistor exhibited a negative photoconductive characteristic which plays an import role in the emulation of natural adaptation behaviors of human visual system. This work represents a great step for developing an artificial visual adaptive system to diverse light environments.

Xiong et al. reported the synaptic emulation with artificial visual systems based on HfS₂-based semiconductor material^[105]. The devices can successfully emulate the LTP and LTD by varying light conditions, as shown in Figs. 11(d) and 11(e). Furthermore, this study further constructed an artificial neural network for MNIST handwritten digit recognition, attaining an accuracy rate of 88.5%. Furthermore, the circularly polarized light can facilitate the recognition of intricate patterns and chiral signals, resulting in a sophisticated information processing capability for artificial visual system. By introducing a cellulose nanocrystal (CNC) composite layer into the device, a cutting-edge visual system was developed with chiral light detection capabilities. The synaptic array is adept at storing and recognizing images with 4 × 8 pixel units, as shown in Fig. 11(f)^[43]. The system is capable of achieving a great intelligent chirality recognition.

Except the visual system, the human body is also equipped with a multitude of sensory systems with pain and tactile perception abilities. Feng et al. reported a neuromorphic device based on a sub-10 nanometer vertical ITO transistor, utilizing a biopolymer electrolyte as the gate dielectric, as shown in Figs. 12(a) and 12(b)^[106]. The device manifests several neuromorphic functions were emulated for nociceptors, including the pain thresholds, memory of prior injuries, and pain sensitization/desensitization. The device not only exhibits the low-voltage operation but also the pain sensitization characteristics.

Besides, Feng et al. also introduced a vertical multi-terminal flexible transient transistor network with an ultra-short channel, as shown in Fig. 12(c)^[107]. The optical transition from STM to LTM is attributed to the photo-gating effect of the device. It achieves pain sensitization using the spatiotemporal color pair mode. This device plays an important role in different domains such as intelligent robotics and biomedical engineering, enhancing the sensitivity and adaptability of the future intelligent machines.

As a fundamental component of human sensory system, the tactile perception serves as the earliest developed, extensively distributed, and intricately complex system, and is adept at converting external stimuli into internal sensations. Tactile perception hinges on the intricate interplay of perceptual processing, sensory refinement, and experiential learning, which profoundly influences our interactions with external environment^[110−112]. Consequently, artificial tactile systems is of great importance for the future applications of intelligent robotics and prosthetics. In recent years, significant progresses have been made in integrating diverse tactile sensors with neuromorphic devices to realize the intelligent tactile perception systems.

Yu et al. reported an artificial tactile receptor using a dual-mode electrolyte-gated synaptic transistor, as shown in Fig. 12(d)^[108]. The device operation is based on two dynamic ion processes (EDL and electrochemical doping), corresponding to volatile and non-volatile behaviors, respectively. The artificial tactile receptor can also realize the pain perception of nervous system. Dong et al. demonstrated a 2D/1D BP-C/CNT heterostructure for artificial photo-olfactory co-sensing synapse, as shown in Fig. 12(e)^[109]. The optical transition from STM to LTM was accomplished, and the dynamic behavior of EPSC was observed, as shown in Fig. 12(f). These devices exhibited significant optoelectronic response and excellent stability, resulting in good STP and LTP characteristics. Moreover, the devices are capable of achieving multisensory integration, holding a high promise for the advanced intelligent sensing applications.

Artificial synapses serve as foundational components in the development of neuromorphic chips, realizing the low-power and efficient neuromorphic computing. The current research on neuromorphic devices has largely focused on electrically-controlled devices. However, at present, the energy power of circuit is still far away from the human brain, which acts as a big obstacle for the future electronic applications. Emerging optoelectronic neuromorphic devices, combining both the photonics and electronics, can offer significant advantages in minimizing energy consumption. This paper reviews the recent advancements in electrolyte-gated neuromorphic optoelectronic devices from the standpoint of operation modes, material systems, and underlying mechanisms, respectively. Although significant progresses have been made, several challenges still remain to be overcomed.

Firstly, the electronic materials still have some issue. (ⅰ) Due to their bandgap limitations, oxide semiconductors normally exhibit a good response to ultraviolet light but respond poorly in the visible and infrared wavelength ranges. Although the 2D materials have an enhanced light response, their large-area stability and reproducibility issues still limit the future electronic applications. (ⅱ) For the point of electrolytes, hazardous organic solvent and process stability still need to be well tackled.

Next is the issues of device array. In the fabrication of transistor arrays, electrolytes have demonstrated both significant advantages and notable challenges. The benefits include low-power regulation, the ability to mimic synaptic behavior, dual control over electrical and optical signals, simple structural design, and suitability for flexible electronics and biocompatible applications. However, several drawbacks of electrolytes must be addressed, such as the slow migration speed of ions, issues with chemical and environmental stability, response lag, ion accumulation effects, and inconsistency in the manufacturing process. Apart from that, several factors still limit the integration of device array.

(1) Device stability. The large-scale integration of optoelectronic neuromorphic devices require excellent uniformity and stability. Currently, the development of optoelectronic neuromorphic devices is still in its early stage with immature fabrication processes. Most of EGTs are susceptible to the influence of water and oxygen in the air, and the metal electrodes may experience corrosion from electrolytes, which may significantly degrade the device performance. Moreover, the organic electrolyte is normally un-compatible with the traditional microelectronic process, such as photolithography and etching. Therefore, the further improvement of process stability and compatibility is very important.

(2) All-optical control. Ideally, optoelectronic neuromorphic devices should possess the ability for all-optical control, meaning that they can directly utilize external optical signals to realize the functional emulations. This would greatly simplify the operational procedures and further reduce its energy consumption. The development of all-optical control devices need both the positive and negative photo-conductance materials. However, the realization of negative photo-conductance materials still keeps to be a great challenge. More importantly, it is hard to realize both the positive and negative photo-conductance phenomenon in a single device.

In addition to the aforementioned factors, the relatively slow ion migration speed also limits the performance of EGTs in high-frequency and rapid-response applications. Ion migration is influenced by factors such as the type of electrolyte, viscosity, and temperature, leading to response delays and reduced transient performance. To address this issue, several strategies could be implemented. First, selecting materials with high ionic conductivity and reducing electrolyte viscosity or developing solid-state electrolytes would have improved ion migration speed and stability. Second, reducing the thickness of the electrolyte layer would have shortened the ion migration path. Optimizing semiconductor materials, such as two-dimensional materials or perovskites, and refining the structural design could have enhanced carrier mobility, thereby accelerating charge transport. Additionally, improving electrode design by using highly conductive materials could have reduced resistance and capacitance effects, thus increasing response efficiency. The introduction of multi-gate structures could have enabled simultaneous control over ion migration and photogenerated carrier transport, further enhancing device speed. Lastly, employing self-assembly and nanopatterning techniques to control ion distribution could have minimized migration barriers, effectively increasing the response rate. With the development of novel electrolyte materials and the integration of other neuromorphic device technologies, EGTs are expected to play a significant role in high-density neural networks, low-power computing, and flexible electronics. While the speed limitation remains a challenge, the inherent advantages of these devices, along with anticipated improvements in performance through technological advancements, position EGTs as highly promising in the field of neuromorphic computing.

Finally, it is the aspect of functional application. Although these devices have found some preliminary applications, their functional capabilities remain relatively simple, focusing on the basic tasks like image detection, preprocessing, and memory. To achieve the real application, more complexity software algorithms and circuit architectures need to be developed in the future research. EGTs have demonstrated unique advantages in neuromorphic applications, particularly in simulating neuronal and synaptic behavior, low-power operation, and multimodal information processing. By regulating conductivity through ion migration, EGTs can achieve synaptic plasticity, making them well-suited for brain-inspired computing and neural network learning and memory. Additionally, EGTs operate with low energy consumption and can retain their state, making them highly suitable for large-scale integration and portable devices. Their dual control capability allows for the simultaneous response to electrical and optical signals, enabling more accurate simulation of biological multimodal information processing. EGTs also feature simple structures, low manufacturing costs, and flexible material choices, facilitating large-scale integration and three-dimensional architecture design. Furthermore, EGTs are ideal for flexible electronics and wearable devices due to their biocompatibility, making them highly suitable for brain−machine interfaces.

In conclusion, as intelligent technology rapidly progresses in human society, the optoelectronic neuromorphic devices present both tremendous development opportunities and significant challenges. The development of high-performance optoelectronic neuromorphic devices based on EGTs is very important for the next-generation of intelligent photoelectric device integration.