The rapid development of humanoid robotics and artificial intelligence has led to an exponential increase in data generation, placing significant demands on current information processing systems. However, most traditional computing systems rely on the von Neumann architecture, which separates memory and processing units^[1]. This architecture requires continuous data transmission between the two units during computation, causing high latency, signal distortion, and reduced operational efficiency^[2]. Furthermore, the performance of system is constrained by limited bus bandwidth and high power consumption^[3]. While Moore's Law suggests that further miniaturization could enhance computational performance^[4], this trend has also introduced challenges, such as electronic drift and signal delays. Consequently, there is an urgent need for novel devices and architectures capable of addressing the increasing demands for efficient data processing.

The human brain, a complex information-processing system comprising hundreds of billions of neurons, performs both memory and processing activities simultaneously^{[5, 6]}. Synapses, acting as linking units between neurons, allow for the rapid and energy-efficient transmission of electrophysiological signals via the migration, reception, and re-release of ions and neurotransmitters^[7]. These synapses are critical components for implementing distributed parallel computing, which enhances processing speed and accuracy. Inspired by the brain design, Carver Mead coined the term "neuromorphic" in the 1980s to achieve high-speed, low-power data processing by emulating the information-processing mechanisms of brain^[8]. Unlike traditional computing architectures, neuromorphic devices incorporate memory and processor units, allowing for efficient computation of large-scale data, resulting in improved transmission efficiency and energy utilization^[9].

Ions provide superior controllability compared to typical electronic transmission devices^[10]. Higher precision and storage densities are achieved by regulating the spatial distribution and charge states of ions^[11]. They can move at lower voltages and powers, resulting in significant energy savings and longer device lifespans^[12]. Furthermore, ionic migration-based devices can perform both storage and logic operations, simplifying circuit design, improving system integration, and enabling multifunctionality in information devices^[13]. Artificial biomimetic neuromorphic devices based on ion migration have been intensively researched and found to imitate biological neuron functions such as learning, memory, and reasoning, indicating substantial potential in information processing applications^[14]. For example, chitosan-gated oxide neuromorphic transistors on polyimide substrates successfully simulated four types of spike-timing-dependent plasticity (STDP) learning rules^[15]. Additionally, flexible triboelectric nano-generators (TENGs) have been used to simulate biological tactile sensing systems, simulating muscle functions such as muscle contraction, fast/slow muscle fiber shift, conscious/unconscious muscle movements, as well as hypersensitivity/allodynia and recovery^[16].

Although numerous developments in neuromorphic devices have been made, most research works predominantly rely on electrical signals for ion control^[17−19]. Due to limitations in bandwidth, density, and connectivity, the devices triggered by electrical signals exhibit single-mode control and signal crosstalk, which hinders the practical applications of neuromorphic technology^[20]. In contrast, optical signals have wide bandwidth, high energy efficiency, and low latency, which enhances capabilities related to large-scale parallel interconnection^{[21, 22]}. In 2001, researchers have discovered that the lasers with specific wavelengths could be used to achieve ion exchange and doping^[23]. For example, 248 nm excimer laser radiation was used to shift silver ions towards the exposed upper surface, forming a high silver concentration region approximately 3 μm thick^[24]. Furthermore, 785 nm femtosecond laser facilitated the migration and reduction of Sm³⁺ ions at the interface between Sm-doped and Cr-doped sodium borate^[25]. With the advent of novel device structures like memristors, the light can significantly affect operating voltages and switching speeds by reducing the activation energy required for Ag ion migration in Ag/AgS₂/Au memristors^[26]. While previous studies have demonstrated ion migration through light exposure, they primarily focused on ion behaviors and device characteristics. In contrast, the developments in optogenetics—a field that use specific wavelengths and intensities of light to control the activation of proteins and regulate neuronal activity—have introduced new opportunities for neuromorphic computing^[27−29]. Optoelectronic neuromorphic devices can achieve multidimensional control by applying optogenetic techniques, combining the advantages of both optical and electrical signals to enable efficient information processing, reduce energy consumption, and accelerate machine learning capabilities. They have been fabricated with various photosensitive materials with ionic migration, such as halide perovskites^[30−34]. Low migration energy promotes ion movements within these materials. Besides, metal oxide semiconductors can easily capture and release charges under illumination, modulating interfacial barriers to mimic artificial synapses^[35−38]. These properties demonstrate that they are promising candidates for applications in artificial visual systems, as well as in simulating associative and non-associative learning simulations.

This review presents the recent advancements in ion-modulation optoelectronic neuromorphic devices, focusing on mechanisms, materials, characteristics, and applications, as shown in Fig. 1. It first elucidates two primary mechanisms: ionic migration control and capture and release of charge through ions. Representative materials and their properties based on different mechanisms are further discussed. Moreover, the applications of the devices in artificial visual systems, neuromorphic computing, and optogenetics are introduced. Finally, the current challenges faced by ion-modulation optoelectronic neuromorphic devices are outlined, offering critical insights for the future development of high-performance systems in this field.

Electric fields can drive ions to migrate through various mechanisms, such as the formation of conductive filaments, charge trapping and detrapping, and the electric double-layer effect^[46−49]. With the continued advancement of optoelectronic neuromorphic devices, extensive research has explored precise ion motion behavior through optical fields^{[50, 51]}. This Section introduces the mechanisms by which optical fields influence ionic migration and facilitate the capture and release of charge in optoelectronic neuromorphic devices.

Light signals have been shown to induce ionic migration in materials such as perovskites, which has been identified as a key factor contributing to reducing efficiency and instability in solar cells^[54−56]. Moreover, researchers have successfully achieved active ion transport in nano-channels by synergistic photo-electric/thermal effects^[57]. Controlling ionic migration via optical fields offers a promising approach to regulating the electrical performance of optoelectronic neuromorphic devices.

Optical signals can directly influence ion migration by modulating potential barriers. Wang et al. fabricated Ag/CH₃NH₃PbI₃/ITO synaptic devices and explored the mechanism of conductive switching filaments formed by iodine vacancies under both illuminated and dark conditions^[39]. As illustrated in Fig. 2(a), light would create a photogenerated electric field (E_ph) from top to bottom electrode when irradiating the vertical type two-terminal device. It was the same with the direction of the external electric field when a positive bias was applied to the silver electrode. The photogenerated electric field enhanced the driving force for iodine vacancy migration, reducing the activation energy compared to conditions without optical signals. Inspired by optogenetics, Lu et al. developed a memristor based on Ag/CH₃NH₃PbI₃/Ag, suggesting that light enhanced the formation energy of iodine vacancies in CH₃NH₃PbI₃. This mechanism suppressed the electrically induced formation of iodine vacancies while promoting their spontaneous annihilation^[58]. Furthermore, Guo et al. demonstrated active ion transport with a rate approximately five orders of magnitude higher than classical diffusion by applying asymmetric light illumination to layered graphene oxide membranes (GOM)^[59]. The photogenerated electrons and holes in the illuminated regions, driven by their concentration gradient, diffused towards the non-illuminated regions, which generated an electric potential difference across the graphene oxide that facilitated ionic transport.

In addition, light signals can also promote ionic migration indirectly through various means such as phonon excitations and thermal effects. Villegas et al. used a micrometric junction between the high-temperature oxide superconductor YBa₂Cu₃O_7−δ (YBCO) and transparent semiconductor (ITO) to construct a photo-memristor^[52], as shown in Fig. 2(b). When the junction was set in the OFF state, the photovoltage (V_OC) generated under illumination facilitated the migration of oxygen ions. In contrast, oxygen ion migration back from the YBCO interface to the ITO was thermally activated in the dark when the junction was set in the ON state. They concluded that illumination activated ionic diffusion due to phonon excitations caused by light absorption. Furthermore, Hao et al. fabricated an optically modulated ion-conductive (OMIC) hydrogel composed of Fe₃O₄ nanoparticles, polyacrylamide (PAAm) networks, and azo-benzene functionalized imidazole (AZIM) salt^[45]. Under near-infrared light stimulation, Fe₃O₄ nanoparticles converted light stimulation into heat, thus controlling the thermal-responsive reversible assembly process of AZIM ions. When the aggregated supramolecular disassemble, the released AZIM ions could act as charge carriers to modulate the conductivity of the OMIC hydrogel, thereby achieving light-controlled conductivity.

In addition to controlling ion migration, optical signals also facilitate the capture and release of charges by ions, thus modulating the barrier at interfaces and resulting in changes to the device conductivity^{[60, 61]}.

Metals and their derived ions, as common charge carriers, can capture and release charges by changing their valence states in response to optical signals. Chen et al. developed an all-optical controlled and self-rectifying optoelectronic memristor (OEM) crossbar array based on a NiO/TiO₂ heterostructure^[40], as shown in Fig. 2(c). Since the band gaps of NiO and TiO₂ were approximately 3.6 and 3.2 eV, respectively, intrinsic excitation was difficult to realize under 480 nm light irradiation with an energy of 2.58 eV. The EPSC was mainly driven by the oxidation of Ni atoms, which released electrons (Ni → Ni⁺ + e⁻). When irradiated with 320 nm ultraviolet light with an energy of 3.87 eV, valence band electrons of NiO and TiO₂ directly entered the conduction band and Ni ions undergone a reduction reaction (Ni⁺ + e⁻ → Ni). The charging and discharging of Ni atoms in the functional layer was identified as the key factor in forming the all-optical modulation memristor. Liu et al. developed a plasmonic optoelectronic memristor using an Ag−TiO₂ nanocomposite film as the dielectric layer, with Au and FTO (fluorine-doped tin oxide) as the electrodes^[62]. The operating mechanism of the device was governed by the oxidation and reduction process of Ag nanoparticles. Under visible light, hot electrons of the Ag nanoparticles were first excited by the LSPR effect, which were subsequently transferred to the conduction band of TiO₂. This process reduced the Schottky barrier at the Ag/TiO₂ interface and enhancing electronic conductivity. On the other side, electrons in TiO₂ were excited from the valence band to the conduction band and recombined with Ag⁺ under UV irradiation, causing decreased conductivity. In addition, Chai et al. fabricated a two-terminal Pd/MoO_x/ITO structure as an ORRAM synaptic device^[63]. Electrons and holes were generated in the MoO_x film upon exposure to UV light. The photo-excited electrons moved into the conduction band of MoO_x, while the photo-generated holes reacted with absorbed water molecules in the MoO_x film to produce protons (H⁺). These photo-generated electrons and protons changed the valence state of Mo ions from Mo⁶⁺ to Mo⁵⁺, forming hydrogen molybdenum bronze (HyMoO_x), leading to the transition of resistance states.

Vacancies are also able to capture and release charges in response to optical signals. Wang et al. reported a novel optoelectronic memristor based on the chlorophyll (Chl) heterojunction, which exhibited synaptic potentiation and inhibition behaviors under 430 and 730 nm light stimulation, respectively^[53]. As shown in Fig. 2(d), the underlying mechanism was attributed to the photo-ionization and deionization of oxygen vacancies at the ZnO/Chl interface. Under 430 nm light, the neutral oxygen vacancies (V_Os) in the ZnO layer were converted to ionized oxygen vacancies (V_O²⁺s), resulting in an increase in electronic conductivity due to excess carriers, which was manifested as photo-induced enhancement. Conversely, photo-generated electrons in Chl combined with V_O²⁺s under 730 nm light, reducing them back to V_Os and leading to photo-induced suppression. Similarly, Wen et al. reported an optoelectronic synaptic device based on a metal halide perovskite (MHP) CsFAMA^[64]. It was proposed that the InO_x and CsFAMA films generated a photoelectric response when light pulses were applied and the photon energy (hν) exceeded the energy gap between the ground state and excited states. Once the light pulse disappeared, the generated electrons tended to migrate to the interface of InO_x and CsFAMA and the photo-generated holes were trapped in the valence band of the CsFAMA layer. Li et al. reported a light-gated memristor based on an ITO/CeO_2−x/AlO_y/Al structure^[65]. By detraping and retrapping of electrons at the interface, Boolean logic functions were successfully implemented within a single device. This was attributed to photon-induced excitation of electrons in the CeO_2−x layer near the AlO_y interface, which increased the concentration of V_O⁺ at the interface.

This Section outlines the primary mechanisms by which the optical field regulates ions in optoelectronic neuromorphic devices. The optical field affects ion migration either by modifying energy barriers or by converting light into other forms of energy. Additionally, the optical field facilitates ions to capture and release charges, which alters the Schottky barrier at interfaces. By modulating ion behavior using these mechanisms, the optical field effectively regulates the device conductivity, enabling neuromorphic functions such as artificial synapses.

The ability to respond to light signals is a crucial criterion for ion-modulation optoelectronic neuromorphic devices. Different materials exhibit varied photoregulatory mechanisms, which significantly influence device performance. For example, halide perovskites can induce ion migration under light irradiation, aswhile cations or holes in metal oxides can capture or release charges in response to specific optical stimuli^{[69, 70]}. This Section provides a detailed overview of optoelectronic neuromorphic devices using various ion-modulation materials and an analysis of their characteristics.

Halide perovskites have emerged as promising candidates for ion migration-based optoelectronic neuromorphic devices due to their low activation energies for ion migration and tunable bandgaps^{[71, 72]}. Moreover, the bandgap of halide perovskites can be modified by adjusting the type and ratio of organic cations or halide anions, which enables precise control over the light absorption range^[73−75]. Blackburn et al. demonstrated persistent photoconductivity (PPC) in perovskite nanocrystal/single-walled carbon nanotube (NC/SWCNT) heterojunctions^[66], as shown in Fig. 3(a). PPC was observed as a common characteristic across all three NC/SWCNT heterojunctions, with the effect lasting for over an hour following a 5-s light pulse. And the energy consumption of device was lowered because of the reduced energy barrier for ion migration under the light pulse. A switching energy of about 740 fJ was achieved at the gate voltage of −20 V and a source−drain voltage of 3 V. Roy et al. utilized FAPbBr₃ as the dielectric layer to construct a vertical Al/FAPbBr₃/ITO memristor^[76]. When exposed to 395 nm UV light at 30 mW/cm², the photo-generated electric field had the same direction with the external electric field, reducing ion activation energy and enhancing the migration of V_Br⁺. The device successfully emulated key synaptic properties, including memory switching, short-term plasticity, and long-term plasticity.

Metal oxides exhibit remarkable light absorption properties with bandgaps primarily in the UV to visible range^{[77, 78]}. Their chemical and thermal stability make them ideal for developing ion-modulation optoelectronic neuromorphic devices^{[79, 80]}. Subramanian et al. presented a photonic memristor based on an Al/SnO_x/FTO structure^[67], as illustrated in Fig. 3(b). The device achieved reliable switching over 100 cycles and responsiveness to a broad spectrum of light wavelengths. Moreover, the SnO_x layer displayed responsiveness to a broad spectrum of light wavelengths, including 254 nm UV, 365 nm UV, 405 nm violet, and 533 nm green light. When exposed to the light stimulus of 0.7 mW/cm², oxygen ions within the SnO_x layer overcame the 0.7 eV migration energy barrier, thereby facilitating the recombination of oxygen vacancies and consequently changing the resistance state by disrupting the conductive filaments.

Ion gels, using ionic liquids as the dispersed phase, exhibit high ionic conductivity and excellent electrochemical stability, making them highly promising for applications in energy storage and ionic skins^[81]. Exposed to external stimuli such as light and heat through specific design, ions move locally between polymer chains and ion coordination sites^{[82, 83]}. Huang et al. demonstrated a self-driven hemispherical retinomorphic eye with elastomeric retina based on an ionogel heterojunction that functions as photoreceptors^[68], as shown in Fig. 3(c). Different migration rates of cations and anions within the heterojunction under illumination led to an imbalanced ion concentration, and an ionic current was produced by the temperature gradient as a consequence of the synergistic effects of photothermal and thermoelectric conversion processes. By selectively doping the ion gel with photosensitive polypyrrole nanoparticles (PPy-NPs) as pigments, the device illustrated ultra-broadband absorption across the ultraviolet−visible−near infrared range (365−970 nm). Furthermore, synaptic behaviors such as SIDP and SNDP were observed by increasing the intensity and number of light pulses.

In addition to ion migration, halide perovskites also exhibit high electron and hole mobility, facilitating efficient separation and transport of photogenerated carriers^{[85, 86]}. The capture and release of charges through halide vacancies renders them suitable for the construction of optoelectronic neuromorphic devices. Li et al. presented a fully optically controlled photoelectric memristor based on the Au/Cs₂AgBiBr₆/Au structure^[41], as shown in Fig. 4(a). The electrons captured by bromine vacancies at the interface of Au/Cs₂AgBiBr₆ could release when exposed to 532 nm (~2.4 eV) green light, leading to a reduction in the width of the Schottky barrier and an increase in device conductivity. In contrast, free electrons from shallow defect levels could be released when exposed to 660 nm (~1.88 eV) red light, which then recombined with V_Br⁺s. The wavelength-dependent persistent photoconductivity successfully mimicked excitatory and inhibitory synaptic plasticity.

Metal oxide heterojunctions leverage the built-in electric field to facilitate efficient separation of photogenerated electron−hole pairs^{[87, 88]}. Metal atoms or oxygen vacancies capture and release electrons under illumination, thereby modulating the interface barrier and consequently altering the resistance of the optoelectronic neuromorphic device^[89]. Chen et al. demonstrated an all-optical modulated optoelectronic multistate memristor crossbar array based on NiO/TiO₂ utilizing the charge and discharge of Ni atoms in the functional layer^[40], as shown in Fig. 4(b). The excitatory postsynaptic current (EPSC) would increase with extended exposure time under irradiation with 480 nm blue light at 6.19 mW/cm² and decrease with 320 nm ultraviolet light at 4.48 mW/cm². Reversible long-term potentiation (LTP) and long-term depression (LTD) were achieved using 320 and 480 nm light. The four-level storage induced by 320/480 nm light with 0.5/1 ks illumination demonstrated retention for over 1000 s, with a device inaccuracy lower than 2 pA under 1 ks of optical modulation, which confirmed excellent uniformity across the array. Li et al. reported an artificial optoelectronic synapse based on an ITO/Nb:SrTiO₃ heterostructure^[84], as shown in Fig. 4(c). The device was found to release electrons from oxygen vacancies at the interface upon illumination, whose current could be modulated by adjusting the wavelength, intensity, and duration of light exposure. Furthermore, the application of an increased number or frequency of light pulses facilitated the transition from short-term memory (STM) to long-term memory (LTM). Besides, Kim et al. developed photonic neuromorphic devices using indium−gallium−zinc-oxide (IGZO), achieving PPC through the ionization of oxygen vacancies under light^[90]. These devices exhibited strong response to both blue and ultraviolet light, successfully simulating critical synaptic functions.

The choice of materials is crucial for device performance. Halide perovskites demonstrate outstanding photoresponse properties across a broad spectrum of wavelengths. Metal oxides provide excellent thermal stability and well-established fabrication processes. Ion gels enable the development of flexible optoelectronic neuromorphic devices. Each of these materials offers distinct advantages and plays a crucial role in various practical application scenarios.

Ion-modulation optoelectronic neuromorphic devices demonstrate significant advantages in emulating synaptic functions through the synergistic interaction of optical and electronic signals^[95−97]. Their ability to provide multi-dimensional control makes them highly promising for applications in a wide range of fields. This Section focuses on the potential for enhancing artificial vision systems, advancing neuromorphic computing, and developing other bionic applications.

Visual information processing is critical for human perception, with over 80% of information acquired from the external world coming through the visual system^{[98, 99]}. However, traditional vision systems face significant challenges, such as high energy consumption and delay caused by the separation of sensing and processing units^[100]. Optoelectronic neuromorphic devices, integrating light sensing and information processing, offer a promising solution to reduce energy consumption and latency^[101]. Huang et al. developed synaptic transistors with ultra-low power consumption (~250 aJ/spike) using flexible alkyl chains to covalently bond cations and anions in an electrolyte^[91]. These transistors exhibited various optoelectronic synaptic behaviors. As shown in Fig. 5(a), after applying "F"-shaped light pulses to the transistor array, the letter "F" was detected and stored in the array after removing the light pulse, demonstrating its retinal-like signal preprocessing capability. Furthermore, when the artificial synapse was used as a memory module within a sensor, the spatiotemporal visual information was mapped through the photon synapse array to construct a reservoir computing (RC) system capable of recognizing the trajectory of a basketball with an accuracy of 95.56%. Liu et al. used two-dimensional perovskite/organic heterojunction to develop an ambipolar synaptic phototransistor for flexible color recognizable visual system^[92]. As shown in Fig. 5(b), the perovskite-organic heterojunction ambipolar SPTs (POASPTs) generated different currents under red, green, blue and NIR light spikes to irradiate the 12 × 5 POASPT arrays, respectively. The arrays clearly distinguished the light colors, displayed four images of different color characters and exhibited memory abilities, which simulated the retinal function of the human eye. Mei et al. developed an organic optoelectronic synapse characterized by high-density, multilevel conductance modulation and low operating voltages (<1 V)^[43]. The current response could be modulated by light intensity and wavelength, enabling the perception and memory of various visual information. A simulated 64 × 64 optical synapse array was used as an artificial retina for human facial recognition, as shown in Fig. 5(c). The artificial retina exhibited an activation rate of over 85% for the target female face after training, while activation rates for other female and male faces remained below 70%, demonstrating the ability of the artificial retina to effectively learn and recognize key visual features.

Park et al. introduced an all-optical bidirectional synapse device that could obtain synaptic potentiation under red light and synaptic depression under green light^[93]. The device realized an accuracy of 90.1% when applied to digital image pattern recognition, as shown in Fig. 5(d). Wang et al. developed a three-dimensional neuromorphic photosensor array based on a vertical graphite/CuInP₂S₆/graphite (Gr/CIPS/Gr) photosensor unit^[102]. The array was laterally extended to 10 × 10 and further fabricated into a vertically stacked 3D 3 × 3 × 3 array. Compared to conventional 2D serial computing, the 3D parallel computing approach significantly increased processing speed and accuracy of handwritten digit recognition as well as reduced power consumption. Liang et al. reported a memristor synaptic device based on the 2D MoTe₂/MoS_2−xO_x Van der Waals heterostructure^[92]. The device acted as both an electronic and photonic synapse, reaching recognition accuracies of 99.3% and 96.5% in handwritten digit recognition tasks, respectively. For more complex image classification, it obtained accuracies of 95.3% (electronic) and 91.5% (photonic), outperforming similar neuromorphic devices in triggering mechanisms and recognition accuracy, as shown in Fig. 5(e). Li et al. introduced a reconfigurable and precisely controllable synaptic device based on the organic−inorganic hybrid halide perovskite Cs_0.05MA_0.15FA_0.8PbI_0.85Br_0.15^[64]. The optoelectronic neural network achieved a fast and high recognition accuracy of 96.1% using optical/electrical pulses of 0.01 s width. Incorporating an additional preprocessing layer with optoelectronic co-stimulation further improved accuracy from 96.1% to 96.5%. These advances highlighted the significant potential of ion-modulation optoelectronic devices in artificial vision systems, offering high-precision, energy-efficient and low-latency recognition for various applications.

Associative learning is an essential cognitive function that enables humans to form connections between different stimuli and produce corresponding responses^[104]. Recently, researchers have proved that this cognitive function can be simulated using optoelectronic neuromorphic devices^[105]. Lin et al. developed an optoelectronic sensor based on the MAPbBr₃/PdSe₂ Schottky junction^[44]. A 520 nm light signal was employed to represent the sound of a bell, while continuous voltage pulses of 100 ms were used to represent food. It was observed that the current remained below threshold when only the light signal was applied, analogous to the dog not salivating. However, after simultaneous application of both light and electrical signals, the current exceeded threshold with the light signal alone, mimicking the dog salivating at the sound of the bell alone. After a period of no voltage stimulation, the current elicited by the light signal was below threshold again, suggesting that the dog had forgotten the conditional response, as shown in Fig. 6(a). Moreover, Zhang et al. demonstrated associative learning in Li-ion doped ZrLiO/InLiO thin-film transistors through the Pavlovian classical conditioned reflex experiments^[106]. Five electrical pulses and optical pulses (UV, 50 mW/cm²) with a width of 20 ms were applied as the unconditional and conditional stimuli, respectively. Before training, the dog did not salivate in response to the bell, but did salivate in response to food. After repeated simultaneous stimulations of the bell and food, the dog began to salivate to the bell alone, indicating the formation of an association between the stimuli. After about 3 min, the dog forgot this association and the bell no longer elicited salivation, representing the decay of the conditioned reflex.

Non-associative learning refers to changes in response to a single stimulus over time without the formation of associations between different stimuli^[107]. This type of learning typically involves processes such as habituation and sensitization. Kim et al. developed an all-oxide-based artificial photonic nociceptor that exhibited allodynia and hyperalgesia when exposed to light signals^[103], as shown in Fig. 6(b). The "uninjured" nociceptor responded weakly to low-intensity UV light before exposure to harmful strong UV light. However, the "injured" nociceptor showed a marked increase in current in response to low-intensity UV light after exposure to strong UV radiation, demonstrating pain sensitization characteristics. Except that, Zhu et al. reported a neuromorphic transistor based on the mixed proton and electron conductor (MPEC). When 30 optical pulses with a duration of 1 s and an interval of 10 ms were applied as strong stimuli, the postsynaptic current (PSC) gradually increased, exhibiting light sensitization behavior. This effect was further enhanced by either shortening the pulse interval or increasing the number of stimulation pulses.

Ion-modulation optoelectronic neuromorphic devices can simulate memory implantation, erasing, and modification regulated by optogenetics in biological synapses. Li et al. developed a 3 × 5 optoelectronic synaptic array using Au/Cs₂AgBiBr₆/Au all-optically controlled memristor^[41], as shown in Fig. 6(c). Based on the properties that 532 nm green light increased conductance and 660 nm red light decreased conductance, the conductance state of each device represents the synaptic weight encoding memory information. Initially, pulse trains of 100 mW/cm² green light stimulated five devices in the second column of the array to write the digit pattern of "1", and the digital feature was reliably stored. Subsequently, red light was applied to erase pixels 5 and 11, followed by green light to write the digit "2" on the corresponding pixels. The digit pattern of "3" was written by irradiating the corresponding pixels with red and green light, which simulated the repeated modification of memory contents by optical techniques. By further constructing an artificial neural network based on a multiple-layer perceptron (MLP), the recognition rates for digits "2" and "3" were 80% and 74%, significantly higher than the 42% and 38% recognition rates achieved when only green light was used to modify the digits, demonstrating the effectiveness of optogenetic tuning in enhancing cognitive abilities. Hao et al. reported a hydrogel whose conductivity could be modulated by optical field to mimic the function of biological synapses^[45]. The hydrogel could act as an information processing unit that perceived different optical stimulus. When a weak stimulus was applied to the OMIC hydrogel, the synaptic weight did not reach the threshold and the optically modulated motion feedback system failed to generate a grasping motion. However, it would be realized when a strong stimulus activated the artificial synapse to a highly active state. Subsequently, a weak stimulus was also able to maintain the artificial synapse in a highly active state, enabling the robotic hand to repeatedly perform grasping motions. This system that experienced a learning stage exhibited different motion feedback when subjected to the same optical input, performing learning-experience behavior.

These findings confirm that optoelectronic neuromorphic devices have significant potential in artificial visual systems, neuromorphic computing, and other bionic applications. In artificial vision systems, the successful simulation of retinal functions enables image recognition and motion detection with high accuracy. For neuromorphic computing, associative and non-associative learning are simulated, efficiently mimicking the learning rules found in the brain. Additionally, ion-modulation optoelectronic neuromorphic devices have shown promising prospects in simulating optogenetics and constructing optically modulated motion feedback systems, suggesting broader potential applications in the future.

This review highlights recent works in ion-modulation optoelectronic neuromorphic devices, focusing on mechanisms, characteristics, and applications. The operating mechanisms are categorized into two types: (Ⅰ) illumination that directly or indirectly regulates ion migration by altering energy barriers or photothermal effects; (Ⅱ) light-assisted charge capture and release via metal ions or vacancies. Various materials, including halide perovskites, metal oxides, and ion gels, have been explored to develop the devices based on different mechanisms. The ion-modulation optoelectronic neuromorphic devices demonstrate superior controllability compared to traditional electrical-driven neuromorphic devices, particularly in the simulation of synaptic functions. This is attributed to the wide bandwidth, high energy efficiency, low latency, and multi-dimensional control methods offered by optical signals. The devices have shown significant potential in areas such as artificial vision systems and neuromorphic computing.

Despite the numerous developments in the field of ion-modulation optoelectronic neuromorphic devices, several challenges remain: (Ⅰ) Though various mechanisms have been proposed, they do not fully elucidate all observed phenomena. For instance, the generation of additional carriers in amorphous oxide semiconductors under light exposure requires further investigation^[108]. (Ⅱ) Conventional optoelectronic materials like oxide semiconductors are limited by their bandgap, restricting their response to specific wavelengths. Strategies such as doping are needed to extend spectral responses from ultraviolet to infrared^[109]. (Ⅲ) The fabrication processes for perovskite materials remain immature, presenting challenges in terms of reproducibility and stability, which hinders large-scale integration. (Ⅳ) Poor stability and uniformity of devices limit recognition accuracy, further restricting their utilization in advanced applications.

To address these challenges and advance the field, several research directions are proposed. (Ⅰ) In-situ characterization techniques are employed to deepen the understanding of ion behaviors and mechanisms within the materials. (Ⅱ) Innovations and optimizations should be conducted in material design and synthesis to enhance device performance and integration capabilities. (Ⅲ) Novel device architectures are explored to leverage the unique properties of ion-modulation optoelectronic materials more effectively. (Ⅳ) Research works focus on implementing more complex neuromorphic functions, moving beyond basic image detection and memory to advanced recognition and processing tasks. By addressing the current challenges and pursuing innovative research directions, the devices are poised to achieve significant advancements and breakthroughs.