The exploration of photosensitive synaptic devices focuses on leveraging the human brain’s ultralow energy consumption, parallel computation, and high fault tolerance in visual perception and learning [1–7]. The development of multi-response and multifunctional synaptic devices is crucial for advancing electronic skins and brain-inspired chips [8–12]. A key step involves effectively developing optoelectronic synaptic devices, which are essential components of intelligent synaptic systems [13–16]. To explore the desired characteristics of photoelectric synapses, research has focused on two device structures: two-terminal devices and three-terminal transistors [17–19]. Transistors have more signal modulation ports, better fault tolerance, and a higher signal-to-noise ratio than two-terminal devices [20,21]. To enhance photoresponsive synaptic properties in transistor-based visual systems, specific materials and structural designs are necessary [22,23]. Moreover, there is an ongoing discussion concerning the mechanism of synapses’ response to light stimulation [24,25]. More research is needed to improve the integration of optoelectronic synapses for stronger connections with artificial intelligence. Developing photoelectric synaptic devices capable of multiple responses is crucial in the era of neuromorphic computation.

By using optical signal bandwidth and interconnection, computational speed in neural networks can be enhanced [26]. It is essential to focus on developing intelligent synapses that respond to light, study their response mechanisms, and simulate multifunctional applications for future artificial intelligence systems. Many research papers have been published on light-stimulated synaptic transistors and their functional simulations, concentrating on inherent photoconductivity or incorporating light-responsive materials like perovskite quantum dots [27]. To improve the light-sensitive characteristics of synaptic transistors, Huang et al. integrated chlorophyll into organic transistors, replicating key functionalities such as synaptic plasticity and image identification [10]. Photonic synaptic devices using a heterostructure of fluorescent silicon quantum dots and monolayer molybdenum disulfide were demonstrated by Wang et al., enhancing charge transfer [28]. The intrinsic photoconductive properties of porous oxide semiconductors enable synapse transistors that respond to light and exhibit pulse-time dependency [29]. Extensive research has been conducted on light-responsive synaptic transistors using hybrid semiconductors and inorganic quantum dots. For example, heterojunctions improve the photoresponsivity and charge separation efficiency in fully printed optoelectronic synaptic transistors [21]. In recent years, versatile photoelectric synaptic transistors have been successfully simulated for tasks such as detecting rotational movement [30,31], sensing pain perception [32,33], controlling emotional simulation and action [34], recognizing gestures [35], and regulating transistors with multiple wavelengths. However, using these devices requires photosensitive materials and complex manufacturing techniques. It is crucial to explore synapse devices, develop material-independent preparation methods, understand their operational mechanisms, and create adaptable gadgets to leverage the advantages of synapse devices and mimic the human brain’s functionalities.

In this study, we developed artificial synaptic transistors by constructing a heterostructure with poly(2-((3,6,7,10,11-pentakis (hexyloxy) triphenylene-2-yl)oxy) ethyl methacrylate) (PMHT) and incorporating ${\mathrm{Al}}_2{\mathrm{O}}_3$ nanoparticles to capture charge carriers efficiently. The PMHT demonstrates exceptional ability to form films and displays remarkable photoelectric synaptic properties. The effective charge separation achieved through photoproduction is ensured by integrating ${\mathrm{Al}}_2{\mathrm{O}}_3$ nanoparticles, enhancing the photo-synaptic plasticity of transistors. The study develops logical functionality, electrical and light synaptic plasticity, UV-induced sunburned skin simulation, and a model for improving learning efficiency via photoelectric cooperative pulses, advancing versatile neuromorphic computing systems.

The schematic diagram of a $\mathrm{PMHT}/{\mathrm{Al}}_2{\mathrm{O}}_3$ heterojunction synapse transistor is depicted in Fig. S1 of Ref. [36]. Figure S2 in Ref. [36] illustrates a cross-sectional view of the device. FTIR spectroscopy of PMHT and $\mathrm{PVA}+{\mathrm{Al}}_2{\mathrm{O}}_3$ composite material is depicted in Fig. S3 of Ref. [36]. The output characteristic and transfer curves of the heterojunction synapse transistor are shown in Fig. S4 of Ref. [36]. To verify the repeatability of device performance, detailed characterization tests were conducted on 20 devices, encompassing output characteristics, transfer curves, and threshold voltages, as depicted in Fig. S5 of Ref. [36]. The insulation ability of PMMA was investigated by testing the leakage current, as shown in Fig. S6 [36]. The data retention characteristics of the device are shown in Fig. S7 [36].

Logical gates are fundamental components in digital circuits and essential for developing neuromorphic circuits [37]. The device’s excellent gate and drain control over the channel current facilitated the development of four logic gates. Figure S8 in Ref. [36] shows the circuit diagram to realize the logic function. Figure S9 in Ref. [36] illustrates the schematic representation of the logical gate. The execution of “AND”, “OR”, “NOR”, and “NAND” functions is shown in Fig. S10 of Ref. [36], while the truth table can be found in Table S1 of Ref. [36].

To enable subsequent cooperative training using photoelectric methods, a comprehensive investigation is carried out to understand the essential characteristics of electrical synapses. Figures 1(a) and 1(b) illustrate the arrangement of devices and the orientation of carrier motion within a synapse under varying gate voltages, both negative and positive. In the presence of a negative gate voltage, there is a tendency for hole accumulation near the channel, leading to an elevation in EPSC. After the elimination of the voltage surge, there is a reduction in hole density as the dissipation of induced voltage occurs. Alternatively, when subjected to a positive gate pulse stimulation, the presence of electronic accumulation leads to the depletion of holes away from the channel. Subsequently, upon removal of the positive gate stimulus, there is a gradual release of these depleted holes, which results in a slow restoration of the initial value for IPSC [38]. Consequently, the continuous increase in EPSC can be observed as a result of the enhanced negative gate voltage stimulation [Fig. 1(c)]. As depicted in Fig. 1(d), the IPSC exhibits a gradual decline (from $-4.86$ to $-0.69\mathrm{nA}$) as the positive gate voltage magnitude increases from 4 to 10 V. It is mainly due to the amplified positive gate voltage pulses generating more clustered electrons, thus improving efficiency [39]. Hence, by manipulating the gate voltage’s direction and magnitude, it is possible to exert additional control over the channel conductance. The PMHT semiconductor is primarily responsible for the higher current generated when applying negative gate voltage as compared to positive voltage stimulations, which is a representative p-type material containing abundant hole charges that exhibit greater sensitivity towards negative gate voltage. The number of pulses, and width- and amplitude-dependent plasticity are shown in Fig. S11 of Ref. [36].

The investigation primarily examines the stability of EPSC triggered by red light [Fig. 2(a)] and UV light [Fig. 2(b)] individually, throughout six cycles. Figure S12 in Ref. [36] shows the error bars and device-to-device variation for EPSC values from 20 devices under two types of light stimulation conditions. Comparatively, UV light induces EPSC with a greater current intensity. The increase in EPSC from red to UV light suggests that higher energy levels result in enhanced exciton separation facilitated by ${\mathrm{Al}}_2{\mathrm{O}}_3$. In addition, the training of various stimuli simultaneously enables synapses to replicate intricate operations akin to Pavlovian dogs. Figure 2(c) illustrates the EPSC following training with red light and UV light. Figure S13 in Ref. [36] shows the statistical diagram of EPSC during alternating red and UV light training. The synaptic facilitation behavior is shown in Fig. S14 of Ref. [36].

In addition, Fig. 2(d) illustrates the synaptic currents regulated by the light pulse counts. Furthermore, Fig. 2(e) demonstrates that as the number of pulses increased from 1 to 40, there was an initial rapid rise followed by a gradual increase in EPSC. This observation suggests that with an increasing number of pulses, the channel carrier reaches a state of saturation. Similarly, the EPSC distribution influenced in Fig. 2(f) by the duration of the light pulse (ranging from 0.2 to 1.0 s in 20 pulses) exhibits a linear correlation with time. Figure S15 in Ref. [36] provides the EPSC distribution influenced by the duration of the light pulse.

Figure 2(g) illustrates the training process of optical synapses, including learning, forgetting, and relearning. During the continuous sequence of 40 ultraviolet light pulses (each lasting for 0.2 s with a 0.2 s interval), the current exhibited an increase to 18.42 nA. Subsequently, after a period of forgetting lasting for approximately 67 s, the current decreased to 6.32 nA. However, following an additional set of 18 relearning pulses, the current was restored to its previous level of 18.42 nA once again. These findings suggest that light synapses demonstrate memory capabilities akin to those observed in human brain learning processes. Reliability and repeatability of learning and relearning processes are analyzed in Fig. S16 of Ref. [36]. Compared with the learning process, the number of pulses required for the relearning process is smaller, which leads to the improvement of relearning efficiency. The essence of this improvement lies in the residual charges of deep traps in ${\mathrm{Al}}_2{\mathrm{O}}_3$, the incomplete orientation relaxation of the PVA matrix, and the pre-formed ion migration paths [40], which collectively constitute the physical-level memory traces. This mechanism of the synergy between non-volatile and volatile successfully simulates the “learning acceleration effect” in biological synapses.

The thickness of the $\mathrm{PMHT}/{\mathrm{Al}}_2{\mathrm{O}}_3$ heterojunction exhibits a nonlinear influence on both the photoelectric response and synaptic behavior of the device through the regulation of the photoelectric-ion coupling mechanism. Regarding the photoelectric response, when the heterojunction is excessively thin ($<80\mathrm{nm}$), light penetration dominates, but limited light absorption occurs, leading to a low concentration of photogenerated carriers and consequently a weak photocurrent response. Conversely, an overly thick heterojunction ($>240\mathrm{nm}$) results in saturated light absorption; however, the prolonged carrier transport path increases recombination probability, attenuating the photocurrent due to the weakening of the built-in electric field. A moderate heterojunction thickness (80–240 nm) achieves a balance between optical absorption and carrier separation efficiency, enabling effective separation of photogenerated electron-hole pairs under the applied electric field and optimizing the photoelectric response. In terms of synaptic behavior, thinner heterojunctions enhance interface coupling, facilitating the modulation of channel conductivity by the electric field and expanding the dynamic range of synaptic weights (conductances). However, an excessively thick heterojunction diminishes the gate voltage’s ability to regulate the channel, resulting in suboptimal synaptic current responses. Based on experimental data analysis, the approximate thicknesses of the PMHT and $\mathrm{PVA}+{\mathrm{Al}}_2{\mathrm{O}}_3$ layers are 67 and 51 nm, respectively. Ultimately, we selected a heterojunction thickness of 118 nm for the $\mathrm{PMHT}/{\mathrm{Al}}_2{\mathrm{O}}_3$ structure.

The effect of ${\mathrm{Al}}_2{\mathrm{O}}_3$ nanoparticles dispersion (3%, 13%, 25%, 40%, 55%) on the performance of $\mathrm{PMHT}/{\mathrm{Al}}_2{\mathrm{O}}_3$ heterojunction artificial synaptic transistors was investigated. The experimental results show that when the dispersion rate is 25%, ${\mathrm{Al}}_2{\mathrm{O}}_3$ nanoparticles show uniform distribution in the PVA matrix, which significantly improves the contact characteristics of the PVA and PMHT heterojunction interface, promotes ion migration, and effectively mimics the plasticity of biological synapses. The surface morphology of $\mathrm{PVA}+{\mathrm{Al}}_2{\mathrm{O}}_3$ composite film with ${\mathrm{Al}}_2{\mathrm{O}}_3$ nanoparticles dispersion rates of 13%, 25%, and 55% is shown in Fig. S17 in Ref. [36].

The human skin is prone to harm caused by UV radiation, leading to issues like skin inflammation and the development of skin cancer. Hence, a model for assessing various levels of sunburn-induced damage on the skin is suggested. The device level includes the mechanisms for identifying the length of time exposed to UV rays, gaps between consecutive exposures to UV light, and effects of exposure on skin harm. Here, we present a schematic illustration of skin sunburn caused by intense UV exposure [Fig. 3(a)], mild skin injury resulting from intermittent UV irradiation, and moderate skin damage due to repeated UV exposure. Furthermore, we investigated the effects of varying UV exposure times ranging from 0.2 to 2 s, as depicted in Fig. 3(b); intervals between UV irradiation, as shown in Fig. 3(c), with $\mathrm{\Delta }t$ ranging from 0.5 to 10; and number of UV pulses applied, from one pulse up to 10 pulses, as illustrated in Fig. 3(d). Notably, all experiments utilized a consistent intensity of UV light at a wavelength of 365 nm and power density of $110\mathrm{\mu W}·{\mathrm{cm}}^{-2}$.

The corresponding changes in postsynaptic current were systematically analyzed and are presented in Fig. S18 of Ref. [36]. A change of the facilitation index as a function of exposure time, irradiation intervals, and pulse number is shown in Fig. S19 of Ref. [36]. The facilitation index was quantitatively correlated with the biological model of skin injury. When the facilitation index was greater than one, the damage occurred. Ultraviolet light damaged the skin barrier function, resulting in increased skin water loss, thus making the skin dry, which corresponded to mild damage. A facilitation index greater than 200 indicates that ultraviolet light causes damage to lipids and collagen, leading to skin aging, which corresponds to moderate damage.

To investigate the long-term stability under UV irradiation, a gate voltage of $-5\mathrm{V}$ was applied to convert the device into the on state, while simultaneously exposing it to UV light. Notably, the strong penetration of 365 nm ultraviolet light may trigger reactions between free radicals and oxygen, thereby degrading the electrical properties of the device. Consequently, the long-term stability was evaluated under a wavelength of 365 nm and a power density of $110\mathrm{\mu W}/{\mathrm{cm}}^2$ (with a drain reading voltage of $-0.5\mathrm{V}$). As shown in Fig. S20 of Ref. [36], the current exhibited attenuation, which can be attributed to high-energy interactions causing main-chain scission, side-chain oxidation, or conjugated structure disruption in organic polymers such as PMHT and PVA. These processes lead to reduced carrier mobility and diminished conductivity. Furthermore, charge traps induced by UV irradiation compromise the long-term retention of synaptic weights.

Reaching optimal learning enables the human brain to achieve swift advancements. In the field of biology, utilizing appropriate learning approaches can lead to a twofold outcome with only half the exertion. To replicate this phenomenon, we devised a straightforward model for gauging learning efficacy by employing photoelectric training with the mentioned transistors. In relation to the idea, there is available evidence that suggests favorable outcomes can be achieved through the concurrent examination and exploration of new information. The two excitation signals employed were an electrical signal consisting of 30 consecutive pulses of $-2\mathrm{V}$ with a duration of 0.2 s, and a light pulse comprising 30 consecutive UV pulses each lasting for 0.2 s. The alterations in EPSC following light and electrical signals (initially, 30 UV light pulse stimuli followed by 30 trained synapses with electrical pulse stimuli, and finally another 30 UV light pulse stimuli) are depicted in Fig. 4(a), while the corresponding peak current $I_{30}$ is shown in Fig. 4(d). These results indicate that the impact of electrical stimulation on light stimulation is minimal, and there is no significant enhancement in current (from 10.61 to11.41 nA) or memory effect on synapses after undergoing electrical training. This procedure, referred to as reviewing old knowledge in Fig. 4(a), involves the utilization of light-electric-light stimulation. The inadequate effectiveness of this instructional approach hinders the progress of human knowledge acquisition, necessitating modifications to enhance learning efficiency [Fig. 4(g), process ①]. In a similar manner, the application of an electric-light-electric stimulation process during training [Fig. 4(b)] resulted in minimal improvement in synaptic current memory [Fig. 4(e), increasing from 31.75 to 33.75 nA]. This indicates that this process primarily facilitates the acquisition of new knowledge rather than significantly enhancing existing memories. Merely undergoing minimal training has a restricted influence on the electrical response, thus, indicating that the effectiveness of this training approach is currently insufficient for achieving rapid improvement [Fig. 4(g), process ②]. In addition, Fig. 4(c) illustrates the investigation of the training process involving $\mathrm{light}-(\mathrm{light}+\mathrm{electric})-\mathrm{light}$. The corresponding peak current $I_{30}$, as depicted in Fig. 4(f), exhibits an increase from 10.68 to 26.89 nA. The significant current rise in present indicates the combined impact of photoelectric concurrent training on intelligent synapses, resulting in notable amplification of current behavior. In the context of biological learning, this process involves the simultaneous acquisition of new knowledge and the revision of previously acquired knowledge, which is considered to be the most effective approach [Fig. 4(g), process ③]. In the process of this procedure, skillful training of intelligent synapses enables control over the conduction state. The concurrent training that incorporates photovoltaic stimulation aids in enhancing the excitatory postsynaptic current and ensuring its prolonged existence. This occurrence has the ability to replicate the logical function of “AND” and mimic the efficient learning process observed in the human brain. It is solely by employing appropriate and optimal techniques for acquiring knowledge that individuals can make significant advancements at a faster pace and on a broader scale. The potential insights for future synaptic systems can be derived from simulating the training and learning capabilities of photoelectric synapses. Analysis of the resistive switching mechanism of optical synapses is shown in Fig. S21 of Ref. [36]. Table S2 in Ref. [36] presents a comparison of heterojunction artificial synaptic transistors from this study with those reported in previous studies.

According to the architecture of our synaptic transistor, it is feasible to configure multiple gates and drains, enabling the simulation of diverse synaptic behaviors with multiple inputs and outputs within a single transistor. This design significantly enhances the scalability of the device. In laboratory settings, hundreds to thousands of transistors can be fabricated in batches. When simulating real-world scenarios, the exact number of connections is contingent upon the functional requirements of synaptic bionics. For the functionalities described in this work, a single transistor connection suffices. The slower response time of synaptic transistors, measured in seconds, remains a significant concern. Such delays may compromise their efficacy in real-time applications. Therefore, enhancing the response time of synaptic transistors represents our next critical research objective.

In biological experiments, using animal models or cell cultures to study the behavior of neurons and synapses involves high experimental costs and time consumption. Processing large-scale experimental data requires a large amount of storage space and data analysis capabilities. However, through heterojunction synaptic phototransistor research, the manufacturing cost of such devices is lower than that of traditional chips or experimental devices, which will reduce the funding requirements of the overall research. Due to its nanoscale channel, its miniaturization property, it is possible to integrate more functions in a smaller space, reducing the need for laboratory space.

Artificial synaptic transistors with photoelectric response are realized using a PMHT heterostructure and ${\mathrm{Al}}_2{\mathrm{O}}_3$ nanoparticles for efficient charge carrier capture. This study analyzes both electrical and optical properties. The synaptic function can be regulated using electrical pulses, light pulses, or a combination of both. Investigations have examined the excitatory and inhibitory properties of electrical synapses and the learning functions of optical synapses. Advanced features, including a sunburn model controlled by UV radiation, a human brain learning efficiency model regulated by photoelectric collaboration, four logic functions (“AND”, “OR”, “NOR”, “NAND”), and memory function controlled by UV pulses, have been effectively demonstrated and simulated. The advancement of neuromorphic computing can be further facilitated by effectively and adaptively managing and simulating photoelectric signals. The use of artificial synaptic transistors in this study will significantly advance device-level simulation for multifunctional synapse integration.