Optoelectronic synapse based on Mg_0.1Zn_0.9O/TiO₂ heterojunction for neuromorphic computing

Abstract

Optoelectronic synapses that integrate sensing and synaptic weight modulation are key enablers for brain-inspired computing. Here, we report an Mg_0.1Zn_0.9O/TiO₂ (MZO/TiO₂)-based optoelectronic synaptic device capable of emulating essential biological synaptic functions, including paired-pulse facilitation, short- and long-term plasticity, as well as potentiation and depression. While electrical stimulation alone leads to volatile synaptic responses, the introduction of 365-nm optical excitation enables nonvolatile behavior, allowing more effective emulation of learning and memory. The device further supports basic arithmetic operations and, when integrated into a LeNet-5 neural network, achieves a recognition accuracy of 96.48% on the MNIST data set. These results demonstrate a promising optoelectronic strategy for advanced neuromorphic computing systems.

Graphical abstract

Keywords

Optoelectronic synapse / Resistive switching / Thin films / Neural networks / Neuromorphic computing

1 Introduction

With the rapid development of artificial intelligence and neuromorphic computing, the construction of brain-inspired information processing systems has emerged as a key research direction in the post-Moore era [1–3]. Neuromorphic computing, inspired by the structure and functionality of biological neural networks, aims to emulate the behavior of neurons and synapses, offering significant advantages over traditional von Neumann architectures in terms of parallelism, learning capability, and computational efficiency [4]. However, the realization of efficient brain-like computing critically depends on the development of novel artificial synaptic devices that exhibit multi-level storage, tunable synaptic weights, and plasticity [5]. Among the various candidates, memristors have attracted considerable attention due to their simple structure, non-volatility, multistate switching behavior, and compatibility with CMOS technology, making them one of the most promising options for constructing artificial synapses [6–9].

Among various memristive material systems, transition metal oxides have attracted extensive attention due to their abundant defect states and excellent electrical and optical properties. Among them, ZnO-based materials have been widely employed in optoelectronic synaptic devices owing to their wide bandgap, high carrier mobility, and remarkable ultraviolet photoresponse characteristics [10–18]. However, the ability of a single ZnO-based functional layer to regulate the separation, transport, and trapping of photogenerated carriers is relatively limited, making it challenging to simultaneously achieve diverse and tunable optoelectronic synaptic plasticity [19,20]. TiO₂ is also a wide-bandgap oxide semiconductor with excellent chemical stability, mature thin-film fabrication processes, and abundant defect states, providing potential advantages in ultraviolet photoresponse and defect-assisted charge modulation. Compared with single oxide systems, constructing an Mg_0.1Zn_0.9O/TiO₂ (MZO/TiO₂) heterostructure introduces additional interfacial band engineering and interface defect states, which are expected to facilitate the separation of photogenerated carriers and regulate their trapping and release processes. Therefore, the MZO/TiO₂ heterostructure provides a feasible material-design strategy for achieving richer and more tunable optoelectronic synaptic behaviors.

In this work, we designed and fabricated an optoelectronic synaptic device based on a MZO/TiO₂ heterostructure and systematically investigated its conductance modulation, synaptic plasticity, and photoresponse characteristics. Mg doping can regulate the band structure, carrier concentration, and defect-state distribution of ZnO, while the introduction of TiO₂ creates a heterointerface that provides additional pathways for regulating the separation, transport, trapping, and gradual release of photogenerated carriers. Therefore, the significance of this device does not lie in claiming absolute performance superiority over all previously reported ZnO- or TiO₂-based devices, but rather in the integration of multiple functionalities within a single device through the synergistic effects of the MZO layer, TiO₂ layer, and their heterointerface, including electrical synaptic plasticity, optically regulated memory, leaky integrate-and-fire (LIF)-like behavior, basic arithmetic operations, and neural-network recognition. This multifunctional integration highlights the potential of the MZO/TiO₂ heterostructure for optoelectronic neuromorphic devices.

Furthermore, compared with purely electrically controlled synapses, optoelectronic synapses cannot only regulate synaptic weights through electrical stimulation but also directly perceive external optical signals, enabling the integration of optical information sensing, conversion, storage, and synaptic weight modulation within a single device. In this work, the MZO/TiO₂ device exhibits partially volatile synaptic responses under electrical stimulation, whereas 365 nm ultraviolet illumination significantly improves the current retention after removal of the stimulus, resulting in a more persistent optoelectronic synaptic response. This dual electrical and optical modulation capability can potentially reduce information transfer between photodetection, storage, and processing units in conventional visual systems, providing a potential device foundation for neuromorphic visual systems integrating sensing, memory, and computing functionalities. Through this work, we aim to explore the potential of the MZO/TiO₂ heterostructure for multifunctional optoelectronic synapses and neuromorphic computing, while further expanding the design strategies for composite oxide heterostructure-based devices.

2 Experimental section

The Mg_0.1Zn_0.9O/TiO₂ (MZO/TiO₂) thin films were fabricated using radio frequency (RF) magnetron sputtering. To eliminate potential surface impurities on the target, the TiO₂ target was pre-sputtered prior to deposition. Specifically, the TiO₂ target was pre-sputtered in an Ar atmosphere for 5 min. Afterward, TiO₂ films were deposited onto cleaned and dried indium tin oxide (ITO) substrates. The sputtering was conducted at an RF power of 100 W, with an initial base pressure of 1 × 10⁻⁵ Pa, an Ar flow rate of 75 sccm, and a sputtering time of 60 min. Following the TiO₂ deposition, the MZO target was also pre-sputtered for 5 min under identical Ar conditions. Subsequently, the MZO layer was deposited directly onto the TiO₂ film at an RF power of 80 W, with the same base pressure (1 × 10⁻⁵ Pa) and Ar flow rate (75 sccm), and a sputtering duration of 20 min. After film deposition, Au top electrodes were deposited onto the MZO layer using a small vacuum coating system to complete the fabrication of the synaptic device.

3 Results and discussion

The schematic comparison between a biological synapse and the Au/MZO/ TiO₂/ITO device structure is shown in Fig. 1a. In this configuration, the Au top electrode serves as the presynaptic neuron, while the functional layers composed of MZO and TiO₂ act as an analog of the synaptic cleft. The ITO bottom electrode represents the postsynaptic neuron. By analogizing conductance modulation to synaptic weight change, synaptic plasticity can be emulated through the application of electrical or optical pulses to regulate the device’s conductance.

Figure 1b displays the cross-sectional scanning electron microscope (SEM) image of the device, in which distinct layers of different materials are clearly observed, confirming the successful fabrication of the heterojunction structure. Figure 1c shows the O 1s X-ray photoelectron spectroscopy (XPS) spectrum, where the peak at a binding energy of 530.45 eV corresponds to lattice oxygen, 531.5 eV to oxygen vacancies (defect oxygen), and 532.25 eV to surface-adsorbed oxygen. Figure 1d presents the Zn 2p XPS spectrum, with peaks located at 1021.5 eV and 1044.5 eV, corresponding to Zn 2p_3/2 and Zn 2p_1/2, respectively. Figure 1e shows the Mg 1s XPS spectrum, where the peak at 1304.1 eV is attributed to Mg 1s. The X-ray diffraction (XRD) pattern of the MZO/ TiO₂ device is shown in Fig. S1 in the Supplementary Information. The MZO diffraction peak corresponds to the (002) plane, indicating that the MZO film has a hexagonal wurtzite structure and forms a heterojunction with the TiO₂ layer [21].

The I–V hysteresis loops of the device are shown in Fig. S2, obtained by cyclic voltage sweeps in the sequence of 0 V → 3 V → 0 V → −3 V → 0 V for 50 consecutive cycles. Synaptic behavior refers to a device’s ability to emulate biological synapses, while synaptic plasticity is a key prerequisite for realizing brain-inspired computing and the hardware implementation of artificial neural networks [22,23]. To investigate the synaptic characteristics, continuous voltage sweeps of ±3 V were applied, as illustrated in Figs. 2a and 2b. With increasing sweep cycles, the device conductance gradually increases (or decreases), indicating progressive potentiation and depression behaviors.

Paired-pulse facilitation (PPF) is an important indicator for evaluating short-term synaptic plasticity. It is characterized by applying two closely spaced voltage pulses with a time interval (Δt) and measuring the corresponding postsynaptic current responses [24]. The PPF behavior of the device is shown in Fig. 2c, with the inset displaying the schematic of the input paired pulses. Experimental results demonstrate that as Δt increases, the PPF index gradually decreases, exhibiting a typical exponential decay trend. This behavior closely resembles the neurotransmitter residual effect observed in biological synapses, indicating that the device can effectively emulate short-term synaptic memory processes. PPF is defined as:

(1)\text{PPF}=\frac{A_2-A_1}{A_1}\times 100\mathrm{\%},

where A₁ represents the postsynaptic current increment induced by the first pulse, and A₂ corresponds to that induced by the second pulse. The PPF data were fitted using a bi-exponential decay function:

(2)\mathrm{P}\mathrm{P}\mathrm{F}=W_1\exp \left(-\frac{\text{Δ}t}{{\tau }_1}\right)+W_2\exp \left(-\frac{\text{Δ}t}{{\tau }_2}\right).

In the bi-exponential decay fitting, A₁ and A₂ are the amplitude constants of the two exponential terms, representing the response intensities on different time scales. Δt denotes the time interval between the two pulses. τ₁ and τ₂ are the relaxation time constants, corresponding to the fast and slow decay components, respectively. The extracted values are τ₁ = 0.3566 and τ₂ = 0.4358. The two relaxation time constants obtained from the double-exponential fitting correspond to the fast and slow carrier relaxation processes in the device, respectively. The shorter relaxation time constant (τ₁) may be associated with the rapid release and recombination of carriers from shallow-level defect states, whereas the longer relaxation time constant (τ₂) may originate from the slow release of carriers trapped in deep-level defects or interfacial trap states within the heterostructure. These two relaxation processes collectively enable the residual carriers generated by the first pulse to enhance the device response to the subsequent pulse, resulting in the typical PPF behavior.

Figure 2d shows the LTP and LTD characteristics of the MZO/TiO₂ neuromorphic device. A series of constant 1 V / −2 V pulses were applied to induce LTP and LTD, respectively. The current increases under stimulation by 1 V pulses, indicating that the device conductance is in a potentiated state. Conversely, the current decreases under −2 V pulses, suggesting that the conductance is undergoing long-term depression. The pulse width and interval were both set to 0.1 s. The device exhibits stable LTP and LTD behavior, maintaining stability over approximately 900 pulses. In addition, the MZO/TiO₂ neuromorphic device is also capable of emulating leaky integrate-and-fire (LIF) neuron behavior, as shown in Fig. S3.

We draw an analogy between the device and the human brain: the increase in current under external stimulus corresponds to the brain’s learning process; the peak current at the end of stimulation represents the level of memory acquired. The subsequent decrease in current after the stimulus is removed is regarded as the forgetting process, and the final stabilized current represents the retained memory after forgetting. As shown in Fig. 2e, a sequence of 30 consecutive voltage pulses (1 V amplitude, 0.2 s width, 0.1 s interval) was applied to the device. This can be considered as 30 rounds of continuous learning, during which the memory level of the device reached 6.6 μA. After a period of time, the memory current returned to its initial level due to forgetting. When another 10 consecutive learning pulses were applied, it was observed that the device required only 6 pulses to reach the same memory level that previously required 30 pulses, indicating a learning enhancement. However, the device exhibited volatile behavior under electrical stimulation, which is insufficient for stable memory applications. Therefore, we further employed a 365 nm light stimulus to excite the MZO/TiO₂ device and found that it exhibited nonvolatile behavior under ultraviolet illumination.

However, after the removal of the external voltage stimulus, the device current gradually returns to its initial level, exhibiting a relatively pronounced volatile behavior. To achieve a longer-lasting synaptic response, we further applied 365 nm ultraviolet illumination to the MZO/TiO₂ device. Unlike the transient conductance modulation mainly induced by electrical stimulation, ultraviolet illumination can excite the generation of photogenerated electron–hole pairs in both the MZO and TiO₂ functional layers. The band alignment of the MZO/TiO₂ heterointerface may facilitate the spatial separation of photogenerated carriers, while oxygen vacancy-related defect states observed from the O 1s XPS analysis and interfacial trap states can capture part of the photogenerated carriers, thereby suppressing rapid carrier recombination. After removal of the optical stimulus, the trapped carriers are gradually released, allowing the device to maintain an elevated conductance state for a certain period and exhibiting an optoelectronic synaptic response with prolonged retention characteristics. Therefore, the observed optically regulated memory behavior may originate from the synergistic effects of the MZO layer, TiO₂ layer, and their heterointerface, involving the generation, separation, trapping, and gradual release of photogenerated carriers.

Optoelectronic synaptic devices emulate the synergistic functions of photoreceptors and neural synapses in the human visual system, enabling the perception, conversion, and transmission of external optical signals into electrical signals for information processing in brain-inspired computing systems, as illustrated in Fig. 3a. The MZO/TiO₂ device exhibits both photosensitivity and tunable conductance, allowing it to mimic the memory and response behaviors of biological neurons under optical stimulation, thereby providing a foundation for developing artificial visual systems [25,26]. Figure 3b presents the I–V curves of the device under dark conditions and under ultraviolet illumination. It is evident that the current under UV illumination is significantly higher than that in the dark, indicating the device’s high sensitivity to UV light.

The excitatory postsynaptic current (EPSC) of the device under light pulse stimulation with varying power densities is shown in Fig. 3c [27]. As the power density increases, the EPSC also increases, and the nonvolatile retention becomes more pronounced. The EPSC responses under light pulses with different pulse widths and different pulse numbers are presented in Figs. 3d and 3e, as well as in Figs. S4 and S5. It can be observed that with increasing pulse width and number of pulses, the EPSC gradually increases, accompanied by enhanced nonvolatile behavior. These results indicate that longer learning duration and more frequent stimulation effectively improve the memory level and enable greater memory retention after a period of forgetting. These results are consistent with the trend described by the Ebbinghaus forgetting curve, which suggests that repeated learning can enhance memory retention. This indicates that the device is capable of mimicking learning, forgetting, and relearning behaviors associated with biological synaptic processes [28,29].

The ultraviolet photoresponse mechanism of the MZO/TiO₂ neuromorphic device mainly originates from the synergistic effects of heterointerface modulation, oxygen vacancy-related defect states, and the trapping/release processes of photogenerated carriers, as illustrated in Fig. S6. TiO₂ has an electron affinity of approximately 4.03 eV, while that of MZO is approximately 3.76 eV [30]. After contact between the two materials, the difference in conduction band positions leads to a conduction band offset of approximately 0.27 eV, inducing interfacial charge redistribution and band bending. Meanwhile, Au forms a Schottky contact with MZO, and the barrier height is theoretically estimated to be approximately 1.34 eV based on the difference in work functions. These interfacial effects collectively influence the carrier transport behavior of the device.

Under dark conditions, the Au/MZO Schottky contact and the MZO/TiO₂ heterointerface induce the formation of a space charge region on the MZO side, where electrons are mainly distributed within the MZO layer, maintaining the device in a low-conductance state. When the device is illuminated with 365 nm ultraviolet light, photons are absorbed by MZO to generate electron–hole pairs. Driven by the built-in electric field at the MZO/TiO₂ heterointerface, photogenerated electrons and holes undergo spatial separation, with electrons migrating toward the TiO₂ layer while holes remain primarily in the MZO layer. This process reduces carrier recombination and enhances the photoresponse. Meanwhile, oxygen vacancy-related defect states in MZO can capture a portion of the photogenerated carriers and gradually release electrons through slow detrapping processes after removal of the optical stimulus, enabling the device to maintain an elevated conductance state. Therefore, the defect-assisted carrier trapping and release processes provide the physical basis for the persistent optoelectronic response and the relaxation behavior of the EPSC that mimics biological synaptic functions.

Arithmetic operations are one of the fundamental functions of the human brain for logical reasoning and information processing. Implementing arithmetic computation based on neuromorphic devices is of great significance for advancing neuromorphic computing systems [31]. In this study, we successfully realized arithmetic responses to electrical pulse inputs using the MZO/TiO₂ synaptic device. Benefiting from its spike number-dependent synaptic plasticity, the device is capable of emulating and performing brain-inspired arithmetic functions such as addition, subtraction, multiplication, and division, providing a new pathway for constructing efficient neuromorphic computing hardware [32,33]. As shown in Fig. 4a, a sequence of electrical pulses with a width of 0.2 s, amplitude of 1 V, and interval of 0.1 s was applied to the device, resulting in spike-number-dependent EPSC responses as presented in Fig. 4b.

Figure 4c demonstrates the arithmetic operation of addition, specifically “6 + 6 = 12”. Based on Fig. 4b, the threshold current for EPSC₁₂ is set at 6.43 μA. The device was first stimulated with six consecutive pulses, followed by another six pulses. The EPSC triggered by the final pulse reaches 6.43 μA, corresponding to EPSC₁₂, thus confirming the result of “6 + 6 = 12”. Similarly, the multiplication operation “4 × 3 = 12” is shown in Fig. 4e. The device was sequentially stimulated with three sets of four consecutive pulses. The final pulse elicited an EPSC of 6.43 μA, again corresponding to EPSC₁₂, confirming the multiplication result.

Figure 4d illustrates the subtraction operation “12 – 8 = 4”. The threshold current is set to EPSC₈, approximately 5.28 μA. After applying 12 consecutive pulses, the number of additional pulses required to again reach the threshold EPSC₈ is counted, yielding the result “12 – 8 = 4”. Figure 4f presents the division operation “12 ÷ 5 = 2…2”. The threshold current is set to EPSC₅, approximately 4.52 μA. Twelve consecutive pulses are applied to the device, and each time the EPSC reaches the threshold value, a division cycle is completed. When the current from the remaining pulses is insufficient to reach the threshold, the process ends. The number of completed cycles corresponds to the quotient, and the number of remaining pulses corresponds to the remainder, yielding the result “12 ÷ 5 = 2…2”.

The above arithmetic operations are achieved based on the device’s integration response to sequential pulses and threshold-based decision processes, which essentially correspond to information accumulation, state retention, and threshold output mechanisms in neuromorphic systems. The output current of the device not only records the historical information of the input pulses but also serves as the criterion for determining the computational results, demonstrating the integration of memory and computation within a single device. Although this demonstration represents a basic functional verification, it provides experimental evidence for the future development of parallel and low-power in-memory computing systems based on memristor arrays.

In the biological nervous system, the strength of synaptic connections can be dynamically regulated by neuronal activity, a phenomenon known as synaptic plasticity, which is considered the biological foundation of learning and memory. LTP enhances synaptic weights by strengthening the coordinated activity between pre- and postsynaptic neurons, while LTD weakens active connections to enable selective forgetting. In neuromorphic devices, consecutive electrical pulses can be applied to gradually increase or decrease the conductance, corresponding to LTP and LTD processes, respectively. As shown in Fig. 5c, the LTP/LTD curves of the device were fitted using a nonlinear exponential function, yielding nonlinearity indices of A_P = 2.28 and A_D = 3.23.

To verify the application potential of the MZO/TiO₂ device in neuromorphic computing, its LTP and LTD characteristics were mapped as synaptic weight update rules and incorporated into a classical convolutional neural network model for pattern recognition simulation [34–36]. Based on this nonlinear synaptic behavior, a brain-inspired weight update function was constructed and embedded into the LeNet-5 network architecture to perform handwritten digit recognition tasks. The structure of the adopted LeNet-5 model is shown in Fig. 5a, which consists of two convolutional layers, two pooling layers, and one fully connected layer. The input is a 28 × 28 grayscale image. The first convolutional layer extracts 16 feature maps of size 28 × 28, followed by max-pooling to obtain 16 feature maps of size 14 × 14. The second convolutional layer further extracts 16 feature maps of size 14 × 14, which are then down sampled to 16 feature maps of size 7 × 7 through another pooling layer. All feature maps are flattened and passed into the fully connected layer to produce classification results for 10 digit categories [37–39]. The MNIST handwritten digit data set is shown in Fig. 5b [40]. After 100 training epochs, the model achieved a recognition accuracy of 96.48% on the test set, as presented in Fig. 5d, fully demonstrating the feasibility and application potential of the proposed neuromorphic device in neuromorphic computing systems.

4 Conclusion

In this work, we successfully fabricated MZO/TiO₂ optoelectronic synaptic devices using magnetron sputtering and demonstrated typical synaptic behaviors, including short-term plasticity, PPF, LTP/LTD, and LIF characteristics. The device exhibited relatively volatile responses under electrical stimulation, whereas 365 nm ultraviolet illumination significantly enhanced the current retention after stimulus removal, resulting in prolonged optoelectronic synaptic responses and more effectively mimicking biological learning and memory processes. In terms of neuromorphic applications, the device enabled basic arithmetic operations, including addition, subtraction, multiplication, and division. Furthermore, by mapping the experimentally measured LTP/LTD characteristics into an artificial neural network, an accuracy of 96.48% was achieved in the MNIST handwritten digit recognition task. These results demonstrate that the MZO/TiO₂ device integrates optoelectronic synaptic plasticity and neuromorphic computing functions within a single device, providing a potential strategy for developing optoelectronic neuromorphic intelligent devices.

Future efforts will focus on further optimizing device structures and fabrication processes, together with systematic evaluations of device-to-device uniformity and carrier dynamics, to gain deeper insights into the charge modulation mechanisms at the MZO/TiO₂ heterointerface. These studies will provide a foundation for the integration of memristor arrays and the development of practical neuromorphic hardware systems.

References

[1] Duan , X. , Cao , Z. , Gao , K. , Yan , W. , Sun , S. , Zhou , G. , Wu , Z. , Ren , F. , Sun , B. : Memristor-based neuromorphic chips. Adv. Mater 36(14), 2310704(2024)

[2] Zhu , Z. , Shui , J. , Wang , T. , Meng , J. : Mechanical properties analysis of flexible memristors for neuromorphic computing. Nano-Micro Lett 18(1), 2(2026)

[3] Guo , T. , Zhang , B. , Wang , X. , Xiao , Y. , Sun , B. , Zhou , Y.N. , Wu , Y.A. : Broadband optoelectronic synapse enables compact monolithic neuromorphic machine vision for information processing. Adv. Funct. Mater 33(49), 2303879(2023)

[4] Liu , Z. , Fang , Y. , Cai , Z. , Liu , Y. , Dong , Z. , Zheng , R. , Shen , Z. , Wu , R. , Qu , W. , Fu , J. , Ru , C. , Wu , Y. , Gu , J. , Liu , Y. , Liu , Q. , Zhao , C. , Wen , Z. : Advanced dual-input artificial optical synapse for recognition and generative neural network. Nano Energy 132, 110347(2024)

[5] Ghazanfar , H. , Rabeel , M. , Nisar , S. , Ahmad , M. , Zulfiqar , M.W. , Sulaman , M. , Kim , H. , Dastgeer , G. , Kim D. : Interface-engineered Gd₂O₃/ZrO₂ bilayer memristor for emulating synaptic plasticity in neuromorphic systems. J. Colloid Interface Sci 709, 139943(2026)

[6] Hu , L. , Yang , J. , Wang , J. , Cheng , P. , Chua , L.O. , Zhuge , F. : All-optically controlled memristor for optoelectronic neuromorphic computing. Adv. Funct. Mater 31(4), 2005582(2021)

[7] Jin , B. , Wang , Z. , Wang , T. , Meng , J. : Memristor-based artificial neural networks for hardware neuromorphic computing. Research 8, 0758(2025)

[8] Dastgeer , G. , Nisar , S. , Rasheed , A. , Akbar , K. , Chavan , V.D. , Kim , D.k. , Wabaidur , S.M. , Zulfiqar , M.W. , Eom , J. : Atomically engineered, high-speed non-volatile flash memory device exhibiting multibit data storage operations. Nano Energy 119, 109106(2024)

[9] Dastgeer , G. , Nisar , S. , Zulfiqar , M.W. , Eom , J. , Imran , M. , Akbar , K. : A review on recent progress and challenges in high-efficiency perovskite solar cells. Nano Energy 132, 110401(2024)

[10] Jang , Y.H. , Han , J.K. , Moon , S. , Shim , S.K. , Han , J. , Cheong , S. , Lee , S.H. , Hwang , C.S. : A high-dimensional in-sensor reservoir computing system with optoelectronic memristors for high-performance neuromorphic machine vision. Mater. Horiz 11(2), 499–509(2024)

[11] Chu , J. , Chen , L. , Zheng , K. , Chen , H. , Song , J. : Optical memristors based on homojunctions of zinc oxide quantum dots and nanowires. ACS Appl. Nano Mater 8(28), 14065–14071(2025)

[12] Wang , Y. , Zheng , J. , Shang , Y. , Guo , D. , Wang , H. , An , Z. , Chen , X. , Huang , R. , Jiang , J. , Sun , K. , Wang , S. : LTP–LTD transformation of unipolar pulse voltage-driven zinc oxide memristors via TiO₂ thin layer incorporation. J. Mater. Chem. C Mater. Opt. Electron. Devices 12(4), 1281–1288(2024)

[13] Laurenti , M. , Porro , S. , Pirri , C.F. , Ricciardi , C. , Chiolerio , A. : Zinc oxide thin films for memristive devices: a review. Crit. Rev. Solid State Mater. Sci 42(2), 153–172(2017)

[14] Zhao , T. , Zhao , C. , Xu , W. , Liu , Y. , Gao , H. , Mitrovic , I.Z. , Lim , E.G. , Yang , L. , Zhao , C.Z. : Bio-inspired photoelectric artificial synapse based on two-dimensional Ti₃C₂T_x MXenes floating gate. Adv. Funct. Mater 31(45), 2106000(2021)

[15] Li , Z. , Li , Z. , Zuo , C. , Fang , X. : Application of nanostructured TiO₂ in UV photodetectors: a review. Adv. Mater 34(28), 2109083(2022)

[16] Dastgeer , G. , Zulfiqar , M.W. , Nisar , S. , Zulfiqar , R. , Imran , M. , Panchanan , S. , Dutta , S. , Akbar , K. , Vomiero , A. , Wang , Z. : Emerging role of 2D materials in photovoltaics: efficiency enhancement and future perspectives. Nano-Micro Lett 18(1), 32(2026)

[17] Nisar , S. , Basha , B. , Dastgeer , G. , Shahzad , Z.M. , Kim , H. , Rabani , I. , Rasheed , A. , Al-Buriahi , M.S. , Irfan , A. , Eom , J. , Kim , D. : A novel biosensing approach: improving SnS₂ FET sensitivity with a tailored supporter molecule and custom substrate. Adv. Sci. (Weinh.) 10(33), 2303654(2023)

[18] Nisar , S. , Dastgeer , G. , Shazad , Z.M. , Zulfiqar , M.W. , Rasheed , A. , Iqbal , M.Z. , Hussain , K. , Rabani , I. , Kim , D. , Irfan , A. , Chaudhry , A.R. : 2D materials in advanced electronic biosensors for point-of-care devices. Adv. Sci. (Weinh.) 11(31), 2401386(2024)

[19] Chen , X. , Hu , W. , Wu , S. , Bao , D. : Stabilizing resistive switching performances of TiN/MgZnO/ZnO/Pt heterostructure memory devices by programming the proper compliance current. Appl. Phys. Lett 104(4), 043508(2014)

[20] Doroody , C. , Sajedur Rahman, K. , Chelvanathan , P. , Adib Ibrahim, M. , Sopian , K. , Amin , N. , Chowdhury , S. , Channumsin , S. : Incorporation of magnesium-doped zinc oxide (MZO) HRT layer in cadmium telluride (CdTe) solar cells. Results in Physics 47, 106337(2023)

[21] Chen , J.Y. , Zhong , W.M. , Ren , Q.Z. , He , A. , Guo , X.B. , Jiang , Y.P. , Liu , Q.X. , Tang , X.G. : Brain-like device simulating dendrites perception process and optical induced excitatory postsynaptic current. ACS Photonics 12(2), 696–704(2025)

[22] Kim , H. , Kim , M. , Lee , A. , Park , H.L. , Jang , J. , Bae , J.H. , Kang , I.M. , Kim , E.S. , Lee , S.H. : Organic memristor-based flexible neural networks with bio-realistic synaptic plasticity for complex combinatorial optimization. Adv. Sci. (Weinh.) 10(19), 2300659(2023)

[23] Sung , S.H. , Kim , T.J. , Shin , H. , Im , T.H. , Lee , K.J. : Simultaneous emulation of synaptic and intrinsic plasticity using a memristive synapse. Nat. Commun 13(1), 2811(2022)

[24] Xia , Z. , Sun , X. , Wang , Z. , Meng , J. , Jin , B. , Wang , T. : Low-power memristor for neuromorphic computing: from materials to applications. Nano-Micro Lett 17(1), 217(2025)

[25] Guo , Z. , Kan , H. , Zhang , J. , Li , Y. : Neuromorphic visual computing with ZnMgO QDs-based UV-responsive optoelectronic synaptic devices for image encryption and recognition. Small 21(15), 2412531(2025)

[26] Li , Y. , Wang , J. , Yang , Q. , Shen , G. : Flexible artificial optoelectronic synapse based on lead-free metal halide nanocrystals for neuromorphic computing and color recognition. Adv. Sci. (Weinh.) 9(22), 2202123(2022)

[27] Tan , Z.H. , Yang , R. , Terabe , K. , Yin , X.B. , Zhang , X.D. , Guo , X. : Synaptic metaplasticity realized in oxide memristive devices. Adv. Mater 28(2), 377–384(2016)

[28] Cui , D. , Pei , M. , Lin , Z. , Zhang , H. , Kang , M. , Wang , Y. , Gao , X. , Su , J. , Miao , J. , Li , Y. , Zhang , J. , Hao , Y. , Chang , J. : Versatile optoelectronic memristor based on wide-bandgap Ga₂O₃ for artificial synapses and neuromorphic computing. Light Sci. Appl 14(1), 161(2025)

[29] Li , R. , Dong , Y. , Qian , F. , Xie , Y. , Chen , X. , Zhang , Q. , Yue , Z. , Gu , M. : CsPbBr₃/graphene nanowall artificial optoelectronic synapses for controllable perceptual learning. PhotoniX 4(1), 4(2023)

[30] Zhang , Y. , Wu , Q. , Zhao , W. , Liu , Y. , Lin , X. , Zhu , Z. , Huang , K. , Liu , X. : The effect of hydrogen on the properties of Mg_0.1Zn_0.9O thin film. J. Appl. Phys 133(15), 155104(2023)

[31] Tan , H. , Liu , G. , Zhu , X. , Yang , H. , Chen , B. , Chen , X. , Shang , J. , Lu , W.D. , Wu , Y. , Li , R.W. : An optoelectronic resistive switching memory with integrated demodulating and arithmetic functions. Adv. Mater 27(17), 2797–2803(2015)

[32] Huang , W. , Hang , P. , Wang , Y. , Wang , K. , Han , S. , Chen , Z. , Peng , W. , Zhu , Y. , Xu , M. , Zhang , Y. , Fang , Y. , Yu , X. , Yang , D. , Pi , X. : Zero-power optoelectronic synaptic devices. Nano Energy 73, 104790(2020)

[33] Wang , J. , Wang , J. , Zhang , J. , Huang , W. , Wang , X. , Zhang , M. : Ultralow-power synaptic transistors based on Ta₂O₅/Al₂O₃ bilayer dielectric for algebraic arithmetic. Adv. Electron. Mater 8(5), 2100922(2022)

[34] Guo , L. , Sun , H. , Min , L. , Wang , M. , Cao , F. , Li , L. : Two-terminal perovskite optoelectronic synapse for rapid trained neuromorphic computation with high accuracy. Adv. Mater 36(27), 2402253(2024)

[35] Yang , Z. , Huo , S. , Zhang , Z. , Meng , F. , Liu , B. , Wang , Y. , Ma , Y. , Wang , Z. , Xu , J. , Tian , Q. , Wang , Y. , Ding , Y. , Hu , X. , Xie , Y. , Fan , S. , Pan , C. , Wu , E. : High-precision multibit opto-electronic synapses based on ReS₂/h-BN/graphene heterostructure for energy-efficient and high-accuracy neuromorphic computing. Adv. Funct. Mater 35(48), 2509119(2025)

[36] Mahata , C. , Ismail , M. , Kim , H. , Kim , S. : Optically tunable synaptic plasticity and memory emulation in Au-nanoparticle enhanced HfTiO_x/Al₂O₃-based photonic memristors. Adv. Funct. Mater 36(3), e10663(2026)

[37] Portner , K. , Schmuck , M. , Lehmann , P. , Weilenmann , C. , Haffner , C. , Ma , P. , Leuthold , J. , Luisier , M. , Emboras , A. : Analog nanoscale electro-optical synapses for neuromorphic computing applications. ACS Nano 15(9), 14776–14785(2021)

[38] Chen , J. , Xu , W. : 2D-materials-based optoelectronic synapses for neuromorphic applications. eScience 3(6), 100178(2023)

[39] Yao , P. , Wu , H. , Gao , B. , Tang , J. , Zhang , Q. , Zhang , W. , Yang , J.J. , Qian , H. : Fully hardware-implemented memristor convolutional neural network. Nature 577(7792), 641–646(2020)

[40] Ahn , H. , Kim , Y. , Seo , S. , Lee , J. , Lee , S. , Oh , S. , Kim , B. , Park , J. , Kang , S. , Kim , Y. , Ham , A. , Lee , J. , Park , D. , Kwon , S. , Lee , D. , Ryu , J.E. , Shin , J.C. , Sahasrabudhe , A. , Kim , K.S. , Bae , S.H. , Kang , K. , Kim , J. , Oh , S. , Park , J.H. : Artificial optoelectronic synapse featuring bidirectional post-synaptic current for compact and energy-efficient neural hardware. Adv. Mater 37(34), 2418582(2025)

Rights & permissions

The author(s)