Experimental and First‐Principles Study of Visible Light Responsive Memristor Based on CuAlAgCr/TiO2/W Structure for Artificial Synapses with Visual Perception

The development of vision bionic systems is indispensable for the perception, memory, and processing of optical signals, which promotes the exploration of efficient visual perception systems. In this work, a simple and novel two‐terminal optoelectronic memristor based on the CuAlAgCr/TiO2/W (CTW) structure is prepared, where the CuAlAgCr high‐entropy alloys are employed as the top electrode for the first time. Before annealing, the CTW optoelectronic memristor exhibited fascinating performance, including uniformly distributed operating voltage, reliable data retention, and a higher switching ratio. Moreover, the optoelectronic memristor can be reversibly switched between volatile and nonvolatile memories by adjusting compliance currents. The CTW optoelectronic memristor annealed in air exhibits various artificial synaptic functions, such as short‐term memory, optical learning, and forgetting behavior under the illumination of the laser. The photo‐response current is increased from nano‐ampere to micro‐ampere level. Furthermore, a logic function unit based on CTW optoelectronic memristor is proposed, which realizes “AND” operation. Furthermore, first‐principles calculations of the CTW structure are performed to describe the influence of photocarriers on the barrier height at the CuAlAgCr‐TiO2 interface, revealing the working mechanism of the CTW optoelectronic memristor. This work has greatly facilitated the development of optically operated artificial synaptic devices and vision bionic systems.


Introduction
Human learning strongly relies on vision, which is a dominant channel to collect knowledge of the external world. Therefore, the success of achieving artificial vision equipment in near future arises from the debut of a visual bionic system to perceive and learn from the external environment. [1,2] Typical vision bionic system includes photodetectors that perceive visual input as digital images, storage units for visual information, and processing units that perform complex image processing tasks. [3][4][5] It should be noted that in the human brain, sensed optical signal that is transformed into an action potential, is transmitted among neurons according to synapses. Therefore, optical synapses become one of the most important prerequisites for realizing vision bionic systems. Following such design criteria, the popularity of humanoid systems with artificial synapses, such as face recognition and industrial robots, have been witnessed during the last decade, excitingly changing citizens' daily life.
However, the mainstream vision system today is based on complementary metal oxide semiconductor (CMOS) integrated circuits bound to conventional Von Neumann architecture. With Moore's law reaching its limits, device performance can be hardly improved by solely reducing the feature size of CMOS integrated circuits. Besides, Von Neumann architecture that separates data processing from storage inevitably causes certain problems, such as memory walls and power consumption walls. These undoubtedly increase the system's power consumption and slow down the system's computational speed. [6] Hence multifunctional brain-inspired devices that integrate sensing, memory, and processing functions are highly anticipated to achieve a more efficient vision bionic system.
Recently, a simple but effective humanoid vision device based on an optoelectronic memristor was reported to realize the integration of both image recognition and memory functions. [7] The optoelectronic memristor can directly respond to the optical stimulus, transfer computing tasks to the optical sensing unit, and process visual information in real time, which greatly improves the system's efficiency. [8][9][10][11][12] Furthermore, operating memristors in an optical manner usually presents several advantages of higher bandwidth, faster speed, lower crosstalk, strong antiinterference ability, and low computational energy consumption, in comparison with its electronic counterpart. [13] Therefore, tremendous efforts have been made to study optoelectronic memristors for their highly efficient neuromorphic applications. [7,9] For instance, Lei et al. reported an optoelectronic memristor based on 2D bismuth iodide (BiOI) nanosheets that exhibited resistive switching performances as well as light-induced synaptic plasticity. [14] However, its photoresponse current stayed at the nano-ampere (nA) level. Such low photoresponse current undoubtedly increased the difficulty in subsequent readout operations that may require very complex peripheral circuits. [15] Paola Russo et al. studied the resistance state switching behavior of resistive random access memory based on ZnO rods exposed to ultra-violet (UV) light while failing in the visible band. This deteriorates its practical applications. [9] Based on previous reports, the current optoelectronic memristor-based vision system is subjected to insurmountable drawbacks, and novel optoelectronic devices with higher performances are urgently desired in order to further push forward the development of the vision bionic system. Both of them obstruct the development of the vision bionic system. In addition, most optoelectronic memristors used traditional metal (i.e., low-entropy metals) electrodes. In recent years, medium-entropy and high-entropy attract much attention because it possesses some advantages of high strength/hardness, outstanding structural stability, and excellent corrosion et al. [16][17][18][19] The high entropy alloys may lead to applications in new photodetectors, lasers, scintillators, and memristors. [20] Inspired by the above challenges and advantages, the CuAlAgCr high-entropy alloy was for the first time adopted as a contact electrode of a newly developed optoelectronic memristor with CuAlAgCr/TiO 2 /W (CTW) structure. The CTW optoelectronic memristor was found to exhibit several advantages such as great endurance (>10 2 ), long retention (>10 3 s), low switching set voltage (0.8 V), and high switching ratio (10 2 ). Without an annealing process, setting different compliance currents (I CC ) can make CTW optoelectronic memristors exhibit either volatile threshold switching (TS) or nonvolatile resistive-switching (RS) behavior, owing to the stability of the conductive filament. Such memristive behaviors were not found in the same sample after the annealing process, which can be retrieved by exposing the annealed sample to the laser light. In contrast to the aforementioned memristors, the CTW optoelectronic memristor enabled the light response at the visible spectrum (i.e., 405-650 nm), triggering the photoresponse current within the micro-ampere (µA) regime. This is likely due to the advent of photogenerated carriers that significantly changed the barrier height at the CuAlAgCr-TiO 2 interface, demonstrated by the corresponding first-principles calculations. Typical biological synapse features, such as long-term potentiation (LTP), paired-pulse facilitation (PPF), optical learning and forgetting behaviors, and short-term memory (STM), were successfully reproduced by the developed CTW optoelectronic memristor via either optical or electrical stimulus. Furthermore, we demon strated a basic logic function unit based on the CTW optoelectronic memristor, which realizes the "AND" operation. This work has greatly contributed to artificial synaptic devices and vision bionic systems with photoelectric operations.

Results and Discussion
For the human cognitive system, external light information can be perceived through the visual organs and transmitted to the visual cortex of the brain through the connected neurons for data storage and computation, as schematically illustrated in Figure 1a. In biological synaptic systems, when the presynaptic membrane is stimulated, the energy of calcium ions and sodium ions triggers the transmission of neurotransmitters to the postsynapse, which is known as the transmission behavior of the synapse. [21] These are similar to the synapse function (analogy characteristics) of CTW optoelectronic memristor (Figure 1b) upon light stimuli. Each unit in the 5 × 5 memristor array ( Figure 1b) can sense the external optical stimulus and store the state after the stimulation, with excellent photoelectric performance, which can integrate sensing, storage, and computing functions in a single unit. Therefore, the synapse functions of the CTW optoelectronic memristors under electrical and optical stimuli are investigated as follows.
The field emission scanning electron microscope (FE-SEM) was used to obtain the cross-sectional morphology of the fabricated device and the CuAlAgCr thin film, resulting in Figure 2a-d. The cross-sectional images of the CTW optoelectronic memristor without and with annealing in Figure 2a,b revealed the thickness of TiO 2 film of ≈30 nm, while the thickness of the CuAlAgCr top electrode (TE), TiO 2 film, and W bottom electrode were found to be ≈100, 30, and 100 nm, respectively. It can also be seen that the annealed CTW optoelectronic memristor exhibited a rougher surface than that of the unannealed CTW optoelectronic memristor. Furthermore, we carried out FE-SEM analysis on the CuAlAgCr film before and after annealing, respectively, as shown in Figure 2c,d. It can be clearly observed that the particle size of the CuAlAgCr film before and after annealing was estimated to be ≈100 nm and ≈30 nm, respectively, indicating a smaller and denser cluster distribution after annealing. The Energy Dispersive Spectroscopy (EDS) www.advelectronicmat.de mapping images of the CTW optoelectronic memristor, as shown in Figure S1, Supporting Information, revealed that the elements of Ag, Cu, Cr, Al, Ti, and W coexisted in the as-prepared devices. Figure 2e presented the X-ray diffraction (XRD) patterns of the CuAlAgCr alloy film before and after annealing, in which the broad peak at a degree of ≈22° corresponds to the amorphous peak of the SiO 2 substrate. [22] Other peaks of XRD patterns indicated the characteristic peaks of CuAlAgCr alloy. It was also informative to note that the full width at halfmaximum (FWHM) of CuAlAgCr alloy film after annealing was narrower than that before annealing, exemplified by an FWHM of 2.18 and 1.26 at a degree of ≈38° before and after annealing, respectively. More intriguingly, a new peak appeared at a degree of ≈77.6°. The above results clearly revealed the fact that the CuAlAgCr film after annealing allowed for larger, more uniform, and less defective grains in comparison with the case before annealing. [2] The electrical characteristics of the CTW optoelectronic memristor before annealing were measured by the semiconductor analyzer (Keithley, 4200A). During the measurement, the electrical stimulus was applied to the CuAlAgCr electrode, and the W electrode was grounded. Figure 3a exhibited the direct current (DC) current-voltage (I-V) measured results of the CTW optoelectronic memristor under different I CC . It can be seen that the CTW optoelectronic memristor showed different memristive switching behaviors (TS and RS) along with different I CC . The CTW optoelectronic memristor led to TS behavior under small I CC (i.e., 10 and 20 µA in Figure 3a) and RS behavior under large I CC (i.e., 70, 80, and 90 µA in Figure 3a), respectively. The initial resistance state of the CTW optoelectronic memristor was in a high resistance state (HRS). When the bias voltage was swept to "SET" voltage (V SET = 0.5 V), the resulting current suddenly increased to I CC (I CC = 10 µA or 20 µA), meaning that the CTW optoelectronic memristor switched from the HRS to the low resistance state (LRS) (i.e., ON). When the positive bias voltage was swept back, the response current suddenly decreased to 0.2 µA and returned to its HRS (i.e., OFF). When the applied voltage entered the negative regime, the CTW optoelectronic memristor still remained "OFF", showing the TS behavior.

www.advelectronicmat.de
When I CC was further increased to 70, 80, and 90 µA, the CTW optoelectronic memristor transited from HRS to LRS at a bias voltage greater than 0.8 V. Such LRS was maintained until the applied bias reached "RESET" voltage (V RESET = −0.6 V), triggering the transition from LRS to HRS. Such a trend clearly reflected the RS characteristic of the designed device.
In order to demonstrate the repeatability of the TS phenomenon, we repeatably performed DC sweeps on the CTW optoelectronic memristor with I CC = 10 µA for 150 cycles. The resulting HRS and LRS among different cycles were extracted and illustrated in Figure 3b. It was found that the memory window (HRS/LRS) was ≈10 2 and can be well maintained, which considerably reduced the potential readout error. The corresponding V SET among different cycles were also extracted and its distribution histogram was depicted in Figure 3c, which is fitted by the Gaussian function. Note that V SET was concentrated at ≈ 0.5 V, showing a uniform switching voltage performance. Such low V SET undoubtedly endowed the designed memristor with low-power consumption. In order to explore the physics that governed its TS behavior, the I-V curves of different cycles (i.e., 30, 60, 90, 120, and 150) @ positive LRS region were fitted, respectively, as shown in www.advelectronicmat.de Figure 3d. It can be observed that there existed a linear relationship between bias voltage and response current, which likely contributed to the conduction mechanism of the LRS to the ohmic conduction. [24] In addition to the TS behavior, we also studied the RS behavior of the CTW optoelectronic memristor, as illustrated in Figure 3e-i. Figure 3e depicted the I-V curves of the CTW optoelectronic memristor for 100 cycles under I CC = 90 µA. The typical I-V curve was highlighted by the red curve, and the arrow indicated the scan direction of the bias voltage. During the positive scan, when the bias voltage was scanned to the V SET of ≈0.8 V, the current of the CTW optoelectronic memristor suddenly increased from 1 µA to 90 µA, shifting its resistance state from HRS to LRS. The optoelectronic memristor can subsequently return to its HRS when the bias voltage reached V RESET of ≈ −0.7 V. Such low voltage can greatly benefit its energy consumption. Furthermore, as shown in Figure 3f, the HRS and LRS retention capabilities of the CTW optoelectronic memristor were measured. LRS1 and LRS2 corresponded to the low resistance state retention characteristics measured at I CC = 80 µA and 90 µA, respectively, using a read voltage of 0.02 V. It was noticed that both HRS and LRS exhibited long retention of 10 3 seconds without any obvious degradation, and were very likely to last longer period according to the current resistance state trend, enabling superb stability. It was also worth noting that the LRS decreases slightly with increasing I CC , as reported in other papers, low I CC induces narrow filaments while increasing I CC can form wide filaments. Therefore, the multiresistance characteristics can be attributed to the width modulation of the conductive filaments. [25] The cumulative distribution of V SET and V RESET for 100 continuous cycles was depicted in Figure 3g, implying the range of V SET from 0.5 V to 0.8 V and the V RESET from −0.6 V to −0.3 V, respectively. The corresponding coefficients of variation (CV) were calculated to be 13.07% and 12.91% for V SET and V RESET , respectively, according to Equation (S1), Supporting Information. In addition, The CV comparison of the proposed CTW optoelectronic memristor with other memristors can be found in Table S1, Supporting Information. The above results encouragingly suggested small cycle-to-cycle variation and excellent repeatability. Additionally, the ratio of HRS to LRS among 100 cycles under RS mode was also calculated in Figure 3h, giving rise to a switching ratio of ≈10 2 . As shown in Figure 3i, the slope of the CTW optoelectronic memristor from HRS to LRS is >7.68 mV dec −1 .
This ultralow switching slope is very useful to implement the array-level operation, due to its larger read voltage margin and faster reading speed. [26] Table 1 summarized the corresponding figures of merit of the recently reported memristor using TiO 2 as the functional layer. Compared with other TiO 2 -based memristors, the TiO 2based memristor with CTW structure showed much lower V SET and V RESET , which was beneficial to low-power storage and computing. Furthermore, the CTW optoelectronic memristor also led to much longer endurance, which was highly important for practical applications.
Similar to its unannealed counterpart, the annealed CuAlAgCr alloy was also found to exhibit excellent optical and electrical properties. The optical characteristics of the CuAlAgCr alloy films were studied in Figure S2, Supporting Information, which revealed the transmittance of CuAlAgCr films after annealing is greatly improved compared with that before annealing. Therefore, on this basis, we further investigated the response of the optoelectronic memristor with CTW structure to the optical signal. The electrical characteristics of the annealed CTW optoelectronic memristor in response to illumination were investigated by applying a sweeping voltage to the top CuAlAgCr electrode. The I-V characteristics of the annealed CTW optoelectronic memristor with and without exposure to illumination were shown in Figure 4. Figure 4a-c showed the I-V curves of the annealed CTW optoelectronic memristor without illumination, with the laser illumination (λ = 405 nm and P = 100 mW) and after removal of illumination for 5 s, respectively. When immune to illumination (Figure 4a), the response current of the optoelectronic memristor remained small, which was likely due to the reduced defects and oxygen vacancies in the annealed TiO 2 film. [33] The resistive switching effect of the CTW optoelectronic memristor was significantly enhanced under the illumination of λ = 405 nm laser compared with that of without illumination, and the response current was significantly increased up to the micro-ampere level. It is intriguing to note that the resulting current in fact decreased when removing the illumination for 5 s (Figure 4c), still much higher than the current value in the dark environment ( Figure 4a). This simply meant that the current response can sustain for a certain time even if the stimulus was removed, thus indicating its nonvolatility. This undoubtedly enabled the annealed memristor to be switched between "ON" and "OFF" states. The aforementioned photoresponse characteristics of the CTW optoelectronic memristor can be further verified by applying a small reading voltage (0.1 V) and detecting the device resistance, as shown in Figure 4d. When the laser (λ = 405 nm) was irradiated to the TE of the CTW optoelectronic memristor, the resulting current of the device increased sharply within 194.7 ms, and the ratio of the current before and after irradiation reached ≈ 10. Figure 4e showed the retention time of the CTW optoelectronic memristor under different wavelengths of laser irradiation. It was found that the CTW optoelectronic memristor exhibited different resistance states under laser irradiation with different wavelengths. By decreasing the wavelengths of the laser (red, green, and UV laser), the photo response current of the CTW optoelectronic memristor gradually increases. This is because with the decrease of the wavelength, the laser energy increased and the transmittance of CuAlAgCr film increased, leading to more electron-hole pairs generated inside the TiO 2 . Therefore, different resistance levels can be achieved by either maintaining the device under dark condition or irradiating it with different wavelengths, realizing its photoelectric-based multilevel storage. Such photoresponse currents for different laser wavelengths and laser powers were shown in Figure 4f, which indicated that either decreasing the wavelength or increasing the optical power can effectively enhance photoresponse current.
The design and manufacture of novel brain-like synaptic devices are important for the implementation of neuromorphic computing systems. In biological synaptic systems (Figure 1b), the neurotransmitters released from the presynaptic membrane, after diffusing through the synaptic cleft, can activate the receptor in the postsynaptic membrane and continuously modulate the synapse weight, which usually refers to synaptic plasticity. [21] Synaptic plasticity has been commonly considered the origin of brain learning and memory, which thus needs to be mimicked accurately in order to build a real brain-like system. [34] To explore the synaptic behavior of the studied CTW optoelectronic memristor, its programming current in response to a variety of applied voltages and pulse number was measured, resulting in Figure 5. It was found that sweeping positive voltage back and forth (0 V → 3 V → 0 V) led to a gradual increase in the resulting current, as illustrated in Figure 5a. In contrast, the programming current however underwent a gradual decrease when applying a negative voltage sweep (0 V → −3 V → 0 V), as reflected in Figure 5b. The I-V responses revealed in Figure 5a,b clearly corresponded to the potentiation and depression behaviors of the biological synapse. To further demonstrate the above speculation, the programming current in response to a series of voltage pulses having a magnitude of 3 V and a width of 50 ms was also assessed, giving rise to Figure 5c. As can be seen from Figure 5c, the programming current continuously boomed by increasing the number of the applied pulses, undoubtedly proving its potentiation phenomenon. Additionally, after the removal of the electrical pulses, the resulting current experienced a slow relaxation for a few seconds. Such degradation promisingly reproduced the shortterm plasticity (STP) of the biological synapse corresponding to a phenomenon that the synaptic potentiation can only last a few seconds or minutes. One typical form of STP is PPF. PPF is a phenomenon that postsynaptic potentials evoked by an impulse are increased when the impulse closely follows a prior impulse.

www.advelectronicmat.de
The ratio of the resulting current from the second pulse to that from the first pulse is called the PPF index, defined as , where I 1 and I 2 are the currents measured after the first and second spike pulse, respectively. As it plays an important role in decoding temporal auditory or visual information, here we simulated the PPF phenomenon in the CTW optoelectronic memristor. When two spike pulses with the same amplitude of 3 V were applied to the CTW optoelectronic memristor, the induced current by applying the second pulse was obviously higher than that generated by the first pulse, clearly reflected in the inset of Figure 5d. The dependence of the PPF index on pulse interval (Δt) was shown in Figure 5d. It was noticed that the PPF index decreased with increasing the pulse interval (Δt from 100 ms to 500 ms), well matching the following exponential function: where, τ is the relaxation time, and C and y 0 are constants . By fitting the experimental data (the red line in Figure 5d), the value of the τ, C, and y 0 are ≈77.74 ms, 194.63, and 9.3, respectively. This tendency is evident in the behavior of human memory, which pertains to the behavior of biological synaptic responses. [35] When the interval of the pulse of perception decrease, the memory effects can be reinforced. Furthermore, the power consumption of the annealed CTW optoelectronic memristor under electrical stimulus was estimated for bio-synapse applications. According to Equation (S2) and (S3), Supporting Information, the power consumption and energy consumption were calculated to be ≈3 µW and 150 nJ, respectively. In addition, the power consumption and energy consumption of various optoelectronic memristors from previous works have been studied and compared, as summarized in Table S2, Supporting Information. It can be observed that the power consumption and energy consumption of the CTW optoelectronic memristor are at an acceptable level.
Another fascinating feature of the designed device arouse from its ability to mimic STP behavior when only subjected to the optical stimulus, resulting in Figure 5e,f. As shown in Figure 5e, the illumination of a 20 mW laser pulse of 2 s made the resulting current gradually increase to a stable value, which subsequently returned to its initial value within 1 s after the light was removed. This encouragingly simulated the biological STP function, which was also known as STM in psychology. It was also possible to achieve the long-term memory (LTM) function by means of the CTW optoelectronic memristor, as demonstrated in Figure 5f. As can be seen from Figure 5f, changing the laser pulse width to 12 s led to a larger current when compared to the pulse with a smaller width. It was observed that when the light stimulus was removed, the light-induced current decayed spontaneously to a certain value that was higher than its initial current without illumination. Such current can maintain at least 5 s without suffering from attenuation, obviously longer than the case with larger illumination width. It is therefore expected that altering the illumination parameters such as intensity or width, can effectively extend its data retention time, thus realizing the transition from STM to LTM. Moreover, these results revealed that the learning, forgetting, and relearning behaviors can also be mimicked in the CTW www.advelectronicmat.de optoelectronic memristor. It was indicated in Figure 5f that applying and removing the light source can effectively boom and attenuate the resulting current, respectively, which corresponds to the "learning" and "forgetting" processes. It should be noticed that only a 3 s period (t 1 ) is needed to achieve the first learning, while the time taken to repeat such learning process (i.e., relearning) was reduced to 0.3 s (t 2 ). Such a trait also exhibited good agreement with the real brain working mechanism. Hence, the designed CTW optoelectronic memristor here enabled a successful imitation of the biological synapse with dual electrical-optical stimulus. Table 2 summarized the corresponding figures of merit of the recently reported optoelectronic memristor achieved via various structures. Compared with other structures, the CTW optoelectronic memristor designed in this work allowed for a high turn-on ratio and a mediumly low light-induced current, thus significantly reducing the power consumption and keeping the readout signal discernible. More importantly, the designed CTW optoelectronic memristor can be operated in the visible band, which is seamlessly compatible with human visual perception applications in contrast to previously reported structures.
In order to explain the electrical-optical duality of the CTW optoelectronic memristor, the resistance-switching mechanism of the CTW optoelectronic memristor was explored. We studied the conduction mechanism of the CTW optoelectronic memristor before annealing. The conduction mechanism was investigated according to the fitting of the I-V characteristic curves in HRS and LRS regimes, as shown in Figure 6. The I-V curve collected at I CC = 90 µA, divided into three segments, was chosen to demonstrate its conduction mechanism, resulting in Figure 6a. Note that to clearly establish the dependence of the resulting current on the applied voltage, a log-log scale was adopted for Figure 6b-d. The HRS regime initially showed a slope of 1.06, indicating a typical ohmic conduction (I ∝ V). This can be attributed to the presence of thermal-induced free electrons. Continuously increasing the applied voltage brought the slope up to 1.99, following Child's law (I ∝ V 2 ). Such dependence of current on the square of the applied voltage may arise from the fact that the injected carriers start to surpass the equilibrium concentration and partly fill the traps in the functional layer. Further increasing the bias voltage makes all traps filled and the generated carriers can be therefore transported to the conduction band. Such a conduction mechanism results in an abrupt increase of resulting current, corresponding to a slope of 4.67 (I ∝ V n , n > 2). The above analysis for the I-V characteristics is consistent with the Space Charge Limited Current (SCLC) conduction mechanism. As a consequence, the resistance-switching mechanism of the unannealed CTW optoelectronic follows the SCLC conduction mechanism. Herein, the relationship between the current density and the bias voltage of the memristor is: [39] where, θ is the ratio of free electrons to electrons captured by the trap, ε 0 is the vacuum permittivity, ε r is the relative permittivity of the dielectric material, μ is the electron mobility, V is the applied voltage and L is the thickness of the film. The slopes of the LRS regime and negative voltage regime were found to be 0.99 and 1.05, respectively. Both slopes were close to 1, which was obviously dominated by the formed conductive filaments, and thus showed typical ohmic conduction (Figure 6d). [24] To experimentally confirm the localized nucleation characteristics of conductive filaments. As depicted in Figure 3a,e, during the SET process, the current abruptly increases, indicating that resistance switching is a result of filament formation. [40] In order to further verify that the conduction mechanism of the unannealed CTW optoelectronic memristor is based on conductive filaments, both experimental and theoretical calculations have been conducted, as illustrated in Figure 6e-g. Figure 6e shows the schematic diagram of cutting the TE into two parts (denoted as TE1 and TE2), which can be employed to study the localized nucleation property of conductive filaments to confirm the filamentary switching. [41] First, the voltage was applied on the TE to make the device in LRS, and its resistance state was read using a 0.1 V voltage (See the red line in Figure 6f). Subsequently, the TE was divided into TE1 and TE2 by a tungsten probe, and the corresponding resistance state of TE1 and TE2 at a 0.1 V read voltage were individually measured, as shown in Figure 6f (blue and black lines). Notably, when the voltage was applied on TE2, the resistance of the CTW optoelectronic memristor exhibited HRS, seen from the black line of Figure 6f. This supports that the conduction state of the CTW optoelectronic memristor is not a result of the interface effect of uniform current distribution, but rather influenced by the conductive filament formed locally. [42,43] The resistive switching behavior of the proposed device with CuAlAgCr TE is likely to be attributed to the formation of Cu-assisted conductive filament through the TiO 2 layer, as demonstrated by conventional electrochemical memory. [44][45][46] To further support our hypothesis, we performed the firstprinciples calculations to evaluate the isosurface charge density for doping of Cu inside the TiO 2 layer, using the Vienna Ab-initio Simulation Package (VASP). The optimized structure with Cu dopants is exhibited in Figure 6g, where silver, red, and blue colors correspond to Ti, O, and Cu atoms, respectively. The resulting isosurface charge density is plotted in Figure 6h, where light blue and yellow colors represent charge Adv. Electron. Mater. 2023, 9, 2201320 www.advelectronicmat.de accumulation and depletion regions, respectively. It is obvious from Figure 6h that charge accumulations are not only present along with the O atoms, but also along with Cu atoms. This implies that the Cu doping greatly improves the charge attraction and thus allows for the formation of the conductive filaments along the c-direction. As a result, subsequent charge transferring along the initially nucleated filament can facilitate its growth and eventually drive it to bridge the TE with the bottom electrode, thus leading to resistive switching. The simulation computational details based on first-principles calculations further support that the conduction mechanism of the unannealed CTW optoelectronic memristor is the conduction filaments comprised of Cu ions. Differing from the case without annealing, the conductivity of the annealed CTW optoelectronic memristor can be significantly modulated upon being subjected to the optical illumination pulse and electrical stimulation pulse. In the initial state, the Schottky barrier (SB) in the CTW optoelectronic memristor resulting from the work function difference of the TiO 2 and Cr (Figure 7a) is formed (see the following first-principle calculations). Since the carriers are trapped by the defect levels, there are no free carriers for conducting, so the annealed CTW optoelectronic memristor is in a high resistance state. This is due to the fact that when the CTW optoelectronic memristor was annealed at 500 °C for 2 h in the air environment, active metals of the high entropy electrode are oxidized, resulting in more oxygen vacancies, which provides the extra defect energy levels. Upon being exposed to laser light (Figure 7b), electrons trapped in the defect energy level can be excited to the conduction band and become free electrons to participate in conduction. Therefore, the conductance of the CTW optoelectronic memristor increases, and the memristor is transformed to the LRS. On the contrary, when the laser source is removed, free electrons may be recaptured by the traps in defect energy level, thereby increasing the SB depletion width (Figure 7c), and the CTW optoelectronic memristor returns to HRS. [8] In order to further verify the conduction mechanism of the CTW optoelectronic memristor, first-principles calculations were performed on the proposed CTW devices. Figure S3, Supporting Information and Figure 7d show the structures of the anatase TiO 2 (101) plane, CuAgAlCr-TiO 2 heterojunction, and Cr-TiO 2 heterojunction before and after optimization respectively. Figure 7e shows the density of states for the TiO 2 , Cr-TiO 2,

www.advelectronicmat.de
and CuAlAgCr-TiO 2 structures respectively. The blue and pink dashed lines represent the top of the valence band of TiO 2 and the Fermi energy levels at two heterojunction interfaces, respectively. Note that TiO 2 is a typical n-type semiconductor and its bandgap with anatase structure is 3.2 eV. [47] According to Equations (3) and (4), the SB at the Cr-TiO 2 interface is calculated to be 1.36 eV, while its counterpart at the CuAlAgCr-TiO 2 interface is 2.25 eV. When subjected to illumination, the electrons in the valence band of TiO 2 can be excited to the conduction band. Since the SB at the Cr-TiO 2 interface is smaller than that of CuAlAgCr-TiO 2 , a large number of electrons are apt to be injected into the Fermi level of Cr, which facilitates the separation of electron/hole pairs in TiO 2 and ultimately improves the device conduction capability. Figure 7f, g exhibit the differential charge density of the Cr-TiO 2 and CuAlAgCr-TiO 2 structures, respectively, and the plane-averaged differential charge densities are depicted on the left part of Figure 7f,g. The yellow and cyan parts represent an increase and a decrease in charge density, respectively. It is obvious to visually discriminate the charge transfer trend at two heterojunction interfaces from resulting differential charge density distributions. It can be seen from Figure 7f,g that the charge transfer ability at the Cr-TiO 2 interface (Figure 7f) is much stronger than that of Figure 7g, and more electrons are gathered around the oxygen atoms at the heterojunction interface, thus forming an interfacial electric field from the metal side to the TiO 2 surface. When suffering from photo-excitation, the electrons inside the TiO 2 conduction band are liable to migrate to the Cr side under the www.advelectronicmat.de aid of the interfacial electric field, thereby reducing the recombination probability of the electron-hole pairs inside the TiO 2 layer. Such a phenomenon can promisingly trigger the generation of more carriers, and remarkably enhance the electrical conductivity of the CTW optoelectronic memristor.
The above results revealed that CTW optoelectronic memristor has an excellent optical synaptic function, but the realization of the vision bionic system function needs to realize the perception ability of external information. To prove this, we constructed a 5 × 5 CTW optoelectronic memristor array to realize the basic function of image sensing. As shown in Figure 8a, each storage unit inside the fabricated array corresponded to one image pixel. To mimic the visual bionics, a reverse bias voltage was first applied to each cell to restore it to HRS. An N-shape optical mask (see Figure S4, Supporting Information) was used under the 405 nm visible light irradiation, as shown in Figure 8b. The N-shape optical mask was implemented here to prevent other image pixels from being stimulated by laser light. It can be obviously observed that the storage units under laser irradiation returned to their LRS. As shown in Figure 8c the memory effects of the letter N were observed. The above demonstrates the image perception capability of the CTW optoelectronic memristor array devices. In addition to its vision analog, we also assessed its potential for in-memory computing applications such as the Boolean-logic function. As shown in Figure 8c, the CTW optoelectronic memristor for "AND" logic had one electrical pulse (Ele) input and one optical pulse (Opt) input, which were considered as two independent inputs. The photo-induced current (E out ) was regarded as the output signal. Figure 8d showed the schematic diagram of the "AND" circuit based on the CTW optoelectronic memristor. The designed circuit only gave rise to logic 1 (above 0.5 µA), when both optical and electrical inputs were present (i.e., triggering LRS); otherwise, the device remained as HRS, thus outputting logic 0 (below 0.5 µA). The corresponding truth table, as shown in Figure 8d, was consistent with the "AND" logic.

Conclusion
In this work, we designed and fabricated an optoelectronic memristor based on a CTW structure. Before annealing, the CTW optoelectronic memristor can achieve a stable and controllable switching between TS and RS. When a small I CC is applied, the CTW optoelectronic memristor exhibits TS characteristics, but when the I CC increases, the CTW optoelectronic memristor exhibits RS characteristics. In addition, it has a low switching set voltage (0.8 V), a larger memory window (10 2 ), and long data retention (10 3 s). Furthermore, typical biological synapse features, such as LTP, PPF, STM, optical learning, and forgetting behaviors were successfully reproduced by the developed CTW optoelectronic memristor via either optical or electrical stimulus. Furthermore, we demonstrated a basic logic function unit based on the CTW optoelectronic Figure 8. a) A microscope image of 5 × 5 CTW optoelectronic memristor crossbar array. b) Schematic structure of 5 × 5 CTW optoelectronic memristor crossbar array, in the illustration with N-shape optical mask under laser illumination. c) Normalized photo-response current of each storage unit in 5 × 5 CTW optoelectronic memristor crossbar array after light stimuli. d) CTW optoelectronic memristor logic implements an "AND" operation, along with its truth table and output current value.
www.advelectronicmat.de memristor, which realizes the "AND" operation. Finally, the CTW optoelectronic memristor can respond to visible light and its photoresponse current is increased to microamps (µA) level, which greatly reduces the difficulty of peripheral circuit design and shortens the distance from a single device to system integration.

Experimental Section
Fabrication of CTW Optoelectronic Memristor: The silicon (Si) wafers with 300 µm SiO 2 were first cleaned sequentially in acetone, ethanol, and deionized water for 15 min using ultrasonic waves, followed by a blow dry with a nitrogen gun. A 100 nm tungsten (W) metal film was subsequently deposited on a 300 µm thick SiO 2 /Si wafer as the bottom electrode by magnetron sputtering, and 30 nm titanium dioxide (TiO 2 ) film was deposited on the W layer as the functional layer. Finally, a 100 nm CuAlAgCr alloy film was sputtered on the TiO 2 film as the TE via magnetron sputtering with a shadow mask. To further study the performance of CTW optoelectronic memristor, the same devices at 500 °C for 2 h in the air environment was annealed.
Characterization and Measurement: The XRD patterns of the TiO 2 and CuAlAgCr films were obtained using a Bruker D8 Advanced diffractometer. The cross-sectional images and element distribution characterization of the CTW optoelectronic memristor were performed using field emission scanning electron microscopy (Apreo2C). The transmission and absorption spectra were recorded by an ultraviolet and visible (UV-vis) spectrum analyzer (PE Lambda950). The I-V characteristics of the CTW optoelectronic memristors were measured using the semiconductor parameter analyzer (Keithley, 4200A-SCS) in the probe station. The optoelectronic memristors described above were fabricated and characterized at room temperature. The CTW optoelectronic memristors were illuminated by the commercial UV, green, and red lasers, which were composed of the laser diode, collimating lens, focusing lens, and other optical modules. The corresponding peak wavelengths are 405 nm, 520 nm, and 650 nm, respectively. Their normal working light-output power was calibrated to 200 mW through a commercial optical power meter with the standard silicon photodetector (S121C, 400-1100 nm, 500 nW -500 mW). A series of neutral density filters were exploited to adjust the light-output power from the various lasers to achieve different light-output powers, which were also calibrated by the optical power meter.
Computational Details: All spin-polarized calculation results were obtained by first-principles calculation, implemented in VASP. The projector augmented-wave (PAW) method implemented in the VASP code was utilized to describe the interaction between the ionic cores and the valence electrons. The generalized gradient approximation (GGA) parameterized by Perdew, Burke, and Ernzerhof (PBE) was employed to describe the exchange-correlation potential in the standard Density Function Theory (DFT) calculations. For the calculations of the unannealed CTW optoelectronic memristor under the electronic stimulus, the cut-off energy for a plane wave is set to 450 eV. The energy criterion is set to 10 −5 eV in the iterative solution of the Kohn-Sham equation. All the structures are relaxed until the residual forces on the atoms have declined to less than 0.02 eVÅ −1 . The Brillouin zone integration is performed using a 2 × 4 × 3 k-mesh.
For the calculations of the annealed CTW optoelectronic memristor under the light stimulus, an energy cutoff of 500 eV and 2 × 2 × 1 k-point with a G-centered k mesh was used for structure optimization. To describe adequately the strongly localized Ti 3d orbitals and obtain an approximate experimental band gap value of anatase TiO 2 , the GGA+U scheme of Dudarev et al. was applied, where the Coulomb U and exchange J parameters were combined into a single effective Hubbard U-parameter U eff = U− J. The U eff value for the Ti centers was chosen according to the previous literature to be 8 eV. [48,49] The fcc-type high-entropy alloy (CuAlAgCr) model (1:1:1:1) was obtained through a comprehensive search using the mcsqs method for Special Quasi-random Structures (SQS) generation. For a given number of atoms in each supercell, SQS represents an approximation of the best periodic supercell for a truly disordered state. Here, the pair-correlation functions were used to generate the model, owing to the tradeoff between the modeling efficiency and the computational consumption caused by the chemical complexity of the quaternary alloy.
Schottky barrier height (SBH) is defined as the difference between the metal Fermi level ( E F m ) and the majority-carrier band edge of the semiconductor at the interface. SBH at the interface of the metal and anatase TiO 2 (101) surface (denoted as metal-TiO 2 ) was calculated by the following equation: [50] ( ) ( )

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.

Data Availability Statement
The data that support the findings of this study are available in the supplementary material of this article.