Fully Photon Controlled Synaptic Memristor for Neuro‐Inspired Computing

The emerging optoelectronic memristive synapses having the advantages of both optics and electronics exhibit a great potential in neuro‐inspired computing, which is a new generation of artificial intelligence. Herein, a light stimulated synaptic memristor (LSSM) based on ZnO/Zn2SnO4 heterostructure is prepared with the characteristics of reversibly tunable conductance states by varying the wavelength of the incident light. The synaptic feature of this fully photon controlled memristive synapse is revealed by potentiation and depression behaviors stimulated by violet and red light pulses, respectively. Similar to biological brain, the device demonstrates the dynamic learning and forgetting behavior. All‐optically driven and bio‐vision inspired image processing function such as contrast enhancement is exemplified. The international Morse code for Arabic numerals (0–9) is also successfully conveyed by patterned light pulses and suggests the device's potential in the field of optical wireless communication for human–machine interface. Classical Pavlovian conditioning (associative learning) is successfully demonstrated through visible light induction. Finally, the device can realize the recognition application of Zalando's article image through the simulation based on Hopfield neural network (HNN). This work provides a promising approach toward optoelectronic neural systems and human–machine interaction technologies.


Fully Photon Controlled Synaptic Memristor for Neuro-Inspired Computing
Saransh Shrivastava, Lai Boon Keong, Sparsh Pratik, Albert S. Lin, and Tseung-Yuen Tseng* DOI: 10.1002/aelm.202201093 consumption and inherent information processing speed limitation. [1] Our biological human brain, which consists of up to 10 11 -10 15 neuron-synaptic junction units, is a more powerful parallel computing system in terms of fault tolerance, energy efficiency, information transmission, and other crucial functions. [2,3] To overcome the von Neumann bottleneck, several artificial synapses based on electronic devices such as memristors, [4] and field effect transistors [5] have been investigated to mimic the basic computational function of nervous system called "synaptic plasticity" [6] by modulating the device's conductance (synaptic weight). The conventional electronic [7] and optoelectronic [8] synaptic devices have utilized pure electrical stimulus and an essential combination of electrical and optical stimuli, respectively, to operate the devices. Some analogous issues of computational latency, higher energy consumption, and hardware redundancy could occur in those devices owing to bandwidth connection density trade-off. [9,10] Such kind of problems make them less tempting. The cognizance of fully photon modulated synaptic memristor was getting researchers' attention in an impressive manner because they can get over the above mentioned issues with the benefits of ultra-fast operation speed, no electrical interconnect power losses, and high bandwidth. [9,11,12] Therefore, a fully light stimulated synaptic device is highly desired.
Regarding the photosensitive materials, two-dimensional (2D) materials, [13] halide perovskites, [14] organic-inorganic hybrids, [15] metal oxides, [16] and sulfides [17] have led to the exciting progress in the field of optoelectronic applications. Among them, metal oxide semiconductors with wide bandgap (>3 eV) combine the properties of high conductivity, amenability for n-type doping, unique electronic structure, and visible light transparency, which offer a variety of their uses in commercialized optoelectronic applications. [18] They have been investigated extensively due to their ecofriendly nature and abundant reserves. ZnO and Zn 2 SnO 4 (ZSO) are two prominently studied metal oxide semiconductors with wide optical bandgap of 3.3 and 3.88 eV, respectively, and have great potential for visible light sensors because of their fascinating optical properties and strong absorption in ultraviolet (UV) region. [19][20][21] The emerging optoelectronic memristive synapses having the advantages of both optics and electronics exhibit a great potential in neuro-inspired computing, which is a new generation of artificial intelligence. Herein, a light stimulated synaptic memristor (LSSM) based on ZnO/Zn 2 SnO 4 heterostructure is prepared with the characteristics of reversibly tunable conductance states by varying the wavelength of the incident light. The synaptic feature of this fully photon controlled memristive synapse is revealed by potentiation and depression behaviors stimulated by violet and red light pulses, respectively. Similar to biological brain, the device demonstrates the dynamic learning and forgetting behavior. All-optically driven and bio-vision inspired image processing function such as contrast enhancement is exemplified. The international Morse code for Arabic numerals (0-9) is also successfully conveyed by patterned light pulses and suggests the device's potential in the field of optical wireless communication for human-machine interface. Classical Pavlovian conditioning (associative learning) is successfully demonstrated through visible light induction. Finally, the device can realize the recognition application of Zalando's article image through the simulation based on Hopfield neural network (HNN). This work provides a promising approach toward optoelectronic neural systems and human-machine interaction technologies.

Introduction
Due to the physical separation of data storage and processing units, von Neumann architecture based conventional computing systems are facing the great challenges of high energy www.advelectronicmat.de All-optically modulated artificial synapses [9,[22][23][24][25][26][27] have been reported in previous works and listed in the Table 1, where excitatory and inhibitory synaptic behaviors have been exhibited in an all-optically pathway by applying a wide range of light spectra from UV to Infrared. Hu et al. recently reported IGZO thin films based bilayer structured (homo-junction) synaptic memristor, [22] where the blue and near infra-red (NIR) light pulses were used for optical SET and RESET behaviors, respectively, in the same device. Ji et al. reported an organic semiconductor based device, which can mimic the potentiation/depression behavior by 660/445 nm light pulses. [27] To the best of our knowledge, for an all-metal oxides based device with different band gap materials (hetero-junction), the repeatability of optically stimulated potentiation and depression phenomena by controlling the wavelengths of visible light only has not been realized till now.
In this work, we fill this gap by proposing a n-n heterojunction (n-ZnO/n-ZSO) based bilayer structured visible light stimulated memristive synaptic device (ITO/ZnO/ZSO/ITO/ Glass). Because of non-complex nature and ease of operation, optical SET and RESET of a memristor seems a convenient process. Therefore, in order to follow those behaviors, we utilize violet, and red light sources, respectively. The below band gap light (≈3.06 eV) induced holes determine the conductance states of the device by de-trapping of electrons from the interfacial oxygen vacancies (V O s). The fabricated device serves as a synaptic emulator, as we successfully manifest long-term potentiation/depression (LTP/LTD) and spike timing dependent plasticity (STDP) by electrical stimulation. When the device is exposed to violet light, we simulate the synaptic transition process from short-term memory (STM) to long-term memory (LTM) by increasing the number, and intensity of applied light pulses. We demonstrate dynamic learning and forgetting process and image processing function such as contrast enhancement of the light stimulated synaptic memristor (LSSM). We also prove that our optoelectronic synapse can react to visible light that represents the International Morse code, in which Arabic numerals (0-9) can induce a distinct post synaptic current (PSC) amplitude response. In addition, the classical Pavlov conditioning is demonstrated in fully optical pathway. The biovision inspired neuromorphic computation is performed by the recognition of 28 × 28-pixel gray scale image, where the constructed Hopfield neural network (HNN) model is successfully simulated with optically driven data. This work underscores a great potential of oxide based optoelectronics for building visible light stimulated neuro-inspired computing system.

Device Structure and I-V Characteristics
A schematic representation of n-ZnO/n-ZSO based two terminal vertical structured synaptic memristor is depicted in Figure 1a. Its cross-sectional Transmission Electron Microscopy (TEM) image captured on the scale of 100 nm (Figure 1b) and 20 nm (Figure 1c) indicate the thicknesses of resistive switching layers, ZnO (10 nm) and ZSO (40 nm). Layer-by-layer distributions of each element (In, Sn, Zn, and O) throughout the device was determined by the Energy Dispersive X-ray Spectroscopy (EDS) mapping and the result is shown in Figure S1 (Supporting Information). This device shows a high transmittance (>85%) over a wide range of visible spectrum (Figure 1d), and inset depicts an optical photograph of the original device. The I-V characteristics ( Figure 1e) indicate that the device is capable to switch its current states in both positive and negative bias regions. A forming process, which was required to switch ON the device, designates its high resistance in the pristine state. The same device can be switched back to OFF state by applying a bias of −1.3 V, called as RESET process. Afterward, the SET process is occurred by sweeping the DC voltage (0→+2 V→0) in positive bias region with a controlled compliance current (CC) of 200 µA. Both SET and RESET processes are quite gradual as well as stable up to 1500 consecutive DC cycles with the switching voltages of +1.22 and −1.3 V, respectively. The gradual resistive switching indicates its analog behavior, which makes it a promising candidate for the emulation of synaptic functions including LTP/LTD, and STDP and neuro-inspired computing applications. [28] In addition, DC endurance characteristics (Figure 1f) was measured during the SET process, where we observe a large ON/OFF current ratio (≈750). This large difference in current values measured at ON and OFF states is attributed to the Redox reaction [29] within the switching layer, which results in the formation and rupture of large oxygen vacancy based conductive filament (CF) in the oxide region, respectively.
This work "-" shows that the characteristics were not measured and shown. www.advelectronicmat.de

Electrical Synaptic Performance
To reveal our memristor's feasibility as an artificial synapse, we stimulate it with a train of electrical AC pulses and consequently, synaptic plasticity, i.e., LTP/LTD is achieved in the form of conductance (synaptic weight) states. The strength of the applied pulse train is depicted in Figure S2 (Supporting Information), and the corresponding resultant states of the device conductance during LTP/LTD for first 10 and last 10 out of total 680 cycles are depicted in Figure 2b,c, respectively. The rest of the dataset (11th to 670th cycles) for same experiment are shown in Note S3 (Supporting Information). The nonlinearities (NLs) of experimentally obtained LTP/LTD curves were calculated by fitting their profiles with Equations (S1)-(S3) (Supporting Information), [30] and the curves are depicted in Figure 2d. STDP is a crucial synaptic function to know about how the information is transmitted by a synapse in a neural network when the sequences of pre-and post-synaptic spikes vary along with the time interval (∆t) between them. To investigate the STDP response of our synaptic memristor, we have designed and applied a pulse scheme ( Figure S3f, Supporting Information) with the combination of asymmetric pre-and postsynaptic spikes. As a result, the change in conductance (∆G) reduces with increasing ∆t (Figure 2e), where LTP is recorded for ∆t < 0, while LTD is attained for ∆t > 0. We employ the following equations for the calculation of ∆t and ∆Gpost pre Here, t post and t pre are firing times of the post-and presynaptic spikes, respectively. G final and G initial are the device conductance after and before applying the spikes, respectively. This type of asymmetric STDP response is referred as anti-Hebbian learning. [31] In our previous work, the dependency of NL of STDP curve on ∆t was already discussed. [32] The dependency of ∆G on the G initial , spike's shape and ∆t, facilitates selfadaptation of the conductance states and emulates stable Hebbian learning in the brain. The STDP curves are fitted well by using Equation (3) exp( Here, C 1 and C 2 are the scaling factors, while τ 1 and τ 2 are the time constants for positive and negative time intervals, respectively. Their calculated values by data fitting are 1.42, −1.25, −130.92, and 79.5 s, respectively.

Optical Response
We have investigated the response of our transparent memristor under the irradiation of two different wavelengths of visible light. The pictorial representation of ZnO/ZSO based LSSM is depicted in Figure 3a. By the irradiation of violet (λ = 405 nm) and red (λ = 633 nm) LASER lights individually, PSC of the device changes (shown in Figure 3b) in a positive manner and exhibits an increase in the synaptic weight (conductance). However, a relatively poor positive photoresponse is observed at red light excitation compared to violet light because of lower photon energy. After removing the light source, initially the PSC decays fast, but after some time, this decay (forgetting) rate becomes slower and slower. PSC does not reach to its initial state even for 5000 s of darkness relaxation time and depicts a long lasting persistent photoconductivity (PPC), [33] which is similar to biological synaptic function. This long PPC under the dark environment is attributed to the slow neutralization of photo-ionized V O s. With the help of this wavelength dependent photo-response, a figure of merit referred as responsivity (R) [34] of our LSSM was calculated and depicted in Note S4 (Supporting Information). A relatively lower responsivity was detected for longer (633 nm)  www.advelectronicmat.de wavelength when compared with shorter wavelength and high energy excitation (≈3.06 eV).

Optical Synaptic Performance
After demonstrating synaptic behavior of our memristor under electrical stimulation (Figure 2), now we investigated its synaptic activities under light exposure. According to psychological memory and forgetting model of human brain [24] (shown in Figure 3c), all the information received by the perceptual organs from the surrounding environment are registered in the human sensory system. Information inherited with attention are transformed into STM and stored for a short period. Herein, our LSSM exhibits stimulation number-dependent learning like a biological synapse. In our experiment (Figure 3d), the device was first exposed to two consecutive violet light pulses and displays an increase in PSC of 5.45 nA before turning off the stimulation. It induces STM behavior in our LSSM, which can be transmuted to LTM via receiving external information repeatedly. Violet light pulses with same intensity, width, and interval but with a larger number (15) demonstrate the LTM behavior with a further increase in PSC to 11.38 nA. This increased PSC relatively takes a longer time to decay than the PSC induced by small rehearsal stimulation number. Thus, STM to LTM transition can be observed in our LSSM with weak (STM) and strong (LTM) potentiation effects through varying the number of stimulations. As such, the Atkinson and Shiffrin human memory model [35] of cognitive learning is followed by our LSSM.
To show a better comparison among the PSCs upon stimulation with different pulse numbers, spike-number-dependentplasticity [27] (SNDP) ratio was calculated by the ratio of PSC values triggered by the 1st pulse to Nth pulse and the result is depicted in Figure S5a (Supporting Information). Besides, the STM to LTM transition can also be achieved by modulation of the pulse intensity. The response of the device to light pulses (width 10 s) with different intensities from 10 to 30 mW cm −2 at the read voltage of 0.1 V is shown in Figure 3e, where the larger PSC with longer memory time is achieved by increasing the pulse intensity. An increase in change of PSC (∆PSC) from 0.5 to 2.26 nA as a function of pulse intensity ( Figure S5b, Supporting Information) is also observed, demonstrating the effect of learning enhancement.
For human brain, learning and forgetting behaviors are very crucial and we can relate them with the LSSM's conductance states during and after the stimulation, respectively. [36] Here, we represented the number of applied violet light pulses as the times of training, while the PSC of the device is regarded as the memory level. It should be noted in Figure 4a (top panel) that as times of training increases through the learning process, the memory level is promoted. After removing the light source, the forgetting process starts (bottom panel Figure 4a), but the memory level does not drop to its initial state. Figure S6a,b (Supporting Information) depicts the PSC of the device as functions of light pulse number and the decay time, respectively. The PSC is reduced by ≈33% after a decay time of 4 min. The demonstration of this image memory function of our LSSM is similar to the dynamic learning and forgetting process of the human brain.
Image processing such as classification, contrast enhancement, and recognition is a vital function of human retina imaging. [36] To enhance the recognition and accuracy of images, contrast enhancement plays a crucial role. After realizing the STM to LTM transition by modulating the pulse intensity, the LSSM also realizes the contrast enhancement function as the human visual system. As we have seen that Figure 3e exhibits the change in PSC of the LSSM after irradiation at various light intensities. When compared with lower intensity, a relatively gradual decay in PSC is observed after irradiating with higher intensity. Because of this feature, Figure 4b depicts the evolution of normalized contrast (or current level) as a function of decay time. The color scale bar of 5 × 5 image array in Figure 4c represents the memory (current) level of the LSSM, whereas all marked positions in this image array (such as 1, 2, 3, 4, and 5) belong to PSCs after irradiation with intensities of 0, 10, 15, 20, and 30 mW cm −2 , respectively. The normalized values of these PSCs are also referred as initial contrasts, 0, 0.34, 0.62, 0.79, and 1, respectively (smallest PSC is normalized to 0 and largest PSC to 1).
Since position "1" in the image array belongs to 0 mW cm −2 intensity (means no irradiation), therefore at this position, no change in image contrast is illustrated in Figure 4c. After terminating the irradiation, the normalized contrast for 10 and 15 mW cm −2 light intensities (Figure 4b) starts to reduce rapidly at the initial phase (0-10 s) of decay time because of rapid decay in PSC after stimulation with lower intensity pulses. Consequently, the initial contrast at positions 2 and 3 gets lower after 10 s decay as illustrated in Figure 4c. On the other hand, a small reduction in normalized contrast after stimulation with 20 mW cm −2 light pulse supports the almost same image contrast (not much lower than initial) at position "4" in Figure 4c. Thus, the difference in PSCs is intensified with decay time, which evolves in an enhanced contrast of input image of "Swastika" (at position 5 with the largest PSC corresponding to 30 mW cm −2 ) by retaining a higher value of PSC during the decay time. This experiment demonstrates that our LSSM have potential application in artificial image processing.
Similar to the memory loss in biological system, we can depict the current decay in our LSSM. The rate of memory loss strongly depends on that how the information was learned by the human brain. In this experiment, the learning (stimulating) condition is affected by varying the light pulse number (such as 3, 5, 10, and 13) during the rehearsals. Figure 4d depicts the memory retention curves during the forgetting process as a function of time after the nth rehearsal, which are fitted well by using the Kohlrausch stretched exponential function [37] ( ) exp where I(t) is the PSC at time t; I 0 the maximum PSC during stimulation; τ the relaxation time constant, and β a stretching exponent ranging from 0 to 1. Here, the I(t)/I 0 decays more gradual after the rehearsal with larger light pulse number and can be referred as memory retention during the forgetting process. This phenomenon relates to the fact that something can be remembered much clear for longer time by human brain if there are more impressions at the learning time. Figure S7 (Supporting www.advelectronicmat.de Information) presents that τ of the forgetting process increases after getting stimulation with more pulse number. For the further characterization of non-volatile multilevel memory feature, a train of violet light pulses with a frequency of 0.2 Hz was applied to program our LSSM. 65 distinct levels of PSC with the LTP synaptic plasticity feature are produced throughout the writing process and shown in Figure S8a (Supporting Information). From the same figure, we extract the first 10 and last 10 states of PSC, which depict the distinct memory levels in Figure S8b,c (Supporting Information), respectively. The accumulation of ∆PSC during this consecutive stimulation is calculated and depicted in Figure 4e along with the visible light dose by using the equation below [38] light dose (J cm ) power density (W cm ) irradiation time (s) To demonstrate the potential of our LSSM as an optical wireless communication method for human-machine interfaces, the International Morse Codes [39] of Arabic numerals (0-9) were used as patterned violet light pulses. We can observe in Figure 4f, every digit of Arabic numerals can induce a distinct PSC response and these responses depict that our system can be implemented to light fidelity in the future.
In mammalian cognitive activities, classical associative learning behavior plays a vital role. [24,40] We optically emulate Pavlov's dog conditioning experiment as an instance of associative learning in our LSSM. Recent works [40,41] indicated that this conditional learning application was realized by combining the input conditions of voltage and light pulses. In this work, the light pulses of 633 and 405 nm wavelengths are used as bellringing (conditioned) and food-sight (unconditioned) stimuli, respectively (Figure 5a). The threshold PSC for salivation is set to 1.75 nA (gray shaded region in Figure 5b-e) and measurement is done after (60 s) removing the light source. At the first step, LSSM is irradiated with a train of 3 pulses of 633 nm (40 mW cm −2 , 0.2 Hz, and 5 s) which mimics the conditioned stimulus (CS). The measured PSC (1.6 nA) is below the threshold level indicating the absence of salivation (Figure 5b). At the next step, LSSM is irradiated with 3 pulses of 405 nm (40 mW cm −2 , 0.2 Hz, and 5 s), which simulates the unconditioned stimulus (US). As accepted, the PSC (2.01 nA) is above the threshold level, which would exhilarate leading to the salivation (Figure 5c). When both the light pulse trains irradiate LSSM simultaneously, then an enhancement in PSC (4.02 nA) much higher than the salivation threshold with long retention appears (Figure 5d). After this training phase, only bell ringing

www.advelectronicmat.de
(conditioned) stimulus is applied and kept the PSC (2.79 nA) above the threshold (Figure 5e). It can be allied with the furtherance of photo induced charge carriers after the training pulses.
To achieve the optical RESET behavior, the violet light irradiated device is subsequently exposed under red (633 nm) light and induced a stronger reduction in PSC compared to dark environment ( Figure S9, Supporting Information). This result indicates that red light can be utilized to RESET the violet light irradiated device. Herein, the LSSM is firstly irradiated with a train of 25 violet light pulses (405 nm, 40 mW cm −2 , 2 Hz, 0.5 s) and an expected increase in device conductance is observed. After removing the violet light source, the same device is exposed under a series of 25 red light pulses (633 nm, 55 mW cm −2 , 2 Hz, 0.5 s), consequently, a change from potentiation to depression is revealed indicating obviously the alloptically controlled characteristics. Figure 6a depicts 15 consecutive cycles of LTP and LTD under irradiations of 405 and 633 nm, respectively. The enlarged view of the second cycle and its corresponding NL during LTP/LTD behavior are shown in Figure 6b,c, respectively. The calculation of NL was done by using Equations (S1)-(S3). These normalized experimental data can be utilized in the calculation of learning efficiency of HNN based on our LSSM.
The bio-vision inspired neuromorphic computation was demonstrated by simulating HNN. [42,43] The weight map simulation is carried out by utilizing the normalized optically induced experimental data of LTP and LTD depicted www.advelectronicmat.de in Figure 6c. In this experiment, we classify 28 × 28 single channel or gray scale image of "Handbag" adopted from Modified National Institute of Standards and Technology (MNIST) fashion dataset [44] by designing an HNN model. The simulation details are explained in our previous reported work [45] and also mentioned in Note S10 (Supporting Information). Each pixel of the input image (Figure 7a) acts as a single synapse. The initial state of the 28 × 28 gray scale image is obtained by shuffling all the 784 synapses (normalized conductance data) and depicted in Figure 7b as noisy image. Figure 7c indicates the recalled picture after the training phase. We calculate the maximum recognition accuracy of ≈91% achieved by the HNN model simulated by optical LTP/LTD pulses after 23 iterations (Figure 7d).

Optical Switching Mechanism
The optical switching mechanism is determined by analyzing the band structures of ZnO and ZSO (Note S11, Supporting Information). ZnO has lower electron affinity and narrower band gap than ZSO. [32,46] When ZnO comes into contact with ZSO, then a type-II heterojunction band structure formed [47] ( Figure S11b ). [48] In our work, under the low band gap excitation (hν < E g ), the charged state of V O s is achievable via trapping the excited holes near the valence band maximum (VBM). This hole mediated transition is theoretically viable where the holes  are possible to be excited from acceptor defects by absorbing photons with energy less than the band gap. [49] Here, O wb 2− is the weakly bound oxygen and O i is the interstitial oxygen. When the device is irradiated with violet light (≈3.06 eV), then V O 0 are photo-ionized into V O 2+ at the ZnO/ ZSO interface (photo-induced V O 1+ are thermodynamically unstable due to higher formation energy) [50,51] and leads to a narrow barrier width (Figure 8a), resulting in an increase in PSC. Therefore, the barrier narrowing is the basic reason behind the optical SET behavior. The concentration of V O 2+ at the ZnO/ZSO interface is determined by X-ray photoelectron spectroscopy (XPS) depth profile analysis and plotted in Note S12 (Supporting Information).
When this violet light irradiated device is further exposed to red light, then two opposite reactions namely photo-ionization and deionization of V O s occur simultaneously and the resultant PSC reckon on the dynamic equilibrium between these two reactions. [22] Since the device was already irradiated with shorter wavelength light, therefore a large amount of V O 0 was already ionized. Consequently, a low concentration of V O 0 would be left at the deep level in the band gap of ZnO for the ionization by the subsequent exposure to longer wavelength light. In this case, the inverse process of the above reaction (i.e., deionization of V O 2+ ) would dominate by trapping free electrons, which were entered into the ZnO conduction band (CB) by passing over the barrier (Figure 8b). Consequently, the barrier region gets broaden and a decrease in the PSC is observed as an optical RESET behavior.

Conclusion
In summary, we fabricated an optoelectronic artificial synapse based on ZnO/ZSO heterostructure which exhibits fully photon controlled reversibly tunable conductance states. The device utilizes trapping and de-trapping of electrons by interfacial V O s. It allows the device to exhibit optical switching behavior where irradiation of visible light pulses with different wavelengths implement LTP (writing) and LTD (erasing) operations. Some basic functions of biological synapse such transition from STM to LTM, and dynamic learning and forgetting behaviors were successfully mimicked by this device. Meanwhile, the emulation of biological behavior through International Morse code and classical Pavlov conditioning were achieved in a fully optical pathway. HNN model was simulated with experimental data of optically driven potentiation and depression characteristics and recognized an image adopted from MNIST fashion dataset with an accuracy of 91% after 23 iterations. A simple two terminal structure with light stimulated current response represents a worthwhile approach for the realization of next generation optogenetics inspired neuromorphic computation.
Device Fabrication: A commercially available ITO coated glass substrate (20 × 20 × 0.7 mm 3 ) was cleaned by ultra-sonication with isopropyl alcohol and deionized water for 10 min each. After cleaning, it was dried by blowing nitrogen gas. A thin film of ZSO material was deposited on this cleaned ITO coated glass substrate by RF sputtering method. For deposition, a pure Argon environment was created and maintained inside the chamber at a work pressure of 1.33 Pascal. In order to complete the bilayer structure, a thin film of ZnO material was deposited above the ZSO layer by using the same sputtering method, but in the mixture of argon and oxygen (2:1 ratio) gases. Afterward, ITO as a transparent top electrode was deposited and patterned by using a shadow mask in the same sputter chamber. All oxide thin films were deposited at room temperature.
Electrical and Optoelectronic Characterization: Electrical and optoelectronic measurements were performed at room temperature and atmospheric pressure by using Agilent B1500A semiconductor parameter analyzer equipped with a probe station. Violet and red color Material Characterization: Thickness of the ZnO and ZSO thin films were obtained by cross-sectional image of the device observed by transmission electron microscope (HRTEM, JEOL JEM-2010F). Layerby-layer distribution and atomic concentration of each element (In, Zn, Sn, and O) were obtained by energy dispersive X-ray spectroscopy (EDS) and X-ray photoelectron spectroscopy (XPS, ULVAC-PHI Quantera SXM), respectively. Absorbance and transmittance spectra were obtained by a UV-vis spectroscopy (Hitachi U-3010).

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