Synaptic plasticity in zinc oxide-based flexible invisible transparent memristor by modulating oxygen concentration

Modern electronic equipment, including computer systems, are built using the traditional von Neumann computing architecture, which has its computational constraints. The conventional CMOS transistors used in these systems make it power consuming, with slower speed of operation for solving complex tasks. Even downscaling of transistors to follow Moore's law¹ is found to be not a solution. To remove the von Neumann bottlenecks and solve the issues of downscaling of devices, research is on the way for making the latest technologies to fulfill the requirement of fast operational speed and economical and minimal power consumption. Also, as the idea of big data advances and vast volume of formulated and unformulated data becomes more complicated, the limit of processing capacity causes high manufacturing cost and degraded power efficiency, with enhanced other costs. An engineering technique called “Neuromorphic Computing (NC)” aims to solve the current computing system's bottlenecks by mimicking the human brain's function with computer processors.² 

NC technique got the inspiration by looking at how the human brain works. There are roughly 10¹¹ neurons in the human brain, and each neuron is connected by synapses. When a signal travels from a pre-synaptic neuron to a post-synaptic neuron, it typically involves the release of neurotransmitters from the pre-synaptic neuron into the synaptic cleft. These neurotransmitters then bind to receptors on the post-synaptic neuron, leading to changes in the post-synaptic neuron's membrane potential and resulting in the generation of an action potential. The process by which the strength of synaptic transmission is altered is called synaptic plasticity. There are two forms of synaptic plasticity—long-term potentiation/depression (LTP/D)—that result in an increase/decrease in the number or sensitivity of neurotransmitter receptors, respectively. To develop NC hardware, a suitable device is needed that can emulate the biological synapse and perform analog computation efficiently. Given this, the memristor is found to be a suitable candidate that plays a fundamental role in NC applications.

After the development of a memristor in the HP laboratory in 2008, many researchers used their ideas to fabricate different memristive devices and proved that the devices fabricated have great potential for applications in various engineering domains. Memristors have a simple structure with an insulator/semiconductor sandwiched between two metal electrodes. By choosing the proper materials for metal electrodes and insulator/semiconductor layer (active layer), it is aimed to have memristive properties such as low set/reset voltages, fast switching speed, high endurance characteristics (LTP/LTD), long retention, paired-pulse facilitation (PPF), and less power consumption. Now, among the different memristors, oxide-based memristors are best fitted in the framework of NC applications. Metallic oxides, namely, ZnO, TiO₂, HfO₂, NiO, TaO_x, SnO_x, WO₃, and SrTiO₃, are commonly taken for the manufacturing of memristors.^3–9 Also, widely used chalcogenide substances like Ag₂S, Cu₂S,^10,11 and ferroelectric materials like Pb₀.₈Ba_0.2ZrO₃, La₂O₃,^12,13 have been investigated in fabricating memristors. The switching operation and synaptic behavior in WO₃/NiO/FTO composite structures are successfully illustrated.¹⁴ ZnO-based memristor is fabricated, which can play the role of artificial synapse and along with synaptic functions such as short- and long-term plasticity.¹⁵ Photoelectric plasticity such as electrical resistive switching and photonic response in the ZnO_1−x/AlO_y hetero-structure has been demonstrated by Hu et al.¹⁶ In a study, Lee et al. grew IZO (indium gallium zinc oxide) on SiO₂ and Si substrates and demonstrated how the apparatus replicated a key synaptic performance.¹⁷ He et al. accomplished electrical habituation and photonic potentiation using a single-layer MOS₂.¹⁸ 

Also, transparent memristors or transristors have wide applications in non-volatile memory, optoelectronic devices, neuromorphic systems, artificial synapses, neural networks, and human visual systems, making them popular at present. Park et al.¹⁹ have developed an organic memristor that is flexible and transparent showing a high optical transparency and can be used as non-volatile memory. Saleem et al.²⁰ have presented a transparent bilayer memristor using indium-doped tin oxide (ITO)/ZnO/HfO_x/ITO, which can be used in an optically synaptic device. Yan et al.²¹ fabricated a flexible egg albumen-based memristor that can act as an artificial synapse. Such bio-based memristors are suffering with their degradable stability with time. Zinc oxide (ZnO) is a multifunctional material with a wide range of applications due to its unique properties, including a wide bandgap (3.37 eV), high exciton binding energy (60 meV), and diverse morphologies. In electronics and optoelectronics,^22,23 ZnO is employed in light-emitting diodes (LEDs), laser diodes, and photodetectors, leveraging its semiconducting properties and efficient ultraviolet (UV) light emission. Its high transparency and conductivity make it an excellent candidate for transparent conducting oxides (TCOs) in photovoltaic cells and display technologies. The diverse applications of ZnO across various fields underscore its significance and the incessant interest in exploring and optimizing its properties for advanced technological applications.

ZnO is generally chosen as the active layer in the memristor because of its high electron mobility and high binding energy, thereby resulting in good transparency. As a result, it is a popular material for transparent devices and widely employed for memristive applications and synapses. So, in our current research, we have created and examined a flexible transristor, taking ZnO as the intermediate layer between the electrodes of indium-doped tin oxide (ITO) and aluminum-doped zinc oxide (AZO). We have analyzed the switching mechanism and other synaptic characteristics of the proposed fabricated device. There are currently limited reports on the study of synaptic characteristics for this transparent and flexible device at varying concentrations of oxygen, which enhances the scope of this work.

Figures 1(a) and 1(b) depict the three-dimensional representation view of the flexible transristor device based on AZO/ZnO/ITO/PEN and its biological synapse counterpart, respectively. The devices are fabricated on the cleaned flexible PEN substrate using conventional RF sputtering techniques. Before being loaded into the chamber, the substrate covered with indium-tin-oxide (ITO) that is about 300 nm thick is first sonicated and cleaned for about 10 min using ethanol and de-ionized (DI) water. It is then dried by purging nitrogen gas. ITO is utilized as a transparent bottom electrode (BE) because of its high transmittance and low resistance. Zinc oxide (ZnO) layers are applied onto ITO/PEN substrates using magnetron sputtering, with adjustments made to the Ar/O₂ gas ratios to control the deposition. At this stage, the Ar/O₂ ratios being used have flow rates of 3/0, 2/1, and 1/1. During all depositions, the working pressure of 20 mTorr and an operating power of 50 W are sternly maintained. Using the shadow mask, a 60 nm thick layer of AZO as a top electrode is also sputtered after the deposition of a 50 nm ZnO layer on the ITO bottom electrode. The AZO film acts as a top electrode having an area of 17.67 μcm².

X-ray photoelectron spectra (XPS) are used to investigate the device composition. The electrical properties of the transristor have been analyzed using an Agilent B1500A semiconductor parameter analyzer at room temperature. During electrical measurements of the device, the direction of the current flow between the top and bottom electrodes is indicative of whether a positive or negative bias is applied. The fabricated device's optical transparency is measured with a UV-vis spectrometer.

Depth profile analysis has been conducted using XPS on the optimized device to examine the composition at different depths. Figure 2(a) illustrates the atomic concentration of the elements O 1s, Zn 2p3, Al 2p, and In 3d5 over sputtering time. From the XPS depth profile, it is observed that during the 157–202 s sputtering period, the concentration of zinc is 55% and the rest is oxygen. It is shown that while the concentrations of indium (In) and aluminum (Al) roughly stay constant, the concentration of zinc (Zn) gradually drops as the sputtering time increases. Zn can, therefore, be discovered from the film's surface. Since AZO is our top electrode, a slight reduction in the atomic concentration of Al with respect to sputtering time is anticipated.²⁴ According to our artificial structure, the presence of various elements at various depths is confirmed by other elements. Figure 2(b) displays the deconvolution of Zn-2p_3/2 of the ZnO film formed at a gas flow ratio of Ar/O₂= 2/1. The presence of an energy peak of Zn 2p3/2 at 1022.6 eV suggests that the majority of zinc atoms within the oxygen-deficient region retain the Zn²⁺ valence state. A Gaussian fit technique is used for the simulation of the peak energy and also to find out the different energy levels of the material. The deconvoluted XPS spectra of O 1s at depth 1 (ZnO: D1), which have been fitted with a Gaussian fit, are shown in Fig. 2(c). The O 1s spectra are well fitted with the three energy levels of O 1s, indicating the O 1s is the combination of three energy peaks. These three sub-peaks correspond to O_I (530.3 eV), O_II (531.05 eV), and O_III (532.25 eV). O_I stands for oxygen ions that are closest to ZnO links, O_II for oxygen ions that are in a deficient area, and O_III for adsorbed oxygen ions that are weakly bound. The proportionate area of O_II/O_total[O_total = O_I + O_II + O_III] ratio is used to estimate the oxygen vacancy-mediated defects, and it is determined to be 77.0%. The process of synaptic conduction is aided by these oxygen vacancy-mediated defects. The sandwiched layer ZnO, at depth 2 (ZnO: D2), is shown in Fig. 2(d). The O_II/O_total at this particular depth is 56.39%. Figure 2(e) represents the O 1s spectra of AZO, which was Gaussian deconvoluted into three peaks. The two peaks that appear at O_I (530.15 eV) and O_II (531.25 eV) are related to the oxygen bounded as (Al–O and Zn–O) and oxygen vacancies, respectively. The third peak at O_III (532.4 eV) is the loosely bound chemisorbed oxygen. Consequently, the ratio O_II/O_total for the AZO top layer is found to be 74.82%. For the bottom layer ITO, O 1s spectra analysis has been done as shown in Fig. 2(f), and deconvoluted peaks are found to be at O_I (530.02 eV), O_II (531.38 eV), and O_III (532.78 eV). The O_II/O_total in this layer results as 27.87%.

The characterization of the constructed devices using their current–voltage (I–V) characteristics is depicted in Fig. 3(a), with +5 and −2 V biases applied for set and reset operations, respectively. In the transristor's operation, the redox process leads to the generation of an electric field. This electric field, in turn, plays a vital role in modulating the transristor's conductivity and thereby controlling its switching behavior. The presence of oxygen vacancies created is confirmed through XPS analysis of the oxide layer. These vacancies then migrate toward the switching zinc oxide upon applying a positive bias to the AZO top electrode (TE) and a negative bias to the indium-doped tin oxide (ITO) bottom electrode (BE). As the concentration of oxygen vacancies in the ZnO layer increases, a conducting filament (CF) develops between the top and bottom electrodes creating a low resistance state (LRS). This phenomenon leads to the transition of the devices from a high-resistance state (HRS), representing the off-state, to an LRS, indicating the on-state. Additionally, the device is switched from an on position, i.e., from LRS to HRS by the re-ionized oxygen ions filling the voids that cause the filament to rupture. First, the switching procedure needs to be started by applying higher positive voltage by forming an initial filament from the pristine state. About +5 V is required as a forming voltage for all the devices. When a positive voltage is applied to the TE, the current flowing through the device gradually rises until it reaches the 1 mA compliance current (CC). Concurrently, the stated device undergoes a transition in resistance level, shifting from an HRS to an LRS. This process is commonly referred to as the “set procedure,” characterized by the formation of a CF. During the reset process, initiated by applying −2 V, the current flowing through the device gradually decreases as the CF is disrupted. Consequently, the device changes its state from set to reset (i.e., LRS to HRS).^25,26 When a positive voltage ranging from 0 to 5 V is applied to the device, a moderate increase in current is observed, particularly for the Ar/O₂ ratio of 2/1, as depicted in Fig. 3(b). However, the current spikes up to the specified CC at 5 V. The device deposited with a gas ratio of 3/0 exhibits an abrupt increase in current upon applying a voltage between 0 and +5 V, as demonstrated in Fig. 3(c). This abrupt current leads to poor synaptic properties. According to the I–V characteristics, bipolar resistive switching can be achieved through conductivity modulation with the application of either a positive or negative voltage for all the devices. The switching curves confirm that current levels vary proportionally to voltage and polarity due to charge trapping and de-trapping processes. This state demonstrates the device's synaptic behavior of LTP and LTD.

Figures 4(a)–4(c) show the device's endurance characteristics at +0.1 V read voltage. The device with a 2/1 gas ratio maintains stability and a comparable ratio throughout set/reset operation over 500 cycles, with no degradation between LRS and HRS, as shown in Fig. 4(b). In contrast, the other two devices experience poor ratio compared to the device fabricated at the gas ratio of 2/1 and suffer with resistance state stability after only 100 cycles when using a sample that is not deposited without oxygen atmosphere, i.e., the gas ratio of 3/0, as illustrated in Fig. 4(c). The I–V characteristics for all cycles of the devices of (a) Ar/O₂= 1/1, (b) Ar/O₂= 2/1, and (c) Ar/O₂= 3/0 are shown in Fig. S1 in the supplementary material. The stability of the states from low resistance and high resistance over time is utilized to determine the retention characteristics of the transristor. Retention is measured by applying the read voltage and recording the corresponding current response from the device for both resistive states. When the Ar/O₂ gas ratio of 1/1 is used in the deposition process, as shown in Fig. 4(a), a significant difference is observed between the two resistance states in the resistive switching devices. On the other hand, Ar/O₂ having 2/1 flow ratio exhibits an excellent retention of 10⁴ s for non-volatile memory and neuromorphic computing, as shown in Fig. 4(d).

Various pulse signals are applied at varying Ar/O₂ ratios to examine the synaptic behavior of the transristor. The reaction of the transristor to applying positive and negative pulses is depicted in Fig. 5 in terms of potentiation and depression. For the synaptic measurement of Ar/O₂= 1/1, as shown in Fig. 5(a), an optimized pulse train is applied with a 10 μs width and +0.92 V amplitude for potentiation/−0.96 V amplitude for depression. The device with Ar/O₂= 2/1 is subjected to an optimized pulse train with a 10 μs pulse width and +0.72 V amplitude for potentiation/−0.81 V amplitude for depression, as depicted in Fig. 5(b). For Ar/O₂= 3/0, a pulse train with 10 μs width and +0.8 V amplitude for potentiation/−0.72 V amplitude for depression is energized, as illustrated in Fig. 5(c). To achieve both depression and potentiation during the read condition, a consistent 0.1 V amplitude and 1 ms pulse width is applied to each device. Ar/O₂= 2/1 improves the potentiation and depression linearity values, competing with the sample for a better fit for neuromorphic application than the other prepared devices. Moreover, by energizing a sequence of pulses, the device replicates potentiation and depression cycles and maintains a stable conductance ratio over multiple cycles (>25 cycles).

The conductance values for the LTD and LTP are shown by G_LTD and G_LTP, respectively. Maximum conductance (G_max), minimum conductance (G_min), maximum pulse number (P_max), controllable variable (A) that provides information on nonlinear behavior, and dependent parameter (B) on variable (A). Equations (1)–(3) are utilized to determine the non-linearity of the transristors in terms of α_p/α_d (potentiation/depression). The results show that for Ar/O₂ with 1/1, 2/1, and 3/0 gas flow ratios, α_p/α_d is 4.11/4.74, 2.31/3.05, and 3.20/4.50, respectively. According to the aforementioned findings, the transristor with an Ar/O₂= 2/1 gas flow ratio has improved non-linearity for both potentiation and depression, making the device appropriate for neuromorphic computing applications. Such improvements in non-linearity in the device prepared at Ar/O₂= 2/1 gas flow ratio may be due to the controlled growth and rupture of filaments formed by oxygen vacancies. The non-linearity value of the optimized device is calculated for three cycles—first, middle, and the last cycles, as shown in Figs. 5(d)–5(f), which is found to be consistent as 2.42/3.07, 2.51/3.15, and 2.42/3.07, respectively.

Figure 6 illustrates the conductance values of the fabricated transristors for different ratios of Ar/O₂. Notably, in Fig. 6(a), the sample with Ar/O₂= 1/1 demonstrates minimal alteration in conductance values for up to six cycles when subjected to consecutive potentiation and depression cycles. On the contrary, while the cycle number increases for Ar/O₂= 3/0, as shown in Fig. 6(b), the conductance value falls. For depression and potentiation, superior 200 pules are administered to each Ar/O₂= 2/1 pair, as depicted in Fig. 6(c). It is evident that for more than 25 cycles, the conductance states remain stable, indicating suitability for synaptic applications. The electrical characteristics of the manufactured gadget are compiled in Table I.

In this work, ZnO due to its unique properties is utilized as a switching film with different oxygen concentrations to fabricate the two-terminal transparent memristor-based synapses. The ZnO layer or the film deposited at Ar/O₂= 2/1 significantly enhances the switching mechanism in the device, as confirmed by the XPS result of Zn 2p_3/2 and O 1s. The optimized device exhibits significant resistance switching behavior, with the lowest V_set/V_reset voltage of 1.4/−2 V. An acceptable endurance exceeding switching cycles of more than 500 is observed, indicating the device's reliability over repeated switching events. The device demonstrates good retention characteristics, with a retention time of 10⁴ s, ensuring stable performance over time. Enhanced linearity of 2.31/3.05 is achieved for potentiation and depression, respectively, indicating precise control over the resistance switching process. The device maintains stability for 25 cycles without noticeable degradation, ensuring consistent performance over multiple operational cycles. With over 90% transparency, the optimized transristor can be integrated with invisible electronic applications, expanding its potential use cases. Therefore, applications involving invisible synapses are particularly suitable for these devices.

See the supplementary material for the IV characteristics of all cycles obtained for the device with Ar/O2 = 1/1, Ar/O2 = 2/1, and Ar/O2 = 3/0 as shown in Fig. S1.

This work was supported by the Department of Science and Technology (DST), Science and Engineering Research Board (SERB), Government of India, under core Project Grant No. CRG/2023/001265.

The authors have no conflicts to disclose.

Asutosh Patnaik: Data curation (equal); Formal analysis (equal); Methodology (equal); Software (equal); Writing – original draft (equal). Arpan Acharya: Formal analysis (equal); Visualization (equal). Kabin Tiwari: Investigation (equal); Visualization (equal). Priyanka Saha: Formal analysis (equal); Validation (equal). Narayan Sahoo: Supervision (equal); Validation (equal); Writing – review & editing (equal). Debashis Panda: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal).

The datasets generated and/or analyzed during this study are not publicly available due to confidentiality but are available from the corresponding author upon reasonable request.