Zinc oxide and indium-gallium-zinc-oxide bi-layer synaptic device with highly linear long-term potentiation and depression characteristics

The electrical properties, resistive switching behavior, and long-term potentiation/depression (LTP/LTD) in a single indium-gallium-zinc-oxide (IGZO) and bi-layer IGZO/ZnO (ZnO: zinc oxide) memristors were investigated for synapse application. The use of the oxide bi-layer memristors, in particular, improved electrical properties such as stability, memristor reliability, and an increase in synaptic weight states. The set voltage of bi-layer IGZO/ZnO memristors was 0.9 V, and the reset voltage was around − 0.7 V, resulting in a low-operating voltage for neuromorphic systems. The oxygen vacancies in the X-ray photoelectron spectroscopy analysis played a role in the modulation of the high-resistance state (HRS) (oxygen-deficient) and the low-resistance state (oxygen-rich) region. The VRESET of the bi-layer IGZO/ZnO memristors was lower than that of a single IGZO, which implied that oxygen-vacancy filaments could be easily ruptured due to the higher oxygen vacancy peak HRS layer. The nonlinearity of the LTP and LTD characteristics in a bi-layer IGZO/ZnO memristor was 6.77% and 11.49%, respectively, compared to those of 20.03% and 51.1% in a single IGZO memristor, respectively. Therefore, the extra ZnO layer in the bi-layer memristor with IGZO was potentially significant and essential to achieve a small set voltage and a reset voltage, and the switching behavior to form the conductive path.


Scientific Reports
| (2022) 12:1259 | https://doi.org/10.1038/s41598-022-05150-w www.nature.com/scientificreports/ concentration, allowing for high-speed operation. Following a high driving current, the increased concentration of indium in the IGZO layer increased mobility, while the gallium concentration controlled the amount of oxygen vacancy 24,25 . ZnO is an excellent material of chemical stability, and has a wide direct bandgap and high transparency properties. However, because donor defects in ZnO materials contribute to n-type conduction, high resistive ZnO switching films are difficult to achieve. The low-resistance state (LRS) layer is positioned on top of the bi-layer structure, resulting in good switching behavior. However, memristor devices for hardwarebased neural networks should consider the linear and symmetric changes in conductance with the number of potentiation and depression pulses. Furthermore, the SET pulse applied as the pre-synaptic input can determine the consumed energy per the weight update for the network training. The relatively high SET amplitudes and long pulse width must be further improved for application in energy-efficient and largescale neuromorphic device arrays. The potentiation and depression characteristics of single-layer memristor devices are frequently nonlinear, resulting in less efficient neural network processing. It is believed that the bi-layer memristor device's structure achieves high linearity of the potentiation and depression characteristics. The electrical properties of the ReRAM with an additional oxide layer between the bottom electrode and the IGZO layers, on the other hand, are yet to be investigated. The use of oxide bi-layers improved electrical properties such as stability, memristor reliability, and an increase in synaptic weight states [26][27][28] . This paper proposed bi-layer IGZO/ZnO memristors to improve the electrical characteristics and synaptic linearity in long-term potentiation/depression (LTP/LTD) characteristics compared with a single IGZO memristor. The conduction mechanism of the charge transport behavior at HRS and LRS was validated by the role of the potential barrier between the bottom electrode and HRS material. To our knowledge, the high linearity in LTP/ LTD characteristics of the bi-layer IGZO/ZnO memristors is more linear than that of other reported devices [29][30][31][32] .

Methods
A single structure of memristor crossbar was fabricated to a square type, as shown in Fig. 1, which could be integrated at high density. The Ti layer, which served as the bottom electrode, was deposited at 100 nm using a radio frequency (RF) sputtering system and patterned using photolithography's lift-off process. Memristors are classified into two types: single IGZO memristors and bi-layer IGZO/ZnO memristors. The single-layer IGZO was deposited using the following RF sputtering system. First, 50 nm-thick IGZO used as an HRS was deposited with the molecular composition of In:Ga:Zn = 1:1:1 at a working pressure of 2 mTorr (Ar gas flow rate: 30 standard cubic centimeters per min (sccm)) without substrate heating (substrate temperature < 50 °C). The 10 nm-thick IGZO for an LRS was then deposited with the molecular composition In:Ga:Zn = 2:2:7, and the other conditions were the same as described above. To control the amount of oxygen vacancy, the oxygen gas flow was adjusted as 20 sccm for HRS and 1 sccm for LRS when the IGZO layers were deposited. For the bi-layer IGZO/ZnO structure, a 50 nm ZnO layer was deposited using an atomic layer deposition system with DEZn (diethylzinc, Zn(CH 2 CH 3 ) 2 ) and an H 2 O of source precursors at an 80 °C chamber temperature, followed by a 10 nm IGZO LRS deposition using an RF sputtering system. Finally, the Ti layer was deposited as the top electrode using the RF sputtering system, which is identical to the bottom electrode process through photolithographic patterning to form a crossbar type device. Figure 1b showed the top-view images of the field emission scanning electron microscopy (FE-SEM, Hitachi, Japan, S-4800) in a 20 μm × 20 μm crossbar structure.
The electrical properties of a single IGZO and bi-layer IGZO/ZnO memristors were measured at room temperature using the Keysight B1500A semiconductor device analyzer. For the transition from HRS to LRS, the current compliance was used. The Keysight B1530 waveform generator/fast measurement unit, which also has current and voltage measurement functions, was used to apply voltage pulses across the single IGZO and bi-layer IGZO/ZnO memristors. Precision and long-term sampling measurements were used to measure the current across the diffusive memristor and to measure the voltage at the output of the waveform generator. The long-term sampling measurements allowed the application of pulse signals down to 100 ns and precise current measurement at a sampling rate of 50 ns. To minimize the disturbance on device conductance, a write pulse with amplitude 2 V was applied across a single IGZO and bi-layer IGZO/ZnO memristor, and then a short duration (20 or 50 ms) was used as the read voltage. Positive/negative pulses were applied at a rate of 50 for single IGZOs and 100 for bi-layer IGZO/ZnO memristors, and the current was measured after each stimulation pulse. All equipment in the setup is controlled by the Agilent Easy Expert software. Figure 2 shows the RS behaviors of 10 cycles in a single IGZO and bi-layer IGZO/ZnO memristors. This result shows the typical bipolar RS characteristics with the bottom electrode grounded in electrical measurements. The memristor devices are initially in HRS mode. A voltage sweep from 0 to 3 V (meaning arrow "1" in the figure) is applied to the top electrode with a compliance current of 1 mA, and the RS behaviors of the single IGZO and bi-layer IGZO/ZnO memristors change from HRS to LRS. The compliance current is set to prevent memristor devices from being permanently damaged by a sudden increase in current levels. Once again, the voltage sweep is applied from 2 to 0 V (meaning arrow "2" in the figure) to measure the current and the RS behaviors of the single IGZO and bi-layer IGZO/ZnO memristors that remain in the LRS and show a high current level. The arrows "1" and "2" in the figure are referred to as "SET, " and they indicate that the RS behaviors of the single IGZO and bi-layer IGZO/ZnO memristors change from HRS to LRS. The process of changing from arrow "3" to "4" in the figure is called "RESET, " and it means that the RS behaviors of the single IGZO and bi-layer IGZO/ ZnO memristors change from LRS (arrow "3") to HRS (arrow "4"). Because conductance values (the inverse of resistance values) are used as synaptic weights in binarized neural networks, these findings can help them. The set voltages of the single IGZO and bi-layer IGZO/ZnO memristors are 1 V and 0.9 V, respectively. In contrast, the reset voltages of a single IGZO and bi-layer IGZO/ZnO memristors are about − 1.8 V and − 0.7 V, respectively.

Scientific Reports
| (2022) 12:1259 | https://doi.org/10.1038/s41598-022-05150-w www.nature.com/scientificreports/ Therefore, in the case of IGZO/ZnO structure, the memristor device can be used as a low operation, which can improve the RS speed because the donor defects in ZnO materials contribute to the n-type conduction. Figure 3 shows the atomic force microscopy results of a single IGZO (1:1:1), IGZO (2:2:7), and ZnO films, and bi-layer memristors for a measurement area of 1 × 1 μm. The root-mean-square (RMS) roughness value of a single IGZO (1:1:1), IGZO (2:2:7), and ZnO films is 0.112 nm, 0.145 nm, and 2.247 nm, respectively. The RMS roughness of a single IGZO memristor and a bi-layer IGZO/ZnO memristor is 0.121 nm and 2.176 nm, respectively. These findings imply that small grains with a nanometric dimension of 33 can be used to observe a typical polycrystalline ZnO film. Figure 4 depicts the results of the Hall measurements on carrier concentration, resistivity, and Hall mobility of a single IGZO (1:1:1), IGZO (2:2:7), and ZnO film. The electrical properties of the carrier concentration, resistivity, and Hall mobility for a single IGZO (1:1:1) and IGZO (2:2:7) films are greater than for a single ZnO film 33,34 . In particular, the resistivity of IGZO (2:2:7) decreases with the increase of gallium (Ga) content because Ga 3+ forms strong bonds with oxygen 33,34 . The resistivity of each layer for a single IGZO and bi-layer IGZO/ZnO memristors is proportional to the reciprocal of the product of carrier concentration N and mobility μ as reported in Ref. 35 . It can confirm the conjecture of controlling the carrier concentration via oxygen-related defects associated with Ga. X-ray photoelectron spectroscopy (XPS, Thermo Fisher Scientific, USA, K-Alpha + ) measurements were performed to investigate the chemical composition of the single IGZO and bi-layer IGZO/ZnO memristors, to verify the proportions of the oxygen vacancy. Figure 5 depicts the XPS analysis result of the oxygen vacancy peak (O1) spectra in the surface after deposition of each single HRS and LRS layer for the single IGZO and bi-layer IGZO/ZnO memristors using the Gaussian peak fitting. The proportions of the O1 of the HRS layer for a single IGZO and bi-layer IGZO/ZnO memristors are 45.2 percent and 38.2 percent, respectively, while the proportion of the LRS layer for both memristors is around 43.4 percent. Because the Ga/Zn ratio determines the oxygen concentration in HRS and LRS layers during IGZO sputter-deposition, increasing the Ga/Zn ratio increases the number of non-oxygen vacancies in the memristor devices, resulting in a lower conductivity layer 36  www.nature.com/scientificreports/ role in distinguishing each HRS and LRS layer for the RS performance in the bi-layer memristor structure. The oxygen vacancies modulate the HRS layer (oxygen-rich) into the LRS layer (oxygen-deficient). The V RESET of the bi-layer IGZO/ZnO memristors is lower than that of a single IGZO, which implied that the oxygen vacancy filaments could be easily ruptured due to the lower oxygen vacancy peak HRS layer.
To verify the mechanism of the RS behaviors for a single IGZO and bi-layer IGZO/ZnO memristors, the corresponding I-V characteristics of the SET and RESETs are plotted in Fig. 6. A linear fitting slope based on experimental data for the single IGZO and bi-layer IGZO/ZnO memristors is close to one, indicating a linear relationship between the current and applied voltage 37 . The charges originating at the metal electrode interface are thought to be trapped by the HRS layer's empty trap sites of IGZO and ZnO. As the electric field across memristor devices increases, the steep current for a single IGZO is followed by a quadratic term (I ∝ V 2 ) with the increase of the injected charges when the conductive filaments form between two electrodes, as shown Fig. 6. When the empty trap sites are gradually occupied completely, the slope of the fitting line decreases by about 2, indicating that the conduction enters the trap-free space charge limited current (SCLC). It implies that the SCLC is dominant because the majority of injected electrons contribute to the current component [38][39][40][41][42][43][44] .
However, the slope of the fitting line at the high electric field for bi-layer IGZO/ZnO memristor is found to be 4.0, which means that the Schottky emission mechanism is dominant. In a high electric field, the Schottky emission mechanism may be caused by oxygen vacancies near the metal/metal-oxide interface. The Schottky emission and ohmic mechanisms dominate the I-V characteristics of the RESET, as illustrated in Fig. 6. The switching behaviors of a single IGZO and bi-layer IGZO/ZnO memristors were controlled by the interface properties due to the Schottky emission mechanism. It should be noted that the SCLC is the primary conduction mechanism in the SET for single-layer IGZO and bi-layer IGZO/ZnO memristors. We conclude the Schottky emission mechanism observed in a single IGZO and bi-layer IGZO/ZnO memristors at the RESET is attributed to the interface barrier. Figure 7 shows the long-term potentiation/depression (LTP/LTD) characteristics with applied positive/negative pulses for an amplitude of 2 V in the two memristors. The positive (2 V, 400 ns) or negative (− 2 V, 400 ns) voltage pulses with the interval time (4.5 μs) are applied on the memristor devices, and then, the current is measured by a reading voltage pulse (0.2 V, 1 μs) after each pulse. Depending on the input spiking signal, the LTP and LTD characteristics exhibit gradual potentiation and depression in synaptic weight, which can be used to determine whether memristors can learn or not. When a potentiating input signal train of positive pulses with an amplitude of 2 V is applied to the top Ti metal of the memristor synapse (pre-neuron), the synaptic weight changes progressively as the current increases, indicating that oxygen vacancies are injected into the RS layer, and this process is then formed between TE and BE for potentiation. It can emulate the potential of oxygen www.nature.com/scientificreports/ vacancies for neuromorphic computing, which enhances the synapse weight by releasing neurotransmitters. When a depressing input signal train consisting of negative pulses with an amplitude of − 2 V is applied to the top metal, the synaptic weight is gradually depressed, and the conductive path formed by the oxygen vacancies moves away from the bottom metal, causing the current to decrease. The nonlinearity for the memristor devices is quantitatively given by Eq. (1) where G is the change in the conductance of memristive devices (equivalently, synaptic weight), and G Linear is the linear change in conductance (determining training accuracy 45 ). The nonlinearity of LTP and LTD characteristics in the bi-layer IGZO/ZnO memristor is 6.77% and 11.49%, respectively, while these are for in a single IGZO memristor is 20.03% and 51.1%, respectively 46 . When the   www.nature.com/scientificreports/ number of synaptic weight states and total plasticity can be increased, oxygen vacancies closed to the metal/metaloxide interface at the bottom electrode in the bi-layer IGZO/ZnO memristor are largely generated, appreciably increasing the effective Schottky barrier height as discussed in Fig. 7. As shown in Fig. 7, donor defects in ZnO materials that contributed to the n-type conduction are expected to have a more linear conductance response during LTD (b). The transition mechanism from conductive filamentary switching for a single IGZO and bilayer IGZO/ZnO memristors is shown in Fig. 8, implying that the mobility difference between the two materials  www.nature.com/scientificreports/ in the bi-layer structure, as discussed in Fig. 4, is critical. Therefore, the high electron conductivity of the ZnO layer in the bi-layer IGZO/ZnO memristor plays an important role in charge carriers to be injected easily under a small set voltage and a reset voltage switching behavior to form the conductive path. Not only the nonlinearity but also the asymmetry ratio is a critical factor in determining the learning accuracy 47 . The symmetric weight update between LTP and LTD in the memristor devices is defined by, where G is the same as Eq. (1), N max is total pulse number between G min and G max , and N x is a certain pulse number, indicating asymmetry ratio is "1" in ideal case 48 . Figure 9 shows the cumulative probability results of asymmetry ratios for the single IGZO and bi-layer IGZO/ ZnO memristors. The average asymmetry ratio for a single IGZO memristor and a bi-layer IGZO/ZnO memristor is 0.877 and 1.20, respectively. The standard deviations normalized to the mean ratio for two memristors, on the other hand, are 0.213 and 0.031, respectively. As a result, the comprehensive nonlinearity and asymmetry results are shown in Table 1, indicating that a bi-layer IGZO/ZnO memristor made by inserting a ZnO layer for the HRS layer is more suitable for ideal synaptic devices. It has the potential to be a desirable option to be used in implementing neural networks in the future.

Conclusion
We investigated the electrical characteristics, RS behavior, and the LTP/LTD of ZnO and IGZO and bi-layer memristors for high-performance synaptic devices. The set and reset voltages of the bi-layer IGZO/ZnO memristors are 1 V, and − 0.7 V, achieving low operating voltage to realize the neuromorphic systems. The oxygen vacancies played a role in the modulation of the HRS layer (oxygen-deficient) and the LRS layer (oxygen-rich) region. The V RESET of the bi-layer IGZO/ZnO memristors was lower than that of a single IGZO, which implied that the oxygen vacancy filaments could be easily ruptured due to the higher oxygen vacancy peak in the HRS layer. When compared to a single IGZO memristor, the nonlinearity and asymmetry ratios of the LTP and LTD characteristics for the bi-layer IGZO/ZnO memristor improved significantly. These findings were significant because they revealed the difference in mobility between two materials in a bi-layered structure. As a result, the ZnO layer's role in the bi-layer IGZO/ZnO memristor was potentially significant and necessary for charge carriers to be injected easily under a small set voltage and a reset voltage, as well as the switching behavior to form the conductive path. . Asymmetry ratio distribution of a single IGZO and bi-layer IGZO/ZnO memristors. The standard deviation of bi-layer IGZO/ZnO memristors shows much more improvement compared to that of a single IGZO memristor, meaning close to ideal synapse device.