In recent years, memory devices are indispensable components to meet the growing demand for smart products, such as cellular phones and tablets. Floating gate memory has been well developed and commonly applied for nonvolatile data storage. However, with the scaling down technology of integrated circuits, the memory device cannot effectively retain electrons and storage information [1]. Recently, memristor has been considered as one promising candidate for current memory device, because of the simple structure, high storage density, fast switching speed [[2], [3], [4]]. Besides, indium-gallium-zinc-oxide (IGZO) becomes one of the most popular oxide materials, which has the potential to be the competitor for thin film transistors (TFTs) in future application of flat-panel displays [[5], [6], [7]]. Despite of the application in display industry, IGZO used as a resistive switching (RS) layer in memristive device has also been investigated [[8], [9], [10]]. A Cu/IGZO/TiN resistive memory device was proposed with excellent pulse operation speed [11]. Doping an appropriate amount of Ru into the IGZO RS layer exhibited enhanced bipolar RS properties and a sufficient OFF/ON ratio of 10⁵ [12]. On the basis of the advantages of IGZO material, if it can be adopted in the memristive device, the integration of IGZO thin film transistors and IGZO memory devices will become compatible and easily accomplish system-on-panel applications. However, for the utilization of IGZO film in memristive device, it can be seen the RS performance of IGZO memristor is not satisfied. If we further integrate memory in next-generation commercial flat-panel display industry, the improvement of performance for IGZO memristor is urgently necessary.

Many methods have been devoted to boost up the performance of memristive devices. The bilayer structure design is proposed to acquire higher performance than the single layer device [13,14]. Besides, it has been investigated that the nitrogen doping could help to improve the uniformity of the RS parameters [15]. As concerned as N doping methods, post-annealing method in N₂ atmosphere, treating with N₂ sputtering atmosphere and supercritical fluid nitridation technique are effective methods [[16], [17], [18]]. Plasma treatment has been proven to be an effective and convenient way to regulate the material property related to oxygen vacancies in metal oxide films [19]. It has been demonstrated by many experiments that for the RS mechanism of binary metal oxide RRAM, the conductive filaments (CFs) formation/rupture is actually a redox reaction [[20], [21], [22]].

In this work, an ultrathin IGZO film with N₂ plasma treatment was inserted to fabricate Pt/IGZO/IGZO:N/TiN device. The material characterization and RS behavior were systemically analyzed. The role of the inserting IGZO:N layer on the RS switching behavior was evaluated and the RS mechanism triggered by inserting film was explored. This work opens the prospect of improving RS performance of IGZO memristive device and provides the potential application on the integration with TFTs to meet the demand of flat panel display development.

The IGZO memristive devices were prepared on TiN/SiO₂/Si commercial substrates, and the 180 nm TiN film was used as the bottom electrode. First, a IGZO target was used to deposit amorphous IGZO film on the substrate. In the sputtering process, the radio frequency power was 50 W, the mixing ratio of the gas in the sputtering chamber was set at Ar:O₂ = 12:12, and the total pressure was kept at 0.65 Pa at room temperature. After a 4 nm IGZO film was deposited, the sample was fixed in the chamber with a closed shutter. Then, the IGZO film can be treated in nitrogen plasma with gas mixing ratio of N₂:Ar = 50:12 and IGZO:N film can be obtained with the plasma treatment time was 30 s. Afterwards, the remaining 18 nm IGZO layer was sputtered. Finally, 200 nm Pt top electrode was sputtered by a direct current sputtering process at a power of 80 W. For the purpose of shaping and completing an individual device, photolithography (ABM inc MA-TSV) and lift-off methods were applied. At the same time, the Pt/IGZO/TiN control device was also fabricated. The thickness of IGZO film was measured by Ellipsometer (Alpha-SE). The RS properties of memristive devices were measured by a semiconductor parameter analyzer (Agilent B1500A) at room temperature. For the electrical test, a voltage bias was applied to the TiN electrode while the Pt electrode was grounded.

Atomic force microscope (AFM) surface scans are carried out on the IGZO and IGZO:N films for analyzing the impact of nitrogen plasma treatment on the surface topography. The images by scanning a 1 × 1 μm² area of the films are presented in Fig. 1. The results reveal that the root mean-square roughness decrease from 0.974 nm of IGZO film to 0.157 nm of IGZO:N film. It indicates that the surface of the IGZO film becomes smooth and dense through nitrogen plasma treatment. Besides, the IGZO:N film seems to consist smaller grain size, which is evenly presented in the film. Therefore, the inserting IGZO:N film presents high quality. It is commonly acknowledged that the film quality may affect the RS performance of memristive device [23].

To examine the elemental composition and chemical bonding states of IGZO films with and without N impurity, XPS measurement is carried out. Fig. 2(a)-(c) depicts the XPS spectra of the In 3d, Ga 3d, Zn 2p peaks for the updoped and the N-doped IGZO films. Compared to the undoped IGZO film, the three metal element peak of the N-doped IGZO film has a slightly blue shift of about 0.2 eV, implying N incorporated in the IGZO film successfully [24]. From Fig. 2(d), an N 1s peak occurred at the binding energy of 396.8 eV. The mole concentration of each element in IGZO:N film is calculated, and the atomic ratio of In:Ga:Zn:O:N is 8.4 %:28.9 %:6.1 %:39.8 %:16.8 %. Fig. 2(e) illustrates the O 1s core level spectra for two IGZO samples. It can be distinctly observed that the O 1s spectra can be divided in two peaks. The main peaks of O 1s occurs at 529.8 and 530.1 eV are originated to the lattice oxygen, which normally refers to the O^2- ions in crystal lattice. For non-lattice oxygen, the binding energy locates at a higher value of 531.1 and 531.2 eV, associating with oxygen-deficient regions [25]. Furthermore, for pure and N-doped IGZO films, the contents of non-lattice oxygen are calculated to be 40.56 % and 50.45 %, respectively. Therefore, the XPS results infer more oxygen vacancies, which may be introduced in the IGZO film by the N dopants [19]. For a memristive device, the content of oxygen vacancies in the RS film influences the electrical characteristics [8,26,27].

To verify the effect of the inserting IGZO:N thin film, I-V curves are measured on both devices. Fig. 3(a) depicts the typical forming process of both devices under 1 mA compliance current. It can be observed that the forming voltage value of IGZO memristor is 3.0 V, while the value of IGZO:N memristor is 1.6 V. Furthermore, we randomly select some samples for each kind of device to investigate the device-to-device variation of forming voltage. It can be concluded that the forming voltage decreases after inserting IGZO:N thin layer. On the basis of O 1s analysis from XPS, it is supposed that N dopant stimulates the oxygen vacancy formation and leads to lower formation energy, which affects the formation of the CFs, so Pt/IGZO/IGZO:N/TiN device exhibits reduced forming voltage [28]. Typical bipolar I-V characteristics for both devices are shown in Fig. 3(c) and (d). For Pt/IGZO/TiN device, the I-V curves reveal a poor uniformity of RS parameters, which fluctuate seriously during the 100 cycles DC sweeping. For Pt/IGZO/IGZO:N/TiN device, it can be seen that the memristor exhibits greatly improved uniformity with stable bipolar RS properties for 100 cycles.

The distributions of high/low resistance states (HRS/LRS) for both devices are shown in Fig. 4(a). Clearly, it can be found that a higher uniformity of HRS and LRS distribution is achieved in Pt/IGZO/IGZO:N/TiN device, while the resistance variation is observed in Pt/IGZO/TiN device. The OFF/ON ratio of IGZO:N device (HRS/LRS~80, read at -0.2 V) is larger than that of IGZO device (HRS/LRS~50). The distribution of set voltage (V_set) is also analyzed in Fig. 4(b). It can be seen that IGZO:N device presents the smaller V_set distribution in the range of 0.65 V to 0.75 V. By contrast, a wide V_set distribution is observed in the range of 0.3 V to 1.3 V for IGZO device. The higher uniformity of resistance and more concentrated distribution of V_set imply that the electrical performance for the Pt/IGZO/IGZO:N/TiN memristor has been enhanced greatly. Normally, uniformity is a significant factor for the future application of memristive device. For the purpose of precisely studying the uniformity of both devices, we randomly chose three samples from each device and discussed the device-to-device variation in 100 cycles. As shown in Fig. 4(c) and (d), according to the length of the individual box, the LRS distribution for both devices still maintains the same level approximately, while the device-to-device HRS variation for IGZO:N device is much smaller.

Reliability tests, including pulse endurance and retention properties, are performed to evaluate the ability of memory device for data storage. For endurance tests, a 1 V pulse amplitude of set process and -1.2 V of reset process is applied, and a 50 μs pulse width is set, respectively. The resistance values are read at -0.2 V. From Fig. 5(a) and (b), the IGZO device without IGZO:N insertion only reveals a stable endurance cycle to 10⁵ cycles, but the IGZO/IGZO:N device presents an excellent endurance cycle up to 10⁷ cycles. Actually, for the IGZO-based memristor, it is difficult to obtain a reliable endurance characteristic, despite numerous efforts having been made [[11], [12], [13],29]. Here, in this work, we insert a 4 nm ultrathin IGZO:N layer with N₂ plasma treatment and acquire satisfied 10⁷ pulse endurance. On the other hand, the retention properties have been studied in Fig. 5(c) and (d). The OFF/ON ratio can be maintained for 10⁴ s under 125 ℃ and Pt/IZGO/IGZO:N/TiN device shows an improved retention property. In summary, the superior characteristics of endurance and retention indicate that the RS behavior with satisfied electrical performance can fulfill the demand of future applications of memristors.

For the purpose of revealing the current conduction mechanism for the IGZO memristor with and without IGZO:N inserting layer, we further explore the possible switching mechanism for both devices, as shown in Fig. 6(a) and (b). From the current fitting analysis, it can be seen from Fig. 6(d) and (f) that the Schottky emission conduction is dominant in HRS at negative voltage, according to the linear relationship between ln (I) and V^1/2, which implies that electrons surpass the barrier between the CF and TiN electrode. Moreover, the Schottky emission equation can be described by:

$J={{\text{A}}^{**}}{{T}^{2}}\exp \left[ \frac{-\text{q}\left( {{\varphi }_{B}}-\sqrt{qE/4\pi {{\varepsilon }_{i}}} \right)}{kT} \right] $

Where A**, q and ϕ_B stand for the effective Richardson constant, the magnitude of the electronic charge and barrier height, respectively. By taking the logarithm of the equation and depicting the curve of ln (I)-V^1/2, the relationship between the intercept and the slope can be expressed as:

$\left| Intercept \right|\propto {{\varphi }_{B}}$

$ BSlope \propto\sqrt{\frac{1}{\varepsilon _{i}d}}$

In addition, we also analyze the current conduction mechanism at LRS. For LRS of both devices, the carrier transport mechanisms are dominated by Ohmic conduction with a slope of about 1, as is shown in Fig. 6(c) and (e).

In order to clarify the role of IGZO:N inserting layer on the enhancement of RS performance, a RS model was proposed and illustrated in Fig. 7. Through a forming process, oxygen vacancies are produced and constitute the CFs to connect Pt and TiN electrode. In the subsequent operation processes, the connect and rupture of CFs are accompanied by the oxidation and reduction reaction near the TiN electrode. For Pt/IGZO/TiN device, the CF is recovered randomly along some possible paths during the set process, corresponding to widely distributed V_set, as depicted in Fig. 7(a). Moreover, the randomness of ruptured distance of CFs results in very poor uniformity in HRS, this will finally lead to the performance failure of the memory cell. Fig. 7(b) presents the set process for Pt/IGZO/IGZO:N/TiN device. It has been proven that N-O bonding energy is larger than O-O bonding energy [30,31]. That is to say, the nitrogen atom can easily capture the oxygen ions and attract them to concentrate around the CF tip. So the IGZO:N inserting layer can act as an oxygen reservoir layer, which makes it easier for the recovery and rupture of the CFs than that in the Pt/IGZO/TiN device. Due to the N atoms help to suppress the random formation of CFs, the CF tips can grow along fixed paths in IGZO:N layer, which will significantly improve the uniformity of distribution of V_set and resistance state, as well as the endurance properties.

In summary, we designed and investigated bilayer Pt/IGZO/IGZO:N/TiN memristive device using simple method with the goal of enhancing RS performance. With inserting a 4 nm N₂ plasma treated IGZO thin film, the memory cell exhibited lower forming voltage, remarkable uniformity of V_set and resistance state, as well as the 10⁷ pulse endurance. An OFF/ON resistance ratio of at least two orders of magnitude was achieved and the resistive states could remain stable up to 10⁴ s at 125 °C. In contrast, the Pt/IGZO/TiN control device presented an unstable RS characteristic and poor endurance. We speculated that the doped N atoms may play the role of capturing oxygen ions and concentrate them around the CF tip, which helped to reduce the randomness of CF connection and rupture. Overall, the optimized bilayer Pt/IGZO/IGZO:N/TiN device showed competitive switching performance, which is promising in the application of future high resolution flat panel displays.

L.Z. and Z.X. contributed equally to this work. This study was supported by the National Key Research and Development Program under Grant 2017YFB0405600; National Natural Science Foundation of China (No. 61774057); The Open Fund of State Key Laboratory on Integrated Optoelectronics (IOSKL2018KF08). The Open Fund of Hubei Key Laboratory of Applied Mathematics (HBAM 201801).