Highly durable and flexible gallium-based oxide conductive-bridging random access memory

The flexible conductive-bridging random access memory (CBRAM) device using a Cu/TiW/Ga2O3/Pt stack is fabricated on polyimide substrate with low thermal budget process. The CBRAM devices exhibit good memory-resistance characteristics, such as good memory window (>105), low operation voltage, high endurance (>1.4 × 102 cycles), and large retention memory window (>105). The temperature coefficient of resistance in the filament confirms that the conduction mechanism observed in the Ga2O3 layer is similar with the phenomenon of electrochemical metallization (ECM). Moreover, the performance of CBRAM device will not be impacted during the flexibility test. Considering the excellent performance of the CBRAM device fabricated by low-temperature process, it may provide a promising potential for the applications of flexible integrated electronic circuits.


Results
For both forming and set processes, the current compliance of 100 μA is crucial to avoid a permanent or hard breakdown. The forming process was required to activate the CBRAM devices under DC sweeping mode. The forming voltage of the as-deposited Ga 2 O 3 CBRAM device was 5.9 V. However, the forming voltages of the device annealed at 200 °C in N 2 atmosphere is significantly decreased to 4.4 V. Figure 2(a,b) show the typical bipolar current-voltage (I-V) curves of the as-deposited CBRAM device and the device annealed at 200 °C on flexible PI, respectively. In the beginning, the low-resistance state (LRS) is achieved when conductive filaments in the Ga 2 O 3 switching layer are produced by Cu ion migration and stacking from the Cu to Pt bottom electrode in the Ga 2 O 3 layer; it's called the set process. Conversely, the high-resistance state (HRS) is achieved when conductive filaments in the switching layer are ruptured by Cu ion migration and stacking from the Pt to Cu electrodes in the Ga 2 O 3 layer; this is called the reset process. The DC endurance of the as-deposited device and the device annealed at 200 °C in N 2 atmosphere are depicted in Fig. 2(c,d), respectively. The performance of the as-deposited device is stable only up to 700 cycling endurance, while the device annealed at 200 °C in N 2 atmosphere exhibits the better endurance cycles over 1.4 × 10 3 . In order to investigate the uniformity of the switching parameters, the statistical variations in set voltage (V set ) and reset voltage (V reset ) for both devices were checked. Figure 2(e) indicates a substantial difference in the statistical distributions of V set and V reset between the devices. The coefficient of variation (CV) expresses the variation as a percentage of the mean, and is defined as the ratio of the standard deviation (σ) to the mean value (μ). The distribution is improved from the as-deposited devices (CV set = 33.8%, CV reset = 58.8%) to the device annealed at 200 °C in N 2 atmosphere (CV set = 29.1%, CV reset = 57.8%). Moreover, the retention characteristics of our device are also studied, as shown in Fig. 2(f). For Ga 2 O 3 device annealed at 200 °C in N 2 atmosphere, HRS and LRS are quite stable, without significant resistance decay (HRS/LRS > 10 5 ) even after 10 4 s operation at room temperature. According to the results mentioned above, the CBRAM annealed at 200 °C in N 2 atmosphere shows good performance and evinces its potential for the memory applications.
To further investigate the physical mechanism for the endurance improvement caused by the post-deposition annealing in N 2 atmosphere, material analysis was performed. The XPS spectra were performed with PHI Quantera SXM, using Al Kα source (a beam power of 25 W and an emission current of 4.025 mA). The XPS spectrum of the O 1 s signal was applied to examine the oxygen binding states at the surface of the Ga 2 O 3 thin film. The XPS result of the sample with Ga 2 O 3 switching layer without annealing process is shown in Fig. 3(a). The O 1 s peak can be fitted by two nearly Gaussian distribution peaks, approximately located at 529.8 (FWHM = 1.6) and 531.1 eV (FWHM = 1.5), respectively. The lower binding energy peak located at 529.8 eV is attributed to oxygen-lattice bonds, which are related to the O 2− ions combined with the Ga atoms in the Ga 2 O 3 compound system. On the other hand, the higher binding energy peak located at 531.1 eV can be attributed to oxygen-vacancy (O V ) bonds. The O V peak is attributed to the oxygen deficient in the Ga 2 O 3 matrix. According to the XPS results shown in Fig. 3(a,b), it is important to notice that the proportion of oxygen-vacancy bonds (14.81%) of Ga 2 O 3 film annealed at 200 °C in N 2 atmosphere is higher than the one without annealing (7.17%). Therefore, the Ga 2 O 3 CBRRAM devices at 200 °C in N 2 atmosphere exhibit good switching behaviors and lower V set and V reset caused by a considerable amount of oxygen vacancies in the oxide layer. This improvement can be attributed to these considerable oxygen vacancies which may lower the energy cost of Cu insertion into the Ga 2 O 3 layer, and lead to the Cu migration in the switching layer 32 . In order to further understand the nature of conductive filaments, the conduction mechanism can be confirmed by the curve fitting. Firstly, we consider a linear temperature dependence of the conductive filament typical of metallic behavior, that is, R(T) = R(T 0 ) [1 + γ(T − T 0 )], where R(T) is the LRS resistance at temperature T, R(T 0 ) is the LRS resistance at room temperature T 0 , and γ is the temperature coefficients of resistance 36,37 . The LRS resistance of the devices is taken from 298 to 358 K and linearly increases with temperature, as shown in Fig. 4. For our devices, the temperature coefficients of resistance (γ) is about 8.77 × 10 −3 K −1 . As shown in Table 1, the result is one order of magnitude more than that of oxygen vacancy assisted filaments (6.03 × 10 −4 K −1 ) 38 . The temperature coefficients obtained are close to the value 1.3 × 10 −2 K −1 for high-purity Cu assisted filaments 39 . Therefore, the formation and rupture of Cu-based conductive filaments is responsible for the resistive switching behavior in our devices.
Good mechanical flexibility is crucial for applications in flexible electronics. Then, the substrate is bent to different curvature radius of 5.0, 3.0, 1.0, and 0.5 cm with both tensile and compressive stresses. After each bending operation, the performance is evaluated by recording I-V curves on five different devices. As shown in Fig. 5(a,b), the R LRS /R HRS values are stable at more than 10 5 even after it is bent with different curvature radius. The statistical data in Fig. 5(c-f) may reveal that V set , and V reset can be kept, almost the same with those of a fresh device. Furthermore, a continuous bending test of up to 10 4 times with curvature radius of 5.0 cm is carried out. Even at 10 4 bends, the R LRS /R HRS can be maintained, as shown in Fig. 5(g,h). These results indicate that a flexible Ga 2 O 3 www.nature.com/scientificreports www.nature.com/scientificreports/ CBRAM device with 200 °C annealing in N 2 atmosphere exhibits good flexibility, which suggests its possibility of commercially viable nonvolatile memory devices.
In summary, an amorphous Ga 2 O 3 CBRAM device annealed at 200 °C in N 2 atmosphere on low-cost flexible substrate is proposed for achieving good memory window, stable retention characteristics and a 2 times enhancement in endurance in this study. Through the current-voltage measurements, the device can be reliably operated in a low set voltage of 1.3 V and a low reset voltage of −0.65 V. The mechanism of the endurance enhancement effects is also ascribed to the generation of a considerable amount of oxygen vacancies in the Ga 2 O 3 layer by N 2 annealing, which may lead to the formation of Cu filaments. Analysis of the XPS O 1 s depth distribution profile has been used to confirm this inference. Therefore, oxygen vacancies play a crucial role in resistive switching mechanism in Ga 2 O 3 CBRAM device. The extraction about the temperature dependence of resistance is an effectual method to examine the nature of the conductive filaments. The temperature coefficient of resistance is about 8.77 × 10 −3 K −1 in this work, which means the devices are ECM-based RRAM with Cu filament. These results indicate that the flexible Ga 2 O 3 CBRAM is attractive for low-cost wearable devices and suitable for the future bendable displays.   www.nature.com/scientificreports www.nature.com/scientificreports/

Methods
The CBRAM using Cu/TiW/Ga 2 O 3 /Pt stacks were successfully fabricated with limited process temperature of 200 °C on a flexible PI substrate. Firstly, the PI substrate was cleaned ultrasonically with ethanol for 5 mins and DI water. After cleaning up a PI substrate, a thin 100 nm SiO 2 buffer layer was deposited on PI substrate by plasma enhanced chemical vapor deposition. Then, 100 nm-thick Pt bottom electrode with 5 nm-thick Ti adhesion layer was deposited by using direct-current (DC) magnetron sputtering. Then, a 20 nm-thick Ga 2 O 3 film was deposited  www.nature.com/scientificreports www.nature.com/scientificreports/ on a Pt/Ti/SiO 2 /PI substrate by radio-frequency magnetron sputtering at room temperature. The deposition pressure was kept at 0.4 Pa, while the ratio of Ar:O 2 gas flow was kept at 1:1. The optimization of Ga 2 O 3 resistive switching layer deposition was applied in this work. Except for the control sample, a post-deposition annealing in N 2 atmosphere is applied at 200 °C for 30 mins. Finally, 1.5 nm-thick TiW barrier layer and 100 nm-thick Cu top electrodes were deposited by using DC magnetron sputtering and patterned with a shadow mask with a diameter of 100 μm. Electrical measurements were recorded by Keithley 4200 semiconductor characterization analyzer. To investigate the endurance improvement, transmission electron microscope (TEM) and X-ray photoelectron spectroscopy (XPS) were used to analyze the material properties.