Fabrication of a resistive switching gallium oxide thin film with a tailored gallium valence state and oxygen deficiency by rf cosputtering process

Abstract

Resistive switching gallium oxide base thin films with tailored oxygen deficiency were fabricated by rf cosputtering of Ga₂O₃ and Cr. XPS and STEM-EDX analyses were used to determine that the resultant film was made of a homogeneous oxide glass layer with mixed valance states of Ga(III)–Ga(I). The amount of Ga(I) and the corresponding oxygen deficiency was precisely controlled because the following redox reaction subsequently progresses within the deposited films: 3Ga(III) + 2Cr(0) → 3Ga(I) + 2Cr(III). The on/off resistance ratio was largely varied by changing the Ga(I) fraction in relation to the oxide ion conductivity, and Ga_0.82Cr_0.18O_1.2 thin film was found to exhibit an optimal switching performance. The film resistance state was tunable by 100's of μs pulse biasing and was incrementally changed by increasing the applied pulse numbers. The strongly time-dependent switching events and area dependent current level of Cr-GaO_x films were distinct from the abrupt switching behavior of the filamentary mechanism TiO_x thin film devices. It was demonstrated that rf cosputtering of the metal oxides and the corresponding oxygen scavenging metals was a powerful technique to design the bulk state resistive switching devices based on nonstoichiometric metal oxide thin films.

---

Introduction

As a potential next generation nonvolatile memory, resistive random access memory (RRAM) with a simple metal/oxide/metal sandwich structure has been studied intensively during the past decades.¹ In the case of well-established Pt/TiO₂/Pt memristor devices, the switching between high resistance state (HRS) and low resistance state (LRS) is triggered by the modification of the tunnel barrier width between a top electrode (TE) and the preformed conductive filament due to the ionic drift within a few nanometre-thick space charge layer in the vicinity of the electrode interface.^2–4

Gallium oxide devices are considered as an ideal candidate for RRAM because of their intrinsic high resistance characteristic and extraordinarily sensitive conductivity towards oxygen concentration.^5–9 In a previous study, we found resistive switching in highly-nonstoichiometric, amorphous gallium oxide thin films and a-GaO_1.1. Originating from the bulk mixed oxide ion-electron conductivity.⁸ The homogeneous migration of oxygen vacancy donors modified the electronic carrier distribution across the films and thus the resistance states of the bulk film varied with the applied field.⁸ Such a bulk-conduction mechanism memristor could realize multilevel states of resistance and a large on/off current ratio due to the continuous tunability of the internal state variable, and therefore is expected to be a dynamical memristor system to emulate synaptic memory behaviour.^10,11 To develop feasible homogeneous resistive switching devices for future neuromorphic applications, it is beneficial to design amorphous gallium oxides with enhanced ion conductivity. The oxide ion conductivity of a-GaO_x must be related with interstitial spaces and local Ga environments of the mixed valence state Ga(III)–Ga(I) oxide glasses. Therefore, it is of fundamental and technological importance to develop the precise fabrication process with tailored oxygen deficiency, namely, Ga(I) fractions in the oxide matrices in order to tune the resistive switching properties of film devices.

Highly-nonstoichiometric amorphous GaO_x thin films have been fabricated by pulse laser deposition (PLD) with a Ga₂O₃ target under low p_O2 conditions.^5,8,9 However, it is difficult to control the yields of Ga(I) suboxides by this method because Ga₂O₃ is not easily reduced under extremely low O₂ pressure at elevated temperature.¹² Lee et al. reported the fabrication of oxygen deficient gallium oxide films by high temperature vacuum annealing of Ga₂O₃/Cr/Ga₂O₃ three layer laminates.⁶ Gallium oxide was efficiently reduced through an interdiffusion reaction because of relatively large negative energy of Cr–O bonding (−1053 kJ mol⁻¹) in comparison to Ga–O bonding (−998.3 kJ mol⁻¹).¹³ Such a redox reaction is useful to prepare Ga(I)–Ga(III) mixed valence gallium oxide. Herein, we successfully fabricated Cr-doped gallium oxide films with a controlled amount of Ga(I) and oxygen deficiency by depositing a homogeneous mixture of Ga(III) oxide moieties and Cr(0) metal atoms. The chromium atoms scavenge the oxygen atoms of Ga(III) oxide moieties and thus the Ga(I)–Ga(III) mixed valence state homogeneous oxide glass was prepared. The films with an optimal amount of Ga(I) fractions revealed remarkable bulk resistive switching with a large on/off ratio according to the efficient oxide ion conductivity, which enabled incremental change of the bulk resistance state by increasing the bias duration or applying short pulse numbers. The switching behavior was compared to the well-established TiO_x thin film base filamentary mechanism switching devices.

Experimental

A 20 nm thick Pt deposited Si wafer (100) was used as the bottom electrode (BE). Nonstoichiometric TiO_x and Cr-doped GaO_x thin films were deposited on a Pt BE by rf sputtering at room temperature. TiO_x thin films were deposited using a Ti target (purity: 99.9%) in a mixed argon and oxygen atmosphere (the ratio of Ar/O₂ was 5 : 1) at 0.4 Pa total pressure at a 150 °C substrate temperature. Cr-doped GaO_x films (Cr-GaO_x) were deposited by cosputtering with Ga₂O₃ and Cr targets in a pure argon atmosphere at a 1 Pa pressure. The sputtering power of Ga₂O₃ was fixed at 50 W and that of Cr was adjusted at 30, 40, 60 and 70 W in order to change Ga/Cr molar ration in the resultant film. The substrate was kept at ambient temperature. Circular metal top electrodes (TE) with a 100–800 μm diameter and 50 nm thickness were sputter-deposited onto the films through a metal shadow mask. The I–V characteristics of the devices were measured at room temperature using source-measure units (KEITHLEY 2601B). In all cases, Pt TE was first swept from 0 V to the anodic region (+1 V), swept back to cathodic region (−1 V) and returned to 0 V so as to complete one voltage sweep cycle. The voltage steps were fixed at 0.1 V.

The chemical composition of Cr-GaO_x films was examined by wavelength dispersive X-ray analysis (WDX) with a JEOL JXA-8530F. For WDX measurements, 800 nm-thick films were deposited onto the Pt/Si wafer by sputtering under the same atmospheric, rf power and temperature conditions as those for the corresponding switching device films. Cross-sectional transmission electron microscopy (TEM) and energy dispersive X-ray fluorescent analysis (EDX) were carried out in a HITACHI HD-2000. The specimens for TEM observation were prepared by a focused ion beam microfabrication (FIB; HITACHI FB-2100). X-ray photoelectron spectroscopy (XPS) was carried out with a JEOL JPS-9010MC with Mg Kα radiation. The depth profile was applied by in situ Ar⁺ ion sputtering. The curve fitting as carried out with a Lorenz–Gaussian function.

Results

First, GaO_x films were fabricated by sputtering a single Ga₂O₃ target. The resulting films were transparent, indicating that the amount of oxygen deficiency in the film is quite small. On the other hand, the uniform, dark-colored thin film could be obtained by cosputtering of Cr and Ga₂O₃. The chemical composition of the films deposited with a Cr sputtering power of 30, 40, 60 and 70 W was Ga_0.96±0.02Cr_0.04±0.02O_1.4±0.1, Ga_0.91±0.04Cr_0.09±0.05O_1.3±0.1, Ga_0.82±0.2Cr_0.18±0.02O_1.2±0.1 and Ga_0.78±0.06Cr_0.22±0.03O_1.2±0.1, respectively, as determined by WDX measurements (Table 1). These films are hereafter referred to as Ga₉₆Cr₄, Ga₉₁Cr₉, Ga₈₂Cr₁₈ and Ga₇₈Cr₂₂, respectively, based on the average metal composition. Fig. 1(a) shows cross-sectional TEM of Ga82Cr18 thin films. The densely-packed films (120 nm thickness) were uniformly formed over a wide area of a Pt/Si wafer substrate. Apparent pinholes and clacks were not observed. STEM-EDX showed chromium was homogeneously distributed throughout the film bulk (Fig. 1(c) and (d)) without segregation, indicating that homogeneous mixtures of chromium and gallium oxide are prepared by the cosputtering. The Ga82Cr18 thin films showed only a halo ring, indicating an amorphous phase (Fig. 1(b)).

Table 1 List of the sputtering power of Cr for Cr-Ga₂O₃ cosputtering deposition and composition and Ga(III)/Ga(I) molar ratios of the corresponding Cr-GaO_x thin films

RF power/W	Name	Composition^a	Ga(III)/Ga(I)^b	Ga(III)/Ga(I)^c
a Determined by WDX.b Determined by XPS.c Calculated by eqn (3).
70 W	Ga78Cr22	Ga0.78±0.06Cr0.22±0.03O1.2±0.1	61/39	58/42
60 W	Ga82Cr18	Ga0.82±0.2Cr0.18±0.02O1.2±0.1	69/31	67/33
40 W	Ga91Cr9	Ga0.91±0.04Cr0.09±0.05O1.3±0.1	77/23	85/15
30 W	Ga96Cr4	Ga0.96±0.02Cr0.04±0.02O1.4±0.1	90/10	93/7

---

Fig. 1 (a) Cross-section TEM image and (b) ED pattern of Ga82Cr12 thin films. Corresponding elemental EDX mapping of (c) Ga and (d) Cr.

Fig. 2 shows Ga 3d and Cr 2p XPS spectra of 120 nm-thick Cr-GaO_x thin films. In all cases, Ga 3d peak was deconvoluted into two peaks: a peak at 20.7 eV due to the Ga(III) state and a peak at 19.6 eV due to Ga(I) state (Fig. 2(a)).^8,14 Ga(III)/Ga(I) molar ratios in Ga96Cr4, Ga91Cr9, Ga82Cr18 and Ga78Cr22 were 90/10, 77/23, 69/31 and 64/36, respectively (Table 1). These values of the molar ratios were consistent with ones calculated from chemical composition by accounting for charge compensation between cations (Ga³⁺, Ga⁺ and Cr³⁺) and anion (O²⁻). On the other hand, Cr 2p XPS spectral features were not changed by chemical composition (Fig. 2(b)). All the films revealed apparent spectral feature of Cr(III) with Cr 2p_1/2 and 2p_3/2 peaks at 587.5 eV and 577 eV, respectively,¹⁵ whereas the features of lower valence state Cr, such as Cr(0) (574.2 eV of 2p_3/2)¹⁶ and Cr(I) (576.3 eV of 2p_3/2),¹⁶ did not appear (Fig. 2(c)).

Fig. 2 (a) Ga 3d and (b) Cr 2p XPS spectra of 120 nm-thick thin films of (a1 and b1) Ga78Cr22, (a2 and b2) Ga82Cr18, (a3 and b3) Ga91Cr9 and (a4 and b4) Ga96Cr4. (c) Ga 3d XPS depth profile of 120 nm-thick Ga82Cr18 thin films. The spectra were stored at the surface (c1) and at the points in 60 nm (c2) and 100 nm depth (c3). Red dots are the observed data and the black lines the simulated.

Ga 3d XPS spectra of Ga82Cr18 were also obtained at about a 50 nm (middle) and 100 nm (near bottom edge) depth by Ar⁺ ion milling (Fig. 2(c)). The spectra show similar features at any depth (Fig. 2(b)), indicating the Ga(I) states were homogeneously distributed throughout the film.

Guo et al. reported that the switching properties of a-GaO_1.2 thin film devices strongly depend on the electrode materials.⁵ Pt was used as a top electrode material in this study in order to avoid oxidation of the electrode by contacting the oxide films.

Resistive switching behavior of Cr-GaO_x films strongly depends on the composition (Fig. 3). The Ga96Cr4 thin films gave two orders of magnitude smaller currents than those of other composition and the hysteretic I–V curves did not appear at any sweep rates (Fig. 3(a)). The others, Ga91Cr9, Ga82Cr18 and Ga78Cr22, involved resistive switching at a 1 and 0.01 V s⁻¹ sweep rate (Fig. 3(b)–(d)), revealing a pinched hysteresis I–V loop with a counter clockwise figure-eight polarity when the voltage on TE was swept in the way of 0 → +1 → 0 → −1 → 0 V at every composition (Fig. 3). A positive voltage RESET the film from the low resistance state (LRS) to the high resistance state (HRS), and with the negative bias the film SET to the LRS, indicating bipolar resistive switching behavior of the films. The electroforming process using a high applied voltage was not required to initiate the switching. These results were consistent with the previous studies of a-GaO_1.1 thin film devices.^5,8

Fig. 3 I–V characteristics of Pt(TE)/Cr-GaO_x film/Pt(BE) devices thickness during a repeated voltage sweep cycle of 0 → +1 → −1 → 0 V at different voltage sweep rates. (a1)–(a3) Ga92Cr4, (b1)–(b3) Ga91Cr9, (c1)–(c3) Ga82Cr18 and (d1)–(d3) Ga78Cr22 thin films with 120 nm thickness. The sweep rate was 100 V s⁻¹ in (a1), (b1), (c1) and (d1), 1 V s⁻¹ in (a2), (b2), (c2) and (d2) and 0.01 V s⁻¹ in (a3), (b3), (c3) and (d3). The cycles were repeated 20 times in (c2) and 10 times in others. The numbered arrows indicate the direction of the switching cycles (counter-figure-eight loops). 1st, 10th and 20th curves are shown by blue, red and green, respectively. The curves of other cycles are shown in black. The current levels in (a1)–(a3) are multiplied by 30.

The hysteretic shape of the I–V curves strongly depends on the voltage sweep rate, e.g. see Fig. 3(b)–(d). The films did not reveal hysteresis loops at fast sweep rates (100 V s⁻¹), but tended to reveal wider hysteretic loops at slow sweep rates. The switching behavior was stationary and did not decay for 10's voltage sweep cycles.

The Ga82Cr18 thin film was distinct from other composition films. The films revealed the large hysteresis loop at 1 and 0.01 V s⁻¹ sweep rates and possess the highest HRS/LRS resistance ratio (on/off ratio), which was about 20 at 0.7 V (Fig. 4(a)). The ratios decreased to about 2 if the Cr dopant concentrations, namely, Ga(I) fractions, were decreased or increased with Ga91Cr9 or Ga78Cr22, respectively.

Fig. 4 (a) On/Off switching ratio vs. Cr content at 0.7 V, measured at of 100 (+), 1 (○) and 0.01 (△) V s⁻¹ voltage sweep rates. (b) TE area scaling of the current of LRS Ga82Cr18 and TiO_x thin films at −0.5 V.

The area dependence of the current level in Ga82Cr18 film showed linear scaling (Fig. 4(b)). This feature discloses that the resistance state of the film below TE was homogeneously changed by responding to the applied field or a number of conducting filaments that were uniformly distributed over the electrode area.

In a separate experiment, Pt/TiO_x/Pt thin film devices by the similar sputtering deposition processes were fabricated, as reported elsewhere.¹⁷ Rutile-type TiO_x thin films with a 100 nm thickness were uniformly formed on the Pt electrode, as confirmed by the TEM and ED patterns (Fig. S1a†). The TiO_x thin film prepared here reveals similar switching behavior as the filamentary mechanism switching TiO_x devices that were reported previously.^2,4 I–V hysteretic shape did not vary by decreasing the voltage sweep rate (Fig. S1b†) in the less than 100 V s⁻¹ range. The current level was independent of the Pt TE area (Fig. 4(b)). This is peculiar to the resistive switching triggered by variation of local states such as formation and deformation of the conductive filament.

It is evident that resistive switching of Cr-GaO_x was quite different from TiO_x. The strong time-dependent switching events and area dependent current level was incompatible with the filamentary mechanism as operating in TiO_x devices. It is concluded that the resistive switching of Cr-GaO_x films was driven by variation of bulk states.^8,18

The bulk switching mechanism suggests that the film resistance can be incrementally adjusted by tuning the duration and sequence of the applied bias.¹⁹ Unlike the devices with abrupt switching characteristics wherein the programming bias controls the final device state,^20,21 each programming pulse controls the relative change of the film resistance in case of Cr-GaO_x devices because donor concentration profiles across the film can be varied sequentially by the migration at each programing bias. Fig. 5 shows the resistance state change of Ga82Cr18 films when a series of 70 identical negative (−0.95 V, 500 μs) pulses were applied to the films, followed by a series of 70 identical positive voltage pulses (0.95 V, 500 μs). The large pulse bias (namely, −0.95 V or 0.95 V) programs the resistance state of the films and a small read bias of 0.1 V measures a response current to represent the resistance state. As expected from the DC characteristics of the device, the application of negative voltage pulses (−0.95 V) incrementally decreased the film resistance, and the application of positive voltage pulses (+0.95 V) incrementally increased the resistance.

Fig. 5 Current response transients of Ga82Cr18 thin films, measured by applying a 500 μs programming bias (+0.95 or −0.95 V) and 100 μs read bias (+0.1 V). Red is the current measured at 0.1 V read pulse after a programmed −0.95 V pulse and blue is the one after a programmed a 0.95 V pulse. The insets show pulse patterns used for the measurements.

Discussion

It is clear from the results of XPS and STEM-EDX measurements (Fig. 1 and 2) that co-sputtered Cr(0) metallic atoms scavenge oxygen from adjacent Ga(III) oxide moieties by the following redox reaction:

2Cr(0) + 3Ga(III) → 2Cr(III) + 3Ga(I) (1)

The chemical composition of the resultant film can be given as follows:

(1 − x)Ga^IIIO_1.5 + xCr⁰ ⇆ (Ga^III_1−yGa^I_y)_1−xCr^III_xO_1.5(1−x). (2)

Then, the charge balance yields as follows:

y = 1.5x(1 − x)⁻¹ (3)

Reaction (3) shows how the amount of oxygen deficiency (1.5x) is related to the Ga(I) fraction (y). The molar fractions of Ga(I) of the films prepared here were in agreement with the values calculated by reaction (3) (Table 1). It is apparent that oxygen deficiency is involved by reaction (2) during cosputtering.

The switching performance of Cr-GaO_x thin films was drastically changed by Ga(I) fractions (Fig. 3). The schematic model of the bulk mechanism switching of gallium oxide base thin films is shown in Fig. 6. The positive bias on small area TE accumulates negatively-charged oxide ions in the vicinity of the TE blocking electrode by extinguishing electron carriers so as to form highly-resistive, oxygen-rich layer near TE as follows (Fig. 6(a)):

Ga(I)O_0.5 + 2e′ + O ⇆ Ga(III)O_1.5. (4)

Fig. 6 Schematic of the bulk mechanism resistive switching in Cr-GaO_x thin film devices. (a) Positive bias and (b) negative bias on TE.

The opposite bias can enrich the electron carriers in the layer again by driving oxide ions towards large area BE (Fig. 6(b)). At fast voltage sweep rates, the resistance ratio between HRS and LRS, namely, the hysteretic width of I–V loop tended to be small because the shift of oxygen vacancy donor distribution cannot be largely changed due to the relatively short bias duration. However, the hysteresis became wider at slower sweep rates due to the large shift in donor distribution by a longer bias duration.

Based on the mechanism, the poor resistive switching performance of low Ga(I) fraction films, such as Ga96Cr4, can be related to the poor oxide ion conductivity because the low oxygen deficiency may not afford sufficient space for the pathway of oxide ion migration. The Cr(III) contents are not beneficial for the oxide ion conduction as the dissociation of Cr–O bonding requires relatively large energy. This might be one of the reasons for the deteriorated switching performance of the high chromium content films, Ga78Cr22 (Fig. 3(d)). The current results suggest that the optimal molar fraction of Cr/Ga is nearly 18/82 in the relation to fast oxide ion conduction in highly oxygen deficient glass matrices.

The resistive switching performance of 120 nm-thick Ga82Cr18 thin films is superior to that of the GaO_x films (x = 1.1–1.2) given by PLD in previous studies.^5,8 The former exhibits the on/off ratio of 20 at 0.7 V; however, the ratios of the later is about 2 at the same voltage. These results reveal that the oxide glass network of the cosputtered Ga82Cr18 films retain the efficient oxide ion conductivity as much as the PLD films. The Ga82Cr18 film exhibits remarkable multiple resistance states as responding to short pulse biasing (Fig. 5), indicating that the significant shift of oxygen vacancy donors for resistance change can be induced by the bias duration of 100's of μs.

The oxide ion mobility in M-doped GaO_x thin films may be largely varied by choice of dopant cations as is the case with various ceramic oxide ion conductors.²² Except for chromium, various metals, such as Mg, Ti, and Al, are potential candidates as an oxygen scavenging dopant of Ga₂O₃ from the view point of M–O bond formation energy. Recently, Chang et al. reported that oxygen-deficient, amorphous WO_x thin film also shows multiple resistive switching probably due to oxide ion conduction.²³ This suggests that not only gallium oxide but also other nonstoichiometric metal oxide systems can be considered as a possible candidate of bulk oxide ion conducting thin films. Most of metals and binary metal oxides can be used as a target material of rf sputtering techniques. It is concluded that cosputtering with a pair of resistive switching oxides and oxygen scavenging metal dopant affords a strong tool to tune the oxide ion conductivity of the highly-nonstoichiometric, resistive switching metal oxide thin films.

Conclusion

In summary, the fabrication method of Cr-doped gallium oxide thin films with a tailored gallium valence state and oxygen deficiencies has been established, based on the reactive cosputtering with Ga₂O₃ and Cr targets. The Ga(I)–Ga(III) mixed valence state homogeneous oxide glass can be formed from a homogeneous mixture of Cr(0) metal atoms and Ga(III) oxides moieties deposited by cosputtering, because Cr(0) scavenges oxygen from Ga(III) oxide due to the relatively large negative energy of Cr–O bonding compared to Ga–O bonding. The Ga82Cr18 thin film with optimal Ga(I) contents can reveal remarkable homogeneous resistance switching due to the efficient bulk oxide ion conduction. The on/off ratio of the film was similar to that of the highly-nonstoichiometric GaO_1.1 thin films prepared by PLD. The film clearly involved multiple resistance state switching wherein the final device state was determined by the bias history, since the oxygen vacancy donor profiles can be incrementally modified by bias duration or the number of applied short bias pulses. The current results open up a new way to design the bulk mechanism resistive switching metal oxide thin films with tailored oxide ion conductivity.

Acknowledgements

This study was conducted at the Hokkaido University, supported by "Nanotechnology Platform" program of the MEXT Japan. CK was supported by the MEXT Japan through the program for Leading Graduate Schools (Hokkaido University “Ambitious Leader's Program”). YA is grateful for financial support for his stay in RWTH Aachen University by the Deutsche Forschungsgemeinschaft (DFG) within the SFB 917 “Nanoswitches”.

References

1. D. B. Strukov, G. S. Snider, D. R. Stewart and R. S. Williams, Nature, 2008, 453, 80; R. Waser, R. Dittmann, G. Staikov and K. Szot, Adv. Mater., 2009, 21, 2632.
2. J. J. Yang, M. D. Pickett, X. Li, D. A. A. Ohlberg, D. R. Stewart and R. S. Williams, Nat. Nanotechnol., 2008, 3, 429; K. K. Min, J. D. Seok and H. C. Seong, Nanotechnology, 2011, 22, 254002; D.-H. Kwon, K. M. Kim, J. H. Jang, J. M. Jeon, M. H. Lee, G. H. Kim, X.-S. Li, G.-S. Park, B. Lee, S. Han, M. Kim and C. S. Hwang, Nat. Nanotechnol., 2010, 5, 148; Y. Sato, K. Kinoshita, M. Aoki and Y. Sugiyama, Appl. Phys. Lett., 2007, 90, 033503; J. P. Strachen, M. D. Pickett, J. J. Yang, S. Aloni, A. L. D. Kilcoyne, G. M. Ribeiro and R. S. Williams, Adv. Mater., 2010, 22, 3573; J. H. Hur, M.-J. Lee, C. B. Lee, Y.-B. Kim and C.-J. Kim, Phys. Rev. B: Condens. Matter Mater. Phys., 2010, 82, 155321.
3. R. Muenstermann, T. Menke, R. Dittmann and R. Waser, Adv. Mater., 2010, 22, 4819.
4. M. D. Pickett, D. B. Strukov, J. L. Borghetti, J. J. Yang, G. S. Snider, D. R. Stewart and R. S. Williams, J. Appl. Phys., 2011, 106, 074508; F. Alibert, L. Gao, B. D. Hoskins and D. B. Strulov, Nanotechnology, 2012, 23, 075201.
5. X. Gao, Y. Xia, J. Ji, H. Xu, Y. Su, H. Li, C. Yang, H. Guo, J. Yin and Z. Liu, Appl. Phys. Lett., 2010, 97, 193501.
6. D.-Y. Lee and T.-Y. Tseng, J. Appl. Phys., 2010, 110, 114117.
7. D. Y. Guo, Z. P. Wu, Y. H. An, P. G. Li, P. C. Wang, X. L. Chu, X. C. Guo, Y. S. Zhi, M. Lei, L. H. Li and W. H. Tang, Appl. Phys. Lett., 2015, 106, 042105; X. B. Yan, H. Hao, Y. F. Chen, Y. C. Li and W. Banerjee, Appl. Phys. Lett., 2014, 105, 093502; D. Y. Guo, Z. P. Wu, L. J. Zhang, T. Yang, Q. R. Hu, M. Lei, P. G. Li, L. H. Li and W. H. Tang, Appl. Phys. Lett., 2015, 107, 032104; J. B. Yang, T. C. Chang, J. J. Huang, S. C. Chen, P. C. Yang, Y. T. Chen, H. C. Tseng, S. M. Sze, A. K. Chu and M. J. Tsai, Thin Solid Films, 2013, 529, 200; C. W. Hsu and L. J. Chou, Nano Lett., 2012, 12, 4247.
8. Y. Aoki, C. Wiemann, C. Fayer, H.-S. Kim, C. M. Schneider, H. I. Yoo and M. Martin, Nat. Commun., 2014, 5, 4473.
9. L. Nagarajan, R. A. DeSouza, D. Samuelis, I. Valov, A. Borger, J. Janek, K.-D. Becker, P. C. Schmidt and M. Martin, Nat. Mater., 2008, 7, 391.
10. J. J. Yang, D. B. Strukov and D. R. Stewart, Nat. Nanotechnol., 2013, 8, 13.
11. T. Chang, S. H. Jo, H. K. Kim, P. Sheridan, S. Gaba and W. Lu, Appl. Phys. A: Mater. Sci. Process., 2011, 102, 857.
12. D. G. Kolman, T. N. Taylor, Y. S. Park, M. Stan, D. P. Butt, C. J. Maggiore, J. R. Tesmer and G. J. Havrille, Oxid. Met., 2001, 56, 347.
13. D. R. Lide, Handbook of Chemistry and Physics, CRC Press, 84th edn, 2003, p. 74.
14. R. Carli and C. L. Bianchi, Appl. Surf. Sci., 1994, 74, 99; C. C. Surdu-Bob, S. O. Saied and J. L. Sullivan, Appl. Surf. Sci., 2001, 183, 126.
15. K. Petkov, V. Krastev and T. Marinova, Surf. Interface Anal., 1992, 18, 487.
16. J. Sainio, M. Aronniem, O. Pakarinen, K. Kauraal, S. Airaksine, O. Krause and J. Lahtinen, Appl. Surf. Sci., 2005, 252, 1076.
17. D. S. Jeong, H. Schroeder and R. Waser, Appl. Phys. Lett., 2006, 89, 082909; D. S. Jeong, H. Schroeder and R. Waser, Electrochem. Solid-State Lett., 2007, 10, G51.
18. R. Meyer, L. Schloss, J. Brewer, R. Lambertson, W. Kinney. J. Sanchez and D. Rimerson, IEEE Proceed. NVMTS2008, 2008, p. 1.
19. Y. B. Nian, J. Strozier, N. J. Wu, X. Chen and A. Ignatiev, Phys. Rev. Lett., 2007, 98, 146403.
20. M. J. Lee, S. Han, S. H. Jeon, B. H. Park, B. S. Kang, S. E. Ahn, K. H. Kim, C. B. Lee, C. J. Kim, I. K. Yoo, D. H. Seo, X. S. Li, J. B. Park, J. H. Lee and Y. Park, Nano Lett., 2009, 9, 1476.
21. S. H. Jo, K. H. Kim and W. Lu, Nano Lett., 2009, 9, 496.
22. S. Grieshammer, B. O. H. Grope, J. Koettgen and M. Martin, Phys. Chem. Chem. Phys., 2014, 16, 9974; S. G. Kang and D. S. Sholl, RSC Adv., 2013, 3, 3333.
23. T. Chang, S. H. Jo and W. Lu, ACS Nano, 2011, 5, 7669; S. H. Jo, T. Chang, I. Ebong, B. B. Bhadviya, P. Muzumder and W. Lu, Nano Lett., 2010, 10, 1297.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5ra21160c

---