Volatile and non-volatile behavior of metal–insulator transition in VO₂ through oxygen vacancies tunability for memory applications

The oxygen vacancies are used widely to control the electrical properties of oxides since they act as charge carriers.^1–5 In the electron-correlated systems like VO₂, they play a more critical role in affecting the metal–insulator transition (MIT).² Oxygen vacancies can affect charge degrees of freedom through meddling electron–electron interaction. This would in turn affect lattice, orbital, and spin degrees of freedom. That is, they disturb the lattice short-range ordering, change the valance state and, hence, electronic band structure, and affect the spin configuration. The effect of such a reconfiguration is changing the physics behind MIT. The combination of Peierls and Mott physics has shown to derive MIT.^6,7 The MIT is triggered by dimerization of V–V ions which lifts the antibonding V–O (π*) orbital.^8,9 The splitting of d$t2g$-orbital into bonding and antibonding then creates a bandgap which is a result of V–V covalent bonding (Peierls physics) and electron–electron correlation (Mott physics).^8,9 By introducing oxygen vacancies, we rearrange the structure to go through the transition with Mott physics domination.^10–16 

For VO₂ as a Mott memory, the “0” and “1” states can be defined by insulator and metal states, respectively. The benefits of using Mott memories include but are not limited to the fact that phase transitions could be triggered at sub-nanosecond timescales^17,18 and the state of devices can be accessed electrically. Theoretically, oxygen vacancies can be used as the “write” stimuli to derive the transition at any temperature below the actual MIT temperature.

This tunability provides us with the opportunity to enable nonvolatile operations and utilize VO₂ in neuromorphic applications. For neuromorphic circuits, the synapses should be nonvolatile, whereas the neurons need to be volatile components.¹⁹ If the transition is triggered by voltage or temperature, the operation is volatile. This is due to the reason that, in both cases, the metallic state cannot be retained after the stimuli are removed. However, if the transition is induced by defects which act as charge carriers, the behavior is nonvolatile due to the persistence of defect presence.¹⁹ 

It is also worth mentioning that VO₂ is a great memory candidate from Moor's law point of view since it is shown that the phenomenon of phase transition is achievable down to almost 4 nm.²⁰ Recently, it was shown that oxygen vacancies stabilize the metallic phase at room temperature (RT) through the ionic liquid gating method as well as high-temperature vacuum annealing.^21–23 The only remaining concern would be whether multiple cycles of adding and removing oxygen vacancies provide reliable and repeatable “0” and “1” states. To test this hypothesis, we investigate the metallicity of VO₂ thin films ≈40 nm in the VO₂/NiO/c-sapphire heterostructures. We introduce oxygen vacancies to the system to investigate the structural changes and behavior of electronic transition in response to that process. The oxygen annealing is then employed to investigate the reversibility of the metallicity. Moreover, the reversible behavior of VO₂ in response to multiple vacuum stages and thereafter oxygen annealing cycles are presented. The annealing is conducted in two different temperatures since this is a diffusion-based process and has an exponential temperature dependency.

We compare the results from as-deposited (pristine), vacuum annealed (VA), and oxygen annealed VO₂/(111)NiO/(0001)Al₂O₃ heterostructures. The pristine samples have been grown above the critical thickness to study solely the effect of introducing defects without pinning the lattice degrees of freedom.^23,24 The vacuum annealing process designed to introduce oxygen vacancies into the system. Thereafter, the oxygen annealing process introduces oxygen to fill out the vacancies. To study the crystallographic properties of vacuum annealed (low and high temperatures, respectively, represented as VA-low T and VA-high T) and oxygen annealed VO₂ thin films, XRD patterns of the pristine and annealed samples are collected and the results are presented in Fig. 1(a). The inset figure represents the full range including NiO and sapphire peaks as well. For the pristine sample, (020) monoclinic M₁ diffraction peak is observed, while for the vacuum annealed samples, a different diffraction peak shows up which is indicative of (200) M₂ monoclinic planes of VO₂.¹⁵ The monoclinic structure in the vacuum annealed samples has a different symmetry belonging to the C2/m space group and is similar to what is reported for the high-pressure monoclinic phase of VO₂.^25,26 The oxygen annealing is employed on vacuum annealed samples and interestingly the peak position switches back to (020) M₁ monoclinic VO₂ as shown in Fig. 1(a). The scanning transmission electron microscopy (STEM) was used to characterize these phases and sheds light on the atomic structure of these thin films. Figure 1(b) represents the low magnification high-angle annular dark-field (HAADF) image of the VO₂/NiO/c-sapphire epitaxial heterostructures where VO₂ is grown above the critical thickness. The atomically sharp interface and crystalline quality of the VO₂/NiO layers are shown in Fig. 1(c). The HAADF image in Fig. 2(a) shows the cross section of the pristine-VO₂/NiO-VO₂ interface, and the pristine-VO₂ higher magnification structure is shown in Fig. 2(b). These micrographs illustrate M₁ monoclinic phase for pristine-VO₂ as it is also shown schematically in the inset figure. Figure 2(c) is a cross section of the VO₂/NiO-VO₂ interface in the vacuum annealed sample at a high temperature (450 °C). The HAADF image in Fig. 2(d) shows an atomic resolution micrograph of the VA-high T. This structure is representative of the M₂ monoclinic phase of VO₂ as also is depicted in the inset figure and confirmed by XRD analysis. Figure 2(e) shows the interface of VO₂/NiO-VO₂ after vacuum annealing at a lower temperature (200 °C). Figure 2(f) captures the atomic structure of VO₂ in the VA-low T sample which belongs to the M₂ monoclinic phase (shown in the inset figure schematically).

The temperature-dependent electrical resistivity measurements for pristine and annealed samples are shown in Fig. 3. The pristine sample shows a typical hysteresis belonging to relaxed VO₂ thin films.²⁷ It is observed that after annealing at a low temperature, the metal–insulator transition amplitude decreases, and the resistivity of the insulating is lower compared to the pristine sample. We hypothesize two reasons for the observed behavior. Since, in the M₂ phase of VO₂, there are two types of V–V chains where only in one of them the dimerization occurs and takes part in changing conductivity, the electrical conductivity behavior is different from the M₁ phase.^28,29 Second, the presence of oxygen vacancies which act as charge carries is not neglectable. This can explain the lower resistivity in the insulating phase of this sample. After the transition to the metallic phase, also, the resistivity is lower compared to the pristine sample which is also justified with the presence of more charge carriers (oxygen vacancies) in the system.

After annealing at high temperature (450 °C), the metal–insulator transition vanishes. The vacancies introduced during the vacuum annealing derive the formation of a metallic M₂ monoclinic phase. It is explained that the transition of tetragonal VO₂ to M₂ phase is a Mott type vs to M₁ which is a combined Mott and Peierls transition.²⁹ We believe that in M₂, the role of Mott transition is dominant and the role of Peierls transition is not as strong as in the M₁ phase. The smaller bandgap and the dominance of the Mott nature of transition in M₂ make it easier to manipulate bandgap by introducing charge carriers. When the carrier concentration introduced to the system is above the Mott criterion at room temperature [$nc=(1/4aH)3$ ≈ 3 × 10¹⁸ cm⁻³], energetically it is favorable for the Mott system to stay in the metallic state. The inset figure represents the full temperature range behavior for this stabilized phase which shows no signs of V₂O₃ or V₂O₅ transition upon the introduction of the vacancies. The base resistivity of the metallic M₂ phase is higher than both the metallic state of the pristine and low-temperature vacuum annealed VO₂ samples. This can be explained due to the presence of too many charge carriers in this sample which increases their collision leading to enhanced resistivity. The introduction of charge carriers enhances the conductivity as explained in the case of the low-temperature vacuum annealed sample. However, the introduction of too many oxygen vacancies creates high-density of scattering centers, which leads to lower conductivity in the metallic phase compared to the metallic phase derived with or without a lower amount of oxygen vacancies. In the low-temperature vacuum annealed sample, the thermodynamic equilibrium vacancy concentration is lower than the high-temperature vacuum annealed sample. Table I compares the carrier concentration in all three samples of pristine, low-temperature vacuum annealed, and high-temperature vacuum annealed. Interestingly, even at low temperatures, the carrier concentration meets the Mott criterion. However, since the transition in VO₂ is not pure Mott and needs Peierls assistance, the higher concentration of charge carriers needed to derive the formation of a stable metallic phase at room temperature. This validates our hypothesis about the slight role of Peierls in MIT transition in the M₂ phase of VO₂. This phenomenon has been also confirmed by the report on VO₂ thin films which were grown below the critical thickness.²³ In this report, the VO₂ structure was strained where a lower concentration of charge carriers could drive the MIT.

Figure 4 depicts the electron energy loss spectroscopy (EELS) of pristine, vacuum annealed samples at low and high temperatures, as well as literature extracted spectra of VO, VO₂, V₂O₃, and V₂O₅ for comparision.^30,31 The oxygen K-edge (O-K) in VO₂ has the t₂g component at 532 eV and the e_g component of the hybridized orbitals which appears as a shoulder at 535 eV.³² Between these spectra, the VO₂ and V₂O₅ spectra are the most similar to each other; however, there still exist differences. It is reported that the intensity of the first unoccupied state decreases as the number of vanadium 3d electron states increases by decreasing valance state.³³ That is, the decrease in the intensity of the 532 peaks is recognized as the decrease of the transition probability of the O 1s electron to the first unoccupied hybridization state between O 2p and V 3d. That explains the higher intensity of the t₂g component in V₂O₅ in comparison to VO₂ owing lower valance state. Even though the upper hybridization states are not very sensitive to the local symmetry, the spectral feature at 544 eV is different in the case of VO₂ and V₂O₅ with VO₂ owning higher intensity. These distinct differences confirm the formation of VO₂ in the pristine and vacuum annealed samples.

The oxygen pre-peak also provides information regarding oxygen vacancy concentration, which is following the trend of more oxygen vacancies leading to lower the valence states of vanadium ions.³⁴ If the valance state of vanadium decreases, the intensity of the O-K pre-peak decreases, and it merges with the O-K main edge since the hybridization between V and O decreases. Thus, there is a shift toward higher eV. The shift of peak around 532 eV to higher eV for oxygen pre-peak in VA-high T represents the presence of a higher concentration of oxygen vacancies in this sample compared to VA-low T as shown by electrical measurements as well. The vacancy concentration is calculated using the EELS technique to be x ∼ 0.20 ± 0.02 in VO_2−x.²³ The chemical shift of the L-edge of 3d transition metal oxides is another indicator of changes in the oxidation state of metal atoms in the oxides. The shift of V 2p_3/2 to lower eV has been reported as a decrease in the valence state which is not observed in the present work and further confirms the absence of V₂O₃ and V₂O₅ formation.^30–32 

The high-temperature vacuum annealed sample with metallic behavior is exposed to the oxygen annealing for the oxygen to get back into its lattice site. It is observed that after oxygen annealing at the same temperature, the sample returns to the M₁ state and shows typical metal–insulator transition of the pristine sample. This behavior is illustrated in Fig. 5. The vacuum and oxygen annealing processes are repeated consecutively for four full cycles (eight annealings). No sign of degradation is observed in the hysteresis or metallic behavior as the heterostructures go through multiple cycles of annealing. The transition temperature for the hysteresis behavior, however, is observed to change to lower temperature and eventually comes back to the known transition temperature for the relaxed bulk VO₂. The yellow and blue lines in Fig. 5 represent the transition temperature for the VO₂ in the first and fourth cycles which is ∼342 K and ∼341 K, respectively. The transition temperatures are reported as the mean between cooling and heating transition temperatures. In the second cycle, the VO₂ transition temperature has decreased to ∼336 K as the blue arrow is pointing toward it, and in the third cycle, the VO₂ transition temperature has increased back toward the equilibrium temperature and equals ∼339 K. In the fourth cycle, the transition temperature is identical to the bulk strain-free structure of VO₂ (341 K), which shows its structure has reached to equilibrium.

In the pristine sample, even though the transition temperature is so close to bulk behavior, there exist defects and misfit dislocations. Upon annealing at high temperatures, the system attempts to get rid of defects employing the diffusion process. We have, also, found out another relaxation process which has taken place at the interface of VO₂/NiO and requires higher energy than what RT provides. The system is demanding to minimize the stress field around misfit dislocations by the formation of stacking faults at the interface as it is captured and illustrated in the HAADF image in Fig. 6. The stacking fault involves two atomic layers, and the presence of steps has provided a nucleation site for stacking fault's formation. This shows that high-temperature annealing provides enough energy to overcome the formation energy of the stacking fault and relax the system further to achieve the bulk transition temperature in the fourth cycle. After the first annealing cycle, the presence of stacking faults in addition to misfit dislocations create stress fields which could cause overall compressive strain. This would lead to a decrease in the transition temperature in VO₂.³⁵ But eventually, after multiple cycles, the dislocation movement using the diffusion process returns the system back to the relaxed state to minimize the total energy.

We successfully controlled the transition in VO₂ thin films by introducing oxygen vacancies. The oxygen vacancies at low temperature annealing derive the M₂ phase of VO₂ which undergoes metal-to-insulator transition with dominant Mott nature. The introduction of oxygen vacancies at high temperatures and concentrations is shown to create metallic behavior in M₂–VO₂. This behavior is reversible through oxygen annealing. After oxygen annealing, the structure goes back to the M₁ monoclinic phase of VO₂. The high-temperature film relaxation mechanisms are also activated throughout these annealings such as the formation of stacking faults at the interface which affects the transition temperature. The structure eventually reached the relaxed state after three cycles.

This on and off hysteresis behavior generated by oxygen vacancies can be exploited in nonvolatile memory devices for neuromorphic applications, where there is no need for an external energy source to retrieve information (metallic and insulting behaviors). Considering the incredibly low power (10–15 W) of the human brain, energy efficiency is critical for neuromorphic computing.

Thin-film heterostructures were deposited using a KrF excimer laser with a frequency of 5 Hz and an energy of 2–2.5 J/cm².^36–38 The NiO layer was grown at 700 °C and 1 × 10⁻⁴ Torr oxygen pressure, and the VO₂ film was deposited above the critical thickness at 550 °C and 1.2 × 10⁻² Torr oxygen pressure.^2,39 The vacuum annealing procedure was performed at 1 × 10⁻⁷ Torr in two different temperatures of 200 °C and 450 °C for 2 h. The reverse process of oxygen annealing was done at 1.2 × 10⁻² Torr in two different temperatures of 200 °C and 450 °C for 2 h.

A PANalytical Empyrean diffractometer using Cu-Kα radiation was used to collect in situ XRD measurements. The in situ structural changes during heating and cooling cycles were monitored using this technique in the 25–120 °C range. The step size of 2θ was set to 0.013° and the count time per step was set to 1 s.

The HAADF images were collected using an FEI Titan 80-300 probe aberration-corrected scanning transmission electron microscope (STEM) operated at 200 kV. Across thin-film interfaces, electron energy loss spectra (EELS) were also acquired at 200 keV with a 28 mrad collection angle and 19.6 mrad convergence angle. Spectra were collected using a 0.25 eV ch⁻¹ dispersion with a ∼35 pA probe current and 0.07 nm pixel size to perform an elemental analysis. A routine peak fitting was applied on V-L and O-K edges using Gaussian profiles for detailed analysis.⁴⁰ The noise reduction in the data is performed by fast Fourier filtering of the data, following the dark reference.

The physical property measurement system (PPMS) by Quantum Design was used to acquire electrical transport measurements over the temperature range of −173 to 110 °C during heating and cooling cycles. The resistivity data were collected using a step size of 0.1 at zero magnetic fields. The carrier concentration of the metallic phases are produced by Hall measurements, and for the insulating phase, it is calculated from R vs T measurements considering the mobility of the metallic state, assuming across the transition the mobility does not change considerably.⁴¹ 

The manuscript was written by A.M. EELS spectroscopy measurements were done by R.S. All authors have contributed to the final version of the manuscript.

This project was supported by the National Science Foundation (NSF) (Grant No. DMR-1304607). The authors acknowledge the use of the Analytical Instrumentation Facility (AIF) at the North Carolina State University, which is supported by the State of North Carolina and NSF. R.S. also acknowledges the support of faculty start-up funding at the Oklahoma State University.

The data that support the findings of this study are available within the article.