The effects of oxygen in spinel oxide Li1+xTi2−xO4−δ thin films

The evolution from superconducting LiTi2O4-δ to insulating Li4Ti5O12 thin films has been studied by precisely tuning the oxygen pressure in the sample fabrication process. In superconducting LiTi2O4-δ films, with the increase of oxygen pressure, the oxygen vacancies are filled gradually and the c-axis lattice constant decreases. When the oxygen pressure increases to a certain critical value, the c-axis lattice constant becomes stable, which implies that the sample has been completely converted to Li4Ti5O12 phase. The two processes can be manifested by the angular bright-field images of the scanning transmission electron microscopy techniques. The transition temperature (Tch) of magnetoresistance from the positive to the negative shows a nonmonotonic behavior, i.e. first decrease and then increase, with the increase of oxygen pressure. We suggest that the decrease Tch can be attributed to the suppressing of orbital-related state, and the inhomogeneous phase separated regions contribute positive MR and thereby lead to the reverse relation between Tch and oxygen pressure.

In the research on oxide superconductors, the oxygen always plays an important role in the superconductivity and their normal state behaviors 1,2 . In copper oxide high-critical temperature (T c ) superconductors, such as Nd 2-x Ce x CuO 4±δ 3-5 , Pr 2−x Ce x CuO 4±δ 6-8 and YBa 2 Cu 3 O 7−δ 9-11 , T c can be greatly improved in a large range by adjusting the oxygen content during the annealing process, as well as the titanium oxide systems, such as SrTiO 3 12,13 and TiO 14,15 . Oxygen has a strong effect not only on superconductivity, but also on many other properties. For instance, the antiferromagnetism 16 and the charge density wave 17 can also be tuned by the oxygen vacancies. Furthermore, the doping and disorder effects induced by oxygen vacancies can cause obviously change on Hall resistance and magnetoresistance (MR) behaviors in the normal state 5,6,18 . Studying the oxygen effects is of great help to understand the mechanism of superconductivity, transport and other properties of the oxide superconductor 9,19,20 .
Among hundreds of spinel oxides, the metallic lithium titanate LiTi 2 O 4 is the only known oxide superconductor, which T c is as high as 13.7 K 21 , discovered by Johnston et al. in 1973 22 . Previous studies have disclosed that LiTi 2 O 4 is a BCS s-wave superconductor with intermediate electron-phonon coupling (λ el-ph ~ 0.65) 23,24 . Nevertheless, an enhanced density of states has been unveiled by magnetic susceptibility 25 and specific heat measurements 23 , indicating that d-d electronic correlations cannot be ignored in this system. Meanwhile, due to the mixed-valence of Ti ions in the frustrated Ti sublattice, LiTi 2 O 4 exhibits complicated spin-orbit fluctuations, which is evidenced by the resonant inelastic soft-x-ray scattering 26 , nuclear magnetic resonance 27 and magnetic susceptibility measurements 25 . Very recently, electrical transport and tunneling spectra measurements on high quality epitaxial [001]-oriented LiTi 2 O 4 films have revealed an orbital-related state below ~50 K, confirmed by a twofold in-plane angular dependent MR, positive MR as well as the relation Δ ~−B 2 28 .
Interestingly, by tuning the oxygen in the process of sample deposition, the phase of the thin film changes from LiTi 2 O 4-δ to Li 4 Ti 5 O 12 along with the superconductor-insulator phase transition 29 . However, this transition seems to happen abruptly, which hinders us from understanding the nature of the transition. Previous work on LiTi 2 O 4 polycrystals has disclosed that the existence of oxygen-site distortion induces prominent changes in the electronic states near E F 30 . In addition, tunneling spectra experiments on LiTi 2 O 4 films of different orientations reveal an anisotropic electron-phonon coupling in this system, which is regarded to originate from the Jahn-Teller distortions enhanced by oxygen vacancies 31 . Nevertheless, it is still unclear what happens in the microstructure of the sample during the transition from LiTi 2 O 4 to Li 4 Ti 5 O 12 . Moreover, the mechanism of the oxygen effects on superconductivity of

Results and Discussion
The θ-2θ XRD spectra of (001) Li 1+x Ti 2-x O 4-δ (0 ≤ × ≤ 1/3) samples grown in different P O2 are shown in Fig. 1(a). The (001)-oriented LiTi 2 O 4-δ thin films are achieved when the films are deposited under P O2 ≤ 10 −6 Torr. Instead, the (001)-oriented Li 4 Ti 5 O 12 thin films are formed at P O2 > 10 −4 Torr. The XRD patterns of the samples in different P O2 are quite similar except that the diffraction peaks gradually shift to higher angle in the LiTi 2 O 4-δ films at larger P O2 . In order to check the crystallization quality of the thin films, we also perform ϕ-scan. In Fig. 1(b), the ϕ-scans of (404) plane of both LiTi 2 O 4-δ and Li 4 Ti 5 O 12 samples display four-fold symmetry with uniformly distributed peaks. From the θ-2θ XRD spectra, we can extract the value of the out-of-plane lattice constant (c-axis) as a function of the P O2 . As seen in Fig. 1(c), when P O2 < 10 −4 Torr, c gradually decreases with increasing P O2 . However, when P O2 is higher than 10 −4 Torr, c is saturated, indicating the complete formation of Li 4 Ti 5 O 12 phase. As a result, a phase transition from LiTi 2 O 4-δ to Li 4 Ti 5 O 12 has been successfully achieved by tuning P O2 during the sample deposition. In order to further study the effects of oxygen on superconducting state and normal state, we systematically measured the resistances of various thin films from LiTi 2 O 4-δ to Li 4 Ti 5 O 12 . The R-T curves of the LiTi 2 O 4-δ thin films with different oxygen pressures are shown in Fig. 2(a). Increasing the P O2 during the deposition, the samples undergo a transition from metal to insulator in the normal state. In Fig. 2(b), the residual resistivity ratio (RRR) decreases monotonically with increasing P O2 . Here, the RRR is defined as the ratio of room temperature resistivity to the resistivity of T c°n set , i.e. R (300 K)/R (T c°n set ), where the T c onset is the critical temperature at the beginning of superconducting transition. We plot the dependence of T c0 on P O2 as seen in Fig. 2(b), and the T c0 of the LiTi 2 O 4-δ thin films is quite stable at P O2 < 5.4 × 10 −6 Torr, whereas it drops rapidly when P O2 > 5.4 × 10 −6 Torr.
To find out the microstructure evolution from LiTi 2 O 4-δ to Li 4 Ti 5 O 12 , we have carried out atomic-resolution STEM measurements on these high-quality samples. Figure 3 shows the ABF images along the [110] direction and the corresponding line profiles in different regions with different types of oxygen vacancies. In Fig. 3(a), the O columns, as indicated by red arrows, are imaged as dark spots due to the absorption nature of the ABF contrast, and the contrast of the Ti columns as indicated by the blue arrows is darker than the O columns based on the ~Z 1/3 contrast mechanism where Z is the atomic number. Thus, in the pristine regions the contrast of O 1 and O 2 is of  approximate equal darkness, and oxygen vacancies as shown in Fig. 3(b) and (c), are imaged as light gray spots. Here, we divided the positions of oxygen atoms into two types, i.e. O 1 and O 2 (see Fig. 3(d)) to describe clearly the distribution of oxygen vacancies, as shown by the red and blue arrows. To visualize the oxygen vacancies clearly, we extracted the line profiles on the oxygen rows as indicated by the yellow and red rectangles.
From the line profile of ABF contrast in Fig. 3(e), we can find that the depth of the ABF contrast valley (darkness) at O 2 positions is lower than that at O 1 as seen in the [110] direction, which means some vacancies exist in the O 2 sites. Similarly, some vacancies at O 1 and O 2 exist in another region as shown in Fig. 3(f). However, these oxygen vacancies have not been observed in the Li 4 Ti 5 O 12 samples 32 . It is known that the LiTi 2 O 4-δ exhibits serious aging effects in forms of polycrystal and single crystal 30 . The LiTi 2 O 4-δ thin films, especially the one deposited in the higher vacuum, are much more stable. It is reasonable to speculate that the samples in higher vacuum will contain more oxygen vacancies. Increasing oxygen pressure will fill these oxygen vacancies and finally turn the superconducting phase to insulating Li 4 Ti 5 O 12 .
The phase evolution with P O2 should inevitably make difference in the electronic states. In LiTi 2 O 4-δ , one concern is about the orbital-related state. Normally, the formation of the orbital order results from the band split near the Fermi level. As for LiTi 2 O 4-δ , the distortion of Ti-O octahedron leads the splitting of Ti 3d band to e g and t 2g band 33 , and the orbital-related state is expected to exist. Although it has been unveiled in previous work, it remains unclear in the origin 25 . One of the evidence is the crossover from the negative MR (n-MR) to the p-MR at T ch ~ 50 K in the normal state. Entering the superconducting state, the orbital-related state interacts with Cooper pairs and results in an unexpected relation between the superconducting gap and the applied magnetic field, i.e. Δ~ −B 2 . This relation implies the coexistence of the superconducting state and the orbital-related state. Therefore, it is deserved to clarify how the oxygen makes the influence on these two states.
To clarify this issue, we finely tune the P O2 around 10 −6 Torr to avoid the Li 4 Ti 5 O 12 phase. Then, we focus on the effects of P O2 on R and MR. In the precise tuning process, the vacuum value is not a good scale due to the limitation of the vacuum gauge. Fortunately, the RRR decreases monotonically with the increase of oxygen pressure, which can reflect the trend of P O2 and the oxygen defects as discussed above. Thus, we use RRR to index the samples, named S1 to S8 with RRR in the range between 5.6 and 1.5. As shown in Fig. 4(a), the T c seems unchanged in the tuning range. For samples S1 to S5, the MR at 35 K changes from positive to negative as seen in Fig. 4(b). By fitting the MR with the Kohler's formula, i.e. MR ~A 0 B 2 2 , the slope A 0 can be obtained for these samples at different temperatures. In Drude model, A 0 is proportional to μ 2 (i.e. μ = eτ/m) with μ the mobility, τ the relaxation time and m the electron mass. With the increase of temperature, the value of A 0 decreases from positive to negative as seen in Fig. 4(c). A negative mobility cannot be understood in this simplified model, and the n-MR is interpreted as the suppression of spin-orbital fluctuations in this system 28 . The T ch from p-MR to n-MR is extracted from Fig. 4(c) and plotted in Fig. 4(d). From S5 to S1, T ch gradually increases with the increase of RRR.
However, further reducing the RRR, the relation between T ch and RRR will be broken. For instance, the MR becomes stronger for the samples deposited under higher P O2 , e.g. S6-S8, and thus the T ch goes up. In this regime, the formation of Li 4 Ti 5 O 12 phase may lead to more boundaries in phase separated samples. Such inhomogeneity in the magnetic field usually exhibits strong p-MR 34,35 . For the samples S1-S5, the p-MR below T ch mainly origins from the orbital-related state since the LiTi 2 O 4-δ phase dominates the transport 25 . We speculate that filling the oxygen vacancies seems to suppress the p-MR but in fact the orbital-related state.
In order to verify this assumption, we should evaluate the effects of oxygen vacancies on Ti valance. Although oxygen vacancies have been detected by STEM, the content of oxygen vacancies cannot be quantified. Therefore, we collected EELS profiles of both LiTi 2 O 4-δ and Li 4 Ti 5 O 12 films. As seen in Fig. 5, the Ti L 2,3 edges, from 2p 1/2 and 2p 3/2 to 3d orbits respectively, split into two peaks in Li 4 Ti 5 O 12 , but not in LiTi 2 O 4-δ . Usually, the splitting of L 2,3 is attributed to the degeneracy lifting of Ti 3d orbits by the crystal field.
The missing split of peaks in LiTi 2 O 4-δ EELS may origin from two reasons. First, the energy gap between e g and t 2g , named Δ e-t , is too small to be discernable in EELS. According to the band calculations, Δ e-t equals to 2.1 eV and 2.4 eV for the ideal structure LiTi 2 O 4 and Li 4 Ti 5 O 12 , respectively 36 . Moreover, four peaks were also observed in Li 7 Ti 5 O 12 , where Δ e-t equals to 1.8 eV 32 . Considering the existence of oxygen vacancies, which may further distort the Ti-O octahedrons, we do not expect a smaller crystal field. Therefore, the change of Δ e-t cannot account for the discernable peak splitting in LiTi 2 O 4-δ . Second, the valance of Ti in ideal LiTi 2 O 4 is +3.5. If large numbers of oxygen vacancies exist in LiTi 2 O 4-δ , the Ti 3.5+ will transform to Ti 3+ . In this condition, the electrons on t 2g band increase, and thereby the hoping possibility from Ti 2p to t 2g is reduced due to the Pauli Exclusion Principle. Consequently, oxygen vacancies will smear out the peaks of Ti 2p to t 2g in EELS.
Based on the EELS results, we can give a reasonable explanation for the suppression of the orbital-related state by filling oxygen vacancies. In general, the formation of the orbital order results from the band split near the Fermi level. As for LiTi 2 O 4 , crystal field splits Ti 3d bands to e g and t 2g bands 33 . The oxygen vacancies in LiTi 2 O 4-δ system, on the one hand, enhance the distortion of Ti-O octahedrons, on the other hand, dope electrons to enhance the electron correlations, which are beneficial to the formation of the orbital-related state. With the filling of oxygen vacancies, the valence of Ti increases and some of Ti sites become empty states, thereby weaken the orbital-related state. Compared to the obviously suppressed orbital-related state, the T c0 of the LiTi 2 O 4-δ thin films is quite stable at P O2 < 5.4 × 10 −6 Torr. Actually, the O 2p bands are far below the Fermi level with weak p-d hybridizations 30,33 . Although the oxygen vacancies induce doping effect and influence on the splitting of Ti 3d bands by the crystal field, the density of states near Fermi surface may not change obviously, and thus the T c0 remains the same.
In conclusion, we studied the evolution from LiTi 2 O 4-δ to Li 4 Ti 5 O 12 with increasing oxygen pressure during the thin film deposition. By transport and STEM measurements, we have disclosed that there are two processes happened during the evolution, i.e. the filling of oxygen vacancies and the forming of Li 4 Ti 5 O 12 . The EELS results of the LiTi 2 O 4-δ and Li 4 Ti 5 O 12 samples provide the evidence that the orbital-related state is suppressed by the filling of oxygen vacancies. The evolution of electronic states by adjusting the oxygen content gives an insight into the interaction between the orbital-related state and the superconductivity in LiTi 2 O 4-δ .

Methods
The (00l)-oriented Li 1+x Ti 2-x O 4-δ (0 ≤ × ≤ 1/3) thin films are grown on (00 l) MgAl 2 O 4 (MAO) substrates by PLD with a K r F excimer laser (λ = 248 nm). Before the deposition, the MAO substrates are annealed at 1000 °C for 5 hours in the air 37,38 to obtain the smooth surface. The sintered Li 4 Ti 5 O 12 ceramic target is used to fabricate the films, with pulse frequency of 4 Hz, energy density of 1.5 J/cm 2 , and deposition temperature of ~700 °C. The deposition rate is determined by measuring the thickness of ultra-thin films using X-ray reflectivity analysis. In this study, we fix the film thickness ~150 nm. After the deposition, all the thin films are quenched to the room temperature in situ.
X-ray diffraction (XRD) is employed to characterize the phase and crystalline quality of Li 1+x Ti 2-x O 4-δ (0 ≤ × ≤ 1/3) thin films. The microstructure is detected by the spherical aberration-corrected scanning transmission electron microscopy techniques (Cs-STEM). The transport properties are measured by the Quantum Design Physical Property Measurement System (PPMS) with the temperature down to 2 K and magnetic field up to 9 T. Samples are etched into Hall bar by the UV lithography and Ar plasma etching technology for the measurement of the resistance properties.