Aligning Oxygen Vacancies Oriently: Electric-Field Inducing Conductive Channels in TiO2 Film to boost Photocatalytic Conversion of CO2 into CO

Oxide semiconductors are widely used in the photocatalytic elds and introducing oxygen vacancies is an effective strategy to reduce the band gap, and consequently, improve their photocatalytic eciency. However, oxygen vacancies in bulk often act as the recombination centers of electron-hole pairs, which would accelerate the recombination of electron-hole pairs and reduce carrier migration rate. Therefore, for achieving excellent photocatalytic performance in oxide photocatalysts, taking good advantage of oxide vacancies is very crucial. In this paper, we propose a strategy of electric eld treatment and apply it in the TiO2 lm with oxygen vacancies to promote the photocatalytic eciency. After treated by an electric eld, conductive channels consisting of oxygen vacancies are formed in TiO2 lm, which makes the resistance greatly decreased by almost 6×103 times. In the photocatalytic CO2 reduction reaction, the yield of CO in the electric-eld-treated TiO2 lm can reach up to 1.729 mmol·gcat-1·h-1, which is one of the best performance among the reported TiO2-based catalysts. This work provides an effective and feasible way for enhancing photocatalytic activity through electric eld and this method is promising to be widely used in the eld of catalysis.


Introduction
With the destruction of the environment and the depletion of fossil fuels, the application of photocatalysis to produce clean energy and useful chemicals has attracted more and more attention 1-4 .
However, the photocatalytic performance is far from satisfying the practical application requirements due to the narrow light response, fast carrier recombination and low charge mobility. Thus, enormous efforts have been made to solve aforementioned problems and explore the photocatalysts with high performance. For example, by doping defects, the band gap of the semiconducting photocatalysts can be reduced, and as a result, the optical response range is broadened, which leads to many more photogenerated carriers to participate in the photocatalytic processes 5,6 ; As for the suppression of charge recombination, metal or non-metal loading and construction of heterojunction are designed to create the built-in electric elds, in which the electrons and holes are separated and transfer to spatially separated locations 7-10 ; In the case of improving the carrier mobility, it is reported that under the action of external elds, the carrier migration can be speeded up, and consequently, more carriers would participate in the catalytic reaction per unit time 11 .
Oxide semiconductors are the most common photocatalysts and introducing oxygen vacancy (OV) is a usually used method to improve their photocatalytic performance, which can be ascribed to the reduced band gap [12][13][14][15][16] . For instance, Mao et al. reported that the black TiO 2 with large amount of OVs could absorb visible and infrared light and broaden the light response range 17 . Huang et al. introduced that OVs would narrow the band gap and improve visible light photocatalytic performance of ZnO 18 . However, in various cases, the bulk OVs would serve as the recombination centers of carriers, which accelerate the recombination of electron-hole pairs and reduce carrier migration rate [19][20][21] . It seems that although the light response range in the oxide photocatalysts can be remarkably broadened by introducing the OVs, their photocatalytic performance needs to be further improved. Therefore, making good use of OVs to promote charge separation and improve carrier mobility simultaneously is a giant challenge for achieving excellent photocatalytic activities in the oxide photocatalysts.
It is reported that under the action of an external electric eld, OVs in some oxide semiconductors can be oriented along the electric eld direction to form conductive channels, and consequently, greatly reduce the system resistivity [22][23][24] . This phenomenon is called as resistance switching (RS) effect 25,26 . The composition of the RS device is a metal/oxide/metal sandwich structure, where the metals act as the electrodes. The mechanism of RS is schematically illustrated in Fig. 1. As shown in Fig. 1a, the original state of the sample is considered a high resistance state (HRS), in which some OVs are scattered in the oxide. Under the action of an electric eld, oxygen ions move to anode electrode and leave the OVs staying at their original location. As the electric eld rises, the number of OVs continues to increase, and the conductive laments (i.e. conductive channels) consisting of OVs are gradually formed along the direction of electric eld. Until the OVs align in the lm from bottom to the top, the device turns into a low resistance state (LRS), which is shown in Fig. 1b. As a result, conductive channels consisting of OVs are constructed in the LRS lm, which can accelerate the carrier migration, and consequently, facilitate the photocatalytic reaction.
TiO 2 has attracted extensive research interests in the eld of photocatalysis owing to several outstanding advantages such as high photocatalytic activity, good photostability, non-toxicity, earth abundance and low cost [27][28][29] . It can show broadened light response range by introducing a lot of OVs 12,13 , but faces the same problems mentioned above. As we know, TiO 2 is a well-studied RS material based on the mechanism of conductive laments 30,31 , which provides a good opportunity for us to study the effect of OVs conductive channels on the photocatalytic performance. In this paper, the TiO 2 lm with some OVs is prepared in a low oxygen pressure environment. After treated by an electric eld, the conductive laments consisting of OVs are formed due to the RS effect, which makes the resistance of TiO 2 lm decreased by almost 6´10 3 times. Thus the OVs conductive laments would provide many highways in bulk to speed up the migration of the photogenerated carriers, which is very meaningful for improving photocatalytic activity. As expected, the electric-eld-treated TiO 2 lm show excellent photocatalytic performance of CO 2 reduction, and the CO production rate reaches up to 1.729 mmol·g cat -1 ·h -1 .

Results And Discussion
Composition and structure of device TiO 2 lm is deposited onto a Pt/SiO 2 /Si substrate using pulsed laser deposition (PLD). Subsequently, Au circular electrodes are deposited using an ion sputtering instrument onto TiO 2 lm with a shadow mask.
The formation process of Au/TiO 2 /Pt/SiO 2 /Si device is shown in Fig. 2a, in which Au and Pt act as the top and bottom electrodes, respectively. It is worth mentioning that Au electrode not only serves as the electrode, but also plays a role of co-catalyst for promoting the reduction of CO 2 to CO 32 . Fig. 2b shows  Supplementary Fig. 1b, in which the weight percentage (wt %) of anatase to rutile is 94% : 6%.

Characteristic of RS effect
The evolution process of RS effect in TiO 2 lm is illustrated in Fig. 3a. The original state of TiO 2 lm is HRS. When the positive sweeping voltage is applied to the device, the current gradually increases and shows a sudden jump at 2.5 V, indicating the switching from HRS to LRS (Set process). This phenomenon corresponds to the formation of OVs conductive laments, which is shown in Fig. 1b. During this process, a compliance current of 10 mA is adopted to avoid permanent dielectric breakdown of the device. As the voltage polarity is changed and the negative sweeping voltage is applied, the device returns to HRS with the current dropping sharply at -1.5 V (Reset process). At this situation, the oxygen ions will move to the opposite direction and combine with the OV, leading to the rupture of the conductive laments. As a result, the device switches to HRS, which is shown in Fig. 1a. The resistance state of the Au/TiO 2 /Pt device can switch between LRS and HRS by changing the voltage polarity, displaying a bipolar RS behavior, which is consistent with the earlier reports of TiO 2 -based RS devices 33,34 . It is worth noting that the RS effect of the device is nonvolatile, i.e., TiO 2 can maintain LRS or HRS after the applied electric eld is switched off, which is very meaningful for the practical applications 35,36 . In order to study the durability of RS behavior, an endurance test on LRS and HRS of the device is performed. As shown in Fig. 3b, both LRS and HRS samples can remain stable without any obvious degradation even after 10 4 s, demonstrating that the device has good retention performance. More importantly, the on/off ratio (R HRS /R LRS ) is larger than 6´10 3 , which implies that TiO 2 in LRS possesses high conductivity and is promising to show excellent photocatalytic performance.

Structure and oxygen vacancy characterization of LRS and HRS samples
The valence state of elements in TiO 2 lm and the presence of OVs are analyzed by X-ray photoelectron spectroscopy (XPS). Fig. 4a shows the XPS spectrum of Ti 2p level for both LRS and HRS samples. The peaks locating at 458.8 and 464.6 eV correspond to the Ti 2p 3/2 and Ti 2p 1/2 bands, respectively, which are typical energy level peaks with Ti 4+ characteristics 37,38 . Additionally, the TiO 2 lm possesses shoulder peaks near 457.6 eV and 463.6 eV at the lower energy sides, which are assigned to Ti 3+ species 39 . As we know, the generation of Ti 3+ is usually associated with the formation of OVs 40 . As shown in Fig. 4b, the shoulder peak at 531.7 eV can be deemed as the defect site with low oxygen coordination (i.e. OV) 41 . Considering the difference of peak areas at 531.7 eV, the OV concentration of LRS sample is larger than that of HRS sample 41  The structural properties of TiO 2 lm are investigated by Raman spectroscopy. As shown in Fig. 4c, the peaks around 143, 195, 635, 399, and 515 cm -1 belong to E g , E g , E g , B 1g , and A 1g modes of anatase, respectively. Moreover, the band at 320 cm -1 is observed, which can be assigned to a two-phonon scattering band of anatase 42 . In addition to the peaks of anatase, the weak peaks at 245 and 430 cm -1 are ascribed to the two-phonon scattering and Eg modes of rutile, respectively 43 . Thus we can deduce that the main phase of TiO 2 lm is anatase, which is consistent with the result of XRD. It is reported that the existence of OVs can be con rmed by Raman shift either 13 . As shown in Fig. 4d, the E g peaks of HRS and LRS samples locate at 142.5 and 144.1 cm -1 , respectively. Therefore, an evident shift of Eg peak is presented in LRS sample, which can be attributed to the increased amount of OVs in LRS sample 44,45 .
Meanwhile, the Eg peak of LRS sample is wider than that of HRS sample, implying that the degree of lattice disorder is increased by more bulk OVs 46,47 .
In order to detect the distribution of the OVs, a two-dimensional Raman spectrogram around 143 cm -1 is investigated in the samples. As shown in Fig. 5, the two-dimensional Raman spectra exhibit different brightness. Here, the change of brightness represents the shift of Raman peak. According to the result of Fig. 4d, the Raman peak of LRS sample with more OVs has higher wavenumber. As a result, the darker regions represent the lower OV concentrations of HRS sample (Fig. 5a) and the brighter regions stand for the higher OV concentrations of LRS sample (Fig. 5b). Combined with the analysis of the RS mechanism, the OVs conductive laments would locate at the brighter regions in Fig. 5b.
The optical properties of HRS and LRS samples are investigated by UV-visible diffuse re ectance absorption spectroscopy in Supplementary Fig. 2a. Compared with the optical absorption spectra of TiO 2 lm without OVs 48 , the spectra of LRS and HRS samples exhibit long wavelength absorption, which extends to visible light and infrared light. In order to exclude the in uence of Au on the absorption peaks, the UV-visible absorption spectrum of TiO 2 lm without Au electrode is characterized in Supplementary   Fig. 2b. It is obvious that the absorption spectrum of pure TiO 2 lm is similar to that of HRS and LRS samples. This result implies that the effect of Au electrodes on the long wavelength absorption can be almost neglected. As a result, the long wavelength absorption of LRS and HRS samples are mainly attributed to the presence of OVs in the TiO 2 lms. Furthermore, the LRS sample demonstrates wider range of light absorption than that of HRS sample due to the more OVs 49,50 .
Photocatalytic performance of CO 2 reduction The photocatalytic performances of the samples are evaluated via CO 2 photoreduction under simulated sunlight irradiation. Prior to irradiation, ultra-pure CO 2 gas is owed into the reaction chamber and the samples are kept in the reaction atmosphere for several hours to ensure that the adsorption of gas molecules is completed. During the photoreduction process, CO is detected as the main product. Fig. 6a shows the time dependence of the production of CO for LRS and HRS samples. It is obvious that with the increase of time, the CO evolution of HRS sample increases almost linearly. As shown in Fig. 6b, the CO production rate of HRS sample is probably 1.007 mmol·g cat -1 ·h -1 . As for LRS sample, it exhibits similar catalytic behavior to HRS sample, but its photocatalytic performance is signi cantly improved. After the irradiation for 4 h, the CO production rate of LRS sample is as high as 1.729 mmol·g cat -1 ·h -1 , which is 172% higher than that of HRS sample. Therefore, the electric eld treatment is a very effective method for achieving excellent photocatalytic performance. As far as we know, the CO yield of LRS sample is one of the best among the reported TiO 2 -based catalysts and the detailed results are listed in Table 1.
Furthermore, LRS and HRS samples can demonstrate excellent stability toward CO 2 photoreduction. As shown in Fig. 6c, no signi cant deterioration is detected for the photocatalytic activity after four cycles of repeated tests.

Photoelectrochemical properties of LRS and HRS samples
The photoluminescence (PL) spectroscopy is used to evaluate the carrier separation/recombination performance 51 . As shown in Fig. 7a, the samples of HRS and LRS display strong emission peaks in the wavelength range of 550-825 nm. It is obvious that the peak intensity of LRS sample is relatively low, indicating that the carrier recombination of LRS sample is slower than that of HRS sample. The timeresolved photoluminescence spectroscopy is measured to reveal the lifetime of carriers. As shown in Fig.  7b, the average lifetime of carriers in HRS sample is 1.50 ns. However, after the treatment of electric eld, the average lifetime in LRS sample increase to 1.69 ns, which is bene cial to enhance photocatalytic performance. The linear voltammetry scan curve of HRS and LRS samples at the cathode is shown in Supplementary Fig. 3. It is observed that the photocurrent of LRS sample is stronger than that of HRS sample, indicating that the electrons of LRS sample have a faster transfer speed and a shorter migration time to the surface owing to the lower resistivity. Fig. 7c shows the transient chronoamperometry of the samples, in which all the samples display constant on-off responses. Obviously, the photocurrent density of LRS sample is 2.0 times higher than that of HRS sample, suggesting that LRS sample has higher electron-hole separation e ciency. The Nyquist plots of the electrochemical impedance spectroscopy are carried out at the open circuit potential, which is shown in Fig. 7d. Compared with the HRS sample, LRS sample shows the smaller hemicycles in the Nyquist plot, suggesting the lower charge transfer resistance and the higher charge mobility 52 .
To gain further insight into the charge-transfer process and the impact of electric eld on the electronic properties of TiO 2 , the electrochemical impedance measurements of LRS and HRS samples are performed at 1 kHz. Supplementary Fig. 4 shows the Mott-Schottky plots of samples, in which the positive slopes can be observed as expected for n-type semiconductors 37 . Different from HRS sample, a substantially shallow slope for LRS sample with higher OV concentration can be obtained, suggesting the increase of carrier density. Here carrier density can be calculated from the slope of the Mott-Schottky plots using the equation (1) 53,54 .
where e 0 is the electronic charge, ε is the dielectric constant of TiO 2 (ε anatase = 55, ε rutile = 170) 55,56 , ε 0 is the dielectric constant of vacuum, N d is the carrier density, and V is the external bias voltage on the electrode. The calculated carrier density of HRS and LRS samples are 5.7´10 18 cm -3 and 4.0´10 19 cm -3 , respectively. Here the carrier density of LRS sample is 7 times higher than that of HRS sample, which is bene cial to improve the electron-hole pair separation and transport e ciency 54 . Based on above analysis, LRS sample possesses quite excellent charge separation and transfer e ciency, which can signi cantly improve the activity of CO 2 reduction.

Photocatalytic mechanism of CO 2 reduction in LRS system
To understand the in uence of OVs on the band gap of the non-conductive lament region (NCFR) and the conductive lament region (CFR), the electronic density of states and band structure of anatase (because 94% of the content in TiO 2 lm has this structure) with OVs are calculated by rst-principles calculations based on density functional theory. Based on the data of OV concentration from the XPS results, the structure models, density of states and band structure of TiO 2 with different OV concentrations are shown in Fig. 8a, Fig. 8b and Supplementary Fig. 5 , the values of band gap for NCFR and CFR are 1.62 eV and 1.5 eV, respectively, suggesting that the light absorption range of CFR is larger. Furthermore, the potentials of conduction band (CB) minimum and valence band (VB) maximum in CFR are lower than those in NCFR. Therefore, a homojunction can be formed between NCFR and CFR in LRS sample, which is shown in Fig. 8c. In this homojunction, the electrons in the CB of NCFR would migrate to the CB of CFR and the holes in the VB of CFR transfer to the VB of NCFR, which leads to the e cient spatial charge separation. Different from the earlier reports 7,57 , the junctions in LRS sample are constructed not only at the surface but also in bulk, which is conducive to the bulk charge separation.
Based on the experimental and theoretical results mentioned above, we can investigate the origin of excellent photocatalytic performance in LRS sample. Fig. 9 shows the schematic illustration of photocatalytic process of CO 2 reduction. Due to the contribution of large amount of OVs in LRS TiO 2 sample, the band gap is reduced, which broadens the light absorption range. As a result, many photogenerated carriers can participate in the photocatalytic reaction, which is one of the reasons for excellent photocatalytic performance. Secondly, after the treatment of electric eld, the OVs align along the electric eld direction and form some conductive laments. As mentioned above, the conductive lament consisting of OVs reduces the band gap and a homojunction can be constructed between CFR and NCFR in LRS sample, which is conducive to promote the separation of electrons and holes. Finally and most importantly, the photogenerated carriers can migrate rapidly to the surface along the conductive laments to participate in the reduction reaction. Under the synergistic effect of above three favorable conditions, an excellent photocatalytic performance of CO 2 reduction is realized in LRS sample.

Summary
TiO 2 lm with OVs is prepared by PLD to investigate the photocatalytic reduction of CO 2 . After the treatment of electric eld, TiO 2 lm can switch from HRS to LRS due to the formation of conductive laments consisting of OVs. The existence of conductive laments brings about three advantages for improving photocatalytic properties of LRS sample. First, the light absorption range is broadened which leads to many more photogenerated carriers to participate in photocatalytic reaction. Second, homojunctions are constructed between CFRs and NCFRs, which promotes space charge separation. And the last is a distinctive advantage, conductive channels based on OVs are built, which accelerates the migration of carriers to the lm surface. This work provides a novel approach for enhancing photocatalytic activity with the treatment of electric eld. Moreover, this method of electric-eld-induced high conductivity in semiconductive/insulative catalysts is promising to be widely used in the catalytic elds.

Methods
Preparation of Au/TiO 2 /Pt device. TiO 2 lm was deposited on Pt/SiO 2 /Si substrate using PLD with a 248 nm KrF excimer laser at 3Hz. The target material has a stoichiometric ratio and the base vacuum of the chamber was below 1.0´10 −5 Pa. During the deposition of the lm, the substrate temperature was maintained at 600 °C, whereas the oxygen pressure was 20 Pa. After deposition, the lm was annealed in situ at 600 °C for 30 min. Circular Au top electrodes with diameter of 100 μm were deposited on TiO 2 lm with a shadow mask using an ion sputtering instrument.
Material characteristics. The crystal structure of the sample was characterized by X-ray diffraction spectrometer (Cu kα, td-3500, Tongda). The thickness of the lm was measured by a eld emission scanning electron microscopy (GeminiSEM 500, ZEISS, Germany). X-ray photoelectron spectroscopy measurement was performed on a Phi5000 VersaProbe (ULVAC-PHI, Japan) using 200 W monochromated Al Ka radiation as the X-ray source. The standard C1s peak was used as a reference for correcting the shift. Raman and PL spectra/mapping were carried out by confocal Raman and photoluminescence techniques with a 532 nm laser as the light source (Alpha 300R, WITec, Germany). The PL lifetime measurements were performed using a PL spectrometer (FLS 920, Edinburgh, UK) at an excitation wavelength of 375 nm. The UV-visible diffuse re ectance spectra (UV-vis DRS) of samples were recorded on an UV-vis-NIR spectrophotometer (TU-1900, Puxi). The electrical properties of the samples were measured with Keithley2410 and a dual probe con guration was performed.
Photocatalytic CO 2 reduction reaction. Photocatalytic reactions were performed in a reactor with a quartz window on the top. The device with an area of 0.5 cm 2 was placed on a Te on catalyst holder without being immersed in water. The volume of reaction system was about 440 ml. Before the irradiation, ultrapure CO 2 gas was continuously owed into the CO 2 -reduction chamber for half an hour to ensure that no impurity gas exists in the reaction system. Subsequently, 0.4 mL of deionized water was injected into the bottom of the reactor. The samples were kept in the reaction atmosphere without irradiation for several hours to ensure that the adsorption of gas molecules was complete. During the photoreaction, the chamber was irradiated with a 300 W Xenon (Xe) lamp. The photocatalytic reaction was carried out for 4 h. The gaseous products in chamber were collected each hour by a sampling syringe (1 mL) and then syringed into the gas chromatography (GC-2014, Shimadzu Corp., Japan) for analysis.
Photoelectrochemical measurements. Photoelectrochemical measurements were carried out with an electrochemical workstation (CHI760D) using a standard three-electrode electrochemical cell.  [58][59][60] . The structures were relaxed until forces were converged within 0.01 eV Å -1 , and the convergence criterion for total energy was 10 -4 eV. The Brillouin zone was sampled with a 5×5×2 k-point grid mesh for the unit cell 61 . The plane wave with a kinetic energy cutoff was 600 eV for the structure optimization.

Data availability
The data that support the ndings of this study are available from the corresponding author upon request.

Competing interests
The authors declare no competing interests. Table   Table 1. Comparison of the CO yield rate of this work with that of other reported TiO 2 -based photocatalysts.

Figure 1
Schematic diagram of resistance switching (RS) mechanism. a high resistance state (HRS) and b low resistance state (LRS).    The distribution of oxygen vacancies. Two-dimensional Raman spectrum at 143 cm-1 of a HRS sample and b LRS sample after etching. The brightness represents the wavenumber. Photocatalytic performance of CO2 reduction. a Time course evolution of CO for 4h with 300 W Xe lamp and b CO yield rate of LRS and HRS samples. c Cycle experiment of photocatalytic CO2 reduction. gcat: The mass of TiO2 lm catalyst.