(Sub)surface mobility of oxygen vacancies at the TiO 2 anatase (101) surface
Philipp Scheiber, Martin Fidler, Olga Dulub, Michael Schmid, Ulrike Diebold, Weiyi Hou, Ulrich Aschauer, Annabella Selloni
((Sub)surface mobility of oxygen vacancies at the TiO anatase (101) surface Philipp Scheiber, Martin Fidler, Olga Dulub, Michael Schmid, UlrikeDiebold,
1, 2, ∗ Weiyi Hou, Ulrich Aschauer, and Annabella Selloni Institute of Applied Physics, Vienna University of Technology,Wiedner Hauptstrasse 8-10/134, 1040 Vienna, Austria Department of Physics, Tulane University, New Orleans, LA 70118, USA Department of Chemistry, Princeton University, Frick Laboratory, Princeton NJ 08544, USA (Dated: April 13, 2018)Anatase is a metastable polymorph of TiO . In contrast to the more widely-studied TiO rutile,O vacancies (V O ’s) are not stable at the anatase (101) surface. Low-temperature STM showsthat surface V O ’s, created by electron bombardment at 105 K, start migrating to subsurface sitesat temperatures ≥
200 K. After an initial decrease of the V O density, a temperature-dependentdynamic equilibrium is established where V O ’s move to subsurface sites and back again, as seen intime-lapse STM images. We estimate that activation energies for subsurface migration lie between0.6 and 1.2 eV; in comparison, DFT calculations predict a barrier of ca. 0.75 eV. The wide scatterof the experimental values might be attributed to inhomogeneously-distributed subsurface defectsin the reduced sample. PACS numbers: 68.37.Ef, 68.47.Gh, 61.72.Cc, 68.35.Dv
Titanium dioxide, TiO , is one of the most versatileoxide materials and finds wide use, e.g. , in energy-relatedapplications such as (photo-)catalysis and solar energyconversion schemes. TiO has also evolved as a popularmodel system for studying the fundamentals of defect-related surface processes at the molecular scale [1, 2].TiO crystallizes in three different structures com-monly named rutile ( D h – P /mnm ), anatase ( D h – I /amd ), and brookite ( D h – P bca ). TiO nanomate-rials can be synthesized with various shapes and func-tionalities using sol-gel and other processing techniques[3]. Although the anatase polymorph is metastable, it iscommonly found in nanomaterials where the crystal sizeis below a few tens of nm. Yet few experimental stud-ies on large single crystals exist [4–8], thus the surfacesof anatase are not as well understood as those of rutile,where processes related to intrinsic defects – Ti intersti-tials (Ti int ) and surface O vacancies (V O ) – have receivedconsiderable attention [9–12].Recently we have found a significant difference betweenthe surfaces of rutile and anatase: at anatase (101), themost stable surface of this polymorph, it is energeticallymore favorable for O vacancies to reside in the bulk thanon the surface [13]. This is in stark contrast to rutile(110), where surface V O ’s form easily under standardpreparation conditions [1]. The preponderance of bulkdefects in anatase was first predicted by DFT calcula-tions, which showed that the formation energy of a sur-face V O is larger than that of a bulk vacancy by about ∼ O ’s are visible at thesurface [16].The observation that surface V O ’s are less stable thanbulk V O ’s is remarkable. An O atom can leave a solidonly through its surface, thus an as-formed surface V O should diffuse into the bulk. The activation energy(E act ) for surface-to-subsurface migration is ∼ O ’s non-thermally by electron bombardment [17], andmonitor their fate with low-temperature and variable-temperature STM. We find that surface V O ’s diffuse tosubsurface sites at temperatures above 200 K. Time-lapseSTM images show a temperature-dependent, dynamicequilibrium concentration of surface defects. The resultspoint towards an activation energy for subsurface migra-tion of a V O that depends on its immediate surroundings.The experiments were carried out in a two-chamberUHV system with a base pressure of 10 − mbar. Unlessnoted otherwise, constant current STM measurementswere performed at 78 K. For STM we typically used pos-itive sample bias voltages between 1.3 and 1.5 V, andtunneling currents between 0.1 and 0.4 nA for STM. Amineral anatase (101) sample was cleaved ex-situ as de-scribed in reference [18]. A clean, almost pristine sur-face was repeatedly prepared by sputtering (1 keV Ar + ,fluence of 6.7 × ions/cm ), annealing in O (p= 5 × − mbar) at 923 K for 30 minutes, and post-annealing in UHV at 973 K for another 10 minutes, seeFig.1(a). To create V O ’s, the surface was irradiated with a r X i v : . [ c ond - m a t . m t r l - s c i ] A p r FIG. 1. STM images (T sample = 78 K) of TiO anatase (101).(a) Freshly-prepared surface. (b) After irradtion with 500 eVelectrons, which creates surface O vacancies (V O ’s). The in-sets (b1, b2) show a magnified experimental and calculatedSTM image of a V O , respectively. After annealing the samplefor 10 minutes to (c) 326 K and (d) 450 K a rastered and thoroughly-outgassed electron gun at acurrent density of 1 µ A mm − (current measured with apositive sample bias of 27 V). Electron bombardment wasperformed in the preparation chamber with the samplekept at 105 K. As is shown below, V O ’s are immobileat this temperature. After irradiation the sample wastransferred into the STM for analysis. To determine thestability of the electron-induced surface defects (Fig. 2),we proceeded as follows: The manipulator in the prepa-ration chamber was resistively heated and equilibrated atthe desired temperature. With a pre-cooled wobblestickthe sample was taken from the cold STM and insertedinto the manipulator, where it was kept for 10 minutes.Then the sample was transferred back into the cold STM.The minimum time between taking the sample from themanipulator and the first usable STM image was also 10minutes. It is important to note (see below) that theinitial V O density was kept constant throughout theseexperiments.The DFT calculations were performed using thePerdew-Burke-Ernzerhof (PBE) [19] functional and theplane wave pseudopotential scheme as implemented inthe Quantum ESPRESSO package [20]. In addition, se-lected spin polarized hybrid PBE0 calculations [21] wereperformed using a mixed localized + plane wave basisset expansion of the electronic states as implemented in CP2KQuickstep [22]. The defected surface was modeledusing 3 × × ) supercells with periodicallyrepeated slabs of three (9.7 ˚A) or four (13.1 ˚A) TiO lay-ers separated by a vacuum of about 10 ˚A. For STM cal-culations, larger 4 × × ) supercells wereused to separate the periodic images. Activation energybarriers were estimated using the Nudged Elastic Band(NEB) [23, 24] method. Other computational details aregiven in the Supplemental Material.The sputtered and annealed anatase (101) surface ischaracterized by trapezoidal islands; their orientationindicates the crystallographic directions of the crystal[25]. Atomically-resolved STM shows rows of oval-shapedspots that extend over both, the Ti and O surfaceatoms [5], oriented along the [010] direction, (see Fig.1(a)). Our sample preparation procedure renders a bulk-reduced sample, as evidenced by a small shoulder in theXPS Ti2p core levels. The surface has a non-uniformappearance in STM, with a long-range corrugation thatdepends strongly on the tunneling conditions as observedpreviously [13]; these are attributed to either intrinsic orextrinsic subsurface defects.TiO is sensitive to electron irradiation, which can beused to create vacancies at the undercoordinated O sitesof the surface [17, 26]. An STM image of an electron-irradiated anatase (101) surface is shown in Figure 1(b).V O ’s appear as extra bright features at regular latticesites, consistent with STM simulations, see the Figs. 1(b1, b2). After exposure to 6.6 × electrons/cm ,the density of such V O ’s amounts to 12 % of a monolayer(ML, where 1 ML is defined as the number of primitiveunit cells, i.e. , 3.8 × cm − ). Assuming a simple,first-order desorption process, we estimate a cross sectionfor electron-induced O desorption of 3 × − cm − .The stability of these surface vacancies was probed byannealing the electron-irradiated sample for 10 minutesas described above. Each heating excursion was per-formed with a freshly-prepared and irradiated surface;the V O densities after the annealing steps are shown inFig. 2. No significant change was observed up to a tem-perature of 200 K; after an anneal to 230 K, the defectdensity decreases significantly. The higher the sampletemperature during the 10 min anneal, the fewer V O ’ssurvive. Above 320 K, new features appear that spanseveral unit cells, one is marked with a black box in Fig.1(c). These features (not taken into account in Fig.2)become more extended when an electron-irradiated sur-face is heated to higher temperatures (Fig. 1(d)), anddisappear completely above 500 K.In addition to heating excursions, we also followedthe fate of single V O ’s in time-lapse images at varioustemperatures. For these measurements we first equili-brated the STM for several hours at a specific temper-ature between 220 and 300 K. Electron bombardment ofthe freshly-prepared sample was again performed at 105K. (At this temperature we do not expect any surface- FIG. 2. (Color online) Stability of surface V O ’s created byelectron beam bombardment. The plot shows the density ofV O ’s after heating the sample to various temperatures for 10minutes, normalized to the initial value after electron irradi-ation at 105 K. The dashed line shows the expected behaviorassuming one E act of 0.75 eV. The full line assumes the trape-zoidal distribution of E act ’s from 0.6 – 1.2 eV displayed in theinset. to-bulk migration, Fig. 2.) The irradiated sample wasinserted into the temperature-stabilized STM, and seriesof images were taken. Fig. 3(a) shows an example ofsuch a time-lapse sequence, taken at T sample = 259 K.One of the defects, marked with an arrow, disappearsand returns to the same spot a few frames later. Wealso observed that defects disappeared at one positionand appeared at another position at the same or – lessfrequently – a neighboring row. The mobility of V O ’s in-creases with temperature, see Fig. 3(c). The total defectdensity, however, remains constant within the time frameof the experiment, see Fig. 3(b).It takes at least 10 minutes between the end of electron-irradiation (at 105 K) and the recording of the time-lapsesequences in our experimental setup. During this timethe total defect density decreases significantly, as shownin Fig. 2. This is the reason why the absolute V O densi-ties in Fig. 3 vary with temperature. On the other hand,the fact that the number of defects stays constant (Fig.3(c)) after the original, rapid decrease gives us confidencethat the data displayed in Fig. 2 indeed show the equi-librium concentrations at the given temperatures, andthat the finite time constants of our experiment do notinfluence the results.By DFT calculations, we estimate that the barrier,E act , for surface-to-bulk migration of V O ’s is 0.75 eV,while it is 1.15 eV for the reverse process (slight dif-ferences with respect to the barriers in Ref. [15] are due FIG. 3. (Color online) Results from time-lapse STM imagesof surface V O ’s on anatase (101). (a) Series of images (4 × ; +1.6 V / 0.2 nA) recorded at T = 259 K; the time be-tween images was 3.2 minutes. The arrows mark a V O thatdisappears and re-appears at the same position. (b) Totaldefect density in time-lapse images; each trace correspondsto a separate experimental run at the sample temperature in-dicated. (c) Defect mobility, represented by the number ofhopping events per defect and frame for different tempera-tures. to the larger surface model used for the present NEB cal-culations). The dashed line in Fig. 2 shows the expectedbehavior if we adapt this E act and a conventionally-usedpre-factor of 10 s − . While the onset of bulk migrationis consistent with the DFT result, the expected decreasewith temperature is much steeper than the measured one.In addition, the dis/re-appearance of the surface V O ’s,which leads to a temperature-dependent, dynamic equi-librium is hard to reconcile with the picture derived fromour DFT calculations: once the sample temperature ishigh enough to overcome the energetic barrier for surface-to-bulk migration, there is little reason for a V O to re-turn back to the surface. One should consider, however,that the calculations were performed assuming an ideal-ized case, i.e. , a perfect anatase slab devoid of any otherdefects except the single V O under investigation. Thisis different from the situation in the experiment, wheresubsurface defects are present at the outset. From titra-tion experiments using O adsorption we estimate thatthe density of Ti int ’s and V O ’s in the near-surface regionof our sample amounts to 2 ( ±
1) % of a ML at the clean,as-prepared surface. The uneven appearance of the STMimages from the clean surface (Fig. 1(a)) is attributedto local band bending effects. Thus at least some of thesubsurface defects are charged; plausibly these exert aconsiderable influence on the energetics and dynamics ofdefects migrating within their neighborhood. It is notunreasonable to assume a range of E act ’s for subsurfacediffusion, as this value will depend on the immediate envi-ronment of each surface V O . The full line in Fig. 2 takesinto account such a scenario, where we assume a trape-zoidal distribution of E act ’s ranging from 0.6 to 1.2 eV,as displayed in the inset of Fig. 2).The time- and temperature dependent behavior ofV O ’s can also be explained with such a range of activationenergies: starting with a certain surface V O concentra-tion, the defects that happen to reside above relativelyperfect region of the sample can disappear into the bulkonce a temperature >
200 K is reached. If another de-fect is present within the selvedge of the crystal, it willaffect the V O and change the activation energy for itsdisappearance into the bulk. It is well possible that thedefect migrates a certain distance in the subsurface regionbefore it pops up again – estimates for lateral diffusionenergies are in the range 1.1 - 1.8 eV (see SupplementalMaterial), hence the V O ’s can appear at different posi-tions, as is observed in the experiment. The extendedfeatures observed in Figs. 1(c, d) suggest that V O ’s ag-gregate in the near-surface region at moderate annealingtemperatures. The temperature dependence of bulk dif-fusion and defect equilbria observed in this work are pos-sibly affected by the initial V O concentration; this couldbe tested in future experiments.The experimental results presented in this work areunequivocal proof for the theoretical prediction that va-cancies are more stable in the bulk than at the surface.This prediction, originally based on DFT-PBE calcula-tions [14, 15], is also supported by results from hybridcalculations which account for the polaronic character ofV O -induced Ti states and are thus considered moreaccurate for the study of defects in TiO [27, 28], seeSupplementary Material. While hybrid calculations arestill too demanding to be used for diffusion barrier deter-minations, DFT+U studies indicate that the barriers forthe hopping diffusion of the Ti polarons are low, typi-cally between 0.1 and 0.3 eV [29–31] Therefore the effectof excess electron localization on V O migration barriersis expected to be relatively minor, as has recently beenshown for H diffusion in anatase. [32]An inspection of the anatase (101) surface structureprovides a simple qualitative rationale for the instability of surface V O ’s: removal of an O gives rise to one five-fold and one highly unstable four-fold coordinated Ti cation, whereas bulk V O ’s have two five-fold coordinatedTi cations. Moreover, the Ti-O bonds are short andstrong, so breaking two Ti-O bonds at the surface isenergetically more costly than to breaking three in thebulk. Clearly, the resulting subsurface defects have to bereckoned with when considering the surface chemistry ofTiO anatase, and some observations have already beeninterpreted along these lines [16, 33]. Subsurface migra-tion automatically results in inhomogeneity within theselvedge of the crystal, which, in turn, affects the acti-vation energies. The dynamic equilibrium of surface Ovacancies will then depend on the presence of intrinsic aswell as extrinsic charged defects. Even at room temper-ature defects come and go from the surface, suggestingthat the chemically active sites change across the surface.Generally, the flow of lattice oxygen (defects) to andfrom the surface is of continued interest in solid-statechemistry, and important in established and emergingtechnologies such as catalysis [34], solid-oxide fuel cells[35] and memristor devices [36]. Direct observation ofsuch defect migration, combined with modeling at theatomic scale can help pave the way for future experimentsthat give insights into the relevant processes.Acknowledgement: This work was supported by theAustrian Science Fund (FWF; Project F45) and theERC Advanced Grant ’OxideSurfaces’. AS acknowledgessupport from DoE-BES, Chemical Sciences, Geosciencesand Biosciences Division under Contract No. DE-FG02-12ER16286. Calculations were performed at the TI-GRESS high performance computer center at PrincetonUniversity. ∗ [email protected][1] U. Diebold, Surf. Sci. Rep. , 53 (2003).[2] C. L. Pang, R. Lindsay, and G. Thornton, Chem. Soc.Rev. , 2328 (2008).[3] X. Chen and S. Mao, Chemical Reviews , 2891(2007).[4] U. Diebold, N. Ruzycki, G. Herman, and A. Selloni,Catal. Today , 93 (2003).[5] W. Hebenstreit, N. Ruzycki, G. S. Herman, Y. Gao, andU. Diebold, Phys. Rev. B , R16334 (2000).[6] N. Ruzycki, G. Herman, L. Boatner, and U. Diebold,Surf. Sci. , 239 (2003).[7] M. Xu, Y. Gao, E. M. Moreno, M. Kunst, M. Muhler,Y. Wang, H. Idriss, and C. W¨oll, Phys. Rev. Lett. (2011).[8] L. Walle, A. Borg, E. M. J. Johansson, S. Plogmaker,H. Rensmo, P. Uvdal, and A. Sandell, J. Phys. Chem. C , 9545 (2011).[9] Z. Dohn´alek, I. Lyubinetsky, and R. Rousseau, Progr.Surf. Sci. , 161 (2010).[10] S. Wendt, P. T. Sprunger, E. Lira, G. K. H. Mad-sen, Z. Li, J. O. Hansen, J. Matthiesen, A. Blekinge- Rasmussen, E. Laegsgaard, B. Hammer, et al., Science , 1755 (2008).[11] C. M. Yim, C. L. Pang, and G. Thornton, Phys. Rev.Lett. , 036806 (2010).[12] Z. Zhang and J. T. Yates Jr., J. Phys. Chem. C ,3098 (2010).[13] Y. He, O. Dulub, H. Cheng, A. Selloni, and U. Diebold,Phys. Rev. Lett. , 106105 (2009).[14] H. Cheng and A. Selloni, J. Chem. Phys. , 054703(2009).[15] H. Cheng and A. Selloni, Phys. Rev. B , 092101 (2009).[16] U. Aschauer, Y. He, H. Cheng, S. Li, U. Diebold, andA. Selloni, J. Phys. Chem. C , 1278 (2010).[17] O. Dulub, M. Batzilln, S. Solovev, E. Loginova, A. Alcha-girov, T. E. Madey, and U. Diebold, Science , 1052(2007).[18] O. Dulub and U. Diebold, J. Phys.: Condens. Matter ,084014 (2010).[19] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev.Lett. , 1396 (1997).[20] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car,C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococ-cioni, I. Dabo, et al., J. Phys.: Condens. Matter ,395502 (2009).[21] J. P. Perdew, M. Ernzerhof, and K. Burke, J. Chem. Phys , 9982 (1996).[22] J. VandeVondele, M. Krack, F. Mohamed, M. Parrinello,T. Chassaing, and J. Hutter, Computer Physics Commu-nications , 103 (2005).[23] G. Henkelman, B. P. Uberuaga, and H. J´onsson, J. Chem.Phys. , 9901 (2000). [24] G. Mills and H. J´onsson, Phys. Rev. Lett. , 1124(1994).[25] X. Gong, A. Selloni, M. Batzill, and U. Diebold, Nat.Mater. , 665 (2006).[26] C. Pang, O. Bikondoa, D. Humphrey, A. Papageorgiou,G. Cabailh, R. Ithnin, Q. Chen, C. Muryn, H. Onishi,and G. Thornton, Nanotechnol. , 5397 (2006).[27] M. Ganduglia-Pirovano, A. Hofmann, and J. Sauer, Surf.Sci. Rep. , 219 (2007).[28] E. Finazzi, C. Di Valentin, P. G., and A. Selloni, J. Chem.Phys. , 154113 (2008).[29] N. A. Deskins and M. Dupuis, Physical Review B ,195212 (2007).[30] P. M. Kowalski, M. F. Camellone, N. N. Nair, B. Meyer,and D. Marx, Physical Review Letters , 146405(2010).[31] N. A. Deskins, R. Rousseau, and M. Dupuis, The Journalof Physical Chemistry C , 7562 (2011).[32] U. Aschauer and A. Sellini, Phys. Chem. Chem. Phys.(2012), accepted.[33] M. Xu, H. Noei, M. Buchholz, M. Muhler, C. W¨oll, andY. Wang, Catalysis Today , 12 (2012).[34] F. Esch, S. Fabris, L. Zhou, T. Montini, C. Africh, P. For-nasiero, G. Comelli, and R. Rosei, Science , 752(2005).[35] J. Suntivich, H. A. Gasteiger, N. Yabuuchi, H. Nakanishi,J. B. Goodenough, and Y. Shao-Horn, Nat. Chem. , 546(2011).[36] R. Waser and M. Aono, Nat. Mater.6