Electronic structure of stoichiometric and Ar+-bombarded ZrO2 determined by resonant photoemission

PHYSICAL REVIEW BVOLUME S2, NUMBER 16 1S OCTOBER 1995-II Electronic structure of stoichiometric and Ar+-bombarded ZrO2 determined by resonant photoemission C. Morant Cantoblanco, E-28049 Madrid, Spain A. Fernandez and A. R. Gonzalez-Elipe Instituto Ciencia de Materiales de Sevilla, Consejo Superior de Investigaciones Cientigcas, Universidad de Sevilla, Departamento de Quimica Inorga'nica, P.O. Box 1115,E41071 -Sevilla, Spain L. Soriano Can toblanco, E-28049 Madrid, Spain A. Stampfi and A. M. Bradshaw Fritz Haber I-nstitut -der Max Planck G-esellscha-ft, Faradaytveg 4-6, 14195Berlin, Germany J.M. Sanz* Cantoblanco, E-28049 Madrid, Spain (Received 2May 1995;revised manuscript received 10July 1995) The electronic properties of thermally grown ZrO& thin films before and after Ar+ bombardment have been studied with resonant photoemission spectroscopy using synchrotron radiation. For stoichiometric Zr02 thin films the experimental valence-band spectra are in good agreement with the calculated density of states for bulk Zr02. For both stoichiometric and Ar+-bombarded Zr02 thin films, resonant photoemission from the valence band was observed when the photon energy was swept through the Zr 4p ~4d transition energy. The resonant profile was found to exhibit amaximum at hv=39 eV, followed by a second well-resolved broad maximum around 50 eV. The feature at 39 eV is consistent with resonant enhancement of the Zr 4d states and has been used to identify those regions of the valence band with an important Zr 4d admixture. The results are in good agreement with the calculated Zr 4d partial density of states. The intensity increase observed at hv-45 — 50 eV is found to be associated with the nonbonding region of the valence band, although aproper interpretation is needed. In addition, it was found that Ar+ bombardment induces electronic states in the band-gap region and changes in the 02p valence band. Three distinct emission bands were identified in the band gap as afunction of the Ar+ dose. They are associated with the formation of oxygen vacancies and mixed oxidation states due to preferential sputtering of the oxygen atoms. Resonant photoemission of these Ar+-bombarded films demonstrates both the cationic character of the band-gap states and the increase of the cationic contribution to the 0 2p valence band. I. INTRODUCTION The main purpose of this paper is to investigate the electronic structure of thermally grown Zr02 thin films as well as the effects induced by Ar+ bombardment using resonant photoemission spectroscopy (RPES). The electronic band structure of Zr02, arefractory transitionmetal oxide (TMO), is of great interest from both the technological and fundamental points of view. Its wide use as ahigh-performance ceramic for nuclear and hightemperature applications, as well as in gas sensing, electro-optical, thin-film, and catalysis applications makes band structure information very important for understanding its physical and chemical properties. Past investigations seem to have been oriented mainly towards practical aspects, 'rather than towards these more fundamental aspects. 'Only afew experiments have probed the electronic structure of Zr02, 'and their interpretation has relied on alimited number of band structure calculations. Moreover, the electronic structure of defects on transition-metal oxides has also become a subject of extensive study. 'Low-energy ion bornbardment is often used to create defects in TMO's, but, because of the difficulty of characterization, its effect on electronic properties has not yet been sufFiciently studied. Many spectroscopic studies have shown that oxygen vacancies cause localized energy levels in the band gap. In most cases it has proven difficult, however, to relate the observed electronic states to specific defects, even though this has been attempted recently in some theoretical approaches. ''To our knowledge only Sobolev et al. ,using aself-consistent scattered-wave method, 0163-1829/95/52(16)/11711(10)/$06.00 11 711 1995 The American Physical Society 11 712 C. MORANT et al. have tried to study the inAuence of oxygen vacancies on the electronic structure of Zr02. The changes in composition and electronic structure induced in Zr02 by 3keV Ar+ bombardment have been studied by core and valence-band x-ray photoelectron spectroscopy (XPS) as reported in arecent paper. The results showed that the electronic structure was fairly sensitive to Ar+ sputtering, atreatment that produces emission features in the band gap and other changes in the O2p region of the valence band (VB), which in turn were ascribed to the formation of oxygen vacancies and suboxides (i.e.,reduction). However, the relatively low photoemission cross section of the VB prevented any definitive conclusions from being drawn as to the electronic structure of the stoichiometric and Ar+-bombarded solid and of the Ar+-modified surface. The phenomenon of RPES in transition metals and their compounds has been the subject of many recent investigations. Alarge resonance in the dphotoionization cross section is known to occur at photon energies close to the threshold for ap~d transition. The resonance process is usually explained as being due to the interference between the direct photoemission process and autoionization of ahighly localized excited state created by photoabsorption. Similar effects are found for the valence bands of TMO's even when those have aformal dconfiguration: It is assumed that the resonance is due to hybridization between the 02p and cation dorbitals. In fact, RPES is used extensively as amethod to isolate the d-state contribution and to understand the VB structure of transition-metal compounds and other complex hybridized systems. For adetailed discussion the reader is referred to the review by Davis and the book by Henrich and Cox. Studies of the resonant phenomenon itself are also interesting since the details of the resonance in some compounds are not fully understood. For example, extra resonances not directly related to the metal dstates have been recently observed in Ti02 (Ref. 37) and MoS2, but asatisfactory explanation has not yet been put forward. In this paper we describe aPES and RPES study of Zr02 thin films before and after Ar+ bombardment aimed at obtaining abetter understanding of its valenceband structure and of the changes induced by Ar+ bombardment. In Sec. II we describe some experimental details and the sample preparation. In Sec. IIIAwe present acomparison of the experimental data with the calculated density of states (DOS) to obtain areliable picture of the valence-band structure. The RPES study of the ZrQ& valence band is presented in Sec. IIIB. Apreliminary account of some of these results appeared elsewhere. 'Section IIICis dedicated to the effects induced by Ar+ bombardment. In Sec. IIIDwe present and discuss the RPES data from ahighly reduced surface (i.e., ZrO„) obtained by Ar+ bombardment of the original surface. Asummary and the conclusions are presented in Sec. IV. II. EXPERIMENTAL DETAILS Apolycrystalline Zr foil of 99.98%%uo purity from Goodfellow Metals has been used throughout this work. The foil was subjected to the usual cleaning procedures by rinsing in acetone, ethanol, and distilled water before it was introduced into the UHV chamber. The surface was then cleaned by Ar+ sputtering and annealed under the conditions previously described in order to avoid segregation of impurities which hinder the oxidation. Thermally grown Zr02 thin films were obtained by heating the foil at -450 Kfor longer than 30 min in an oxygen atmosphere (I'o — 10 Torr). In some cases, after long periods of subsequent exposure to the residual gas some OH adsorption was observed, but could be easily removed by heating at 373 Kfor short periods. In order to induce surface defects and changes in stoichiometry the sample was bombarded with Ar+ ions using an ion gun operated at 0.5and 2keV and current densities of — 1pA cm The photoemission experiments were performed on the TGM3 monochromator at the Berlin synchrotron radiation source (BESSY) using photon energies between 30 and 70 eV. At these energies the escape depth of the photoelectrons is expected to be less than 10 A. The angle of incidence of the p-polarized light was 45 with respect to the sample normal. The spectra were recorded at normal emission with an ADES-400 electron energy analyzer. The overall resolution (monochromator + spectrometer) was typically 600 meV. Overview spectra were measured in the constant-retarding-ratio mode, whereas the valence-band spectra in the 0~E~ ~14 eV binding-energy range were taken in the constantanalysis-energy mode with apass energy of 20 eV. For a more detailed description of the monochromator the reader is referred to the BESSYuser's handbook. All the valence-band spectra presented in this paper have been normalized to the photon Aux as measured at a gold mirror at the entrance of the spectrometer chamber. In most cases abackground has been subtracted from the spectra using standard methods. III. RESULTS AND DISCUSSION A. Electronic structure of ZrO& Figure 1shows an overview spectrum of thermally grown ZrOz recorded at hv=50 eV. The spectrum shows the Zr 4p doublet at 30.6eV, the broad 02s peak at 21.7eV, and the valence band peaked at 5.3eV with a clearly distinguishable shoulder at higher energies. The binding energy is referred to the Fermi level of metallic zirconium. Figure 2shows the 0— 14 eV binding-energy (E~) range measured with hv=35 eV. For comparison we have included the total density of states, as well as the partial oxygen and zirconium densities of states, calculated by Soriano et aI. for cubic zirconia using the LSW method. The calculated DOS was obtained in terms of the Zr and 02p partial DOS's weighted according to their respective cross sections. For amore detailed description of this procedure see Ref. 8. The overall shape of the experimental curve is well explained by the calculated bulk DOS. The valence band exhibits two distinct structures at -9and -6eV binding energy, which according to the 52 EI.ECTRONIC STRUCTURE OF STOICHIOMETRIC AND Ar+-. . . 11 713 tI) 12 10- Zr02, hu=508V EF 8- I— (0 Z4- LLI I— Z2- 0- I 50 IIIII 40 30 20 10 BINDING ENERGY (eV) FIG. 1. Photoemission spectrum measured at hv=50 eV from stoichiometric Zr02. ES os 0- I— GO Z UJ I— Z band structure calculation correspond to the bonding O 2pcr and nonbonding 02pm. orbitals, respectively, and contain an important zirconium contribution. The Zr contribution is more significant in the higher-binding- energy region, where the bonding 02p orbitals are expected to have alarger admixture of Zr orbitals, and illustrates the importance of the covalent interaction (estimated to be about 24%) in the bonding of Zr02. The width of the valence band is estimated to be -4.8eV from the energy separation of the points of maximum slope on each side of the band. The valence-band maximum (VBM) is located at — 3.8eV below the Fermi level as determined by extrapolation of the tangent on the low-binding-energy side of the 02p band. This value is in good agreement with that determined by other electron spectroscopies" and optical gaps reported in the literature. ''" Apart from the almost classic calculation of Morinaga, Adachi, and Tsukuda based on adiscrete variational (DV) Xa cluster method, various other attempts have been made to study the electronic structure of ZrO2. Using the linear-muftin-tin-orbitals (LMTO) method Medvedeva et QI. have determined the energy band structures of Zr02 and other TMO's in order to investigate trends in the band gap and cohesive energy. The band structure of cubic and tetragonal zirconia have been calculated by Jansen using density-functional theory. The structural and electronic properties of cubic and tetragonal zirconia were also explored by Orlando et al. using aperiodic ab initio Hartree-Fock method. Zandiehnadem, Murray, and Ching performed calculations for the three stable phases of Zr02 (i.e.,monoclinic, cubic, and tetragonal) using the orthogonalized linear combination of atomic orbitals (OLCAO) method, and concluded that the bonding is very similar for the three crystal structures with only minor differences in the width of the band and the band gap. In addition to the general good agreement aresult common to all these calculations is the existence of an important covalent bonding contribution, which has been confirmed experimentally with 0 ls x-ray absorption spectroscopy' (XAS) and resonant photoemission. '" Adetailed discussion of these calculations and their comparison with other approaches as well as with experimental photoemission (PES) and bremsstrahlung isochromat (BIS) spectra can be found in Refs. 8and 13. We note that the photoemission intensity of the PES spectra reported here does not fall to zero within the band gap even for the freshly prepared surface. Weak emission in the band gap extends from the top of the valence band up to the Fermi level, showing amaximum at Ez =2.5eV. An expanded view of their contribution is shown in Fig. 2. Usually, such bands are ascribed to trapped electrons in oxygen vacancies and other surface defects' caused by the preparation method used (i.e., heating at low oxygen pressures). However, by means of aself-consistent scattered-wave method, Sobolev et al. have predicted that oxygen vacancies in perfect Zr02 crystals are expected to cause trapped electrons in local states within the gap at binding energies between 1.5and 2eV, in good agreement with the band observed in this study. Aprevious XPS study has quantified this effect giving an electron surface density of 6X 10' cm which is equivalent to amonolayer of oxygen vacancies. It will be shown below that the intensity of this feature can be substantially increased by ion bombardment. B. Resonant photoemission in ZrIO2 Dos os 14 12 10 86420 BINDING ENERGY (eV) FIG. 2. Calculated bulk DOS (solid line) (see text and Ref. 8 for details) and experimental valence-band photoemission spectrum (points) at hv=35 eV for stoichiometric ZrO2. Lower curves: Calculated Zr and 0partial densities ofstates (Ref. 8). The process of resonant photoemission has been extensively used as amethod for isolating the cationic contribution to the valence band of transition-metal compounds. The energy distribution curves (EDC's) from the valence band of Zr02 were thus measured at photon energies between 30 and 70 eV as shown in Fig. 3. They indicate that the relative intensities of the two features observed in the valence band are very sensitive to the photon energy due to aresonance process associated with the Zr 4p threshold. Although the resonance manifests itself C. MORANT et al. III'LII'III'I'II Zr02 Eh(eV) F70 10- a) Zr02 EF 60 50 5- 45 42 40 39 38 36 33 30 ~iIIIfl 14 12 10 86420-2 BINDING ENERGY (eV) FIG. 3. Valence-band photoemission spectra of Zr02 as a function ofthe photon energy. over the whole valence band at photon energies between 36 and 50 eV, the 02p bonding region on the highbinding-energy side of the valence band is enhanced at photon energies around 39 eV. In the case of the spectrum taken at hv= 30 eV the unusually high intensity of the gap states is due to the contribution of the Zr 4p core levels excited with second-order light. In order to study separately the behavior of the bonding and nonbonding contributions to the valence band, we followed aprocedure previously used by Zhang, Jeng, and Heinrich in their study of resonance effects in Ti02. According to these authors the valence-band spectra can be analyzed in terms of three Gaussians, labeled "bonding, ""nonbonding, "and "overlapping, "the widths and energies of which are summarized in Table Ifor the whole series of spectra. Atypical fit is shown in Fig. 4(a) for the VB spectrum measured at hv=50 eV. Obviously the high- and the low-binding-energy components involve predominantly the bonding and nonbonding orbitals, respectively, whereas the component labeled "overlapping" helps to get abetter fit to the valence band. Amore quantitative insight into the resonance effects TABLE I. Best-fit parameters binding energy (Eb) and full width at half maximum (FWHM) for the line shape of the valence band at different photon energies. Three and two Gaussians were used for Zr02 and ZrO, respectively. The mean, standard deviation (SD), and median value are given for each parameter. 0- LU Z10- 0- I'I'I'I'1'I'II 14 12 10 86420-2 BINDING ENERGY (eV) FIG. 4. Analysis of the valence-band photoemission spectrum at hv=50 eV using three and two Gaussians for ZrO& (a) and ZrO (b), respectively. can be gained by integrating the areas associated with the different components of the valence band and thus generating spectra equivalent to the usual constant-initial- state (CIS) curves. The photon energy dependence of the total intensity of the valence band has thus been depicted in Fig. 5together with that of the intensity of the bonding, nonbonding, and overlapping components and of the emission band in the gap. The profiles are rather complex and quite unlike the ideal Fano-like curves. In the case of the total valence-band intensity (curve a) the pro61e is characterized by arise at 36 eV with amaximum at hv=39 eV, and by asecond broad maximum at energies between 45 and 50 eV. The intensity of the bonding component (curve b) reaches the main maximum I'I'I'I'I'I'I'I'I e 35— 30- 25- N20— I— 15 -~ 10- Eb {eV) Mean SD FWHM {eV) Median Mean SD Median 0- Bonding 8.74 Nonbonding 5.67 Overlap 7.14 Bonding 8.63 Nonbonding 6.51 0.06 0.07 0.07 0.09 0.20 Zr02 8.75 5.68 7.15 ZrO 8.65 6.60 2.32 2.09 1.89 2.53 2.57 0.05 0.03 0.01 0.07 0.04 2.31 2.10 1.89 2.57 2.57 IIiIiI~I I I IiIsI 30 35 40 45 50 55 60 65 70 PHOTON ENERGY (eV) FIG. 5. Photon energy dependence of the integrated intensity of the total valence-band emission (0) {curve a) and of the band-gap emission {0)(curve e) as well as of the bonding (A) (curve b), nonbonding (0)(curve c), and overlapping (Q) (curve d) components. 52 ELECTRONIC STRUCTURE OF STOICHIOMETRIC AND Ar+-. . . 11 715 at approximately 39 eV followed by abroad plateau which extends up to 50 eV. On the other hand, the resonance profile of the nonbonding component (curve c) consists of asmall bump at 39 eV and agradual rise up to -50 eV to fall off at higher energies, so that in this case the maximum enhancement of the intensity is produced at photon energies around 45— 50 eV. The overlapping component (curve d) shows no clear evidence of resonant behavior, but shows rather constant intensity up to 50 eV photon energy. The profile corresponding to the states in the band gap (curve e) suggests aweak resonance at 39 eV, but because of the low intensity and possible errors in the background subtraction procedure an unambiguous characterization is difficult. The lack of asystematic picture in the observation of the two resonances (i.e.,at 39 and 50 eV) may be linked to the fact that none of the peaks used in the fitting procedure actually correspond to pure bonding or pure nonbonding orbitals. In order to follow the changes observed in the valence-band shape in ameaningful way, it is useful to compute the 39/36 eV difference spectrum by subtracting the EDC measured at 36 eV (off resonance) from that measured at 39 eV (on resonance). The result is labeled b in Fig. 6. The curve labeled acorresponds to an XPS valence-band spectrum (i.e.,hv= 1253.6eV) from Ref. 26. For comparison we have included several curves labeled from cto fwhich represent the Zr and 0PDOS's obtained by different methods. Curves cand fare the calculated Zr and 02p partial density of states, respectively (see Fig. 2) scaled by the relative photoemission cross sections at hv= 1253.6eV. Since the Zr 5s and Zr 5p contributions are negligible, we will use curve cto represent the Zr 4d density of states. Curves dand eare intended to represent the 02p density of states. Curve d I'I'I'I'I'I'I'I'I Zr02 a) XPS Zr 4d b) c) 02p d) I'I'I'I'I'I'I'I'I 14 12 10 86420-2 BINDING ENERGY (eV) FIG. 6. Curve a: XPS spectrum of the valence band from stoichiometric Zr02. Curve b: Zr 4d contribution to the valence band as derived from the resonance at hv=39 eV. Curve c: Calculated Zr partial density of states (from Ref. 8). Curves df: 02p contribution — to the valence band as determined from (curve d) photoemission at hv=70 eV, (curve e) the XPS spectrum (a) after subtraction of ib), and (curve Pelectronic structure calculations (Ref. 8). is the valence band of Zr02 measured at hv=70 eV (see Fig. 3), which is expected to refiect the 02p PDOS since the ratio of the 02p and Zr 4d photoemission cross sections is &30 at that photon energy. "Curve eis the result of the subtraction of the Zr contribution as represented by spectrum bfrom the experimental XPS valence band (i.e.,curve a) (see details below). Resonant photoemission of the Zr 4d states in Zr02 is expected to occur when the photon energy is tuned over the Zr 4p ~48 transition energy at — 31 eV. The 4p~4d transition creates an excited state [Zr 4p 4d']' which remains highly localized on the cation site. The excited state can then relax in adirect recombination process which involves the Zr 4d states participating in the valence band; i.e.,[Zr 4p 4d 'VB"]~Zr 4p 4d VB" '+e, which is the same final state as in conventional photoemission. In this manner there is a possible interference between both processes, giving rise to the resonant enhancement of those parts of the spectrum with an important cationic character. In the present study Fig. 6clearly demonstrates that this is the case for the resonance observed at hv=39 eV. The good agreement existing between the 39/36 eV difference spectrum, curve b, and the calculated Zr 4d PDOS, curve c, clearly confirms that the phenomenon can be used to identify the Zr-derived states in the valence band. Furthermore, the difference spectrum bprovides experimental evidence that the partial covalent character of the Zr- 0bond is larger in the bonding region than in the nonbonding part of the valence band. Therefore, the bonding contribution should resonate more strongly than the nonbonding one in good agreement with the experimental observation (see Fig. 5). The resonance at hv=39 eV is then consistent with the resonant photoemission of the Zr 4d states and with the assumption of apure intra-atomic relaxation of the excited state. It is important to mention here that the resonant photoemission of the dstates at 39 eV occurs at an 8.4eV higher energy than the Zr 4p binding energy (E~=30.6 eV) as measured by photoemission (see Fig. 1). Similar delays between the threshold (Zr 4p binding energy) and the resonance energies have also been observed in metallic Zr (— 10 eV), 'as well as in the zirconium nitrides ZrN (— 10 eV), and Zr3N4( — 11.5eV), and seem to be acommon observation in the lighter transition metals and their compounds. In general, the binding energy of the [4p 4d "+' ]*excited state measured from resonant photoemission profiles agrees well with the results of other measurements such as optical absorption or electronenergy-loss spectroscopy (EELS). For Zr02, the Zr 4p core losses have been shown to exhibit amaximum intensity around 42 eV,"in reasonable agreement with the energy measured by resonant photoemission. Figure 6also allows the comparison between the spectrum assigned to the Zr 4d states and the valence-band spectrum obtained by XPS. Interestingly, it is observed that the XPS spectrum is more intense around the top of the valence band (i.e.,the nonbonding region). Clearly this is due to the 02p-derived states, because the cationic contributions from Zr 5s and Zr 5p states are negligible. The O2p states cannot be neglected: although their cross 11 716 C. MORANT et al. 52 section at XPS energies is four times lower than that of the Zr 4d orbitals, the much larger 02p partial DOS compensates. We have attempted to extract the 02p contribution from the experimental XPS valence band by subtracting the Zr 4d contribution determined by resonant photoemission. The result is labeled as ein Fig. 6 and shows good agreement with the calculated partial 0 2p density of states depicted in the same figure as f. This agreement is not surprising because the difference was obtained by subtracting aproperly scaled difference spectrum, for which the scaling factor was determined by aprevious fitting of the experimental XPS valence band in terms of the calculated partial densities of states and the respective cross sections as tabulated by Yeh and Lindau, so that the relative 02p/Zr 4d contributions to the valence band were determined according to theoretical calculations. At present we do not have asatisfactory explanation for the broad resonance occurring at an energy near 50 eV. Recently, Heise and co-workers ''performed a RPES and photoelectron diffraction (XPD) study of the TiOz(110) surface to determine the Ti 3d and 02p valence-band partial densities of states. The RPES results show different displacements of the resonance maxi- Ina for different features of the VB, which the authors suggest are associated with different symmetry-resolved initial and final states, although the details of the assignment remain unclear. Assuming apurely intra-atomic direct recombination mechanism, the resonance could be associated with transitions of the Zr 4p — +Zr 5sp type, where a4p electron is excited into the empty 5sp band allowing aresonance with the cationic Zr 5s states involved in the valence band of Zr02. Obviously, the Zr 5p states should not play any role since transitions 4@~5p are dipole forbidden. In fact, this mechanism was previously used to explain extra resonances in RPES experiments on Ti02 (Ref. 37) and MoSz. The profile of the resonance indicates that in that case the [Zr 4p 5s']* excited state lies 9eV above the [Zr 4p d']* state, in good agreement with previous XAS experiment' which locate the Zr 5' band -9eV above the empty dband. However, the calculated Zr 5s PDOS does not appear to be intense enough to give rise to the observed enhancement. Unfortunately, the EELS spectra of Zr02 do not show any distinguishable feature which could be associated with the resonance at 45 eV; thus either the corresponding excited state is not available with EELS or the excitation cross section is very low. C. Ar+ bombardment effects Figure 7shows valence-band spectra in the 0— 14 eV binding-energy range, taken at hv=39 eV (i.e.,on resonance), for the initial ZrOz surface (lowest curve) and after various degrees of Ar+ bombardment. In addition, the band-gap region (4 eV ~EiI ~0) after subtraction of the main O2p valence band is displayed on an expanded scale on the right-hand side of Fig. 7. The figure clearly shows that the electronic structure of ZrO2 is sensitive to bombardment with Ar+ ions of 2keV and that the spectroscopic changes introduced by low-energy sputtering 2keV Ar+ Zr02 15 Sputtering time (min. )I EF ','.EF I ~, I~ I0 I I I .i I 130 i~'~— 50 Ie ~~ ~~ if l —13 ~~ ~~ j 0 I~IcI~giIaI~II~II, I 10 505432 10-1-2 BINDING ENERGY (eV) FIG. 7. Photoemission spectra (hv=39 eV) of the valence band of Zr02 (left) and band-gap emission on an expanded scale (right) as afunction ofthe sputtering time with 2keV Ar+ ions. begin to appear immediately upon bombardment. The spectra obtained from afresh Zr02 surface as well as after the first three minutes of sputtering exhibit a broad structure of constant shape in the band gap, the intensity of which increases with sputtering time, and whose shape remains unchanged. Its emission maximum is found at E~=2.5eV and its origin was explained in Sec. IIIAin terms of oxygen vacancies. At higher Ar+ doses the structure is observed to develop two additional features around 1.7and 3.7eV which also grow upon further Ar+ bombardment. In order to follow the different features separately the emission in the band gap was first analyzed by accounting for the broad emission at 2.5eV. Difference spectra were generated by removing the maximum possible contribution of the original structure (i.e.,0min) from the spectra at different sputtering times, so that the other two features could be clearly distinguished. These difference spectra are displayed in the inset of Fig. 8labeled with the scaling factor Fwhich was used for each spectrum. Above 3min sputtering time (i.e.,— 1X10' ions cm ) two peaks at 3.8and 1.2eV can be clearly distinguished, and their growth can be followed as the sputtering process proceeds. The integrated intensities of the respective features have been depicted in Fig. 8as a function of the sputtering time, together with the total intensity of the emission band in the gap. The intensity of the whole emission band induced by Ar+ ions of 0.5keV has also been included for comparison. The figure shows that the intensities of the different features and of the whole emission band increase progressively until astationary state is reached for sputtering times above 60 min (i.e.,2.2X10' ions cm ). What is the origin of the three different gap emissions which can be produced by Ar+ bombardment? The first type is the broad emission at 2.5eV binding energy, 52 ELECTRONIC STRUCTURE OF STOICHIOMETRIC AND Ar+-. ..11 717 M8 7- Cg ~6- I— M5- Z'. LLI I— Z' (0 I—3- 0 UJ LL 2 Cl Sputtenng Time (min. )."'. E, F 130 '.: !' ~~3.S IS .-' 2.3 0)~- 14 e4 2 0-2 ~BINDING ENERGY (ev) kev I— (f) LLj ~0 0 o eo ~o eo 10 ~3 I 0~ oe 0~ o~ oe .. 'Zr02 b o~ 0'I 0~ I'I'I'I'I'I'I'I'I 14 12 10 86420-2 BINDING ENERGY (eV) 0- 50 100 150 FIG. 9. Valence-band photoemission spectra from both stoichiometric Zr02 and reduced ZrO measured at hv=39 eV. SPUTTERING TIME (ITiin) FIG. 8. Intensity of the band-gap emission as afunction of the sputtering time with 2keV (~)and 0.5keV (0) Ar+ ions. For 2keV Ar+ ions the intensities of three diferent contributions at 1.2eV (V), 2.5eV (o), and 3.8eV (4) have been included. Inset: band-gap emission after subtraction of the spectrum at 0min scaled with afactor F. which forms after any degree of reduction of the surface and is present even on afresh Zr02 surface. It has been assigned to electrons trapped on oxygen vacancies that populate the 4d orbitals of adjacent Zr atoms. The rneasured energy of this band is in good agreement with the calculations by Sobolev et al. using aself-consistent scattered-wave method. Furthermore, Munnix and Schrneits' have also shown that an emission band around 1.3eV observed in Ti02 could be accounted for by subsurface oxygen vacancies. In fact, it seems to be ageneral finding in TMO's that oxygen vacancies induce electronic states in the band gap. ' At higher Ar+ dose the increase in the concentration of oxygen vacancies produces more complex defect structures and even significant changes of stoichiometry. This is probably the origin of the two sharp features observed only after heavy reduction of the surface (i.e., high-defect-density surfaces). The high-binding-energy feature at 3.8eV we assign to bulk Zr-0 vacancy complexes associated with the formation of lower oxidation states (i.e.,Zr +, Zr +). Furthermore, the band can be related to an optical absorption at 3.31 eV observed in cubic zirconia and attributed to sixfold-coordinated Zr + by Azzoni and Paleari using electron paramagnetic resonance. In contrast, the weakest band observed at 1.2 eV requires still heavier reduction to be unambiguously distinguished. It is tentatively ascribed to the formation of small clusters of Zr+ and even Zr in the limiting species of the sputtering process when Ar+ bombardment produces much higher defect densities, and more electrons populate the Zr 4d orbitals. In fact, the formation of mixed oxidation states (i.e.,Zr +, Zr +, Zr+, and Zr ) has been demonstrated by XPS. The cationic character of all these band-gap states is confirmed by the resonant photoemission spectra shown below. Although the most significant effect of the Ar+ bombardment is the growth of the emission features in the gap, the 02p valence band is also observed to undergo changes. Figure 9compares the valence-band photoemission spectra for the stoichiometric and severely reduced surfaces. In general, we observed adecrease in the valence-band width and an increase in its intensity due to the presence of additional reduced cations, which increase the dcontribution. The valence band maximum moves 0.45 eV away from the Fermi level (i.e, band bending) as aconsequence of the reduction of the surface. In fact, it is well established that electrons trapped in oxygen vacancies occupy localized empty orbitals of the cations, pushing up the Fermi level and bending the whole electronic structure to higher binding energies. Figure 10 shows photoemission spectra of ZrO after its exposure to increasing amounts of 02 at temperatures between 298 and 473 K(as labeled). The figure shows rather clearly that an oxygen exposure of 100 Lat 298 K causes asignificant decrease in the intensity of the gap states. On the contrary, practically no change is observed in the shape of the 02p band. Only after exposures above 300 Lat 423 Kdoes the valence-band spectrum recover the shape of that for Zr02. Thus, although generation of states in the band gap and modification of the valence-band shape occur simultaneously during Ar+ bombardment, the second effect remains even after removal of the oxygen vacancies by 02 adsorption at low temperatures. This fact suggests that the shape could be related to the amor phization and loss of the threedimensional order of Zr02 by Ar+ bornbardrnent. Furthermore, according to these results the modified structure characterized by ahigher covalency (see below) is preserved until the structure is annealed above 423 K. In order to observe any possible inAuence from the ion energy we performed asimilar detailed study of the effects induced by 0.5keV Ar+ ions. The effects are qualitatively very similar to those discussed above for 2keV Ar+ ions, but not so intense. After asimilar sputtering time we observed that the steady state reached at 0.5keV consists of an altered layer with considerably fewer defects than that obtained by sputtering with 2keV Ar+ ions (Fig. 8). In the following resonant photoemission experiments on the reduced surface using 0.5keV ions are described. 11 718 C. MORANT et al. ~I~III~I~I~I~ EF +100 L02 473K ~4 +100 L02 423K +100 L02 298K Ar' 2keV I'I'IIII'I 14 12 10 86420 BINDING ENERGY (eV) ~I FIG. 10. Photoemission spectra (hv=39 eV) of the valence band of ZrO (2 keV Ar+ ions) after oxygen exposure at increasing temperatures. D. Resonant photoemission of ZrO„ Figure 11 shows an overview photoemission spectrum taken after bombardment with 0.5keV Ar+ ions for 135 min (i.e.,steady state). Compared with the spectrum of Zr02 in Fig. 1the effects of Ar bombardment are clearly observable in all the bands including the Zr 4p, the 0 2s, the 02p, and the emission band in the gap. The Zr 4p peak is broadened, suggesting that Zr species having valence states lower than 4+ are now present. The Zr 4p/0 2s intensity ratio increases upon ion bombardment, confirming the preferential loss of oxygen. An estimation ZrOx EF hv (eV)~E 'la jW~ 60 50 of the surface composition in terms of the Zr 4p and 02s intensities gives an average stoichiometry ZrO» which is in good agreement with the average composition (-ZrO) of the altered layer caused by 3keV Ar+ ions, as determined by XPS. In Fig. 12 we show the valence-band photoemission spectra of ZrO„ in the 0— 15 eV bindingenergy region, measured at photon energies in the range 30~hv~60 eV. On the right-hand side of the figure, the band-gap emission has been depicted on an expanded scale. Again, the intensity and shape of the gap states measured at hv= 30 eV are strongly affected by the contribution of the Zr 4p core levels excited with secondorder light and will not be considered in the quantitative analysis. Effects associated with the resonant process are clearly observable and have been quantitatively analyzed in a similar manner to that for stoichiometric Zr02 in Sec. IIIB. Figure 13 shows the dependence of the photon energy with respect to the intensity of the valence band and the emission in the band gap. Again we show the behavior of the respective intensities of two Gaussians labeled as "bonding" and "nonbonding. "In this case, however, the valence-band spectra could be analyzed in terms of only two of such curves, the widths and energy positions of which are summarized in Table I. Atypical fit is shown in Fig. 4(b). The Gaussian labeled as nonbonding is assigned to the lowest-binding-energy cornponent, whereas the bonding one has been assigned to the highest-binding-energy component. The resonance profiles of Fig. 13 show abehavior similar to those measured for stoichiometric Zr02. In particular, all the profiles are characterized by two maxima around 40 eV and 45— 50 eV. Whereas the intensity of the bonding component reaches its main maximum at 40 eV, we note that the resonance profile of the nonbonding component shows abump at that energy followed by an ~10 C 8 CQ 6 g) 4- UJ z I— 0- ZrO„, hU=50eV EF 15 I 10 50 45 42 41 —40 38 37 36 ~o ~'!M~+ /0 Crw ~Q ) IIIII~I 54321 0-1-2 IIII I II 50 40 30 20 10 BINDING ENERGY (eV) FICx. 11. Photoemission spectrum measured at hv=50 eV from ZrO& after bombardment with 0.5keV Ar+ ions (135 min). BINDING ENERGY (eV) FICx. 12. Photoemission spectra of the valence band (left) and the band-gap emission (right) from ZrO„(see Fig. 11)as afunction of the photon energy. 52 ELECTRONIC STRUCTURE OF STOICHIOMETRIC AND Ar+-. . . 11 719 45— l/l 40 35— lg 30— I— 25— Z', UJ I- 20— Z'. 15— 10— 0I~I~ I ~II~I~I 30 35 40 45 50 55 60 PHOTON ENERGY (eV) FIG. 13. Photon energy dependence ofthe integrated intensity of the total valence band (~)(curve a) and the band-gap emission (~)(curve e) as well as of the bonding (A) (curve b) and nonbonding (0)(curve c) components. absolute maximum at -50eV. The resonance at 40 eV is associated with the enhancement of the Zr 4d states involved in the valence band, while the origin of the resonance at 45— 50 eV is unknown. The resonant behavior of the states in the band gap is more clearly observed in this case due to the increased intensity in this spectral region. The maximum at 40 eV clearly confirms the cationic character of these states. All the profiles show that the resonance at 40 eV has broadened significantly compared with the sharp peak observed for stoichiometric ZrOz. As in Sec. III8we calculated difference curves in order to point out those regions of the spectra which are enhanced when the energy of resonance is reached. The resulting curve obtained after subtracting the photoemission spectra measured at 36 eV (off resonance) from that measured at 40 eV (on resonance) is shown in Fig. 14. This figure also includes the XPS valence-band spectrum ZrO„ XPS '~ I'I'I'I'I'I'I't'I 14 12 10 86420-2 BINDING ENERGY (eV) FICr. 14. Zr 4d contribution to the valence band as determined from the resonances at hv=40 eV. For comparison we have included the XPS spectrum of the valence band of ZrO„ from Ref. 26. of ZrOz bombarded with 3keV Ar+ ions published in Ref. 26. According to the interpretation given above the 40/36 eV difference spectrum should reflect the Zr 4d partial density of states in ZrO; unfortunately, there are not band structure calculations for comparison. Comparing with the equivalent spectrum for ZrOz (see Fig. 6) indicates asignificant increase in the Zr 4d contribution, mainly in the low-binding-energy region of the valence band (i.e.,the nonbonding region), as aresult of the sputtering. The increase in the cationic contribution clearly suggests asignificant increase in covalency due to the preferential loss of oxygen atoms and the corresponding reduction of the cation. This effect is also confirmed by comparing the 40/36 eV difference curve, labeled Zr 4d in Fig. 14, with the XPS valence band of ZrO obtained after bombardment with 3keV Ar+ ions. Contrary to the comparison made in Fig. 6for Zr02, we observe in the case of ZrO reasonable agreement between the XPS valence-band spectrum and the Zr 4d contribution represented by the difference curve 40/36 eV. This indicates that the 02p contribution to the valence band of ZrO is quantitatively less important than for Zr02. The small differences should be accounted for by the 02p contribution. Acomparison between the analysis of the XPS valence band for Zr02 (Fig. 6) and ZrO (Fig. 14) leads to the conclusion that whereas the 02p component is significantly reduced in intensity the Zr 4d contribution increases over the width of the valence band but more strongly in the nonbonding region. Although the cationic component is probably overestimated, the data indicate that the creation of oxygen vacancies and the heavy reduction of the oxide by Ar+ bombardment causes a significant decrease in the number of ionic Zr-0 bonds and aconsiderable increase in covalency for the modified surface. IV. SUMMARY AND CONCLUSIONS The electronic structure of thermally grown Zr02 thin films before and after Ar+ bombardment was studied with resonant photoemission spectroscopy using synchrotron radiation. Good agreement has been found between the valence-band spectra of thermally grown Zr02 thin films and the calculated density of states for bulk Zr02. For both stoichiometric and Ar+-bombarded Zr02 thin films, resonant photoemission from the valence band was observed when the photon energy was tuned through the Zr 4p — +4d transition energy. Arather complex dependence of the total valence-band intensity on the photon energy was found. The resonance profile exhibits amaximum at hv= 39 eV followed by asecond broad structure at 45— 50 eV. The intensity enhancement observed at 39 eV is consistent with the resonant photoemission of Zr 4d states involved in the valence band. In fact, the phenomenon has been used to identify those regions of the valence band with an important Zr 4d character and also to estimate the degree of Zr 4d/0 2p hybridization. The second resonance at photon energies between 45 and 50 eV is associated with the nonbonding region of the valence band; however, its origin remains unexplained. In addition, we have studied the changes that occur in