X-ray photoemission spectroscopy and scanning tunneling ... - INFN

X-ray photoemission spectroscopy and scanning tunneling spectroscopy study on the thermal stability of WO3 thin films S. Santuccia) Department of Physics and Unita´ INFM, University of L’Aquila, 67100 L’Aquila, Italy

C. Cantalini Department of Chemistry and Materials, 67040 Monteluco di Roio, L’Aquila, Italy

M. Crivellari, L. Lozzi, L. Ottaviano, and M. Passacantando Department of Physics and Unita´ INFM, University of L’Aquila, 67100 L’Aquila, Italy

共Received 29 September 1999; accepted 6 December 1999兲 In this work the surface electronic and structural properties of about 150 nm thick WO3 films, deposited in high vacuum by thermal evaporation onto Si substrates, have been studied in ultrahigh vacuum 共UHV兲 by means of x-ray photoemission spectroscopy 共XPS兲 and scanning tunneling microscopy 共STM兲/spectroscopy. After deposition these films have been annealed in atmospheric oven for 24 h at different temperatures 共300 and 500 °C兲 to stabilize the film morphology. XPS measurements to follow W 4 f , O 1s peaks and the valence band, have been performed on these samples both as prepared and after a re-annealing in UHV at temperatures ranging from 50 to 350 °C. The UHV re-annealing procedure strongly modifies the W 4 f peak of both the as deposited and the 300 °C/24 h treated samples, and produces an increase of metallic states at the Fermi edge. Instead, the 500 °C/24 h sample, after heating in UHV shows substantial stability of the nearly stoichiometric WO3 phase. Using STM in UHV we have investigated the morphology of the samples at room temperature and after the annealing at elevated temperatures up to 350 °C. In particular, we have taken I – V curves on typical grains of the polycrystalline sample. Our findings on the electronic structure of samples close to the Fermi level are in agreement and allow a clearer understanding of the findings from the parallel XPS study. © 2000 American Vacuum Society. 关S0734-2101共00兲01504-9兴

I. INTRODUCTION Tungsten trioxide (WO3) has been studied for a long time because of its electro-optic, electrochromic, ferroelectric, and semiconducting properties.1–3 More recently tungsten trioxide has been used as a sensing material in metal–oxide– semiconductor 共MOS兲 gas sensors. The interactions of gases such as H2S, NO, and NH3 with sintered powder,4 thick film,5 and thin film6–12 of WO3 induce changes in electrical conductivity. Thin sensing films are attractive because they can be fabricated economically using existing planar technology while maintaining precise control of film properties. The microstructure and morphology of metal oxide films are believed to have a large influence on the sensitivity, selectivity, and stability of the sensor.13,14 However, a direct relationship between film microstructure and gas response has not yet been clearly established in the WO3 system. Fabrication of WO3 films with well-defined microstructure, crystallinity, and orientation is required in order to study how microstructure influences the electronic properties of this class of materials. Tungsten trioxide films have been synthesized by many methods including sol gel,15 chemical vapor deposition,16 and physical vapor deposition.7,10,17,18 Our approach is to fabricate WO3 films using low deposition rates 共0.05 nm/s兲 in a high vacuum 共HV兲 environment (10⫺7 Torr) on silicon substrates using thermal evaporation a兲

Electronic mail: [email protected]

1077

J. Vac. Sci. Technol. A 18„4…, JulÕAug 2000

and submit them to thermal treatments in oxygen in order to study their microstructural stability. In previous work19 we showed by scanning tunneling microscopy 共STM兲 measurements how the granular structure of these films is modified by thermal annealing in oxygen. The occurrence of electronic states near the Fermi edge, in the bulk band gap, on WO3 and on other semiconductor oxides is generally attributed to oxygen deficiencies and can be divided in two categories: the first is intrinsic or bulk oxygen deficiencies, i.e., during preparation the WO3 film becomes amorphous and deficient in oxygen.20 The thermal treatment up to 300 °C in oxygen induces recovery of the correct stoichiometry and crystallization in the monoclinic phase, the stable phase at room temperature. At higher temperatures, with small changes in the atomic positions as well as in unit cell dimensions and angles, a transition to the orthorhombic phase is observed.1 The second category is electronic states at oxygen deficient surfaces. These states which may again be intrinsic on the surfaces of WO3 single crystals may be more densely induced by electron or ion bombardment21 and in the case of thin amorphous and polycrystalline films may be dependent on the growth conditions and postgrowth treatments. But, to our knowledge, little work has been done in the past on the generation of surface states in thin WO3 films induced by thermal treatments in ultrahigh vacuum 共UHV兲. The present study is focused on studying how the in vacuum thermal

0734-2101Õ2000Õ18„4…Õ1077Õ6Õ$17.00

©2000 American Vacuum Society

1077

1078

Santucci et al.: XPS and STM study on the thermal stability of WO3

treatment modifies the electronic states at the surface of thermally evaporated and in air annealed thin WO3 films by looking at the strong influence that these states may have in the electronic transport properties and at the behavior of these films towards the interacting gases. To conduct this study we used high resolution x-ray photoemission spectroscopy 共XPS兲 and in vacuum scanning tunneling spectroscopy 共STS兲 techniques. On these samples we have followed, as a function of the UHV annealing temperature, the evolution of the W 4 f and of the valence band 共VB兲 states by means of XPS spectroscopy, and of the I – V curves, acquired by means of the STS technique. The variation of the density of states, as a function of the annealing temperature, obtained by both these techniques has been compared.

1078

chemical etching with an electronic control optimized to minimize the tip apex radius.22 No further in vacuum treatment was necessary except for occasional in situ cleaning performed by operating the tip in the field emission regime 共⫾10 V, 50 nA兲 for several seconds on a fixed point of the sample. The STS measurements were performed by taking I – V curves on a 100⫻100 pixel grid on a scanned frame about 200⫻200 nm wide. Thus each I – V spectrum presented in this article is the average of 10 000 curves. The measurements were taken on the samples cooled down to room temperature after annealing in situ on the STM stage by direct resistive heating of the Si共100兲 substrate of the WO3 thin film. The temperature was determined according to calibration curves for direct heating of Si samples of fixed thickness, width, and height.23

II. EXPERIMENT A. Sample preparation

WO3 films, about 150 nm thick, were deposited by high vacuum thermal evaporation onto a Si substrate, whose native SiO2 thin film had been removed by HF etching. The pressure during the evaporation was about 1⫻10⫺6 Torr and the deposition rate, controlled by a quartz microbalance, was about 0.1 nm/s. The samples were then annealed at different temperatures 共300 and 500 °C兲 for 24 h in an atmospheric oven. The increasing and decreasing heating ramps were 1 °C/min. In the following these samples will be called 300 °C/24 h and 500 °C/24 h samples, respectively. They will be compared with the as deposited one, that is, the sample that has not been annealed in the oven after preparation. B. XPS measurements

The XPS measurements were performed in a custom UHV apparatus equipped with a monochromatic Al x-ray source and a hemispherical analyzer. The overall resolution, with about 5 eV pass energy, was about 0.2 eV, as determined at the Ag Fermi edge measured at room temperature. All the reported binding energy data have been calibrated using both the Au 4 f 7/2 peak 共84.0 eV兲共a small gold dot has been deposited, for XPS calibration, onto each sample兲 and the residual carbon present on the surface, positioned at 285.0 eV. The carbon present is graphitic and it is not bonded either to oxygen and to tungsten. The XPS measurements were performed on these three samples both as prepared, i.e., after insertion into the UHV chamber, and after 1 h of UHV annealing at different temperatures and subsequent cool down to room temperature. The pressure during the UHV heating procedure grew to 5⫻10⫺8 Torr. The heating temperature was checked by a thermocouple fixed on the back side of the sample. C. STMÕSTS measurement

The STM/STS measurements were carried out with the VT100 STM system from Omicron VakuumPhysik. The system operates in UHV with a base pressure of 1 ⫻10⫺10 Torr. The STM tips were prepared via dc electroJ. Vac. Sci. Technol. A, Vol. 18, No. 4, JulÕAug 2000

III. RESULTS AND DISCUSSION A. XPS

In Fig. 1共a兲 the W 4 f peak of the as deposited sample is shown. On going from the bottom the W 4 f spectra, as a function of the UHV annealing temperature, are shown. For the measurement at room temperature a well resolved double peak is present due to 4 f 7/2 and 4 f 5/2 . The position of the highest one (4 f 7/2) is at about 35.8 eV. This energy is compatible with the presence of the WO3 compound.24 The broad structure at about 41 eV is due to the 5p 3/2 . By increasing the UHV annealing temperature an evident shoulder in the low binding energy side, located at around 33.6 eV, appears and, at the same time, there is a general broadening of the other features. A stable W 4 f shape is obtained starting from a 300 °C annealing temperature; after that no sizeable variation increasing the annealing temperature up to 700 °C has been observed 共not reported here兲. In Figs. 1共c兲 and 1共e兲 the W 4 f spectra recorded at different sample temperatures on the 300 °C/24 h and 500 °C/24 h samples are shown. The 300 °C/24 h sample already shows in the bottom spectrum 关room temperature 共RT兲 measurement兴 the presence of a small bump in the low binding energy side 共located at 34.5 eV兲 evidently not affected by the thermal treatments, while the structure at 33.6 eV evolves with the annealing temperature in the same way as the as deposited sample. For this sample all spectra show structures that are less broadened than that in the as deposited sample. Instead, for the 500 °C/24 h sample, very few variations as a function of the annealing temperatures are observed 关Fig. 1共e兲兴. Only a slight increase of the tail in the low binding energy side 共34.5 eV兲 and of the width of the peaks is observable, while no evidence of the 33.6 eV feature observed in the other two samples was found. The presence of low binding energy features has been already observed in crystalline WO3 submitted to Ar⫹ bombardment21 or in oxidized tungsten foils.24 These new features were attributed to different final states, that is 5d 1 共33.5 eV兲 and 5d 2 共32.1 eV兲, which correspond to W5⫹ and W4⫹ nominal oxidation states21 and also to W2⫹, that is, the WO compound 共32.1 eV兲,24 while the main peak 共35.8 eV兲 is due to the 5d 0 final state 共corresponding to W6⫹ oxidation

1079


1079

FIG. 1. XPS spectra as a function of the UHV annealing temperature of 共a兲 W 4 f and 共b兲 valence band of the as deposited sample; 共c兲 W 4 f and 共d兲 valence band of the 300 °C/24 h sample; 共e兲 W 4 f and 共f兲 valence band of the 500 °C/24 h sample.

state兲. These binding energy values are influenced both by the chemical state and by the screening of the hole after ionization. This screening can change from metallic type to dielectric type and seems to depend on the oxygen JVST A - Vacuum, Surfaces, and Films

concentration.24 The features in the low binding energy side of our spectra are compatible with the presence of a 5d 1 final state, while there are no indication of other oxidation states. In Figs. 1共b兲, 1共d兲, and 1共f兲 the valence band spectra of

1080


the as deposited, 300 °C/24 h and 500 °C/24 h samples as a function of the UHV annealing temperature are reported. These spectra can be divided in three parts: the first one is very close to the Fermi edge and it is due to the tungsten 5d states, the second one between 2 and 10 eV binding energy contains contributions from both W 5d and O 2p states, and the third part, at about 22 eV binding energy, is determined by O 2s states.21,25 All the spectra have been normalized to the O 2s peaks. The most important variations are observed in the as deposited 关Fig. 1共b兲兴 and 300 °C/24 h 关Fig. 1共d兲兴 samples. Upon going from the room temperature measurement up to 300 °C the increase of the feature at the Fermi level, that is, the W 5d states, is clearly observable. In the crystalline WO3 sample, a semiconductor with a band gap of about 2.6 eV,21 the 5d states are empty and no states are observed at the Fermi level.25 We interpret the presence of these states at the Fermi level in our samples as due to missing oxygen atoms and, therefore, to a reduction of tungsten. This reduction is in agreement with the evolution of W 4 f signals. For our samples the evolution of both the 4 f level and of the valence states 共and in particular the states close to the Fermi edge兲 increasing the annealing temperature is completely different from that observed on WO3 crystalline samples submitted to Ar⫹ bombardment21 which also produces the presence of low binding energy side components in the valence band spectra. In such work successive UHV annealing at about 580 °C restored the initial surface oxygen concentration almost completely, and with the tungsten metallic peak in the valence band close to the Fermi level disappeared. Instead, in our case, we observed a growing formation of reduced metallic tungsten states with increasing UHV annealing temperature both for the as deposited and for the 300 °C/24 h samples while substantial stability was observed for the 500 °C/24 h sample which presents a small broad structure at the Fermi level that is not affected by the UHV annealing. The different behavior in the valence band evolution induced by the thermal treatments in UHV of our samples with respect to the WO3(100) crystal 共used in Ref. 21兲 is probably due to the different morphologies of the sample surfaces. On the flat surface of the crystalline sample the recovered oxygen may come away from the bulk of the sample. Instead, as previously reported,19 our samples show a morphology composed of grains whose size depends on the annealing temperature in the atmospheric oven. In fact, the grain size remains at around 50 nm for the as deposited and 300 °C/24 h samples while the 500 °C/24 h sample shows coalesced grains having well rounded shapes with mean sizes of over 150 nm. The increase of the W 5d states at the Fermi level may be due to the release of oxygen adsorbed at the grain boundaries of the first two samples while the 500 °C/24 h sample, which experienced a monoclinic to orthorhombic phase transition, presents only a weak and stable presence of W 5d states due to its intrinsic defective structure, not affected by temperature. B. STS

In Fig. 2 we show the STS spectra acquired on the three differently prepared samples. In particular the curves on the J. Vac. Sci. Technol. A, Vol. 18, No. 4, JulÕAug 2000

1080

left side are the as acquired I-V spectra, while those on the right side are the corresponding calculated normalized conductance curves. According to the work of Feenstra et al.26 the (dI/dV)/(I/V) experimental data give direct information on the surface local DOS 共LDOS兲 of the sample under investigation. It is worth recalling that for a metallic surface 共ohmic behavior at the Fermi level兲 the normalized conductance curve is symmetric with respect to the Fermi level and reaches its minimum value 共unity兲 at 0.0 V.26 Thus the value of unity can be regarded as a threshold value between metallic behavior and semiconducting, and can be used as a reasonable parameter to estimate the band gap from STS normalized conductance spectra. The information that can be derived is rich and comparison with XPS is also enlightening. There are some qualitative trends that can undoubtedly be evidenced. In general the spectra measured at RT and after annealing the samples in UHV at 100 and 150 °C 共black, green, and red curves, respectively in Fig. 2兲 show the same behavior and group together. In particular, in the 300 °C/24 h sample the black, green, and red I-V curves are almost superimposed. The slight differences that can be observed in the current values can be hardly assigned to thermal effects, since the RT curve 共black兲 lies between the 100 °C 共green兲 and the 150 °C 共red兲 ones, indicating the absence of any trend. Correspondingly, the normalized conductances 共black, red, and green兲 have the same overall shape with a marked symmetric minimum around E f . The band gap can be estimated from these curves to be about 0.8 eV. The behavior of the I-V spectra measured again at RT, 100, and 150 °C in the 500 °C/24 h sample is qualitatively strikingly similar. There, the black, red, and green curves show very slight quantitative differences and the corresponding normalized conductances 共in the lower right panel of Fig. 2兲 almost overlap. In this case the gap is wider and can be estimated to be about 1.2 eV. Likewise, in the upper right panel of Fig. 2 one can notice that the normalized conductances relative to 150 and 200 °C for the as deposited sample also show a gap having the same width as that observed at RT, 100, and 150 °C in the 300 °C/24 h sample. Thus, we can affirm that samples which have been treated in UHV with mild annealing show a band gap in their STS spectra. This gap opens when passing from the 300 to the 500 °C/24 h samples. The opening of the gap can be ascribed to an overall increase in the quality of the film. Besides, a structural phase transition occurs at 290 °C and this can also explain the observed narrowing of the gap. A remarkable change is observed in all the I-V spectra measured after strong annealing of the samples 共above 200 °C兲 in UHV. The major feature in the I-V spectra is the sudden variation of the current when crossing the Fermi level 共0 V bias voltage兲. This can especially be noticed in the 250 and 350 °C curves 共pink and yellow兲 in the as deposited sample 共STS spectra of the upper two panels兲. The two curves are symmetric, and they closely resemble a step function especially with regard to the 350 °C I-V curve. In particular, in the 250 °C curve 共the as deposited sample, upper left panel兲 the I-V curve deviates from a pure step line shape

1081


1081

FIG. 2. I-V spectra and normalized conductances (dI/dV)/(I/V) of the as deposited 共upper panel兲, 300 °C/24 h 共middle panel兲 and 500 °C/24 h 共lower panel兲 samples as a function of the UHV annealing temperature.

outside the 共⫺2, 2兲 eV range and recovers the behavior of the blue 共200 °C兲 I-V curve; the saturation current in this bias range is about 0.8 nA 共positive or negative兲. The yellow curve instead is merely step like, with a saturation current 共positive or negative兲 of 4.2 nA. These I-V spectra indicate the appearance of states symmetrically populating a narrow energy window centered at the Fermi level. Within that energy range the I-V curves follow purely ohmic behavior 共linear dependence of the current with voltage兲. This peak of the DOS at the Fermi edge can be singled out in the two norJVST A - Vacuum, Surfaces, and Films

malized conductance curves reported correspondingly in the upper right panel. Qualitatively this occurrence can be also verified in the 300 °C/24 h sample 共middle panels, blue, pink, and yellow curves, 200, 250, and 350 °C respectively兲. Here the saturation current increases from the 200 to the 250 °C spectra and then decreases with a further increase of the UHV annealing temperature to 350 °C. The corresponding normalized conductance curves all show the same symmetric structure at zero bias 共the Fermi level兲. The quantitative differences in

1082


this set of three I-V curves can be explained as follows: with an increase in the annealing temperature from 200 to 250 °C there is likely a corresponding increase of the DOS at the Fermi level, with a further increase of the temperature likely a partial structural phase transition is occurring in the WO3 film with a certain percentage of crystallites changing their phase. Accordingly, as comparison with the (dI/dV)/(I/V) curves of the lower right panel suggest, the normalized conductance curve is somewhat of a mixture of those typical of the 500 °C/24 h sample with the presence of the peak of the DOS at the Fermi level. Qualitatively, the same behavior is maintained in the 500 °C/24 h sample after UHV annealing at 200 °C 共blue curve in the lower left panel兲. We can thus affirm that, regardless of the sample preparation procedure, a strong postannealing in UHV 共above 200 °C兲 induces the appearance of states in a narrow energy window centered at the Fermi level. The I-V curves assume the shape of a step function with a saturation current value which is dependent on the sample preparation and on the postannealing temperature. It is worth noting that, especially in the case of the as deposited sample, fine tuning of the postannealing temperature can yield whatever saturation current at least in the range observed 共1–4 nA兲. The possible utilization of this procedure for the implementation of electronic devices should be surely addressed in further studies. As we already mentioned, the normalized conductance spectra show good qualitative agreement with the VB XPS data. A comparison can be carried out in the 共⫺4, 0兲 eV range. The observed appearance of the peak near the Fermi edge in the XPS spectra is clearly related to the step in the I-V curves of Fig. 2. STS gives additional information because it shows the presence of a symmetric contribution of the sample DOS above the Fermi edge. Our fundamental understanding of the electronic properties around the Fermi level of the WO3 samples can be complemented by a more ‘‘device oriented’’ point of view. In fact, for polycrystalline samples of WO3 关or ZnO, TiO2, BaTiO3, 共Ref. 27兲兴 the so called Schottky barrier model is widely accepted for the description of the electrical behavior dictated by the presence of grain boundaries. IV. CONCLUSIONS WO3 thin films prepared by high vacuum thermal evaporation onto silicon substrate and treated for 24 h in oxygen at various temperatures and re-annealed in UHV present evidence of an increase of metallic states on the surface starting from UHV annealing temperatures higher than 150 °C. Only the sample treated in air at 500 °C is not affected by the UHV annealing, confirming that it is this treatment that produces its stability and as a consequence the reproducibility of its electrical properties when employed as a sensing element

J. Vac. Sci. Technol. A, Vol. 18, No. 4, JulÕAug 2000

1082

for gas sensors.12 Furthermore, the possibility, shown in this work, to tune by UHV thermal annealing metallic states on the surface of samples annealed in oxygen at temperatures lower than 500 °C is intriguing and the study both towards improving the gas sensitivity and selectivity of these new WO3 ‘‘surfaces’’ and towards developing new devices applications will be the subject of further experimental work. 1

E. Salje, Acta Crystallogr., Sect. A: Cryst. Phys., Diffr., Theor. Gen. Crystallogr. 31, 360 共1975兲. 2 Ph. Labbe, in Diffusionless Phase Transitions and Related Structures in Oxdes, edited by C. Boulesteix 共Trans Tech, Zurich, Switzerland, 1992兲. 3 A. Hjelm, C. G. Granqvist, and J. M. Wills, Phys. Rev. B 54, 2436 共1996兲. 4 H. Lin, C. Hsu, H. Yang, P. Lee, and C. Yang, Sens. Actuators B 22, 63 共1994兲. 5 M. Ando, S. Suto, T. Suzuki, T. Tsuchida, C. Nakayama, N. Miura, and N. Yamazoe, Chem. Lett. 5, 335 共1994兲. 6 A. Bryant, M. Poirier, G. Riley, D. L. Lee, and J. F. Vetelino, Sens. Actuators 4, 105 共1983兲. 7 Z. Xu, J. F. Vetelino, R. Lee, and D. C. Parker, J. Vac. Sci. Technol. A 8, 3634 共1990兲. 8 D. J. Smith, J. F. Vetelino, and R. S. Falconer, Sens. Actuators B 13, 264 共1993兲. 9 I. Ruokamo, T. Ka¨nkka¨inen, J. Huusko, T. Ruokanen, M. Bolmberg, and H. Torvela, Sens. Actuators B 18–19, 486 共1994兲. 10 M. D. Antonik, J. E. Schneider, E. L. Wittman, K. Snow, and R. J. Lad, Thin Solid Films 256, 247 共1995兲. 11 B. Fru¨hberger, M. Grunze, and D. J. Dwyer, Sens. Actuators B 31, 167 共1996兲. 12 C. Cantalini, H. T. Sun, M. Faccio, M. Pelino, S. Santucci, L. Lozzi, and M. Passacantando, Sens. Actuators B 31, 81 共1996兲; 35, 112 共1996兲. 13 M. J. Madou and S. R. Morrison, Chemical Sensing with Solid State Devices 共Academic, San Diego, 1989兲. 14 H. T. Sun, C. Cantalini, L. Lozzi, M. Passacantando, S. Santucci, and M. Pelino, Thin Solid Films 287, 258 共1996兲. 15 C. Cantalini, M. Z. Atashbar, Y. Li, M. K. Ghantasala, S. Santucci, W. Wlodarski, and M. Passacantando, J. Vac. Sci. Technol. A 17, 1873 共1999兲. 16 N. Djourelov, D. Gogova, and M. Misheva, Thin Solid Films 347, 302 共1999兲. 17 L. J. LeGore, O. D. Greenwood, J. W. Paulus, D. J. Frankel, and R. J. Lad, J. Vac. Sci. Technol. A 15, 1223 共1997兲. 18 S. Santucci, L. Lozzi, M. Passacantando, S. Di Nardo, A. R. Phani, C. Cantalini, and M. Pelino, J. Vac. Sci. Technol. A 17, 664 共1999兲. 19 S. Santucci, L. Lozzi, M. Passacantando, L. Odorisio, and C. Cantalini, J. Vac. Sci. Technol. A 17, 1639 共1999兲. 20 S. Lee, H. M. Cheong, J. Zhang, A. Mascarenhas, D. K. Benson, and S. K. Deb, Appl. Phys. Lett. 74, 242 共1999兲. 21 R. A. Dixon, J. J. Williams, D. Morris, J. Rebane, F. H. Jones, R. G. Egdell, and S. W. Downess, Surf. Sci. 399, 199 共1998兲. 22 J. P. Ibe, P. P. Bey, Jr., S. L. Brandow, R. A. Brizzolara, N. A. Burnham, D. P. DiLella, K. P. Lee, C. R. K. Marrian, and R. J. Colton, J. Vac. Sci. Technol. A 8, 3570 共1990兲. 23 The Variable Temperature STM User’s Guide, Ver. 1.3, 5 May 1999, Omicron Instruments, p. 95.3. 24 F. J. Himpsel, J. F. Morar, F. R. McFeely, R. Pollak, and G. Hollinger, Phys. Rev. B 30, 7236 共1984兲. 25 R. D. Bringans, H. Ho¨chst, and H. R. Shanks, Phys. Rev. B 24, 3481 共1981兲. 26 R. M. Feenstra, J. A. Stroscio, and A. P. Fein, Surf. Sci. 181, 295 共1987兲. 27 H. T. Sun, L. Y. Zhang, and X. Yao, J. Am. Ceram. Soc. 76, 1150 共1993兲.