1. Introduction

Ferroelectric materials are promising candidates for new devices, such as FeRAM non-volatile random access memories (Suzuki, 1995), and much attention is paid to the control of the growth of high-quality thin films of such compounds. Aurivillius phases like SrBi₂Nb₂O₉ (SBN) or SrBi₂Ta₂O₉ (SBT) (Rae et al., 1992) are specially good candidates for such devices because these materials are resistant to fatigue during cycling (Araujo et al., 1995; Scott et al., 1996), in contrast to more classical and simple compounds, such as the well known PbZr_xTi_1-xO₃ (PZT). However, they present a complex structure, built from the stacking, along the c axis, of double perovskite blocks and (Bi₂O₂) layers, leading to a large unit cell [space group A2₁am, a = 5.5189, b = 5.5154, c = 25.1124  Å for SBN (Ismunandar et al., 1996)] and a marked anisotropy. High-quality c-axis-oriented epitaxial thin films of these two compounds are routinely grown, especially on the well matched (100)SrTiO₃ perovskite substrate (Nagahama et al., 1999a; Legrand et al., 1999; Duclère et al., 2002). However, because the polarization vector P is parallel to the (Bi₂O₂) layers (Shimikawa et al., 2000), along the a-axis direction, this orientation is not suitable for the design of devices based on simple capacitor structure. Thus, it is highly desirable to control alternative orientations, taking advantage of the use of specific substrates. In-plane c-axis orientation has been reported on (110)MgO (Moon et al., 1999) and (110)SrLaAlO₄ (Legrand, 2000) substrates: the polarization vector would then be distributed towards both in-plane and perpendicular directions, because the a and b unit-cell constants are practically identical. Another approach is to tilt the c axis about 45° from the substrate surface, exactly as in the well known example of (103)YBa₂Cu₃O₇ superconductor films (Guilloux-Viry et al., 1993). For this purpose, (110)SrTiO₃ substrate presents itself immediately as a good candidate, and SBN or SBT films have been successively grown on it. However, different growth orientations have been reported, either the (116) one (Lettieri et al., 2000; Nagahama et al., 1999b) or a mixture of the (115) and (116) orientations (Garg et al., 2000).

In this study, we have used pulsed laser deposition (PLD) to grow SBN films on (110)SrTiO₃. The films have been structurally characterized in detail, especially by X-ray diffraction (XRD) in various modes (-2, scans and scans), and by their electron channeling patterns (ECPs).

2. Experimental

Thin films have been grown on 0.5  mm thick (110)SrTiO₃ (Crystal GmbH) single-crystal substrates using pulsed-laser deposition, with the same process parameters previously used for the growth of c-axis-oriented (SBN) on (100)SrTiO₃ (Duclère et al., 2002): a pulsed laser operating at a wavelength of 308  nm (SOPRA 520 Xe-Cl excimer laser) with a fluence of 3-4  J  cm^-2, a substrate holder temperature of 973  K, an oxygen pressure of 0.3 × 10²  Pa and a target-substrate distance of 4  cm. Homemade targets, obtained by the ceramic route, were 30% (in weight) Bi₂O₃ enriched, to compensate bismuth oxide volatility and then to achieve stoichiometry control, as previously reported (Duclère et al., 2001b). The SBN films reported here have a typical thickness of 250  nm.

Phase analysis and film growth direction have been determined from the XRD -2 scans recorded with a two-circle CGR Theta 2000 diffractometer working with Cu K₁ radiation (curved quartz single-crystal front monochromator). The two motions can be disconnected, an essential configuration in the case of highly textured or epitaxial films, which additionally allows the measurement of the mosaicity level by recording XRD (or ) scans. For this purpose, the lateral beam divergence can be limited down to less than 0.02° by mechanically adjustable slits.

In-plane ordering was checked in situ by reflection high energy electron diffraction (RHEED) using a Cameca electron gun working at 10  keV energy, and ex situ by electron channeling patterns (ECPs) recorded on a Jeol JSM 6400 scanning electron microscope. RHEED patterns give access to the surface periodicity along the perpendicular to the electron beam and probe typically a depth of about 2  nm. ECPs display a stereographic projection of the film planes along the pole aligned with the microscope axis, due to a systematic drop of backscattered electron yield, each time the electron beam reaches a Bragg angle for its associated wavelength (0.077  Å at 25  keV), and is a very sensitive method to probe any in-plane misorientation (Perrin et al., 1992). In-plane orientation was determined from XRD scans recorded on a four-circle Philips PW 3020 texture diffractometer.

Additional information about the microstructure of the films was obtained by high-resolution scanning electron microscopy (field-effect emission gun Jeol SEM JSM-6301F). Substrate in-plane axes were located by standard backscattered Laue photographs taken with the unfiltered radiation of copper.

Structural calculations and modeling were carried out with the CaRine software (Bourias & Monceau, 1998).

3. Results

3.2. In-plane ordering

Fig. 1 displays a typical RHEED pattern, taken along the [10]_S substrate direction. The presence of streaks is obviously the signature of in-plane ordering of the film. The h separation of these streaks is very close to that observed at the same azimuth for the substrate before the layer was grown. From the classical relation d = D/h (where D is the distance between the sample and the phosphor screen), the d-spacing associated to the probed atomic rows is 3.9  Å, i.e. the half-diagonal of the basal (a,b) plane of SBN unit cell. This result means that the film is epitaxially grown onto (110)SrTiO₃, with the relationship:

The spotty reinforcements are indicative of a quite rough surface on an atomic scale.

Figure 1 A typical RHEED pattern of a (116)SBN film, taken with the electron beam towards the [10]Substrate direction.

The epitaxial growth is clearly evidenced also by the ECP photograph displayed in Fig. 2(a). The observation of sharp stripes is indicative of high-quality in-plane ordering within the probed depth, of about a few hundred Å. Like the pattern of the (110)SrTiO₃ substrate, taken before film deposition (Fig. 2b), this pattern displays, as expected, a twofold symmetry. However, it can be noticed that the general shape of the ECP of this film is quite unusual, when compared for instance with Fig. 2(b) or with the ECPs of c-axis-grown SBN thin films (Duclère et al., 2002); ECPs usually exhibit stripes crossing at the point representative of the probed pole, perpendicular to the substrate surface. In the case reported here, direct indexation of the ECP is difficult. Thus, in order to help interpretation, the stereographic projection, along the 116 pole, of the traces of planes has been calculated. For clarity, this model has been restricted to planes nearly perpendicular to the (116) one and presenting large structure factors [for this purpose, X-ray intensities have been used instead of electron diffraction ones as the atomic scattering amplitudes for electrons are roughly proportional to the atomic scattering factors for X-rays (Lonsdale, 1962)]. The result is schematized in Figs. 3(a) and 3(b); it is in complete accord with the experimental orientation of the stripes and position of the poles shown in Fig. 2(a). Furthermore, the stripe widths, a measurement of the value of the Bragg angle associated with a given plane, have been calculated from structural data; they are in full agreement with the widths measured from the angular scale reported on the photographs, e.g. : calculated 2.28°, measured 2.29°; : calculated 1.44°, measured 1.46°.

Figure 2 Wide-angle electron channeling patterns of: (a) a (116)SBN film with single family, (b) the (110)SrTiO3 substrate (where the principal stripes are indexed), and (c) a (116)SBN film with two families.

Figure 3 (a) Modelled stereographic projection of SBN along the 116 pole, where the angular field of 25  keV ECP is delimited by the dashed circle. (b) An enlargement of the latter, with the indexation of the traces of the planes and selected poles. For reference, the 001 pole of the substrate is shown in (a).

From comparison of the patterns before and after film deposition, taken at the same azimuth, it appears that

and, taking into account the (116) orientation, the following in-plane relationships are deduced:

(refer also to RHEED results), as well as the alternative [110]_Film || [001]_Substrate, because the a and b unit-cell constants are in practice non-discriminable and thus equivalent at this level, and, along the perpendicular in-plane direction,

These relations mean that, referring to the substrate as reference, the standard unit-cell of SBN is rotated by 45° and then tilted by about 45° relative to the substrate surface, as schematized in Fig. 4.

Figure 4 A scheme of the bi-epitaxial growth of the two families, labelled A and B: (a) the model (116)F || (110)S; (b) growth on {100}S facets. For clarity, the half-volume subcell based on SBN perovskite blocks (meaning a and b = aSBN21/2/2 and bSBN21/2/2) has been drawn.

However, the ECP shown in Fig. 2(a) is not actually representative of all our films, and we have also observed characteristic patterns like Fig. 2(c). It is clear that this pattern is the superimposition of two individual patterns of the type of Fig. 2(a), rotated by 180° with respect to each other, implying a mirror plane containing the [001]_S substrate axis. This behaviour is the signature of a twin and is indicative of the bi-epitaxial growth of two crystallite families, labelled A and B in Fig. 4. In fact, this situation was clearly expectable from the symmetry of the two materials, because the [10]_S and [10]_S directions of the substrate are obviously strictly identical, as well as the [33]_F and [1]_F directions of the film, and thus the two in-plane orientations are a priori equiprobable.

Situations intermediate between those corresponding to Figs. 2(a) and 2(c) are often encountered. XRD -scan measurements have been carried out in order both to confirm the epitaxial relationships and to derive a more quantitative distribution of the two families. The probed planes were (200)_S for the substrate (2 = 46.48°; = 45°) and (00 10)_F for the film (2 = 35.725°; = 47°). Fig. 5 summarizes the results. Fig. 5(a) refers to the substrate diffraction and shows, as expected, two intense peaks at azimuths 180° apart from each other. Fig. 5(b) is the XRD scan of the film corresponding to Fig. 2(a). It shows one intense peak (labelled A), at the same azimuth as that of the peaks of the substrate, as expected from geometrical considerations and the prominence of one family. But a weak peak emerges at 180° azimuth, corresponding to a small amount of the B family. The intensity ratio of the two peaks gives directly the volume ratio of the two families, because the same structure factor is involved. The distribution is about 90/10, which gives an effect that is hardly discernible on ECPs, as in the pattern of Fig. 2(a). In contrast, Fig. 5(c), relative to the film corresponding to Fig. 2(c), shows two peaks of similar intensity, with a ratio close to 50/50, again in good agreement with qualitative ECP results. Finally, Fig. 5(d) is an illustration of an intermediate situation, where the volume ratio of the A and B families is about 70/30.

Figure 5 XRD patterns in texture mode ( scans) of (a) the substrate as reference, (b) a single-family (116)SBN grown on the substrate, (c) a two-equiprobable-families film, and (d) an intermediate case. Note that the origin is arbitrary and azimuthal comparison is significant only for patterns (a) and (b).

3.3. Morphology

Fig. 6 is a typical field-effect gun scanning electron microscope photograph of the thin films discussed in this report. The grains are quite large, facetted and strongly anisotropic: they are elongated along the substrate direction [001]_S. The roughness is noticeable and explains the dotty RHEED patterns. The morphology of these (116)SBN films is very similar to that of the (103)YBa₂Cu₃O₇ superconductor films (Rossel et al., 1994) which present a comparable growth orientation.

Figure 6 A typical SEM photograph (field effect gun instrument) of the surface of a (116)SBN film.

4. Discussion

All the above results agree that our SBN films, grown by PLD on (110)SrTiO₃ substrates, are epitaxial and (116) oriented. The c axis of the SBN structure is tilted by about 45° with respect to the substrate surface, a situation that is very similar to the example of (103)YBa₂Cu₃O₇ films grown on the same (110)SrTiO₃ substrate (Guilloux-Viry et al., 1993). However, in the case of the latter cuprate, the occurrence of two equally distributed families is almost systematically observed. In fact, it has been shown that, in this example, a vicinal substrate (about 2° miscut angle) is able to favour one given family markedly (Guilloux-Viry et al., 1993). Such a mechanism could be considered in the case of our (116)SBN films, but all the substrates appeared to be cut parallel to the (110)SrTiO₃ plane, at least within the accuracy of backscattering Laue diffraction photographs. Therefore, up to now, no correlation can be established between the volume ratio of the two families and substrate vicinality.

This addresses the question of the actual interface between the two materials, on the atomic scale. Two growth models can be accounted for, depending on the actual terminal plane of the substrate, (110) plane or {100} facets, which could theoretically be differentiated from -scan experiments carried out on a sample containing the two families A and B, and oriented with [10]_Film parallel to the diffractometer axes.

For the first model, there is a quite small local mismatch along the [001]_S and the [10]_S substrate directions (nearly 0 and 5.7%, respectively), but it is accompanied by low atomic density and a long-range incoherent interface along [10]_S. In this model, the rocking curve of the 22 12 peak would be unique, while that relative to the 115 peak would be split in two components separated by 10.30°, i.e. twice the (116)(115) angle.

The second hypothesis, based on the presence of a {100} facetted terminal surface, appears more convincing, because it corresponds to a direct matching of the perovskite blocks of SBN with the perovskite structure of the substrate, the local interface being the (100) plane for the two materials, a feature which reproduces locally, on each facet, the growth model of the standard c-axis-oriented SBN films on (100)SrTiO₃, with a very small misfit (0.15%). In this case, taking into account the 47.01° angle between the (116) and (001) planes of SBN, the rocking curves of the 115 and 22 12 peaks would exhibit two components, split by 2(5.15 + 2.01) = 14.32° and 2(47.01 - 45) = 4.02°, respectively.

Fig. 7 reports our experimental data. The rocking-curve splittings are 10.75 and 0.4°, respectively. This result favours the first model, although there is a small but significant angular deviation between the (110)SrTiO₃ and the (116)SBN planes: 0.2° (to be compared to the accuracy of our measurements, 0.02°). This result contrasts with data reported by Zurburchen et al. (2001), who concluded on the second model. However, they pointed out that their substrates were facetted, as a result of thermal treatment.

Figure 7 XRD rocking curves of a film with two nearly equivalent families, relative to the (a) 115 and (b) 22 12 reflections.

Finally, Suzuki et al. (1999) have reported an HRTEM study carried out on a cross-sectioned thin film of the tantalum SBT isostructural compound, also grown on (110)SrTiO₃. They observed a deviation of about 2° between the directions [100]_S and [001]_F, corresponding exactly to the first hypothesis suggested above, namely the coincidence of (116)_F and (110)_S, although the displayed images show that their substrate is clearly facetted.

From all the above observations, the growth mechanism could be postulated as follows: the film could nucleate either on the {100}_S substrate facets, or on the (110)_S plane. Surface-energy calculations suggest that in fact the (100) and (110) faces of SrTiO₃ would have very close energy (Woensdregt, 2001) and thus the prominence of one of these effective terminal planes would be related to very subtle parameters, which may be the explanation of the different observations. In both cases, the structure could be forced to bend at the early stages of growth by a twinning effect, the twin boundary (schematized in Fig. 8) being the (11)_F plane as reported by Suzuki et al. (1999). In fact, the latter plane has a significant atomic density and makes a remarkable angle of 90.4° with (116)_S. Then, the driving force of the final orientation [i.e. the nearly (116) orientation] of the bulk thin film would be the twinning of the SBN material, with the twin boundary normal to the substrate surface. Due to this mechanism, only the final orientation of the film can be derived from XRD, not the initial nucleation and growth at the interface.

Figure 8 Scheme of the twin boundary of the two families A and B viewed along the [001]S substrate direction, highlighting the importance of the (11) plane. Oxygen atoms have been removed for clarity.

5. Conclusion

In summary, PLD gives access to high-quality epitaxial thin films of the ferroelectric material SrBi₂Nb₂O₉ on (110)SrTiO₃ substrates. These films are nearly (116) oriented, as shown by several complementary methods of structural characterization, namely -2 XRD, ECP and -scan texture XRD analysis. The two latter techniques have evidenced the epitaxial growth of the films and shown the coexistence of two crystallite families, rotated by 180° with respect to each other, and which can coexist in various proportions. However, the driving force determining the relative weight of these two families appears very subtle and is still unknown.

The nearly (116) orientation allows a non-zero component of the polarization vector along the normal to the substrate, which is a good characteristic for future application in capacitor-like devices. The next step will be the growth of heterostructures including a bottom electrode compatible with the structures of both the substrate and SBN: possible candidates are platinum and (103)YBa₂Cu₃O₇, which can be grown efficiently on (110)SrTiO₃. The corresponding heterostructures have been grown recently, but secondary-ion mass spectrometry (SIMS) measurements suggest some interdiffusion in the two systems (Duclère et al., 2001b,c), which can be strongly reduced, at least in the case of a (110)Pt electrode, provided that the deposition temperature is strictly controlled.