Pii: S0167-5729(98)00005-3

Experimental determinations of the atomic structure of insulating oxide surfaces and metal/oxide interfaces are scarce, because surface science techniques are often limited by the insulating character of the substrate. Grazing incidence X-ray scattering (GIXS), which is not subject to charge effects, can provide very precise information on the atomic structure of oxide surfaces: roughness, relaxation and reconstruction. It is also well adapted to analyze the atomic structure, the registry, the misfit relaxation, elastic or plastic, the growth mode and the morphology of metal/oxide interfaces during their growth, performed in situ. GIXS also allows the analysis of thin films and buried interfaces, in a non-destructive way, yielding the epitaxial relationships, and, by variation of the grazing incidence angle, the lattice parameter relaxation along the growth direction. On semi-coherent interfaces, the existence of an ordered network of interfacial misfit dislocations can be demonstrated, its Burger's vector determined, its ordering during in situ annealing cycles followed, and sometimes even its atomic structure can be addressed. Careful analysis during growth allows the modeling of the dislocation nucleation process. This review emphasizes the new information that GIXS can bring to oxide surfaces and metal/oxide interfaces by comparison with other surface science techniques. The principles of X-ray diffraction by surfaces and interfaces are recalled, together with the advantages and properties of grazing angles. The specific experimental requirements are discussed. Recent results are presented on the determination of the atomic structure of relaxed or reconstructed oxide surfaces. A description of results obtained during the in situ growth of metal on oxide surfaces is also given, as well as investigations of thick metal films on oxide surfaces, with lattice parameter misfit relaxed by an array of dislocations. Recent work performed on oxide thin films having important physical properties such as superconductivity or magnetism is also briefly reviewed. The strengths and limitations of the technique, such as the need for single crystals and surfaces of high crystalline quality are discussed. Finally, an outlook of future prospects in the field is given, such as the study of more complex oxide surfaces, vicinal surfaces, reactive metal/oxide interfaces, metal oxidation processes, the use of surfactants to promote wetting of a metal deposited on an oxide surface or the study of oxide/liquid interfaces in a non-UHV environment. @ 1998 Elsevier Science B.V. All rights reserved 1 Tel.: +33 4 76 88 35 58; fax: +33 4 76 88 51 38; e-mail: [email protected]. 2 Laboratory associated with the Joseph Fourier University of Grenoble. 0167-5729/98/$ see front matter © 1998 Elsevier Science B.V. All rights reserved PII: S0167-5729(98)00005-3 6 G. Renaud/Surface Science Reports 32 (1998) 1-90 1 . I n t r o d u c t i o n Oxide surfaces [1-5] and metal/oxide interfaces [6-10] are involved in various technologically important areas such as composites, protective coatings, thin film technology, electronic as well as nuclear combustible and waste packaging, heterogeneous catalysis, gas sensors and the glass industry. Interfaces where either the metal or the oxide has magnetic properties are of growing interest, since they may find applications in future devices for magnetic recording. The electrical, mechanical, chemical or thermal properties of many technologically important devices are often intimately related to the structure, composition and morphology of internal metal/oxide interfaces, which in turn depend on the structure of the oxide surface. It is now a widely accepted idea [11] that scientific computing will become a major research field in the future. It will allow predicting the properties of materials before they are elaborated, and thus development of new technological materials with specific properties may be expected. However, before this is achieved, the theoretical models used in numerical simulations need to be very detailed and have proven accuracy. Whereas theoretical models allowing the prediction of the surface or interface structure are now fairly well developed in the case of metals and semiconductors, much less is known about oxide surfaces because of their ionic character, and about metal/oxide interfaces because of the complexity of the bonding between such dissimilar materials. The interracial energy is composed of competing terms, which often have the same order of magnitude. In particular, an important component is believed to be the image interaction [12]. In the recent years, the matter has drawn a number of theoretical contributions [ 13-15] and the development of more powerful computers resulted in improvements in the microscopic modeling of these surfaces and interfaces [16-18]. One of the major objectives of experiments on oxide surfaces and metal/oxide interfaces is to determine some parameters that allow testing the theoretical models. Among the questions one would like to address on oxide surfaces, one may ask: Is the surface "ideal", or does it have defects? Is it stoichiometric or does it have vacancies? Are there steps, regular or not, on the surface? Is the surface free of any contaminants (like carbon)? How to remove them? What is the equilibrium structure and morphology of the surface: is it a simple truncation of the bulk, or is it relaxed or reconstructed? How well do the available theories predict this structure? How is it affected by defects or by different surface treatments, for instance annealing in oxygen partial pressure or ion sputtering? What are the different reconstructions undergone by an oxide surface upon reduction by heating in UHV? Can these different surface reconstructions be predicted theoretically? Are polar surfaces (according to the definition of Tasker [19]) unstable or can they be stabilized by a reconstruction? How does the surface evolve upon exposure to different gases (H2, CO, NO, H20, CO2, NO2, etc); does the surface dissociate water? Are there general trends between similar surfaces? The questions one would like to address on metal/oxide interfaces are also numerous: Is the interface reactive? G. Renaud/Surface Science Reports 32 (1998) 1-90 7 What is the equilibrium morphology of the interface? What is the growth mode: layer-by-layer (2D or Van-der-Merwe), 2D followed by 3D (StranskiKrastanov), three-dimensional (3D or Volmer-Weber), or 2D for a fraction of the first monolayer, followed by 3D (2DI) [20]? How are the growth mode and the structure of the metallic overlayer affected by the characteristics of the oxide surface listed above? What are the growth kinetics and how is the growth mode modified by changing the substrate temperature or the flux of incoming atoms? Can the growth be modified by a surfactant, like oxygen or carbon monoxide? What are the structure and morphology of the growing film, and their evolution as a function of thickness, substrate temperature and incoming flux? When does the film start to coalesce; when does it become continuous? Is there epitaxy? What are the orientational relationships between the growing film and the substrate? What are the epitaxial sites and the interfacial distance between the last oxide plane and the first metal atomic plane? How is the lattice parameter mismatch between the substrate and the adsorbate accommodated as a function of the film thickness? Are there growth defects such as stacking faults, twins or dislocations in the growing film? When and why do they appear? More generally, what is the crystalline quality of the film? Can it be improved either with the help of a surfactant or by changing the growth temperature or by postannealing'? Metal/oxide interfaces may be prepared by very different routes, such as internal oxidation of alloys, hot pressure bonding, sputtering, laser ablation, etc. We will restrict ourselves here to those prepared under ultra high vacuum (UHV) conditions, by molecular beam epitaxy (MBE) on well-characterized and clean oxide surfaces. More and more studies have been devoted to oxide surfaces and metal/oxide interfaces prepared under "clean" conditions in the recent years. A few model surfaces have been selected both by theoreticians and by experimentalists, either because of their simplicity, because of their important applications, or because of their availability as large single crystals. Among them, 0~-A1203, MgO, TiO2, ZnO and SrTiO3 surfaces are the most studied. Theoreticians have also chosen model metal/oxide interfaces, like Ag/MgO(00 1) or Pd/MgO(00 1), and minimized the interfacial energy with respect to two structural parameters, the epitaxial site and the interfacial distance, for which they crucially need experimental determination. Other oxides have also been studied to a less extent, such as: CaO, CeO2, ZrO2, Cr203, Fe203, Fe304, WO2, NiO, CuO, Ga203, SnO2, etc. Despite this fast growing interest, often little is known of the structure and morphology of oxide surfaces and metal/oxide interfaces because of the difficulty to quantitatively characterize insulator surfaces and interfaces with the usual surface science techniques based on electron beams or tunneling effects. The atomic force microscope (AFM) usually lacks the required atomic resolution to characterize the atomic structure of insulators. Alternative techniques exist, like the scattering of beams of neutral atoms, He for instance, but they are only sensitive to the top atomic layer of the surface, and thus cannot probe buried interfaces. One way of avoiding charging effects is to prepare ultra-thin oxide films (for instance of A1203, Fe203, Fe304, Cr203, NiO, CoO, etc.) on conducting substrates, thus allowing the use of all surface science techniques, including the scanning tunneling microscope (STM). However, these thin films often have a large density of defects (vacancies, interstitials, dislocations, 8 G. Renaud/Surface Science Reports 32 (1998) 1-90 domain walls, different phases and variants, grain and sub-grain boundaries, etc.), which may completely change their electronic properties and thus also their macroscopic properties. It is thus often mandatory to work on single crystals either to preserve given properties, such as a large gap and a high resistivity, or to control the type and the density of defects which may play a dominant role on the interfacial properties. We will try to show in this review the power of the grazing incidence X-ray scattering (GIXS) technique to characterize single-crystal oxide surfaces and metal/oxide interfaces. Since its discovery [21] GIXS has emerged over the past decade as a new tool for studying the structure of surfaces and interfaces [22-30]. This probe has several advantages with respect to the more conventional electron based surface techniques such as low-energy electron diffraction (LEED) and reflection high-energy electron diffraction (RHEED). X-rays interact weakly with matter, so that in most cases, a simple quantitative analysis based on a single scattering (kinematic) calculation can be performed, while multiple-scattering effects dominate electron diffraction. Moreover, X-rays penetrate deeply in matter, enabling the study of buried interfaces. In addition, the scattering cross section is very well known for all atoms. X-ray diffraction peaks can be measured with very high resolution and over a large intensity range, thus enabling detailed lineshape analyses that are not accessible with other diffraction techniques. GIXS also benefits from the huge amount of work performed in conventional 3D crystallography, and hence allows the determination of the atomic structure of surfaces and interfaces with high accuracy. GIXS is especially suited to investigate the structure and morphology of oxide surfaces, because, unlike most surface techniques, it is not subject to charge build-up due to the insulating character of the surface. The large intensity of today's synchrotron radiation sources allows getting measurable diffraction from less than one monolayer (ML) of material. The grazing incidence geometry can drastically reduce the X-ray penetration in matter, down to ~ 25 ,~, and thus the unwanted bulk elastic and inelastic (i.e. Compton, Resonant Raman, Fluorescence, Thermal Diffuse) scattering with respect to the measured surface or interface elastic scattering [26]. Even under incidence conditions different from total external reflection, the surface sensitivity can be achieved by measuring diffraction rods that are specific to the surface, such as non-integer reflections arising from a surface reconstruction. This review is organized as follows. The basis of X-ray scattering under grazing incidence will first be summarized. The diffraction by a surface will next be considered, with methods to quantitatively determine its atomic structure (roughness, relaxation or reconstruction), and to analyze the degree of order using lineshape analysis. The diffraction by an epilayer/substrate interface will next be recalled, with the possibilities to determine the registry and the lattice parameter relaxation on coherent interfaces, or analyze the interfacial network of dislocations in the case of semi-coherent interfaces. A few experimental considerations will next be given, in particular the sample requirements to obtain reliable quantitative measurements. Several recent determinations of the atomic structure, relaxation or reconstructions of clean oxide surfaces will next be described. They will be followed by a review of several in situ studies of the first stages of formation of metal/oxide interfaces, and next by the investigations of thick metal films on oxides, exhibiting a network of interracial dislocations that plastically relax the lattice parameter mismatch. A quick overview of the characterization of diverse oxide thin films by GIXS will next be given before concluding on the present and future possibilities of the technique applied to the field of oxide surfaces and metal/oxide interfaces. G. Renaud/Surface Science Reports 32 (1998) 1-90 9 2. X-ray scattering by surfaces and interfaces 2.1. Grazing incidence X-rays 2.1.1. Grazing incidence geometry The grazing incidence X-ray scattering geometry, shown in Fig. 1, is identical to the threedimensional case, except that, in order to decrease the bulk scattering contributions, the incident X-ray beam, of wavevector ki, is kept at a glancing incidence angle ~i with respect to the surface. The scattered beam, of wavevector kf, is detected at an angle oLf with respect to the sample surface and at an in-plane angle 20 with respect to the transmitted beam. The momentum transfer is defined as Q = k f k i , and is often decomposed into two components, Q// and Q±, respectively, parallel and perpendicular to the surface. The absolute value of Q± ms a function of o~i and c~f: Q± = k (sin o~i + sin O~f); k = 27r/A, where A is the wavelength. When Oq and O~f are very small, Q ~ Q//, the scattering plane is nearly parallel to the surface, and diffracting net planes are perpendicular to it. The scattering geometry being defined by the incident beam and detector directions, one has only to rotate the sample about its surface normal to bring these net planes into diffraction condition, which occurs when they make an angle 0 with respect to both the incident and the scattered beam. In this way, the long-range periodicity parallel to the surface is probed. It is often useful to measure the scattered intensity as a function of Q±, which is often achieved by increasing o~f while keeping grazing incidence. 2.1.2. Refraction of X-rays at grazing angles Because the incidence angle is small, it is necessary to consider the effects of refraction at the surface [31 ]. The refractive index, n, of matter for X-rays is slightly less than unity,