Crystallization dynamics and interface stability of strontium titanate thin films on silicon

Nonstoichiometric SrTiO3 thin films were fabricated by different thin-film deposition methods. The impact on the oxide/silcon interface stability as well as the crystallization onset temperature is investigated.


Introduction
Metal-insulator-metal stacks are a promising concept for nonvolatile memories based on resistive switching (Sawa, 2008;Waser et al., 2009). Perovskite-type transition metal oxides are favorable materials for the required thin insulating layers, because of their wide band gaps (Kahn & Leyende, 1964;Benthem et al., 2001) and comparatively high dielectric constants (Samara, 1966), as well as mixed ionic and electronic conductivity (Baiatu et al., 1990). Strontium titanate is studied for its potential in resistive random access memory applications (Sawa, 2008;Waser & Aono, 2007), since it exhibits a metal-insulator transition with a change in electrical resistance of over several orders of magnitude (Watanabe et al., 2001;Beck et al., 2000). Investigations of the resistive switching mechanisms and improvements of the device performance focus mostly on amorphous thin films (Kü geler et al., 2011;Yan et al., 2010;Jung et al., 2010;Kang et al., 2013;Liu et al., 2013), because they offer lower leakage currents compared to polycrystalline layers which have grain boundaries as leakage paths and non-isotropic electrical properties (Wilk et al., 2001). However, resistive switching behavior is also found in strontium titanate bulk crystals Wojtyniak et al., 2013), as well as epitaxially grown and polycrystalline thin films (Szot et al., 2007;Shibuya et al., 2010;Menke et al., 2009;Choi et al., 2005;Sun et al., 2011). Proposed mechanisms are governed by conductive filament formation resulting from redistribution of point defects. These are also known to influence the crystalline structure of the material (Hanzig et al., 2013(Hanzig et al., , 2015. Likewise, the valence state of titanium in strontium titanate switches reversibly under the influence of an external electric field (Leisegang et al., 2009;Hanzig et al., 2014). Therefore, the question arises, how do microstructure and crystallization influence the switching mechanism and stability in polycrystalline thin films? The starting point for such investigations is the equilibrium phase diagram of the SrO-TiO 2 quasi-binary system (Levin et al., 1964). At a composition of 50% SrO and 50% TiO 2 , SrTiO 3 crystallizes in the cubic structure with space group Pm3m. Stoichiometry deviations towards higher Sr content lead to the formation of the homologous series of Ruddlesden-Popper (RP) phases SrO(SrTiO3) n (Ruddlesden & Popper, 1957, 1958 with space group I4=mcm. The RP phases are composed of perovskite unit cells that are shifted by [ 1 2 1 2 0] after n layers, thereby introducing an additional SrO plane. Ab initio calculations show that RP phases have electronic properties comparable to those of SrTiO 3 , but reveal tunable permittivity (Bhalla et al., 2000) and band gap . Ti excess in the phase diagram leads to the formation of TiO 2 precipitates in an SrTiO 3 matrix. The ternary phase diagram Sr-Ti-O (Tanaka et al., 2003) predicts additional strontium titanate derived phases, e.g. Magneli phases (Andersson et al., 1957), in the case of oxygen deficiency. For thin layers of a ternary oxide the microstructure and in turn the optical and electric properties do not depend solely on the given stoichiometry. Nucleation and growth of crystalline phases are also influenced by the substrate through lattice mismatch and roughness. Since SrTiO 3 may be used as a high dielectric constant gate oxide in CMOS-based devices, the interfaces of these thin films in contact with silicon are still under extensive investigation. In particular, numerous experimental and theoretical studies on the stability of binary oxides (Hubbard & Schlom, 1996;Gutowski et al., 2002), ternary oxides with perovskite structure (Goncharova et al., 2006) and especially scandates on Si have been carried out (Sivasubramani et al., 2006;Adelmann et al., 2008;Copel et al., 2010). At temperatures above 770 K, the thermal stability of the interface of the oxide thin film and silicon substrate decreases (El Kazzi et al., 2007) and silicon interdiffusion interferes with the crystallization of the thin-film material system. The decomposition of ternary oxide thin films has been observed (Adelmann et al., 2008), as well as cation diffusion into the substrate accompanied by silicate formation (Copel et al., 2010). Even for epitaxial SrTiO 3 on Si, extensive investigations regarding interface instability have been reported (Goncharova et al., 2006;Delhaye et al., 2006;Hu et al., 2014).
Although stoichiometric SrTiO 3 precursor material has been used in the present study, we show here that deviations from the ideal SrTiO 3 stoichiometry are introduced already during the deposition process. Further, we report on the crystallization behavior of Sr-rich and Sr-deficient strontium titanate thin films during annealing under ambient atmosphere, including an in situ analysis of the growth kinetics of the cubic SrTiO 3 phase in these films. Finally, special emphasis is put on the thermal stability of the interface between the transition metal oxide and silicon.

Materials and methods
Strontium titanate thin films were prepared via electron beam evaporation (EBE) with an Edwards Auto 500 utilizing coarse-grained and ball-milled strontium titanate obtained from CrysTec GmbH, Berlin. Further fabrication of SrTiO 3 thin films was done by radio frequency magnetron sputtering (RF SP) on a Bestec UHV magnetron sputtering system with argon plasma using a strontium titanate target (purity 99.95%) from Testbourne Ltd, England. All substrates were (001)oriented monocrystalline Si wafers with a native oxide surface layer of approximately 2 nm thickness. For reasons of comparability no substrate heating was applied. Process parameters are summarized in Table 1.
The elementary composition of all thin-film samples was obtained from wavelength dispersive X-ray fluorescence (XRF) using a Bruker S8 Tiger spectrometer employing a rhodium source and LiF200, XS-55 and PET monochromator crystals, covering a spectral energy range from 60 keV down to 0.491 keV. A full fundamental parameter approach was adopted for all calculations, as implemented in the software ML Quant (Bruker AXS, Karlsruhe, Germany.). The Ti/Sr ratio of the material residues in the crucible after evaporation was checked by energy dispersive XRF (EDX) using a JEOL JSM-6400 scanning electron microscope equipped with a Noran EDX detector. Layer thicknesses were determined from spectroscopic ellipsometry (SE) with a Sopra GES 5E and X-ray reflectivity (XRR) using a Seifert HZG4. Layer density was evaluated from XRR with the pyxrr (Richter, 2014) analysis software. Depth-resolved stoichiometry analysis was performed with X-ray photoelectron spectroscopy (XPS) on a Thermo Fisher Scientific Escalab 250Xi. High-resolution transmission electron microscopy (HRTEM) was carried out using a 200 kV analytical high-resolution transmission electron microscope (JEOL JEM 2200 FS) equipped with an in-column filter to improve image quality by removing inelastic scattered electrons. In addition, TEM imaging was employed for depth calibration of the XPS compositional profile. In situ temperature-dependent grazingincidence X-ray diffraction (GI-XRD) was measured at beamline E2 of the DORIS III storage ring (HASYLAB at DESY, Hamburg). Measurements were carried out using monochromatic light of 12 480 eV energy (approximately 1 Å wavelength). The diffraction patterns were recorded at a constant 2 detector angle of 15 with a Dectris Mythen onedimensional photodiode array and an exposure time of 150 s, while the sample surface was inclined by an angle ! of 2.5 with respect to the incident beam. Sample heating was facilitated by an Anton Paar DHS 1100 oven with a carbon dome under atmospheric conditions and controlled by a Eurotherm 2604 temperature controller. Interfering carbon dome reflections were cleared from the diffraction patterns by a copper aperture, thus limiting the temperature to a maximum of 1223 K. Afterwards, angle calibration of the diffraction data was performed using GI diffraction data from a Philips X'PERT thin-film system (PW3020 goniometer, 0.4 equatorial collimator and planar Ge monochromator) with copper radiation in a 2 range of 10-120 . Depth-resolved X-ray diffraction data were collected at glancing angles ! of 0.125-8.45 to obtain the distribution of crystalline phases within the samples. The development of the SrTiO 3 phase properties during temperature evolution was determined by Rietveld analysis utilizing Bruker's TOPAS 4.2 software with a combined 1/x and sixth-order polynomial background. To enable convergence of the fit routine, secondary phases were taken into account as peak phases.

As-deposited strontium titanate thin films
The layer stoichiometries, thicknesses and densities of the RF SP and EBE samples as determined from XRF, SE and XRR are summarized in Table 2. The RF magnetron sputtered sample shows strontium deficiency with an Sr/Ti ratio of 0.8. All thin films prepared by electron beam evaporation reveal a distinct excess of strontium with an Sr/Ti ratio of 1.9. Thus, the residue of the material in the evaporation crucible was analyzed by EDX. The results show that the residue is inhomogeneously depleted of Sr, with the Ti/Sr ratio approximately matching the Sr/Ti ratio of the EBE films. Both kinds of samples reveal similar densities of 3.70-3.85 g cm À3 , which is much lower than for single-crystal strontium titanate with a density of sc ¼ 5:13 g cm À3 (Abramov et al., 1995). As characterized by XRD, all as-prepared layers are amorphous.

In situ crystallization
The EBE samples were heated at rates of 2, 4 and 8 K min À1 , whereas only the rate of 4 K min À1 was used to crystallize the RF SP sample. Fig. 1 shows the in situ GI-XRD data of EBE-and RF-prepared samples, subjected to different temperature ramps. Reflections attributed to the cubic SrTiO 3 phase are indicated in these graphs. The broad reflection present at a 2 angle of 36.8 in all patterns is due to the silicon substrate [Umweganregung of the Si 311 reflection (Tö bbens et al., 2001)]. In the EBE samples, superimposed on the amorphous background, weak reflections of the cubic SrTiO 3 phase (Abramov et al., 1995) appear shortly after starting the X-ray diffraction measurement at 473 K. Despite the different heating rates, all EBE samples show comparable crystallization behavior. Starting with the vanishing of the amorphous background (see black dashed lines in Fig. 1), the intensities of the reflections of the cubic SrTiO 3 phase first increase rapidly. This stage is followed by a plateau of constant

Figure 1
Crystallization of SrTiO 3 thin films monitored by GI-XRD through a temperature ramp of (a) 2 K min À1 , (b) 4 K min À1 and (c) 8 K min À1 for Sr-rich (EBE) and (d)  intensity that is accompanied by the consecutive emergence and disappearance of additional phases (see white dashed and dash-dot lines in Fig. 1). Fig. 2 summarizes the intensity evolution of the SrTiO 3 002 reflection for all annealing experiments. At elevated temperatures, a narrow temperature window arises, displaying strong reflections exclusively from the cubic strontium titanate phase. Above the latter temperature range, the diffraction patterns show similar reflections attributed to secondary phases. The deviations in onset and dissolution temperature of the additional phases correlate with the particular magnitude of the heating rate (see Table 3). This formation and dissolution of secondary phases also influences the lattice parameter and crystallite size of cubic SrTiO 3 (see Figs. 3a and 3b). Analysis of the respective SrTiO 3 reflection intensities shows no deviations from the theoretical intensity distribution, thus indicating randomly oriented crystallites. In the case of the RF SP sample, cubic SrTiO 3 initially crystallizes at 739 K. Here the reflections from SrTiO 3 reach their maximum in intensity at a temperature of 763 K (see Fig. 2), with the vanishing of the amorphous background. In the entire temperature range, the diffraction pattern of the RF SP sample contains two very weak signals at 2 values of 22.9 and 24.9 , which are assign-able neither to SrTiO 3 nor to the experimental setup of the beamline, because they are present in ex situ X-ray diffraction patterns too. The logarithmic scaling emphasizes the signals which are almost indistinguishable from the background in individual diffraction patterns. In order to increase the signalto-noise ratio for phase matching and indication, diffraction patterns were summed through the whole temperature range. Unfortunately, phase matching was unsuccessful. However, a few expected phases like rutile and anatase were excluded with certainty. Details of the temperature dependence of the lattice parameter a and crystallite size were obtained by a basic Rietveld analysis with Bruker TOPAS, starting at the crystallization temperature for the respective layers. The thin film prepared by RF magnetron sputtering reveals an initial lattice constant of 3.925 Å at 739 K, whereas for the EBE samples it is 3.945 Å . In the measured temperature range, the RF SP sample lattice parameter stays nearly constant. In contrast, those of the EBE samples undergo strong alteration at temperatures up to 1173 K, while additional phases form and disappear. All samples show lattice parameters larger than those reported for pure SrTiO 3 , with a = 3.905 Å (Abramov et

Figure 3
Temperature dependence of (a) lattice parameter and (b) crystallite size of SrTiO 3 extracted from in situ GI-XRD with different temperature ramps of 2, 4 and 8 K min À1 for Sr-rich (EBE) and 4 K min À1 for Srdepleted layers (RF SP), determined by Rietveld refinement using TOPAS. Dashed lines highlight the relative thermal lattice expansion calculated from the SrTiO 3 thermal expansion coefficient (de Ligny & Richet, 1996). Prominent points in the development of the EBE samples' crystallization are highlighted by the letters A, B and C.

Figure 2
The evolution of the 002 reflection intensity with increasing temperature as extracted from the Bruker TOPAS SrTiO 3 hkl output files. Intensities are scaled with respect to layer thicknesses and the global count maximum.
al., 1995), which can be attributed to deviations of the Sr/Ti ratio from 1.0 (Brooks et al., 2009). The low density found for all samples is mainly due to microporosity. Focusing on the temperature intervals where only SrTiO 3 reflections are detected in the EBE diffraction patterns (compare pure SrTiO 3 in Table 3) a distinct drop of the lattice parameter is observed (see Fig. 3a), while at the same time the crystallite size increases (see Fig. 3b). At the crystallization temperature the RF SP sample crystallite size is about 24 nm and grows steadily with increasing temperature up to 29 nm (see Fig. 3b). Among the EBE thin films the crystallite size evolves similarly. Their initial and final crystallite sizes vary from 29 to 25 nm and 32 to 30 nm, respectively. Fig. 3(b) indicates a drop of the size of the SrTiO 3 crystallites with the appearance of additional phases. The inverse behavior can be seen at temperatures where additional phases disappear. The final crystallite sizes are of comparable magnitude, independent of the fabrication method. For all EBE samples the irreversible formation of at least one secondary phase is observed above 1132.7, 1150.2 and 1175.2 K, respectively.

Secondary phase investigation
The formation of additional phases during crystallization motivates a closer examination of the thin films with HRTEM. Fig. 4 displays the cross-section images of an electron beam evaporated thin film (a) in comparison to an RF-sputtered sample (d) after annealing up to 1223 K. Whereas the RF SP sample exhibits one well defined layer, the EBE sample is inhomogeneous, with at least two distinct layers on top of the substrate, which was oxidized in the formation process. To obtain a compositional profile of an EBE sample, depthresolved photoelectron spectroscopy was employed, averaging over a sample area of 1.13 mm 2 (see Fig. 5). As with HRTEM, a decomposition into two layers of 165 and 140 nm thickness was detected, exhibiting Sr:Ti:O and Si:Sr:O ratios of 1:1:3. Within the topmost layer, HRTEM micrographs at a magnification of 400k show crystalline areas in both EBE and RF SP samples (Figs. 4b and 4e). Reflections in the respective Fourier transformation are assigned to the cubic room-temperature phase of strontium titanate ( Fig. 4c and 4f).  binding energies were determined to be 102 eV, indicating the formation of silicates (Shutthanandan et al., 2002). Underneath, a silicon oxide film has developed during annealing, with Si 2p binding energies of 99.5 and 104.0 eV, typical for SiO 2 bonding (Shutthanandan et al., 2002). In combination, these measurements give evidence of silicon diffusion into the Sr-rich oxide layer, leading to an increase of the film thickness up to 305 nm with homogeneous content of oxygen and strontium across the uppermost layers. To complement the extremely local TEM images with volume average information and to identify the depth distribution of secondary phases, angle-resolved GI-XRD was conducted (see Fig. 6) at glancing angles ranging from 0.125 to 8.450 , which correspond to a penetration depth of 2.95 (2.93) nm to 3.61 (5.41) mm in SrTiO 3 (SrSiO 3 ) when using copper radiation and the densities listed in Table 2

Discussion
In spite of the use of stoichiometric precursor materials, different physical vapor deposition techniques nevertheless result in thin films with deviating stoichiometry within the binary system SrO and TiO 2 . Their composition was determined to be either Sr-deficient Sr 4 Ti 5 O 18 or Sr-and O-rich Sr 2 TiO 7 (see Table 2), which clearly indicates the different impact of the deposition method on film stoichiometry. Referring to the binary phase diagram of SrO and TiO 2 (Levin et al., 1964), samples should crystallize in the form of cubic SrTiO 3 with segregation of TiO 2 or Ruddlesden-Popper phases (Ruddlesden & Popper, 1957, 1958, respectively. In the case of Sr-rich layers from electron beam evaporation the deviation is attributed to preferential ablation of SrO due to its lower evaporation enthalpy from the SrTiO 3 melt compared to TiO 2 (Dam et al., 1996). In contrast, the presence of Sr-depleted layers resulting from magnetron sputtering is attributed to preferential sputtering of the stoichiometric SrTiO 3 target. Because of the comparable atomic masses of Ar and Ti as compared to Sr, the latter is sputtered less effectively.
For clarity, Fig. 7 depicts the crystallization behavior of Srrich samples during annealing up to 1223 K. The background intensity in the XRD data is related to the presence of an amorphous phase (Fig. 7a). Its dissolution during annealing coincides with increasing structural ordering in the layers. Hence the variation in the temperature at which this background intensity vanishes is attributed to different degrees of structural and compositional ordering in the as-deposited layers. Consistently, an even higher degree of chemical order in the as-deposited layer can be achieved by using atomic layer deposition, which further increases the crystallization temperature (Rentrop et al., 2015). Because of the combination of low particle energy and inhomogeneous evaporation, only a randomly distributed fraction of the EBE layers' volume exhibits stoichiometry close to SrTiO 3 . Here the crystallization of the cubic SrTiO 3 phase starts almost at the beginning of the temperature treatment, leaving a strontiumoxide-rich matrix in the surrounding region (see Fig. 7b).
Owing to the small peak width, Rietveld refinement results in relatively large crystallites, and the weak peak intensities suggest a low diffracting volume of these seed crystallites, Depth-resolved gracing-incidence X-ray diffraction of an Sr-rich SrTiO 3 (EBE) thin film (Cu radiation). The related glancing angles are indicated on top of each diffraction pattern. Reflections from cubic SrTiO 3 (Abramov et al., 1995) are labeled with stars and those from the substrate (Tö bbens et al., 2001) by the letter S.

Figure 5
Depth-resolved XPS measurement of an Sr-rich SrTiO 3 (EBE) thin film crystallized using a heating rate of 2 K min À1 up to 1223 K. The depth axis is calibrated according to Fig. 4(a).
leading to a large error in the determination of the particle size. The fact that the initial lattice parameter is larger than the reported value of the cubic SrTiO 3 phase (Abramov et al., 1995) suggests a nonstoichiometry either to the Sr-or to the Ti-rich side (Brooks et al., 2009). Further increase of the annealing temperature to approximately 573 K allows successive crystallization of several phases from the initially amorphous matrix (see Fig. 1a-1c), causing a drop in crystallite size (Fig. 1b). The tabulated (de Ligny & Richet, 1996) and observed coefficient of thermal expansion do not match (see Fig. 3a). In detail, the rapid changes in lattice parameter marked A, B, and C in Fig. 3(a) correspond to the formation of secondary phases (see Fig. 7c-7e). Since the stoichiometry of the final SrTiO 3 is nearly ideal we assume that the lattice expansion is governed by the relaxation of nonstoichiometry (Brooks et al., 2009). In contrast, the cation nonstoichiometry in the SrTiO 3 crystallites has to remain constant if the lattice parameter rises as the temperature approaches the prominent points A, B and C in Fig. 3(a). In terms of thermodynamics the final state is probably still not fully equilibrated because of the unusually low lattice constant for this temperature.
At elevated temperatures two processes have to be discussed. First, the oxygen and strontium mobility is large enough to precipitate Sr and O at the interface to the substrate (see Fig. 4a), leaving an approximately 100 nm-thick layer composed of phase-pure SrTiO 3 on top (see Fig. 5). In between, an intermixing zone with gradients of titanium and silicon is thereby created. Second, SrTiO 3 as well as SrO on Si are both reported to be thermodynamically unstable at higher temperatures (Hubbard & Schlom, 1996;Reiner et al., 2010;El Kazzi et al., 2007). Therefore, a reaction of excess Sr with SiO 2 to form SrO by degradation of the SiO 2 layer on top of the substrate is possible, and even in lower-temperature regions the presence of SrTiO 3 , SrO and Si can lead to formation of SrSiO 3 (Hubbard & Schlom, 1996). In the case of strontium oxide, it is reported that the conversion of SrO at the interface with Si at higher temperatures first induces Sr 2 SiO 4 formation, prior to the latter's conversion to a phase close to SrSiO 3 as the temperature is further increased (El Kazzi et al., 2007). Furthermore, the surface of the Si substrates is oxidized by oxygen originating from the thin film, increasing the SiO 2 layer thickness to approximately 25 nm as determined by XPS and TEM, starting at an initially natural silicon oxide layer. Possibly, the oxygen stems from the surrounding atmosphere as well. The observed growth of SiO 2 is consistent with the prediction of the basic oxidation model proposed by Deal & Grove (1965). Subsequently, a diffusion of silicon from the substrate establishes the newly formed silicate layer (Fig. 7g). Here, the XPS compositional profile between 200 and 300 nm from the layer surface indicates Si reaching a homogeneous content of 20%, framed by a decreasing percentage towards SrTiO 3 as well as SiO 2 (see Fig. 5). However, we were unable to match impurity phases to entries of the ternary system Sr:Si:O in the common powder pattern databases. As the Srdeficient thin film shows no additional crystalline phases in the diffraction pattern up to 1223 K without a visible decomposition in the TEM cross section, it is concluded that the additional titanium stabilizes SrTiO 3 on top of the silicon substrate.

Conclusion
We have reported the preparation of initially amorphous SrTiO 3 thin films by electron beam evaporation and RF magnetron sputtering. In situ X-ray diffraction of Sr-rich and Sr-depleted layers during annealing in air up to 1223 K was performed using synchrotron light. Clear differences in the crystallization onset of cubic SrTiO 3 have been observed. During heat treatment, electron beam evaporated samples exhibit a number of unknown secondary phases, and transmission electron microscopy confirms the formation of an additional layer between the SrTiO 3 film and the Si substrate. By means of X-ray photoelectron spectroscopy the composition was evaluated to be SrSiO 3 . For RF-sputtered samples the interface to the substrate is chemically stable and solely the cubic SrTiO 3 phase is formed during annealing. The reactivity at the interface between substrate and thin film as well as formation of additional phases resulting from silicon interdiffusion is triggered by stoichiometry deviations and thus by the preparation method. This work has been performed within the Cluster of Excellence 'Structure Design of Novel High-Performance Sketch of the supposed evolution of the crystallization for Sr-rich samples (EBE): (a) as-deposited amorphous layer with an Sr/Ti ratio of 1.9, (b) growth of SrTiO 3 seed crystals, (c) an enlarged number of SrTiO 3 crystallites accompanied by the first secondary phase, (d) and (e) appearance of alternative secondary phases, whereas the former phase vanishes, ( f ) presence of crystallites from cubic SrTiO 3 only, and (g) separation into two distinctive layers comprising SrTiO 3 and SrSiO 3 , respectively. The amorphous SiO 2 layer thickness increases. All temperatures are exemplarily extracted from the sample with 2 K min À1 heating rate.
Materials via Atomic Design and Defect Engineering (ADDE)', which is financially supported by the European Union (European Regional Development Fund) and by the Ministry of Science and Art of Saxony (SMWK). We acknowledge Thermo Fisher Scientific for kindly performing X-ray photoelectron spectroscopy measurements. We also thank the DFG for financial support within PAK 215. FH is grateful to the project 'Dritte Sä ule Hochschulpakt' for covering travel expenses. Parts of this work were done within the joint research project 'CryPhysConcept -mit Kristallphysik zum Zukunftskonzept elektrochemischer Energiespeicher' (03EK3029A), which is financially supported by the Federal Ministry of Education and Research (BMBF) and the HGF-funded Virtual Institute Memriox (VH-VI-422). CR acknowledges financial support by the BMBF within project 05K10OF1.