Al1−xScxN Thin Films at High Temperatures: Sc-Dependent Instability and Anomalous Thermal Expansion

Ferroelectric thin films of wurtzite-type aluminum scandium nitride (Al1−xScxN) are promising candidates for non-volatile memory applications and high-temperature sensors due to their outstanding functional and thermal stability exceeding most other ferroelectric thin film materials. In this work, the thermal expansion along with the temperature stability and its interrelated effects have been investigated for Al1−xScxN thin films on sapphire Al2O3(0001) with Sc concentrations x (x = 0, 0.09, 0.23, 0.32, 0.40) using in situ X-ray diffraction analyses up to 1100 °C. The selected Al1−xScxN thin films were grown with epitaxial and fiber textured microstructures of high crystal quality, dependent on the choice of growth template, e.g., epitaxial on Al2O3(0001) and fiber texture on Mo(110)/AlN(0001)/Si(100). The presented studies expose an anomalous regime of thermal expansion at high temperatures >~600 °C, which is described as an isotropic expansion of a and c lattice parameters during annealing. The collected high-temperature data suggest differentiation of the observed thermal expansion behavior into defect-coupled intrinsic and oxygen-impurity-coupled extrinsic contributions. In our hypothesis, intrinsic effects are denoted to the thermal activation, migration and curing of defect structures in the material, whereas extrinsic effects describe the interaction of available oxygen species with these activated defect structures. Their interaction is the dominant process at high temperatures >800 °C resulting in the stabilization of larger modifications of the unit cell parameters than under exclusion of oxygen. The described phenomena are relevant for manufacturing and operation of new Al1−xScxN-based devices, e.g., in the fields of high-temperature resistant memory or power electronic applications.


Introduction
Thin films of scandium-substituted aluminum nitride (Al 1−x Sc x N or 'AlScN') are a pioneering new class of ferroelectric materials with wurtzite-type structure and unidirectional polarization reversal resulting in square-like hysteresis loops [1][2][3][4][5]. Their ultra-stable remnant polarization and high coercive fields promise ferroelectric field-effect transistorbased non-volatile memory devices maintaining large memory windows combined with high access speed, high endurance and low energy consumption [6]. In this respect, the integration of AlScN on Si has been demonstrated recently for ferroelectric field-effect transistors as well as GaN technology-based structures, e.g., for high electron mobility transistors [7][8][9][10] with integrated memory functions. Further, thin films of AlScN are potential candidates for high-temperature actuation and sensing applications in harsh environments (T > 500 • C) or for high-temperature non-volatile memory due to remarkable temperature (#1) Thin films of Al 0.73 Sc 0.27 N (thickness 400 nm) were deposited on high-temperaturestable metal electrode Mo(110)/AlN(0001)/Si(001) templates using an Oerlikon MSQ200 multi-source pulsed-DC sputter chamber and previously reported processes [26,27]. These films show a typical highly c-axis-oriented fiber texture (the 0002 XRD rocking curve (XRC) full-width at half-maxima (FWHM) is~1.6 • ) with a small number of misoriented grains [26]. Surface capping of the films with a 100 nm layer of amorphous SiN x was applied to protect the film from oxidation during high-temperature treatment.
(#3) Epitaxial thin films of Al 0.6 Sc 0.4 N (thickness of 400 nm) with c-axis-oriented columnar growth on a template of Mo(110)/AlN(0001)/Al 2 O 3 (0001). The films were deposited according to the procedure described in [29] and capped with a 100 nm thick layer of SiN x for passivation during annealing experiments.

X-ray Diffraction Experiments
The temperature-dependent microstructural changes in all thin film samples were investigated via XRD experiments on as-deposited samples as well as during and after in situ experiments. The structural properties of the as-grown epitaxial films (#2) were analyzed on a PANalytical X'Pert Pro MRD diffractometer equipped with a multi-layer mirror and a 2 × Ge 220 monochromator, providing Cu-Kα 1 radiation (λ = 154.06 pm). XRD experiments on the sample sets #1 and #3 were conducted on a Rigaku SmartLab diffractometer operated under similar conditions. To characterize the as-deposited and annealed samples, phase analyses and mosaic tilt analyses were performed by recording 2θ/θ-scans and rocking curve ω-scans of the AlScN 0002-reflection.
To study effects of high-temperature annealing in (#2) Al 1−x Sc x N/Al 2 O 3 films as a function of Sc concentration, we performed in situ XRD measurements in the temperature range of 30-1000 • C. The experiments were conducted at a base pressure of ∼6 × 10 −3 mbar achieved inside a graphite dome heating furnace (Anton Paar, Ostfildern-Scharnhausen, GermanyDHS 1100 Domed Hot Stage) placed in a PANalytical Empyrean diffractometer equipped with a PIXcel hybrid detector (Malvern Panbalytical, Kassel, Germany). 2θ/θ and ω/2θ diffraction scans were recorded after 20 min hold time at selected temperatures for approximately 60 min of dwell time. The heating rate was set to 150 • C/min. Lattice parameters were estimated from scans of the symmetrical 0004-reflection and asymmetric 10-15-reflection with shallow (−) and steep (+) angles of incidence ω, similar to the technique described by Herres et al. [30]. For the analysis of sample set #1 and #3, simple 2θ/θ-scans were conducted during in situ heating experiments under similar conditions using an AntonPaar (DHS 1100) hot stage placed on the Rigaku diffractometer and as described in [11].
In addition to the in situ examinations, we also conducted an ex situ annealing experiment to compare the high-temperature behavior of samples exposed to and protected from oxygen-contaminated environments during high-temperature annealing. Here, a Al 0.73 Sc 0.27 N/Mo(110)/AlN(0001)/Si(001) sample was placed inside a self-built tube furnace keeping a quartz tube under vacuum (3-4 × 10 −7 mbar). Initially, the quartz tube was annealed without the sample to 800 • C together with Ti getter material to reduce the amount of adsorbed oxygen on the tube walls. The actual annealing experiment was started at a base pressure of 1.2 × 10 −6 mbar and the temperature was increased to 800 • C in steps of 150 • C and cooled down to room temperature in a single step.

Results
The Results section is structured in two parts. In this first part, in situ annealing experiments on the fiber textured Al 0.73 Sc 0.27 N/Mo(110)/AlN(0001)/Si(001) (sample set #1) thin films are discussed with respect to intrinsic and extrinsic effects of observed anomalous thermal expansion. In the second part, the effect size of the thermal expansion behavior and the microstructural changes to epitaxial thin film samples of Al 1−x Sc x N/Al 2 O 3 (0001) are analyzed depending on Sc concentration (0 < x < 0.40) (sample set #2). Structural data recorded before and after annealing are discussed and data recorded during annealing will be compared. Eventually, we conclude on our hypotheses considering data from sample #3.

Part A: Intrinsic and Extrinsic Effects of Anomalous Thermal Expansion in AlScN Thin Films
The exceptional robustness of the wurtzite-type structure and its ferroelectric polarization switching have been recently demonstrated for 400 nm Al 0.73 Sc 0.27 N/Mo(110)/AlN(0001)/ Si(001) thin films during and after high-temperature treatment up to 1100 • C [11]. However, the experimental data suggested a minor degradation of the crystalline quality during the process, which is illustrated by highly comparable types of data collected on Al 0.73 Sc 0.27 N thin films from the identical wafer as presented in Figure 1. The evolution with temperature of the Al 0.73 Sc 0.27 N 0002-reflection profile centered at 2θ~36 • shown in Figure 1a displays the decrease in the maximum diffracted intensity of the Al 0.73 Sc 0.27 N component and the development of a hump as a shoulder at lower diffraction angles of the 0002-reflection. These observations indicate the reduction in the sizes of coherently scattering domains of the Al 0.73 Sc 0.27 N phase and the formation of a top oxide layer of (Al,Sc)(N,O) x , which is supported by energy-dispersive X-ray spectroscopy measurements on a scanning electron microscope before and after annealing (not shown). No further peaks indicating oxide formation were observed in the temperature-dependent 2θ/θ-scans (30-100 • C) (not shown). Further, the temperature dependent 0002-reflection profiles in Figure 1a reveal the 0002reflection of the AlN seed layer at a temperature of 1000 • C (see arrows) at higher diffraction angle with respect to the intensity maximum. This could be explained by differences in thermal expansion coefficients of AlN and AlScN, which are known for temperatures up to 400 • C [17]. However, when treating AlScN films at high temperatures, a transition in thermal expansion is observed. The comparisons between the thermal shifts in the 0002-reflection maxima for Al 0.73 Sc 0.27 N and AlN thin films and the Mo(110)-reflection of the Mo layer underneath Al 0.73 Sc 0.27 N are presented in Figure 1b. Here, the relative change in the lattice spacings ∆d[T]/d 0 reveals strong non-linearity in the thermal lattice expansion of the Al 0.73 Sc 0.27 N film. The curve for the Al 0.73 Sc 0.27 N film (green curve in Figure 1b) shows a non-linearity in thermal expansion changing from a linear low-temperature branch to a high-temperature branch at a transition temperature of about T tr~6 00 • C. The linear slope of the low-temperature branch is (∆c[25 • C-600 • C]/c 0 )/∆T~4.4 × 10 −6 / • C and increases by a factor of 3 to (∆c[700 • C-1000 • C]/c 0 )/∆T~13.7 × 10 −6 / • C in linear approximation to the high-temperature branch. Any influence of the underlayer is excluded by comparison to the relative change in the underlayer's Mo(110)-reflection which shows highly linear expansion. Further, an AlN film deposited under identical conditions was heated and shows a linear expansion of (∆c[25 • C-1100 • C]/c 0 )/∆T~3.5 × 10 −6 / • C. Due to the rather rough linear approximation over the entire temperature interval, the value for AlN is somewhat smaller but still consistent with literature data of 4.2 [20- cation-oxygen networks forming in metallic glasses [33] or small structural transition (Ba0.5Sr0.5)TiO3 induced by a change in oxygen site occupancy [34]. Reversible phenom are commonly observed in non-stoichiometric perovskite-based ion-conducting ceram which show pronounced chemical expansion depending on the oxygen partial press and temperature [35,36].
Parallel to the strong oxidation of the AlScN film, the interaction with oxygen spec and the intrinsic defect structure of the material could lead to the strong thermal exp sion and irreversible changes. In this respect, the fiber textured columnar films could tentially provide a Sc-and oxygen-enriched grain boundary structure [37], which co promote pathways for atmospheric oxygen species into the material.  To our knowledge, this strongly non-linear transition has not been observed to date for AlScN thin films. In the case of many oxide materials, non-linear thermal expansion behavior can be discussed in the context of oxygen-or oxygen vacancy-related effects termed 'chemical expansion' [32]. These chemical expansion phenomena can be nonreversible or reversible in nature. For instance, non-reversibility is observed in the case of cation-oxygen networks forming in metallic glasses [33] or small structural transitions in (Ba 0.5 Sr 0.5 )TiO 3 induced by a change in oxygen site occupancy [34]. Reversible phenomena are commonly observed in non-stoichiometric perovskite-based ion-conducting ceramics which show pronounced chemical expansion depending on the oxygen partial pressure and temperature [35,36].
Parallel to the strong oxidation of the AlScN film, the interaction with oxygen species and the intrinsic defect structure of the material could lead to the strong thermal expansion and irreversible changes. In this respect, the fiber textured columnar films could potentially provide a Sc-and oxygen-enriched grain boundary structure [37], which could promote pathways for atmospheric oxygen species into the material.
These preliminary experiments suggest the presence of residual oxygen contamination of the annealing atmosphere inside the graphitic dome placed in the diffractometer. Hence, the Al 0.73 Sc 0.27 N film was capped with a 100 nm thick SiN x layer to protect the film surface from oxidation-dependent effects, allowing us to investigate the purely intrinsic material contribution to the transition in thermal expansion behavior. With this experimental design, new in situ annealing experiments were conducted up to 1100 • C. The evolution of reflection profiles during the first and second annealing cycle and the corresponding ∆c(T)/c 0 plots are shown in Figure 2. The reflection profiles depicted in Figure 2a exhibit neither an oxide hump, nor a strong decrease in the reflection intensity, which is a sign of improved structural stability due to avoiding surface oxidation. Instead, a negligible XRD reflection broadening is observed by the increase in background intensity at higher diffraction angles at 550 • C, which could be due to changes in the average crystallite size, accumulation of defects and local lattice strains. Upon further annealing, no further changes in the reflection profile ( Figure 2b) are observed for a second temperature cycle, indicating completed activation of any intrinsic processes until the applied temperature of 1100 • C.
at Δc/c0~0.5 × 10 , which is much smaller than for the uncapped film. In a second temperature cycle (blue curve in Figure 2c), no anomalous thermal expansion is observed, consistent with the reflection profiles.
The comparison of both experiments suggests that oxidation effects or the interaction of material defects with oxygen species play a prominent role for high-temperature lattice expansion at >800 °C and the irreversible change in respective lattice parameters at room temperature. In this study, the first sample was introduced into the in situ XRD analysis when performing two consecutive heating cycles. The second sample was placed into a quartz tube furnace which was evacuated to 10 −7 mbar. Both The corresponding ∆c(T)/c 0 plots for the first and second full temperature cycle are displayed in Figure 2c. In the first cycle (red curve), the thermal expansion in the low-temperature regime is consistent with uncapped films with (∆c [ Figure 1b) and literature reference data (6.38 × 10 −6 K −1 ) for Al 0.68 Sc 0.32 N/Al 2 O 3 [17]. However, a new and purely intrinsic regime at intermediate temperatures is observed with (∆c[450 • C-650 • C]/c 0 )/∆T~22.2 × 10 −6 / • C with much higher expansion and lower transition temperature T tr~4 50 • C. This value of thermal expansion is almost double compared to the uncapped film at high temperatures with T tr > 600 • C. In the high-temperature regime of >650 • C, the expansion slows down and reverses its sign (∆c[750 • C-1100 • C])/c 0 )/ ∆T~−1.5 × 10 −6 / • C. After cooling down to room temperature, linear thermal expansion over the entire temperature regime with (∆c[1100 • C-30 • C]/c 0 )/ ∆T~−6.0 × 10 −6 / • C is observed and a small irreversible change in the c lattice parameter remains at ∆c/c 0~0 .5 × 10 −3 , which is much smaller than for the uncapped film. In a second temperature cycle (blue curve in Figure 2c), no anomalous thermal expansion is observed, consistent with the reflection profiles.
The comparison of both experiments suggests that oxidation effects or the interaction of material defects with oxygen species play a prominent role for high-temperature lattice expansion at >800 • C and the irreversible change in respective lattice parameters at room temperature. This hypothesis is supported by the comparison of in situ XRD experiments with oxygen in the annealing atmosphere and ex situ XRD experiments without available oxygen in the atmosphere performed on uncapped films of Al 0.73 Sc 0.27 N/Mo(110)/AlN(0001)/Si(001). In this study, the first sample was introduced into the in situ XRD analysis when performing two consecutive heating cycles. The second sample was placed into a quartz tube furnace which was evacuated to 10 −7 mbar. Both samples were treated with identical temperature profiles. In the first heating cycles, a maximum temperature of 800 • C was applied, which is about the temperature at which the positive slope of the intrinsic expansion reverses, whereas a maximum of 1000 • C was used in the second in situ heating hysteresis. After the first ex situ cycle, a second temperature cycle was performed in situ to evaluate the 2θ shifts in the 0002-reflection at higher temperatures.
The reflection profiles and plots of the relative thermal expansion are summarized in Figure 3. Figure 3a shows the respective reflection profiles at selected stages during the experiment. As expected, no oxide hump is observed after the ex situ annealing (blue line), suggesting a purely intrinsic thermal expansion behavior. For the in situ annealed sample (red line), the oxide hump is observed. However, after performing the first cycles, both samples show an irreversible change in lattice parameter of about ∆c/c 0~1 × 10 −3 in the plots of the relative thermal expansion shown in Figure 3b. This could indicate that the observed oxidation does not have a major effect on the expansion at intermediate temperatures and that the intrinsic contribution is dominating. By stopping the first cycle at 800 • C, it is assumed that the intrinsic effects on the anomalously high positive thermal expansion have all been activated and no further reaction would be observed. Indeed, this holds true for both samples in the second cycle up to 800 • C (golden and turquoise lines). By passing the 800 • C mark, comparable thermal expansion is observed when residual oxygen is supplied by the annealing atmosphere which results in a strong oxidation and large irreversible lattice changes ∆c/c 0~3 × 10 −3 after annealing.
positive slope of the intrinsic expansion reverses, whereas a maximum of 1000 °C was used in the second in situ heating hysteresis. After the first ex situ cycle, a second temperature cycle was performed in situ to evaluate the 2θ shifts in the 0002-reflection at higher temperatures.
The reflection profiles and plots of the relative thermal expansion are summarized in Figure 3. Figure 3a shows the respective reflection profiles at selected stages during the experiment. As expected, no oxide hump is observed after the ex situ annealing (blue line), suggesting a purely intrinsic thermal expansion behavior. For the in situ annealed sample (red line), the oxide hump is observed. However, after performing the first cycles, both samples show an irreversible change in lattice parameter of about Δc/c0~1 × 10 −3 in the plots of the relative thermal expansion shown in Figure 3b. This could indicate that the observed oxidation does not have a major effect on the expansion at intermediate temperatures and that the intrinsic contribution is dominating. By stopping the first cycle at 800 °C, it is assumed that the intrinsic effects on the anomalously high positive thermal expansion have all been activated and no further reaction would be observed. Indeed, this holds true for both samples in the second cycle up to 800 °C (golden and turquoise lines). By passing the 800 °C mark, comparable thermal expansion is observed when residual oxygen is supplied by the annealing atmosphere which results in a strong oxidation and large irreversible lattice changes Δc/c0~3 × 10 −3 after annealing. In summary, anomalous high thermal expansion and related irreversible lattice changes have been observed upon thermal activation by annealing of AlScN thin films. The experiments suggest an intrinsic material specific contribution activated at intermediate temperatures of >600-800 °C which is superimposed with extrinsic effects acting in parallel to oxidation of the films at temperatures exceeding 800 °C, if not protected by a surface covering layer. In summary, anomalous high thermal expansion and related irreversible lattice changes have been observed upon thermal activation by annealing of AlScN thin films. The experiments suggest an intrinsic material specific contribution activated at intermediate temperatures of >600-800 • C which is superimposed with extrinsic effects acting in parallel to oxidation of the films at temperatures exceeding 800 • C, if not protected by a surface covering layer.

Part B: Discussion of Scandium Concentration in Al 1−x Sc x N(0001)/Al 2 O 3 (0001) Thin Films
In this section, the discussion is turned towards epitaxial c-axis columnar grown thin films of 1 µm Al 1−x Sc x N on Al 2 O 3 (sample set #2). The recorded XRD 2θ/θ-scans for 1 µm thick Al 1−x Sc x N (0 < x < 0.40) films before and after thermal annealing are shown in Figure 4a. All diffractograms of the as-grown sputtered Al 1−x Sc x N thin films demonstrate exclusive c-axis orientation as well as fixed in-plane orientation described by epitaxial relationships (0001)AlScN//(0001)Al 2 O 3 and (10-10)lScN//(11-20)Al 2 O 3 , respectively [17]. After annealing in situ (with oxygen atmosphere) to 1000 • C, only the 000 ( = 2, 4, 6)-reflections are observed, indicating decent temperature stability of the wurtzite-type phase for all examined Sc concentrations x.  In further analysis of the in situ annealing experiments, the temperature-dependent changes in the reflection profiles are followed individually. In Figure 5, thin films of Al0.77Sc0.23N and Al0.60Sc0.40N are compared to demonstrate the difference between intermediate and high Sc contents. When comparing the reflection profiles, it is directly apparent that the Sc content influences the structural stability and the activation temperature of the observed degradation effects. The evolution of both the symmetric 2θ/θ 0002-and asymmetric ω/2θ 10-15-reflection profiles shows a strong degradation of the initial crystalline quality. The loss of structural coherence is most pronounced in the low-intensity asymmetric reflections, which limits the precise calculation of in-plane parameters at higher temperatures and after annealing. Concerning the XRD reflection broadening, for Sc x = However, a considerable XRD reflection broadening of the 000 -reflection profiles is observed after annealing for Sc concentrations x ≥ 0.23. This is paired with a shift in the reflection position to lower 2θ values. The broadening of the 000 -reflections features asymmetry, with the right tail of the Bragg reflection becoming extended relative to the left tail. No hump on the 0002-reflections is observed, which could indicate a better stability against oxidation for the epitaxial films with fewer grain boundaries. Indeed, rocking curve measurements of the 0002-reflection (Figure 4b) confirm this broadening and provide evidence for a structural degradation by the reduction in maximum diffracted intensity.
The XRD reflection broadening is mainly attributed to reduced sizes of coherently scattering domains, the accumulation of defects and local lattice strain. A shift in the diffraction angle to lower values of 2θ in consequence of annealing is related to the expansion of the c lattice parameter, as discussed in part A. However, the exact origin of the reflection broadening remains speculative without a structure model and Rietveld refinement. Instrumental broadening of the reflections can be neglected due to the high mosaicity of the AlScN films. A detailed analysis of the in-plane and out-of-plane lattice parameter changes using data from the symmetric 2θ/θ-scans and asymmetric ω/2θ 10-15(−) and 10-15(+)-reflection scans [30] is performed and the results are summarized in Figure 4c and Table 2. From the comparison, it is apparent that the high-temperature annealing induces irreversible changes to the lattice parameters of Al 1−x Sc x N. These changes manifest in the reduction in the a parameter and the increase in the c parameter in combination with the degradation of the overall crystal quality (broadening of FWHM). A clear trend is visible in the magnitude of the effect which seems to scale with Sc concentration x indicating an increasing instability of high-Sc alloys at elevated temperatures. For instance, in the case of low-Sc content Al 0.91 Sc 0.09 N films, the irreversible change in the c lattice parameter is ∆c/c 0 0.04% in contrast to high-Sc content Al 0.60 Sc 0.40 N films showing ∆c/c 0~0 .5%, associated with a relative broadening of FWHM of 2.2% and 15.7%, respectively. We note that the film with x = 0.23 shows a very large unexpected change in the a parameter, but follows the general trend regarding the other parameters. The structural origin of the observed reflection broadening in high-Sc content films could also be related to the competition between the hexagonal wurtzite-type phase and the cubic rocksalt-type structure when approaching 46% Sc [23][24][25]38]. High-temperature annealing could result in local phase destabilization and formation of nanosized cubic domains in Sc-enriched regions, e.g., at defect sites or grain boundaries [37]. Such nanosized domains, as well as the migration of defects, e.g., dislocations, will lead to reflection broadening and asymmetry by diffuse scattering and the formation of low-intensity shoulders in diffraction patterns [39,40]. Typically, we would expect a phase transition to be reversible, but here the strong and non-reversible increase in lattice parameters seems to indicate irreversibility.
In further analysis of the in situ annealing experiments, the temperature-dependent changes in the reflection profiles are followed individually. In Figure 5, thin films of Al 0.77 Sc 0.23 N and Al 0.60 Sc 0.40 N are compared to demonstrate the difference between intermediate and high Sc contents. When comparing the reflection profiles, it is directly apparent that the Sc content influences the structural stability and the activation temperature of the observed degradation effects. The evolution of both the symmetric 2θ/θ 0002and asymmetric ω/2θ 10-15-reflection profiles shows a strong degradation of the initial crystalline quality. The loss of structural coherence is most pronounced in the low-intensity asymmetric reflections, which limits the precise calculation of in-plane parameters at higher temperatures and after annealing. Concerning the XRD reflection broadening, for Sc x = 0.23, the elevation of the background tails starts at a temperature of 850 • C, whereas for Sc x = 0.40 these features are already observed at 700 • C and with larger magnitude ( Figure 5).
Micromachines 2022, 13, 1282 10 of 16 0.23, the elevation of the background tails starts at a temperature of 850 °C, whereas for Sc x = 0.40 these features are already observed at 700 °C and with larger magnitude ( Figure  5).  [30] for more information on the method.
As before, the temperature-dependent changes in the reflection positions are used to calculate the relative thermal expansion of both c and a lattice parameters. The resulting thermal expansion is illustrated in Figure 6. Here, the relative changes in the c and a lattice parameters over the temperature range 25-1000 °C are presented for the investigated films in Figure 6a,b, respectively. In this analysis, the transition between the two expansion regimes is evidenced for all Al1-xScxN compositions but not for AlN. Note that the films are not protected against available oxygen species, hence, intrinsic and extrinsic processes adding to the thermal expansion are superimposed. The transition from almost linear and low-value expansion at moderate temperatures into large values of expansion at high temperatures is observed for a and c. This expansion of the lattice is isotropic which is visible from the almost constant c/a ratio plotted in Figure 6c. The transition temperature Ttr seems to be related to the Sc content of the film, as well as the magnitude of the expansion. Roughly estimated values from the plots are Ttr ~ 850 °C for Al0.91Sc0.09N and Ttr ~ 550 °C for Al0.60Sc0.40N films as indicated by the vertical dotted lines in Figure 6b. In detail, the As before, the temperature-dependent changes in the reflection positions are used to calculate the relative thermal expansion of both c and a lattice parameters. The resulting thermal expansion is illustrated in Figure 6. Here, the relative changes in the c and a lattice parameters over the temperature range 25-1000 • C are presented for the investigated films in Figure 6a,b, respectively. In this analysis, the transition between the two expansion regimes is evidenced for all Al 1−x Sc x N compositions but not for AlN. Note that the films are not protected against available oxygen species, hence, intrinsic and extrinsic processes adding to the thermal expansion are superimposed. The transition from almost linear and low-value expansion at moderate temperatures into large values of expansion at high temperatures is observed for a and c. This expansion of the lattice is isotropic which is visible from the almost constant c/a ratio plotted in Figure 6c. The transition temperature T tr seems to be related to the Sc content of the film, as well as the magnitude of the expansion. Roughly estimated values from the plots are T tr ∼850 • C for Al 0.91 Sc 0.09 N and T tr ∼550 • C for Al 0.60 Sc 0.40 N films as indicated by the vertical dotted lines in Figure 6b. In detail, the thermal expansion behavior in the low-temperature regime (<550 • C) is highly comparable for all Sc concentrations, in agreement with previous studies within this temperature range [17]. In strong contrast, the high-temperature regime is characterized by a manifold increase in the expansion, depending on the Sc concentration (up to ∼8-fold for Al 0.60 Sc 0.40 N). After cooling back to room temperature, the lattice parameters show irreversible changes as discussed for Figure 4 and Table 2.
thermal expansion behavior in the low-temperature regime (<550 °C) is highly comparable for all Sc concentrations, in agreement with previous studies within this temperature range [17]. In strong contrast, the high-temperature regime is characterized by a manifold increase in the expansion, depending on the Sc concentration (up to ~8-fold for Al0.60Sc0.40N). After cooling back to room temperature, the lattice parameters show irreversible changes as discussed for Figure 4 and Table 2. For direct comparison with fiber textured samples from set #1, XRD measurements during two temperature cycles were conducted for the Al0.68Sc0.32N film, cf. Figure 7. The displayed temperature cycle and reflection profiles are highly congruent to the data recorded on the fiber textured thin films with similar composition. An irreversible change of Δc/c0~3.5 × 10 −3 remains after annealing at 1000 °C ( Figure 7a) and no further changes are observed in the second temperature hysteresis, congruent with the measured reflection profiles shown in Figure 7b,c. For direct comparison with fiber textured samples from set #1, XRD measurements during two temperature cycles were conducted for the Al 0.68 Sc 0.32 N film, Figure 7. The displayed temperature cycle and reflection profiles are highly congruent to the data recorded on the fiber textured thin films with similar composition. An irreversible change of ∆c/c 0~3 .5 × 10 −3 remains after annealing at 1000 • C (Figure 7a) and no further changes are observed in the second temperature hysteresis, congruent with the measured reflection profiles shown in Figure 7b,c.
Micromachines 2022, 13, 1282 11 of 16 thermal expansion behavior in the low-temperature regime (<550 °C) is highly comparable for all Sc concentrations, in agreement with previous studies within this temperature range [17]. In strong contrast, the high-temperature regime is characterized by a manifold increase in the expansion, depending on the Sc concentration (up to ~8-fold for Al0.60Sc0.40N). After cooling back to room temperature, the lattice parameters show irreversible changes as discussed for Figure 4 and Table 2. For direct comparison with fiber textured samples from set #1, XRD measurements during two temperature cycles were conducted for the Al0.68Sc0.32N film, cf.

Discussion
The described experiments reveal high-temperature effects related to structural degradation and irreversible anomalous non-linear thermal expansion behavior in AlScN thin films depending on Sc content and film microstructure. The non-linear thermal expansion is described by a transition from an initial expansion regime to a fast-expanding hightemperature regime which is divided into intrinsic and extrinsic contributions, dependent on the availability of oxygen. Available oxygen is believed to lead to the oxidation of the material dependent on the film microstructure (oxidation was only observed for fiber textured films with higher density of grain boundaries) but also enhances the expansion effect size at T > 800 • C and stabilizes the increased lattice parameters. The effect size of high-temperature expansion scales with the Sc content of the films, which could be related to a higher defect density leading to the increased destabilization and easier oxidation. A further indicator for the scaling defect density is the increasing magnitude of the peak broadening, which was discussed as an indicator for defect movement as well. Besides dislocations or grain boundaries, point defects such as nitrogen vacancies V N are one major type of defect in AlN and AlScN thin films. Previous studies by Harris et al. have demonstrated that the wurtzite-type structure of AlN can incorporate up to 0.75 at% oxygen under thermal equilibrium conditions by substitution of N (O N ) associated with the formation of Al vacancies (V Al ) [41]. Further insights into the structure relation of AlN films and oxygen are provided by DFT modeling by Gasparotto et al. [42] showing that oxygen can have significant influence on the lattice parameters.
The discussed examples show potential interrelations of oxygen-induced defect structures and changes to the lattice parameters. As observed in this work on AlScN thin films, such phenomena could provide valid explanation for the discussed intrinsic and extrinsic contributions to the anomalous thermal expansion at high temperatures. In addition, Sc has strong affinity for oxygen [43][44][45], and the intrinsic oxygen contamination of AlScN thin films is supposed to scale with the Sc content as well. In this respect, Sc-and oxygen-enriched grain boundaries were already evidenced [37] and the photoluminescence emission of low-Sc films (x = 0.05) was already revealed to be dominated by oxygen defects [46]. Further, modeling of the point defects in rocksalt AlScN resulted in preferable defect complexes of substitutional and interstitial oxygen (O N + O i ) [47]. The observed oxygen content-dependent destabilization of the lattice in Al-O-N is also consistent with the increasing degradation of the crystal quality with Sc concentration and the formation of oxide phases.
To support the above discussion, we designed a third sample of a SiN x -capped AlScN film with high concentration of Sc x = 0.4 grown epitaxially on a sapphire substrate, i.e., SiN x /Al 0.6 Sc 0.4 N(0001)/epi-Mo(110)/AlN(0001)/Al 2 O 3 (0001). For such systems, previous work [29] demonstrated that the special epi-Mo(110) electrode provides a growth template for AlScN with one-dimensional single crystalline properties and low defect density. The following assumptions are made: First, the SiN barrier should protect the thin film from oxidation and interaction of oxygen species with the internal defect structure at high temperatures. Second, by choosing a high amount of Sc, any effect magnitude is expected to be large. Third, this AlScN thin film exhibits high crystalline quality with an FWHM~0.7 • (from XRC) indicating highly oriented columnar grains originating from epitaxial growth and a low defect density. Hence, the effect of any intrinsic, defect-driven anomalous expansion should be strongly limited. Indeed, the recorded thermal expansion during the first temperature cycle shown in Figure 8 shows almost linear expansion featuring no anomalous behavior. This observation, although not expected so clearly, provides supporting evidence for the above hypothesis. In summary, the irreversible changes in the lattice parameters of Al1-xScxN/Al2O3 when exposed to atmospheric oxygen present in low-vacuum conditions are comparable to fiber textured systems. The estimated transition temperatures between the low-temperature expansion and high-temperature expansion regimes and the magnitude of the superimposed intrinsic and extrinsic contributions show a clear trend with the Sc content of the films. With increasing Sc content, the transition temperature is decreased and the effect sizes of thermal expansion and degradation of crystallinity are increased. The anomalous thermal expansion can only be activated by temperature once. Further, the microstructure and possibly the defect structure seem to impact the stability to oxidation of the films, which seems not to be interrelated with the extrinsic part of high-temperature thermal expansion. However, at very high temperatures >800 °C, hypothetically, oxygen species could diffuse into the material via defects and grain boundaries and interact with thermally activated defect sites to further drive the lattice expansion and act as obstacles to stabilize the expanded lattice. Scandium is known to destabilize the AlN lattice by its larger cation size and structural preference for octahedral coordination. Hence, increasing its concentration in AlN could induce higher defect densities, which are seemingly possible to activate at lower temperatures by the flattened energy landscape.

Conclusions
The thermal stability and temperature-induced effects of AlScN thin film samples with different microstructures based on the growth template were investigated by in situ XRD. The degradation of the crystalline quality and a remnant lattice expansion in the cdirection of up to 0.5% for epitaxial Al0.60Sc0.40N(0001)/Al2O3(0001) were observed as a function of Sc concentration. There is first evidence that the remnant expansion is related to the activation of intrinsic defects and the films' oxygen affinity at elevated temperatures, which are accounted for as intrinsic and extrinsic sources of anomalous thermal expansion. The understanding of the exact details of these phenomena provides opportunity for further investigations on the exact type of defect structures using more advanced methods, such as positron annihilation spectroscopy [48][49][50]. The detailed understanding of the intrinsic defect structures and temperature activation effects is of high importance for the integration of sputtered AlScN layers into sandwich structures, e.g., for In summary, the irreversible changes in the lattice parameters of Al 1−x Sc x N/Al 2 O 3 when exposed to atmospheric oxygen present in low-vacuum conditions are comparable to fiber textured systems. The estimated transition temperatures between the low-temperature expansion and high-temperature expansion regimes and the magnitude of the superimposed intrinsic and extrinsic contributions show a clear trend with the Sc content of the films. With increasing Sc content, the transition temperature is decreased and the effect sizes of thermal expansion and degradation of crystallinity are increased. The anomalous thermal expansion can only be activated by temperature once. Further, the microstructure and possibly the defect structure seem to impact the stability to oxidation of the films, which seems not to be interrelated with the extrinsic part of high-temperature thermal expansion. However, at very high temperatures >800 • C, hypothetically, oxygen species could diffuse into the material via defects and grain boundaries and interact with thermally activated defect sites to further drive the lattice expansion and act as obstacles to stabilize the expanded lattice. Scandium is known to destabilize the AlN lattice by its larger cation size and structural preference for octahedral coordination. Hence, increasing its concentration in AlN could induce higher defect densities, which are seemingly possible to activate at lower temperatures by the flattened energy landscape.

Conclusions
The thermal stability and temperature-induced effects of AlScN thin film samples with different microstructures based on the growth template were investigated by in situ XRD. The degradation of the crystalline quality and a remnant lattice expansion in the c-direction of up to 0.5% for epitaxial Al 0.60 Sc 0.40 N(0001)/Al 2 O 3 (0001) were observed as a function of Sc concentration. There is first evidence that the remnant expansion is related to the activation of intrinsic defects and the films' oxygen affinity at elevated temperatures, which are accounted for as intrinsic and extrinsic sources of anomalous thermal expansion. The understanding of the exact details of these phenomena provides opportunity for further investigations on the exact type of defect structures using more advanced methods, such as positron annihilation spectroscopy [48][49][50]. The detailed understanding of the intrinsic defect structures and temperature activation effects is of high importance for the integration of sputtered AlScN layers into sandwich structures, e.g., for ferroelectric field-effect transistor memory capacitors or ferroelectric tunnel junctions. That is especially relevant with respect to downscaling of the layer thickness and high-temperature operation, where oxidation of the functional layer has to be prohibited and an anomalously high expansion of the AlScN crystal lattice puts increased stresses on any neighboring crystalline layers to avoid device failure via crack formation or delamination. In conclusion, our investigation emphasizes the requirement of a low density of material defects in AlScN thin films when operating at high temperatures and the benefit of integration with a protecting top layer, e.g., a temperature-resistant electrode such as Mo or NbN.