Structural characterization of epitaxial ScAlN films grown on GaN by low-temperature sputtering

ScAlN has garnered substantial attention for its robust piezoelectric and ferroelectric properties, holding promise for diverse electronic device applications. However, the interplay between its structural attributes and physical properties remains poorly understood. This study systematically elucidates the structural characteristics of epitaxial ScAlN films grown on GaN by low-temperature sputtering. Correlations between Sc composition, lattice constants, and film strains were revealed utilizing high-resolution X-ray diffraction, reciprocal space mapping, and machine learning analyses. Our machine-learning model predicted c-axis lattice constants of ScAlN grown on GaN under various conditions and suggested that sputtering permits coherent growth over a wide compositional range. These findings advance the understanding of ScAlN and provide valuable insights for the research and development of novel ScAlN-based devices.


E
[3] Notably, Akiyama et al. demonstrated a substantial enhancement in the piezoelectric constant of Sc-doped AlN, which was five times that of pure AlN. 4) The heightened piezoelectric response, combined with the high-temperature stability of ScAlN, has driven its widespread adoption in various piezoelectric devices.Additionally, Fichtner et al. revealed the ferroelectric nature of ScAlN, where its polarization direction can be altered through external voltage application. 5)Leveraging the robust piezoelectric and ferroelectric characteristics of ScAlN holds promising potential for diversifying the functionality of nitride semiconductor devices.[13][14][15][16][17][18][19][20][21][22][23][24][25][26][27] Despite its significant potential, ScAlN remains a relatively new material with various unresolved questions regarding the interplay between its structural attributes, such as crystal structure, lattice constants, and physical properties.Notably, ScAlN predominantly crystallizes into a wurtzite structure at lower Sc compositions, and a Sc composition of 18% results in a lattice constant that matches that of GaN.Theoretical calculations suggest that increased Sc composition may induce a transition from the wurtzite to a layered ScN structure. 28)owever, the exact dynamics of this structural transition remain elusive.Furthermore, as the growth method of ScAlN has not yet been standardized, exploration continues across a variety of growth techniques, including molecularbeam epitaxy (MBE), metal-organic chemical vapor deposition (MOCVD), and sputtering.Consequently, the lack of explicit theories and guidelines for achieving single-crystal films presents significant obstacles to establishing the device structure and fabrication process for novel electronic devices employing ScAlN, such as FeFETs.Machine learning is a useful way to reveal the ScAlN growth mechanisms hidden behind various experimental results.For example, machine learning models can be used to virtually run experimental conditions that have not actually been performed and predict the properties of the thin films that will be grown.This study systematically elucidates the structural characteristics of epitaxial ScAlN grown by sputtering at low temperatures with the aid of machine learning.Sputtering offers high degree of freedom in growth conditions due to its capability to grow films at lower temperatures.This versatility makes it a suitable technique for exploring a variety of growth parameters of ScAlN.In addition to academic significance, addressing these issues can catalyze technological advancements and unlock new applications in electronics.
Using a sputtering method, we epitaxially grew ScAlN films on GaN/sapphire template substrates prepared by MOCVD.The sputtering was performed using a mixture of purified N 2 and Ar gases.Sc and Al were supplied from independent sputtering sources, with the Sc composition of the ScAlN films controlled by adjusting the power supplied to each target.The Sc composition and lattice constants of the ScAlN films were determined using high-resolution X-ray diffraction (XRD), reciprocal space mapping, and scanning electron microscopy with energy-dispersive X-ray spectroscopy.Information on the ScAlN films, including film thickness, composition, crystal growth method, growth temperature, and lattice constants, was collected from the literature 1,22,[29][30][31][32][33][34][35][36][37][38] and compiled into a database. Sbsequently, this database was analyzed using machine learning based on neural networks.For the machine learning analysis, we employed the Sony Neural Network Console as our software platform.The input parameters for the model, consisting of standardized values for Sc composition, growth temperature, and growth method, were designed to predict the c-axis lattice constant.To counteract the effects of overfitting due to the limited data set, we incorporated weight decay through L2 regularization.This study achieved a reduced epitaxial growth temperature of 300 °C [the lowest curve in Fig. 2(a)].The c-axis lattice constants, calculated from the peak positions of ScAlN 0002, are summarized in Fig. 2(b).For ScAlN films grown at temperatures below 450 °C, the c-axis lattice constant increases monotonically with increasing Sc composition.However, the ScAlN films grown at 600 °C decrease the caxis length as the Sc composition exceeds 0.20.This finding suggests that the c-axis lattice constant of ScAlN does not follow Vegard's law, indicating the involvement of complex factors in determining the c-axis lattice constant.The influence of strain requires further investigation, including a detailed examination of the a-axis lattice constants.
Figure 3(a) shows the 10 15 reciprocal space mapping of Sc 0.06 Al 0.94 N (27 nm)/GaN grown at 450 °C.The reciprocal lattice points of ScAlN are separated into two components: one set represents coherent growth on GaN, and the other indicates lattice relaxation.Peaks originating from these two components were observed in the 0002 2θ/ω scan of this sample (not shown).The Sc 0.06 Al 0.94 N exhibits properties similar to AlN owing to the lower Sc composition, resulting in lattice relaxation.However, some regions still exhibit coherent growth.In Fig. 3(b), the 10 15 reciprocal space mapping of Sc 0.17 Al 0.83 N (69 nm) grown at 600 °C is presented, which shows fully coherent growth on GaN.With the increase in the Sc concentration, the lattice constant of ScAlN matches that of GaN, promoting the coherent growth of ScAlN.
We conducted reciprocal space mapping on ScAlN/GaN samples grown under various conditions using sputtering to investigate the conditions under which ScAlN shows coherent growth.The results are shown in Fig. 4. In the graph, the filled circles indicate coherent growth, whereas the open circles represent samples with lattice relaxation.The graph shows that at low Sc composition, lattice relaxation was observed for ScAlN.However, coherent growth occurs without lattice relaxation for ScAlN with high-Sc composition, specifically when the growth temperature was below 600 °C.Coherent growth remains achievable even with Sc composition exceeding 0.30.Epitaxial growth of ScAlN on GaN by MBE shows that the region where coherent growth is possible is limited to a Sc composition of approximately 0.20. 37)These results suggest that sputtering deposition enables growth at low temperatures and is favorable for coherent growth, even for ScAlN films with high-Sc concentrations.
Next, we discuss the c-axis lattice constants of ScAlN. Figure 5 presents the plot of the c-axis lengths of our data and those reported in the literature for ScAlN against its Sc composition.The c-axis lattice constants of ScAlN are not uniquely determined and exhibit variations.This variation suggests a deviation from Vegard's law.However, when focusing on the plots for ScAlN grown above 700 °C, the caxis lattice constant decreases as the Sc composition exceeds 0.10.Ambacher et al. proposed a model that describes lattice constants of ScAlN with a quadratic function. 39)The red line represents the c-axis lattice constant of fully relaxed ScAlN, and the blue line represents the c-axis lattice constant of ScAlN coherently grown on GaN, calculated using the quadratic model.High-Sc-composition ScAlN exhibits a shorter lattice constant than the red line, indicating that these    22,[29][30][31][32][33][34][35][36][37][38][39] (open circles) and our sputtering experiments (filled circle).Blue and red lines represent coherent and fully relaxed growth models, respectively.

011002-3
© 2023 The Author(s).Published on behalf of The Japan Society of Applied Physics by IOP Publishing Ltd ScAlN samples deform into crystal structures other than the wurtzite structure.These results highlight the potential of determining structural characteristics (coherent growth feasibility, crystal structure) from Sc composition and c-axis lattice constants.
The compositional dependence of the c-axis lattice constant in ScAlN could not be explained by existing theoretical models.Therefore, we constructed a machine learning model for predicting the c-axis lattice constant based on growth conditions.The predicted c-axis lattice constants of ScAlN grown using sputtering and MBE are plotted in Fig. 6.The growth temperatures were set to typical values, specifically 450 °C for sputtering [Fig.6(a)] and 600 °C for MBE [Fig.6(b)].The c-axis lattice constant was predicted for film thicknesses of 10 and 200 nm.The dashed lines represent lattice constants, which were calculated based on the quadratic model, illustrating relaxed and coherently grown scenarios.When comparing these two graphs, it becomes evident that sputtering closely follows the theoretical line based on the quadratic model over a wide compositional range compared to MBE.The blue coherent line aligns with Sc composition around 0.20, indicating that sputtering promotes coherent growth.Furthermore, in the MBE case, the red line for the machine learning model (quasi-fully relaxed) diverges from the line for the quadratic model for Sc compositions exceeding 0.20.In contrast, no such deviation is observed in the sputtering case.The machine learning model indicates that high-Sc-composition ScAlN grown using sputtering crystallizes into the wurtzite structure.The advantage of employing machine learning in this study is that we can collect data from a variety of sources with different conditions and fit them in a complementary manner.It is noteworthy that only the curve for c-axis lattice constant predicted by machine learning for a particular set of conditions (at low temperatures by sputtering) can be explained by the quadratic function model.

Conclusion
This study highlights an elongation of the c-axis lattice constant as the Sc composition increases for ScAlN when coherently grown on GaN.This behavior could be accurately described by a quadratic model that deviates from Vegard's law.Notably, the quadratic model provides a good fit for ScAlN films grown by sputtering at low temperatures.In contrast, ScAlN films with a high-Sc composition, grown at elevated temperatures, deviate from this quadratic model, suggesting the existence of phases other than the wurtzite structure.The machine learning model developed in this study offers the potential to reveal features and correlations that elude theoretical analysis, assisting in the prediction and analysis of the structural properties of ScAlN.Furthermore, the feature information database generated in this study could serve as a valuable resource for advancing the research and development of novel devices that utilize ScAlN.
(a) Fig. 6.Compositional dependence of the c-axis lattice constants of ScAlN grown on GaN (a) by sputtering at 450 °C and (b) by MBE at 600 °C, as predicted using a machine learning (ML) model.The dashed lines represent the lattice constants of coherent and relaxed ScAlN on GaN, as determined using the quadratic model (same as those in Fig. 5).Our experimental data are also plotted as filled circles in (a).

011002-4
© 2023 The Author(s).Published on behalf of The Japan Society of Applied Physics by IOP Publishing Ltd

Figure 1
Figure 1 shows the atomic force microscope (AFM) images of the ScAlN film surface.Cracks are present on the surface of Sc 0.03 Al 0.97 N, epitaxially grown at 700 °C [Fig.1(a)].The presence of cracks could be attributed to the tensile stress exerted on the ScAlN film by the GaN/sapphire substrate.The structural characteristics of this film are similar to AlN owing to the relatively low Sc composition, suggesting that increasing the film thickness might be challenging.The film grown at 600 °C exhibited an atomically flat surface, with a Sc composition ranging from 17% to 31% [Figs.1(b)-1(d)].Additionally, the growth of Sc 0.17 Al 0.83 N at 450 °C did not degrade the surface flatness [Fig.1(e)].As shown in Fig. 1(f), the reflection high-energy electron diffraction (RHEED) pattern of Sc 0.17 Al 0.83 N, grown at 450 °C, exhibited streaks with spots, indicating the epitaxial growth of wurtzite-type ScAlN.The sputtering method demonstrated the successful growth of flat ScAlN films at temperatures below 600 °C.Growing these films at temperatures below 600 °C could be significant technologically and industrially, as it may expand material options and reduce manufacturing costs.Figure 2(a) shows the XRD 2θ/ω scans for ScAlN films grown on GaN/sapphire by sputtering.The 0002 peak of ScAlN grown at 700 °C exhibits broadening, suggesting inhomogeneous growth of ScAlN with varying Sc compositions.Lowering the growth temperature to 600 °C facilitates the epitaxial growth of ScAlN with relatively high-Sc compositions, such as Sc 0.31 Al 0.69 N and Sc 0.25 Al 0.75 N. The peak intensity of Sc 0.17 Al 0.83 N epitaxially grown at 600 °C significantly increases, indicating a high crystalline quality.

Figure 2 (
Figure 1 shows the atomic force microscope (AFM) images of the ScAlN film surface.Cracks are present on the surface of Sc 0.03 Al 0.97 N, epitaxially grown at 700 °C [Fig.1(a)].The presence of cracks could be attributed to the tensile stress exerted on the ScAlN film by the GaN/sapphire substrate.The structural characteristics of this film are similar to AlN owing to the relatively low Sc composition, suggesting that increasing the film thickness might be challenging.The film grown at 600 °C exhibited an atomically flat surface, with a Sc composition ranging from 17% to 31% [Figs.1(b)-1(d)].Additionally, the growth of Sc 0.17 Al 0.83 N at 450 °C did not degrade the surface flatness [Fig.1(e)].As shown in Fig. 1(f), the reflection high-energy electron diffraction (RHEED) pattern of Sc 0.17 Al 0.83 N, grown at 450 °C, exhibited streaks with spots, indicating the epitaxial growth of wurtzite-type ScAlN.The sputtering method demonstrated the successful growth of flat ScAlN films at temperatures below 600 °C.Growing these films at temperatures below 600 °C could be significant technologically and industrially, as it may expand material options and reduce manufacturing costs.Figure 2(a) shows the XRD 2θ/ω scans for ScAlN films grown on GaN/sapphire by sputtering.The 0002 peak of ScAlN grown at 700 °C exhibits broadening, suggesting inhomogeneous growth of ScAlN with varying Sc compositions.Lowering the growth temperature to 600 °C facilitates the epitaxial growth of ScAlN with relatively high-Sc compositions, such as Sc 0.31 Al 0.69 N and Sc 0.25 Al 0.75 N. The peak intensity of Sc 0.17 Al 0.83 N epitaxially grown at 600 °C significantly increases, indicating a high crystalline quality.

Fig. 4 .
Fig. 4. Coherent growth conditions for ScAlN/GaN samples grown by sputtering.Filled circles represent coherent growth, whereas open circles indicate lattice relaxation.Film thickness is annotated above each plot.