Sputter epitaxy of ScAlN films on GaN high electron mobility transistor structures

ScAlN, an alloy of AlN and ScN, has recently garnered significant attention due to its unique physical characteristics such as a significantly enhanced piezoelectric coefficient, high spontaneous polarization, and ferroelectricity. These properties are not observed in conventional group III nitrides.^1–5 These properties make ScAlN highly promising for a broad range of applications in advanced electronic and optoelectronic devices. In 2009, Akiyama et al. reported that adding Sc to AlN increases its piezoelectric coefficient by more than several times.⁶ The addition of Sc makes the arrangement of atoms within the crystal more susceptible to changes under stress and reduces the elastic constant C₃₃, leading to an increase in the piezoelectric coefficient.⁷ This remarkable enhancement is particularly beneficial for film bulk acoustic resonator devices, which operate at resonant frequencies in the GHz range and are vital for the development of fifth-generation (5G) wireless systems.

In 2019, Fichtner et al. reported that ScAlN films with up to 40% Sc composition exhibited ferroelectric properties.⁸ Moreover, the ferroelectricity of ScAlN has been confirmed by various groups.^9–13 These findings suggest that the ferroelectricity of ScAlN can be used to implement ferroelectric gates in transistors, potentially enabling the fabrication of ferroelectric field-effect transistors (FeFETs).^14–16 FeFETs are promising for nonvolatile memory applications, where the ability to retain data without power is highly advantageous. The introduction of ferroelectric properties into hexagonal group III-nitride semiconductors paves the way for advanced memory and logic devices.^1,2

Furthermore, ScAlN is a promising barrier layer material for GaN high electron mobility transistors (HEMTs). ScAlN can be lattice-matched with GaN¹⁷ and exhibits spontaneous polarization comparable to that of AlN,¹⁸ potentially enabling strain-controlled barrier layers and generating a high-density two-dimensional electron gas (2DEG) within the GaN channel. Lattice matching is essential for minimizing the generation of defects at the interface, which can significantly impact electron mobility and overall device performance. Various Sc compositions for lattice matching with GaN, ranging from 11% to 20%,^17,19,20 have been reported; however, the precise value for lattice matching remains uncertain. This uncertainty highlights the need for further investigation to determine the ideal Sc composition to maximize device efficiency and reliability.

To date, ScAlN has been epitaxially grown using various techniques, including sputtering,^21–24 metalorganic chemical vapor deposition (MOCVD),^25–29 and molecular-beam epitaxy (MBE).^{10,19,30–40} Each of these methods offers distinct advantages and limitations. MOCVD and MBE are well-known for producing high-quality single-crystal films with excellent structural and electrical properties. However, these techniques often require high processing temperatures and complex setups. In contrast, sputtering offers a cost-effective and scalable approach with lower processing temperatures, making it suitable for industrial applications.⁴¹ Despite these advantages, the properties of ScAlN epitaxially grown on GaN by sputtering are not yet fully understood. In our previous research, we epitaxially grew ScAlN on GaN/sapphire substrates with Sc compositions up to 30% using the sputtering method and performed structural characterization of ScAlN films using machine-learning analysis.²² Additionally, we conducted spectroscopic ellipsometry characterization of ScAlN epitaxially grown on GaN bulk substrates.²⁴ Nevertheless, it remains uncertain whether ScAlN produced by sputtering can be effectively utilized as a barrier layer in HEMTs. In this study, we epitaxially grew ScAlN on AlGaN/GaN HEMT structures and evaluated their structural and electrical properties. This work aims to address the existing knowledge gaps regarding the performance of sputtered ScAlN and to explore its potential as a viable material for enhancing the characteristics of GaN HEMTs.

Using the sputtering method, we epitaxially grew ScAlN thin films with various Sc compositions, ranging from 5% to 20%, on AlGaN (∼3 nm)/AlN (∼1 nm)/unintentionally-doped GaN/6H-SiC template substrates (AlGaN/GaN HEMT structures) fabricated via MOCVD. The Al composition in the AlGaN layer was approximately 20%. The growth chamber was evacuated using a turbomolecular pump to achieve a background pressure of 4 × 10⁻⁸Pa, and high-purity N₂ and Ar gases (99.9999%) were introduced through a purifier to ensure a clean environment. Sputtering was conducted in a mixed N₂ and Ar gas atmosphere at a growth temperature of 600 °C and a growth pressure of 1.0 Pa. The nitrogen gas ratio in the sputtering pressure was set to 10%. Independent 2-inch-diameter targets with purities of 99.9% (Sc) and 99.999% (Al) were used, with power levels adjusted in the range of 20–35 W for Sc and 25–45 W for Al, and sputtering pulse frequencies of 4 kHz for Sc and 10 kHz for Al. A pulsed DC power supply was employed, with voltages of −250 V applied to the Al target and −650 V to the Sc target. The Sc composition (5%–20%) and growth rate (0.7–1.4 nm min⁻¹) were controlled by adjusting the power supplied to the targets, resulting in film thicknesses of 20–25 nm. It was confirmed that variations in growth rates did not affect the crystallinity of the thin films. The lattice constants of the ScAlN films were determined using high-resolution x-ray diffraction (XRD), and the Sc composition was measured using a transmission electron microscope (TEM) equipped with an energy-dispersive x-ray spectrometer. The electrical properties of the fabricated ScAlN/AlGaN/GaN HEMT structures were evaluated using Hall effect measurements. Indium electrodes were formed at the four corners of the sample and connected to the terminals of the Hall effect measurement system using gold wires.

Figure 1 shows the atomic force microscopy (AFM) images of ScAlN surfaces with Sc compositions of 7%, 8%, 10%, and 15%, grown on an AlGaN/GaN HEMT structure. Each surface consists of grains several tens of nanometers in size. Epitaxial growth was performed under nitrogen-rich conditions, resulting in insufficient migration of adatoms on the substrate surface and producing a densely packed grain surface morphology. These nitrogen-rich conditions were deliberately chosen to prevent the formation of Sc or Al droplets on the film surface. The root mean square (RMS) surface roughness values were 0.28, 0.24, 0.43, and 0.75 nm for Sc compositions of 7%, 8%, 10%, and 15%, respectively. The RMS values increased gradually with higher Sc compositions. To achieve smoother surfaces, it is necessary to optimize growth conditions to enhance precursor adatom migration on the surface. For example, reducing the growth rate or decreasing the supply of nitrogen plasma to levels that avoid metal droplet formation could be effective.

Figure 2(a) shows the XRD 2θ/ω scan of ScAlN films epitaxially grown on the AlGaN/GaN HEMT structure using the sputtering method. For all samples, diffraction peaks corresponding to AlGaN 0002, GaN 0002, and 6H-SiC 0006 were observed, along with a diffraction peak for ScAlN 0002. As the Sc composition increased, the diffraction angle of ScAlN 0002 shifted to lower angles, indicating an increase in the c-axis lattice constant with increasing Sc composition. This shift suggests that the introduction of Sc atoms causes lattice expansion, and in-plane stress further contributes to elongation along the c-axis. The gradual shift observed in the XRD patterns confirms that the lattice constants vary consistently with Sc composition, driven by mechanisms more complex than those described by Vegard's law.

To investigate the in-plane lattice constants of the ScAlN films, reciprocal space mapping (RSM) of the (10 $1¯$5) plane was performed, as shown in Fig. 2(b). The q_x coordinates of the ScAlN (10 $1¯$5) plane with Sc composition ranging from 5% to 20% matched those of the GaN (10 $1¯$5) plane. This q_x alignment indicates that the a-axis lattice constants of these ScAlN films are coherent with those of GaN, suggesting that the films are coherently grown on the underlying AlGaN/GaN HEMT structure. Additionally, the RSM results confirmed that AlGaN grew coherently on GaN. The sharpness of the diffraction peaks for ScAlN indicates the high crystallinity of the epitaxial ScAlN films. The absence of additional peaks or significant peak broadening suggests that there are no major secondary phases or significant stress relaxation within the ScAlN films. As the Sc composition increased, the reciprocal lattice points of ScAlN (10 $1¯$5) shifted toward lower q_z values, indicating a reduction in the interplanar spacing of the c-plane. This observation is consistent with the peak shift observed in the XRD 0002 measurements. When using SiC substrates, the diffraction peaks of SiC and ScAlN 0002 often overlapped, hindering the accurate determination of lattice constants. In such cases, RSM provides a more reliable method for extracting lattice constants.

These results demonstrate that high-quality, coherent ScAlN films can be grown on AlGaN/GaN HEMT structures using the sputtering method. This method enables precise control of the Sc composition in ScAlN and the corresponding lattice constants, thereby enabling the optimization of material properties for specific device applications. The coherent growth and high crystalline quality achieved in this study suggest that ScAlN could enhance the performance of GaN-based electronic devices.

The closed circles in Fig. 3 represent the c-axis lattice constants of ScAlN calculated from the reciprocal lattice points of ScAlN (10 $1¯$5). It is well-established that the lattice constants of ScAlN do not follow Vegard's law. Ambacher et al. proposed that the lattice constants of ScAlN can be modeled using a quadratic function.⁴² We used this model to calculate the c-axis lattice constants for coherently grown and relaxed ScAlN on GaN, represented by the blue and red lines, respectively. The experimental data closely align with the quadratic function model for coherently grown ScAlN on GaN. An exception is observed for the diffraction angle of Sc_0.08Al_0.92N, which deviates from the expected trend, as shown in Fig. 2(a). Although this sample is coherently grown and the corresponding values are expected to lie on the blue line, its lattice constant is longer than the predicted value. This suggests that the elastic constant may be smaller than anticipated. We hypothesize that unintended variations in experimental conditions could have contributed to this discrepancy, and further investigations are planned to identify the underlying causes. Additionally, Fig. 3 includes the c-axis lattice constants of ScAlN grown by sputtering^21,22,24 and MBE^{33,35,37,40,43–46} as reported in the literature. The sputtering data closely follow the coherent growth line up to an Sc composition of 20%. Beyond this composition, the lattice constants shift along the relaxed line; this is characteristic of ScAlN films grown by sputtering.

Figure 4(a) shows an annular bright-field scanning transmission electron microscopy (ABF-STEM) image of the Sc_0.20Al_0.80N/AlGaN/AlN/GaN HEMT structure. The image shows that the wurtzite-structured ScAlN is epitaxially grown on the AlGaN layer. The ScAlN/AlGaN interface is abrupt, indicating that ScAlN grows coherently from the first layer. Additionally, the abruptness of the AlGaN/AlN/GaN interfaces is maintained, demonstrating the robustness of the HEMT structure against the epitaxial growth of ScAlN via sputtering. The ScAlN epitaxial layer inherits the polarity of the HEMT structure and exhibits metal polarity, as shown in the magnified images of ScAlN [Fig. 4(b)] and GaN [Fig. 4(c)]. No regions of nitrogen polarity are observed within the ScAlN layer, further confirming the uniformity of the growth process.

Finally, Hall effect measurements were conducted for the ScAlN/AlGaN/GaN HEMT structures, and the sheet carrier density (n_s) and mobility (μ) of the samples are summarized in Fig. 5. Two types of AlGaN/GaN HEMT structures (HEMT A and B) were prepared. The n_s and μ values were 3.8 × 10¹² cm⁻² and 1330 cm² V⁻¹ s⁻¹, respectively, for HEMT A, and 6.1 × 10¹² cm⁻² and 1410 cm² V⁻¹ s⁻¹, respectively, for HEMT B. For the sample with Sc_0.06Al_0.94N epitaxially grown on HEMT A, the sheet resistance was 1087 Ω □⁻¹, the n_s was 6.1 × 10¹² cm⁻², and μ was 935 cm² V⁻¹ s⁻¹. The growth of Sc_0.06Al_0.94N slightly increased the n_s, independent of the HEMT substrate. We conducted simulations using 1D Poisson⁴⁷ to calculate the band profiles and electron density of ScAlN/AlGaN/AlN/GaN heterostructures. The results, presented in the supplementary material (Fig. S1), show that 2DEG forms at two interfaces. The sum of the 2DEG densities at these interfaces was used as the sheet carrier density and added to Fig. 5 for comparison with experimental results. The discrepancy between the theoretical and experimental 2DEG densities was found to be independent of Sc composition. Additionally, the defect-free calculation revealed that the polarization of Sc_0.06Al_0.94N (spontaneous and piezoelectric) was 1.38 × 10⁻⁵ C m⁻². Reducing this polarization to 0.40 × 10⁻⁵ C m⁻² in the simulation resulted in a 2DEG density matching the experimental value of 6.3 × 10¹² cm⁻². This reduction in polarization remained consistent for Sc_0.09Al_0.81N and Sc_0.15Al_0.85N, indicating that the amount of polarization reduction does not vary with Sc composition. Consequently, the corresponding decrease in 2DEG density also remains constant. These findings suggest the presence of defects in the ScAlN layer or at the regrown interfaces, with a consistent defect density regardless of the Sc composition. Although the exact defect density has not been directly measured, future investigations using secondary ion mass spectrometry and STEM are planned to characterize these defects and confirm their independence from Sc composition. These studies will provide further insights into the mechanisms responsible for the observed trends in 2DEG density.

Additionally, the μ decreased across all samples after the deposition of ScAlN. Notably, the electrical properties of ScAlN/GaN heterostructures grown by MBE and MOCVD are superior to those of the samples in this study. For instance, ScAlN (Sc composition of 8.6%)/GaN grown by MOCVD achieved n_s of 2.5 × 10¹³ cm⁻² and μ of 914 cm⁻² V⁻¹ s⁻¹.²⁶ Similarly, a previous study on ScAlN/AlGaN/GaN structures grown by MBE reported n_s of approximately 8 × 10¹² cm⁻² and μ of 1025 cm⁻² V⁻¹ s⁻¹ for an Sc composition of 30%.¹⁴ Although the dielectric films deposited atop the ScAlN layer varied between these samples, their electrical properties appear to differ significantly from those of ScAlN grown by sputter epitaxy. Despite the high structural quality of the sputtered samples, specific defects are likely present, degrading their electrical properties. These defects, which may not be significant in films grown by MBE and MOCVD, are likely introduced by the Sc source material used in the sputtering process. The observed reduction in n_s and μ with increasing Sc composition suggests that these defects play a key role in the degradation of electrical performance. Importantly, the defect density appears to remain consistent regardless of the Sc composition, as indicated by the 1D Poisson simulations. These findings suggest that polarization-induced charges in the ScAlN layer are screened by structural or chemical imperfections, such as impurities or grain boundaries.

Here, the n_s and μ of ScAlN/(AlGaN)/GaN structures grown by different methods are compared (see Fig. S2 in the supplementary material). The data indicate that for ScAlN/GaN heterostructures grown by MBE and MOCVD, both μ and n_s decrease with increasing Sc composition, consistent with the results obtained for sputtered ScAlN in this study. Notably, the MBE data include ScAlN/(AlGaN)/GaN heterostructures with an interlayer at the AlGaN/GaN interface, which exhibit significantly higher mobility compared to the general trend. While this study also includes samples with an interlayer, their mobility remains lower than that of the MBE-grown samples. This discrepancy suggests the presence of impurities or other structural imperfections in the sputtered samples that are absent in the MBE-grown structures. This issue may not be intrinsic to sputtering itself but could instead stem from limitations of the current setup. Improvements such as chemical cleaning or in situ cleaning may help reduce impurities and imperfections, offering a pathway to enhance sample quality.

In this study, ScAlN was epitaxially grown on AlGaN/AlN/GaN HEMT structures using a sputtering method, and its structural and electrical properties were evaluated. ScAlN films with Sc compositions ranging from 5% to 20% were coherently grown on the AlGaN/AlN/GaN HEMT structure, with the c-axis lattice constant of ScAlN increasing as the Sc composition increased. The change in the c-axis lattice constant can be explained using a quadratic function model. STEM observations revealed that the wurtzite-structured ScAlN inherited the metal polarity of the underlying HEMT structure and grew epitaxially. For ScAlN grown on AlGaN/AlN/GaN HEMT structures, a slight increase in n_s was observed at a 6% Sc composition, whereas higher Sc compositions led to a decrease in n_s. Reduced μ was observed across all Sc compositions. Further investigation into the factors causing these changes, such as impurities, could help optimize the 2DEG properties of ScAlN/GaN structures. These insights could contribute to broader applications of sputter epitaxy of ScAlN films in various devices, including piezoelectric, ferroelectric, and high-frequency GaN HEMTs.

See the supplementary material for additional data and analysis that support the findings of this study. Figure S1 illustrates the band profile of ideal ScAlN/AlGaN/AlN/GaN heterostructures without defects. Figure S2 compiles data reported by other groups on MBE- and MOCVD-grown samples. The characteristics of the heterostructures fabricated in this study are summarized in Table S1.

This study was partially supported by the JSPS KAKENHI (Grant No. JP23KK0094) and the Izumi Science and Technology Foundation. We extend our gratitude to Professors T. Iida and H. Kunioka for their invaluable assistance with the Hall effect measurements and to Y. Wakamoto for his help with the 1D Poisson calculations.

The authors have no conflicts to disclose.

Tomoya Okuda: Data curation (lead); Formal analysis (equal); Resources (equal); Validation (equal); Visualization (equal); Writing – original draft (lead); Writing – review & editing (equal). Shunsuke Ota: Data curation (supporting); Formal analysis (supporting). Takahiko Kawahara: Conceptualization (supporting); Data curation (supporting); Validation (equal); Visualization (equal); Writing – review & editing (equal). Kozo Makiyama: Conceptualization (supporting); Data curation (supporting); Formal analysis (supporting); Validation (equal); Visualization (equal); Writing – review & editing (equal). Ken Nakata: Conceptualization (supporting); Data curation (supporting); Formal analysis (supporting); Validation (equal); Visualization (equal); Writing – review & editing (equal). Takuya Maeda: Conceptualization (supporting); Validation (equal); Visualization (equal); Writing – review & editing (equal). Atsushi Kobayashi: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Funding acquisition (lead); Methodology (lead); Project administration (lead); Resources (equal); Supervision (lead); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.