Heteroepitaxial growth of Ga₂O₃ on 4H-SiC by liquid-injection MOCVD for improved thermal management of Ga₂O₃ power devices

Ultrawide bandgap (UWBG) semiconductors represent very promising electronic materials for next generation power electronics, greatly extending capabilities of currently used wide bandgap (WBG) semiconductors such as GaN and SiC. Competitive material properties, controlled n-type doping, and available native substrates resulted in recent increased research focus in Ga₂O₃, even surpassing that of other typical UWBG materials, i.e., AlN, AlGaN, or diamond. High expectations of UWBG Ga₂O₃ over WBG GaN and SiC stem from favorable comparison of well-established figures of merit (FOMs). These allow us to benchmark device limits when a particular material is used for device applications. Typical FOMs used in power electronics are, e.g., Johnson’s FOM (JFOM = E_br²⋅v_sat²/4⋅π²), Baliga’s FOM (BFOM = ɛ⋅μ⋅E_br³), and Huang’s material FOM (HMFOM = E_br⋅√μ), where E_br represents the material’s breakdown electric field (Ga₂O₃: >8 MV⋅cm⁻¹, GaN: 3.5 MV⋅cm⁻¹, SiC: 2.0–2.5 MV⋅cm⁻¹),^1–3, v_sat is electron saturation velocity (Ga₂O₃: 2.0 × 10⁷ cm s⁻¹, GaN, SiC: 2.0–2.5 × 10⁷ cm s⁻¹),^1,2,4,5 ɛ is the dielectric constant, and μ is the mobility of charge carriers. JFOM predicts semiconductor radio-frequency capabilities (Ga₂O₃: 2844, GaN: 1089, SiC: 278)^4,6 while BFOM predicts semiconductor power-switching capability (Ga₂O₃: 3214, GaN: 846, SiC: 317).^4,7 HMFOM reflects suitability for high-frequency power switching applications (Ga₂O₃: 12, GaN: 10, SiC = 7).^8,9 Even though inferior in the case of expected achievable electron mobility¹⁰ and similar saturation velocity, Ga₂O₃ offers almost 2.5–3× higher breakdown field compared to GaN, and SiC secures its potentially strong position in the field of power electronics.

Apart from high breakdown electric field and high carrier saturation velocity at high voltages, high power densities, or high switching frequencies, the thermal conductivity of the base device material plays an important role in keeping the on-state device temperature within the reasonable limits for reliable long-term operation.¹¹ Unfortunately, low and anisotropic lattice thermal conductivity represents one of the key issues in Ga₂O₃ device technology. Compared to WBG GaN (170 W/m K)¹ and SiC (∼300–400 W/m K)² or UWBG AlN (∼260–370 W/m K)^12,13 and diamond (∼2000 W/m K),^14–16 bulk thermal conductivity of Ga₂O₃ is substantially lower;¹⁷ e.g., α-Ga₂O₃: ∼10 W/m K, β-Ga₂O₃: from ∼9 W/m K in [100] direction to ∼27 W/m K in [010] direction, κ-Ga₂O₃: ∼11 W/m K.

Six Ga₂O₃ crystal phases with slight differences in their material properties were identified: α (rhombohedral), β (monoclinic), γ (cubic—defective spinel), δ (cubic—bixbyite), ɛ (pseudo-hexagonal), and κ (orthorhombic).^3,18,19 Monoclinic β-Ga₂O₃ is the only thermodynamically stable phase; others are metastable and can typically convert to β-Ga₂O₃ at higher temperatures.³ As a result, bulk growth of single crystal Ga₂O₃ was only demonstrated for monoclinic β phase; typical range of used methods include Czochralski, edge-defined film-fed growth, floating zone, or vertical Bridgman.^10,20,21 Epitaxial growth of thin films of all Ga₂O₃ phases is typically done using standard techniques,²² i.e., halide vapor-phase epitaxy,^18,23–25 molecular-beam epitaxy,^26–30 atomic layer deposition,^31–33 pulsed-laser deposition,^34,35 and by variety of chemical vapor deposition (CVD) methods—metalorganic chemical vapor deposition (MOCVD),^32,36,37 mist-CVD,^38–43 or liquid-injection MOCVD (LI-MOCVD).^44,45

The LI-MOCVD growth technique used in this study represents a low-pressure variation of MOCVD where the precursor is delivered in liquid form; powdered precursors are typically dissolved in a suitable organic solvent. The growth method is similar to low-pressure chemical vapor deposition (LP-CVD) previously studied by other groups.^46–50 The main difference is that LP-CVD utilizes the solid gallium precursor, typically in a form of high purity Ga pellets⁴⁸ instead of Ga-based metalorganic. In the case of LI-MOCVD, the liquid nature of precursor solution offers great versatility in used chemical sources compared to other CVD techniques. Another key difference is the temperature range used for growth of Ga₂O₃. LP-CVD growth of Ga₂O₃ is typically achieved at temperatures of >800 °C^46,47,51 and shows high growth rates of ∼1–10 μm/h.⁴⁶ The LI-MOCVD technique uses lower growth temperature in the range of ∼550–700 °C, which allows the successful synthesis of metastable α- and κ-Ga₂O₃ polymorphs besides the thermodynamically stable monoclinic β-Ga₂O₃ produced by LP-CVD, however at much lower growth rates.^46,48

Homoepitaxial growth on native β-Ga₂O₃ substrates was shown on various substrate orientations, e.g., (010), (001), (100), and $( 2 ¯ 01)$⁠.^52,53 Heteroepitaxy of Ga₂O₃ targeted MgO^54,55 as a foreign substrate material in several studies; however, an intense research effort was devoted predominantly to the use of affordable sapphire of various orientations to achieve high quality films.^43,56,57 The use of silicon substrates usually led to amorphous or polycrystalline Ga₂O₃ layers.^58–60 Single crystalline β-Ga₂O₃ on Si was demonstrated using the catalyst-modified vapor-liquid-solid method; however, the thickness of the resulting layers was only ∼15 nm.⁶¹ 

While a clear advancement in the field of Ga₂O₃ heteroepitaxy was achieved, it is expected that heterointegration of Ga₂O₃ with high thermal conductivity materials such as SiC, AlN, or diamond will be necessary for achieving high power outputs of Ga₂O₃ devices, as maintaining their on-state device temperature at acceptable levels results in improved device reliability and lifetime.^62,63 In fact, there was an increased recent effort in heterointegration of Ga₂O₃ and SiC, AlN, and diamond using either wafer bonding^64–66 or thin film growth techniques.^{29,42,67–71} In this work, we focus on the heteroepitaxial growth of β- and κ-Ga₂O₃ thin films on 4H-SiC substrates for improved thermal management of Ga₂O₃ power devices using a low-pressure liquid-injection MOCVD method. The detailed analysis of structural and thermal properties of prepared Ga₂O₃ layers is presented and discussed.

Monoclinic β- and orthorhombic κ-Ga₂O₃ thin films were grown on Si-terminated highly resistive 4H-SiC substrates using custom-built low-pressure liquid-injection metalorganic chemical vapor deposition (LI-MOCVD) system with a horizontal hot-wall quartz reactor. Schematics of the used deposition system and prepared samples are shown in Fig. 1. More information on this method can be found elsewhere.⁴⁴ Gallium (III) acetylacetonate [Ga(acac)₃] or tris(2,2,6,6-tetramethyl-3,5-heptanedionato) gallium (III) [Ga(thd)₃] dissolved in toluene (0.02 mol/l) were used as Ga precursors and as a liquid solution injected into the evaporation part of the LI-MOCVD unit kept at 180 °C using electronically controlled electromagnetic injection microvalve. O₂ and Ar were used as reaction and carrier gases, respectively. Silicon tetraethyl orthosilicate (TEOS) was used to achieve n-type doping in all prepared samples; 0.1 mol. % TEOS concentration in the precursor solution was used. Growth parameters are summarized in Table I. The number of injections was 5000, and the thickness of grown layers was evaluated by ellipsometry or x-ray reflectivity.

Room temperature (RT) van der Pauw measurement was used to evaluate the resistivity (ρ) of prepared Ga₂O₃ layers. Ohmic contacts for this measurement were manufactured using 60 nm Ti/100 nm Au e-beam-evaporated metallic stack followed by alloying using rapid thermal annealing in forming gas (FGA, 10% H₂, 90% N₂) atmosphere at 550 °C for 30 min. When no doping was used, all Ga₂O₃ layers regardless of the grown phase were highly resistive (ρ > 10⁵ Ω cm). Si-doped β-Ga₂O₃ layers showed resistivity of ∼1.6 Ω cm, while Si-doped κ-Ga₂O₃ remained highly resistive.

X-ray photoelectron spectroscopy (XPS) was employed to provide depth elemental composition profiling. XPS signals were recorded using a Thermo Scientific K-Alpha XPS system equipped with a microfocused, monochromatic Al Ka x-ray source (1486.68 eV). An x-ray beam (6 mA × 12 kV) of 400 μm size was used. The spectra were acquired in the constant analyzer energy mode with a pass energy of 200 eV for the survey. Narrow regions were collected using the pass energy of 50 eV. Charge compensation was achieved with the system flood gun. The depth profile analysis was done using ion gun (1.4 μA of 3 keV Ar⁺ ions over 14 mm²) sputtering. The Thermo Scientific Avantage software 5.9931 was used for digital data acquisition and processing. Spectral calibration was determined by automated calibration routine and internal Au, Ag, and Cu standards. The surface composition in atomic % was determined from the integrated peak areas of the detected atoms (I_A) and the respective sensitivity factors (s_A).

X-ray diffraction (XRD) was used to evaluate the prepared Ga₂O₃ layers. Bruker D8 DISCOVER diffractometer equipped with the x-ray source with a rotating Cu anode operating at 12 kW was used. XRD measurements were performed in parallel beam geometry with a parabolic Goebel mirror in the primary beam; resulting beam divergence was ∼0.03°. Beam size of 1 × 6 mm² was used for acquisition of symmetrical ω/2θ scans. During the measurements, the samples were tilted by an angle of 0.5° away from the precise SiC 0004 diffraction position to suppress the strong diffracted intensity from the substrate. φ scans of the selected Ga₂O₃ and SiC diffractions were acquired to evaluate the azimuthal ordering of the layer structure. To decrease the effect of defocusing, the beam size for these measurements was reduced to 1 × 2 mm² and a parallel-plate collimator with the angular acceptance of 0.35° was inserted into the diffracted beam.

To analyze the surface and the microstructure of Ga₂O₃ layers, a scanning electron microscope (SEM) FEI Quanta in secondary electrons detection mode and a transmission electron microscope (TEM) JEOL JEM-1200EX were used, respectively. Thin plan-view specimens were prepared by a combination of conventional mechanical lapping an Ar⁺ ion-beam thinning from the substrate side of selected samples using a liquid nitrogen-cooled sample holder; the cross-sectional TEM specimens were prepared using focused ion beam technique.

Thermal characterization was performed using pump-probe-based transient thermoreflectance (TTR) on β- and κ-Ga₂O₃ layers; full details of the technique can be found elsewhere.^72,73 To achieve sufficient sensitivity to Ga₂O₃ thermal properties, specific samples targeting Ga₂O₃ thicknesses close to 50 nm were prepared. In the case of β-Ga₂O₃ (thickness of ∼53 nm), only the sample grown using the Ga(acac)₃ precursor was evaluated. The κ-Ga₂O₃ sample (thickness of ∼45 nm) was prepared using the Ga(thd)₃ precursor. These Ga₂O₃ samples were coated with 100 nm Au on a 10-nm thick Ti adhesion layer as a heat transducer. A 532 nm continuous wave laser was used to monitor the surface reflectivity while a 355 nm pulsed laser periodically heated the surface (Au transducer). The monitored change in the reflectivity is linearly proportional to the temperature on the surface. The thermal properties of different layers of the sample were then extracted from the TTR signal. A three layer 2D heat diffusion model of β-Ga₂O₃ or κ-Ga₂O₃, i.e., transducer/Ga₂O₃/SiC, was used to fit the measured TTR traces.⁷² Three different locations were measured on each sample to confirm that they produce similar TTR traces. At each measured location, measurements were repeated three times and then averaged to increase the signal to noise ratio. The Monte Carlo analysis was performed (500 times) to provide a distribution of target values to fitted parameters.

Surface morphology and root-mean-square (RMS) surface roughness were evaluated by atomic force microscopy (AFM) using NT-MDT NTEGRA Prima in tapping mode. Optical bandgap determination of prepared Ga₂O₃ layers using the optical transmittance/absorption method was attempted; however, it could not be resolved due to strong absorption in the SiC substrate.

Ga₂O₃ stoichiometry and growth-related carbon content were investigated using XPS on sample A (β phase). The XPS survey spectrum was recorded from the surface of sample A. Only signals related to C, O, and Ga were observed [Fig. 2(a)]. Detailed scans including C, O, and Ga signals (Fig. S1 in the supplementary material)⁸⁸ were recorded from the surface of sample A and from the volume of grown Ga₂O₃ films after 12 s of etching by in situ Ar⁺ ion sputtering to remove surface contamination. Ga and O signals were attributed to Ga₂O₃. Assuming most intense signals—Ga 3d at ∼21 eV [Fig. S1(a) in the supplementary material]⁸⁸ and O 1s at ∼531 eV [Fig. S1(b) in the supplementary material]⁸⁸, predominantly stoichiometric Ga₂O₃ was concluded (Table S1 in the supplementary material)⁸⁸. Weak Ga 3d signal at ∼19 eV [Fig. S1(a) in the supplementary material]⁸⁸ and weak O 1s signal at ∼533 eV [Fig. S1(b) in the supplementary material]⁸⁸ correspond to a negligible Ga-suboxide content that may be related to disordered Ga₂O₃ grain boundaries.

Surface C-related signal likely corresponded to adventitious carbon contamination and decreased below the instrument detection limit (<0.1 at. %) after surface etching [Fig. 2(b)]. Low C-related signal suggest that C content in Ga₂O₃ was within reasonable limits and did not cause detrimental Ga₂O₃ lattice alteration (see XRD results in the following section). Since the used LI-MOCVD growth method uses relatively low pressure (∼0.1–1 Torr), carbon incorporation in Ga₂O₃ layers may be significant as was observed in, e.g., MOCVD-grown GaN.⁷⁴ However, this seems unlikely in our case; we expect the majority of carbon produced from used solvents and organometallics after their thermal decomposition did not incorporate into the Ga₂O₃ layer and left the LI-MOCVD unit, possibly reacting with the gaseous oxygen. We note that while not resolved by XPS, the amount of carbon in grown Ga₂O₃ may be still sufficient to significantly contribute to various charge trapping phenomena and needs to be further evaluated.

XRD confirmed single-phase $( 2 ¯ 01)$ β-Ga₂O₃ in samples A and B grown using Ga(acac)₃ and Ga(thd)₃, respectively. In case of samples A and B, only reflections of $( 2 ¯ 01)$ Ga₂O₃ lattice planes, i.e., $2 ¯01$⁠, $4 ¯02$⁠, and $6 ¯03$ diffractions were observed. As reported elsewhere, phase identification of orthorhombic κ-Ga₂O₃ using only XRD can provide inconclusive results.⁴⁵ However, as discussed later, using a combination of XRD and TEM, we conclude single-phase κ-Ga₂O₃ with a minor inclusion of β-Ga₂O₃ in sample C grown using Ga(thd)₃. Corresponding wide symmetric ω/2θ XRD scans are shown in Fig. 3; rocking curves (ω scans) are shown in Fig. S2 in the supplementary material.⁸⁸ Approximately twofold larger full width at half maximum (FWHM) of corresponding rocking curves for sample grown using Ga(acac)₃ than those grown using Ga(thd)₃ precursors was observed, indicating improved crystal quality of the latter, likely related to lower observed growth rates when Ga(thd)₃ was used (c.f. Table I). This behavior is consistent with our previous findings for β-Ga₂O₃ grown on sapphire substrates.⁴⁴ Similar to previous reports on sapphire substrates,^44,57,75,76 six in-plane domain orientations were observed in our samples as confirmed by XRD φ scans shown in Figs. 4(a) and 4(b). Due to the negligible difference between samples A and B, only φ scans of the former are shown in Fig. 4(a). We note that in sample C, both major orthorhombic κ phase and minor parasitic β phase showed this behavior.

Based on our XRD results, LI-MOCVD growth of monoclinic β-Ga₂O₃ on 4H-SiC at 700 °C using Ga(acac)₃ precursor was in general less sensitive to O₂ flows used. O₂ flows of >600 SCCM led to a repeatable phase-pure synthesis. However, when Ga(thd)₃ was used, a mixture of β and κ phases was often observed. Best results in repeatable growth of phase-pure β-Ga₂O₃ was achieved at similar conditions to those observed when Ga(acac)₃ precursor was used, i.e., 700 °C growth temperature and O₂ flow >600 SCCM.

Growth of orthorhombic κ-Ga₂O₃ was observed only when a Ga(thd)₃ precursor was used. For all investigated κ-Ga₂O₃ growth conditions, some degree of parasitic β phase inclusion was observed. Figure 5 illustrates the contribution of parasitic β-Ga₂O₃ phase in LI-MOCVD-grown κ-Ga₂O₃ films as a function of growth temperature (560–700 °C) and O₂ flow (200–800 SCCM). Using measured ω/2θ XRD scans, diffractions at 2θ angles close to 60°, i.e., at the locations corresponding to respective $6 ¯03$ and 006 diffractions of β- and κ-Ga₂O₃, were approximated by Gaussian peak fitting, and the κ/β-Ga₂O₃ peak height ratio was plotted. The most significant suppression of the parasitic β-Ga₂O₃ phase was observed at a growth temperature of 600 °C and an O₂ flow of 800 SCCM. In this regard, varying the growth temperature and O₂ flow showed a relatively narrow κ-Ga₂O₃ growth window, i.e., growth temperature of ∼600–630 °C and O₂ flow of ∼700–800 SCCM. We expect that the parasitic β-Ga₂O₃ may be further suppressed when higher O₂ flows are used; however, 800 SCCM represented the achievable limit for the used LI-MOCVD setup. A combination of lowest used growth temperature and O₂ flow resulted in unsuccessful Ga₂O₃ growth.

A growth of relatively large faceted grains is characteristic for β-Ga₂O₃ layers [Fig. 6(a)]; the faceting is slightly more pronounced for the thicker sample A grown using Ga(acac)₃ than for the sample B grown using Ga(thd)₃. Electron diffraction from a plane-view TEM specimen of sample A (Fig. 6) confirmed the epitaxial relation observed using XRD, i.e., $( 2 ¯ 01)[102]β - G a 2 O 3||(0001)[ 01 1 ¯ 0]$ 4H-SiC as well as the presence of six in-plane domain orientations. Assuming this epitaxial relationship, Fig. 6(b) schematically depicts the possible orientations of β-Ga₂O₃ (red, yellow, and green rectangles) on the SiC substrate (black triangles). Only three β-Ga₂O₃ orientations (rectangles) are depicted in the growth schematic. Due to the low symmetry of the monoclinic β-phase, shown rectangles do not represent identical orientations after twofold rotation, i.e., each rectangle in fact represents two different domain orientations.

The long diffraction arcs observed in Fig. 6(d) reflect misorientations of diffracting planes in Ga₂O₃ domains. These, however, do not directly express the misorientation of domain lattices because the observed electron diffraction pattern with seemingly sixfold rotation symmetry is in fact composed of many electron diffraction patterns originated in individual domains, which do not have the sixfold symmetry. The ideal angle between $( 2 ¯ 01)$ and (512) planes is 91.66° and between (512) and (020) planes is 61.72°. Thus, the origin of the widening of diffraction arcs observed in SAED and also in the XRD β-phase φ-scan maxima [Fig. 4(a)] can be attributed to a combination of real domain misorientation and low symmetry of the monoclinic phase that consequently results in nonsymmetric diffraction patterns.

The sample C (κ-Ga₂O₃) showed much smoother surface and the Ga₂O₃ layer consisted of considerably smaller domains [Fig. 7(a)] compared to sample A. All TEM observations confirmed the epitaxial relation $(001)[010]$ κ-Ga₂O₃// $(0001)[ 01 1 ¯ 0]$ 4H-SiC. The bright field TEM image [Fig. 7(a)] was taken with the specimen oriented precisely with the SiC [0001] zone axis parallel to the electron beam; thus, the observed Moiré patterns created by double diffraction reveal mutual in-plane and out-of plane misorientations of individual domains. In the SAED pattern shown in Fig. 7(b), the strongest principal diffractions are indexed preferentially, while the indexing of weak intensity diffractions is illustrated by the green indices belonging to only one domain orientation. All weak diffractions corresponding to every domain orientation together with additional spots related to double diffractions create a rather complex diffraction pattern [Fig. 7(b)]. These weak diffraction spots in all orientation domains confirm the presence of κ-phase in our layers as they cannot be attributed to the hexagonal ɛ-Ga₂O₃ phase.⁴⁵ The schematic in Fig. 7(c) illustrates the mutual orientation of six different domain orientations and their relationship to the 4H-SiC substrate. Only three rectangles are shown in the schematic as the κ-phase symmetry group Pna2₁ does not contain twofold rotation axis parallel to the [001] direction. The XRD analysis revealed the presence of a minor epitaxially grown β-phase. Its occurrence is also visible in the plan-view SAED pattern and illustrated in a magnified detail shown in Fig. 7(d) as arc-shaped weak diffraction spots located between 1210 SiC and 330 and 060 family of κ-phase diffractions [labeled by green arrow in Fig. 7(d)]. Due to clarity, the β-phase diffraction spots are not indexed in Figs. 7(b) and 7(d); however, they are the same as those in Fig. 6(d), which can be used as a reference. Green arrow in Fig. 7(d) denotes the 020 family diffraction spots; 512 β-phase diffraction spots are superposed to 060 and 330 family spots belonging to the κ-phase. Because of their location, it is not possible to distinguish the β-Ga₂O₃ domain distribution in the layer using conventional dark field (DF) image technique. The weak arc spots labeled by the red arrows in Fig. 7(d) were created by double diffraction.

Similar to the plan-view TEM analysis, sample C cross-sectional electron diffraction patterns can be indexed according to the orthorhombic κ-phase Ga₂O₃ [Fig. 8(c)].

As can be seen in the cross-sectional TEM images in Figs. 8(a) and 8(b), the κ-phase Ga₂O₃ in the sample C layer exhibited columnar growth of thin domains, following the initial growth of an intermediate layer of variable thicknesses. A detailed study on the structure of this intermediate layer as well as its role in the growth of κ-phase Ga₂O₃ will be published elsewhere. The thin κ-phase Ga₂O₃ domains were visualized by dark field technique using 013 diffraction [Fig. 8(b)].

Observed domain structure hindering the crystal quality of the κ-Ga₂O₃ layer in sample C is likely caused by the large lattice mismatch between Ga₂O₃ and 4H-SiC. For example, the mismatch along [100] κ-Ga₂O₃ and $[ 10 1 ¯ 0]$ 4H-SiC is approximately 5.4%, leading to a domain structure resulting from strain relaxation.⁷⁷ The single-domain κ-Ga₂O₃ growth was only achieved on an exotic ɛ-GaFeO₃ substrate,⁷⁸ suggesting that a suitable buffer layer may improve crystal properties of Ga₂O₃ on SiC. However, the domain structure similar to our case was observed when the AlN buffer was used for κ-Ga₂O₃ layers.⁷⁸ 

AFM confirmed the granular surface in all three studied samples [Figs. 9(a)–9(c)]. β-Ga₂O₃ (samples A and B) showed similar respective RMS surface roughness of 6.5 and 6.8 nm. Sample A, grown using Ga(acac)₃, showed more uniform surface coverage than sample B grown using Ga(thd)₃. Despite of the same number of injections during the LI-MOCVD growth, the thickness of the sample B was ∼60% smaller compared to sample A, i.e., a slower growth and/or longer nucleation phase could have resulted in the observed less uniform surface coverage and can possibly be mitigated by prolonged growth. Much lower surface roughness was observed for κ-Ga₂O₃, i.e., ∼0.9 nm. Surface line profiles of samples A, B, and C positioned along the dashed lines displayed in corresponding AFM scans are also shown in Fig. 9.

Low surface roughness of κ-Ga₂O₃ (∼0.9 nm) is a promising result that can enable, e.g., κ-Ga₂O₃/III-N heterostructure growth, where low interfacial roughness is desirable. κ-Ga₂O₃ is expected to provide strong polarization charge⁷⁹ and forming heterostructures with III-N materials could enable Ga₂O₃ devices with a 2D electron gas (2DEG) channel.^80,81 Thermal stability of this metastable polymorph will, however, play a crucial role in any kind of heterointegration attempts, limiting the growth of available barrier materials atop the κ-Ga₂O₃.

Figure 10 shows the measured TTR signals and best fitting results for the three-layers model; thermal properties of the samples included Au, Ga₂O₃, and the 4H-SiC substrate. Literature values were used for density and heat capacity of all layers. Thermal conductivities of Ga₂O₃ and SiC were the fitting parameters. Thermal boundary resistance (TBR) of Au(Ti)/Ga₂O₃ and Ga₂O₃/SiC interfaces were lumped into the Ga₂O₃ layer. This is because the thermal conductivity of Ga₂O₃ is comparable to that of the interlayer between Ga₂O₃ and SiC when the thickness of the Ga₂O₃ is only at the tens of nanometer level;⁸² this makes the TBR value challenging to distinguish. Nevertheless, lumping the TBR with the thermal conductivity of the Ga₂O₃ layer has only limited effect on the fitted resulting value of the Ga₂O₃ thermal conductivity as its thickness is much greater (5–6×) than that of the interface layer. Estimated transducer/Ga₂O₃ TBR was close to, e.g., Au/Si, indicating a negligible error on the fitting and determined resulting Ga₂O₃ thermal conductivity.

Figure 11 shows a sensitivity analysis of the TTR signal to a 10% change in the thermal conductivity of Ga₂O₃ and TBR of Ti/Ga₂O₃ and Ga₂O₃/SiC interfaces. The sensitivity analysis demonstrates the effect of each parameter on the TTR signal. The sensitivity analysis shows distinct sensitivity time scales between the thermal conductivity of Ga₂O₃ and other layers, and other parameters including the heat capacity, allowing it to be accurately fitted. It is important to note that the out-of-plane thermal conductivity of β-Ga₂O₃ and κ-Ga₂O₃ is assumed dominant; the in-plane thermal conductivity is of low sensitivity in the model for such a thin Ga₂O₃ film. In case the TBR values (Au/Ga₂O₃ and/or Ga₂O₃/SiC) were assumed in the model, there would be large related uncertainty because their sensitivities in the sensitivity analysis would be much smaller than those of the Ga₂O₃. We note that the surface roughness can affect the reflected TTR signal, which could represent a concern, e.g., for the β-Ga₂O₃ sample (RMS surface roughness of ∼7 nm). However, owing to the high sensitivity to the Ga₂O₃ layer as shown in the sensitivity analysis (Fig. 11), sufficient TTR signal-to-noise ratio was obtained to achieve good fit to the experimental data (Fig. 10).

The Monte Carlo error analysis was employed to estimate the error bar in the thermal conductivity of Ga₂O₃ by repeating the model fitting 500 times (Fig. S3 in the supplementary material).⁸⁸ The initial values of fitting parameters of the sample and the laser were randomly varied within 5% standard deviation for each time. Thermal conductivities of the 53 nm-thick β-Ga₂O₃ and 45 nm-thick κ-Ga₂O₃, i.e., samples A and C, were 2.13 + 0.29/−0.51 and 1.23 + 0.22/−0.26 W/m K, respectively, consistent with the reported predictions for various Ga₂O₃ thicknesses.⁸³ 

To assess the influence of heterointegration of LI-MOCVD-grown Ga₂O₃ and highly thermally conductive SiC substrate on the channel temperature of Ga₂O₃-based transistors, thermal simulation of an example of metal–oxide–semiconductor field-effect transistor (MOSFET) was made. An Ansys finite element thermal simulation was used to predict the temperature distribution in a β-Ga₂O₃/SiC MOSFET with various thicknesses of the Ga₂O₃ layer. The peak channel temperatures were compared between devices on SiC and on Ga₂O₃ substrates. Thickness-dependent thin-film Ga₂O₃ thermal conductivity of 5, 12.9, 13.2, and 14 W/m K at 300 K for 500 nm, 2 μm, and 6.5 μm Ga₂O₃ and Ga₂O₃ substrate, respectively, based on our measurements and extrapolations,⁸³ was used in the simulation. Temperature-dependent thermal conductivities of the Ga₂O₃ layer and 4H-SiC substrates were adopted from previous studies.^84,85 The thickness of the gate metal and the Ohmic contacts were taken as 150 and 160 nm, respectively; the thermal conductivity of gold was taken as 315 W/m K. A 20 nm-thick Al₂O₃ layer was considered as the gate insulator. Gate-drain, gate-source, and gate lengths were taken as 6, 6, and 2 μm, respectively, same as previously reported in actual devices.⁸⁶ Heat generation in the transistor channel (length of 1.5 μm) was defined at the drain edge of the gate contact to a depth of 50 nm. Ambient temperature was 25 °C.

Figure S4 in the supplementary material⁸⁸ shows a comparison for a homoepitaxial Ga₂O₃ device and the device integrated with SiC for various thicknesses of the Ga₂O₃ device layer. The peak temperature was simulated for power densities ranging from 1 to 5 W/mm. The Ga₂O₃/SiC device thermal resistance of ∼9.7 mm-K/W is much lower compared to ∼190 mm K/W for Ga₂O₃ homoepitaxy when a thin Ga₂O₃ layer (<2 μm) is used, because the heat source is very close to SiC, resulting in efficient heat extraction. This means the thinner Ga₂O₃ layer can increase the ability of heat removal of the device even though the thinner Ga₂O₃ layer results in a drop in its thermal conductivity. Device thermal resistance of ∼9.7 mm K/W for simulated Ga₂O₃/SiC MOSFET with a thin Ga₂O₃ layer (<2 μm) is comparable to the GaN/Si high electron mobility transistor (HEMT, ∼10 mm K/W) and ∼2× higher compared to GaN/SiC HEMT (∼5 mm K/W) when similar device layouts are considered.^63,87

Figure 12 shows a comparison of device depth temperature profiles at the device center assuming various thicknesses of the Ga₂O₃ layer. Dissipated power density of 3 W/mm was assumed. The 500 nm and 2 μm Ga₂O₃ layer has negligible effect on the total device thermal resistance, while the 6.5 μm Ga₂O₃ accounts for almost four times higher device thermal resistance. The homoepitaxial Ga₂O₃ device showed ∼18 times higher total thermal resistance at 3 W/mm compared to the structure with a thin (<2 μm) Ga₂O₃ layer on the SiC substrate.

The interlayer between the Ga₂O₃ and SiC substrate in the simulation was approximated by a layer with the thickness of 10 nm and a thermal conductivity of 2.1 W/m K, the same value measured by the TTR for the 50 nm-thick β-Ga₂O₃ layer. The thermal conductivity of the interlayer was fixed for different Ga₂O₃ layer thicknesses used in the simulation. Thermal simulation assuming 5× lower and higher thermal conductivity of this interlayer was also performed; however, only <1% temperature difference was observed. This indicates that for the chosen device design and dissipated power density range, the thermal boundary resistance between the LI-MOCVD-grown Ga₂O₃ and 4H-SiC does not significantly contribute to the thermal performance of a real Ga₂O₃ device, regardless of the Ga₂O₃ thickness.

A successful epitaxial growth of thin monoclinic β- and orthorhombic κ-Ga₂O₃ films on highly thermally conductive 4H-SiC substrates using LI-MOCVD was demonstrated. Ga(acac)₃ and Ga(thd)₃ were used as gallium precursors; β-Ga₂O₃ was prepared using both precursors, while only Ga(thd)₃ enabled the growth of κ-Ga₂O₃. Regardless of the used precursor, best results for β-Ga₂O₃ were achieved at a growth temperature of 700 °C and O₂ flows in the range of 600–800 SCCM. In contrast, based on the used growth conditions, κ-Ga₂O₃ showed various degrees of parasitic β phase inclusion. Relatively narrow growth window for κ-Ga₂O₃ was identified, and best results were achieved for a growth temperature of 600 °C and an O₂ flow of 800 SCCM. Complete suppression of the parasitic β phase may be achieved for O₂ flows of >800 SCCM. Undoped Ga₂O₃ layers were highly resistive regardless of their phase (ρ>10⁵ Ω cm); Si-doped β-Ga₂O₃ layers showed a low resistivity of ∼1.7 Ω cm, while Si-doped κ-Ga₂O₃ layers remained highly resistive. Complex domain orientation structures of grown β- and κ-Ga₂O₃ films were revealed by XRD and TEM analyses. Stoichiometric Ga₂O₃ films were confirmed by XPS; minor Ga-suboxide content was attributed to disordered grain boundaries. Despite relatively low growth pressure, negligible growth-related C contamination (<0.1 at. %) was observed.

Transient thermoreflectance was used to determine the thermal conductivities of 53 nm-thick β-Ga₂O₃ and 45 nm-thick κ- Ga₂O₃ layers, i.e., 2.13 + 0.29/−0.51 and 1.23 + 0.22/−0.26 W/m K, respectively, in good agreement with previous reports. Computer thermal simulations of an example β-Ga₂O₃/SiC MOSFET device with various Ga₂O₃ layer thicknesses (0.5–6.5 μm) were prepared; results for dissipated power density in the range of 1–5 W/mm were compared to the homoepitaxial device. Thin (<2 μm) Ga₂O₃ layers have negligible contribution to the total device thermal resistance, which was found to be ∼9.7 mm K/W, comparable to GaN on Si technology and about twice as much as typical values for GaN/SiC HEMTs. Devices with 6.5 μm Ga₂O₃ showed almost four times higher thermal resistance, and the homoepitaxial device showed approximately 18 times higher thermal resistance at 3 W/mm compared to devices with 2 μm or less thick Ga₂O₃ layer on the 4H-SiC substrate. This suggests that using a thinner Ga₂O₃ layer is more important for reducing the device thermal resistance than its thermal conductivity, even though it is lower for thinner Ga₂O₃ layers. Thermal boundary resistance between Ga₂O₃ and SiC was found to have negligible impact on the simulated device temperature. Approximating the Ga₂O₃/SiC TBR with the TTR-determined thermal conductivity of ∼50 nm-thick β-Ga₂O₃ layer (2.1 W/m-K) and varying this value in the range from five times lower (0.42 W/m K) to five times higher (10.5 W/m K) produced <1% difference in simulated device temperatures. Therefore, we can conclude that for the investigated conditions, TBR between the LI-MOCVD-grown Ga₂O₃ and 4H-SiC does not significantly contribute to the thermal performance of Ga₂O₃ MOSFET, regardless of the Ga₂O₃ thickness.

Our results show great potential of heterointegration of Ga₂O₃ and SiC for improved thermal management and reliability of future Ga₂O₃-based high power devices. While other substrate materials may offer other advantageous properties such as reduced cost (Si) or even higher thermal conductivity (diamond), relatively low lattice mismatch between β-Ga₂O₃ and 4H-SiC [∼1.3% with monoclinic (010) plane] makes SiC preferable to Si (∼20.8%) and diamond (∼14.8%)²⁹ for growth of thicker Ga₂O₃ films. To take full advantage of the outstanding Ga₂O₃ material properties, high crystal quality layers will be needed. As a next step in the Ga₂O₃/SiC epitaxy, we expect vicinal SiC substrates may bring considerable improvement.

We acknowledge the funding from Slovak Research and Development Agency (Grant Nos. APVV-20-0220 and SK-CN-21-0013), from Slovak Grant Agency VEGA (Grant No. 2/0100/21), and from Slovak Academy of Sciences Dokto Grant No. APP0424. This work was performed during the implementation of the project Building-up Centre for advanced materials application of the Slovak Academy of Sciences, ITMS project code 313021T081 supported by Research & Innovation Operational Programme funded by the ERDF. M.K. was supported by the Royal Academy of Engineering through the Chair in Emerging Technologies Scheme. The thermal characterization was supported as part of the ULTRA, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences at the University of Bristol under Award No. DE-SC0021230.

The authors have no conflicts to disclose.

Fedor Hrubišák: Conceptualization (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Kristína Hušeková: Investigation (equal); Methodology (equal); Resources (equal); Writing – review & editing (equal). Xiang Zheng: Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Visualization (equal); Writing – review & editing (equal). Alica Rosová: Formal analysis (equal); Investigation (equal); Methodology (equal); Visualization (equal); Writing – review & editing (equal). Edmund Dobročka: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). Milan Ťapajna: Conceptualization (equal); Funding acquisition (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal). Matej Mičušík: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). Peter Nádaždy: Investigation (equal); Methodology (equal); Writing – review & editing (equal). Fridrich Egyenes: Investigation (equal); Methodology (equal); Writing – review & editing (equal). Javad Keshtkar: Investigation (equal); Methodology (equal); Writing – review & editing (equal). Eva Kováčová: Investigation (supporting). James W. Pomeroy: Methodology (equal); Writing – review & editing (equal). Martino Kuball: Funding acquisition (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). Filip Gucmann: Conceptualization (equal); Funding acquisition (equal); Investigation (equal); Project administration (equal); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are available within the article and its supplementary material.⁸⁸