Epitaxial growth of GaN on β-Ga₂O₃ via RF plasma nitridation

The power electronics industry has a wide and ever-expanding breadth of applications, which continue to demand robust, ultra-wide bandgap (UWBG) materials to facilitate improved operation at higher voltages, temperatures, and with faster switching speeds. Gallium oxide has a high Baliga figure of merit (BFOM) of 3444 (relative to silicon), availability of shallow donors, and relatively low-cost substrates, making it a promising candidate in comparison to other WBG and UWBG semiconductors like GaN, AlN, and diamond.^1–3 Gallium oxide's high BFOM is due, in part, to its high critical electric field (E_c), which prevents breakdown during high- voltage operation, allowing for improved efficiency. Low substrate costs result from compatibility with traditional bulk crystal growth methods which show rapidly advancing quality with commercially available sizes already in the 4 in. range. Therefore, Ga₂O₃ makes for a cost-effective material, which promises greater efficiency than alternatives.^4–8 

There has been a great deal of success in unipolar Ga₂O₃ devices. Schottky barrier diodes (SBDs) have been demonstrated with high current densities and at elevated temperatures, with BFOMs up to 0.6 GW/cm².^9–11 Unfortunately, without p-type Ga₂O₃, conventional power device designs cannot be realized, and Ga₂O₃ will continue to lag behind the GaN state of the art (27 GW/cm²).¹² Acceptor activation energy and hole effective masses are very high, so room temperature dopant activation is very difficult and any carriers that are generated do not contribute to conduction.¹³ While this can be used for edge termination type structures, it does not work for active regions of bipolar devices. Therefore, heterojunctions with p-type materials are necessary.

NiO and Cu₂O have had success in heterostructures with Ga₂O₃, demonstrating high rectification and decent breakdown voltages, though they still fall far short of theoretical limits.^14,15 GaN has garnered interest as an alternative, owing to its low lattice mismatch to Ga₂O₃, developed processing methods, high crystallinity, high E_c, and reliable p-type doping.¹⁶ Some work has already been done in this area, with successful demonstration of ultraviolet photodiodes by depositing Ga₂O₃ on p-type GaN.^17,18 p–n heterojunction diodes have also been fabricated using mechanically exfoliated Ga₂O₃ onto GaN via the Scotch tape method.¹⁹ However, these devices do not take advantage of Ga₂O₃ as a substrate.

Far fewer concrete results have been published for devices with GaN on Ga₂O₃. Research in this area has had some successes with metalorganic chemical vapor deposition (MOCVD)-grown unintentionally doped (UID) GaN having x-ray diffraction rocking curve (XRC) full-width at half-maximum (FWHMs) as low as 330 arc sec.^20,21 Quantum well structures have been demonstrated as well as simple devices; however, p-type GaN on Ga₂O₃ is expected to be difficult via MOCVD due to the necessity of a high-temperature anneal. Epitaxial growth of p-GaN on n-GaN with a β-Ga₂O₃ substrate has been demonstrated for high-performance optical devices, though these devices employed many micrometers of n-GaN for strain relief, removing the p–n junction from the heterojunction.²² As such, these structures are not suitable for power electronics that seek to take advantage of the benefits of β-Ga₂O₃.

While some sort of buffer layer is needed to provide strain relief, a much thinner one must be used for p–n heterojunctions. Buffer layers can also be achieved through the conversion of Ga₂O₃ to GaN via exposure to nitrogen species at elevated temperatures. Initial nitridations were done using NH₃, but this has been shown to result in severe crystalline damage due to reactions between hydrogen and oxygen during the nitridation process.^23,24 Alternatively, others have demonstrated conversion to and subsequent epitaxial growth of GaN via nitrogen plasma in plasma-assisted molecular beam epitaxy (PAMBE), though there was limited investigation of the different nitridation and growth parameters and no electrical characterization of the films.²⁵ 

At present, this plasma nitridation process is the most promising method of growing device-quality epitaxial p-type GaN on Ga₂O₃. In this work, the mechanism and process variables of nitrogen plasma-based nitridation were explored to create a suitable template for high-quality GaN-on-Ga₂O₃ growth. Further, GaN-on-Ga₂O₃ films were grown, and their defect densities were explored as a function of the nitridation layer properties. A discussion is presented outlining the path toward high-performance devices.

(–201) oriented β-Ga₂O₃ substrates were used for low lattice mismatch to wurtzite GaN (h-GaN). Substrates purchased from Novel Crystal Technology, Inc. were grown via edge-defined film-fed growth (EFG) and were Sn-doped to 1.8 × 10¹⁸ cm⁻³. Nitridation and growth were done in a custom III–N MBE system. Substrates were cleaned ex situ via sonication in solvents followed by a 5 min etch in the piranha solution, then loaded into the PAMBE system. Prior to nitridation, substrates were heated to 750 °C to thermally desorb any contaminants. For nitridation, plasma power, nitrogen flow rate, and substrate temperature were varied. Nitridation time was fixed to 3.5 h. UID n-type GaN growths were done in gallium-rich conditions at 740 °C. The GaN growth rate was approximately 250 nm/h. Reflective high-energy electron diffraction (RHEED) patterns were acquired throughout the nitridation and growth processes.

As will be described in greater detail later, since the rate of nitridation is significantly slower $(∼ 100 ×)$ than the nitrogen-limited growth rate of GaN, the surface is saturated by the nitrogen plasma. As a result, the plasma composition plays a greater role than its overall flux. Therefore, plasma power density (PPD) is discussed as a compound variable (defined as plasma power divided by nitrogen flow rate) describing more accurately the chemistry of the process.

X-ray diffraction (XRD) scans, XRCs, reciprocal space maps, and x-ray reflectivity (XRR) measurements were all acquired on a Bruker D8 Advance system. XRCs captured ω-scans of the GaN (0002) peak. Atomic force microscopy (AFM) maps were obtained on an Asylum Research MFP-3D system. Scanning TEM (STEM) images and energy dispersive x-ray spectroscopy (EDX) maps were acquired on an FEI Themis Z electron microscope.

Previous reports of nitridation observed via RHEED an initial surface conversion to cubic GaN with a subsequent transition to (0001) oriented h-GaN over time.²⁵ This is hypothesized to be due to the cubic surface structure of the (100) Ga₂O₃ used, stabilizing the c-GaN phase for very thin layers, which then relaxed to the more stable h-GaN beyond a certain thickness. Figure 1(a) shows the transition as observed on (–201) oriented β-Ga₂O₃. As the surface was exposed to nitrogen plasma, the monoclinic β-Ga₂O₃ surface reconstruction was replaced with that of h-GaN. As indicated, the cubic phase was not observed for this orientation. The lack of c-GaN or off-axis orientations for (–201) Ga₂O₃ was confirmed in XRD where only the substrate peaks and the GaN (0002) peak appeared, as shown in Fig. S1 in the supplementary material.

For times longer than 3 h, roughening of the surface was observed. This is detailed in Fig. 1(b). Prior to nitridation, the chemically cleaned surface had an RMS roughness of 137 pm, which was further improved upon thermal treatment in situ. Afterward, as shown, it had an RMS roughness of 311 pm for a 3.5 h nitridation, showing steps and pits. There are two main effects at work. The first is ion bombardment from the plasma, which is known to contribute to surface roughening.²⁶ While the nitrogen ion content is low and, therefore, the rate of roughening is slow, the effect can become noticeable for long exposure times. The second mechanism of surface roughening is the slow thermal decomposition of newly formed GaN. This results in mobile Ga atoms on the GaN surface, which diffuse until they reach an appropriate site to react with. From this process, steps and pits form on the GaN surface, causing roughness. The thermal decomposition of GaN works in parallel with adsorption, diffusion, and reaction of impinging atomic nitrogen to form the nitridation layer. The process is outlined in Fig. 1(c). From this proposed mechanism, the thickness of the nitridated layer should depend then on the substrate temperature and PPD for a fixed nitridation time.

In Fig. 2(a), nitridation layer thickness and RMS roughness are plotted against inverse temperature for fixed plasma conditions. With decreasing temperature, roughness decreased due to the reduced GaN decomposition rate. Resulting thicknesses first increased with temperature, then gradually decreased again. The inflection point of the curve shows that for these plasma conditions, enhanced decomposition limits thickness for higher temperatures (>700 °C) while reduced diffusivity limits thickness at lower temperatures (<700 °C). This agrees with observations that the GaN decomposition rate in MBE increases dramatically above 700 °C.²⁷ Below this temperature, the change in thickness and roughness trends follows similar behavior as they are both thermally activated processes.

When fully saturated, the surface concentration of nitrogen species depends only on their concentrations in the impinging flux, rather than their total magnitude. Figure 2(b) shows how the thickness of the nitridation layer varied with PPD at different temperatures. Thickness increased with PPD, indicating that the species of nitrogen plasma that is most conducive to nitridation is more prevalent at higher plasma power densities. From characterization of the nitrogen plasma source used here, the atomic nitrogen concentration increased linearly with PPD in the regime of interest, shown in Fig. S2 in the supplemental material. This suggests that atomic nitrogen is the species primarily responsible for the nitridation process. For our system, increasing the PPD beyond 600 W/SCCM had diminishing returns as shown in Fig. S3 of the supplemental material.

The time rate of thickness increase is nonlinear and depends on the diffusion profile. For short nitridations, the error function profile is expected corresponding to the diffusion equation solution for a constant source. For long nitridations, a linear profile is expected as a steady state is approached. Therefore, the rate of thickness increase starts off high and mostly linear, then decreases to zero as the rate at which atomic nitrogen reaches the nitridation front slows to the rate of GaN decomposition, putting a cap on the thickness. This agrees with experimentally observed behavior. The terminal thickness was, therefore, scaled linearly by surface atomic nitrogen concentration as described in Fig. 2(b) and, in a more complicated manner, by temperature as described in Fig. 2(a). Roughness was seen to increase exponentially with time as shown in Fig. S4 of the supplemental material, which gives a range to work in where there is low roughness and fast nitridation.

High-quality vertical devices require a low threading dislocation density, since threading screw dislocations (TSDs) are attributed to be the main source of leakage current in GaN devices.^28–31 The GaN(0002) XRC full width at half-maximum (FWHM) of the grown film, $β$⁠, can provide a relative estimate of the TSD density, $ρ s$⁠.^32,33 Therefore, it is desirable to relate $β$ to the controllable parameters of nitridation layer thickness, nitridation temperature, and film thickness.

Critical thickness in heteroepitaxy is defined as the thickness to which the film can grow before misfit dislocation formation occurs to relieve misfit strain. The number of dislocations per unit length, $κ$⁠, along the strained direction is given as $κ= ε r/ b e$⁠, where $b e$ is the burgers vector and $ε r$ is the residual strain.³⁴ This equation is simplified with the assumption that there is only one highly strained axis and that the misfit dislocations are edge-type. This was shown via TEM observations and is described in more detail later. Additionally, we consider film thicknesses are well beyond the theoretical critical thickness $(∼ 3 nm)$⁠. As such, the misfit dislocation density is no longer changing strongly with thickness.

The total misfit strain is reduced to $ε r$ due to the strain relief contribution of the nitridation layer. In the nitridation layer, lattice strain is proportional to the nitrogen concentration as the film transitions between the two lattice parameters. Both EDX in STEM and the theoretical composition profile from the diffusion equation suggest that the nitrogen concentration approximately linearly decreases from the surface. Analysis presented by Tersoff says that for a compositionally graded junction, dislocations will form to relieve strain in a region of the film and the remainder of the film will be dislocation-free.³⁵ The boundary between the two regions depends on the film strain energy and the dislocation formation energy. For a linearly graded junction, residual strain is proportional to the square root of the rate of change of the strain, or $ε r ∝ ε ′ 1 / 2 = ( ε t / t n ) 1 / 2$⁠, where $ε t$ is the total strain between layers dictated by the lattice constant mismatch and $t n$ is the nitridation layer thickness.³⁵ 

To relate $β$ to $κ$⁠, we first consider the mosaic block model where the film is made up of many sub-grains separated by small-angle grain boundaries, each having a finite size, tilt, and twist. Misfit dislocations then must intersect with the grain boundaries and, therefore, have the potential to thread to the surface without changing the burgers vector, with number $n=κd$⁠, where d is the average grain diameter. Assuming the grains are approximately rectangular, the number of threading edge dislocations per unit area, $ρ e∼κd/ d 2=κ/d$⁠. These dislocations cause mosaic twist, $ω$⁠, with $ω∝ b eκ$⁠, resulting in $ρ e∝ω/ b ed$⁠. This is comparable to the well-known equation for grain-boundary localized dislocation density of $ρ e=ω/2.1 b ed$⁠.^36,37

To validate this, F and $h c$ were extracted for different temperatures via fitting the critical thickness plots [Fig. 3(a)]. For a clearer physical interpretation, the effect of nitridation layer thickness was multiplied out, giving $F t n$ and $h c/ t n$⁠. This leaves only the contribution of average grain size for each term, which are both plotted vs $1/kT$ in Fig. 3(b).

There is a strong correlation between the model and the measurements. It was seen that lower temperature and thicker nitridation layers resulted in lower FWHMs and, therefore, fewer threading dislocations. This agrees with the previous observations that lower temperature increases average grain size [Fig. 3(b)] and that $ε r∝ t n − 1 / 2$⁠. The results also suggest avenues for further improvement. By decreasing the temperature to 550 °C, a FWHM of 136 arc sec was obtained for a 500 nm film. To the authors' knowledge, this is a record compared to the aforementioned 330 arc sec achieved by MOCVD.²⁰ 

XRD reciprocal space mapping of the GaN(104) reflection was conducted to provide a reference for the mosaic twist angles and lateral coherence lengths; a table of values is presented in Paper I under review.³⁸ Via the extrapolation method presented by Srikant et al., twist angles were found to vary from 11.7 to 18.1 mRad for nitridation temperatures of 600 and 700 °C, respectively.³⁹ Edge dislocation densities were calculated to be in the mid-low 10¹⁰ cm⁻² range, typical of heteroepitaxial GaN films. Further study is needed to explore the impact of these dislocations on electron density and carrier transport, which will be investigated in detail in future work.

Figure 4(b) shows a high-resolution STEM image of the interface. The epitaxial relationship between the Ga₂O₃ (010) and the GaN (100) directions can be observed, as well as the transition from the shorter Ga₂O₃ b lattice constant (0.304 nm) to the longer GaN a lattice constant (0.319 nm). This difference results in misfit strain, with the misfit dislocations being aligned along the GaN (120) zone axis (out of plane). These misfit dislocations are more clearly seen in Fourier filtered images, as seen in Fig. S5 of the supplemental material. It was observed that the structural transition between the GaN and Ga₂O₃ crystals is very abrupt, with mixing limited to 1–2 monolayers. However, lattice distortion was visible both above and below the interface, suggesting that nitrogen and oxygen mix on a larger length scale. A dark region was seen below the interface, which for HAADF-STEM indicates a relative reduction in the diffraction signal. While this is a product of the nitridation process, the cause is not entirely understood. It is possible that it is due to a decreased density of diffracting species, lower crystallinity, or locally thinner material. It is tentatively attributed to either the presence of an amorphous oxynitride species that is distributed into the substrate or the overlapping of different grains or both. Immediately above the interface, a uniform region of GaN capped by a thin distorted region was observed, which then transitioned into the bulk of the film. At present, it is assumed that this distorted GaN is the true boundary between the grown film and the nitridation layer.

The dark regions were localized to the interface and were laterally discreet, showing that the nitridation layer is not continuous, but rather made up of sub-grains, as previously proposed. Figures 4(c) and 4(d) show lower magnification STEM images of the nitridation layer at the two different nitridation temperatures. The higher temperature nitridation of Fig. 4(c) showed more dense but shorter dark regions, supporting that they may be due to the overlap of two or more sub-grains. The more uniform regions in Fig. 4(d) also seemed to produce fewer dislocations in the region above the nitridation layer.

These observations combined with Fig. 3(b) verify the temperature dependence of the grain size. The grains should form through nucleation of the GaN phase, with the average grain area being inversely proportional to the number of grains. Therefore, the temperature dependence of the grain size should be related to the nucleation rate. Kinetic studies are currently being conducted by our group to better understand the mechanism behind this process.³⁸ Low temperature buffer layers resulting in higher quality epitaxy has also been observed in the nitridation of Al₂O₃, with better grain coalescence observed at lower temperatures.⁴⁰ Further analysis via XRD and 4D-STEM can provide a more quantitative description of the relationship between strain relief and mosaic parameters and is currently being investigated by our group.

With a relationship established between the nitridation parameters and the film FWHM, a connection must be made to the screw dislocation density and device leakage current. If screw dislocations are approximately randomly distributed rather than confined at grain boundaries, their density can be estimated by $ρ s∼ ( β / ( 2.1 b s ) ) 2$⁠. To achieve a dislocation density suitable for high performance devices, a FWHM around 50 arc sec is, therefore, desirable. From the model, this should be achievable by further decreasing the temperature, however, the nitridation rate also slows dramatically at lower temperatures, resulting in diminishing returns due to $ε r∝ t n − 1 / 2$⁠. More complex methods of nitridation are, therefore, being investigated to maximize the grain size while still achieving an appreciable nitridation thickness. Application of more advanced methods is currently being investigated for the purposes of growing high-quality p-GaN on Ga₂O₃, with some preliminary success being achieved. Mg incorporation has been achieved as well as hole activation, as shown in Fig. S6 of the supplemental material. Improvement to defect density and surface quality are still being investigated for p–n diodes.

This work has demonstrated conversion of (−201) β-Ga₂O₃ to (0001) h-GaN through a nitridation process. A GaN-decomposition/N-diffusion controlled conversion of Ga₂O₃ to GaN was observed. Significant improvement of buffer layer performance was achieved through modulation of plasma power density and control of substrate temperatures, as characterized by XRD and TEM of grown films. Controllably improved material quality was achieved, allowing for record-low FWHMs for GaN-on-Ga₂O₃. Continuing buffer layer improvement is being investigated through kinetic studies and more advanced XRD and TEM methodologies. Further, the application to p-GaN on n-Ga₂O₃ is being investigated to enable high performance devices.

See the supplementary material for further information regarding the full range XRD of the nitridation layer, plasma source characterization, PPD limitations, time evolution of surface roughness, Fourier filtered imaging of misfit dislocations, and C–V of p-GaN on Ga2O3.

This work was financially supported by the Office of Naval Research Award No. N00014-21-1-2544 (PM: Lynn Petersen) and the Kim-Fund of the University of Illinois (Grant No. UIUC-933008-633134). Device processing was carried out using the facilities at the Holonyak Micro and Nanotechnology Laboratory, and materials characterization was carried out in part in the Materials Research Laboratory Central Research Facilities, University of Illinois. Discussions with Professor Jian-Min Zuo are gratefully acknowledged.

Distribution statement A. Approved for public release: distribution unlimited.

The authors have no conflicts to disclose.

Frank P. Kelly: Conceptualization (equal); Data curation (equal); Formal analysis (lead); Investigation (equal); Writing – original draft (lead); Writing – review & editing (equal). Matthew M. Landi: Conceptualization (equal); Data curation (equal); Formal analysis (supporting); Investigation (equal); Writing – review & editing (equal). Riley E. Vesto: Formal analysis (supporting); Writing – review & editing (equal). Marko J. Tadjer: Conceptualization (supporting); Resources (supporting); Writing – review & editing (supporting). Karl D. Hobart: Conceptualization (supporting); Resources (supporting). Kyekyoon Kim: Conceptualization (equal); Formal analysis (supporting); Funding acquisition (lead); Project administration (lead); Resources (lead); Supervision (lead); Writing – review & editing (equal).

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