WorksPhysics and Astronomy
Role of carbon and hydrogen in limiting -type doping of monoclinic (AlGa)O
Abstract
We use hybrid density functional calculations to assess n-type doping in monoclinic (AlGa)O alloys. We focus on Si, the most promising donor dopant, and study the structural properties, formation energies and charge-state transition levels of its various configurations. We also explore the impact of C and H, which are common impurities in metal-organic chemical vapor deposition (MOCVD). In GaO, Si is an effective shallow donor, but in Al Si acts as a DX center with a (+/-) transition level in the band gap. Interstitial H acts as a shallow donor in GaO, but behaves as a compensating acceptor in n-type Al. Interpolation indicates that Si is an effective donor in (AlGa)O up to 70% Al, but it can be compensated by H already at 1% Al. We also assess the diffusivity of H and study complex formation. Si-H complexes have relatively low binding energies. Substitutional C on a cation site acts as a shallow donor in GaO, but can be stable in a negative charge state in (AlGa)O when x>5%. Substitutional C on an O site (C) always acts as an acceptor in n-type (AlGa)O, but will incorporate only under relatively O-poor conditions. C-H complexes can actually incorporate more easily, explaining observations of C-related compensation in GaO grown by MOCVD. We also investigate C-H complexes, finding they have high binding energies and act as compensating acceptors when x>56%; otherwise the H just passivates the unintentional C donors. C-H complex formation explains why MOCVD grown GaO can exhibit record-low free-carrier concentrations, in spite of the unavoidable incorporation of C. Our study highlights that, while Si is a suitable shallow donor in ALGO alloys, control of unintentional impurities is essential to avoid compensation.
Introduction
Monoclinic (-) is a wide-band-gap material (4.76–5.1 eV [1, 2, 3, 4]) with a high breakdown field (6-8 MV/cm) [5]. Despite its wide band gap, can be controllably -type doped; together with the availability of high-quality yet low-cost substrates these properties render - highly promising for applications in high-power electronics and UV optoelectronics [6, 7, 8].
Effective control of the carrier concentration by doping with shallow donors is essential for device applications. This typically requires that the donor impurity has a low ionization energy and that compensation can be avoided. Si, Ge and Sn have been demonstrated to be effective shallow donors in with modest ionization energies ( meV) [9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21]. In addition to these dopants, first-principles calculations have identified interstitial hydrogen () and carbon on a Ga site () as shallow donors [20, 22]. Compensation of the shallow donors can occur due to native defects [23, 24] or impurities that act as acceptors, or due to self-compensation if the dopant can occur in different configurations. One form of self-compensation involves the formation of substitutional species on either cation or anion sites. For example, carbon in can occur either on the Ga site (), acting as a donor, or on the O site (), acting as an acceptor [22]. Self-compensation can also occur by formation of so-called centers, in which an impurity expected to act as a shallow donor exhibits a large lattice relaxation and becomes negatively charged by trapping electrons, thus effectively acting as a deep acceptor. In , Si, Ge, and Sn do not form centers [16, 20, 25]. However, the likelihood of formation of a center increases as the band gap increases, as is well known for AlGaAs [26] and AlGaN alloys [27].
Alloying with Al raises the band gap of [28, 29]. Heterojunctions of and exhibit a high-density two-dimensional electron gas, which is at the core of field-effect transistors [30]. Modulation doping is required for high mobility, raising the issue of whether -type doping of can be achieved. Addressing this issue requires investigating whether dopants that are effective for remain shallow donors in , and whether compensation will occur. In addition, recent experiments on films grown by metal-organic chemical vapor deposition (MOCVD) indicated that control of doping might be challenging; it was found [31] that doping with Si failed to result in -type doping below a threshold Si concentration. Above the threshold, an abrupt enhancement of the carrier concentration was observed. Identifying the origin of the compensation in alloys is therefore important for further improvements in electronic devices.
Varley et al. [21] recently reported a first-principles study of -center formation for a large number of candidate donor impurities in . They found that Si emerged as the best candidate, since it continues to act as a shallow donor up to high Al concentrations in the alloy. In the present work we therefore focus on Si as the donor; we perform a more detailed study of its behavior as a center, and investigate other sources of compensation and potential complex formation.
Our study, based on hybrid density functional theory (DFT), addresses Si impurities in monoclinic AlO (hereafter denoted as -AlO), and reveals a new configuration of the state. Compensation by native point defects has been previously addressed [23, 24]. Here, we perform an in-depth study of compensation due to other impurities, particularly carbon and hydrogen. Both of these impurities are readily incorporated during MOCVD growth. Carbon on a cation site [22] and hydrogen [20] are both shallow donors in GaO, but possess (0/−) or (+/−) charge-state transition levels in the band gap in AlO. By interpolating between the end compounds, GaO and AlO, we can estimate the position of these defect levels in (AlGa)O alloys, and determine the Al composition at which the onset of compensation occurs. [Note that we use the term “defect” to denote both native point defects and impurities.] Our results show that Si is an effective donor in (AlGa)O up to 70% Al, but it can be compensated by hydrogen already at 1% Al, or by carbon on cation sites at 5% Al.
Given that hydrogen is expected to be quite mobile, we also perform a study of its migration properties. We further assess the possibility of forming hydrogen-related defect complex with Si and C. Si–H complex formation is not a major concern, since the complexes have low binding energies. Binding energies are much higher for C–H complexes; in fact, C–H behaves almost as a fixed entity, with properties very similar to a nitrogen impurity. C–H complexes can act as compensating acceptors in (AlGa)O alloys, but only when the Al content exceeds 56%; otherwise the complexes are mainly neutral and hydrogen passivates the unintentionally carbon donors. This may explain why MOCVD-grown GaO can exhibit record-low free-carrier concentrations [32], in spite of the probably unavoidable incorporation of carbon.
Substitutional carbon on an oxygen site (C), finally, always acts as an acceptor in -type (AlGa)O, irrespective of Al concentration. C will incorporate only under relatively cation-rich (oxygen-poor) conditions; we find that C–H complexes can actually incorporate more easily, explaining experimental observations of carbon-related compensation in GaO grown by MOCVD [33, 34].
The paper is organized as follows. In Sec. II, the calculation details, the definition of defect formation energies, and the physics of centers are introduced. The main results for Si, C, and H and consequences for carrier compensation are presented in Secs. III A, III B, and III C, respectively; complexes are discussed in Sec. III D. Section IV concludes the paper.
Methodology
Computational details
We perform DFT calculations using the projector augmented wave method (PAW) [35] implemented in the Vienna -initio Simulation Package (VASP) [36, 37]. We focus on the monoclinic phase of GaO (denoted -GaO) and AlO (-AlO). The structure, illustrated in Figure 1(a), contains two types of cation sites: the tetrahedral site (denoted as I) and the octahedral site (denoted as II). In addition, there are three types of O atoms: three-fold coordinated O(I) (on a shared corner of two edge-sharing AlO octahedra and one AlO tetrahedron), three-fold coordinated O(II) (on the shared corner of one AlO octahedron and two AlO tetrahedra), and four-fold coordinated O(III).

Figure 1. (a) Conventional cell of monoclinic AlO. Two inequivalent Al sites are labeled: tetrahedrally coordinated Al(I) in aqua and octahedrally coordinated Al(II) in purple. For the O sites, threefold-coordinated O(I) are in red, threefold coordinated O(II) in purple, and fourfold-coordinated O(III) in gray. (b) Supercell constructed by creating a multiple of the conventional monoclinic cell of -AlO, with one tetrahedrally-coordinated Al replaced by Si. Light blue, red and dark blue spheres denote Al, O and Si atoms, respectively. Structural visualization was performed using VESTA [38].
A plane-wave energy cutoff of 400 eV was employed and Brillouin-zone integration was carried out using the -centered k-point mesh for the primitive cell and for the supercell. The PAW potentials correspond to the valence-electron configurations for Ga, for Al, and for O. To correctly describe the electronic structure as well as charge localization we use a hybrid functional, specifically the functional of Heyd, Scuseria, and Ernzerhof (HSE) [39], with a mixing parameter of . This produces a band gap of 4.83 eV for -GaO and 7.41 eV for -AlO, in good agreement with the experimental band gaps of 4.80 for -GaO and 7.40 eV [40] for -AlO. An ordered AlGaO alloy is also investigated, with all Al atoms on octahedral sites and all Ga atoms on tetrahedral sites [28]. The use of an ordered structure serves as an approximation; the
aRef. 41; bRef. 42; cRef. 2; dRef. 3; eRef. 43; fRef. 40.
computation of structures representative of disordered alloys would be computationally prohibitive. We note that for the 50% Al content, ordered AlGaO was found to be a line compound due to the preference for Al to occupy octahedral sites [28]. Spin polarization is included and full structural optimizations were performed. The optimized computed structural parameters of -GaO, AlGaO and -AlO are summarized in Table I; they compare favorably with experiment.
| GaO | AlGaO | AlO | ||||
| Calc | Expt | Calc | Expt | Calc | Expt | |
| 12.14 | 12.21a | 11.86 | 12.00a | 11.66 | 11.85e | |
| 3.02 | 3.04a | 2.94 | 2.98a | 2.88 | 2.90e | |
| 5.78 | 5.81a | 5.69 | 5.73a | 5.57 | 5.62e | |
| 103.77 | 103.87a | 104.25 | 104.03a | 104.04 | 103.83e | |
| −10.22 | −11.29b | −13.87 | −16.11 | |||
| 4.83 | 5.81 | 7.41 | 7.40f | |||
| 4.87 | 4.76c, 4.88d | 5.89 | 7.74 |
Table I. Structural parameters (lattice parameters, Å; angle , degrees), formation enthalpy per formula unit (, eV/f.u.), and band gaps (, eV) for monoclinic GaO, AlGaO and AlO. Experimental results are also listed for comparison.
A 120-atom supercell was constructed for the point-defect calculations, as shown in Figure 1(b). Selected tests were performed using 160-atom supercells. Tests for larger supercells are too expensive to perform with the hybrid functional, but supercell size convergence was further checked using the generalized gradient approximation of Perdew, Burke, and Ernzerhof (PBE) [44] for Si and C impurities in supercells containing up to 1280 atoms. Transition levels were found to change by less than 0.1 eV. For each point defect, we perform multiple calculations with different symmetry-breaking initial local distortions around the defect in the initial state, and optimize the atomic positions until the Hellmann-Feynman forces are lower than 5 meV/Å. All the supercell calculations are consistently performed at the lattice parameters obtained with the same plane-wave energy cutoff, as listed in Table I.
Migration barriers are calculated using the climbing-image nudged elastic band (cNEB) method [45]. To mitigate computational cost we perform one-shot HSE calculations for the migration barriers (): we use PBE [44] in the cNEB calculations, followed by static HSE total energy calculations based on the initial and barrier geometries. We tested the accuracy of this approach for the migration of H in AlO and GaO along the [10] direction; full HSE calculations yield migration barriers that are within 0.1 eV of the results obtained using one-shot HSE calculations.
After determining the migration barrier of a defect, we can estimate the temperature at which the defect becomes mobile with transition state theory [46]. This temperature is denoted as an “annealing temperature”, above which the defect is in thermodynamical equilibrium. In transition state theory [46], the rate at which the defect hops to a neighboring site can be expressed as
where is the Boltzmann constant and the migration barrier. The prefactor is an effective frequency associated with the vibration of the defect; for the H interstitial in both GaO and AlO, we choose , which is approximately the dominant O-H vibrational frequency GaO [23, 47]. We estimate the annealing temperature as the temperature at which the rate [48]. We note that the annealing temperature is not very sensitive to the choice of .
Formation energy
The formation energy of a defect is used to assess the likelihood of the presence of the defect and its concentration. For Si substituting on an Al site (Si), the formation energy is calculated as
where is the total energy of one Si in charge state in the supercell, is the total energy of the bulk supercell, and is the Fermi energy, referenced to the valence-band maximum (VBM). is a finite-size correction term for charged defects [49, 50]. We adopt the previously reported values for the dielectric constants of GaO, AlGaO, and AlO [21]. The chemical potentials are referenced to the elemental phases, e.g., and .
The Al and O chemical potentials have to fulfill the stability condition for bulk AlO:
where is the formation enthalpy of bulk -AlO, as shown in Table I. The calculated formation enthalpies of -GaO and AlGaO are listed in Table I as well. corresponds to Al-rich conditions, and to O-rich (Al-poor) conditions. For purposes of presenting results for the impurities, we choose the chemical potentials to correspond to the solubility limit, i.e., the highest value of the chemical potential that is compatible with formation of other phases that can result from interactions of the impurity with the host elements. For , this correspond to equilibrium with SiO:
where is the calculated formation enthalpy of SiO. For C and H, the upper limits depend on the host chemical potentials: the limits for correspond to for Al-rich and for Al-poor conditions; the limits for correspond to for Al-rich and for Al-poor.
To be able to comment on the behavior in alloys, we also calculate formation energies in . The limiting phases for Si and H are the same as those in the case. The limits for in correspond to graphite for Ga-rich and for Ga-poor conditions. We note that this is different from Ref. 22, where diamond was assumed as the limiting phase under Ga-poor conditions.
The charge-state transition level between charge states and , denoted as , is calculated based on the formation energies:
Stability of DX centers
Following Chadi and Chang [26], we label the neutral and positively charged donors on the substitutional site and , respectively. is used to denote the configuration in the negative charge state. The formation of a center leads to self-compensation, as described by the process
The center can be characterized by the effective correlation parameter , defined by
Stability of the center is typically characterized by a negative value of . However, using as the descriptor of the center can be problematic because of the challenges involved in accurately calculating the energy of the neutral charge state when the level (and hence the Kohn-Sham state for ) is near or above the conduction-band minimum (CBM). Instead, we will use the charge-state transition level as a descriptor: our criterion will be that the center is stable if the level lies below the CBM.
Results and discussion
Silicon
The formation energies of Si on the tetrahedral () and octahedral () sites in - are shown in Figure 2(c,Figure 2d) for cation-rich and cation-poor conditions. For completeness and consistency, we also show the formation energies of Si in in Figure 2(a,Figure 2b); these have been calculated previously [20] and our results are consistent with the previous results. Note that in Figure 2 (and all subsequent formation-energy figures) we keep
Figure 2. Formation energy versus Fermi level for Si and C impurities in (a)-(b) -GaO and (c)-(d) -AlO. (a) and (c) are for cation-rich, and (b)-(d) for cation-poor conditions. The grey area indicates the conduction band of -GaO.
the horizontal-axis scale the same for - and -, which facilitates comparisons of formation energies. In principle, we could also take the valence-band alignment between - and - into account to more accurately compare charge-state transition levels [51, 52]; however, since the valence-band offset is relatively small [28], we do not include this alignment.
Similar to -, Si prefers to be incorporated on the tetrahedral site in -: is lower than by 0.80 eV. In - Si behaves as a shallow donor on either site, i.e., only the positive charge state is stable for all Fermi-level positions within the band gap. In - a charge-state transition level occurs in the band gap, at 0.62 eV below the CBM for and at 0.43 eV below the CBM for (see Table II). Table II also lists values for the other transition levels. For , there is a small region of stability for the neutral charge state; we found and transition levels at 6.77 eV and 6.81 eV, respectively, corresponding to eV. For , the level lies above the level and we have a negative- center (i.e., the neutral charge state is never thermodynamically stable), with eV.
For Fermi levels in the upper part of the gap, is in a negative charge state, associated with the formation of a center. For Si we found that two locally stable configurations exist. The local geometry and the charge density of the localized electrons are illustrated in Figure 3. Figure 3(a) corresponds to the ground state

Figure 3. Local structure and localized charge density for negatively charged configurations of (a) ground-state Si, (b) metastable-state Si, and (c) Si in -AlO. Dark blue spheres denote Si, light blue Al, and red O. Percentage changes of bond lengths, referenced to bonds in bulk AlO, are indicated. Isosurfaces of the charge density () of the localized state in the centers are shown in green.
| -GaO | (+/0) | (+/−) | (0/−) | ||
| C | 5.20 | 4.92 | 4.65 | −0.54 | |
| C | 4.32 | 5.05 | 5.77 | 1.46 | |
| H | 5.47 | 4.84 | 4.21 | −1.25 | |
| H | 5.47 | 5.20 | 4.91 | −0.55 | |
| H | 4.98 | 5.01 | 5.04 | 0.07 | |
| H | 5.71 | 5.56 | 5.42 | −0.29 | |
| AlGaO | (+/0) | (+/−) | (0/−) | ||
| Si | 6.39 | 6.19 | 6.00 | −0.39 | |
| Si | 6.065 | 6.46 | 6.85 | 0.79 | |
| C | 4.93 | 4.81 | 4.69 | −0.24 | |
| C | 4.05 | 4.97 | 5.90 | 1.86 | |
| H | 5.70 | 4.67 | 3.65 | −2.05 | |
| H | 5.92 | 5.69 | 5.45 | −0.47 | |
| H | 5.60 | 5.21 | 4.83 | −0.77 | |
| H | 5.69 | 5.70 | 5.71 | 0.02 | |
| -AlO | (+/0) | (+/−) | (0/−) | ||
| Si | 6.77 | 6.79 | 6.81 | 0.04 | 70% |
| Si | 7.20 | 6.98 | 6.76 | −0.44 | 81% |
| C | 4.96 | 4.95 | 4.95 | −0.01 | 5% |
| C | 4.23 | 5.07 | 5.90 | 1.95 | 51% |
| H | 5.83 | 4.83 | 3.84 | −1.99 | 1% |
| H | 6.51 | 6.36 | 6.41 | −0.10 | 37% |
| H | 5.78 | 5.70 | 5.61 | −0.16 | 13% |
| H | 6.11 | 6.12 | 6.13 | 0.03 | 41% |
Table 2. Charge-state transition levels (eV) Eq. (5) and effective correlation parameter (eV) Eq. (7) for various impurities in -GaO, ordered AlGaO, and -AlO. The neutral and negative charge states that are used to compute charge-state transition levels and all correspond to localized states. Values for Si in GaO are missing since the localized neutral and negative charge states cannot be stabilized. We also list (%), the Al concentration in (AlGa)O corresponding to the onset of behavior.
of Si, while Figure 3(b) represents a metastable state that is 0.40 eV higher in energy. The metastable state [Figure 3(b)] features the type of broken bond that is often considered characteristic of a center, resulting in a dangling bond on the Si atom that is filled with two electrons. The ground state of Si [Figure 3(a)] also displays significant bond distortions, both in the SiO tetrahedron and its nearest AlO; however, no clear bond breaking occurs. Instead, the main effect of the distortion is to reduce the Si-Al distance from 2.88 Å (the distance in bulk -AlO) to 2.32 Å, creating a two-electron bond between Si and Al, as clearly seen in Figure 3(a). Octahedrally coordinated Si also forms a center, shown in Figure 3(c). Three Si-O bonds are broken in the SiO octahedron, and two electrons are localized on the Si dangling bond. Tests for the Si center in a 160-atom supercell yielded very similar local geometries, with bond lengths differing by less than 0.01 Å. The 160-atom supercell calculations yield a (+/−) charge-state transition level that is within 0.05 eV of the result using a 120-atom supercell.
The formation of the silicon center indicates that Si cannot be a shallow donor in -AlO. To gain insight into its effectiveness in (AlGa)O alloys, we evaluate the position of the (+/−) transition level and compare it with the CBM in (AlGa)O alloys. The negative charge state employed to compute the (+/−) transition level has to correspond to a localized state. Such a localized state can be stabilized in AlO and ordered AlGaO alloy but not in GaO. Therefore, the (+/−) transition levels in alloys are interpolated (or extrapolated towards GaO) based on the values in AlO and ordered AlGaO alloy (see values in Table II). The values of the CBM are interpolated based on those in GaO and AlO, with the addition of bowing using a bowing parameter of 0.93 eV and assuming all of the band-gap bowing occurs in the CBM [28].
The resulting values of , the Al concentration at which the (+/−) level moves below the CBM (i.e., the center becomes stable) are listed in Table II. We found Si incorporating on the tetrahedral site will be an effective donor over a wide range of Al alloying, up to 70% Al (see Figure 4). This value is somewhat lower than the reported in Ref. 21. The difference can be attributed to the fact that our calculated level in AlO (AlGaO) is 0.12 eV (0.37 eV) lower than that in Ref. 21, due to the fact that we identified a lower-energy structure for the center. In addition, we find that Si on the octahedral site is also an effective donor in (AlGa)O, up to 81% Al (see Figure 4).

Figure 4. The levels of Si, Si, C, H, H, H, H, and the level of C obtained from calculations in AlO, GaO and ordered AlGaO alloys. The notation Si, resp. Si, denotes substitutional Si on the tetrahedral (I), resp. octahedral (II) cation site, and similarly for C and C. The charge-state transition levels for Si and Si in GaO are missing since the localized neutral and negative charge states cannot be stabilized. The dashed lines connecting the calculated values are just a guide to the eye. The CBM/VBM band offsets as a function of the Al concentration for alloys were obtained from Ref. 28.
Our treatment of defect behavior in the alloy by performing linear interpolation (Figure 4) is obviously an approximation. In the case of (AlGa)O alloys, it is likely to be quite reliable, due to the similarity in the properties of Al and Ga cations; indeed, we have found that the geometries of defect configurations are very similar in -GaO and -AlO. Some degree of validation is also obtained from the results for transition levels explicitly calculated for AlGaO alloys. As shown in Figure 4, the results for the alloy generally fall on the line that linearly interpolates between -GaO and -AlO; the small deviations that occur for C and H impurities amount to less than 0.15 eV, which is gratifying agreement. We also note that this interpolation procedure has been demonstrated to be effective for obtaining transition levels in AlGaN alloys [27, 53]. Indeed, experiment [54] indicated that the O transition level in AlGaN alloys varies linearly with alloy concentration, supporting the validity of linear interpolation.
Carbon
Incorporation
Carbon is a commonly observed background impurity in semiconductors, particularly in MOCVD-grown films. In GaO, carbon on the tetrahedral site is a shallow donor [22, 55]. Here we investigate whether the increase in band gap in (AlGa)O will cause the carbon level to drop below the CBM, turning carbon into a compensating acceptor. We also study whether carbon on the oxygen site could cause compensation.
Lyons et al. [22] investigated carbon impurities on the anion, cation, and interstitial sites in GaO and showed that under most conditions, the carbon impurity prefers the substitutional tetrahedral cation site. For a consistent comparison, we reproduced the formation energies of carbon impurities on the cation sites in GaO [see Figure 2(a,Figure 2b)] and our results agree with Lyons et al. [22] under the same condition. Similar to the case of GaO, C prefers to occupy the tetrahedral site (C) in monoclinic AlO, with a formation energy 2.10 eV lower than that of C at the VBM (see Figure 2). At the CBM, the formation energy of C is 2.03 eV lower than that of C. The behavior of C as a compensating center
is broadly consistent with previous results for carbon in -AlO, the corundum phase [56]. Figure 2 also shows that under Al-rich conditions the formation energy of C is high, significantly decreasing when we move to Al-poor conditions. However, carbon incorporation in MOCVD-grown samples is likely not governed by thermodynamic equilibrium, but rather determined by unintentional incorporation due to incomplete dissociation of metal-organic precursors. In the following we therefore examine both C and C.
Carbon on the tetrahedral site
The small atomic size of carbon leads to bond-breaking and a variety of possible geometries, as depicted in Figure 5. Focusing first on C, already in the positive charge state we find two competing configurations of C. In the most stable state (not depicted), C forms four C-O bonds, and the oxygen cage shrinks, yielding four C-O bonds with bond lengths ranging from 1.40 to 1.44 Å. In the metastable configuration [see Figure 5(a)], which is only 87 meV higher than the ground state, the C-O(III) bond is broken, and the remaining three C-O bonds have similar lengths (1.28, 1.30, 1.30 Å) and almost lie in the same plane, with nearly 120° angles between the bonds. Carbon effectively forms an -bonded configuration with its three O neighbors. This 3-fold coordination was also found for C in GaO [22], with very similar C-O bond lengths (1.27–1.30 Å). The 3-fold-coordinated configuration is the lowest-energy configuration in GaO and AlGaO.

Figure 5. Local structure and localized charge density for (a) metastable 3-fold-coordinated configuration of , (b) , (c) , (d) , (e) and (f) in -. Brown spheres denote C, light blue Al, and red O. Isosurfaces of the charge density (0.01 ) of the localized electrons in , , and are shown in green.
For the neutral charge state of C, we found that we can stabilize as many as five distinct atomic configurations, starting from initial structures in which different C-O bonds are broken in a CO tetrahedron. The most stable configuration corresponds to breaking of the C-O(III) bond, forming a nonplanar CO cluster, as shown in Figure 5(b). The C-O bond lengths are 1.35, 1.39, 1.39 Å, while the distance between C and O(III) is 2.82 Å. This configuration is similar to the 3-fold coordination for the positive charge state, but the carbon has moved beyond the plane of its oxygen neighbors. As seen in Figure 5(b), the C has moved towards a next-nearest Al(I) neighbor in the same (010) plane, reducing the C-Al distance from 3.5 Å (for C on the nominal Al site) to 2.6 Å. In the process, the charge of the localized state becomes confined between C and that next-nearest-neighbor Al atom. The fact that this state is localized clearly indicates that C is not a shallow donor in θ-AlO. Indeed, the corresponding Kohn-Sham state lies 4.42 eV below the CBM.
Another configuration in which the same C-O(III) bond is broken can be stabilized with a distance of 2.38 Å between C and O(III); this configuration is only 0.05 eV higher in energy. Other configurations involve breaking a C-O(I) bond (0.80 eV higher than the ground state) or breaking a C-O(II) bond (0.91 eV higher than the ground state). The geometry in which all four C-O bonds in the CO tetrahedron are maintained is also a local minimum, 0.89 eV higher in energy than the ground-state C-O(III) bond-breaking configuration.
In the negative charge state, the most stable state is similar to the ground state of C, with a broken C-O(III) bond [see Figure 5(c)]. Other configurations can be stabilized, but with energies that are at least 1.3 eV higher than the ground state. Figure 5(c) depicts the charge density of the localized state. Comparison with Figure 5(b) indicates that with the addition of an extra electron a C-Al bond has formed with the next-nearest-neighbor Al(I). The C-Al distance is reduced to 2.0 Å after relaxation. This is another example of a DX configuration in which a cation-cation bond forms, but we note the distinction with the DX configuration of Si shown in Figure 3(a): Si bonds with a nearest-neighbor tetrahedral host cation, while C bonds with a next-nearest-neighbor tetrahedral host cation. A similar bonding configuration for the negative charge state of carbon on the tetrahedral site is also observed in GaO and AlGaO.
Carbon on the octahedral site
Turning now to C, a variety of competing local geometries with different broken bonds are also present in the positive, neutral and negative charge states due to the small atomic size of carbon. In the positive charge state, the ground state corresponds to bond breaking of two C-O(III) bonds and one C-O(I) bond in the nominal octahedron, leaving the remaining three C-O bonds (1.28, 1.28, 1.30 Å) to form a 3-fold coordinated configuration that is similar to the metastable configuration of C [see Figure 5(d)]. We also observe a competing metastable state of C with a different 3-fold coordinated configuration that exhibits bond breaking of all three C-O(III) within the octahedron. This metastable state is 0.20 eV higher than the ground state. This 3-fold coordinated configuration with three broken C-O(III) bonds becomes the ground state in the neutral charge state [see Figure 5(e)]. The remaining three C-O bonds form a nonplanar CO cluster, with C-O bond lengths of 1.37, 1.37, 1.41 Å. The charge of the localized state, dominated by a -orbital character, is confined between C and the center of the octahedron. Another (metastable) 3-fold coordinated configuration with bond breaking of two C-O(III) and one C-O(II) is 0.28 eV higher.
In the negative charge state of C, the most stable configuration is 2-fold coordinated, with two equivalent C-O(I) bonds (1.40, 1.40 Å) [see Figure 5(f)]. The localized states exhibit roughly equal - and -orbital character and the associated charges are confined near C, with an appreciable weight at the center of the octahedron. We also find another competing 2-fold coordinated metastable configuration, with one C-O(I) bond (1.26 Å) and one C-O(II) bond (1.91 Å) remaining, and with an energy 0.18 eV higher.
Carbon on cation sites as a compensating center
When discussing substitutional C in (AlGa)O alloys, C could be replacing either a Ga or an Al atom; in this section (as in Figure 4) we use the generic notation C to denote either C or C, and C to denote either C or C. C is a negative- center across GaO, AlO and the ordered (AlGa)O alloy (see Table II). We can evaluate its (+/−) transition level in the relevant range of (AlGa)O alloy compositions by interpolating between GaO and the AlGaO alloy, with consideration of the band bowing [28] (see Figure 4). For C, we find that the impurity starts acting as an acceptor in (AlGa)O alloys already at 5% Al. C is not a negative- center, but it can still act as compensating acceptor in -type AlO (see Figure 2). To estimate the onset of C as a compensator in (AlGa)O alloys, we interpolate the (0/−) transition levels between AlO and the AlGaO alloy (see Figure 4), finding that C starts acting as an acceptor at 51% Al.
Our results indicate that unintentional C incorporation is a potential explanation for the lack of doping efficiency observed at low concentrations of Si in MOCVD-grown alloys. Since unintentionally incorporated C(I) acts as a compensating acceptor in with , the Si concentration needs to exceed the concentration of C(I) in order for Si to actually provide -type doping.
Carbon on oxygen sites,
In , it was found that acts as a compensating acceptor when the Fermi level is high in the gap [22]. Reference [22] reported results for a single O site and only for O-rich conditions; in Figure 6(a,Figure 6b) we report a full set of results for in . Under conditions relevant for -type doping, the formation energy of is lowest when the Fermi level is at the CBM under Ga-rich conditions; however, even under these extreme conditions, this formation energy is still 2.69 eV.
Figure 6. Formation energy versus Fermi level for C in (a)-(b) β-GaO and (c)-(d) θ-AlO. (a) and (c) are for cation-rich, and (b)-(d) for cation-poor conditions. Carbon incorporation on three possible O sites is considered: O(I), O(II), and O(III). The grey area indicates the conduction band of β-GaO.
The formation energy of can be significantly lower in , as shown in Figure 6(c,Figure 6d). For Fermi levels low in the gap (which may be difficult to attain in ), prefers the 4+ charge state. At higher Fermi-level positions, the neutral, 1−, and 2− charge states become more favorable. For Fermi levels close to the CBM, is the energetically most favorable state.
We can therefore conclude that for -type doping conditions in both and , acts as a compensating acceptor. The high formation energy indicates it will not be a concern in , but it could more easily incorporate in alloys. However, as evident from Figure 6, this sensitively depends on chemical potentials, with cation-rich (oxygen-poor) conditions posing more of a problem. We will return to this issue in Sec. III D 4.
Hydrogen
Hydrogen is another impurity that is commonly unintentionally incorporated, particularly in MOCVD-grown samples. We investigate H substituting on the O site (, , and ) and H on the interstitial site () in -; their formation energies are shown in Figure 7(c,Figure 7d). For comparison, we also show our calculated formation energies in - [see Figure 7(a,Figure 7b)], which compare well with previous reports [20]. In -, all these configurations were found to be shallow donors [20], but in - they are stabilized in the negative charge state and thus act as acceptors when the Fermi level is high in the gap.
Figure 7. Formation energy versus Fermi level for substitutional and interstitial hydrogen in (a)-(b) β-GaO and (c)-(d) θ-AlO. (a) and (c) are for cation-rich, and (b)-(d) for cation-poor conditions. The grey area indicates the conduction band of β-GaO.
Substitutional hydrogen,
In , substitutional favors incorporation on the O(I) site. Incorporation of a proton on the O(II) [O(III)] site is 0.29 [0.80 eV] higher in energy (see Figure 7). The same site preference is also observed in the ordered AlGaO alloy: total energies of H and H are 0.08 eV and 0.33 eV higher than that of H, respectively. In GaO, H is also most favorable, and H and H are slightly higher in energy [20].
Under Al-rich conditions, H has a low formation energy ( eV) in AlO, meaning it is easy to incorporate. H and H behave as negative- centers with eV and eV, and a charge-state transition level at 1.05 eV and 1.71 eV below the CBM (see Table II and Figure 7). H is almost a negative- center, with , and and charge-state transition levels at 6.11 eV and 6.13 eV, indicating a tiny (0.02 eV) range of stability of the neutral charge state. The formation of negative- centers is due to the pronounced energy lowering in their negative charge state, induced by major local structural distortions. The local structures are illustrated in Figure 8(a–Figure 8c); two electrons are localized around H and nearby Al sites. In the case of H [Figure 8(a)], the H atom is still relatively close to the O(I) site, but the configurations of H [Figure 8(b)]

Figure 8. Local structure of (a) (b) , (c) and (d) in -. Light blue, red and gray spheres denote Al, O and H atoms. The position of the O atom that was removed in was labeled using a black dashed circle. Isosurfaces of the charge density () of the localized electrons are shown in green.
and H [Figure 8(c)] are more like complexes of an oxygen vacancy plus a nearby interstitial H.
H, H, H will act as acceptors in -type (AlGa)O alloys due to the emergence of the transition level in the band gap starting as a critical Al composition . Interpolating charge-state transition levels between GaO and the ordered AlGaO alloy, the onset Al concentrations for H, H, H acting as compensating centers are 37%, 13% and 41%, respectively (see Figure 4 and Table II).
Interstitial hydrogen, H
Turning now to interstitial H in AlO, we find that the lowest-energy configuration in the neutral and positive charge state has H bonded to the three-fold coordinated O(I). In the negative charge state, H prefers to sit near two Al(I) atoms; the local geometry is illustrated in Figure 8(d). The local configurations of the , 0, and state of H are similar to those in GaO [20]. In the neutral charge state, the Kohn-Sham state of H lies about eV below the CBM. In the negative charge state, the Kohn-Sham state is about 1.3 eV above the VBM, as a result of the formation of two H–Al bonds.
As seen in Figure 7 and Table II, the level of in lies at 4.83 eV above the VBM. is a negative- center, with a large magnitude of eV. Knowledge about the position of the levels in - and - allows us to estimate the band lineup between these materials, according to Ref. 57; this leads to an alignment where the VBM of - lies 0.01 eV below the VBM of -. We can compare this result with values obtained by alignment with the vacuum level derived from surface calculations; depending on the surface orientation, this procedure produces values ranging from to eV [58]. The overall agreement with the hydrogen alignment is gratifying.
Figure 7 shows that the formation energies of in are quite low, suggesting that will readily incorporate, and in -type material it will act as a compensating acceptor. Following our interpolation procedure based on the case of and ordered alloy, the transition level of will already appear in the band gap of alloys at an Al concentration of 1% (see Figure 4).
Migration of interstitial hydrogen
Since hydrogen is expected to be quite mobile, in order to assess whether hydrogen can play a role in compensation it is important to also assess migration barriers. We investigated the migration barrier () of using the PBE functional and a one-shot HSE method, as described in Sec. II A. The results are listed in Table III. The annealing temperatures at which H becomes mobile for and charge states and for various crystallographic directions, calculated using Eq. (1), are also included in Table III. As expected, the migration of in the monoclinic structure is anisotropic. In , we find the migration barrier of along the [10] direction to be 0.28 eV, in good agreement with the 0.34 eV value reported in Ref. 20. The migration path is illustrated in Figure 9(a). Migration along other directions has higher barriers. Along [100] [Figure 9(b)] the barrier is as high as 2.95 eV, and along [1] Figure 9(c), the barrier is 1.73

Figure 9. Potential energy for H migration along the (a) [10], (b) [100], (c) [100] directions in -GaO. Various configurations along the migration path are shown as insets. Green and red spheres denote Ga and O atoms, respectively. Note the very different scales for the potential energy surfaces between (a) and (b) or (c).
Table 3. Calculated migration barriers (, eV) of and in and along three crystallographic directions. The calculated annealing temperature (, K) is listed, assuming a hopping rate of 1 s.
eV. The annealing temperature for migration along [10], [100] and [1] directions are 101, 1062, and 623 K, respectively.
Our calculated results for migration paths and barriers for in are quite similar to the results in ; migration is also strongly anisotropic (see Table III). The low migration barrier suggests that can move along the [10] direction (which is the most commonly used growth orientation) even at temperatures well below room temperature.
For , the migration barriers are generally higher. Particularly striking is that migration in now has the highest barrier along [10] (see Table III). We attribute this to the fact that in the barrier geometry for , the distance between H and the nearest Ga atoms is quite large (2.1 Å). Since it is Coulombic attraction between and the cations that stabilizes the structure, the lack of strong bonding at the saddle point raises the energy. In contrast, in the corresponding distance is only 1.8 Å, providing for a stronger interaction with the Al atoms and lowering the barrier height. While the migration barriers for are higher than for , the annealing temperatures in Table III indicate that may be mobile at temperatures at which growth or processing of and alloys is typically performed.
Figuring out precise migration barriers for in alloys may not be as simple as interpolating the values in GaO and AlO. The barriers to be overcome will be determined by the local environment. If, at a particular Al composition, a “percolation path” exists that allows hydrogen to follow favorable atomic arrangements, the lower of the two migration barriers will likely apply. In the absence of such a path, the higher of the two barriers will need to be overcome.
We note that the mobility of H will be reduced by binding to positively charged impurities, in particular donor dopants. Formation of a Si–H complex will increase the stability of H in the lattice and add to the overall activation energy for hydrogen motion. Therefore interstitial H, in addition to substitutional H, may play a role in compensation of shallow donors in (AlGa)O.
Complexes
Since hydrogen can easily incorporate and has relatively low migration barriers, we need to assess the possibility of complex formation between hydrogen and other impurities, either with the intentionally incorporated Si donor, or with unintentional impurities such as C.
Sication–H complexes
We first discuss complexes with Si on the cation site; as discussed in Sec. III A, we found that Si is an effective donor in (AlGa)O alloys up to 70% Al incorporation for Si(I) and 81% Al incorporation for Si(II). The formation energy of Si–H complexes in GaO is shown in Figure 10(a) for Ga-rich and Figure 10(b) for Ga-poor conditions. We find that similar to the Si impurity, the Si–H complex also has lower energy when Si occupies the tetrahedral cation (I) site.

Figure 10. Formation energy versus Fermi level for isolated Si impurities, H interstitials, and Si–H defect complexes in (a)-(b) -GaO and (c)-(d) -AlO. (a) and (c) are for cation-rich, and (b)-(d) for cation-poor conditions. Dashed lines denote thermodynamic instability of the defect complex, as explained in the text. The grey area indicates the conduction band of -GaO.
The formation energy of Si–H complexes under Ga-poor conditions is simply a rigid shift to higher energies of the formation energy under Ga-rich conditions [Figure 10(a,Figure 10b)]; our discussion about the properties of complexes applies to both. We find that when the Fermi level is high in the gap, Si(I)–H complexes are stable in the neutral charge state, indicating that the Si donor has been passivated. We also find that Si(I)–H has a (2+/0) charge-state transition level at 0.41 eV below the CBM. The formation of a Si–H complex might seem unexpected, given that the constituents, Si and H, both prefer to occur in the positive charge state throughout the band gap of GaO, as evident from Figure 10(a). Our calculated formation energy of point defect Si and H in GaO agree well with the previous study [20]. However, as seen in Table II, the (+/−) level of H in GaO lies at 4.84 eV above the VBM, i.e., just 0.01 eV above the CBM. It is therefore to be expected that for Fermi levels near the CBM, H could be stable enough to lead to formation of a complex with Si, particularly when Coulomb attraction is taken into account.
This is confirmed by evaluation of the binding energy of the complex, defined as
where a positive value of the binding energy signifies a stable, bound complex. This yields a binding energy of 0.59 eV. This value (which is smaller than the formation energy of each of the constituents) indicates the complex will not be stable during growth at high temperatures, since entropy favors the isolated entities (see Ref. 59). In addition, formation after growth is not very likely since will predominantly occur in the positive charge state, which is repelled by .
One may then wonder about the stability of . Indeed, a binding-energy calculation in this case yields a negative value for ( eV), i.e., the sum of the formation energy of the constituents, and , is smaller than the formation energy of the complex. We show the 2+ charge state of this complex with dashed lines in Figure 10, to indicate that this is a locally stable configuration that is, however, thermodynamically unstable.
The notion that results from bonding between and is consistent with the local geometry of the complex: we found in Sec. III C 2 that prefers to bind to two cation atoms on tetrahedral sites, and the configuration of is such that the hydrogen atom is located close to a Ga(I) and to the Si(I) atom as seen from Figure 11(a). This is very similar to the geometry shown in Figure 8(d). For , the bond length of Si(I)-H and Ga(I)-H are 1.53 Å and 1.77 Å, respectively.

Figure 11. Local structure of (a) , (b) , (c) and (d) in . Green, red, gray, blue and brown spheres denote Al, O, H, Si and C atoms, respectively.
If the complex does form during cooldown, one may wonder about its stability at room temperature. The activation energy for dissociation can be estimated by adding the migration barrier for the mobile hydrogen species to the binding energy of the complex. All of the migration barriers are large enough (Table III) to make this activation energy sufficiently high and keep the complex stable at room temperature. Using the lowest migration barrier for in GaO in Table III, we would estimate an activation energy of 0.59+1.57=2.16 eV, which would correspond to an annealing temperature of 778 K. However, since the (2+/0) transition level is so close to the CBM, an alternative dissociation mechanism could occur in which electron excitation to the CBM converts the complex to a 2+ charge state, after which the dissociation will proceed more easily both because of the lower migration barrier of (Table III) and because of the Coulomb repulsion between and .
Venzie et al. [60] recently reported that exposure of Si-doped GaO to a hydrogen plasma led to passivation of the Si donor, as evidenced by a reduction in conductivity and an increase in mobility. The mobility increase indicates that the concentration of ionized-impurity centers is reduced, consistent with the formation of a complex (as opposed to mere compensation, where the compensating centers are spatially separated from the ionized donors). Venzie et al. [60] also proposed a geometry for the complex, which suggests that H binds to an O(I) neighbor of the Si donor; our calculations show that this configuration is 3.23 eV higher in energy than the stable geometry where H is bonded to two cations [see Figure 11(a)]. The configuration with cation-H bonds leads to vibrational frequencies significantly lower than the 3477.6 cm attributed to the Si-H complexes[60], suggesting that these observed modes may be due to other types of complexes that form upon hydrogenation.
The formation energy of the Si–H complex is higher, but otherwise its characteristics are very similar to those of the Si–H complex. The (2+/0) level of Si–H in GaO occurs at 0.35 eV below the CBM (compared to 0.41 eV for Si–H), and the binding energy is 0.47 eV. The local geometry of is similar to that of , with a hydrogen binding to two nearby tetrahedral cation sites [see Figure 11(b)]. The H-Ga bond lengths are both 1.67 Å. This reinforces the argument that a complex in is related to the point defect.
Similar complexes form in , as seen in Figure 10(c). The level of the complex occurs at eV (3.39 eV below the CBM), and the binding energy of the complex is eV. For , we find the transition level at 4.31 eV above the VBM (3.10 eV below the CBM), with a binding energy eV.
Based on our calculated numbers, we can assess whether complex formation with hydrogen has an impact on conductivity in Si-doped alloys. We have found the neutral complex to be stable in both and for Fermi levels close to the CBM, and hence we expect the complex to also be stable in the alloys, with a binding energy that increases with increasing Al content. Indeed, the experiments of Venzie et al. [60] indicated that intentional hydrogenation can produce the complex in . However, as discussed above, the relatively modest binding energy indicates the complex will not be formed at the growth temperature. Complex formation would therefore need to occur during cool down. Since this requires bringing close to a Si impurity in the positive charge state, the increasing stability of as the Al concentration increases makes complex formation more likely. If complex formation would indeed occur, annealing the sample would serve to dissociate the complex and activate the Si donor, as is well known for hydrogen-related complexes in other semiconductors [61].
and –H complexes
For completeness, we also investigated and –H complexes in both and . Formation energies are shown in the Supplemental Material (SM) [62]. itself is stable in positive charge states when the Fermi level is low, but for Fermi levels close to the CBM and behave as compensating acceptors, while is neutral. However, the formation energies are so large in -type material (even under the most favorable O-poor conditions) that they are unlikely to form or play a role in carrier compensation. –H complexes also act as acceptors when is high in the gap, but again their formation energies are so large that they are unlikely to form.
–H complexes
Figures 12(a,b) show the formation energies of C–H complexes in . Similar to the –H in , a charge-state transition level of –H occurs in the band gap, at 1.18 eV below the CBM, despite the fact that the individual impurities and are single donors and occur in the positive charge state throughout the band gap. Following a similar definition as in Eq. (8), we can define a binding energy relative to and , finding eV. This much greater value of the binding energy and lower position of the transition level for the –H complex as compared to the –H complex indicates that a different type of bonding occurs. As pointed out in Sec. III D 1, can be viewed as a combination of and , with the H atom bonding to two tetrahedral cation sites [Figure 11(a)]. In , on the other hand, the H atom binds to C within the tetrahedral oxygen cage, forming a strong C–H bond with a bond length of 1.06 Å [Figure 11(c)]. The strength of this bond, which is related to the small atomic size of C, makes the C–H combination behave as a unit, similar to a nitrogen atom, and lowers the overall formation energy. For the –H complex, the charge state is actually thermodynamically stable, i.e., it is 0.24 eV lower in energy than the sum of the formation energies of and . In the donor configuration of the –H complex, H is bonded to a neighboring oxygen atom within the tetrahedral oxygen cage instead of to C. This is also true for other donor (i.e., positively charged) configurations of the –H complex in both and .
Similar behavior is observed for –H: a transition level occurs at 2.25 eV below the CBM, and the binding energy of the neutral complex is 4.00 eV. In this case, the charge state is also thermodynamically stable, i.e., it is 1.13 eV lower in energy than the sum of the formation energies of and [see Figure 12(a,Figure 12b)]. Again, a C–H bond with a bond length of 1.06 Å forms within the octahedral oxygen cage for [see Figure 11(d)]. This C–H bond formation thus indicates a unique structural feature of the C–H complex regardless of which cation site C substitutes on.
Figure 12. Formation energy versus Fermi level for isolated Ccation impurities, H interstitials, and Ccation–H defect complexes in (a)-(b) -GaO and (c)-(d) -AlO. (a) and (c) are for cation-rich, and (b)-(d) for cation-poor conditions. The grey area indicates the conduction band of -GaO.
Figure 12(a) shows that –H complexes are lower in energy than isolated impurities under -type doping conditions, and this is true over most of the range of Ga chemical potentials. In addition, because of their high binding energy, such complexes may well be present during growth. Measurements of carbon concentrations in MOCVD-grown samples, in which both carbon and hydrogen are likely to be unintentionally incorporated, may thus reflect the presence of such complexes rather than isolated donors. Since the complexes are neutral, they would not impact the conductivity of the sample. This implicit passivation of carbon impurities, which would otherwise act as shallow donors, may well explain why MOCVD-grown GaO samples can exhibit free-carrier concentrations as low as cm [32].
We now turn to –H complexes in AlO. Figure 12(c,Figure 12d) show the formation energy of the complexes
in AlO in Al-rich conditions. The type of bonding described for GaO is also present in AlO, but now we see that for high in the gap also a negative charge state can be stabilized, which indicates that the complex could potentially act as an acceptor and cause compensation. However, we note that the transition level of the complex occurs at a higher position in the band gap than the transition of the isolated . The level of –H occurs at 0.90 eV below the CBM, and a level is also present, at 0.13 eV below the CBM. Similarly, and levels of –H are also observed close to the CBM, at 1.20 and 0.33 eV below the CBM. We note that the formation energy for –H in monoclinic AlO is similar to that in corundum AlO, except that the charge state of –H is absent in corundum [63]. The –H bonding in monoclinic AlO is also very similar to that in the corundum phase [63], implying that the C–H bonding within an octahedral O cage [as in Figure 11(d)] is a host-independent structural feature.
We verified that the –H complexes are thermodynamically stable (i.e., have positive binding energies) in both the neutral and negative charge states. The only exception is , which has a negative binding energy. Formation of this complex is highly unlikely, anyway, since it would require the Fermi level to be within 0.13 eV of the CBM.
The stability of the Ccation–H complexes in the neutral and negative charge states in both GaO and AlO implies they will also be stable in (AlGa)O alloys. The question is whether the behavior as a compensating acceptor that we find in AlO can occur in (AlGa)O alloys. To assess this, we calculated the position of the transition level in GaO and combined this information with the results for the level in AlO. This allows us to conclude that the Ccation–H complexes can only occur in a negative charge state (i.e., act as acceptors) when the Al concentration exceeds 56% for –H or 64% for –H.
Co–H complexes
We discussed carbon on the oxygen site, , in Sec. III B 5; now we turn to its complexes with hydrogen. As shown in Figure 13(a), Co–H has very low formation energy in GaO, particularly when the Fermi level is high in the gap and under Ga-rich conditions. Regardless of the O site [O(I), O(II), O(III)] on which carbon is substituting, and charge-state transition levels are present in the band gap. We verified that higher negative charge states (such as ) are not stable for Fermi levels in the band gap. In all these complexes, hydrogen is always bonded to the C atom, with a bond length between 1.07 Å and 1.09 Å. Again, we note that the C–H unit behaves very similarly to an N impurity: Co–H and substitutional N have similar and transition levels in the gap, and both behave as acceptors in -type material [64].

Figure 13. Formation energy versus Fermi level for Co impurities, H interstitials, and Co–H defect complexes in (a)-(b) -GaO and (c)-(d) -AlO. (a) and (c) are for cation-rich, and (b)-(d) for cation-poor conditions. Dashed lines denote thermodynamic instability of the defect complex, as explained in the text. The grey area indicates the conduction band of -GaO.
The behavior of Co–H in AlO is very similar: it also only exhibits (+/0) and (0/−) levels in the gap, and the complex has low formation energies under Al-rich conditions and when the Fermi level is high in the gap. The binding energies of the (Co–H)− complexes are all large in both GaO and AlO: all binding energies exceed 3 eV, with the exception of the (Co(III)–H)− complex, where the binding energy is 1.76 eV.
We propose that the Co–H complexes may have a distinct impact on MOCVD-grown GaO. Seryogin et al. [33] grew GaO using a trimethylgallium (TMGa) precursor, which is known to potentially lead to increased C incorporation. They found that films grown with a lower O/TMGa ratio were significantly more resistive and also contained significantly more carbon. These results indicate that carbon is behaving as a compensating acceptor. The authors of Ref. [33] pointed to a computational study [65] that had found C to be a center in GaO, with a (+/−) level 0.81 eV below the CBM. That result is very different from ours, probably due to the particulars of the band-gap correction used in Ref. [65]. In any case, if C is the culprit, moving from O-rich to
O-poor conditions should reduce the C incorporation, which is not consistent with the observed increase in C concentration. We think it is much more likely that C is involved, which indeed acts as a compensating acceptor with a concentration that should increase under O-poor conditions, as seen in Figure 6. However, that figure also shows that the formation energy of C in the negative charge state is still relatively high, even under the most favorable (extreme O-poor and C-rich) conditions. Figure 13 indicates that this energy can be lowered by complexing with hydrogen, which should of course also be abundantly available during MOCVD growth.
The involvement of Co–H complexes is also consistent with the experiments of Alema et al. [34], who found that adding HO vapor to the oxygen source decreased the carrier concentration, both in unintentionally doped (UID) and lightly Si-doped samples. Hydrogen could either passivate donors or form compensating acceptors. Donor passivation may occur by forming complexes with the Si donor (though in Sec. III D 1 we noted this was not very likely) or with unintentional C donors; for compensating acceptors, Co–H complexes are most likely, and indeed, Ref. [34] reported an increase in compensating acceptors rather than a decrease in donor concentration upon adding HO.
Conclusions
We have reported a comprehensive investigation of silicon donors in (AlGa)O and the various ways in which they may be compensated. For Si in -AlO we performed a detailed study of center formation. Two stable Si configurations were identified, with the configuration involving bonding to two cations and no broken bonds being most energetically favorable. Interpolating the (+/−) charge-state transition levels between -GaO and -AlO, we find that Si is an efficient donor over a wide range of Al concentrations in (AlGa)O alloys, up to 70% Al.
We also investigated the behavior of carbon and hydrogen impurities, which are commonly unintentionally present, particularly in MOCVD. Even though these act as shallow donors in -GaO, they become deep centers in (AlGa)O alloys. Based on interpolation of the (+/−) transition levels, we find that C(I) acts as a compensating acceptor already at 5% Al, and H already at 1% Al.
Additionally, complex formation between hydrogen and either Si or C may occur. We found that Si–H complexes have relatively low binding energy; if they form, they can probably quite easily be removed by annealing. C–H complexes are very stable and electrically neutral in -type GaO, and may explain why unintentionally incorporated carbon does not affect the carrier concentration.[32] C–H complexes do act as acceptors in (AlGa)O alloys, but only if the Al concentration exceeds 56%. Co–H, finally, was found to act as an acceptor over the entire range of alloy compositions, behaving very similarly to a substitutional impurity and explaining experimental observations of carbon-related compensation in grown by MOCVD [33, 34].
Our findings indicate that the presence of hydrogen or carbon may interfere with controlled low-level Si doping in alloys, and our detailed results can be used to devise growth or processing conditions to avoid this.
Acknowledgments
The authors acknowledge Hongping Zhao for fruitful discussions and preliminary experimental results. S. M. is much indebted to Andrew J. E. Rowberg for stimulating discussions and valuable comments on the manuscript. The work was supported by the GAME MURI of the Air Force Office of Scientific Research (FA9550-18-1-0479). This work was partially performed under the auspices of the U.S. DOE by Lawrence Livermore National Laboratory (LLNL) under contract DE-AC52-07NA27344 and partially supported by LLNL Laboratory Directed Research and Development funding under project number 22-SI-003 and by the Critical Materials Institute, an Energy Innovation Hub funded by the U.S. DOE, Office of Energy Efficiency and Renewable Energy, Advanced Manufacturing Office. Work at NRL was supported by the Office of Naval Research through the Naval Research Laboratory’s Basic Research Program, and made use of DoD HPCMP resources. Use was made of computational facilities purchased with funds from the National Science Foundation (NSF) (CNS-1725797) and administered by the Center for Scientific Computing (CSC). The CSC is supported by the California NanoSystems Institute and the Materials Research Science and Engineering Center (MRSEC; NSF DMR 1720256) at UC Santa Barbara. This work also used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by the National Science Foundation under Grant No. ACI-1548562, and the Frontera resources at the Texas Advanced Computing Center (TACC) at The University of Texas at Austin (NSF OAC-1818253).
References
- [1]H. Tippins, Phys. Rev. 140, A316 (1965).
- [2]T. Matsumoto, M. Aoki, A. Kinoshita, and T. Aono, Jpn. J. Appl. Phys 13, 1578 (1974).
- [3]C. Sturm, R. Schmidt-Grund, C. Kranert, J. Furthmüller, F. Bechstedt, and M. Grundmann, Phys. Rev. B 94, 035148 (2016).
- [4]A. Mock, R. Korlacki, C. Briley, V. Darakchieva, B. Monemar, Y. Kumagai, K. Goto, M. Higashiwaki, and M. Schubert, Phys. Rev. B 96, 245205 (2017).
- [5]M. Higashiwaki, K. Sasaki, A. Kuramata, T. Masui, and S. Yamakoshi, Appl. Phys. Lett. 100, 013504 (2012).
- [6]R. Suzuki, S. Nakagomi, Y. Kokubun, N. Arai, and S. Ohira, Appl. Phys. Lett. 94, 222102 (2009).
- [7]T. Oshima, T. Okuno, N. Arai, N. Suzuki, S. Ohira, and S. Fujita, Appl. Phys. Express 1, 011202 (2008).
- [8]F. Alema, B. Hertog, P. Mukhopadhyay, Y. Zhang, A. Mauze, A. Osinsky, W. Schoenfeld, J. S. Speck, and T. Vogt, APL Materials 7, 022527 (2019).
- [9]Z. Feng, A. F. M. Anhar Uddin Bhuiyan, M. R. Karim, and H. Zhao, Appl. Phys. Lett. 114, 250601 (2019).
- [10]Y. Zhang, F. Alema, A. Mauze, O. S. Koksaldi, R. Miller, A. Osinsky, and J. S. Speck, APL Materials 7, 22506 (2019).
- [11]N. Son, K. Goto, K. Nomura, Q. Thieu, R. Togashi, H. Murakami, Y. Kumagai, A. Kuramata, M. Higashiwaki, A. Koikitu, S. Yamakoshi, B. Monemar, and E. Janzen, J. Appl. Phys. 120, 235703 (2016).
- [12]A. Parisini and R. Fornari, Semicond. Sci. Technol. 31, 035023 (2016).
- [13]N. Ma, N. Tanen, A. Verma, Z. Guo, T. Luo, H. Xing, and D. Jena, Appl. Phys. Lett. 109, 212101 (2016).
- [14]M. Higashiwaki, A. Kuramata, H. Murakami, and Y. Kumagai, J. Phys. D 50, 333002 (2017).
- [15]N. Moser, J. McCandless, A. Crespo, K. Leedy, A. Green, A. Neal, S. Mou, E. Ahmadi, J. Speck, K. Chabak, N. Peixoto, and G. Jesse, IEEE Electron Device Lett. 38, 775 (2017).
- [16]T. A. Neal, S. Mou, S. Rafique, H. Zhao, E. Ahmadi, J. S. Speck, K. T. Stevens, J. D. Blevins, D. B. Thomson, N. Moser, and Others, Appl. Phys. Lett. 113, 62101 (2018).
- [17]M. Orita, H. Ohta, M. Hirano, and H. Hosono, Appl. Phys. Lett. 77, 4166 (2000).
- [18]T. Oishi, K. Harada, Y. Koga, and M. Kasu, Jpn. J. Appl. Phys. 55, 30305 (2016).
- [19]M. Higashiwaki, K. Sasaki, T. Kamimura, M. Hoi Wong, D. Krishnamurthy, A. Kuramata, T. Masui, and S. Yamakoshi, Appl. Phys. Lett. 103, 123511 (2013).
- [20]J. B. Varley, J. R. Weber, A. Janotti, and C. G. Van de Walle, Appl. Phys. Lett. 97, 142106 (2010).
- [21]J. B. Varley, A. Perron, V. Lordi, D. Wickramaratne, and J. L. Lyons, Appl. Phys. Lett. 116, 172104 (2020).
- [22]J. L. Lyons, D. Steiauf, A. Janotti, and C. G. Van de Walle, Phys. Rev. Appl. 2, 064005 (2014).
- [23]J. B. Varley, H. Peelaers, A. Janotti, and C. G. Van de Walle, J. Phys. Conden. Matter 23, 334212 (2011).
- [24]M. E. Engbertsen, A. Y. Kuznetsov, B. G. Svensson, G. Alfieri, A. Mihaila, U. Badstübner, A. Perron, L. Vines, and J. B. Varley, APL Mater. 7, 022510 (2019).
- [25]H. von Bardeleben and J. Cantin, J. Appl. Phys. 128, 125702 (2020).
- [26]D. Chadi and K.-J. Chang, Phys. Rev. Lett. 61, 873 (1988).
- [27]L. Gordon, J. L. Lyons, A. Janotti, and C. G. Van de Walle, Phys. Rev. B 89, 085204 (2014).
- [28]H. Peelaers, J. B. Varley, J. S. Speck, and C. G. Van de Walle, Appl. Phys. Lett. 112, 242101 (2018); Appl. Phys. Lett. 115, 159901 (2019). (erratum)
- [29]J. B. Varley, J. Mater. Res. 1 (2021).
- [30]Y. Zhang, C. Joishi, Z. Xia, M. Brenner, S. Lodha, and S. Rajan, Appl. Phys. Lett. 112, 233503 (2018).
- [31]H. P. Zhao, private communication.
- [32]F. Alema, Y. Zhang, A. Osinsky, N. Orischin, N. Valente, A. Mauze, and J. S. Speck, APL Materials 8, 021110 (2020).
- [33]G. Seryogin, F. Alema, N. Valente, H. Fu, E. Steinbrunner, A. T. Neal, S. Mou, A. Fine, and A. Osinsky, Appl. Phys. Lett. 117, 262101 (2020).
- [34]F. Alema, Y. Zhang, A. Mauze, T. Itoh, J. S. Speck, B. Hertog, and A. Osinsky, AIP Advances 10, 085002 (2020).
- [35]P. E. Blöchl, Phys. Rev. B 50, 17953 (1994).
- [36]G. Kresse and J. Hafner, Phys. Rev. B 48, 13115 (1993).
- [37]G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996).
- [38]K. Momma and F. Izumi, J. Appl. Crystallogr. 44, 1272 (2011).
- [39]J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003); J. Heyd and G. E. Scuseria, J. Chem. Phys. 124, 219906 (2006).
- [40]R. Franchy, G. Schmitz, P. Gassmann, and F. Bartolucci, Appl. Phys. A 65, 551 (1997).
- [44]J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996).
- [45]G. Henkelman, B. P. Uberuaga, and H. Jónsson, J. Chem. Phys. 113, 9901 (2000).
- [46]G. H. Vineyard, J. Phys. Chem. Solids 3, 121 (1957).
- [47]P. Weiser, M. Stavola, W. B. Fowler, Y. Qin, and S. Pearton, Appl. Phys. Lett. 112, 232104 (2018).
- [48]A. Janotti and C. G. Van de Walle, Phys. Rev. B 76, 165202 (2007).
- [49]C. Freysoldt, J. Neugebauer, and C. G. Van de Walle, Phys. Rev. Lett. 102, 016402 (2009).
- [50]C. Freysoldt, J. Neugebauer, and C. G. Van de Walle, Phys. Status Solidi B 248, 1067 (2011).
- [51]J. L. Lyons, A. Janotti, and C. G. Van de Walle, Phys. Rev. B 89, 035204 (2014).
- [52]J. L. Lyons and C. G. Van de Walle, npj Comput. Mater. 3, 1 (2017).
- [53]M. McCluskey, N. Johnson, C. G. Van de Walle, D. Bour, M. Kneissl, and W. Walukiewicz, Phys. Rev. Lett. 80, 4008 (1998).
- [54]C. Skierbiszewski, T. Suski, M. Leszczynski, M. Shin, M. Skowronski, M. Bremser, and R. Davis, Appl. Phys. lett. 74, 3833 (1999).
- [55]A. Bouzid and A. Pasquarello, Phys. Status Solidi RRL 13, 1800633 (2019).
- [56]M. Choi, J. L. Lyons, A. Janotti, and C. G. Van de Walle, Appl. Phys. Lett. 102, 142902 (2013).
- [58]S. Mu, H. Peelaers, Y. Zhang, M. Wang, and C. G. Van de Walle, Appl. Phys. Lett. 117, 252104 (2020).
- [60]A. Venzie, A. Portoff, C. Fares, M. Stavola, W. B. Fowler, F. Ren, and S. J. Pearton, Appl. Phys. Lett. 119, 062109 (2021).
- [61]C. G. Van de Walle and J. Neugebauer, Annu. Rev. Mater. Res. 36, 179 (2006).
- [62]See Supplemental Material at [URL will be inserted by publisher] for the formation energy of SiO and Si–H complexes in Ga₂O₃ and Al₂O₃.
- [63]M. Choi, A. Janotti, and C. G. Van de Walle, ACS Appl. Mater. Interfaces 6, 4149 (2014).
- [64]H. Peelaers, J. L. Lyons, J. B. Varley, and C. G. Van de Walle, APL Mater. 7, 022519 (2019).
- [65]S. Lany, APL Mater. 6, 046103 (2018).



