Spectroscopic investigations of band offsets of MgO|Al_xGa_1-xN epitaxial heterostructures with varying AlN content

The potential of AlGaN/GaN heterostructure power devices is well known for high-power, high-frequency, and high-temperature operation because of the device channel's high carrier density and the high mobility of the 2D carrier gas. However, the inherent polarization of the interface also makes normally off operation difficult to achieve. One route for enabling normally off operation while also reducing gate leakage^1,2 is gate oxide integration, i.e., metal oxide semiconductor high electron mobility transistors (MOSHEMTs). GaN-based MOSHEMTs (and, in particular, AlGaN-based MOSHEMTs) present many challenges to oxide integration due to their large band gap (ranging from 3.4 eV for GaN to 6.2 eV for AlN) and hexagonal basal plane symmetry. The ideal oxide for MOSHEMTs would (1) provide a large dielectric constant, (2) be chemically compatible with both GaN and AlN, (3) have a suitable growth morphology on a hexagonal template, and (4) provide a large enough band offset (typically ≥ 1 eV) to both the conduction band and valence band to mitigate leakage current across the oxide|nitride interface. Satisfying all of these requirements simultaneously is difficult, and many oxides investigated thus far fail to achieve at least one of these goals. For example, in the case of two commonly investigated gate dielectrics for GaN, SiO₂ has a relative permittivity of 3.9, while SiN_x fails to provide sufficient valence band offsets³ and is therefore unsuitable for p-type devices. Another popular choice, high κ oxides, such as Sc₂O₃,⁴ Gd₂O₃,⁵ and La₂O₃,⁶ have unsuitable band offsets for both the conduction band and valence band.

One possibility for satisfying all requirements for the oxide in AlGaN electronics is to use rocksalt oxides such as MgO,⁷ CaO,⁸ or solid solutions of the two (Mg_xCa_1−xO, or generically, MCO),^9,10 the latter of which can be lattice-matched directly to GaN and Al_xGa_1−xN compositions. Compared to other dielectrics investigated thus far for AlGaN, MgO has a high dielectric constant of 9.8, a large band gap (7.8 eV), is chemically compatible with GaN up to temperatures of 900 °C,⁷ and when grown in certain oxidant environments, can match the GaN morphology, providing smooth oxide|nitride interfaces.¹¹ Though the band offsets of MgO to GaN have been measured,^12,13 the band offsets of MgO to AlGaN with increasing AlN content are still unknown. For MOSHEMT device applications, it is essential to know the band offsets of MgO on Al_xGa_1−xN to ensure that sufficient offsets exist as the nitride band gap increases with increasing AlN content. Previous work by Craft et al. has shown that the valence band offset for MgO on GaN¹² and CaO on GaN¹⁴ are comparable. Thus, studying the band offsets of MgO on Al_xGa_1−xN should also provide guidance for CaO as well as MCO, all of which are excellent candidates for AlGaN/GaN passivation and gate oxides.

GaN (0001) and AlGaN (0001) films were deposited on (0001) sapphire substrates mis-cut 0.2° toward the m-plane by metal-organic vapor phase epitaxy in a Veeco D-125 short-jar reactor using trimethylgallium, thrimethylaluminum, and ammonia precursors, and H₂ and N₂ carrier gases.¹⁵ The unintentionally doped GaN film was deposited at 1050 °C to 2.5 μm thick on top of a 30 nm GaN nucleation layer grown at 900 °C. AlGaN films were grown at 1050 °C on an AlN buffer layer to thicknesses of 1.5 μm and 1.3 μm for Al_0.28Ga_0.72N and Al_0.67Ga_0.33N films, respectively. AlGaN films were silicon doped to an n-type carrier concentration of 1 × 10¹⁸cm⁻³ and 4 × 10¹⁸cm⁻³ for Al_0.28Ga_0.72N and Al_0.67Ga_0.33N films, respectively.

MgO films were grown on GaN and AlGaN epilayers by molecular-beam epitaxy (MBE) in a custom-built ultra-high vacuum (UHV, base pressure < 6 × 10⁻¹⁰Torr) oxide MBE chamber. Prior to growth, GaN and AlGaN surfaces were cleaned following the sequential order: a 5 min acetone ultrasonic bath, a 5 min methanol ultrasonic bath, and finally a 95:5 H₂O:HF etch followed by a deionized water rinse. After ex situ cleaning, GaN and AlGaN films were mounted to molybdenum platens using silver paint and were heated to 500 °C—monitored using an optical pyrometer—for 30 min under vacuum. Magnesium metal (Alpha Aesar, 99.95% purity) was evaporated from an e-Science effusion cell with a boron nitride crucible. Prior to growth, substrates were cooled to 300 °C. Sample growth was monitored in-situ using reflection high-energy electron diffraction (RHEED) with a STAIB Instruments electron source operated at 14.9 kV and 1.5 A. Molecular oxygen was introduced via a directed gas nozzle modulated with a leak valve to O₂ background pressure of 1 × 10⁻⁶Torr. MgO deposition rates were maintained at 4 Å/min.

Band offsets were measured by X-ray photoelectron spectroscopy (XPS), using the procedure described by Waldrop and Grant.^16–18 A Kratos Axis Ultra DLD instrument was used with monochromatic Al Kα radiation (1486.7 eV). Analyzer pass energies of 80 eV and 20 eV were used for survey and high-resolution spectra, respectively. High-resolution spectra were taken with 25–50 meV step sizes and 100 ms dwell times over 25–40 cycles, depending on the signal of the core lines. Valence band spectra were recorded at 10–30 meV, 100 ms, and up to 50 cycles. Base pressure was maintained at <5 × 10⁻⁹Torr. The hemispherical analyzer was used in Hybrid mode with an elliptical spot size of 300 × 700 microns. All XPS spectra were analyzed using CasaXPS (Casa Software Ltd.). Peaks were fit using a linear background and the maxima of the dominant fitted component were used as the peak position. Charge neutralization was not required; however, the samples were exposed to electron irradiation from a neutralization filament during analysis to limit differential charging artifacts. The dominant C 1s peaks were aligned to 284.6 eV for each set of spectra. Valence band maxima were determined from a complementary error function fit.¹⁹ To obtain band offsets for MgO on GaN, Al_0.28Ga_0.78N, and Al_0.67Ga_0.33N, three samples were measured for each substrate composition: a bare semiconductor substrate, a 5 nm MgO film, and a 50 nm MgO film. A 5 nm MgO film is much thinner than the coalescence point for MgO on GaN (∼12 nm)¹² and allows for emission from both the semiconductor substrate as well as the MgO film to allow core levels of both to be measured in one sample. A 50 nm MgO film is used to ensure that the signal from the semiconductor is attenuated so that the valence band maximum for “bulk” MgO can be determined.

Figure 1 shows RHEED images for each bare substrate and terminal images for 50 nm MgO films at 300 °C. The transmission diffraction terminal patterns demonstrate a three-dimensional growth mode for MgO on all Al_xGa_1−xN compositions. This islanding growth mode is consistent with several reports of MgO on GaN^7,11 and is expected for 111-oriented MgO on 0001 GaN. At 50 nm, MgO should be fully relaxed on both AlGaN and GaN. This is confirmed by the decreasing d-spacing along ⟨$112¯0$⟩ from GaN to Al_0.67Ga_0.33N while the d-spacing measured for 50 nm of MgO on each substrate remains constant.

Representative X-ray photoelectron spectra for MgO on 67% AlGaN are shown in Figure 2. Valence band offsets were determined using the relative positions of metal core level (CL) peaks: Ga 3s – Mg 2p, Ga 3p – Mg 2p, and Al 2p – Mg 2p and the valence band maximum (VBM) positions of MgO and Al_xGa_1−xN. Offsets were calculated over the three measurements for each substrate according to the following formula:¹⁶ 

where CL is the core level chosen for the semiconductor substrate (Ga 3s, Ga 3p, or Al 2p) and SC_VBM is the valence band maximum of the semiconductor. The same procedure was repeated for MgO on 28% AlGaN and MgO on GaN (excluding the Al 2p core-level calculation). Example spectra from each substrate are shown in the supplemental material.²⁰ Figure 3 shows valence band offsets and subsequently calculated conduction band offsets for MgO on each substrate. Valence band offsets of 1.65 ± 0.07 eV, 1.36 ± 0.05 eV, and 1.05 ± 0.09 eV were found for MgO on GaN, Al_0.28Ga_0.72N, and Al_0.67Ga_0.33N, respectively. These results follow a similar trend as lanthanide oxides^5,6 as well as TiO₂¹ on Al_xGa_1−xN, which also show a decreasing valence band offset with increasing AlN content. This trend is expected as the band gap of Al_xGa_1−xN increases from x = 0 to x = 1 from 3.4 eV to 6.2 eV. However, these values are much larger than the valence band offsets obtained for La₂O₃, Gd₂O₃, SiN_x, or TiO₂, all of which show a small to negligible offset to GaN and AlGaN.^1,3,5,6

Conduction band offsets are calculated assuming a band gap of 3.4 eV for GaN, 4.05 for 28% AlGaN, and 5.12 for 67% AlGaN for a bowing parameter, b, of 0.69 eV.²¹ For an MgO band gap of 7.8 eV, conduction band offsets are calculated as 2.75 eV, 2.39 eV, and 1.63 eV for MgO on GaN, Al_0.28Ga_0.72N, and Al_0.67Ga_0.33N, respectively. Results for valence band offsets and conduction band offsets of MgO to each substrate are provided in Table I. Therefore, for all AlGaN compositions investigated, both the conduction band offset and the valence band offset to MgO are >1 eV. This implies that MgO should be attractive for both n- and p-type Al_xGa_1−xN devices up to the 67% AlN composition investigated while also offering potential advantages in growth mode (in the correct oxidant environment), dielectric constant, and chemical compatibility. Compositions beyond 67% AlN were not investigated, but it is reasonable to assume that the valence band offset trend will continue to decrease below 1.05 eV, and this should be taken into account when MgO is considered for higher aluminum content AlGaN devices. To this point, valence band offsets for MgO on AlN have been calculated from charge neutrality levels to be 0.2 eV and measured by Yang et al. as 0.22 eV.²² Both the measured and calculated value for MgO|AlN suggests that higher aluminum content substrates may make it difficult to achieve a ≥1 eV valence band offset, however, the conduction band offset should still be sufficient.

We note, as discussed above, that the valence band offset of MgO on GaN has been investigated previously by Craft et al.¹² and Chen et al.¹³ who found values of 1.2 ± 0.2 eV and 1.06 ± 0.15 eV, respectively. These values represent a discrepancy of ∼0.45 eV ± 0.2 eV from the measurement in this study. However, it is interesting to note that the valence band offset measured in this work is closer to Robertson's theoretical prediction of 1.59 eV,^23,24 based on charge neutrality level (CNL) calculations of GaN and MgO using a Schottky pinning factor of 0.71. We speculate that the differences in measured band offsets between these reports (this work and References 12 and 13) of MgO|GaN band offsets could stem from the use of different GaN substrates in each case. Thus, it is likely that the Ga:N surface stoichiometry may be different for each GaN substrate, which could change the CNL barrier heights of GaN and, as a consequence, change the valence band offset to the oxide. For example, Reddy et al.²⁵ measured a CNL value for Ga-polar n-type GaN to be 2.7 eV; almost 0.4 eV higher in the gap than the 2.32 eV CNL calculated by Schliefe et al.^26,27 This difference in CNL was attributed to differences in surface stoichiometry between experiment and theory, though the influence of stoichiometry on CNL was not directly measured.²⁵ This difference in CNL between theoretical values and those measured by Reddy et al. is of a similar magnitude as the two valence band offsets measured in this study and in previous works. The surface Ga:N ratio of GaN grown by MOCVD on sapphire can vary significantly, e.g., ranging from 1.3:1 to 3.2:1 in one study, depending on the dwell time of the GaN film in the reactor (growth time and cooling time).²⁸ We speculate that a difference in Ga:N surface stoichiometry of a similar magnitude can lead to similar differences in valence band offsets. Additionally, it has been shown that the valence band maximum of GaN changes by 0.35 eV for different surface preparations and relative oxygen fraction.^3,29 Therefore, we believe it is reasonable to assume that these surface characteristics of GaN—surface stoichiometry and oxygen content—can influence the band offset across the oxide|nitride interface, and without stoichiometry or O 1s spectral data for each substrate from previous experiments, the cause for the range in reported MgO|GaN band offsets cannot be determined after the fact. Independent of these differences, however, in all cases of measured band offsets between MgO|GaN, and in this particular study, MgO|Al_xGa_1−xN up to 67% AlN, the valence band and conduction band offsets observed should be suitable for MOSHEMT device optimization.

In summary, epitaxial 111-oriented MgO films were prepared on 0001-Al_xGa_1−xN on sapphire substrates via molecular-beam epitaxy. Valence band offsets of MgO to AlGaN with aluminum compositions ranging from 0% to 67% were reported. Valence band offsets of 1.65 ± 0.07 eV, 1.36 ± 0.05 eV, and 1.05 ± 0.09 eV were found for MgO on GaN, Al_0.28Ga_0.72N, and Al_0.67Ga_0.33N, respectively. This yields conduction band offsets of 2.75 eV, 2.41 eV, and 1.64 eV for MgO on GaN, Al_0.28Ga_0.72N, and Al_0.67Ga_0.33N, respectively. All band offsets measured are >1 eV, offering a promising option for gate dielectrics and passivation layers for both GaN and AlGaN-based devices.

Band offset measurements of MgO on AlGaN were supported by the Laboratory Directed Research and Development (LDRD) program at Sandia. Band offset measurements of MgO on GaN were supported by the U.S. Department of Energy's Office of Electricity Delivery and Energy Reliability (OE) Energy Storage Program managed by Dr. Imre Gyuk. The authors acknowledge Rudeger H. T. Wilke for critical review of this manuscript. Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-AC04-94Al85000.