Nitride surface chemistry influence on band offsets at epitaxial oxide/GaN interfaces

GaN-based high electron mobility transistor (HEMT) devices are promising candidates for next-generation power electronics owing to their wide bandgaps and high channel mobilities. However, achieving reliable and consistent normally-off performance in GaN HEMTs still eludes the community. One attractive approach to achieving enhancement mode (normally-off) operation is the incorporation of a gate oxide into the HEMT stack, creating a metal oxide semiconductor-HEMT, or MOSHEMT, structure.^1,2 In this embodiment, the gate oxide can enable normally-off operation while simultaneously aiding in the suppression of gate leakage. The challenge for GaN MOS-structures, however, is twofold: first, the oxide must have a sufficiently large bandgap to offset that of GaN (3.4 eV), and second, these offsets should be consistent to aid in device reliability. Generally, many oxides have been shown to satisfy the first requirement; oxides from Al₂O₃,³ rare earth oxides,^4–7 to rocksalt oxides,^8–12 etc., all demonstrate the necessary >1 eV conduction band offset. However, where these oxide/nitride interfaces sometimes fail is in the consistency of the band offset. Literature reports for epitaxial oxides on GaN differ by ± 0.56 eV for Gd₂O₃,^4,6,7 ± 0.4 eV for Sc₂O₃,^13–15 ± 1.27 eV for La₂O₃,^5,14 and ± 0.6 eV for MgO^8,10,12 These variations are generally outside the error associated with the measurement itself and highlight the lack of consistency for band offsets measured for oxides on GaN.

While there are many possibilities for these differences, one candidate is the GaN epilayer itself. To date, there is not a consensus on the best method to prepare GaN surfaces for oxide growth. Furthermore, depending on the growth procedure for the GaN films, the surface chemistry can vary widely in stoichiometry and in oxygen and hydroxide coverage.¹⁶ These inconsistencies can ultimately affect the surface electronic charge near the oxide/GaN interface, resulting in deviations in interface band bending.^17,18 Consequently, we predict that the GaN surface chemistry is also a source of valence and conduction band offset discrepancies. To highlight the role of the substrate in widespread literature band offset inconsistencies, epitaxial MgO was grown on five GaN epilayers from different sources. MgO growth and GaN cleaning procedures were identical in all cases. MgO was selected as a case study because it provides a chemically stable interface with GaN,^19,20 and the bandgap (7.8 eV) makes it an attractive gate oxide choice for both GaN and AlGaN devices. However, the general trends observed should hold for any chemically abrupt oxide/nitride interface.

Five Ga-polar (0001)-oriented GaN epilayers on sapphire were obtained from four separate suppliers: three were grown in-house (N1, N2, and S), and the remaining two epilayers were purchased from Saint Gobain Lumilog (L) and MTI Corporation (M), respectively. GaN films grown at Sandia (S) and North Carolina State University (N) were unintentionally doped (UID) and deposited by MOCVD on (0001)-oriented sapphire substrates miscut by 0.2° toward the m-plane. NCSU films, N1 and N2, were 2.8 to 4.6 μm thick, respectively; details of the growth can be found elsewhere.^21,22 The Sandia-grown GaN epilayer was 2.5 μm thick.²³ To highlight the changes in GaN epilayers, this film labeled S was purposely grown with high threading dislocation density, as will be discussed later. Commercially obtained n-type (silicon doped) Lumilog GaN was 3.5 μm thick with a carrier concentration of 1.9 × 10¹⁸ cm⁻³, and UID MTI GaN was 5 μm thick. Growth details of epilayers L and M were not available.

Prior to MgO growth, all GaN epilayers were chemically treated following the sequential order: 5 min acetone ultrasonic bath, 5 min methanol ultrasonic bath, 1 min 5:95 HF:deionized (DI) H₂O, DI H₂O rinse, and a N₂ blow dry.²⁴ Following ex situ chemical cleaning, GaN was mounted to a molybdenum platen using silver paint and heated to 500 °C in vacuum (< 1 × 10⁻⁹Torr). The substrate was then cooled to the MgO growth temperature (300 °C), and MgO films were grown by reactive molecular-beam epitaxy, as described elsewhere.¹² 

Valence band offsets and surface chemistries were measured by X-ray photoelectron spectroscopy (XPS) using the procedure described by Waldrop and Grant.^25–27 Valence band maxima (VBM) were fit with a complementary error function using CasaXPS (Casa Software Ltd.).²⁸ GaN dislocation densities were measured using high resolution X-ray diffraction with a PANalytical Empyrean diffractometer with Cu Kα radiation.²⁹ A double-bounce germanium hybrid monochromator prefix module with a 1/32° divergence slit was used for incident beam optics. A 0.18° parallel plate collimator with a proportional detector was used on the diffraction side.

Surface stoichiometries (Ga:N) of the five GaN epilayers (N1, N2, S, L, and M) were measured by XPS, and data are presented in Table I. Surface Ga:N ratios greater than unity are common for MOCVD-grown GaN,¹⁶ and Ga:N ratios of 1.3:1.0¹⁸ to 4.0:1.0³⁰ have been reported. Surface Ga:N stoichiometries in this experiment ranged from 1.33:1 to 1.46:1. We emphasize that the epilayers are from different vendors and were grown under different conditions. Thus, it is likely that conditions such as growth time,¹⁶ temperature, and gallium super-saturation affect the final Ga:N stoichiometry and account for the differences observed across the five epilayers. However, because growth details are not available for all epilayers, we cannot pinpoint the exact cause of stoichiometry differences.

The atomic concentration of surface oxygen to gallium, nitrogen, carbon, and hydroxide for each GaN epilayer was also measured and is shown in Table I and Fig. 1(a). There does not appear to be a direct correlation of excess gallium with the oxygen content. A gallium-rich surface may allow for additional absorption of surface oxygen through a Ga_xO_y-type native oxide or hydroxide coverage.^18,31,32 However, other studies have shown that oxygen impurities formed during GaN growth may preferentially segregate to threading dislocations on the surface and in the bulk.^33,34 As discussed later, the epilayers used in this work possess threading dislocation densities spanning more than one order of magnitude: 9.3 × 10⁷ cm⁻² to 1.1 × 10⁹ cm⁻². However, as will be discussed, no general trend was observed when comparing threading dislocation density or excess gallium with the measured band offsets to MgO. Thus, changes in dislocation density and excess gallium from epilayer-to-epilayer may complicate any general trends of oxygen and excess gallium across the epilayers used.

To understand the effect of the surface chemistry on the GaN surface electronic charge, the band bending of each bare GaN surface was measured. All GaN spectra were corrected for charging by aligning to the C 1s peak position. The position of the core level (CL) peaks, for example, Ga 3d, relative to the GaN valence band maximum (VBM) is a constant (17.24 ± 0.03 eV)^25–27 of GaN and did not change with surface stoichiometry or the impurity concentration across three measurement points on all five GaN epilayers.⁴ Therefore, because this value is constant across all GaN used, the band bending at the charge neutrality level of GaN can then be given by the change in the CL position of the Ga 3d peak, where a decrease in the Ga 3d CL position indicates upward band bending. Figure 1(b) shows the measured Ga 3d CL positions for each GaN epilayer; error bars associated with three measurement locations performed for each epilayer are provided. As shown, as the GaN surface atomic oxygen concentration increases, the Ga 3d binding energy decreases, indicating upward band bending at the GaN surface, depicted in Fig. 1(c). This is expected for Ga-polar GaN, where the polarization charge points along the c-axis, creating compensating bound positive surface charges. Excess negative charge is created through surface chemistry variations to compensate this bound charge, which may arise through Ga_xO_y-type coverage and then create upward band bending, as measured.

These results are consistent with previous research, where a decreasing GaN VBM with increasing surface oxygen has been observed previously by Garcia et al.³¹ Assuming that the gallium CL position and the VBM were constant in the previous work (although this is not explicitly stated), this would indicate that the GaN band bending was also found to increase with surface oxygen. Additional evidence for surface chemistry affecting the band structure was observed by Wu et al.,³⁵ where a GaN VBM change of 0.5 eV was found for a “clean” GaN surface versus a GaN surface measured “as-loaded.” It is likely that the “as-loaded” surface contained contamination, nonstoichiometry, and/or surface states that may have led to these differences.³⁵ No surface chemistry information was reported in the Wu study, but these changes in VBM are of the same magnitude as the changes found in the present study.

Since surface chemistry affects GaN surface band bending, it is reasonable then to assume that these changes in surface chemistry may also affect the valence band offset (VBO) to an oxide overlayer, such as our case study of MgO, and account for at least part of the variability found for band offsets of oxides on GaN in the literature. To test this hypothesis, band offsets for MgO on each GaN epilayer were measured for at least three locations on each sample. Three samples were prepared and measured for each GaN epilayer: bare GaN, a 5 nm MgO/GaN film (below the 12 nm coalescence point for MgO on GaN),¹⁰ and a 50 nm MgO/GaN film (thick enough to attenuate photoelectrons from GaN). Using the relation in Eq. (1), offsets were calculated:

where CL is the core level used for GaN (Ga 3s, Ga 3d, and Ga 3p were all used for statistics) and VBM is measured for “bulk” MgO or GaN. Both Mg 2p and Mg 2s were also used. Example fits for the VBMs and Ga 3d and Mg 2p CLs for the N1 substrate are shown in Figs. 1(a)–1(d) (supplementary material).

Table I and Fig. 2(a) show the effect of the atomic concentration of surface oxygen on the VBO of MgO to GaN. It is evident that the interface chemistry changes given in Table I have an influence on the resultant VBO to MgO; the VBO decreases from 1.68 ± 0.02 eV to 1.29 ± 0.04 eV for an initial GaN surface oxygen concentration increase from 0.9% to 3.4%, respectively. We emphasize that all GaN epilayers used were cleaned in the same manner, all MgO films were grown identically, and surface chemistry and VBOs were measured in the same manner for at least three locations per sample. However, despite the experimental consistencies, valence band and conduction band offsets between MgO and GaN vary by ∼0.40 eV (one order of magnitude larger than experimental error bars). Thus, changes in GaN surface chemistry (∼2.5 atomic concentration of oxygen) appear to affect both the GaN surface band bending and the VBO to MgO.

Once the interface is formed, the data presented in Fig. 2(a) suggest that the role of surface oxygen of GaN is altered. Before MgO growth, an increase in the surface oxygen concentration resulted in an increase in upward band bending of bare GaN. After MgO growth, however, a decrease in the VBO is observed with the oxygen content, suggesting a downward band bending in GaN. We note that a VBO decrease could also be caused by an upward bending of MgO, although the corresponding CL shifts between the 5 nm MgO and 50 nm MgO samples were not observed. Thus, we propose that the downward band bending that occurs once the interface is formed is a direct consequence of the chemically abrupt interface of MgO on GaN.¹⁹ As MgO grows on wurtzite (0001)-oriented GaN along the [111]_MgO direction, it comprises sequential alternating layers of O^2– and Mg²⁺, and we have shown previously through reflection high-energy electron diffraction studies that the first MgO monolayer of MgO growth on GaN scavenges the GaN surface oxide.²⁰ Therefore, it is likely that once the first monolayer of MgO forms, the terminating oxygen layer of the GaN native surface is consumed by the first monolayer of MgO and therefore becomes inconsequential to the resultant charges associated with the interface. In the current XPS measurement, the depth of the measurement is estimated to be ∼3–5 nm. Therefore, the oxygen signal detected by XPS before growth consists of both the oxygen on the GaN surface as well as the oxygen soluble within the first 3–5 nm of GaN. Such subsurface oxygen has been shown in calculations to substitute for nitrogen in the lattice and is a proposed source of unintentional n-type conductivity of GaN.^33,36–38 Consequently, once the GaN surface oxygen is consumed by the MgO monolayer, only the charges compensating oxygen impurities in the top few nanometers of the GaN are then left at the interface. These associated negative charges inside GaN should result in downward band bending of GaN and a concomitant decrease in the valence band offset to MgO, as has been measured. Additionally, this hypothesis is further supported by theoretical calculations of MgO on GaN band offsets, which align the branch-point energies using an electric-dipole term reported by Mönch which show VBOs closest to the values obtained for GaN with the lowest oxygen concentration.³⁹ 

In an attempt to measure the oxygen concentration just beneath the surface (first 3–5 nm), time-of-flight secondary ion mass spectrometry (TOF-SIMS) and Rutherford backscatter spectroscopy (RBS) measurements were also taken on the GaN epilayers. Data from each technique are included in Figs. 2(a) and 2(b) (supplementary material). TOF-SIMS measurements were made using a TOF-SIMS.5 instrument manufactured by IONTOF GmbH of Münster, Germany. Depth profiles were acquired by collecting negative secondary ions from a 100 × 100μm² area with a 25 kV, Bi₁⁺ probe at 2.9 pA in a HC-BUNCHED mode interleaved with a 40 nA, 500 eV Cs⁺ beam sputtering a crater 300 × 300μm² in area. For depth profile quantification, a standard was supplied by North Carolina State University. This standard of GaN was implanted with 1 × 10¹⁵ at./cm² of hydrogen, 1 × 10¹⁵ at./cm² of carbon, 3 × 10¹⁵ at./cm² of oxygen, and 1 × 10¹⁵ at./cm² of silicon. Detection limits of 4 × 10¹⁸ at./cm³ for oxygen, 3 × 10¹⁸ at./cm³ for hydrogen, 7 × 10¹⁷ at./cm³ for carbon, and 2 × 10¹⁷ at./cm³ for silicon were determined. In TOF-SIMS, the first 3–5 nm is typically unreliable due to convolution with the surface content. However, we note that the SIMS measurements do show about half an order of magnitude higher oxygen concentration at the near surface (1.5 nm deep) for the lower VBO MgO:GaN interface (“M”) than for the highest VBO (N1). RBS measurements of the concentration of oxygen near the surface of GaN were also measured using 3.04 MeV alpha particles with the detector at an angle of 165°. However, for RBS, a sharp resonance in the scattering cross section occurs at the energy associated with oxygen, which significantly decreases the sensitivity of the measurement. Very little oxygen was observed, but an upper limit of the oxygen concentration was determined to be 0.7 at. %. Thus, neither technique was able to discern the near-surface concentration information. It is likely that angle-resolved XPS can provide information on the depth of oxygen impurities; however, for this experiment, these measurements were not available. However, in future work, this could be one avenue to identify the depth resolution of oxygen in various GaN epilayers.

For completeness, the influence of hydroxide and the carbon content of the surface were also investigated along with oxygen, but no general trends with VBO were observed. Additionally, dislocation densities of each GaN epilayer were measured using HR-XRD²⁹ to determine if other GaN properties outside of surface chemistry may also affect the resulting band offset to MgO. Dislocation densities range from 9.3 × 10⁷ cm⁻² to 1.1 × 10⁹ cm⁻², but as shown in Fig. 2(b), there does not appear to be a direct correlation with VBO. It is likely that dislocation densities in GaN play a role in setting the VBO to the oxide overlayer; however, in this study, large changes in the oxygen concentration may overwhelm the effect of the dislocation density alone.

Valence band offsets of MgO on GaN have been measured previously by Craft et al.,¹⁰ Chen et al.,⁸ and Paisley et al.¹² who found values of 1.2 ± 0.2 eV, 1.06 ± 0.15 eV, and 1.65 ± 0.07 eV, respectively, a range spanning 0.59 eV. We note that the values previously measured by Paisley et al.¹² were from samples grown on the same substrate as those of substrate “S” for this work. The films for this previous work were grown one year prior, and the measured values are identical, within error, to those reported in the present study (1.65 +/– 0.07 eV). Therefore, the differences between the values measured on different substrates are not likely to be due to any slight variations that may occur from run-to-run in the MBE. Instead, we speculate that this wide range of values stems from the range of starting GaN epilayer chemistries on which MgO was grown. This range of VBOs is significant considering that most oxides studied for GaN have reported VBOs or CBOs within the range of 1.0 − 2.5 eV of GaN. If we consider that these values may contain an uncertainty of at least  ±0.6 eV depending on the surface stoichiometry of the GaN used, then this places many oxides used in the literature for GaN close to the 1 eV desired limit. Additionally, these differences have implications for device reliability as well. The threshold voltage of the MOSHEMT device is related to the conduction band offset of the oxide/nitride interface, and thus, any instability in the conduction band offset would yield an instability in the threshold voltage.

In conclusion, these results suggest that for band offset information to be consistent, GaN growth and point defect concentrations must be carefully controlled and reported. Large changes in MgO/GaN VBOs (∼0.4 eV) are measured for different GaN starting epilayers despite identical cleaning and oxide growth conditions. A clear trend in GaN band bending and VBO with the oxygen impurity concentration was observed. This study indicates that care should be taken when extracting interfacial electronic information from the literature for oxide/GaN interfaces. It is likely that the same tabulated data approach that is commonly used for band offsets of oxides on silicon, GaAs, and SiC may not be a viable option for GaN.

See supplementary material for XPS valence band offset fits for MgO films grown on the GaN epilayer, N1, and data from TOF-SIMS and RBS measurements.

Band offset measurements and surface chemistry 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. NCSU GaN film growth was supported by NSF program DMR-1508191. Dislocation density measurements were supported by ARO program W911NF-14-0285. The authors sincerely thank and acknowledge Dr. Anthony Rice for his critical review of this manuscript. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy's National Nuclear Security Administration under Contract No. DE-NA0003525.