Unintentionally doped impurities formed in the microstructures of free-standing GaN grown with facets were studied using confocal magneto-photoluminescence (PL) microscopy. Donor-bound exciton related peaks in PL spectra and their magnetic behavior allowed us to distinguish typical donor impurity atoms, such as silicon and oxygen. Combining this technique with confocal microscopy also revealed the spatial distribution of the impurities. The results showed that angled facets tend to incorporate oxygen. Moreover, even facets angled at a few degrees with respect to the (0001) surface cause a noticeable change in oxygen incorporation on the order of 1 × 1016 cm−3.
GaN is a key semiconductor material used in the development of high power electronic devices and light emitting diodes that operate in the region between ultraviolet and green light. Recent progress in crystal growth techniques has led to the generation of high quality GaN crystals;1,2 however, the non-negligible concentration of impurities in GaN can affect device quality and lifetime.
Oxygen is a typical unintentional shallow donor in GaN whose incorporation efficiency depends on the facet growth.3–9 When the Fermi energy increases with an increase in oxygen concentration, the concentration of gallium vacancies (VGa) as well as the complexes between VGa and oxygen on nitrogen sites (VGaON) also increases because the formation energy decreases,10–12 where VGa and VGaON act as deep acceptors. Moreover, VGa- and VGaON-related complexes induce undesirable luminescence,10–14 such as yellow luminescence (YL), and VGa-defect complexes can lead to nonradiative recombination centers (NRCs);15–17 both reduce the internal quantum efficiency of near-band-edge emission by shortening the nonradiative recombination lifetime. Unintentionally, oxygen-doped GaN also reduces the transport properties due to an unwanted parasitic channel.18 Therefore, controlling the oxygen concentration is vital for fabricating effective devices. However, although many studies have analyzed the concentration of impurities (specifically oxygen) for {0001}, non-polar, or semi-polar facets,3–9 fewer studies have analyzed vicinal facets inclined by several degrees to particular crystallographic facets.
Photoluminescence (PL) spectroscopy of excitons has previously been used to evaluate GaN crystals because the PL spectra of excitons trapped by impurities are sufficiently sharp to be described by a hydrogen model when the doping concentration is low. Furthermore, both the PL intensity and peak energy are sufficiently sensitive to analyze the types and concentrations of impurities. Thus, PL spectroscopy is suitable for studying pure GaN with an extremely low concentration of impurities.
In this article, we examine the GaN epitaxial layer with a low impurity concentration by using confocal magneto-PL microscopy. Specifically, we compare flat (0001) regions containing extremely low impurity concentrations with a micrometer-scale microstructure consisting of several facets. PL spectra are then analyzed to identify the type of unintentionally doped impurities. We find that facets angled by a few degrees with respect to (0001) incorporate a non-negligible amount of oxygen on the order of 1 × 1016 cm−3.
A GaN sample with a (0001) plane was purchased from Sumitomo Electric Industry. An un-doped GaN epitaxial layer with a thickness of 4–6 μm was grown by metalorganic vapor phase epitaxy (MOVPE). This layer is unintentionally, yet naturally n-doped; the nominal carrier concentration is 0.5–5 × 1016 cm−3. A 1 to 3-μm-thick GaN buffer layer was used prior to the growth of the epitaxial layer. The substrate was grown by advanced dislocation elimination by epitaxial-growth with inverse-pyramidal pits (A-DEEP).1 This technique provides extremely low threading dislocations (TDs) in a large area by intentionally creating regularly arranged regions (cores). The TD densities inside and outside of the cores were 107–109 cm−2 and <106 cm−2, respectively.1 Hexagonal or dodecagonal inverse-pyramidal pits were formed by facets on the surface of the substrate.1
A hexagonal microstructure was observed on the surface of our sample, which reflected the shape of the hexagonal inverse pyramidal pits formed on the substrate (Fig. 1). The size of the structure was ∼70 μm, and the depth was ∼800 nm.
(a) Surface morphology of a hexagonal microstructure formed in un-doped GaN observed by atomic force microscopy. (b) Surface angle calculated from the arctangent of the magnitude of the gradient vector19 obtained from Fig. 1(a). Isolated metals on the left side of the microstructure in Fig. 1(b) are markers used to identify the microstructure during PL measurement.
(a) Surface morphology of a hexagonal microstructure formed in un-doped GaN observed by atomic force microscopy. (b) Surface angle calculated from the arctangent of the magnitude of the gradient vector19 obtained from Fig. 1(a). Isolated metals on the left side of the microstructure in Fig. 1(b) are markers used to identify the microstructure during PL measurement.
The hexagonal microstructure was composed of features inclined at an angle of a few degrees with respect to the (0001) facet, i.e., vicinal (0001) surfaces [Fig. 1(b)]. Therefore, we speculate that the inclined features observed in the sample are composed of c-planes and either or both non-polar or semi-polar facets, as well as growth facets. In the remainder of this paper, we term these inclined features as angled facets.
The TDs in the initial substrate reach the epitaxial layer, and the TD density of the hexagonal microstructure is higher than that of the flat region, corresponding to the initial substrate.1 However, the TDs tend to coalesce with an increase in growth thickness; specifically, the density in the cores is expected to decrease in the higher density region.20
To study how impurities are incorporated into the GaN and the hexagonal microstructure, we used confocal PL microscopy, in which an optical fiber works as a pin hole.21 We used a frequency doubled Ti:sapphire laser with a wavelength of 350 nm and a 4-ps pulse width as a light source. The laser light was focused on the sample with a spot diameter of ∼0.6 μm after passing through a single-mode fiber. The excitation power was ∼8 μJ/cm2. The PL from the sample was collected by an objective lens and transferred to a 30-cm monochromator through a multi-mode fiber. PL spectra were detected by a cooled charge-coupling detector (CCD). The energy resolution of the system was approximately 1.14 ± 0.11 meV, and the standard uncertainty in the photon energy of a PL peak was approximately 0.16 meV.22 As discussed later, this allowed us to resolve the emissions related to excited states of neutral donors under a magnetic field. We performed the measurement at 4 K to eliminate both thermal distribution and carrier diffusion. The sample was set in a vacuum. A magnetic field B can be applied up to 14 T perpendicular to the (0001) plane (i.e., a Faraday configuration). The sample can be moved by piezoelectric stages located at the 4-K region, and the PL spectrum can be obtained point by point. Then, 60 × 60 μm2 PL intensity maps can be constructed by raster scanning the sample. The excitation power density is stable within ±11% during the mapping measurement.
The PL spectra have several peaks in both the hexagonal microstructure and the flat region (Fig. 2). Sharp emissions are typically assigned as transitions of donor- and acceptor-bound exciton states. A donor-bound exciton state is a four-particle state in which two electrons and a hole are bound to a +1 charged donor.23 The initial donor-bound exciton state relaxes into a neutral donor state by emitting a photon, i.e., D0X → D0 + photon. Here, we denote this transition using D0X:D0, where D0 corresponds to the type of neutral donor, which is either Si0 or O0. The final state is either the n = 1 ground state or the excited states n = 2, 3, … of the neutral donor. Here, n is the principal quantum number when the neutral donor is regarded as a hydrogen model. The predominant PL emissions from the hexagonal microstructure and flat region peak at 3.4727 eV and 3.4736 eV, respectively (Fig. 2). As discussed later, these peaks are attributed to the transitions from donor-bound excitons to the n = 1 neutral donor state, D0XA:D01s, where XA denotes an exciton consisting of an electron and an A-valence band hole and 1s denotes an orbital of the neutral donor. These predominant peak energies are higher than the PL peak energy (3.4710–3.4720 eV)24,25 from a strain-free GaN,24,25 suggesting that our sample has slight compressive stress. The transitions to n = 2, 3, …, known as two-electron transition satellite (TES)24–28 and labeled O0XA:O02s, Si0XA:Si02s, and Si0XA:Si02p, are also observed at approximately 3.45 eV. Peaks originating from acceptor-bound excitons (A0XA) and free excitons (XA) are also visible in both spectra. Notably, the photon energies of these peaks, 3.4681 eV and 3.4800 eV, are the same for the hexagonal microstructure and the flat region. This suggests a negligible difference in the band gap and strain effect between the hexagonal microstructure and the flat region. Small peaks between D0XA:D0 and XA observed in the hexagonal microstructure and the flat region may correspond to the transitions of donor-bound excitons related to a B-valence band hole (D0XB:D01s).
PL spectra measured from the hexagonal microstructure (blue) and the flat region (red). Peaks are assigned from the peak energy and the energy shift in the magnetic field (Fig. 3) as discussed in the text. Donor-bound exciton related peaks are labeled (initial state):(final state). O0 and Si0 correspond to the neutral donors of oxygen and Si, respectively. XA and XB represent a free exciton, which is the bound state of an electron, and an A- and a B-valence band hole, respectively. 1s, 2s, and 2p represent orbitals obtained by treating the neutral donors as a hydrogen model.
PL spectra measured from the hexagonal microstructure (blue) and the flat region (red). Peaks are assigned from the peak energy and the energy shift in the magnetic field (Fig. 3) as discussed in the text. Donor-bound exciton related peaks are labeled (initial state):(final state). O0 and Si0 correspond to the neutral donors of oxygen and Si, respectively. XA and XB represent a free exciton, which is the bound state of an electron, and an A- and a B-valence band hole, respectively. 1s, 2s, and 2p represent orbitals obtained by treating the neutral donors as a hydrogen model.
To verify the PL peak assignment, we obtained the PL under a magnetic field (Fig. 3). The photon energies of the predominant donor-bound excitons (D0XA:D01s) do not depend on B (data not shown). In contrast, the TES transition peaks labeled Si0XA:Si02s, Si0XA:Si02p, and O0XA:O02s show clear B-dependence [Fig. 3(a)]. To quantitatively determine the spectral behavior under B, we used a hydrogen model.25,28 We calculated the theoretical values using the variational method,29 assuming an effective electron mass of 0.22m0, which is close to the typical value in GaN,25,28,30 where m0 is the intrinsic electron mass, and effective Rydberg constants (R*) of 34.1 meV and 30.9 meV for the hexagonal microstructure and flat region, respectively. The B-dependence of the TES transition peaks agrees well with the theoretical values [shown by blue and red dashed lines, respectively, in Fig. 3(b)]. This suggests that the TES transition peaks from the hexagonal microstructure and the flat region originate from oxygen- and Si-bound excitons, respectively.25,27,30,31 The B-dependence of the peak energies from 1s to 2p0,1 and 3d1 also agrees well with the theoretical values calculated with the same effective Rydberg constants combined with an offset [Fig. 3(b)]. Each offset includes the energy splitting of neutral donor states,31 but predominantly reflects the energy difference between the ground (l = 0) and the first excited (l = 1) donor-bound exciton states,25 where l is the angular momentum of the hole rotator states.23,27 The values of these offsets, i.e., −1.5 meV and −1.0 meV for the hexagonal microstructure and flat region, respectively, agree well with those reported in previous literature.25
(a) PL spectra under B = 0 T, 5 T, 10 T, and 14 T perpendicular to the (0001) plane. Blue and red lines correspond to the spectra measured in the hexagonal microstructure and the flat region, respectively. The spectra are offset for clarity. (b) Energy differences between the peak energies of D0XA:D01s and the TES transitions as a function of B are taken from Fig. 3(a). Red and blue dashed lines denote the theoretical calculation conducted by the variational method assuming that neutral donors are treated as a hydrogen model.29 Each 2s line is fitted by E = (E2s − E1s)R* to determine the effective Rydberg constants (R*), where E2s and E1s are calculated as a function of B by assuming an effective electron mass of 0.22m0.
(a) PL spectra under B = 0 T, 5 T, 10 T, and 14 T perpendicular to the (0001) plane. Blue and red lines correspond to the spectra measured in the hexagonal microstructure and the flat region, respectively. The spectra are offset for clarity. (b) Energy differences between the peak energies of D0XA:D01s and the TES transitions as a function of B are taken from Fig. 3(a). Red and blue dashed lines denote the theoretical calculation conducted by the variational method assuming that neutral donors are treated as a hydrogen model.29 Each 2s line is fitted by E = (E2s − E1s)R* to determine the effective Rydberg constants (R*), where E2s and E1s are calculated as a function of B by assuming an effective electron mass of 0.22m0.
To study the spatial distribution of donors, we obtained 60 × 60-μm2 PL maps at B = 0 T (Fig. 4). The map of integrated intensity over the entire spectrum from 3.44 eV to 3.51 eV (i.e., panchromatic PL intensity) is homogeneous over the scanned area except for the boundary between the hexagonal microstructure and the flat region [Fig. 4(a)]. This suggests that external quantum efficiency, which is a product of the electron excitation probability, the internal quantum efficiency, and the photon extraction efficiency do not depend on the facets under these slightly high excitation conditions. Although the TD density is high in the hexagonal microstructure, TDs do not strongly affect the external quantum efficiency (even under low excitation conditions ∼60 nJ/cm2, data not shown).
Spatial maps of PL intensity (a) integrated over spectra obtained from 3.44 eV to 3.51 eV and integrated over the energy window, (b) 3.4783–3.4815 eV (XA), (c) 3.4721–3.4727 eV (), (d) 3.4733–3.4739 eV (), (e) 3.4456–3.4482 eV (), and (f) 3.4493–3.4519 eV () at B = 0 T. The maps of (c), (d), and (f) include part of the peaks of , , and , respectively, due to their overlap. Each map represents a 60 × 60-μm2 area (61 × 61 pixel size) on the sample.
Spatial maps of PL intensity (a) integrated over spectra obtained from 3.44 eV to 3.51 eV and integrated over the energy window, (b) 3.4783–3.4815 eV (XA), (c) 3.4721–3.4727 eV (), (d) 3.4733–3.4739 eV (), (e) 3.4456–3.4482 eV (), and (f) 3.4493–3.4519 eV () at B = 0 T. The maps of (c), (d), and (f) include part of the peaks of , , and , respectively, due to their overlap. Each map represents a 60 × 60-μm2 area (61 × 61 pixel size) on the sample.
In contrast to the panchromatic PL image, the integrated PL map of each PL peak shows a clear difference, which reflects the hexagonal microstructure [Figs. 4(b)–4(f)]. The PL intensity of the free exciton XA is weak in the hexagonal microstructure [Fig. 4(b)]. This is consistent with the high TD density in the hexagonal microstructure, which affects the internal quantum efficiency. The PL intensity of excitons bound to oxygen, O0XA:O01s and O0XA:O02s, is strong in the hexagonal microstructure [Figs. 4(c) and 4(e)], whereas the PL intensity of those bound to Si, Si0XA:Si01s and Si0XA:Si02s, is strong in the flat region [Figs. 4(d) and 4(f)]. These results suggest that oxygen and Si tend to be incorporated in the hexagonal microstructure and the flat region, respectively.
The surface of the hexagonal microstructure consists of c-planes and either or both non-polar or semi-polar facets. The facets are angled at a few degrees with respect to (0001) [Fig. 1(b)]. Oxygen is more likely incorporated by non-polar and semi-polar facets than by polar (0001) facets;3,4,7,9 thus, the PL intensity from oxygen-bound excitons is strong in the hexagonal microstructure.
In order to quantify the oxygen and Si concentrations, we perform secondary ion mass spectrometry (SIMS) analysis. In the hexagonal microstructure, oxygen concentration is 6.4 × 1016 cm−3, whereas that in the flat region is 1.2 × 1016 cm−3 [Fig. 5(a)]. These values are consistent with oxygen related PL maps in which higher intensities in the hexagonal microstructure are detected [Figs. 4(c) and 4(e)]. In contrast, Si concentration in the hexagonal microstructure 1.7 × 1016 cm−3 is lower than that in the flat region 3.2 × 1016 cm−3 [Fig. 5(b)], and Si related PL maps also agree with the Si concentrations [Figs. 4(d) and 4(f)]. It has been reported that Si concentration in GaN can be reduced by intentional oxygen doping32 because the binding energy of VGaON is larger than VGa bound to Si on the Ga site complex (VGaSiGa).10 Our results show similar trends, however, revealing that unintentional and relatively small increase in the oxygen concentration ∼5 × 1016 cm−3 can reduce the Si concentration on the order of ∼1 × 1016 cm−3. Furthermore, both oxygen and Si concentrations are consistent with the fact that a noticeable increase in the PL width and decrease in the PL peak energy are not observed. When doping concentration increases, the PL peak width increases through inhomogeneous broadening,32 and the PL peak red-shifts due to band gap renormalization.32,33 The PL peak width in the microstructure and the flat region calculated from PL spectra (Fig. 2) are 1.4 meV and 1.1 meV, respectively, and the peak energy difference between O0XA:O01s and Si0XA:Si01s is 0.9 meV.
SIMS depth profiles of (a) oxygen and (b) silicon. Dashed lines denote background levels. In the hexagonal microstructure and the flat region, representative 20 × 20-μm2 areas are analyzed.
SIMS depth profiles of (a) oxygen and (b) silicon. Dashed lines denote background levels. In the hexagonal microstructure and the flat region, representative 20 × 20-μm2 areas are analyzed.
It should be noted that the GaN layer studied here is a homoepitaxial layer grown on a GaN substrate. Due to large differences in the lattice constant and thermal expansion coefficient between GaN and sapphire, GaN epilayers grown on a sapphire substrate typically exhibit a high TD density and rough surface morphology. Therefore, the fact that the oxygen concentration in the GaN epilayer grown on the GaN substrate is lower than that on the sapphire substrate34 should reflect the macroscopic characteristics. However, we conclude that an increase in the oxygen concentration on the order of 1 × 1016 cm−3 can be caused by the microscopic features, even when the epilayer is grown on the GaN substrate. For future applications that require extremely low oxygen concentrations, ideal two-dimensional growth will be required. Furthermore, we speculate that residual oxygen in the reactor was the dominant source of oxygen in this study because we did not use sapphire as the initial substrate.
In conclusion, this research examined a state-of-the-art pure GaN material with low donor concentration by low-temperature magneto-PL microscopy. Analysis of the donor-bound exciton levels revealed the types of donors incorporated in both the hexagonal microstructure and the flat region. Our results indicated that a facet angle, even one of a few degrees with respect to (0001), leads to a non-negligible amount of oxygen incorporation on the order of 1 × 1016 cm−3.
DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author upon reasonable request.
The authors are grateful to K. Hashimoto and Y. Niimi for experimental support. This work was supported by a Grant-in-Aid for Scientific Research (Grant Nos. 17H01037 and 19H05603) from the Ministry of Education, Culture, Sports, Science, and Technology (MEXT), Japan, the Asahi Glass Foundation, and the Center for Spintronics Research Network.
REFERENCES
The equations of c-plane and surface of the sample are z = 0 and , respectively, where zAFM is height of the sample measured by AFM Fig. 1(a).
As a sample of statistical calculations, 6 lines originating from a Ne lamp are used. Sample mean and sample standard deviation of the FWHMs are 0.114 nm and 0.005 53 nm, respectively. We estimate spectral reolution to be approximately 0.114 ± 0.011 nm. Sample mean and sample standard deviation of wavelength differences between measured and specification values are 0.007 70 nm and 0.008 22 nm, respectively. We approximate the standard uncertainty in the wavelength of a PL peak by 0.0159 nm. A conversion factor to the photon energy at 3.44 eV is 9.94 meV nm−1, according to .