Parasitic p–n junctions formed at V-pit defects in p-GaN

GaN and its solid solutions are currently widely used for fabrication of light emitting devices (LEDs) operating in the visible and UV spectral range and in high-power devices, such as rectifiers and field effect transistors.^1,2 However, in spite of rapid progress in the development of GaN growth technology with low dislocation densities, many optical and electrical properties of extended defects in this material are not fully understood.³ One of the most common defects in GaN-based structures is a so-called V-pit, a hexagonal pit formed by threading dislocations when they cross the surface of the growing layer.⁴ The facets of these V-pits are formed by six {10−11} planes inclined in respect to the (0001) growth surface at 61° angle.^5–8 Often, but not always, these V-pits can have a flat bottom.^6–9 These defects are believed to be partly responsible for the GaN-based LEDs demonstrating good quantum efficiency, despite a very high density of threading dislocations exceeding ∼10⁸ cm⁻² when grown on alien substrates (sapphire, SiC, Si).⁴⁻¹⁰ The explanation commonly offered is that the quantum wells (QWs) in LEDs are inclined on the facets of the V-pit and are thinner than on the flat surface and thus have a higher bandgap. As a result, recombination mainly occurs in the narrow-bandgap QWs portions on the flat surface and the impact of nonradiative recombination via threading dislocations is strongly diminished.⁴⁻¹⁰ 

For that reason, it is often considered advantageous to have as many threading dislocations as possible to be terminated with V-pits and the dimensions of the pits to be large.⁴⁻¹¹ The situation is not as straightforward as that: In concentration (and hence the bandgap of the GaN/InGaN QW) on the facets of the V-pits can be measurably different from the flat portion of the structure, electrically active impurities can be segregated on the facets of the V-pits, the hole injection efficiency into the QWs can be different in the vicinity of V-pits, and the density of nonradiative recombination defects can vary near the V-pits.^10,12−15 Detailed understanding of the properties of V-pits is necessary to optimize the performance of GaN-based LEDs.

The properties of V-pits have been studied in detail for n-GaN, where it has been shown that the pit facets are easily decorated with oxygen and thus show a higher shallow donor density than in the flat portions of the films. The result is a strong increase of the leakage current of the Schottky diodes.¹³ It was also observed that enhanced shallow donor density on the facets of the V-pits leads to a red shift of the band edge photoluminescence (PL) peak, more pronounced for the facets than for the flat bottom.^6,7 For GaN/InGaN QW structures, it has been noted that the In concentration on the facets of the V-pits can be higher than in the flat (0001) oriented part of the film, which can overpower in some cases the increase of the bandgap due to the thinning of the QW on the facets and result in either blue shift or red shift of the QW PL peak near the V-pit and to possible negative effects of the V-pits on QW luminescence efficiency.¹⁰⁻¹² Among the growth factors, it has been noted that the addition of In surfactant can influence the density and dimensions of the V-pits in QW structures,¹⁶ while increasing the temperature of the barrier growth can increase the dimensions of the V-pits.¹⁷ However, virtually nothing is known about the optical and electrical properties of the V-pits in p-type GaN. In this paper, we present the results of V-pit studies performed by cathodoluminescence (CL) and electron beam induced current (EBIC) measurements. For comparison, the results of similar measurements are presented for V-pits in freestanding n-GaN crystals.

The samples used were (a) a p-GaN film grown by metalorganic chemical vapor deposition (MOCVD) and (b) an unintentionally doped (UID) n-GaN freestanding crystal grown by hydride vapor phase epitaxy (HVPE). The p-GaN film was grown on basal plane sapphire and consisted of 1.6 μm-thick unintentionally doped (UID) n-GaN buffer, and 2 μm-thick layer doped with Mg to the Mg concentration of (3–4) × 10¹⁹ cm⁻³ determined by secondary ion mass spectrometry (SIMS) calibration. The film was grown in the same way as the p-GaN emitter layer in our standard blue LED process fabrication. The electrical properties of the UID n-GaN buffer were measured after a calibration run with no p-GaN deposition. These measurements showed that the net shallow donor concentration at the surface, as determined by capacitance–voltage (C–V) profiling in UID n-GaN buffer, was below 10¹⁶ cm⁻³.

The HVPE sample was acquired from Kyma Technologies Inc. (USA) and had thickness of 650 μm, the net donor concentration of 5 × 10¹⁶ cm⁻³, and the threading dislocation density below 10⁷ cm⁻². Electrical properties and deep electron and hole traps spectra were studied previously and are described in detail elsewhere.^18,19

The CL studies were carried out in a scanning electron microscope JSM-6490 (Jeol, Japan) with the MonoCL-3 system. A Hamamatsu photomultiplier was used as a detector. Measurements were carried out with beam energy of 10 keV and beam current of 1–10 nA in the temperature range from 80 to 300 K. EBIC measurements were carried out at room temperature in the scanning electron microscope JSM-840 (Jeol, Japan) with beam current about 0.1 nA and beam energy in the range from 15 to 39 keV. The EBIC collection efficiency was taken as the EBIC current I_c normalized by the product of the probing electrons energy E_b and beam current I_b, I_c/(E_b × I_b). Experimental setups are described in detail elsewhere.^3,20

For the MOCVD p-GaN sample, we also performed electrical and deep trap spectra studies including (a) current–voltage (I–V) measurements in the dark and under illumination with a high-power (optical output power density 250 mW/cm²) GaN-based LED with peak wavelength of 365 nm and (b) capacitance vs frequency (C–f) measurements (frequency range 20 Hz–1 MHz) in the dark and under illumination with high-power LEDs with peak wavelengths from 950 to 365 nm and optical power density 250 mW/cm², capacitance vs voltage (C–V) measurements in the frequency range 20 Hz–1 MHz in the dark and under illumination, deep level transient spectroscopy with electrical (DLTS) or optical (ODLTS) injection. For these measurements; for the EBIC characterization, the contacts were prepared by e-beam evaporation through a shadow mask of Ni/Au (50/200 nm) and rapid thermal annealing at 600°C for 30 s in nitrogen, followed by room temperature deposition of 30-nm-thick, 1 mm in diameter Ni dots. The Ni/Au provides the Ohmic contact, while Ni forms a Schottky diode to p-GaN. Experimental setups are described elsewhere.²¹⁻²³ 

I–V characteristics of the MOCVD p-GaN measured at 295, 130, and 400 K are shown in Fig. 1. The figure also shows the I–V characteristics measured at room temperature and 130 K with 365 nm LED excitation (photon energy 3.4 eV) (at 400 K the photocurrent was strongly masked by the increased dark current). The plus sign in the voltage axis in this figure corresponds to positive voltage applied to the Ni Schottky contact. There are three key points from the data: (a) the current for two voltage directions is similar, (b) the temperature dependence of the dark current is quite weak (the activation energy is close to 0.1 eV, the actual dependence not shown for the lack of space), and (c) the open-circuit voltage is positive, as would be the case if the Ni Schottky diodes were shorted and the signals were determined by the two p-GaN/n-GaN junctions turned face-to-face, with the contribution of the junction with the thinner Ni electrode dominant because of the lower area and better contact transparency (in the case of I–V measurements under illumination). As shown in Sec. III B, the V-pit density in the p-GaN film is quite high and a credible explanation of the lack of rectification at the Ni/p-GaN Schottky diode seems to be the strong leakage via the V-pits, as in the case of n-type GaN.¹³ 

Capacitance–voltage results also confirm the lack of rectification by the Ni Schottky diode and indicate the charged center concentration to be 9.3 × 10¹⁷ cm⁻³, with the built-in voltage close to 3 eV, as is often observed in GaN p–n junctions.²³ Illumination with high-power 365 nm wavelength LED increased the photo-concentration to 10¹⁸ cm⁻³ (Fig. 2). The measured concentration is too high to correspond to the space charge boundary movement with applied voltage in the n-GaN buffer of the film, so that the measured concentration is most likely due to the p-GaN side of the junction. The reason is that the n-side is almost depleted down to the ∼(0.5–1) μm adjacent to the sapphire substrate where the donor concentration is very high²³ so that the space charge region boundary movement with changing voltage mostly occurs in the p-type portion of the p–n junction.

The DLTS spectrum of the sample measured at 10 kHz to alleviate problems due to the high series resistance²¹ is shown in Fig. 3. The dominant feature of the spectrum is the peak near 300 K due to a prominent majority carriers trap (the capacitance decreases with time after the application of the forward bias pulse).²⁴ The activation energy is 0.85 eV and capture cross section is 8.9 × 10⁻¹⁴ cm² if the signal comes from the p-GaN region, and the trapped charges are holes, as suggested by I–V and C–V measurements. The parameters of these deep acceptors are close to the characteristics of gallium-vacancy-related deep acceptor complexes with shallow donors (V_Ga-D)²⁻ or to C acceptors on N site, with C_N⁻ expected to be major deep hole traps in n-GaN.²³ The V_Ga-related defects are not expected to be present in high concentrations in p-GaN because of their high formation energy when the Fermi level is close to the valence band edge,^25,26 but the C_N⁻ acceptor traps have been detected in p-GaN by DLTS.²⁷ Another possibility²⁸ is a deep donor with a transition level near E_v + 0.85 eV, which has been predicted for a complex of nitrogen vacancy donors V_N³⁺ with Mg acceptors, (V_N-Mg)²⁺ and can be easily formed in p-type material. In principle, these two scenarios can be distinguished by measuring CL spectra: theory predicts a broad luminescence line in the yellow–green region of 2.3–2.5 eV for the C_N⁻,^25,26 whereas for (V_N-Mg)²⁺ deep donor, the respective luminescence band should be lying in the red spectral region and centered near 1.8 eV.²⁸ However, detailed CL spectra measurements allowing to decide which of the defects is observed have not yet been done for the red–green spectral range.

The same hole traps could also be clearly seen in ODLTS spectra measured with an excitation pulse provided by illumination with high-power LED with photon energy 3.4 eV (Fig. 3). In addition, an electron trap with a peak near 210 K and the level near E_c − 0.45 eV and electron capture cross section of 1.6 × 10⁻¹⁶ cm² could be detected. Such traps have been previously reported in some n-GaN films.²³ 

Secondary electron (SE) imaging of the MOCVD p-GaN sample and the freestanding HVPE n-GaN sample are shown in Figs. 4 and 5. For the p-GaN, there is a high density (10⁶–10⁷ cm⁻²) of hexagonal V-pits with flat bottom and in-plane dimensions of about 1 μm, which allows us to calculate their depth as being below 1 μm (Fig. 4). In freestanding HVPE n-GaN, the V-pit density does not exceed 10² cm⁻²; however, their size is larger than 100 μm (Fig. 5).

The room temperature CL spectra measured in HVPE n-GaN are shown in Fig. 6. On the facets of the V-pits, the CL intensity increases and the CL peak position shifts to lower energies, which correlates with the results published in Ref. 6. The spectrum measured far from a defect can be described with one Gaussian peak centered at 3.4 eV with width of 50 meV. The spectrum in the V-pit center can be fitted with two Gaussian peaks exp[−(E − E_m)²/2σ²] with energies E_m of 3.375 and 3.32 eV and σ equal to 45 and 80 meV, respectively. The spectrum in the V-pit side facet can be fitted also with two Gaussian peaks with E_m of 3.25 and 3.05 eV and σ equal to 95 and 100 meV, respectively. The Schottky diodes made on V-pits of the HVPE sample demonstrate a practically Ohmic behavior, i.e., they are shorted by V-pits. This can indicate an increased donor concentration in V-pits and correlates with oxygen donor segregation on the surface framing the V-pit.⁶ The origin of the emission bands observed is unknown; however, different bands are dominant for the V-pit-free “bulk” area and for the facets and bottom regions of V-pits. Thus, it can be assumed that different defects are accumulated in the V-pit facet and bottom regions. The emission band at 3.05 eV is close to that produced by dislocations gliding at room temperature.^15,20

Due to a smaller size of V-pits in the MOCVD p-GaN sample, it was difficult to separate the spectra from their edges and centers; therefore, the spectra measured present some average ones. Typical CL spectra obtained in p-GaN on the V-defect and far from it, obtained at 80 and 300 K, are shown in Fig. 7. In the CL spectra of bulk p-GaN, there is no clearly defined luminescence line associated with the near-band-edge transition. This is in good agreement with the results of Refs. 29 and 30 at high magnesium concentrations. A small peak close to the near-band-edge transition is present in some V-pit spectra; however, it can come from the n-GaN buffer because in pits, the e-beam can penetrate deeper into the sample. The spectra measured on V-pits are shifted to higher energies when compared to those on the defect-free region. This trend is opposite to the case of n-GaN. However, fitting the spectra with Gaussian peaks shows that in the range from 2.5 to 3.6 eV, they can be described by superposition of three peaks with E_m of 3.28 and 3.2 eV for both V-pits and defect-free region, and the most intense peak at 3.03 eV for V-pits and 2.98–3 eV for the defect-free region. E_m for all peaks is practically independent of temperature in the range 80–300 K. The relative intensities of Gaussian peaks normalized by 3–3.03 eV peak intensity for V-pit and two defect-free regions are shown in Fig. 8. While the positions of Gaussian peaks in defect-free regions are the same across the sample, their intensities varied in different regions. Therefore, the data for defect-free regions are shown in Fig. 8 for two typical cases.

Thus, it can be assumed that the emission in both defect-free and V-pit regions is determined by the same defects. However, in V-pits, the relative intensities of high energy peaks are higher, which explains the overall spectral shift to higher energies. The emission peaks with similar E_m have been already observed in GaN with high Mg concentration.²⁹ Nevertheless, some differences remain. As reported in Ref. 29, the intensity of the UV band at 3.2 eV practically disappears at room temperature, and at 300 K, the blue band at 3 eV dominates in the spectrum. As seen in Fig. 8, the opposite behavior is observed in our work, i.e., relative intensity of UV bands increases at room temperature. Contrary to observations in Ref. 29, the position of the blue band is practically independent of the temperature. The origin of the blue and UV luminescence bands in p-GaN (Mg) is still under debate. Theoretical analysis^25,26 suggests that the blue line near 3 eV is due to a strongly lattice-coupled quasi-deep isolated Mg acceptor center, whereas the UV line near 3.2 eV could be related to a transition involving a donor state of the Mg–H complex with the transition level near E_v + 0.13 eV. One can easily envisage the change in relative concentrations of such centers in the V-pit-free and V-pit regions of the p-GaN film.

In the EBIC mode, V-pits in p-GaN produce a bright contrast, i.e., the collected current increases at them (Fig. 9). The current direction was the same as would be observed for the p-GaN/n-GaN junction with the Ni Schottky diode effectively shorted, most likely by the V-pits, as already inferred in Sec. III A from I–V and C–V measurements. The induced current is collected by the p–n junction between the n-type buffer layer and the p-type epilayer. As follows from Fig. 9, V-pits essentially enhance the excess currier collection. Pits usually increase the collected current due to the suppression of backscattering electron emission, which in turn increases the absorbed energy. However, in this case, the current increase cannot exceed 15%–20%. In the samples studied, the contrast exceeds 50% at 35 keV and the increase is even larger at lower beam energies; therefore, this purely “geometric” mechanism can be excluded.

The dependence of collected current I_c normalized by the product of beam energy (E_b) and current (I_b) on electron range R is shown in Fig. 10. The electron range is calculated as R = 0.47 × R_B, where R_B = 13.2 × E_b^1.75 nm is the electron Bethe range.^31,32 Far from V-pits, the collected current starts to increase for R > 1.5 μm. That means at smaller R, the electrons cannot reach the depletion region of p–n junction, which is in agreement with the p-layer thickness. With the increase of R, the number of excess carriers generated near the depletion region increases; therefore, the collected current also increases. To check if the collected current increase is determined by the layer thickness decrease in the pits, the curve in Fig. 10 is shifted by 1000 nm (the maximum pit depth). The collected current values in the pit are still larger than far from the pit, and the normalized EBIC current dependence on depth is different from the one observed in defect-free regions. Thus, the EBIC results allow us to assume that in V-pits, conductivity changes to n-type and the p–n junction formed around V-pits increases the collection efficiency. The outer radius of this n-type cylinder can be quite small; therefore, when the excess carriers are generated far from the buffer layer, the probability to reach the buffer layer decreases, which determined the current decrease at small R. The p–n junction formation around V-pit defects well correlates with the increase of oxygen concentration near them because oxygen is known to be a shallow donor in GaN.

Our results strongly suggest that V-pits in p-GaN can short Schottky diodes, similarly to what is observed in n-GaN. The reason is likely related to oxygen donor segregation on the side facets and the bottom of the V-pit funnel. In p-GaN, judging by the EBIC collection efficiency results, this is accompanied by the formation of a parasitic p–n junction around the V-pit. In GaN/InGaN multi quantum well (MQW) LEDs, the thickness of the p-GaN emitter layer is typically below 0.2 μm¹ and thus lower than the possible depth of the V-pit. Therefore, the high density of V-pits in such layers can lead to excessive leakage of the p–n junction and considerable loss of injection efficiency.

We also observe that, in the vicinity of V-pits, the relative intensity of Mg-related centers giving rise to the blue band near 3 eV and the UV band near 3.2 eV is seriously redistributed in favor of the latter, suggesting, based on theoretical predictions, that the V-pit region is enriched by the Mg–H complexes that have survived the Mg activation annealing.

The work at NUST MISiS was financially supported by the Ministry of Science and Higher Education of the Russian Federation in the framework of Increase Competitiveness Program of NUST “MISiS” (No. К2-2020-040). The work at IMT RAS was supported in part by the State Task No 075-00355-21-00. The work at Korea University was supported by the Technology Innovation Program (No. 20004479, Micro LED chip and module fabrication technology with integrated field effect transistor for IoT application) funded by the Ministry of Trade, Industry & Energy (MOTIE, Korea). The work at UF was sponsored by NSF DMR (No. 1856662 (James Edgar)].

The data that support the findings of this study are available within the article.