Deep trap spectra of Sn-doped $α$-Ga₂O₃ grown by halide vapor phase epitaxy on sapphire

Ga₂O₃ and related ternary solid solutions are promising next-generation materials for use in high-power devices. This is due to their excellent properties, including band gaps close to 5 eV, electric breakdown field about 3 times higher than for SiC and GaN, and electron saturation velocity similar to SiC and GaN.^1–4 The main focus in Ga₂O₃ materials and device research has been on the thermodynamically stable, monoclinic β-polytype, for which large diameter, high crystalline quality bulk crystals, epitaxial films, and heterojunctions exist.^1–5 This has led to demonstrations of prototype power rectifiers, field effect transistors, and solar-blind photodetectors.^2,4,5

However, other Ga₂O₃ polytypes are also of interest. The most important among those seems to be the α-polytype with corundum crystalline structure.¹ The band gap of α-Ga₂O₃ is higher than β-Ga₂O₃, about 5.2 eV vs 4.6–4.8 eV.^1,2 The higher symmetry of the rhombohedral corundum structure is an additional bonus compared to the monoclinic structure of β-Ga₂O₃, where the anisotropy of many properties is well documented.^6–8 Although α-Ga₂O₃ is not thermodynamically stable, its formation energy is only slightly higher than for β-Ga₂O₃ and formation of the single crystalline α-polytype can be achieved under nonequilibrium growth conditions. An additional attractive feature is the compatibility of α-Ga₂O₃ growth with deposition on other corundum-structured materials, most importantly sapphire, which is cheap, readily available in large diameters, and with excellent crystalline quality. A problem with all Ga₂O₃ polytypes is the lack of suitable p-type dopants in Ga₂O₃.^2–5 The way around this proposed so far is the use of Ga₂O₃ heterojunctions with wide-band gap materials with controlled p-type conductivity.² Although the formation of n-type α-Ga₂O₃ heterojunctions with p-type wide-band gap materials, such as copper oxide, nickel oxide, SiC, and GaN, have been reported,⁹ the processes involved are technologically challenging due to the difference in the lattice parameters, type of crystalline structure, and the necessity to overcome the formation of highly defective interfacial layers.⁹ High growth temperature of β-Ga₂O₃ also seems to present difficulties when combining gallium oxide with other wide-band gap materials allowing effective p-type doping.⁹ In that sense, the corundum structure is more compatible with growth on these materials, a better matching of in-plane lattice constants of α-Ga₂O₃ with advanced hexagonal structure wide-band gap materials, such as SiC or GaN; a typically lower growth temperature of α-polytype can make high-quality heterojunction preparation easier. It also helps that related corundum-structured oxides, such as α-(Ga_1−xIr_x)₂O₃ or α-(Ga_1−xRh_x)₂O₃,^10,11 have been reported to possess well-defined p-type conductivity. Moreover, because of the similarity in symmetry and the relative closeness of in-plane lattice parameters for the (00.1) plane of α-Ga₂O₃ and (111) plane of diamond, it could prove to be possible to grow high-quality heterojunctions between the n-type α-Ga₂O₃ and p-type or semi-insulating (111) diamond. If successful, this would help to overcome another serious drawback of Ga₂O₃, the relatively low thermal conductivity^1–5 that is a disadvantage for power devices.

There is still the challenge of growing single polytype α-Ga₂O₃ films with high crystalline quality and controlled doping, given that growth is on alien substrates, most commonly sapphire. Recently, high crystalline quality single polytype α-Ga₂O₃ films were grown by mist chemical vapor deposition (MIST CVD), a version of metalorganic chemical vapor deposition (MOCVD), on basal plane sapphire.^10,11 The crystalline quality can be further improved by using halide vapor phase epitaxy (HVPE) epitaxial lateral overgrowth (ELO) for masked growth on sapphire.¹² Dislocation densities <5 × 10⁶ cm⁻² in the window regions of such structures have been reported, although growth of α-Ga₂O₃ was competing with growth of the hexagonal ε-polytype.¹² Even without ELO, the crystalline quality of α-Ga₂O₃ films grown by HVPE on sapphire improves as the film thickness increases from ∼2 µm to 12 µm.¹³ 

Pulsed HVPE growth of α-Ga₂O₃ on sapphire produces films with screw dislocation densities of (2–4) × 10⁵ cm⁻² and edge dislocations on the order of 10⁹ cm⁻².¹⁴ Doping with Sn donors in MIST CVD produced n-type material with shallow donor densities controlled from 10¹⁷–10¹⁹ cm^−310,11 while undoped films prepared by either MIST CVD or by different versions of HVPE were highly resistive.¹⁴ These developments have made it possible to demonstrate α-Ga₂O₃-based power rectifiers with attractive performance: low on-resistance of 0.1–0.4 mΩ cm² and breakdown voltage of 531–855 V,^10,11 although the authors had to use lift-off and transfer of the grown device structures to overcome the heat control problems. However, little is known about the deep traps, diffusion lengths of nonequilibrium charge carriers, and luminescence spectra of doped and undoped α-Ga₂O₃ films, in contrast to abundant data for the β-polytype.^23–26 In undoped semi-insulating (SI) α-Ga₂O₃ films, deep traps near E_c − 1 eV pinning the Fermi level, deep traps with levels near E_v + 1.4 eV, and shallower electron traps near E_c − 0.3 eV and E_c − 0.6 eV, were reported.¹⁴ In this article, we present deep trap and minority carrier transport studies for Sn-doped α-Ga₂O₃ films grown by HVPE on basal plane sapphire.

Ga₂O₃ films were grown on c-plane sapphire substrates in a hot wall atmospheric pressure HVPE reactor with a multizone horizontal furnace. GaCl and oxygen were used as precursors. Sn was used as the n-type dopant. The GaCl vapor was synthesized in situ upstream in the reactor by the reaction of metallic gallium (99.9999%) and gaseous hydrogen chloride (99.999%). The yield of GaCl formation is estimated as >80%. The GaCl vapor was transported to the deposition zone of the reactor held at 500 °C, where it reacted with O₂ to produce Ga₂O₃. The HCl flow through the source was varied from 0.1 to 0.4 slm. The oxygen flow was constant at 0.2 slm. Argon was used as a carrier gas and the total gas flow through the reactor was up to 10 slm. Under these conditions, the deposition rate was 5–10 µm/h. The deposition time was adjusted to obtain Ga₂O₃ films with thickness 2–12 µm. Specifically, the Sn-doped films had thickness ∼2 µm. Three samples with n-type doping of about 10¹⁷ cm⁻³, 5 × 10¹⁷ cm⁻³, and 10¹⁹ cm⁻³ were prepared. After growth, the substrates were cooled to room temperature under a flow of Ar.

The samples were characterized by ω − 2Θ x-ray scans that showed all the films to be single-phase α-Ga₂O₃ with (00.1) orientation. Structural characterization included high resolution x-ray diffraction (HRXRD) rocking curves full width at half maximum (FWHM) for the (00.6) reflections,¹⁵ surface morphology observation in secondary electron (SE) mode of scanning electron microscope (SEM), microcathodoluminescence (MCL) at 90 and 300 K, electron beam induced current (EBIC) imaging of the surface with deposited Schottky diodes, and diffusion length measurements using the EBIC collection efficiency dependence on accelerating voltage of the probing SEM beam. Experimental setups were described previously.^15–17 The model used to determine the diffusion length from EBIC has been developed for Ga₂O₃¹⁶ and used for diffusion lengths in HVPE-grown β-Ga₂O₃.^18,19

Electrical measurements were performed using semitransparent 1-mm diameter Ni Schottky diodes and Ti/Au Ohmic contacts.¹⁴ These measurements involved capacitance-voltage (C-V) measurements at 20 Hz–1 MHz in the dark and under monochromatic illumination with high-power light emitting diodes (LEDs) with wavelengths from 365 to 940 nm, capacitance vs frequency (C-f), admittance spectra (AS),²⁰ deep level spectra with electrical (DLTS), or optical (ODLTS) injection.^21–24 

All samples had thickness ∼2 µm, with visually smooth surfaces. The results of HRXRD FWHM of the (00.6) HRXRD rocking curves measurements were close to 0.35°–0.43° (Table I), indicating a dislocation density >10⁹ cm⁻². For the most lightly Sn-doped sample, sample 1, EBIC imaging showed large dark defects, also seen in SE images. These are most likely growth pores. In addition, there are a large number of small dark defects, forming a cellular structure likely due to dislocation grain boundaries [Fig. 1(top)]. The density of these defects is slightly higher and the number of pores much lower for the heavily Sn-doped sample [Fig. 1 (bottom)]. Thick (10–12 µm) undoped semi-insulating samples demonstrate better characteristics, with FWHM typically 0.12°,⁹ i.e., three times lower than for the present set.

All Sn-doped samples showed clear n-type conductivity from C-f and C-V results. The C-f characteristics showed a plateau at low frequencies and roll-off due to the effect of series resistance [Fig. 2(a)]. Current-voltage measurements of the three samples also indicated high series resistances, with the saturation current density J_s = 7.1 × 10⁻⁷ A/cm², ideality factor 2.1, reverse current density at −2 V of 4 × 10⁻⁴ A/cm², series resistance R_s = 2.1 × 10³ Ω for the least heavily doped sample, sample 1, and saturation current density 3.2 × 10⁻⁵ A/cm², ideality factor 2.5, reverse current density at −2 V of 5.3 × 10⁻² A/cm², series resistance R_s = 210 Ω for most heavily doped sample, sample 3. This suggests that meaningful C-V and DLTS measurements have to be performed at frequencies below 10–20 kHz, which poses a problem with standard DLTS spectrometers but can be handled in our DLTS setup.²² For the two more heavily doped samples, the impact of leakage current on capacitance measurements at low frequencies limits the voltages to which the profiles can be measured and the temperatures to which the deep trap spectra can be reliably obtained. The 1/C² vs voltage plots were linear when measured at frequencies corresponding to the C-f plateau [see Fig. 2(b) obtained for sample 1 at 10 kHz]. The net shallow donor concentration, N_d, calculated from these characteristics was 1.1 × 10¹⁷ cm⁻³ for sample 1, 5 × 10¹⁷ cm⁻³ for sample 2, and 4.8 × 10¹⁹ cm⁻³ for sample 3 (see Table I). The voltage offset of the 1/C² vs V that should be close to the Schottky barrier height²¹ was 2.3 V at room temperature, slightly increased to 2.35 V at 80 K, and slightly decreased to 2.1 V at 400 K for the most lightly doped sample 1 [Fig. 2(b)]. The observed barrier height changes most likely reflect the change of the band gap with temperature.

The two lower doped samples showed photocapacitance for photon energies higher than 2 eV (for the heavily doped sample 3, the effects were qualitatively similar but reliable quantitative measurements were difficult because of the low photocapacitance compared to the dark capacitance). At low temperatures, the photocapacitance became persistent, i.e., not returning to the dark value for a long time. Figure 3(a) shows 80 K concentration profiles measured in sample 1 in the dark and under illumination with photons of various energies produced by high-power GaN-based LEDs (output power ∼250 mW for all LEDs). Figure 3(b) shows the spectrum of photogenerated carriers calculated as the photoconcentration in the nearly flat region of the profiles [we took it for consistency at depth of 0.2 µm in Fig. 3(a)] from which the dark concentration is subtracted. The photocapacitance did not change even after keeping the samples in the dark for an hour.

Two mechanisms produce such persistent photocapacitance changes at low temperatures. First, if traps with levels below the Fermi level in the space charge region (SCR) are photoionized, they can only return to the steady state by thermal emission of holes, which happens slowly at low temperatures. Second, if deep centers possess a barrier for capture of electrons, these electrons, once removed from the centers by light, can only return to their host centers at temperatures high enough to overcome the barrier for capture. These two situations can be distinguished by applying a forward bias pulse supplying electrons to the space charge region. In the first case, the photogenerated charge can be wiped out by flooding the SCR with electrons during the forward bias pulse. In the second case, the photogenerated charge will not be removed.^18,23,24 As seen in Fig. 3(c), with 3.4 eV photon excitation, both situations are encountered: after illumination, the concentration is persistent (the persistent photocapacitance curve, PPC), but can be partially quenched by application of a high forward bias (+2 V), long (2 min) pulse (PPC + 2 V curve). The spectral dependence of concentrations of such PPC and PPC + 2 V centers is shown in Fig. 3(b). For photons with energies >2 eV, centers with a barrier for capture of electrons are excited (all three photogenerated concentrations are close to each other), while, for photons with energies >2.7 eV, there appears a growing number of PPC electrons that can be partially quenched with forward bias application and thus are due to ionization of deep centers in the lower half of the band gap.

Some idea of the temperatures at which the persistent photocapacitance vanishes can be derived from C-T dependences, measured in the dark and after illumination at low temperature. These are compared for sample 2 in Fig. 3(d). The PPC curve obtained after illumination with 365 nm (3.4 eV) LED returns to the dark curve at ∼400 K. A more quantitative estimate of the energy barrier involved can be obtained from ODLTS spectra measurements with optical pulses provided by the 3.4 eV photons. The spectrum for sample 1 is shown in Fig. 4. A wide hole-trap-like ODLTS peak (the capacitance after the excitation pulse decreases with time, a negative peak in the DLTS/ODLTS spectra according to the convention adopted here) is seen. It is comprised of two peaks, with activation energies 0.3 and 0.6 eV. Clearly, 3.4 eV photons cannot excite hole traps near E_v + 0.3 eV or E_v + 0.6 eV in α-Ga₂O₃ with band gap close to 5 eV. Thus, we assume the transitions in these measurements are due to persistent photoexcitation of deep traps with barriers for capture of electrons ∼0.3 and 0.6 eV (for such centers, the capacitance after excitation will slowly decrease with time, as in the case of true hole traps²⁴).

According to spectral measurements of photocapacitance summarized by Fig. 3(c), there should also be deep traps with optical ionization energy >2.7 eV that do not possess a barrier for capture of electrons and can be quenched by the application of forward bias pulse. Such traps could not be detected in our ODLTS spectra measured to 470 K, most likely because of the very high thermal ionization energy of ∼2.3 eV of such centers. This is similar to β-Ga₂O₃ films grown by HVPE or MOCVD.^23,24 In that material, there were also hole traps at E_v + 1.4 eV attributed to Ga vacancies.^18,23,24

The spectra of deep electron traps in the upper half of the band gap of the epilayers were studied by DLTS measurements performed at ∼10 kHz, which prevented reliable spectra collection for biases higher than 1 V for sample 2 and higher than 0.5 V for sample 3 because of high leakage current. It also limited the temperatures to which DLTS spectra could be reliably interpreted to 350 K for sample 2 and ∼300 K for sample 3. The most detailed measurements could be done for the lightly doped sample 1. Figure 5(a) shows DLTS spectrum obtained for quiescent bias of −3 V, with voltage pulsed to 0 V for 3 s. The abscissa axis shows the DLTS signal, ΔC/C, multiplied by 2N_d × F⁻¹ where N_d is the shallow donor concentration obtained from C-V, ΔC = C(t₁) − C(t₂), t₁ and t₂ are the time windows for processing the DLTS spectra, C is the steady-state capacitance, and F⁻¹ is the spectrometer function.²¹ For temperatures corresponding to peaks in DLTS spectra, the magnitude of the peak in these coordinates is equal to the concentration of the centers without accounting for the λ-correction.²¹ 

Two prominent electron traps are present: A (energy E_a = 0.6 eV, electron capture cross section σ_n = 3 × 10⁻¹⁵ cm²) and B (E_a = 1.1 eV, σ_n = 1.2 × 10⁻¹³ cm²). The concentrations of both traps with account for the λ-correction are given in Table I and are >10¹⁴ cm⁻³. For more heavily doped sample 2, the reverse bias had to be decreased to −1 V, and even so, for temperatures close to 400 K, reliable measurements were not possible. The same A traps, but with a much higher concentration, and additional deep traps C (E_a = 0.85 eV, σ_n = 1.2 × 10⁻¹⁶ cm²) could be observed. Even for the most heavily doped sample 3, DLTS measurements were possible for temperatures <300 K, although the applied voltage had to be decreased to −0.5 V to avoid excessive leakage. The same A traps, but with a much higher concentration, could be observed. Additionally, another shallow electron traps D (E_a = 0.23 eV, σ_n = 3 × 10⁻¹³ cm²) were detected [the concentrations in Fig. 5(a) are divided by 10 for sample 3 to bring all DLTS spectra to scale]. Figure 5(b) presents the Arrhenius plots of the trap signatures for traps A, B, C, and D detected in our experiments. Table I gives the trap concentrations in sample 3 with account for λ-correction. From Fig. 5 and Table I, it can be seen that increasing the Sn doping steadily increases the concentration of the A electron traps and introduces new traps C and D. The dominant traps A and B and the trap C are quite similar to deep traps observed in semi-insulating α-Ga₂O₃ grown by pulsed HVPE.¹⁴ The energy levels and concentrations of the A and B traps in the lightly doped sample 1 are not very different from the dominant traps in similarly doped β-Ga₂O₃ grown by HVPE.^{18,19,24,25,27}

The diffusion length of the nonequilibrium charge carriers determined from fitting the EBIC signal collection efficiency I_c/(I_bE_b) dependence on beam energy E_b¹⁶ yields diffusion length L_d = 0.15 µm (Fig. 6), lower than the typical values in HVPE grown β-Ga₂O₃ (0.4–0.5 µm)^16,18,19 (I_c is the EBIC current, and I_b is the probing electron beam current of SEM, 1 nA in this case). Since the dislocation density in our samples is high, it cannot be ruled out that the diffusion length is limited by the dimensions of dislocation bounded cells as is the case for GaN with dislocation densities >5 × 10⁸ cm⁻².¹⁷ More studies are needed to check whether the diffusion lengths are limited by the density of specific deep traps, as observed for low dislocation density HVPE β-Ga₂O₃ films.¹⁸ DLTS/ODLTS measurements have to be conducted on α-Ga₂O₃ with various dislocation densities and compared to L_d measurements by EBIC.

The evolution of MCL spectra of the α-Ga₂O₃ with changing Sn doping and temperature were also examined. Earlier publications^27,28 describing photoluminescence (PL) or MCL spectra of α-Ga₂O₃ refer to powder material. The spectra were dominated by a broad band extending from ∼2 eV (625 nm) to ∼3.6 eV (344 nm), with a sharp peak near 3.445 eV (360 nm) whose magnitude increased with decreasing temperature. Figure 7(a) compares the room temperature MCL spectra of samples 1 and 3. These spectra were broad and consisted of a superposition of several peaks. For comparison, we also show in Fig. 7(a) the spectrum taken for high quality undoped semi-insulating layer. Lowering the measurement temperature from 300 to 90 K did not change the peak positions, but increased the MCL intensity and changed the relative intensities of the peaks, suggesting this involves donor-acceptor pairs (DAP) recombination with different shallow and deep donors and deep acceptors [Figs. 7(b) and 7(c) compare the room temperature and 90 K spectra of samples 1 and 3]. Deconvolution of the low temperature MCL spectra of the two samples into several Gaussian peaks yielded the peak positions as 339 nm (3.66 eV), 389 nm (3.19 eV), and 423 nm (2.93 eV) for the low doped sample 1 and 339 nm (3.66 eV), 428 nm (2.89 eV), 441 nm (2.81 eV), 489 nm (2.54 eV), and 540 nm (2.296 eV) for heavily doped sample 3. For the undoped SI sample, the Gaussian bands peaked at 385 nm (3.22 eV), 425 nm (2.92 eV), and 439 nm (2.82 eV). In neither sample did we observe band-edge luminescence. The attribution of the MCL bands to specific transitions cannot be reliably done at this point, but the shortest wavelength band at 339 nm (3.66 eV) is only observed for the two Sn-doped samples and may be related to DAP transitions involving Sn donors. The bands peaked near 385–389 nm (∼3.2 eV) and 423–428 nm (∼2.9 eV) are common for all samples, and the band near 439–441 nm (∼2.8 eV) is observed for heavily Sn-doped sample 3 and for the SI sample, while the long-wavelength band near 540 nm (2.3 eV) is specific to highly Sn-doped sample 3. Similar broad MCL spectra are often observed for β-Ga₂O₃ films and attributed to recombination involving polaronic hole states in the valence band and different kinds of DAP transitions.^29,30

The concentrations of deep traps in Sn-doped n-type α-Ga₂O₃ are the same order of magnitude as for β-Ga₂O₃ films and crystals,^{18,19,24–26} suggesting electron trapping-detrapping and current collapse in α-Ga₂O₃ rectifiers and FETs are not expected to be more severe than in β-Ga₂O₃. The trap levels also scale with those in the β-polytype: the concentrations of the dominant A traps increases for increased donor doping, similar to the E1 electron traps in β-Ga₂O₃.^19,25,26 Heavy doping that promotes point defect formation in α-Ga₂O₃ generates shallower electron traps D, similar to the effect of proton irradiation for β-Ga₂O₃ films grown by HVPE.¹⁸ The optical ionization thresholds of the deep acceptors detected in photocapacitance in α-Ga₂O₃ are considerably higher than in β-Ga₂O₃,^18,24 which probably reflects the higher band gap of α-Ga₂O₃.

The work at NUST MISiS was supported in part by Ministry of Education and Science of the Russian Federation in the framework of Increase Competitiveness Program of NUST MISiS (No. K2-2017-068). The samples were grown by Perfect Crystals LLC. The work at IMT RAS was supported by the State Task No. 075-00475-19-00. The work at UF was sponsored by Department of the Defense, Defense Threat Reduction Agency, No. HDTRA1-17-1-011, and monitored by Jacob Calkins and also by No. NSF DMR 1856662 (Tania Paskova).