Optical signatures of deep level defects in Ga₂O₃

The ultra-wide bandgap semiconductor Ga₂O₃ has now become a promising candidate for next generation high power electronics.^1,2 The ability to grow Ga₂O₃ by multiple techniques and to manufacture large-size, high-quality bulk wafers has led to rapidly expanding applications^3–12 chiefly because its large bandgap enables high breakdown electric fields estimated at 7–8 MV/cm, yet n-type doping ranging from intrinsic to degenerate is achievable.¹¹ These properties can be impacted by deep level defects that compensate free carriers, increase scattering that reduces carrier mobility,¹³ form gap states¹⁴ that “pin” Fermi levels,^15,16 and initiate trapping that limits breakdown voltage. Multiple defect features that vary with growth and processing have been reported,^17–26 but the nature of native point defects in Ga₂O₃ is relatively unexplored. We have now used nanoscale depth-resolved cathodoluminescence spectroscopy (DRCLS) and surface photovoltage spectroscopy (SPS) with plasma processing and neutron irradiation, combined with temperature-dependent Hall effect measurements, to correlate the spectral features of deep level defect gap states in Ga₂O₃ with their physical and donor/acceptor nature. The energy levels associated with these features display close correspondences with those predicted theoretically and measured using capacitance spectroscopies.

We used DRCLS and SPS to measure the energy level positions of electrically active defects within the outer tens of nanometers of Ga₂O₃ grown by various techniques, including low pressure chemical vapor deposition (LPCVD),^27–29 molecular beam epitaxy (MBE),³⁰ pulsed laser deposition (PLD),³¹ and edge-defined film-fed growth (EFG).⁸ We used surface remote oxygen plasma (ROP) processing,³² neutron irradiation,³³ and forming gas (FG) anneals to manipulate deep level defects, i.e., oxygen vacancies (V_O) and gallium vacancies (V_Ga), from the free semiconductor surface into the bulk, measuring their depth distribution with several nanometers or less resolution.

We used SPS^34–36 to measure the energy levels of defects within the Ga₂O₃ bandgap and DRCLS^37–39 to measure the luminescent transitions involving these defects and their bandgap—normalized intensity variations with excitation beam energy E_B and depth, i.e., defect profiles, of Ga₂O₃ as-received and after each processing treatment. For β-Ga₂O₃ grown by LPCVD, MBE, PLD, and EFG, DRCL and SPS spectra exhibit common optical features (supplementary material) corresponding to transitions into and out of multiple deep level defect states within the bandgap. Individual spectra exhibited multiple peak features that were deconvolved using Fityk (V.0.9.8) and MagicPlot (V.2.7.2) to reveal Gaussian peaks on a linear baseline across the spectral range detected. Figure 1 illustrates spectra of an LPCVD-grown sample after 1¾ h ROP treatment. Four major, broad features at 2.5, 3.0, 3.5, and 3.8 eV appear in each spectrum, likely corresponding to transitions between gap states and the conduction or valence band. The 4.5–4.7 eV shoulder represents the β-Ga₂O₃ conduction-to-valence band transition, consistent with the conclusions of previous studies.³ 

Figure 2 shows SPS features of MBE-grown β-Ga₂O₃. The contact potential difference (cpd) increase at 4.7–4.8 eV represents band flattening due to bandgap excitation, similar to the 4.5–4.7 eV bandgap feature in DRCL spectra. The four negative Δcpd slope changes at 2.4, 3.0, 3.5, and 4.0 eV signify photo-population of four defect states from the valence band. Two positive Δcpd slope changes at 2.0 and ∼3.8 eV indicate photo-depopulation of two deep level states to the conduction band.

We combined DRCLS with near-surface chemical processing and bulk neutron irradiation to characterize the physical nature of the deep level defects. We used Monte Carlo simulation software CASINO (V. 2.4.8.1) to model the depth dependence of electron-hole excitation in DRCLS.⁴⁰ The CASINO⁴¹ simulations yield the final relative densities of electrons at the end of their multiplication and cascade, where they finally lose energy to impact ionization and electron-hole pair creation. For beam energies E_B = 1, 2, 3, 4, and 5 keV, excitations have Bohr-Bethe maximum range R_B = 20, 40, 80, 120, and 160 nm, respectively. Staged exposure to ROP caused systematic spectral changes identified with filling of oxygen vacancies.^32,42 From deconvolved peak areas similar to those in Fig. 1 normalized to the ∼4.5–4.7 eV near band edge (NBE) area, Fig. 3(a) shows 3.5 eV depth profiles for the LPCVD Ga₂O₃ as-received, after 1¾ h ROP, and after 3¼ h ROP. Of all the samples after ROP treatment, the LPCVD samples showed the most prominent ROP effect. The area ratios for each profile are renormalized to the E_B = 5.0 keV bulk ratios since all ROP changes occur at shallower depths. For as-received Ga₂O₃, the 3.5 eV feature is nearly constant with depth, indicating a uniform 3.5 eV defect distribution. After 1¾ h ROP, the 3.5 eV depth profile at 80 nm decreases by 33% toward the surface. After 3¼ h total ROP, it decreases over 65% from 100 nm to 10 nm. In contrast, the analogous depth profiles for the 2.5 eV and 3.0 eV normalized defect areas show no effect after 1¾ h ROP treatment. The decreasing 3.5 eV surface content and deeper penetration suggest that this feature is related to β-Ga₂O₃ oxygen vacancies that are filled by activated oxygen atoms entering the lattice from the surface. Furthermore, this 3.5 eV area ratio increases with N₂ annealing (not shown), consistent with O outdiffusion and increased oxygen vacancy-related defect formation, similar to the case of other metal oxides.⁴² 

We used neutron irradiation to identify DRCLS peaks associated with Ga vacancies. Previously, electron paramagnetic resonance (EPR) studies showed that neutron irradiation produced V_Ga defects.⁴³ Figure 4 shows normalized DRCL spectra of a 1¾ h ROP treated LPCVD β-Ga₂O₃ epilayer after neutron irradiation at the Ohio State University Research Reactor (Columbus, OH) under nearly identical conditions to those reported previously. Similar to Fig. 1 spectra but on a linear scale, Fig. 4 shows the two most prominent defect features at 3.0 and 3.5 eV. Peaks at 2.5 and 3.8 eV are almost invisible due to linear scaling. The appearance of two V_Ga defect emissions may be related to two inequivalent Ga sites in the Ga₂O₃ unit cell, tetrahedrally and octahedrally coordinated Ga(I) and Ga(II), respectively.⁴³ After neutron irradiation, both the 2.5 eV and 3.0 eV peak intensities increase significantly when compared to the 3.5 eV peak—an average of 20% for the 2.5 eV peak and 30% for the 3.0 eV peak. These increases are evident for both 0.5 keV near-surface and 5.0 keV bulk excitation in Figs. 3(a) and 3(b), respectively. When normalized to the NBE emission area for E_B = 0.5–5 keV, depth profiles reveal corresponding 2.5 eV and 3.0 eV defect increases with neutron irradiation that extend continuously from the surface to the bulk, consistent with neutron penetration throughout the sample (see supplementary material). Neutron irradiation increases the 2.5 eV normalized defect by ∼90% consistent with a neutron irradiation increase of an E_C- ∼2 eV deep level transient spectroscopy (DLTS) level by ∼10¹⁶ cm⁻³.⁴⁴ Unlike these features, the 3.5 eV normalized defect decreases by >2× over the same depth range. A defect density calculated using transient SPS^35,45 yields a density for the E_V+ 2.4 eV transition of ∼5 × 10¹⁵ cm⁻³ for LPCVD Ga₂O₃ before irradiation, ∼5× higher than trap densities measured for a similar level in EFG Ga₂O₃ by DLTS.⁴⁴ Overall, these irradiation-induced 2.5 eV and 3.0 eV emission increases indicate that both are related to gallium vacancies in Ga₂O_3.

Hall effect measurements of the neutron irradiated Ga₂O₃ after forming gas anneals provided information on the donor/acceptor nature of the 2.5 and 3.0 eV features. Initially, both as-grown and irradiated LPCVD β-Ga₂O₃ samples were too resistive for accurate measurements. However, after each sample was annealed for 10 min at 750 °C in forming gas (FG), free carrier density n and mobility μ strongly increased for the irradiated sample but not for the as-grown one. A theoretical fit of Hall-effect n vs. T⁴⁶ after irradiation and anneal gave donor and acceptor concentrations N_D = 4.6 × 10¹⁷ and N_A = 2.7 × 10¹⁷ cm⁻³, respectively. By comparison, we have found typical background values of N_D and N_A in bulk Ga₂O₃ to be 2.5 × 10¹⁷ and 0.5 × 10¹⁷ cm⁻³, respectively. (Since our irradiated sample and the bulk Ga₂O₃ were grown by different techniques, the agreement between the increases in N_D − N_A is merely fortuitous. In general, n is not equal to N_D − N_A but instead depends on compensation ratio K = N_A/N_D and activation energy, which are not the same for these two samples.) Annealing above 500 °C improves Hall measurement accuracy and increases both n and μ, which can happen if acceptors such as V_Ga are preferentially destroyed or passivated during the annealing process. Reliable mobility values starting at 500 °C in 50 °C increments up to 750 °C were 3.5, 5.0, 6.5, 10.1, 11.7, and 11.3 cm²/V s. Mobility values for Ga₂O₃ vary greatly with growth and processing treatments, but literature values of mobility for neutron irradiated Ga₂O₃ are not readily available.⁴⁷ Interestingly, DRCL spectra after irradiation and FG anneal exhibit a peak increase at 2.5 eV (possibly passivated Ga vacancies) but a decrease at 3.0 eV (possibly unpassivated Ga vacancies).

Figure 2 photo-stimulated population and depopulation SPS threshold energies and matching Fig. 1 CL transition energies correspond to specific energy levels in the Ga₂O₃ bandgap. Figure 5(left) shows an energy level diagram with these SPS and CL optical transitions. Horizontal black lines indicate energy levels corresponding to 2.0 eV photo-depopulation (blue) and 2.4, 3.0, and 3.5 eV photo-population (red) SPS transitions. The 2.0 eV photo-depopulation transition and 3.0 eV population transition could be complementary, i.e., their respective energies sum to ∼4.9 eV, the bandgap. Darker red arrows signify CLS transitions, which match the energy levels determined by SPS almost exactly as indicated by dotted lines. Without a clear energy level assignment, the 3.8 eV CL feature in Fig. 1 could correspond to a transition 3.8 eV below the conduction band or 3.8 eV above the valence band, the latter close to a defect level reported using deep level optical spectroscopy (DLOS).⁴⁸ 

These levels can be compared in Fig. 5(right) with those predicted theoretically for transition energies between above mid-gap charge states calculated for V_O, V_Ga, and V_Ga-H. Also shown are deep level gap states measured experimentally by deep level transient/optical spectroscopies (DLTS/DLOS).⁴⁸ Energy levels are calculated for V_O⁺ in the geometry of the V_O⁰ configuration for three different lattice configurations (VdW)⁴⁹ using density functional theory (DFT) with hybrid functionals that provide the β-Ga₂O₃ bandgap observed experimentally.⁵⁰ While unambiguous identification of a specific configuration requires additional microscopic evidence and notwithstanding possible screening effects,⁵¹ Fig. 5 shows that the 3.5 eV SPS photo-population transition and the 3.5 eV CLS transition agree closely with the 3.52 eV V_O⁺ (III) energy transition level predicted theoretically.⁴⁹ While not definitive, the V_O⁺ (III) configuration is more thermodynamically stable that its V_O⁺(I) and V_O⁺(II) counterparts for intermediate-doped Ga₂O₃.

Also shown in Fig. 5(right) are energy levels calculated for Ga vacancies V_Ga and their hydrogenated V_Ga-H counterparts.¹⁴ None of these transition energies overlap the 3.5 eV V_O-related emission. Instead, the 2.5 and 3.0 eV spectral features appear to correlate with the 2.55 eV V_Ga-H(II) and either the 2.9 eV V_Ga-H(I) or 3.0 eV V_Ga(II), the latter being thermodynamically less stable than its hydrogenated counterpart.¹⁴ These alignments indicate that the 2.5 eV and 3.0 eV features in Fig. 1 are consistent with V_Ga-related defects while the 3.5 eV peak is due to V_O-related defects. Again, unambiguous identification of a specific configuration requires additional microscopic evidence. Overall, all five DRCLS features correspond to energy levels measured by SPS, while nearly all SPS transitions correlate with energy levels predicted theoretically or measured by DLTS/DLOS.

In summary, we used DRCLS and SPS to measure the effects of near-surface plasma processing and neutron irradiation on native point defects in LPCVD-grown β-Ga₂O₃. The near-surface sensitivity and depth resolution of these optical techniques enabled us to identify spectral changes associated with removing or creating these defects, leading to spectral correlations with one V_O- and two V_Ga-related energy levels in the β-Ga₂O₃ bandgap. In addition, these results suggest processing methods to reduce V_O-related densities by ROP processing and V_Ga-related densities by FG anneals. Defect reductions can be monitored spatially in three dimensions by DRCLS as processing parameters are varied systematically. The combined near-surface defect detection and processing of Ga₂O₃ suggests an avenue for characterizing and reducing native point defects in β-Ga₂O₃ and other semiconductors.

See supplementary material for representative measurements across growth types and pre- and post-irradiation depth profiles.

L.J.B. and H.G. acknowledge AFOSR Grant No. FA9550-18-1-0066 with thanks to Tamura Corp. H.Z. acknowledges NSF Grant No. DMR-1755479. DCL's work supported in part by AFOSR Award No. FA9550-RY18COR800.