Deep level defects in low-pressure chemical vapor deposition grown (010) β-Ga₂O₃

Interest in β-phase gallium oxide (β-Ga₂O₃) has been growing tremendously in recent years due to its great potential for power devices and RF electronics. The major factors causing this surge are its ultra-wide bandgap (∼4.6–4.8 eV),^1–3 large breakdown fields,^4,5 ease of n-type doping,^6–8 availability of melt-grown substrates,^9–11 high tolerance to radiation damage,^12,13 and high predicted device figures of merit compared with GaN and SiC.^5,14 The epitaxial growth efforts have mainly focused on molecular beam epitaxy (MBE),^15–18 metal–organic chemical vapor deposition (MOCVD),^19–21 and halide vapor phase epitaxy (HVPE),^22–24 which have yielded impressive performance metrics resulting in state-of-the-art RF and power devices. Among these growth methods, a noteworthy MOCVD grown epitaxial layer with an electron mobility of 194 cm²/V s at room temperature and background doping of 7.7 × 10¹⁴ cm⁻³ was reported by Feng et al.,²⁵ and similar results have been reported by other groups using MOCVD, HVPE, and MBE.^20,26–29 Recently, low-pressure chemical vapor deposition (LPCVD) films have yielded room-temperature electron mobilities as high as 156 cm²/V s, a wide range of controllable doping, with growth rates up to 40 μm/h, indicating this growth method could help achieve high throughput epitaxial growth with very high quality.^30–33 This high electronic quality coupled with the fast growth rates and subsequently thick β-Ga₂O₃ layers possible with the LPCVD technique creates a great opportunity for future multi-kV vertical devices that need thick, low-doped drift layers. In fact, Schottky barrier diodes (SBD) with breakdown voltages of up to 435 V have been reported using low doped LPCVD grown β-Ga₂O₃.^34,35

However, in comparison with other epitaxial growth methods, little is known about deep level defects, their physical sources, and their relative impact on trapping and transport properties of β-Ga₂O₃ grown by the LPCVD growth technique. Feng et al. reported in their transport studies that the low-temperature mobilities for LPCVD-grown material are limited by scattering, which points to the presence of high acceptor concentrations.³³ It was also reported that unintentionally doped (UID) LPCVD epilayers were insulating or highly resistive, implying either strong compensation of background donors by acceptors or low concentration of background donors during UID growth.^31,33 Such observations provide direct motivation to explore the presence and sources of deep states in β-Ga₂O₃ grown by LPCVD, which is the focus of this effort. In this work, deep-level optical spectroscopy (DLOS) and deep-level transient spectroscopy (DLTS) are employed to quantify the distribution of bandgap states and their individual concentrations through the entire ∼4.8 eV bandgap of LPCVD-grown β-Ga₂O₃. The LPCVD growth conditions used here are nominally identical to those that yielded outstanding transport properties mentioned above.^30–34 

A study was devised to investigate the presence of defect states within the entire bandgap of β-Ga₂O₃ grown by LPCVD. Two samples were grown on Sn-doped (010) β-(3–5 × 10¹⁸ cm⁻³) conductive substrates at a temperature of 1050 °C in an in-house built horizontal LPCVD system utilizing metallic gallium and O₂ gas as growth sources.^31,33 Prior to LPCVD growth, the β-Ga₂O₃ substrate was dipped in HF followed by solvent cleaning using acetone, isopropyl alcohol (IPA), and de-ionized (DI) water. The films were grown at ∼20 μm/h growth rates, and the thickness was ∼5 μm. The n-type Si-doping was achieved using a constant flow of SiCl₄ balanced with argon. Earlier reports on the LPCVD growth technique by Feng et al.³³ revealed that high growth rates could possibly hamper the introduction of Si. To compensate for this effect, two samples were grown with varying silicon tetrachloride (SiCl₄) flow rates set to targets of 0.2 and 0.5 SCCM; named samples A and B. The oxygen flow rate was maintained at 30 SCCM, and the argon flow rate for dilution was at 200 SCCM. The chamber pressure was maintained at 1–2 Torr during the entire growth period. All other growth conditions are discussed in detail by Feng et al.³³ 

After LPCVD growth, both samples A and B were cleaned with acetone, IPA, and DI water. Semi-transparent 8 nm Ni Schottky contacts of 290 × 290 μm² were fabricated on the LPCVD grown epilayer and an ohmic stack of Ti/Al/Ni/Au on the back side of the conducting substrate to facilitate defect spectroscopy. A mesa etch is typically used to reach the highly doped lateral conduction layer for topside ohmic contacts when using insulating substrates. Here, the highly doped substrates made it easier to deposit a large area of backside ohmic contact. The measured capacitance did not show any dispersion with frequencies and a near unity ideality factor from current density–voltage (J–V) characteristics [Fig. 1(a)] indicating high quality contacts. The devices are separated by ∼30 μm, and the depletion depth is <0.1 μm, so there is no interaction between different diodes on the same sample. The device schematic is shown in the inset of Fig. 1(a). Following device fabrication, the diodes were tested using capacitance–voltage (C–V), conductance–voltage (G–V), J–V, and internal photoemission (IPE) for Schottky barrier heights. The extracted barrier height from IPE was 1.4 eV, consistent with what we have observed for other growth techniques like MOCVD and MBE.^21,36–40

The fabricated samples yielded dozens of high-quality Schottky diodes with similar J–V characteristics having ideality factors of 1.1 ± 0.05 and leakage currents below 10⁻⁷ A/cm². A representative set of J–V data is shown in Fig. 1(a). The current saturation at a higher forward bias is due to the back contact on the substrate. Uniform net doping [N_D-N_A] concentrations of 2 × 10¹⁶ and 8 × 10¹⁶ cm⁻³ were measured by C–V from the samples A and B, respectively, as shown in Fig. 1(b). This trend is tracked by the measured Si concentration from SIMS that is shown in the same figure. The net doping extracted from C–V suggests the presence of compensation in each sample since Si donor ionization energy is 15–40 meV^6,20,29,31 and should be fully ionized at room temperature.

After characterizing the essential diode characteristics, an in-depth assessment of trap states commenced. DLTS measurements can typically determine the thermal activation energies of traps within ∼1 eV of the conduction band for our standard measurement temperature range of 10–400 K. The DLTS measurement conditions used here consisted of applying a “fill pulse” of 0 V to the Schottky diode for a duration of 10 ms to fill traps, followed by the application of a “quiescent state” reverse bias at −2 V for 2 s, during which capacitance transients resulting from thermal emission of electrons from traps could be digitally obtained. The capacitance transients were recorded from 10 to 400 K in steps of 0.1 K and analyzed using a conventional double boxcar method. The full details of the measurement can be found elsewhere.^21,40,41 The two LPCVD samples grown at different flow rates each exhibited a single trap in their DLTS spectra shown in Fig. 2(a), with an activation energy of E_C-0.8 eV as determined by the Arrhenius plot in Fig. 2(b). The trap concentration for each sample was accurately calculated using the “lambda” correction to account for the volume of the depletion region in which the occupancy of the specific trap level is modulated by the DLTS biasing,⁴² revealing very similar E_C-0.8 eV trap concentrations of 8 × 10¹⁴ and 9 × 10¹⁴ cm⁻³ for the samples A and B, respectively. Figure 2(b) compares the Arrhenius behavior of the DLTS measured LPCVD trap with prior work that showed this state to be associated with Fe impurities configured as a Fe_Ga substitutional defect, often referred to as the E2 defect, which has been widely reported.^{24,36,43–46}

The remaining portion of the bandgap was investigated using DLOS. DLOS is based on direct photoemission of trapped carriers, which enables the characterization of traps that are too deep in the bandgap for the thermal emission based DLTS technique to observe.^41,47 Here, the trapped electrons in deeper states are optically emitted by a monochromatized light source used to supply incident photons with controlled energies between 0.5 and 5 eV in 0.02 eV steps of known intensity. Prior to photoionization at each energy, the traps are filled by electrons with a 0 V fill pulse for 10 s, followed by a thermal settling time after which the sample is exposed to the monochromatic light for 300 s, during which the resultant capacitance transient is measured. From the photo-capacitance transient, optical cross-sections of any traps present can be obtained as a function of photon energy. Analysis of the photo-capacitance transient by fitting to the Passler model⁴⁸ enables the extraction of both the equilibrium energy level and the Franck–Condon (D_FC) energy due to possible local lattice relaxation associated with each detected defect state. Concentrations of each state are obtained by measuring the steady state photo-capacitance associated with each defect state, and by using the lighted C–V (LCV) method, which we have found to be valuable for analyzing concentrations of defect states with small optical cross sections that can limit the ability to fully saturate that state.^12,36

Figure 3 shows the measured optical cross sections and their fits to the Passler model for samples A and B, noting that both samples revealed nearly identical spectra.⁴⁸ Table I compiles the relevant energies for each defect from these fits, along with trap concentrations that were obtained using LCV measurements with a 3-h photo-saturation for each state. Comparing the concentrations measured from the LCV, it is evident that DLOS detected states did not show a significant dependence on the SiCl₄ flow rate, similar to the DLTS findings.

DLOS and lighted CV measurements have helped identify the position and concentration of these acceptor-like states in the middle and lower parts of the bandgap. Several of these states have been previously observed for other growth methods, which warrants discussion. For instance, the E_C-2.0 eV state has been reported in β-Ga₂O₃ grown by both MOCVD and MBE and has been shown to be a dominant source of carrier compensation.^13,21,37,38 In studies on the effects of high energy proton and neutron irradiation, the concentration of this defect strongly correlated with radiation fluence, which implies its physical source is likely to be of intrinsic character.^12,13,49 Extensive density functional theory (DFT) studies have predicted that a defect state at E_C-2.0 eV associated with the $2VGa1−Gai$ defect complex is likely to form, corroborating the post-radiation DLOS results, which was further supported by observations made by high resolution transmission electron microscopy.^{7,13,21,37,38,50,51} Future studies evaluating how this gallium vacancy-related defect state is influenced by LPCVD growth conditions are warranted, and this is presently under investigation. The E_C-4.4 eV state seen here has also been universally seen for all β-Ga₂O₃ growth methods reported to date. Its source remains an area of active investigation, and while there has been some speculation that hole self-trapping could play a role,⁵² far more work is required to better understand its source.^21,38

The most significant difference observed for LPCVD-grown β-Ga₂O₃ compared with DLOS studies of β-Ga₂O₃ grown by other growth methods is the energy level at E_C-3.6 eV (with a D_FC of 0.6 eV). This is the dominant trap in the LPCVD defect spectrum and, to the best of our knowledge, has not been previously reported. Its position in the bandgap implies it could behave as a strong acceptor-like defect. SIMS measurements were done on both samples A and B to explore if there is a connection with common impurities such as Mg, N, C, or Cl; in all cases, the concentrations of these impurities were lower than the measured E_C-3.6 eV trap concentration, indicating that its source is either a more unusual impurity or related to an intrinsic defect source. Table I shows that its concentration is not significantly affected by the small range of SiCl₄ flow rates used in this study of traps in LPCVD material; however, there have been reports of DFT calculations that predict oxygen vacancies (⁠$VO+[III])$ or hydrogenated gallium vacancies (V_Ga − nH) could create energy levels near E_C-3.6 eV.^7,51 Given its important for LPCVD trap distributions in the bandgap, work is ongoing to investigate the source of this dominant LPCVD defect level in more detail.

For a point of reference, Fig. 4 presents a comparison of defect state distributions across the entire bandgap for the LPCVD-grown β-Ga₂O₃ samples with different SiCl₄ flow rates and our prior MOCVD-grown material.²¹ Energy levels come from the optical cross section fits and concentrations are from LCV for each state, except for the DLTS-detected state at E_C-0.8 eV. In addition, admittance spectroscopy (AS) measurements made on the LPCVD grown β-Ga₂O₃ samples did not reveal any additional states, which is unlike the prior MOCVD work, where AS revealed relatively high concentrations of a trap at E_C-0.12 eV. There is a clear different in defect distributions between LPCVD and MOCVD material. Whereas the E_C-3.6 eV state is new, the V_Ga-related defect at E_C-2.0 eV is lower in concentration and other than a small concentration of the Fe_Ga [(E2) trap, the upper part of the bandgap is quite clean. Now, considering that the trap distributions and total trap concentrations in both LPCVD samples are very similar 7.8 × 10¹⁵ and 8.8 × 10¹⁵ cm⁻³ for the samples A and B, respectively], it is unlikely that the strong reduction in carrier concentration shown in Fig. 1(b) is influenced by differences in carrier compensation. This supports the SIMS data in Fig. 1(b), showing that the high growth rate LPCVD provides very good control of both intentional Si doping and the desired carrier concentration in the range reported in this work, and background doping is not appreciably influenced by inadvertent defect incorporation. This is a key finding for future applications of LPCVD grown β-Ga₂O₃.

In conclusion, an initial detailed investigation of bandgap states in LPCVD-grown β-Ga₂O₃ layers reveals total trap concentrations are on par with state-of-the-art MOCVD grown material.^21,38 This is consistent with earlier reports that identical LPCVD growth conditions result in outstanding electron transport properties^33,53 and promising vertical high voltage Schottky barrier diodes.^34,35 Specifically, the LPCVD-grown material reveals a low concentration of residual Fe impurities that give rise to the well-known E_C-0.8 eV (E2) trap related to Fe_Ga defects, and two other commonly reported defect states at E_C-2.0 eV and E_C-4.4. eV. The primary difference in the trap spectrum of LPCVD-grown β-Ga₂O₃ is a new trap at E_C-3.6 eV that is the dominant trap within the bandgap of LPCVD-grown β-Ga₂O₃, with concentrations in the mid-10¹⁵ cm⁻³ range. Its position in the bandgap implies it could be a source for significant carrier compensation, which is something to be explored in future work. Currently, new efforts based on these findings are ongoing to investigate how this defect state depends on growth conditions and to identify its physical source. The findings presented here support the great promise of the LPCVD growth method in delivering thick, low defect background β-Ga₂O₃ epitaxial layers that are of interest for future multi-kV vertical devices and very sensitive radiation detectors.

The authors acknowledge the funding support from the Air Force Office of Scientific Research No. FA9550-18-1-0479 (Ali Sayir, Program Manager) and from the US Air Force Radiation Effects Center of Excellence, Grant No. FA9550-22-1-0012. This material is based on research sponsored by the Air Force Research Laboratory and Strategic Council for Higher Education under agreement FA8650-19-2-9300 and the Ohio Department of Higher Education and Strategic Council for Higher Education under Ohio House Bill 49 of the 132nd General Assembly. The author acknowledges the funding support from Department of Energy/National Nuclear Security Administration under Award Number(s) DE-NA0003921. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the funding provider.