In this work, conventional deep-level transient spectroscopy (DLTS) and high-resolution Laplace-DLTS (L-DLTS) have been used to characterize deep-level traps in (010) β-Ga2O3 epilayers grown by metal organic chemical vapor deposition on native Sn-doped substrates. Two types of epilayers have been studied, one doped with silicon during growth to about 1.5 × 1017 cm−3 and the other type was unintentionally doped (UID). Electrical measurements were conducted on Au and Pt Schottky barrier diodes. In the Si-doped samples, only one electron trap with emission activation energy of 0.42 eV (E0.42) and concentration of (6–8) × 1013 cm−3 has been detected. In the UID samples, in addition to the E0.42 trap, two other traps with activation energies for electron emission of 0.10 eV (E0.10) and 0.53 eV (E0.53) have been observed. Dependencies of electron emission rate (eem) on the electric field (E) as well as concentration-depth profiles {NT(W)} have been measured and analyzed for the E0.10 and E0.42 traps. The eem(E) dependence for the E0.10 trap is characteristic for a donor energy level, while that for the E0.42 trap indicates an acceptor level. The NT(W) dependencies show non-uniform spatial distributions of both the E0.10 and E0.42 traps in the UID samples, with the concentration of the E0.10 trap dropping from about 1 × 1015 cm−3 at 1.5 μm from the surface to about 2 × 1013 cm−3 at 0.5 μm, which indicates out-diffusion from the substrate or interface into the epilayer as a likely source. The results obtained are compared with the literature, and possible origins of the detected traps are discussed.
I. INTRODUCTION
Thermally stable beta phase gallium oxide (β-Ga2O3) is a polymorph of Ga2O3 that has gained a significant amount of interest in recent years due to its potential in deep ultraviolet (DUV) optoelectronic devices1–3 and high-power electronics.4,5 The ultra-wide bandgap (UWBG) of ∼4.7–4.9 eV6–8 results in large breakdown fields of about 8 MV/cm alongside a high Baliga figure of merit (BFOM)9 in comparison to traditional wide bandgap semiconductors such as gallium nitride (GaN) and silicon carbide (SiC).10 The ability to produce low-cost, large area native substrates using melt-growth techniques11 enables homoepitaxial deposition of β-Ga2O3 device layers, thereby facilitating the production of high-quality material with low threading dislocation densities (TDDs).12,13
To effectively control and develop β-Ga2O3 devices, the origin and electronic properties of defects present in the grown material, especially in the active layers must be studied and understood. In the last decade, there have been significant efforts to identify such electrically active defects, with Irmscher et al.14 publishing the first deep level transient spectroscopy (DLTS) study on (100) orientated β-Ga2O3. Since then, several studies have reported the presence of many traps in the Ga2O3 bandgap.15–25 The obtained results have been reviewed recently by Wang et al.19 and Labed et al.21 Among the reported traps, those labeled E1, E9, and E10 traps in Ref. 19 are of interest in our work, since in the β-Ga2O3 samples that we have studied, three traps with electronic signatures close to these have been observed.
The trap with an energy level located near 0.6 eV from the conduction band edge (Ec–0.6 eV) is commonly referred to as E1.14–25 The E1 DLTS signal has been detected in bulk crystals grown by the Czochralski (CZ) method14 and edge-defined film-fed growth (EFG),15,22 in addition to homoepitaxial epilayers grown via molecular beam epitaxy (MBE) and hydride vapor phase epitaxy (HVPE).16,17,23,24 Polyakov et al.20 have analyzed the electric field dependence of electron emission rates for E1 and argued that the center is a deep donor, with SiGa1-H and SnGa2-H defects being proposed as likely candidates. This suggestion was supported by theoretical modeling results.20 The involvement of hydrogen is corroborated in a study by Langørgen et al.,25 which determined that the concentration of the E1 level increases with the introduction of hydrogen during annealing processes. Further hybrid-functional calculations25,26 revealed other H-related defects, involving singly hydrogenated Ga–O divacancies , which have charge-state transition levels that are consistent with measured activation energies for electron emission from E1. However, considerable ambiguity remains on the specific defect responsible for this emission. An earlier study by Ingebrigtsen et al.22 showed that E1 is not generated by proton irradiation, suggesting but not confirming that E1 is not related to native lattice defects. This theme is revisited in a review by Labed et al. who relate E1 to Si or metallic impurities.21 Thermodynamic transition level calculations16 highlighted a Fe3+/Fe2+ level at Ec–0.59 eV for iron situated on the tetrahedral gallium site (FeGa1). Furthermore, a strong correlation in the concentration of a defect with an energy level at Ec–0.55 eV with the presence of Cr in Ga2O3 samples has been reported.27 However, the Fe- and Cr-related energy levels in the upper half of the Ga2O3 gap are thought to be deep acceptor levels,16,27 which is not consistent with the results on the donor nature of the E1 trap.20 In summary, the inconsistencies in the observed experimental properties suggest that the E1 label has been attached to different traps with similar electron emission properties.
The so-called E9 and E10 traps, with trap energy levels at approximately Ec–0.40 and −0.12 eV below the conduction band edge, respectively, are less commonly reported in DLTS studies of β-Ga2O3.19 E10 has only been observed experimentally using admittance spectroscopy (AS)28 and Hall effect measurements.29,30 Neal et al.29 have argued that incomplete ionization of this unintentional donor level increases the specific on-resistance, and decreases the breakdown voltage of β-Ga2O3-based devices. Supplementary density functional theory (DFT) calculations29 suggest a host of potential defects could be responsible for this level, with Si on the Ga(II) site proposed as a strong candidate, which is supported by scanning tunneling microscopy (STM) results31 that demonstrate Si atoms occupy both Ga lattice sites. This association with Si was supported by Ghadi et al.28 as a Ec–0.12 eV state was detected using AS in the Si-doped MOCVD material. Furthermore, DFT calculations also show that antisites, interstitials, and various extrinsic impurities (H, Si, Ge, Sn, F, Cl) can possess relatively shallow donor levels in Ga2O3,32 ruling out simple oxygen and gallium vacancies as possible sources of E10, as they are deep donors and acceptors, respectively, according to the results of ab initio calculations.32,33 E9 has been reported in DLTS studies within as-grown metal–organic chemical vapor deposition (MOCVD)28,34 and plasma-assisted MBE (PAMBE)17 Ga2O3 epilayers on (010) orientated Sn-doped substrates. Ghadi et al.28 deduced the E9 DLTS signal was consistent with a simple point defect. Subsequent work by McGlone et al.34 investigated the effect of proton irradiation on the E9 trap and determined that its concentration is insensitive to irradiation and the defect is, therefore, likely to be extrinsic in nature.
In this work, conventional deep-level transient spectroscopy (DLTS)35,36 and high-resolution Laplace-DLTS36,37 (L-DLTS) is used to characterize deep-level traps in MOCVD-grown β-Ga2O3 epilayers deposited on native Sn-doped substrates. This analysis is applied to Au Schottky diodes on both Si-doped and unintentionally doped (UID) epilayers. DLTS has been applied to investigate the E10 level, which is partially responsible for the unintentional doping of β-Ga2O3. Moreover, dependencies of electron emission rate (eem) on the electric field (E) have been measured and analyzed, as well as the concentration-depth profiles for both the E10 and E9 traps. Results show a strong eem(E) dependence for the E10 trap, indicating its donor nature and that the E9 level exhibits the characteristics of a deep acceptor.
II. EXPERIMENTAL DETAILS
An Agnitron Agilis 100 MOCVD reactor was used to grow ∼2 μm thick epilayers of β-Ga2O3 on commercially available Sn-doped (010) substrates from NCT. Thickness was measured by reflectometry on sapphire satellites and a cross-section imaging of a cleaved sample. Growth temperature was 840 °C with TEGa (triethylgallium) and O2 gas precursors. To lightly dope the epilayer (estimated ∼1.5 × 1017 cm−3), silane was used as the impurity source. To facilitate all electrical measurements, including DLTS, capacitance–voltage (C–V) and current–voltage (I–V) analysis, vertical Schottky barrier diodes (SBDs) were fabricated. Prior to any metal deposition, the samples were cleaned using a standard solvent cleaning procedure of acetone, isopropyl alcohol, and de-ionized (DI) water. Following pre-treatment with organic solvents, the samples were cleaned in a 10% HCl solution then soaked in H2O2, held at 85 °C. Each step was followed by a DI water rinse. A backside Ohmic contact was prepared via plasma sputtering of Ti and thermal evaporation of Au to form a Ti/Au (30/100 nm) metal stack, which was subsequently annealed at 450 °C in a N2 atmosphere for 5 min. Finally, front side Schottky contacts were deposited via plasma sputtering or thermal evaporation to produce arrays of Pt or Au (200 nm) circular diodes, each with a diameter of 0.45 mm. These showed almost identical characteristics and all the measurements reported here are on Au SBDs, which had slightly lower reverse leakage.
III. EXPERIMENTAL RESULTS
Figures 1 and 2 show the results of the initial static I–V and C–V measurements conducted on the Au Schottky diodes for both wafers. The current density–voltage (J–V) characteristics for each sample, illustrated in Fig. 1, showcase high quality Schottky diodes, with a reverse leakage current approaching the detection limit of the HP 4140A pA meter. Capacitance–voltage measurements were performed from −12.5 up to 0 V at 295 K on the diodes from each wafer, and the results of the subsequent analysis are shown in Fig. 2. C–V measurements were conducted at 1 MHz. The phase angle over the whole temperature range used was between 87° and 89°, verifying that the resistive component was insignificant. All calculations of depletion depths and uncompensated donor concentrations have been carried out using a static dielectric constant (ɛ) value of 10.9.38 For the UID sample (blue curve), the depth profile of the uncompensated donor concentration [Nd (W)] shown in Fig. 2(b) is non-uniform, which correlates with a non-linear dependence on the applied reverse bias, as illustrated in Fig. 2(a). In the case of the Si-doped sample, the measured donor concentration is uniform throughout the probed region of the epilayer, suggesting there is a homogeneous distribution of Si-dopant atoms within the crystal.
Figure 3 shows the conventional DLTS spectra for both samples, across a temperature range from ∼40 up to ∼400 K. A constant reverse bias (Ub) of −2.0 V was applied with a fixed pulse bias (Up) of −0.1 V. This enables the depletion region to be probed from 0.65 to 0.85 μm from the surface for a diode on the UID sample and from 0.12 to 0.2 μm for the Si-doped sample. The spectra were recorded with a rate window (en) of 20 s−1 and a filling pulse length (tp) of 100 μs. Each DLTS peak is labeled with its associated activation energy for electron emission relative to the conduction band (Eem) written in the subscript.
The Y-axis in Fig. 3 represents the 2 × (ΔC/Cb) × Nd × f values, where ΔC is the amplitude of the capacitance transient, Cb is the capacitance measured at the applied reverse bias, Nd is the concentration of uncompensated donors, and f is the correction factor, which takes into account the depletion depths at bias and pulse voltages.36,39 The so-calculated Y values at the peak maxima in the spectra are equal to the average trap concentrations in the probed regions. Both DLTS spectra reveal a sharp peak (E0.42) with its maximum at approximately 190 K, which is consistent with the peaks observed in the literature17,28,34, referred to as E9.19 In the sample with the UID β-Ga2O3 epilayer, two additional peaks were detected. DLTS signals are observed at about 50, 190, and 270 K, with emission activation energies of 0.10 eV (E10), 0.42 eV (E9), and 0.53 eV (E1), respectively.
To accurately extract the electronic signatures of these traps, high-resolution L-DLTS was applied to produce emission spectra at fixed temperatures, which were stable within ±10 mK. Figure 4 shows the L-DLTS spectra recorded for the E9 trap observed in the wafer with the Si-doped epilayer. The spectra can then be used to generate an Arrhenius plot to accurately determine the activation energy for electron emission from a trap as well as the apparent capture cross section (σapp). The occurrence of relatively sharp well-separated single peaks in the L-DLTS spectra measured at each temperature indicates a few things:36,37 (i) the signal is related to electron emission from a point defect with a well-defined energy level, (ii) the inhomogeneous strain within the sample is negligible, and (iii) the effect of the electrical field on electron emission is not strong.
The electric field dependencies for the E9 and E10 traps were studied using “double” L-DLTS and conventional DLTS, respectively.36,37,41 The most common version of this technique requires two filling biases, with a fixed reverse bias, and the difference between two transient signals recorded for different pulse voltages is being analyzed. Hence, restricting the scan to a narrow window in the depletion region, the electric field varies with varying applied pulse voltages. It is found that the eem(E) dependence of the E9 trap is relatively weak in comparison to that of the E10 trap. In fact, the significant eem(E) dependence of the E10 level resulted in strong broadening of the emission signal in the L-DLTS spectra, thereby preventing the application of “double” L-LDTS. Figure 7 shows the eem(E) dependence of E10 measured using the conventional “double” DLTS technique. A constant Ub of −10.0 V was applied and three W ranges with different E values were probed by setting three different double filling pulse biases (Up1–Up2) with the difference between pulse voltages fixed at 2 V. A rate window of 200 s−1 was utilized with a filling pulse length of 100 μs. The analysis shows a significant shift in the DLTS spectra to lower temperatures, with strong broadening of the measured signal as the electric field increases. This strong eem(E) dependence is typical of a donor level,42 which is consistent with the measured σapp derived from the Arrhenius analysis for E10.
Figure 8 shows the relatively weak eem(E) dependence for the E9 trap measured with the use of double L-DLTS. The linear dependence of ln(eem) on the square of the electric field (E2) is consistent with the phonon-assisted tunneling mechanism, which is characteristic for electron emission from negatively charged defects in n-type semiconductors.43 This implies the defect responsible for the E9 level is an acceptor.
IV. DISCUSSION
First of all, it should be mentioned that only one deep-level electron trap, E0.42, with a relatively low concentration, NT < 0.001ND has been observed in the upper quarter of the gap in Si-doped MOCVD grown β-Ga2O3 epilayers, which we have studied. This result is consistent with previously reported results on deep-level defects in Ga2O3 epilayers, which were grown by the MOCVD method.27,33 Furthermore, our L-DLTS results on the E0.42 trap (Fig. 4) indicate a negligible level of inhomogeneous strain in the MOCVD epilayers. These findings confirm the suggested high potential of MOCVD for manufacturing high performance β-Ga2O3 devices. In this section, possible origins of the detected deep-level traps are considered.
A. E0.53 (E1) trap
The E0.53 trap has been clearly detected only in the UID samples, which we have studied. The concentration of the trap was small, <1013 cm−3, in these samples. Because of its low concentration, we could not carry out a careful investigation of the concentration-depth profile and eem(E) dependence for the trap. The electronic signatures of the E0.53 trap are close to those for the E1 trap, which is frequently reported in the literature on deep-level traps in β-Ga2O3.14–25 Possible origins of the E1 trap have been discussed in the Introduction.
B. E0.42 (E9) trap
The electronic signatures and concentrations of the E0.42 trap in the Ga2O3 samples, which we have studied, are close to those for the E9 trap, which was detected in β-Ga2O3 epilayers grown by the MOCVD and PA-MBE techniques.17,27,33 It has been argued in the literature that the source of the E9 trap is related to a single point defect of extrinsic nature.27,33 Our results are consistent with these arguments. Furthermore, the results on the electric-field dependence of electron emission from the E0.42(E9) trap (Fig. 7) indicate that the trap is likely to be related to an acceptor level, however, the value of the apparent capture cross section, 5.5 × 10−14 cm2, is very large for an acceptor level. We have tried to carry out direct capture cross section measurements for the E0.42 trap, which have been only partially successful, as we could not observe a substantial decrease in the E0.42 signal magnitude even upon the application of the shortest possible filling pulses, 40 ns, using our equipment. From this experiment, it was possible to deduce only a lower limit of the capture cross section, which is about 1 × 10−16 cm2. Concentration of the E0.42 trap is found to be a few times 1013 cm−3 in both the Si-doped and UID samples with a smooth decrease toward the surface in the UID sample. This decrease can be interpreted as an indication of potential out-diffusion of impurity atoms, which are responsible for the E0.42 trap, from the Sn-doped substrate into the epilayer.
C. E0.10 (E10) trap
The E0.10 trap has only been detected in the UID samples. The activation energy for electron emission from the trap is close to those for the E10 trap observed previously in β-Ga2O3 epilayers.27–29 It should be mentioned that there was almost no freezing out of carriers in the diodes on both the Si-doped and UID samples down to 35 K, which was the lowest temperature used. The results on the electric-field dependence of electron emission from the E0.10 (E10) trap (Fig. 7) indicate that the trap is a donor level, which is consistent with the large value of the apparent capture cross section, 1.25 × 10−13 cm2. The NT(W) dependence for the E0.10 trap shows non-uniform spatial distribution of the trap in the UID samples (Fig. 6), with the concentration of the trap dropping from about 1 × 1015 cm−3 at 1.4 μm from the surface to about 2 × 1013 cm−3 at 0.5 μm. A comparison of the ND(W) and NT(W) dependencies (Figs. 2 and 6) indicates that the E0.10 trap is partially responsible for the free electron concentration in the UID sample at room temperature. It can be suggested that impurity atoms from the substrate or interface are transported into the epilayer either by diffusion or carried forward on the growth front. This is a possible source of the E0.10 trap. Considering the growth temperatures, we have analyzed the spatial profile of E10 from this work and found that it is not consistent with the results for Sn diffusion into β-Ga2O3 epilayers from the Sn-doped substrate as reported by Frodason et al.,44 implying a different species is responsible if diffusion is the mechanism. Such an extended NT(W) dependence for E10 (Fig. 6) indicates the involvement of a defect with a significantly higher diffusivity than that derived for Sn. Alternatively, H-related defects could be candidates for the E10 trap. They are shallow donors while being sufficiently mobile,45,46 and they can be easily incorporated provided the substrate is grown in a H-rich environment. The UID layer is 2 μm thick; by extrapolating the E10 profile shown in Fig. 6 to the epi-substrate interface indicates a source concentration of E10 of ∼1016 cm−3. Accumulation of Si at the epi-substrate interface is not unusual during MOCVD growth of gallium oxide47,48 and SIMS measurements on samples grown under similar conditions in the same reactor as our samples show an accumulation of Si at the interface in excess of 1018 atoms cm−3. If Si is carried forward during growth and incorporated in the growing layer, it is possible that E10 originates from a fraction of the Si occupying the thermodynamically less preferred octahedral Ga(II) site as discussed in the Introduction. However, when we look at the DLTS spectrum shown in Fig. 3 of the sample intentionally doped with Si, there is no detectable concentration of E10 in the layer, indicating all the Si atoms are on the thermodynamically preferred tetrahedral Ga(I) site. The reasons for this are not at all clear, so the suggested assignment of the structural and chemical identity of E10 is to a large degree speculation. Further work and analysis are required to identify the defect responsible for the E10 trap.
V. CONCLUSIONS
In conclusion, we have analyzed deep levels present in Si-doped and UID β-Ga2O3 epilayers grown via MOCVD on native Sn-doped substrates. Conventional and Laplace DLTS techniques have been applied to Au Schottky barrier diodes to determine properties of electron traps with energy levels at about Ec–0.10 eV, −0.42 eV, and −0.53 eV, commonly referred to E10, E9, and E1, respectively.19 Both epilayers possess the E9 trap, which is consistent with the reported DLTS analysis on the MOCVD grown material.28,34 The electric field dependence of E9 implies it is likely to be an acceptor level. The E1 and E10 traps have been observed only in the UID doped epilayer. Our findings indicate that E10 is an unintentional donor level. The non-uniform trap concentration profiles of both E10 and E9 indicate their origin is likely related to the diffusion of some impurity atoms from either the substrate or interface into the epilayers. The absence of E10 in the Si-doped epilayer suggests that Si is occupying the thermodynamically preferred Ga(I) site. Contrastingly, the presence of E10 in the UID epilayer implies that the Si atoms originating from the interface could also occupy the octahedral Ga(II) site. Further work is needed to reveal the structure of the defects, which are responsible for all of the discussed traps.
ACKNOWLEDGMENTS
This work has been supported in Manchester by the EPSRC-UK under Contract Nos. EP/T025131/1 and EP/S024441/1. This work has also been supported in Bristol by the EPSRC-UK under Contract No. EP/X015882/1. The work of Martin Kuball was supported by the Royal Academy of Engineering through the Chair in Emerging Technologies Scheme.
AUTHOR DECLARATIONS
Conflict of Interest
The authors have no conflicts to disclose.
Author Contributions
C. A. Dawe: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Visualization (equal); Writing – original draft (lead); Writing – review & editing (equal). V. P. Markevich: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Supervision (lead); Validation (lead); Visualization (equal); Writing – original draft (supporting); Writing – review & editing (equal). M. P. Halsall: Conceptualization (supporting); Formal analysis (supporting); Funding acquisition (lead); Investigation (supporting); Project administration (equal); Resources (equal); Supervision (lead); Validation (equal); Writing – review & editing (equal). I. D. Hawkins: Resources (lead); Supervision (equal); Validation (equal); Writing – review & editing (equal). A. R. Peaker: Formal analysis (supporting); Funding acquisition (lead); Investigation (supporting); Project administration (supporting); Resources (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal). A. Nandi: Resources (lead); Validation (equal); Writing – review & editing (equal). I. Sanyal: Formal analysis (supporting); Resources (lead); Validation (equal); Writing – review & editing (equal). M. Kuball: Funding acquisition (lead); Project administration (equal); Resources (lead); Supervision (lead); Writing – review & editing (equal).
DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author upon reasonable request.