Anisotropy of hydrogen plasma effects in bulk n-type β-Ga₂O₃

The behavior of different dopants and impurities in β-Ga₂O₃ is being actively studied due to the growing importance of this wide-bandgap semiconductor as a promising material for next generation high-power devices and solar-blind devices.^1,2 The properties of hydrogen draw much attention because it is a common residual impurity in bulk and epitaxial gallium oxide. Theoretical calculations suggest that hydrogen in β-Ga₂O₃ should be a shallow hydrogen-like donor and that interstitial hydrogen can also easily form complexes with gallium vacancy acceptors V_Ga³⁻ that are major native defects in β-Ga₂O₃.^3–5 Under strongly nonequilibrium conditions, hydrogen acceptors can become dominant species, even though their formation energy is high.⁶ Experimental studies using local vibrational mode (LVM) spectroscopy and IR absorption/reflectance spectroscopy^7,8 suggest that high temperature treatments in molecular hydrogen indeed facilitate shallow donor formation and formation of V_Ga–H complexes.^7,8 However, this process seems to be sensitive to the crystal orientation and is mostly observed for the (−201)-oriented samples which is believed to be a manifestation of anisotropy of the low-symmetry monoclinic structure of β-Ga₂O₃ with many properties measurably differing for various surface orientations.⁹ When hydrogen is introduced from a hydrogen plasma, it has been observed that, under mild plasma treatment conditions, hydrogen does not penetrate deep into β-Ga₂O₃ films and enhances the shallow donor concentration in the near-surface regions.¹⁰ On the other hand, hydrogen plasma treatment under conditions promoting the formation of heavy damage in the near surface region gives rise to hydrogen penetration quite deep into the samples and to strong compensation or passivation of the surface region, rendering it semi-insulating.^11,12 It is possible that under such conditions, hydrogen is the form of an acceptor species, forming complexes with shallow donors.¹⁰ 

It is of interest to examine if the effects of such “harsh” hydrogen plasma treatment conditions will be dependent on crystal orientation. In this paper, we report a study of the effects on bulk n-Ga₂O₃ samples with (−201) and (010) orientations of treatment in high-power hydrogen plasmas with parameters similar to the ones used previously.^10–12 We show that the results of such hydrogen plasma treatment are highly anisotropic, the (−201) orientation facilitating strong enhancement of the shallow donor concentration at the surface, while the (010) orientation undergoes a marked decrease of the net shallow donor concentration near the surface. These results have additional importance in view of a recent report that high-temperature treatment of β-Ga₂O₃ in molecular hydrogen can convert the surface to p-type, presumably owing to the formation of shallow acceptor complexes of hydrogen with V_Ga.¹³ Processes occurring during hydrogen plasma treatment at relatively low temperature and during high temperature annealing in H₂ in sealed ampoules are different and have to be individually studied in detail. In this paper, we point to possible relevance of the model proposed in this molecular hydrogen treatment paper to explain the diversity of results obtained in our plasma experiments.

The samples used were bulk n-type β-Ga₂O₃ crystals grown by edge-defined film-fed growth (EFG) from Tamura Corp. (Japan).¹⁴ One of the crystals had (−201) orientation, the other was (010)-oriented. Both samples were doped with Sn and had shallow donor concentration of ∼3 × 10¹⁷ cm⁻³. The thickness of the samples was 650 μm. The back side was mechanically polished, with the front side chemo-mechanically polished to epi-ready quality.

Hydrogen plasma treatment was performed in an Inductively-Coupled Plasma (ICP) reactor (PlasmaLab 100 dual, Oxford Instruments Technology, UK) at 330 °C for 30 min, at a pressure of 36 mTorr. The ICP RF power was 1500 W, the RF power applied to the chuck was 75 W, and the bias on the chuck was 298 V. These treatment conditions are similar to those used previously.^10–12 

Each sample was cut into two pieces, one used as a reference, and the other treated in H₂ plasma. Ohmic contacts to all samples were prepared by Ti/Au e-beam evaporation on the back surface preliminarily treated in Ar plasma. After Ti/Au deposition, the contacts were rapid thermal annealed in N₂ at 500 °C to facilitate the formation of good Ohmic contacts.^10–12 This procedure was done prior to the H plasma treatment to avoid any high temperature annealing stage. Schottky diodes were formed by E-beam deposition of 30-nm-thick circular Ni contacts on the cleaned front surface of the reference and H plasma treated samples.

The samples were characterized by capacitance–voltage (C–V) profiling, capacitance–frequency C–f, and deep level transient spectroscopy (DLTS)¹⁵ measurements in the temperature range of 100 K–500 K in a cold-finger liquid nitrogen cryostat (Cryotrade company, Russia) using the setup described previously.^10,16,17

Specifically, C–V profiling and C–f measurements were done using an E4980A LCR meter in the frequency range of 20 Hz–1 MHz (KeySight Technologies, USA). The same instrument was used in conjunction with a pulse generator 33500B (KeySight Technologies, USA) for DLTS spectra measurements following the procedure in which capacitance relaxation curves were measured and stored at each temperature point with the temperature step close to 0.1 K and the spectra could be built using any chosen set of time windows t₁, t₂. Alternately, both the relaxation curves at different temperatures and DLTS spectra for the chosen t₁, t₂ time windows could be reproduced by modeling using the concentrations, energy level positions, and capture cross sections of the traps as the fitting parameters. The great advantage of this system is that it allows to measure DLTS spectra for frequencies from 10 kHz to 1 MHz and thus to minimize detrimental effects of the series resistance, which often are important in wide-bandgap materials.^15–17 

The two starting samples were designated sample (−201) and sample (010) for the two orientations used, while after the H plasma treatment, they are referred to as sample (−201)-H and sample (010)-H. Figures 1(a) and 1(b) compare the room temperature and 150 K C–f characteristics of, respectively, sample (−201) and sample (−201)-H, sample (010) and sample (010)-H before and after H plasma treatment. For the (−201) orientation, the H plasma treatment dramatically increased the capacitance, indicating an increased density of shallow donors, while the H plasma treatment of the (010)-oriented sample strongly decreased the capacitance pointing to a prominent decrease of the net shallow donor concentration.

Figures 2(a) and 2(b) compare the room temperature C–V characteristics presented as 1/C² vs voltage V plots, respectively, for the (−201) and the (010) samples before and after H plasma treatment. For the (−201) sample, the Schottky barrier height was of ∼1.5 eV, and the net shallow donor concentration calculated from the slope of the 1/C²–V plot at 10 kHz was 2.7 × 10¹⁷ cm⁻³. After the H plasma treatment, the shallow donor concentration increased to 2.6 × 10¹⁸ cm⁻³ while the voltage cut-off decreased to 0.86 eV, indicating both increased net shallow donor density and the high density of deep traps at the surface.¹⁷ 

For the (010) sample, the Schottky barrier height was 1.75 eV and the shallow donor density was 3.2 × 10¹⁷ cm⁻³. After the H plasma treatment, the net shallow donor density determined from the slope of the 1/C²–V plot decreased to 3.2 × 10¹⁶ cm⁻³, while the voltage offset was very high, indicating the presence of a highly compensated region near the surface.¹⁷ The charge concentration profiling in this sample after the H plasma treatment showed the space charge region (SCR) boundary at even slightly positive voltages up to 1 V extended up to 0.9 μm, pointing to a very low (below 10¹⁵ cm⁻³) net concentration of donors in that part of the SCR in the dark. For reverse voltages, the net shallow donor concentration was still very low down to the maximum depth of 1.2 μm that could be probed by C–V profiling before the reverse current started to increase and handicap the accuracy of capacitance measurements. Thus, the width of the region in which plasma treatment led to a very strong decrease in the net shallow donor density in the (010)-H sample is ∼1 μm, similar to that previously reported for similarly H plasma treated (010)-oriented films prepared by halide vapor phase epitaxy (HVPE).^10,12

An issue is whether the fully depleted region in the dark C–V profile had some deep centers remaining that could be excited with below-bandgap energy photons and whether such centers were the same as in the sample before treatment. Figure 3(a) shows that, for photon energies starting with 2.3 eV, the charge concentration near the edge of the space charge region of sample (010)-H started to increase. As the number of photogenerated carriers increased with increased photon energy, the charge concentration probed by C–V profiling increased, and the SCR boundary position at 0 V moved closer to the surface [compare the profiles obtained with high-power light emitting diode (LED) illumination for the peak LED photon energy of 2.3 eV, 2.8 eV, 3.1 eV in Fig. 3(a)] so that, with 3.1 eV illumination, the space charge region boundary extended to about 0.4 μm from the surface and the net charge density was ∼2 × 10¹⁵ cm⁻³. It stayed near that level when the sample was kept in the dark after illumination [what is generally called Persistent Photocapacitance (PPC)] and the concentration could not be quenched by an application of a forward bias pulse of 2 V for 10 s [“PPC 3.1 eV + 2 V pulse” dashed line in Fig. 3(a)].

Before the plasma treatment, the photoinduced charge concentration spectra had the dependence shown in Fig. 3(b). For photon energies above 2.3 eV, the signal was persistent after the light termination and could not be quenched by the application of a forward bias pulse of +2 V for 10 s, suggesting that respective centers had a high barrier for capture of electrons. The spectrum of photoinduced charge N_photo showed three threshold energies near 1.3 eV, 2.3 eV, and 3.1 eV, with concentrations of deep traps that could be depopulated by illumination being, respectively, close to ∼10¹⁵, 5 × 10¹⁵, and 6 × 10¹⁵ cm⁻³. For all energies above 2.3 eV, the photocapacitance persisted after the light termination (PPC) signal. For photon energies between 2.3 eV and 3.1 eV, the persistent photoinduced concentration could not be quenched by the application of the forward bias pulse of 2 V indicating that the centers with an optical ionization threshold of 2.3 eV had a high barrier for the capture of electrons, which is now a well-established feature of these centers.^17–20 For higher photon energies, part of the signal attributable to the ionization of deep traps with an optical ionization threshold of 3.1 eV could be quenched by the application of the forward bias pulse, suggesting that the PPC signal for these traps was the result of their large ionization energies and the absence of free electrons in the SCR region available for recapture by holes on the 3.1 eV centers.^17,20 Obviously, after the hydrogen plasma treatment, the density of deep traps susceptible to depopulation by light became lower in the near surface region indicating that they have been either passivated by hydrogen or converted to other states that could not be excited by photons used (energies from 1.3 to 3.4 eV).

For deep trap spectra, we used samples (010) and (010)-H as more amenable to quantitative comparisons. Figure 4(a) compares the spectra measured with a reverse bias of −2 V and a forward bias pulse of 0 V for both samples. Before the H plasma treatment, the dominant feature of the spectra was a prominent peak near 325 K corresponding to the center with level E_c-0.82 eV and electron capture cross section σ_n= 1.8 × 10⁻¹⁴ cm². This is the well-known E2 trap ascribed to the acceptor Fe^3+/Fe²⁺ charge transfer level near E_c-(0.78–0.82) eV.^5,21–25 Another prominent peak in the spectrum in Fig. 4(a) is the well documented center E3 with the level near E_c-1.06 eV and σ_n= 4.9 × 10⁻¹⁴ cm².^5,20,26,27 The y axis in Fig. 4 is 2N_d*ΔC/C*F⁻¹ which, for the peaks in DLTS spectra, gives the trap concentration without accounting for the λ-correction¹⁵ (N_d is the shallow donor density, ΔC is the transient capacitance difference at time windows t₁ and t₂ during the capacitance decay measurement in DLTS at temperature T, C is the steady-state capacitance, and F⁻¹ is the spectrometer function converting ΔC to the full value of the capacitance decay magnitude¹⁵).

After the H plasma treatment, the peak belonging to the Fe acceptors was suppressed, while the dominant center became the peak due to a center with a level near E_c-0.74 eV that is similar to the well-known E2* trap due to native defects in Ga₂O₃.^5,20,21 These traps are often observed in HVPE films, and their concentration increases after irradiation with high energy particles.^5,20,21,28 In Fig. 4(a), we show for comparison the spectrum measured for a neutron irradiated HVPE film, where the E2* peak is clearly observed alongside the E2 peak. In EFG samples, the E2 peak is commonly broadened on the low-temperature side by the presence of the E2* peak.^5,21 Fitting of the actual spectra of sample (010) before hydrogenation with model spectra calculated assuming the presence of three major peaks E2*, E2, and E3 allows to accurately reproduce the measured spectra if the E2* level is placed at E_c-0.724 eV (σ_n= 3.4 × 10⁻¹⁵ cm²), E2 at E_c-0.81 eV (σ_n= 1.2 × 10⁻¹⁴ cm²), and E3 level at E_c-1 eV (σ_n= 1.3 × 10⁻¹³ cm²). The relative amplitudes of the peaks of the centers deduced from this fitting procedure are 0.106 for E2, 0.034 for E2*, and 0.017 for E3 centers [Fig. 4(b)].

After the H plasma treatment, we also observed the emergence of a prominent peak near E_c-0.6 eV similar to the well-known peak E1 in β-Ga₂O₃^5,21,28 and a small peak due to a trap with an energy level near E_c-0.3 eV similar to the one observed after proton and α-particles irradiation, so-called E7 center^20,28 [such a peak can also be seen in the spectrum of the neutron irradiated HVPE sample in Fig. 4(a)]. Based on the results of modeling in Fig. 4(b), the hydrogen plasma treatment slightly decreases the concentration of the E2* trap (the starting value around 3.4 × 10¹⁵ cm⁻³), but this should be taken with a great degree of caution because of the strong variations of the net donor density with depth after hydrogenation [in building Fig. 4(a), the donor density was taken to be 3.2 × 10¹⁶ cm⁻³] and the fact that the measured concentrations in the starting and H plasma treated samples refer to different depths (∼0.16 μm before H treatment, ∼1 μm after H treatment). The concentration of the E2* trap does not appear to be seriously affected in sample (010)-H. We cannot conclude anything about the E3 traps after H plasma treatment because DLTS spectra could not be reliably measured at temperatures approaching 400 K due to leakage current increasing with temperature.

For the (−201) sample, DLTS spectra comparisons are more difficult because of the high density of shallow donors that precludes reliable measurements after the H treatment for temperatures higher than 350 K and limits the reverse voltage at which the spectra could be measured to −1 V. From comparison of the spectra before and after H treatment (Fig. 5), the E2 Fe related acceptor peak was again strongly suppressed, while the concentration of the E2* center in the surface region of the plasma treated sample was greatly enhanced, although again the quantitative comparisons are not very straightforward because of the great differences in shallow donor’s concentration profiles (measurements before H plasma treatment refer to the region ∼0.2 μm-deep whereas, after the treatment the SCR region boundary is confined to the very near-surface region).

A reduction in the Fe related E2 deep level is observed for both orientations, leaving the E2* level more easily discernable. The E2* concentration increases in the sample with (−201) orientation. The DLTS measurements before and after plasma treatment differ due to changes in charge carrier concentrations, which implies different volumes are measured, and one needs to be careful to compare defect concentrations from these measurements. The quantitative conclusions regarding the concentration of the E2* level in hydrogen plasma treated samples are difficult to arrive at due to the fact that, in the (010)-H sample we are only able to probe in the dark the very edge of the SCR where the shallow donor concentration switches from strongly passivated region with hydrogen to the bulk region not strongly affected by hydrogen. For the (−201)-H sample, the shallow donor concentration is so high that only the region immediately near the surface can be probed. Thus, comparisons with the profiles measured in the starting samples [Fig. 6 shows profiles for the two samples at 380 K and 450 K corresponding to the temperatures of the major peaks; for the (010)-H sample the profiles are actually shown in Fig. 3(a)] do not help because, for the (010) sample, they probe a much more shallow region, while for the (−201) sample, a much deeper region. Also, accurately estimating the λ-corrections for these two sets of samples after hydrogen treatment is not straightforward. For the Fe E2 center, the conclusions are valid because the respective peak vanishes.

It was also interesting to examine the properties of shallow centers determining conductivity after the H plasma treatment for the (−201) sample, in which the hydrogenation led to a dramatic increase of the donor density. The activation energies of shallow donors in hydrogen treated samples (−201)-H and (010)-H were determined from the temperature dependence of forward current in I–V characteristics at +1 V of the Schottky diodes. These energies correspond to the temperature dependence of the series resistance of the film. Figure 7(a) shows these I–Vs at room temperature, and Fig. 7(b) shows the temperature dependence of the forward current at 1 V. From these measurements of the temperature dependence of the forward current of the Schottky diode at 1 V where it was governed by the series resistance, the ionization energies of the donors were estimated as 22 meV. In contrast, in sample (010)-H, the ionization energy of the donors determining the resistivity was 0.19 eV. Such donors were also observed in admittance spectra of both samples before H plasma treatment and in sample (010)-H after H plasma treatment [for sample (−201)-H admittance spectra measurements were not informative because of the high leakage current].

The first question is why H-incorporation or the presence of native defects is strongly dependent on crystal orientation. The symmetry of monoclinic β-Ga₂O₃ is low, the structure of surfaces forming different facets (the type, arrangement and density of atoms terminating the surfaces) are different, suggesting that the surface energies of different facets could be different, the energies necessary to cleave the crystal should be orientation dependent, incorporation of impurity atoms, their mobility in different directions could be different. Theoretically and experimentally, a prominent anisotropy of thermal conductivity, dielectric constants, adsorption energy of atomic species, energy necessary for cleaving for different directions has been demonstrated.²⁹ In thermal annealing in hydrogen, a clear anisotropy has been reported.^1,2,8,9

When hydrogen is introduced into Ga₂O₃ from a high-power ICP plasma known to produce considerable surface damage, the results critically depend on surface orientation. For (−201) oriented samples, hydrogen plasma treatment predominantly introduces shallow donors in the near-surface region, passivates the dominant E2 acceptors due to the Fe³⁺/Fe²⁺ charge transition level near E_c-0.82 eV, and enhances the concentration of the E2* native defects with levels near E_c-0.74 eV. Because of the high density of H-related shallow donors, it is difficult to estimate the hydrogen penetration depth based on purely electrical measurements used in the present work, but previous secondary ion mass spectrometry (SIMS) profiling of deuterium introduced into bulk (−201)-oriented samples treated in deuterium plasma under similar conditions suggest the penetration depth to be ∼1 μm.^10,11

For the (010) samples, hydrogen introduction leads to a marked decrease of the net shallow donor density down to about 1 μm, passivates the E2 Fe^2+/Fe³⁺ acceptors, and slightly decreases the concentration of the E2* centers present before hydrogenation. Judging by the results of C–V profiling, the top ∼0.9 μm of the sample has a very low density of residual donors, but the electron concentration in this surface region can be persistently increased by illumination by light with a photon energy above 2.3 eV.

These results are in good correspondence with previously reported effects caused by high-power hydrogen plasma treatment of the (010)-oriented HVPE films.^10,12 Hydrogen plasma treatment at the same temperature and during the same time, but under mild conditions has been reported to introduce additional shallow surface donors, but only at a low concentration and to a very small depth.¹⁰ These results were explained under the assumption that harsh hydrogen plasma conditions create a lot of surface damage that facilitates the introduction of hydrogen in the thermodynamically not favorable form of H⁻ acceptors that are highly mobile and can readily form neutral complexes with shallow donors.¹⁰ Under mild hydrogen plasma treatment conditions, hydrogen is introduced predominantly as a thermodynamically stable H⁺ donor species with lower mobility. These H⁺ donors give rise to the increase in the concentration of donors in the near-surface region.

However, the observations described above are not easily reconciled with the explanation in Ref. 10. For example, we see that, even with harsh H plasma treatment conditions, this plasma exposure still increases the net donor density near the surface of the (−201) sample, but compensates or passivates shallow donors near the (010) surface. This is not easy to explain if we assume that under harsh hydrogen plasma conditions hydrogen is introduced preferentially in the form of H⁻ acceptors. Then, if hydrogen is introduced as an acceptor, it would not passivate the Fe acceptors. The hydrogen penetration depth estimated from SIMS for the samples with (−201) orientation is close to the depth of the hydrogenated portion estimated from C–V profiling in (010) samples, suggesting that the same kind of hydrogen species is involved in both cases. This movement is facilitated by the formation of surface damage defects and handicapped when such surface damage is minimized.¹⁰ It looks as though some more complicated processes involving hydrogen interaction with native defects created during plasma treatment occur. A comparison with the recently published results of high-temperature molecular hydrogen annealing in sealed ampoules¹³ might provide clues for qualitatively understanding the nature of the processes involved. There, it is claimed that the main effect of high temperature treatment in molecular hydrogen is the filling of the triply negatively charged Ga vacancies V_Ga³⁻ with different numbers of H⁺ ions.¹³ In the material with a high density of V_Ga³⁻, the main effect is the formation of acceptor complexes of (V_Ga–2H)⁻ with a lowered acceptor ionization energy. If thermal annealing or hydrogen plasma treatment enhances the starting density of V_Ga deep acceptors, such a process will lead to compensation of n-type conductivity or, if the (V_Ga–2H)⁻ acceptors are shallow, can even convert the surface to p-type.¹³ On the other hand, in samples with hydrogen ion density much higher than V_Ga density, the Ga vacancy forms a complex with four H⁺ ions and becomes a shallow donor (V_Ga–4H)⁺ (with ionization energy about 20 meV according to the estimate of Ref. 13) and thus enhances the n-type conductivity.

If one assumes that different numbers of Ga vacancies are produced on the (−201) and (010) surfaces upon harsh plasma treatment and different H ion concentrations are incorporated into these differently oriented samples, one can end up in one case with predominantly (V_Ga–4H)⁺ donors that are 20 meV deep, as for our treatment of the (−201) samples, while on the other hand be left with predominantly (V_Ga–2H)⁻ acceptors compensating the starting n-type conductivity due to Sn donors, as for our (010) sample. In both cases, the mobile species will be H⁺ ions that form neutral complexes with Fe acceptors. This would explain the observed passivation of the E2 Fe acceptors for both orientations. The question regarding the interaction with the E2* traps is less clear and requires a better understanding of the nature of these traps, which are related to native defects. Even if the assumption regarding the depth of the (V_Ga–2H)⁻ acceptors and the possibility of formation of the (V_Ga–4H)⁺ donors is not confirmed, the observed asymmetry of hydrogen treatment effects for different orientations could still be understood if the defect generation rate and the hydrogen incorporation rates are different so that V_Ga concentration is higher for the (010) surface and the surface H⁺ concentration is higher for the (−201) surface. Then, in one case, passivation will be observed for the (010) surface due to (V_Ga–nH) acceptor complexes, while for (−201) orientation, the increase of the surface donor density will be provided by the excess of H⁺ shallow donors over the (V_Ga–nH) acceptors. The mobility of hydrogen then will be in all cases determined by the hopping of hydrogen between the interstitial sites and trapping by vacancies. Whether p-type conductivity can indeed be obtained due to the (V_Ga–2H)^– acceptors becoming shallow¹³ could be checked with hydrogen plasma treatments of surfaces preconditioned by annealing in oxygen or Ga overpressures.¹³ If successful, these experiments could provide an easier way of obtaining n-type or p-type conductivity in the surface layers of β-Ga₂O₃ than the method using high temperature anneals in sealed ampoules.¹³ However, annealing in hydrogen in sealed ampoules creates surface damage and possible contamination (in most cases, the surface region of the samples in similar annealing experiments is removed^8,9). The underlying processes in plasma treatment and in annealing in molecular hydrogen in sealed ampoules are different, the former treatment being much further from equilibrium conditions. Another issue is that our pool of Ga₂O₃ samples is limited, hence the experiments were performed on one sample of each orientation cut into two pieces—one reference, one hydrogen treated. The redeeming factor is that the studied samples were commercial crystals that generally show consistent properties when analyzed by different groups.

Finally, this treatment has been done on n-type samples, and it is interesting to speculate on the expected effect if the treatment is carried out on nonconductive samples. We do not fully understand how the hydrogen diffusion is affected by the position of the Fermi level that determines the switch between the H⁺ and H⁻ species. We expect that, for the (−201) surface, the sample should become more n-type under our hydrogen plasma treatment conditions. For (010) semi-insulating samples, one might observe the samples becoming conducting n-type if the hydrogen concentration is sufficient to passivate Fe acceptors. All these require more studies and a reasonably diverse set of samples grown under well understood conditions.

The work at NUST MISiS was supported in part by the Russian Science Foundation under Grant No. 19-19-00409. The work at UF was sponsored by the Department of the Defense, Defense Threat Reduction Agency (No. HDTRA1-17-1-011) monitored by Jacob Calkins and also by the National Science Foundation (NSF) (No. DMR 1856662) (Tania Paskova).