Hydrogen Release During Isochronal Annealing

The retention of implanted ²H within GaN was monitored by NRA during a sequence of 1-hour vacuum anneals at temperatures incremented by ∼100°C. These experiments were carried out on specimens that had been implanted with 10¹⁵, 10¹⁶ or 10^{17 2}H/cm² at room temperature and 50 keV to produce the calculated [Reference Ziegler, Biersack and Littmark9] depth profiles shown in Fig. 1. The NRA was performed with an incident ³He energy of 0.85 MeV, which, using published results for the nuclear cross section [Reference Möller and Besenbacher7] and He stopping powers [Reference Daudin, Rouvière and Arlery10], is calculated to result in the depthdependent differential cross section shown in Fig. 1. The nuclear-reaction yield under these conditions provides a quantitative measure of the amount of ²H remaining within the implanted region.

Fig. 1. Calculated depth profiles of implanted ²H and the differential nuclear cross section for 0.85-MeV ³He with a scattering angle of 135°.

Results from these experiments are presented in Fig. 2, where the retained fraction of ²H is plotted versus temperature. The anneal at the highest temperature, 980°C, produced significant loss of material from the GaN film as observed by RBS, so that the corresponding datum does not provide a quantitative measure of ²H release by diffusion. In the case of the lowest implantation dose, effects of the analysis beam on the release were observed, so each measurement was made at a new location on the specimen.

Fig. 2. Release of ion-implanted ²H from GaN during isochronal annealing.

For all three doses, the ²H remained localized in the implanted region during the release rather than assuming a diffusion-broadened profile. This is seen from the representative concentration-versus-depth profile in Fig. 3, which was measured in the sample implanted with 10^{17 2}H/cm² after the anneal sequence had reached 886°C and about 2/3 of the ²H had been lost. This profile was obtained by measuring the proton yield from the nuclear reaction as a function of incident ³He energy and performing a deconvolution [Reference Myers, Caskey, Rawl and Sisson8], giving a distribution represented by contiguous straight-line segments. The depth resolution of this method is about 0.2 µm for GaN, which suffices to exhibit the property of interest.

Fig. 3. Depth profile of ion-implanted ²H in GaN after implantation of 10^{17 2}H/cm² at 50 keV and isochronal annealing extending to 886°C.

The release of ²H is seen from Fig. 2 to occur in approximately the same temperature range for implanted concentration from hundredths of an atomic percent to several at.%, notwithstanding fundamental differences in the state of the ²H at the lower and higher concentrations that are discussed below.For the lowest dose, the temperature range of release is consistent with earlier work where H was implanted to comparable concentrations in undoped material [Reference Zavada, Wilson, Abernathy and Pearton1,Reference Pearton, Wilson, Zavada, Han and Shul3]. The absence of diffusion broadening of the implanted peak during the release indicates that the H exists predominantly in bound states rather than being in mobile solution. The nature of these states is addressed in the remainder of this paper.

H₂ Bubble Formation with Wall Chemisorption of H at Implanted Concentrations ∨1 at.%

When the implanted concentration of H was ∨1 at.%, H₂ bubbles formed within the GaN. After annealing in the temperature range of H release, TEM showed most of these bubbles to be polyhedra with a single (0001) facet and 6 {10

1} facets, a shape which reduces to a 6-sided pyramid when the hexagonal basal-plane facet is equilateral. Representative high-resolution images in plan view and cross section are shown in Fig. 4. With no noted exceptions, the base of the polyhedral bubbles was toward the surface of the sample and the apex toward the sapphire substrate. Such an absence of inversion symmetry along [0001] indicates that the facets have a preferred polarity. The identification of this polarity requires determination of the polarity of the epitaxial film. Although this is challenging, several groups have concluded on the basis of convergent-beam electron diffraction and other measurements that smooth MOCVD GaN on sapphire has N atoms above Ga atoms along [0001]-oriented bonds, or, equivalently, that the unreconstructed top surface of the film is terminated by Ga atoms [Reference Ponce, Bour, Young, Saunders and Steeds11-Reference Liliental-Weber, Richter, Washburn, Pakula, Baranowski, Grzegory, Porowski and Yang14]. This implies that the (0001) and {101} internal surfaces of the observed bubbles, both comprising close-packed Ga-N bilayers, have N-terminated polarity. Since the bubbles formed after growth, we conclude that these surfaces were energetically preferred.

Fig. 4. High-resolution TEM images of bubbles in GaN: (a) plan-view in [0001] orientation and (b) cross-section in[11

0] orientation. The implanted layer was injected with 2×10^{16 1}H/cm² at room temperature, and the specimen was then annealed at 886°C for 1 hour.

In view of the relatively high reactivity of N and the transfer of electronic charge from Ga to N in GaN, a preference for N-terminated surfaces is surprising unless these surface are H-passivated. We obtained evidence for such passivation from IR vibrational spectroscopy, as discussed more fully elsewhere [Reference Seager, Myers, Petersen, Han and Headley5]. Under conditions of implantation where bubbles formed, but not otherwise, we observed new IR absorption peaks at 3182 and 3216 cm⁻¹ for ¹H, as shown in Fig. 5, and at 2364 and 2386 cm⁻¹ for ²H. These absorptions are ascribed to N-H centers, based both on comparisons with materials known to contain N-H and on the size of the departure of the isotope shift from √2. The frequencies lie above those reported when the concentration of implanted H is ∧0.1 at.% [Reference Weinstein, Song, Stavola, Pearton, Wilson, Shul, Killeen and Ludowise2], a regime where we did not observe bubble formation and the H is believed bound to defects. Moreover, the integrated strength of the absorptions in Fig. 5 corresponds to about 4 % of the implanted H being in the IR-active state, and this is consistent with estimates of the number of chemisorption sites on the bubble walls. We consider these findings strongly indicative of H passivation of the bubble walls.

Fig. 5. Infrared absorption peaks ascribed to ¹H chemisorbed on bubble walls. The specimen was implanted with 2×10^{16 1}H/cm² in each of 5 layers and then annealed at 600°C for 1 hour.

In view of the relatively high reactivity of N and the transfer of electronic charge from Ga to N in GaN, a preference for N-terminated surfaces is surprising unless these surface are H-passivated. We obtained evidence for such passivation from IR vibrational spectroscopy, as discussed more fully elsewhere [Reference Seager, Myers, Petersen, Han and Headley5]. Under conditions of implantation where bubbles formed, but not otherwise, we observed new IR absorption peaks at 3182 and 3216 cm⁻¹ for ¹H, as shown in Fig. 5, and at 2364 and 2386 cm⁻¹ for ²H. These absorptions are ascribed to N-H centers, based both on comparisons with materials known to contain N-H and on the size of the departure of the isotope shift from √2. The frequencies lie above those reported when the concentration of implanted H is ∧0.1 at.% [Reference Weinstein, Song, Stavola, Pearton, Wilson, Shul, Killeen and Ludowise2], a regime where we did not observe bubble formation and the H is believed bound to defects. Moreover, the integrated strength of the absorptions in Fig. 5 corresponds to about 4 % of the implanted H being in the IR-active state, and this is consistent with estimates of the number of chemisorption sites on the bubble walls. We consider these findings strongly indicative of H passivation of the bubble walls.

During isochronal annealing sequences similar to those of Fig. 2, the strength of the IR absorptions shown in Fig. 5 decreased before the total amount of ²H measured by NRA. This is seen in Fig. 6, where both quantities are plotted versus anneal temperature. We infer that the chemisorbed state is less stable than the molecular gas under the conditions of these experiments.

Fig. 6. Decrease of H-related IR absorptions during isochronal annealing, compared with the loss of total H measured by NRA.

Isothermal Release from Bubbles and the Diffusivity-Solubility Product of H

In order to characterize further the thermal release of H from H₂ bubbles, we used NRA to measure the amount remaining during isothermal vacuum annealing at 794 and 886°C. The specimens were implanted with 1×10^{17 2}H/cm² at 50 keV, and the areal density of ²H within the implanted layer was monitored using a ³He energy of 0.85 MeV, replicating conditions depicted in Fig. 1.The results obtained at 886°C are presented in Fig. 7. Depth profiling at selected points in the sequence, using procedures described for Fig. 3, showed that the retained H was always localized in the implanted layer rather than developing a diffusion-broadened distribution.

Fig. 7. Release of ion-implanted ²H from GaN during isothermal annealing at 886°C.

From the TEM observations discussed above, we conclude that the release shown in Fig. 7 was predominantly from H₂ gas within bubbles. Furthermore, we assume, in accord with theoretical predictions [Reference Neugebauer and Van de Walle15], that the transport of the H through the matrix occurred by dissociation of the molecule followed by atomic diffusion. Then, using the approximation of steady-state diffusion, the variation with time of the retained areal density of H atoms, Λ, is given by

where D is the effective diffusion coefficient, N_H0 is the solubility of mobile H in local equilibrium with the H₂ gas at some reference pressure P₀ small enough for the gas to behave ideally, P* is the fugacity of the H₂ gas in the bubbles, and X is the diffusion distance to the surface, taken to be single-valued for simplicity. The fugacity can be defined in terms of the chemical potential and related through it to the equation of state:

where μ₀ is the chemical potential at the reference pressure and V(P) is the volume per H₂ molecule at pressure P as given by the equation of state. At lower pressures where the ideal-gas equation of state is applicable, P* → P. In order to complete the description of the system, a relationship must be specified between P and ∧. This is accomplished by noting that

where the function V(P) is given by the equation of state and V_b is the combined bubble volume per unit sample area. Solution of Eqs. (1)-(3) shows that that ∧^1/2 decreases linearly with time at lower pressures where V(P) follows the ideal equation of state, while a more rapid initial transient is predicted if the starting pressure is high enough for the gas to behave non-ideally. Such behavior is evident in the inset of Fig. 7, where the initial part of the measured isotherm is plotted as ∧^1/2 versus t. The curve through the data represents a solution of Eqs. (1)-(3) in which the amplitude of the initial transient was fitted by adjusting V_b or, equivalently, the initial bubble pressure P₁, and the variation at longer times was fitted by adjusting the diffusivity-solubility product D×N_H0 for P₀ = 1 bar. In this way we obtained P₁ = 6 GPa and D×N_H0 = 2.2×10⁶ atoms cm⁻¹ s⁻¹ for T = 886°C; the analogous experiment at 794°C yielded P₁ = 9 GPa and D×N_H0 = 2.0×10⁵ atoms cm⁻¹ s⁻¹.

It is apparent from the semilog plot in Fig. 7 that a small fraction of the implanted H, ∼5 %, was released much more slowly than predicted by Eqs. (1)-(3). We consider two possible explanations for this. First, the residual H might conceivably exist in some bound state more stable than H₂ gas within the bubbles. A difficulty, however, is that we did not find IR absorptions associated with the H retained to long times, nor did this H exhibit the channeling effects in NRA that would be expected for a unique atomic position relative to the host lattice; hence, the bound state in question apparently does not involve N-H or Ga-H bonding. If, alternatively, the strongly retained H is imagined to exist as H2 molecules, there is the problem of identifying a location where the molecules are more stably bound than the open volumes of the bubbles.

A second interpretation, and one which we currently consider more viable, is that the release takes place from H₂ bubbles over the entire anneal sequence, but the diffusivity-solubility product D×N_H0 decreases with time due to a shift in the Fermi level as the H concentration in solution decreases. Theoretical calculations reported for the cubic, zincblende variant of GaN [Reference Neugebauer and Van de Walle15] and a recent treatment of the hexagonal, wurtzite phase [Reference Wright16] both indicate that dissociated H in solution exists predominantly as H⁺ or H⁻, depending upon the position of the Fermi level, with the neutral atom always being less stable. Additionally, the positive ion is predicted to be much more mobile than the negative ion. Our tentative interpretation of the release isotherm is then that, during ∼95% of the release, the solution concentrations of H⁺ and H⁻ exceed the concentrations of other electrically active species, and as a result the Fermi level is stabilized at a position such that [H⁺] ∼ [H⁻], with D×N_H0 being nearly constant as a result. In the last stages of the release, however, the H₂ pressure in the bubbles, and hence the H chemical potential, decreases until the H concentration in the lattice is less than that of donors unrelated to H; at this point, the Fermi level rises, reducing the fraction of mobile H⁺ and increasing that of the less-mobile H⁻.

Bubble-Related Defect Microstructure and Its Evolution with Temperature

By using ion-channeling analysis in conjunction with TEM to characterize implantation-related defects, we elucidated the atomic processes associated with bubble formation and defect annealing. Figure 8 shows channeling results for a specimen that was implanted with 10^{17 2}H/cm² at 50 keV and room temperature and then subjected to a series of 1-hour vacuum anneals at increasing temperature. The exhibited RBS spectra were produced by 2-MeV ⁴He incident along the [0001] axis with the sample at room temperature, and an unchanneled, or random, spectrum is included for reference. Backscattering yield from the implanted region appears at channels in the range 90-120, while the surface corresponds approximately to channel 140.

Fig. 8. Channeled RBS spectra from GaN before and after ²H implantation and annealing, obtained using 2-MeV ⁴He.

A direct-scattering peak appears in the channeled spectrum after implantation and before annealing, as indicated in the figure. As discussed more fully elsewhere [Reference Wampler and Myers4], the presence of this peak means that Ga atoms are removed from substitutional lattice sites into the open [0001] channel. Analysis of the area under the peak indicates that the areal density of Ga interstitials is 3.0×10¹⁶ cm⁻² [Reference Wampler and Myers4]; this is equal within experimental uncertainty to the areal density of Ga atoms displaced by bubble formation, which we calculate from the known H areal density and H₂ gas pressure to be 3.4×10¹⁶ cm⁻². This finding is consistent with the following model of bubble formation: the equally numerous vacancies and interstitials produced by collisional atomic displacements mostly undergo annihilation by recombination; however, a small fraction of the vacancies agglomerate and collect H to form high-pressure H₂ bubbles, thereby constraining the recombination and leaving a corresponding number of interstitials within the lattice.

Annealing in the temperatures range 511-709°C is seen in Fig. 8 to remove the direct-scattering peak and introduce a dechanneling step. This indicates a return of Ga atoms to substitutionality, accompanied by the formation of extended defects with large strain fields to produce the observed dechanneling with relatively little direct scattering. Transmission electron microscopy after this evolution showed the predominant extended defect to be a (0001) stacking fault containing one extra Ga-N bilayer and bounded by a partial dislocation, a configuration discussed in detail elsewhere [Reference Liliental-Weber, Kisielowski, Ruvimov, Chen, Washburn, Grzegory, Bockowski, Jun and Porowski17]. A part of one such stacking fault is seen in the high-resolution cross-section TEM image of Fig. 9. Analysis of the amplitude of the dechanneling step gives the density of the stacking faults. The result indicates that only a small fraction of the Ga interstitials present before annealing, about 3%, agglomerate in this way [Reference Wampler and Myers4]; the remainder undergo annihilation, presumably by migrating to the surface of the film.

Fig. 9. High-resolution cross-section TEM image showing a part of one stacking fault in H-implanted GaN after annealing at 886°C.

Hydrogen Behavior in the Absence of Bubbles at Concentrations ∧0.1 at.%

When GaN was implanted with 50-keV ²H at room temperature to a dose of 10¹⁵ cm⁻², ion-channeling analysis revealed that a large fraction of the ²H occupied a well-defined location relative to the host lattice. Figure 10 shows the nuclear-reaction yield from the ²H as a function of angle from the [0001] direction, with the elastic backscattering yield from Ga superimposed for comparison; these yields were produced by 0.85-MeV ³He and 2-MeV ⁴He, respectively. The detected H site is believed to be defect related, since the amplitude of the channeling peak rises markedly during initial impingement of the ³He analysis beam, and rises more rapidly when that beam is tilted from the channeling orientation so that more atomic displacements are produced. Numerical simulations of the channeling indicate that the narrow peak in the nuclear-reaction yield can be produced only by ²H that is displaced into the central region of the open [0001] channel [Reference Wampler and Myers4]. During a succession of 1-hour anneals at increasing temperature, this peak remained essentially unchanged after 411°C, but decreased in amplitude with increasing temperature from 511°C upward, becoming poorly resolved by 809°C.

Previously, ¹H implanted to similar concentrations in GaN was found by IR spectroscopy to have local vibrational modes in the frequency range 2983-3150 cm⁻¹ [Reference Weinstein, Song, Stavola, Pearton, Wilson, Shul, Killeen and Ludowise2]. These modes were assigned with confidence to N-H centers, which were plausibly hypothesized to be located in the Ga vacancy. The absorption spectrum evolved with increasing anneal temperature and disappeared at about 700°C. The similarity with our experiments raises the possibility that the same H state was probed by the two techniques; this is not necessarily so, however, since the fraction of the implanted H contributing to the observed IR peaks may have been as small as 10% [Reference Stavola18]. In any case, the Ga vacancy does not appear viable for the trap observed by channeling, since H tetrahedrally bonded to a neighboring N would not protrude into the [0001] channel. It therefore seems appropriate to consider other defects, including those of interstitial type, as possible binding sites.

Since the H state or states probed by the ion channeling and IR spectroscopy disappear after annealing at temperatures ∧800°C, these are apparently not the states from which the final release near 900°C occurs. The last state of the H is also believed not to involve bubbles or extended defects, since, in cross-section TEM of GaN implanted with 10^{15 2}H/cm² at 50 keV and then annealed at 886°C for 1 hour, we did not observe such microstructural features produced by the implantation. Conceivably, H₂ forms within vacancy-type defects at the most elevated temperatures.

Fig. 10. Channeling scans after implanting 10^{15 2}H/cm² into GaN at room temperature.