Ion Implantation into Nonconventional GaN Structures

Abstract

Despite more than two decades of intensive research, ion implantation in group III nitrides is still not established as a routine technique for doping and device processing. The main challenges to overcome are the complex defect accumulation processes, as well as the high post-implant annealing temperatures necessary for efficient dopant activation. This review summarises the contents of a plenary talk, given at the Applied Nuclear Physics Conference, Prague, 2021, and focuses on recent results, obtained at Instituto Superior Técnico (Lisbon, Portugal), on ion implantation into non-conventional GaN structures, such as non-polar thin films and nanowires. Interestingly, the damage accumulation is strongly influenced by the surface orientation of the samples, as well as their dimensionality. In particular, basal stacking faults are the dominant implantation defects in c-plane GaN films, while dislocation loops predominate in a-plane samples. Ion implantation into GaN nanowires, on the other hand, causes a much smaller density of extended defects compared to thin films. Finally, recent breakthroughs concerning dopant activation are briefly reviewed, focussing on optical doping with europium and electrical doping with magnesium.

1. Introduction

Group III nitride semiconductors, namely, GaN, AlN, InN and their ternary and quaternary compounds, became famous beyond the research community due to their application in light-emitting diodes (LEDs) and laser diodes for lighting and data storage. Their fame culminated in the attribution of the Nobel Prize in Physics in 2014 to Isamu Akasaki, Hiroshi Amano and Shuji Nakamura for the development of the blue light-emitting diode (LED) in the 1990s [1]. Today, nitride LEDs are the base technology for highly efficient white solid-state lighting and advanced photonics applications [2]. Many of these new applications rely on non-conventional structures, such as those incorporating non-polar nitrides (i.e., with a- or m-plane surface instead of the conventional c-plane structures) or nanowires (NWs) [3,4,5]. In this context, non-polar devices avoid efficiency losses due to the quantum-confined Stark effect. The electrical fields induced in c-plane quantum wells due to the polar crystal structure, reduce the electron–hole overlap and, thus, the probability of radiative carrier recombination, in particular, for long-wavelength emissions in the green and red spectral regions [6]. The advantages of devices based on NWs include their high crystalline quality compared to heteroepitaxial layers since the 3-dimensional (3D) structures allow for efficient strain relaxation minimizing the number of dislocations, as well as improved light extraction [7].

Less known but equally promising is the potential of these wide bandgap semiconductors for the development of high-temperature, high-power and high-frequency electronic devices [8]. Thanks to their wide bandgap and high electron mobility, III-nitrides are expected to outperform silicon power devices in terms of breakdown voltage and on-resistance, respectively, opening the possibility of huge energy savings. Novel applications of GaN power devices range from the automotive sector to smart grids. High switching speeds enable technologies, such as LiDAR (light detection and ranging) and space communications. Nevertheless, new design and processing techniques need to be developed and optimised for GaN power electronics to become competitive on a large industrial scale.

To date, doping for commercial devices is carried out during the growth. However, for the insertion of devices in integrated circuits and a more versatile device design, selective-area doping techniques, such as ion implantation, are highly desirable. While ion implantation is a key technology in the silicon industry, it is far from being established for the processing of nitrides. This is mainly due to the difficulties in efficiently removing implantation damage during post-implant thermal annealing. In fact, the main drawback of the ion implantation technique is the large number of lattice defects that are formed when a material is bombarded with energetic and heavy ions. Thermal annealing is required to remove defects and activate the implanted dopants. Despite the difficulties in annealing GaN, many breakthroughs were reported quite recently by applying ion implantation for the processing of diodes, transistors and even laser structures [9,10,11,12,13]. In many of these applications, ion beams are used for implant isolation [10,11,14] and to form current apertures [12,13] (thus taking advantage of implantation defects to modify the materials’ properties), but much progress has also been made concerning p-type doping using magnesium implants [15,16,17]. Thanks to these advances, several device designs were realised, including ion implantation steps, both in lateral [9,17,18,19] and vertical devices [20,21,22], and metal–oxide–semiconductor field-effect transistors (MOSFET) were fabricated using an all ion implantation process [23,24]. Of particular interest are vertical devices, which allow for higher breakdown voltages and overall smaller structures, for example, in current aperture vertical electron transistors (CAVETs). The current blocking layer (CBL) is a key feature in these devices, providing a barrier to current flow in the vertical direction, leaving only a small aperture for vertical flow. Si implantation was used to define this aperture in a p-GaN CBL, which was either fabricated during growth or by Mg-implantation [20,21]. Current apertures defined using ion implantation were also applied successfully for the fabrication of vertical cavity surface-emitting lasers (VCSELs) [12]. Al implantation was used in this case to define the vertical current path, showing that implantation defects are sufficient to block the current in the implanted region. Implanting impurities, such as Fe, leading to deep levels further improves the isolating properties and temperature stability [10,14]. Indeed, implant isolation was used in several III-nitride devices, such as for edge termination in GaN Schottky diodes and GaN/Si p-n diodes, allowing for suppressing leakage along the side walls and increasing the breakdown voltage [25,26]. In several other reports, implant isolation was further used to reduce leakage via the substrate [27,28]. Interestingly, detailed carrier profiling suggests that ion implantation deactivates carriers far beyond the depth of the damaged region seen in transmission electron microscopy images or predicted using Monte Carlo simulations [29].

First reports on ion implantation in GaN date back to the 1970s [30]. However, intense research only started around the millennium change when high-quality GaN heteroepitaxial films became readily available. Early research on ion implantation focused on such c-plane GaN films and was described in several comprehensive reviews [31,32,33,34]. The main conclusions can be summarised as follows: Due to efficient dynamic annealing effects, point defects can already recombine during the implantation keeping implantation damage low. In fact, a complete loss of single-crystalline order is only observed for implantation at cryogenic temperatures and very high fluences [35] or due to chemical effects, e.g., for fluorine implantation [36]. Nevertheless, the high mobility of point defects, even at low temperature [37], also leads to their interaction with each other or with native defects and, consequently, to the formation of thermally very stable point defect clusters and extended defects, such as stacking faults and dislocation loops. Indeed, these are the dominant types of defects in ion-implanted GaN, as well as in AlN and AlGaN alloys [38,39,40,41,42]. Extended defects and defect clusters are difficult to remove during post-implant annealing and can act as traps for free charge carriers, hampering the proper operation of devices [16].

In the past decade, fundamental research on implantation effects in III-nitrides has shifted from simple c-plane GaN films to more complex structures, including films with different surface orientations, ternary compounds or nanostructures. This review summarises the contents of a plenary talk, given at the Applied Nuclear Physics Conference, Prague, Czech Republic, in 2021, and focuses on recent results, obtained at the Laboratory of Accelerators and Radiation Technologies (LATR) of Instituto Superior Técnico (Lisbon, Portugal) [43], on ion implantation into such nonconventional GaN structures. These include non-polar GaN films grown on the a- and m-plane (as compared to the conventional c-plane) of the wurtzite lattice [44], as well as nanostructures, such as nanowires [45]. Although a complete review would go beyond the scope of this paper, some of the most promising reports on dopant activation found in the literature are also highlighted. In particular, ultra-high temperature annealing efficiently activates optical rare earth dopants, as well as electrical dopants.

2. Materials and Methods

Figure 1a shows the thermodynamically most stable wurtzite crystal structure of GaN and the three major crystallographic planes, namely, (11$\stackrel{-}{2}$0), (0001) and (10$\stackrel{-}{1}$0), or the a-, c- and m-plane, respectively. For the present study, 3–10 μm thick GaN films were grown on sapphire or SiC substrates using metal–organic vapour-phase epitaxy (MOVPE) or hydride vapour phase epitaxy (HVPE) (Figure 1b). For the study, presented in Section 3.1, ion implantation of 300 keV Ar ions and in situ Rutherford backscattering spectrometry/channelling (RBS/C) analysis using 1.4 MeV He⁺ were performed at 15 K at the University of Jena, Germany [44].

NWs were grown using molecular beam epitaxy (MBE) on silicon substrates. Figure 1c shows a typical scanning electron microscopy (SEM) image of an as-grown sample. The NWs were grown vertically aligned along the c-axis and show m-plane side facets [46]. They had typical lengths of 2.5 μm and the width of the NWs was around 50–100 nm.

As an example of optical doping, NWs, as well as thin-film c-plane GaN, were implanted with 300 keV Eu ions (Section 3.2 and Section 3.3). Besides RBS/C, structural characterisation was performed using transmission electron microscopy (TEM), and X-ray diffraction (XRD) and optical activation of Eu³⁺ ions were studied using photoluminescence (PL) and nano-cathodoluminescence (CL).

3. Results and Discussion

In the following, some case studies on ion implantation in GaN are discussed. In Section 3.1, a detailed study of argon implantation in GaN thin films with different surface orientations illustrate the complex dynamics of defect accumulation in GaN due to efficient dynamic annealing effects [44]. Section 3.2 discusses ion implantation in GaN NWs [45]. Finally, Section 3.3 addresses doping via ion implantation and annealing. It focuses on optical doping via Eu-implantation, which has been an important research topic at LATR, Instituto Superior Técnico (Lisbon, Portugal), but recent advances found in the literature concerning electrical doping are also highlighted.

3.1. Implantation Damage Formation in a-, c- and m-Plane GaN

Typical RBS/C spectra for the a-plane GaN sample implanted using different Ar-fluences are shown in Figure 2a. The backscattering yield of the as-grown sample is very low. Minimum yields well below 2% were measured close to the surface for all three sample materials, evidencing excellent crystalline quality along the growth direction. As the fluence increases, the backscattering yield also increases, revealing the displacement of Ga-atoms from their lattice site. High dechannelling yields beyond the implanted region (below ~900 keV in Figure 2a) hint at the formation of extended defects [47] similar to what is observed for c-plane GaN [48]. After correction for this dechannelling yield, the fraction of displaced lattice atoms was extracted as a function of depth [44]. The maximum value of this relative defect level is plotted in Figure 2b as a function of the fluence for all three sample materials. Strong dynamic annealing effects are obvious from these damage build-up curves. Implantation damage does not increase linearly with the fluence but occurs in several regimes (marked in Figure 2b).

The curves are well fitted (lines in Figure 2b) using the defect accumulation and amorphisation model by Hecking [49]. According to this model, the linear increase of damage in regime I is due to defect formation in well-separated collision cascades. In regime II, the individual collision cascades start to overlap, leading to an increased recombination of vacancies and interstitials and an almost stable defect level. In regime III, the formation of extended defects and clusters leads to a strong increase in the defect level, which saturates again in regime IV. A full loss of the single crystalline order is observed in regime V for very high fluences above 1 × 10¹⁶ at/cm². Similar curves were measured by several groups in c-plane GaN and AlGaN [38,50,51], nevertheless, it should be noted that, in some cases, chemical effects, as well as cascade density, can considerably change the defect accumulation in GaN [36,52,53].

Concerning the effect of different surface orientations, regimes I and II are similar for the three studied orientations and defect profiles agree well with ballistic models, such as Monte Carlo SRIM (stopping and ranges of ions in matter) [54] simulations [44]. For higher fluences (regimes III and IV), distinct behaviours are observed for the three samples with strikingly lower defect levels in a-plane GaN (note the logarithmic scale in Figure 2b). This effect was also confirmed for heavy ion gold and rare earth ion implantation [55,56]. In this regime, a large number of extended defects and clusters are formed. Two different scenarios could explain the surprising difference in the RBS/C results. On the one hand, it is possible that the same defects are formed but the efficiency with which they are detected depends on the chosen channelling directions. Thus, displaced atoms may be shadowed by the atomic rows along a certain direction and exposed to the analysing beam for channelling along another axis.

On the other hand, the low backscattering yield can reflect a real effect, i.e., the number of defects in a-plane GaN is considerably lower than in c- and m-plane GaN and/or different defect types are present.

To differentiate between these two scenarios, TEM was performed on samples corresponding to regime IV, implanted using a fluence of 8 × 10¹⁵ at/cm². Examples for TEM images of c- and a-plane samples are shown in Figure 3. They reveal that indeed the nature of extended defects is different in the two samples. While basal stacking faults, with the typical light contrast parallel to the surface (Figure 3a), are dominant in c-plane GaN, a-plane GaN shows a high concentration of dislocation loops with a [11$\stackrel{-}{2}$0] character (Figure 3b). The defect profiles derived from RBS/C were superimposed onto the TEM images and show a good agreement with the observed contrast for the m-plane sample. For the a-plane sample, pronounced surface damage was identified using both techniques. In deeper regions, TEM shows a layered structure with dislocation loops clearly observed until a depth of approximately 250 nm, while the region between 250 and 350 nm depth shows the typical contrast of defect clusters. This distinct defect morphology is not resolved in the RBS/C spectra. The extracted defect profile corresponds to atoms that are displaced from their lattice sites. These displaced atoms, mainly due to defect clusters, lead to direct backscattering of the alpha particles, while the distortion of the lattice by dislocation loops gives rise to a high dechannelling yield. In future work, it will be interesting to implement a model for these defects in a Monte Carlo code, which would allow for a more quantitative analysis of RBS/C spectra in the presence of different defect morphologies [57]. Nevertheless, the depth range at which defects are visible was the same for both techniques. It is worth mentioning that a similar layered structure was observed using TEM in ion-implanted c-plane GaN, where stacking faults were formed at the depth of nuclear energy deposition, while deeper regions were free of stacking faults but instead present defect clusters [58].

3.2. Ion Implantation in GaN Nanowires

In Section 3.1, the complex damage build-up processes, occurring in GaN thin films upon ion implantation and the striking differences in defect dynamics for different surface orientations, were discussed. The influence of the surface is expected to have an especially strong impact on the damage mechanisms in nanostructures with a high surface-to-volume ratio. Here, 300 keV europium implantation into GaN NWs was studied [45].

The NW axes were tilted away from the incoming beam direction and the sample rotated in the azimuthal direction during the implantation in order to avoid ion implantation into the growth substrate, warrant a more homogeneous distribution of ions and suppress bending of the NWs due to defect induced strain. The c-plane GaN thin films were implanted simultaneously for comparison.

The distribution of Eu ions and implantation defects in NWs is more complex than in thin films. The results of Monte Carlo simulations, showing the Eu and defect distributions after implantation of 3 × 10¹⁵ at/cm², are presented in Figure 4 for two different types of NWs: (a) 50 nm diameter and an areal density of 2 × 10⁹ cm⁻², and (b) 100 nm diameter and an areal density of 7 × 10⁹ cm⁻². It is clear that the final distributions depend critically on the density of the NWs and their diameter. In particular, the side facets are shadowed by surrounding NWs and the distribution of ions and defects with respect to the NW axis depends on the NW diameter. For the same implantation parameters, the first case of thinner and less dense NWs (Figure 4a) leads to homogeneous doping and defect formation in the entire NW. In contrast, for the second case of thicker and denser NWs, Eu and damage distributions are less homogeneous. In the latter case, Eu ions are confined to the top ~500 nm. A high Eu concentration is found in the top 100 nm of the NW due to ions entering through the top surface. A similar doping depth is expected for the simultaneously implanted GaN thin films. The ion range measured from the lateral surface of Eu entering through the side facets is smaller than the radius of the NWs leading to low Eu and defect concentration near the NW core and higher concentrations close to the rim.

RBS/C is not suitable for extracting implantation defect profiles in NWs. With a beam diameter of ~1 mm, the spectrum would average over a large number of NWs and due to the small tilts between individual NWs, the ion channelling is affected [59]. To study the effect of implantation on structural properties, XRD was employed. XRD allows for measuring the lattice parameters of crystals with very high resolution and is thus the technique of choice to assess strain induced by implantation defects. It is now commonly accepted that strain is one of the main driving forces for defect transformation from point defects to extended defects and eventually to amorphization [38,60,61]. Indeed, the results, presented in Section 3.1, showed that these defect transformations take place even at 15 K, suggesting that thermal effects are negligible, which agrees with the almost identical damage build-up curves at 15 K and room temperature [35].

XRD 2θ−θ scans of the 0004 reflection (with θ being the Bragg angle, i.e., the angle between the incoming beam and the diffracting crystal plane) for GaN thin films and NW samples implanted using different Eu fluences are presented in Figure 5a,b, respectively [45]. The position of the main Bragg peak, corresponding to the c-lattice parameter of the as-grown material, is marked by a dotted line. It is also seen in all implanted samples arising from the unimplanted deeper layers of the crystals. For the thin films, ion implantation leads to the formation of a satellite peak at a smaller 2θ angle than the main Bragg peak. This satellite peak reveals an expansion of the c-lattice parameter of the wurtzite structure due to strain caused by the implantation defects. With increasing ion fluence, the satellite peak shifts to lower angles/higher strain until, above a fluence of 5 × 10¹⁴ at/cm², the XRD pattern remains unchanged. In this regime, the scattered intensity is very low due to the heavily damaged crystal and the XRD technique becomes insensitive to further defect formation. The maximum strain values measured in these thin films are around 1.5% (see the upper x-axis scale in Figure 5a).

For the case of the NWs (Figure 5b), a similar expansion of the implanted volume occurs, leading to a satellite peak, which strongly overlaps with the Bragg peak of unimplanted material.

Two main differences should be pointed out compared to the results in thin films. First, the intensity ratio satellite/main Bragg peak is much higher in the NW samples than in the thin film references. This is expected since, as shown in Figure 4, the damaged volume compared to the volume of unimplanted material is higher in NWs where ions can enter through the side facets. Second, and more interestingly, the maximum strain induced in NWs stays below 0.5%. This value is much lower than in thin films, even those implanted using the lowest fluence of 1 × 10¹³ at/cm². These results suggest that strain is more effectively dissipated in 3D NWs than in bulk material.

TEM high-angle annular dark-field (HAADF) images of a single NW implanted using a fluence of 1 × 10¹⁵ at/cm² and annealed at 1000 °C are shown in Figure 6.

HAADF imaging is sensitive to the atomic number of the material (Z-contrast), thus the dark spots visible in Figure 6a are signs of thinner sample areas. Despite a high surface roughness induced by implantation/annealing, the high-resolution image in Figure 6b reveals the high crystalline quality of the NW core and the absence of the high density of extended defects typical for thin films under similar implantation and annealing conditions. In summary, one can conclude that implantation damage accumulation in NWs follows different processes than in bulk material. Strain in NWs is efficiently relaxed and one can speculate that this absence of strain inhibits the formation of the dense stacking fault network that is typical for ion implantation in GaN thin films.

3.3. Doping GaN via Ion Implantation

3.3.1. Optical Doping with Rare Earth Ions

Implanted GaN Thin Films

At LATR, Instituto Superior Técnico (Lisbon, Portugal), the most studied application of ion implantation in GaN is doping with optically active rare earth (RE) ions. Due to their partially filled 4f electron shells, which are shielded by outer filled shells, the intra-4f transitions of trivalent RE give rise to narrow and temperature-stable emission lines covering a wide range of the electromagnetic spectrum from infrared to ultraviolet [62]. In particular, by doping with different RE ions, light emission in the entire visible range can be achieved using Er or Tb for green, Eu or Pr for red and Tm for blue emissions. The work focused mostly on the red emission from Eu³⁺ since conventional III-nitride LEDs show very low efficiency for these long wavelengths. Recently, first successful realisation of an Eu-implanted LED structure, based on AlN nanowires with an axial p-n junction was reported, albeit still requiring relatively high voltages [63]. The first low-voltage GaN:Eu LED was reported in 2008 and was based on GaN:Eu layers doped during growth [64]. Although in situ doping avoids the drawback of implantation damage, it is prone to phase separations or clustering [65] and does not support lateral patterning.

Early work on RE implantation in GaN was reviewed in [66]. Besides the structural investigation of implantation damage, the main fields of interest were the lattice site location of RE ions in GaN, as well as their optical activation via thermal annealing. Concerning the former, one could establish that Eu is preferentially incorporated into the substitutional Ga-site or slightly displaced from it [67,68]. However, almost complete substitution is only achieved for low implantation fluences since they avoid high damage levels [45].

Concerning thermal annealing, it soon became clear that high annealing temperatures are the key to efficient optical activation of Eu³⁺. However, without appropriate protection of the surface, nitrogen starts to out-diffuse at temperatures above 800 °C, leading to the dissociation of GaN [69]. Indeed, using simple protection by applying a proximity cap (i.e., placing another GaN sample face to face with the implanted sample) and low-pressure N₂ annealing atmospheres, the maximum achievable temperature was ~1000 °C [70,71]. Further research on different annealing conditions and surface protection led to several new “generations” of GaN:Eu samples with an intensity increase of the Eu³⁺ emission of three orders of magnitude. Figure 7 gives an overview of the photoluminescence spectra and emission intensity of these various sample generations. Generation 1 corresponds to Eu-implanted GaN annealed in N₂ at ambient or low pressure. In these conditions, both furnace annealing and rapid thermal annealing can be performed up to a maximum temperature of about 1000 °C [71].

In generation 2, samples are protected using a ~10 nm thick, epitaxially grown AlN cap, allowing for annealing up to 1300 °C, above which, cracks form in the AlN cap, leading to the dissociation of the underlying GaN [72,73]. Generation 3 employs ultra-high nitrogen pressures (in the GPa range), allowing for high-temperature high pressure (HTHP) annealing up to 1400 °C [74,75]. It is interesting to note that the Eu profile is remarkably stable after annealing in these extreme conditions. However, for very high fluences, a redistribution of implanted RE is often observed [76]. For the case of Eu-implanted GaN, RBS/C results suggest that the change in the Eu profile is due to the high defect level, in particular, the dissociation of the heavily damaged surface layer [71], but reaching the solubility limits may play a role at very high annealing temperatures.

Generation 4 allows for doubling the room temperature Eu emission intensity once more using the same 1400 °C HTHP annealing but implanting into p-type Mg-doped GaN. This increase is attributed to the formation of a new optically active centre composed of an Eu–Mg cluster with a high excitation cross-section at room temperature for excitation above the GaN bandgap [77]. Finally, in generation 5, an improved furnace set-up allows for HTHP annealing at temperatures as high as 1600 °C [78].

Implanted GaN Nanowires

In the GaN NWs, discussed in Section 3.2, implanted Eu could be optically activated using 1000 °C rapid thermal annealing [45]. Higher annealing temperatures or longer annealing times are difficult to achieve without damaging the NWs due to their high surface-to-volume ratio, which facilitates the out-diffusion of nitrogen. Figure 8 shows nano-cathodoluminescence (CL) measurements on a single implanted NW performed in a transmission electron microscope [45]. The CL spectra in Figure 8a, taken at different spots along the NW length, show the typical Eu³⁺ red emission at ~620 nm from the tips of the NWs, while the base shows broad band yellow luminescence (YL) and near-band-edge emission due to donor–acceptor pairs (DAP). The mapping of the Eu-related luminescence (Figure 8c) agrees well with the MC simulations shown in Figure 4, with the most intense emission arising from the top 500 nm of the NWs where most of the Eu ions are coming to rest.

Interestingly, the luminescence intensity is higher in these NW samples than in thin-film samples implanted and annealed simultaneously. Furthermore, in thin films, a strong luminescence quenching is observed with increasing implantation fluence due to the high density of defects. In NW samples, the luminescence intensity increases monotonically with the implantation fluence, however, not at the pace expected from the increase in Eu-concentration [45]. Indeed, the measurement of the lattice site location of Eu in GaN films and nanowires using the extended X-ray absorption fine structure (EXAFS) technique revealed that Eu is predominantly incorporated on undisturbed substitutional sites only for low fluences (~1 × 10¹⁴ at/cm²). For higher fluences, an increasing number of Eu ions is found in a highly disturbed EuN-like environment where Eu is surrounded by six nitrogen atoms (instead of four, as expected for a substitutional Ga-site in GaN) [45].

3.3.2. Electrical Doping

The main electrical dopants in GaN are Si and Mg for n- and p-type doping, respectively. Si is a shallow donor, and activation percentages of ion-implanted silicon above 90% were reported quite early, typically employing a proximity cap or oxide or nitride capping layers to allow for annealing above 1000 °C [79,80]. In contrast to Si, Mg is a relatively deep dopant (with an acceptor level of 0.160 eV above the valence band), which theoretically limits the ionisation at room temperature to values smaller than 10%. Only recently, the activation of ion-implanted Mg close to this theoretical limit was reported [81,82]. Emission channelling experiments showed that implanted Mg preferentially occupies substitutional Ga sites, suggesting that low activation fractions are due to compensation by defects rather than due to interstitial incorporation [83]. Therefore, like for the optical activation of Eu, discussed in Section 3.3.1, the way to success passes through an optimisation of the annealing conditions where temperatures well above 1000 °C are required. Anderson et al. [81] used an AlN cap combined with multi-cycle annealing with a peak temperature around 1350 °C to achieve activation close to the theoretical limit. Similar multi-cycling annealing was used by Meyers et al. who furthermore showed that Mg/N co-implantation yields improved p-type conductivity, presumably due to the suppression of nitrogen vacancy formation [84]. The conversion of a low-dislocation density n-type GaN layer to p-type using Mg-implantation and the formation of a p-n junction was achieved after annealing at 1230 °C using a SiN cap [85]. Interestingly, activation of implanted Mg in nitrogen polar GaN was achieved without the need for protective capping or ultra-high pressures and allowed the fabrication of p-n junction diodes [15]. This was attributed to the increased thermal stability of this surface termination, where every N-atom at the surface is bonded to three Ga-atoms.

Similar to the Eu-activation discussed in the previous section, the best results of Mg activation were achieved by HTHP annealing at temperatures as high as 1480 °C. Acceptor activation and mobilities were shown to be close to those of epitaxial films and the surface quality was improved in comparison to AlN-capped samples [82]. In contrast to what might be expected, the high activation fraction does not necessarily imply defect-free material. Indeed, HTHP annealing at 1300 °C and 1480 °C led to the formation of extended defects after annealing. For the lower temperature, the dominant extended defect types are interstitial-type stacking faults and polarity inversion domains with Mg segregation at their boundaries. For the higher temperature, vacancy-type extended defects were observed instead [86]. The improved p-type conductivity for the higher temperature was then explained by a lower density of vacancies and their complexes which would compensate Mg acceptors, as well as reduced Mg segregation.

Eu profiles, described in Section 3.3.1, remained stable during HTHP annealing up to ~1450 °C within the uncertainties of the RBS technique. In contrast, acceptor dopants, such as Mg and Be, were shown to diffuse at these temperatures [87]. Narita et al. [88] suggested that hydrogen plays an important role in the Mg diffusion process. They determined a relatively low diffusion constant of D_Mg–H = 7 × 10⁻¹⁴ cm²s⁻¹ for HPHT annealing at 1200 °C by studying diffusion from an in situ doped GaN:Mg layer into homoepitaxial GaN with a low dislocation density. For heteroepitaxial samples with higher threading dislocation density, the diffusion of implanted Mg during annealing at 1100 °C using an AlN/SiN cap is strongly dependent on the dopant concentration, with it being faster for concentrations above 1 × 10¹⁹ at/cm³. Moreover, for these high concentrations, annealing led to the formation of Mg-rich clusters [89]. Under similar annealing conditions (less than 60 min at 1100 °C using AlN-capped heteroepitaxial GaN), no diffusion of implanted Si was observed [90]. However, diffusion coefficients for silicon reported in the literature vary by several orders of magnitude and depend on the capping layer, annealing conditions and sample quality; furthermore, several distinct diffusion mechanisms can take place simultaneously [90,91,92]. Nevertheless, active dopant distributions were shown to be in good agreement with the total Si depth profiles [93].

Understanding and quantifying the diffusion of electrical dopants in GaN during post-implant annealing will be an important prerequisite for the use of ion implantation in device processing at an industrial scale. It would be interesting to study annealing techniques out of thermodynamic equilibrium, such as flash-annealing or laser-annealing in order to suppress the diffusion of dopants. Nevertheless, the recent progress in the activation of p- and n-type dopants after implantation is encouraging for the development of new device designs based on ion implantation.

4. Conclusions

In general, GaN is considered a very radiation-resistant semiconductor. This is confirmed by the strong dynamic annealing and high amorphisation thresholds revealed in the work described above and in numerous publications. However, defect dynamics upon ion implantation are very complex, leading to a mixture of different defect types, including point defects and their clusters (ranging from very small, e.g., V_GaV_N di- and tri-vacancies, to large clusters), as well as stacking faults and dislocation loops. This review highlights some examples of how the sample structure itself can influence the damage accumulation processes upon ion implantation. In particular, it was shown that the surface orientation of GaN layers influences the nature of extended defects created for high-fluence implantation. Basal stacking faults in the c-plane are dominant in c-plane GaN, while dislocation loops with a [11$\stackrel{-}{2}$0] character predominate in a-plane material. The latter also shows lower levels of randomly displaced atoms, suggesting more efficient dynamic annealing. Interestingly, extended defect formation upon ion implantation is suppressed in GaN NWs, which also shows lower defect-induced strain levels. Further defect transformation and dopant diffusion can occur during annealing, in particular, at the high temperatures needed for dopant activation in GaN. The effect of all these defect types on optical and electrical properties needs to be thoroughly understood in order to clear the way for the use of ion implantation as a routine processing technique for III-nitride devices. In this process, it should be kept in mind that there are “good” and “bad” defects. For doping, typically, the objective will be complete annealing of implantation defects or their transformation into electrically inactive defects. For implant isolation, on the other hand, the formation of thermally stable defects is a prerequisite to ensure stable properties, even after further device processing steps at high temperatures.