Analytical electron microscopy study on gallium nitride systems doped with manganese and iron

Modulated structures of gallium nitride (GaN) doped with transition metal ions (here Fe, Mn) are investigated by analytical (scanning) transmission electron microscopy to gain insight into the structural arrangement and chemical composition of the material, known to be critically correlated to the magnetic response and hence the functionality of these technologically relevant systems. Three classes of samples are considered: (i) homogeneous (dilute) (Ga, Mn)N; (ii) δ-Mn-doped (Ga, δ-Mn)N and phase separated (Ga, Fe)N, containing Fe-rich nanocrystals. The combination of various microscopic techniques employed, allows for a quantitative determination of the distribution of the magnetic ions in the samples, providing essential information on the structural and chemical asset of these systems.


Introduction
Gallium nitride (GaN) based material systems are currently playing a key technological role, particularly in the field of optoelectronics and lighting for the future. It is expected that the addition of magnetic capabilities through doping with transition metals or rare earths, will extend the functionality of these materials systems to applications in spintronics, quantum information and magneto-optics [1][2][3][4]. A number of studies carried out over recent years, have highlighted how the distribution of magnetic dopants in the semiconducting matrix substantially affects the structural, magnetic, electrical and optical properties of the resulting material. In particular, the recent advancements in the fabrication and characterization of magnetically doped nitrides, have made clear that depending on the growth conditions and eventually on the post-growth treatment, the introduction of magnetic ions can essentially result in: (i) a dilute (homogeneous) system [5]; (ii) a chemically inhomogeneous system (chemical phase separation) [6]; (iii) a crystallographically inhomogeneous system (crystallographic phase separation, secondary phases, embedded nanocrystals) [7]; (iv) a system containing functional magnetic complexes [8].
With these premises, the control over the formation, structure, arrangement and effect of inhomogeneity on the Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. chemical and physical behavior of the host material is essential for the design and functionality of the next generation of devices.
Recently, it has been demonstrated that superexchange interactions account for ferromagnetism in (Ga, Mn)N in the absence of free carriers [5,9]. In order to reliably exploit this material system in functional devices such as e.g. a new generation of spin filters, a control of the effective Mn incorporation in GaN and of the interfaces of (Ga, Mn)N in heterostructures is necessary [10].
Particularly in the case of embedded magnetic nanocrystals, it is essential to establish the correlation between the growth parameters and the local chemical composition and size of the nanostructures, considering that these figures have been shown to determine the magnetic properties of the whole system. For instance, in the case of Fe-rich nanocrystals embedded in GaN, it is found that by modifying the composition of the nanostructures, it is possible to tune the system from ferromagnetic to antiferromagnetic up to room temperature. Here the formation of magnetic Fe-rich nanocrystals paves the way to the realization of non-volatile flash-memories and tunable ferromagnetic/antiferromagnetic spintronic devices based on these hybrid systems [11].
In this work, electron microscopic techniques are applied to dilute (Ga, Mn)N systems, to two-dimensional (2D) layers of (Ga, Mn)N embedded in GaN, and to planar arrays of Ferich nanocrystals embedded in GaN.

Material and methods
All the samples considered are fabricated by metalorganic vapor phase epitaxy on c-sapphire, according to the procedures reported elsewhere and have been thoroughly characterized with respect to their magnetic and optical properties via both conventional and synchrotron radiation-based techniques [5][6][7]. The microscopic and chemically-sensitive investigations of the structures are the focus of this work.
For the dilute (Ga, Mn), 2D (Ga, Mn)N layers and planar array of Fe-rich nanocrystals, schematic cross sections and their scanning transmission electron microscope (STEM) image counterparts (oriented along the GaN [11][12][13][14][15][16][17][18][19][20] zoneaxis) are shown in figure 1. In the homogenous Mn co-doping process, a layer thickness of 700 nm is realized with a nominal concentration of 1.55 at% of Mn atoms within GaN. The 2D (Ga, Mn)N layers, having a total thickness of 150 nm, are obtained through δ-doping, a process realized by periodically providing Mn flow during GaN growth. Hereby, the starting point has been a sample with an averaged concentration of nominal 1.30 at% of Mn atoms within GaN. Samples of (Ga, Fe)N above the solubility limit of Fe, with crystallographic and chemically phase separated regions present, are investigated in addition. The growth conditions are given in [8]. The nominal Fe concentrations are 0.25% and 0.5%, respectively, found by secondary ion mass spectrometry (SIMS) in each case.
All cross-section specimens have been fabricated by mechanical polishing utilizing a semi-automated 8″ Platen MultiPrep system from Allied High Tech. Ion-thinning has been performed by using a PIPS-system (Gatan Inc.) with Ar ions at 3, 4 kV voltage and 4°incident angle until perforation of the film. The final polishing step has then been performed on a Gentle Mill system (Technoorg Linda) at 500 eV with an incident angle of 14°for 30 min. This procedure has been repeated typically two to three times while observing the milling quality in the transmission electron microscope (TEM) until the specimen surface has exhibited a sufficient amorphous-free surface at the desired specimen position as visualized in figure 2, exemplified for the nanocrystal case.
The TEM and STEM investigations have been performed on different instruments depending on the equipment needed. Specimen orientations have been either along GaN [11][12][13][14][15][16][17][18][19][20] or close to this orientation, unless the sample has been tilted towards the energy dispersive x-ray spectroscopy (EDXS) detector, merely for maximum signal-collection. Conventional bright field TEM imaging investigations have been carried out on a Tecnai 200 kV FEG microscope from FEI, equipped with a Gatan GIF Tridiem electron energy loss spectroscopy (EELS) spectrometer [12] and an EDAX Phoenix Si(Li) x-ray detection system. This system has been used to record the energy filtered transmission electron microscopy (EFTEM) images as jump-ratio maps from the Mn L 23 edge with a slit-width of 30 eV collected at 620 eV and 655 eV. Furthermore, this system has been utilized for the quantitative analysis of the Fe-rich nanocrystals via EELS and EDXS reported later in this work. X-ray spectra have been subjected to a ζ-factor analysis, using the software-package written by Watanabe [13][14][15]. For the evaluation of concentrations, the peak intensities Kα signals of N, O, Ga and Fe have been extracted by removing the background through bilinear interpolation, left and right of the respective line. Data acquisition has been performed by scanning a region of interest (50 × 50 pixels) across the central part of the individual nanocrystals. Hereby, a dwell-time of 15 μs and a beamcurrent ranging from 150 pA to 250 pA have been used, accounting better for beam-induced modifications of the nanocrystal. Each nanocrystal has been measured only once to avoid erroneous measurements on possibly beam-modified material. In the case of visible heterogeneities in the crystal, individual, spatially resolved spectra have been recorded as spectrum images. The specimens investigated (with nominal Fe concentration of 0.25% and 0.5%) have been tilted 15°t owards the x-ray detector to avoid strong channeling effects and maximize signal collection efficiency.
The collection angles for EELS are 15.9 mrad for a GIF Tridiem installed at the Tecnai 200 kV FEG instrument (utilized for nanocrystal quantification) and 40.4 mrad for a DualEELS capable GIF Quantum [16] installed at the TITAN instrument operated at 300 kV (utilized for both Mn-doped GaN quantifications), respectively. Processing of the EELS data has been carried out with DigitalMicrograph (Gatan Inc.) applying regular EELS quantification schemes, however, accounting also for multiple scattering by low-loss deconvolution. For the quantification of the EELS signals the K edges of N (as well as O) and the L 23 edges of Ga and Fe have been collected. The background is modeled by a power-law function and Hartree-Slater cross-sections are employed for conversion into concentrations.
High resolution STEM investigations have been carried out on a 60-300 kV probe-C s -corrected TITAN microscope. Corrector tunings have always been performed until an aberration-free zone of 19 mrad or better has been reached, giving a spatial resolution better than 1 Å. Annular bright field (ABF) imaging has been utilized as an alternative technique in order to harness its sensitivity for structural changes and its possibility to image lighter N atoms [17][18][19][20][21]. To obtain ABF images it has been paramount to avoid carbon surface contamination and to reduce surfaceamorphization with the procedures described earlier. In our setup the ABF image has been registered with an annular detector as described in [20], collecting the direct beam diffraction disk within an angular range between 10 and 20 mrad for a beam convergence of 20 mrad. A beam current of of ∼80 pA and a pixel time of 20 μs have been used. Corresponding HAADF images have been recorded with an angular range from 26 mrad to 95 mrad.

Mn-doped GaN
To examine the structural homogeneity of the Mn-doped GaN samples, aberration-corrected HR-STEM investigations have been carried out. In figure 3(a) the interface between the GaN buffer and the doped part of the homogenously doped sample are shown. Dislocations originating at the interface substrate/ GaN-buffer and propagating into the Mn-doped layer can be observed (in addition to a minor contamination spot in the field of view). All analytical measurements herein reported have been carried out at least 100 nm away from the dislocations. Structurally, at the interface GaN/(GaMn)N no significant lattice mismatch or crystallographic phase separation despite the rather high doping level (figure 3(b)) has been found. A geometrical phase analysis (GPA) [22] of the GaN/(GaMn)N interface has revealed strain at the interface being in the order of <1% (maps not shown). A faint brightness difference between the GaN/(GaMn)N interface can be identified due to changes in average Z, because of Mn having been mixed homogeneously into the otherwise nominally undoped GaN. An increase of ∼2% in brightness can be identified when moving from GaN into (GaMn)N.
This intensity variation becomes more pronounced in figure 4, where the situation for the Mn δ-doped GaN is reported. A multi-layered structure is formed, leading δ-Mn regions. As evidenced in figure 4, the layers exhibit irregular periodicity and width.
High resolution HAADF and ABF images (figures 5(a) and (b)) have also been subject to a GPA analysis, and reveal no strain above 1%, indicating the coherent incorporation of Mn within the δ-doped region. A line-profile extracted from the HAADF signal ( figure 5(c)) shows a sinusoidal-type contrast variation between bright and dark lines as indicated by the dashed line. To shed more light onto the contrast variation, element specific EFTEM jump ratio imaging of the Mn L 23 edge has been carried out, as displayed in figure 6(a).
Here the onset of the δ-doping (white dashed line) and the subsequent layering is clearly visible. The reason for the enhanced Mn signal at the onset of the δ-doped region (surface) is not yet completely clear but deemed to stem from segregation and/or initial nucleation of the doped layer.
From the extracted profile ( figure 6(b)) of the Mn distribution image two aspects emerge: (i) approximately every 3 nm a layer has been introduced through the δ-process and its width is in the range 2-5 nm. (ii) Furthermore, when integrating perpendicular to the layers, an intensity ramp is detected. This is due to a thickness variation of the TEM foil, as determined by measuring the inelastic mean free path by means of EELS. The total relative thickness variation in the indicated area is going from 0.4 to 0.6.  Spectroscopy has been carried out at several randomly chosen sample positions within the Mn-doped regions of GaN, and the EELS Ga L 23 , N K, O K and Mn L 23 edges have been quantified. Oxygen-detected in noticeable but similar amounts in all specimens-has been accounted for in the data analysis. To study the origin of the oxygen signal and particularly its impact on the quantification, a cross-section extracted from an analyzed specimen has also been fabricated using a focused ion beam (FIB) system. Combined STEM EELS and EDXS line-scans from the protective platinum towards GaN (shown in the STEM image of figure 7(a)) across the former (specimen)  For the analysis of the Mn delta-doped GaN, locations have been chosen such to assure that the beam has been placed either on or in between two bright lines. The average Mn concentration on bright layers is found to be 2.9 ± 0.5 at%, with Ga: 47.3 ± 1.4 at%; N: 46.8 ± 2.2 at% and O: 3.2 ± 3.2 at%. The average value of Mn in between the nominally doped layers is 1.3 ± 0.5 at% (with Ga: 47.6 ± 1.6 at%; N: 44.6 ± 2.4 at% and O: 6.5 ± 3.7 at%), being about 50% less compared to the doped layers but not vanishing. The high concentration variation of both layer-types reflects the heterogeneous deposition of Mn contrasting the homogenously doped sample. Due to this strong local inhomogeneity between the lines and within a line itself it is not feasible to give a meaningful average dopant concentration as for the homogenous sample.

Fe-doped GaN
The perspective functionalities of transition metal doped nitrides are expected to be realized not only with the dilute/ homogeneous systems analyzed in the previous section, but also in phase separated materials such as e.g. (Ga, Fe)N. Particular growth conditions (as detailed in [6]) have led to a very narrow size-distribution of the nanostructures at a defined depth in the sample, as shown in figure 2(a), aggregating in a planar arrangement perpendicular to the growth direction. A comprehensive analysis of the structures and of their magnetic properties is reported in reference [23]. The HAADF imaging ( figure 8(a)) reveals a typical 2-side facetted nanocrystal exhibiting a length of 25 nm and a height of approximately 17 nm. Figure 8 To derive the composition of the nanocrystals, several of them (in samples with nominal doping concentrations of 0.25% and 0.5%) have been investigated and EELS as well as EDXS spectrum images have been recorded. Spectra collected from regions surrounding the nanocrystal (figure 9(b)) show no detectable presence of Fe confirming a confinement of the dopant within the nanocrystal. In a few non-typical cases, the nanocrystals display elemental inhomogeneity, as visible in figure 9. The Fe L 23 map (figure 9(a)) and the respective spectra (figure 9(c)) indicate intensity differences within the nanocrystal, pointing to ∼20% change in Fe concentration. This modulation is counterbalanced by an equal amount of Ga, where Fe is depleted (spectra not shown). Although these effects can be present, the overall variation in Fe content from crystal to crystal is by far larger.
Since absolute concentration measurements with each technique have shortcomings, we have performed data analysis in a relative manner with each technique comparing the two phases as shown in table 1. In this way systematic errors associated with low detection sensitivities and absorption for light elements in EDXS (N, O), and the uncertainties in the inner-shell ionization cross-sections for heavier elements (Ga, Fe) in the case for EELS can be avoided.
The general picture-averaging over all analyzed nanocrystals-is similar for the two techniques and shows similar trends. The Ga and N concentrations decrease to accommodate the large amounts of dopant and the extra oxygen. The Fe concentrations are on average about 2.5 times larger in the 0.5% Fe-doped sample than in the 0.25% Fe-doped one, yielding a ratio which is ∼25% higher than the nominal one (with slight differences in Fe signal between the techniques).    amount of oxygen in the 0.25% doped sample is as low as 0 at% and go up to 7 at%, whereas in the 0.5% specimen, the oxygen level reaches ∼9 at% and is never less than 2 at%.
On average, the O concentration is about twice as high in the 0.5% doped case as in the 25% one, including and assuming equal amounts of surface oxygen in both samples. The deviation between EELS and EDXS measurements is likely due to uncertainties in peak extraction at low energies for EDXS. The correlation between having more oxygen and increasing dopant concentrations falls short when looking at the individual nanocrystals, both in the 0.25% and in the 0.5% case Fe and O levels vary largely, indicating the formation of complex phases. In an attempt to separate out surface oxygen, in order to get an approximate value of the nanocrystal composition, a Ga 2 O 3 layer has been assumed (being the only source of oxygen) and a stoichiometric fractional intensity of Ga (scaled to oxygen) has been subtracted from the nanocrystal. By back-adding different amounts of GaN matrix to the result and upon comparison with assumed γ′-GaFe 3 N and γ′-Fe 4 N phases, no reasonable match has been achieved for a statistically significant number of nanocrystals hinting at a complex formation of mixed nitride oxides. To better account for the embedding of the nanocrystals in the matrix in a quantitative analysis, one has to make assumptions about the volumes of the crystal and correlate it with an absolute volumetric evaluation, e.g. by means of tomography. Alternatively, a particle extraction preparation process might be considered, taking into account that in this case of etching, consequent reduction of the nanocrystal size and modification of their stoichiometry can occur.

Conclusions
Modulated semiconductor structures require microscopic characterization techniques to unravel the distribution and chemistry of magnetic dopants, which critically determine the magnetic properties that are decisive for the performance of perspective devices such as spin filters (homogenous and 2D layers) or antiferromagnetic spintronic elements (nanocrystal based).
In this study, we have investigated structure and chemistry of different transition metal doped GaN systems, relevant for new technologies. Depending on the formation process, homogenous doping or layering in the case of Mn, and coherent formation of ordered Fe-rich nanocrystals in the case of Fe have been obtained. High-angle annular dark-field STEM as well as annular bright field in combination with analytical techniques such as EFTEM and quantitative EDXS  and EELS spectrum imaging have been chosen to investigate structure and homogeneity. Whereas the continuously doped samples show an even level of dopant material and negligible lattice modifications, the δ-doped material reveals a largely fluctuating amount of dopant, forming modulated, linear structures. Even in untreated regions, doping material is found to be significantly present. The introduction of a barrier layer is likely to foster the confinement of dopant to the desired regions.
In systems containing planar arrays of Fe-enriched coherently embedded nanocrystals, the modulation of growth parameters (substrate temperature and flow ratio) allows producing embedded nanocrystals with composition-and consequently magnetic response-on demand and studies in this direction are currently being undertaken. A success in this direction will give access to a system that can be tailored to be either ferromagnetic or antiferromagnetic at up to room temperature-depending on the phase of the nanocrystalsopening remarkable functionalities for both ferromagnetic and antiferromagnetic spintronics.