Correlating cathodoluminescence and scanning transmission electron microscopy for InGaN platelet nano-LEDs Open Access

Solid-state lighting using III-nitrides is crucial in energy-efficient tunable emission devices with applications in, e.g., household lighting, lasers, and self-emissive displays.^1–4 The III-nitrides are particularly important due to their direct bandgap, covering a broad emission spectrum from ultraviolet to near infrared.^5,6 Furthermore, the use of III-nitrides in quantum well (QW) and multiple QW (MQW) geometries with well-defined dimensions and compositions is a way to tune the emission and improve the emission control.⁷ By including QWs, the emission intensity is significantly improved due to an increased overlap between electrons and holes.⁸ In order to create efficient emitters in the visible spectrum, heterostructures of ternary InGaN with different In contents are needed. However, InN and GaN have significantly different lattice constants. This difference may introduce structural defects that can cause non-radiative recombination, which affects the emissive properties of the material and lowers the overall efficiency of the device.^9,10 The structures used in this study are the basis for devices targeting the next generation self-emissive displays, where pixel sizes of a few μm² or smaller are required.⁴ This requires the fabrication of nano-LEDs.¹¹ We have introduced a bottom-up approach to fabricate these nano-LEDs with potentially higher efficiency than conventional top-down devices on the nanoscale.¹² 

For III-nitrides, misfit and threading dislocations (TDs) are commonly formed at interfaces between heteroepitaxially grown nitride layers and substrates, between which the lattice mismatch is even larger than among the III-nitrides. The misfit dislocations are confined close to the interface but often result in TDs that propagate into the subsequent layers.¹³ The presence of TDs is known to play a crucial role in reducing the efficiency of conventional planar blue and UV emitting LEDs.¹⁴ Additionally, basal stacking faults (BSFs), a layer stacking order irregularity in the structure, can cause mismatched regions, for instance, caused by interface steps on a nonisomorphic substrate or multiple seeding with relative stacking mismatch. When regions of different stacking meet, in plane, a stacking mismatch boundary (SMB) is formed,¹⁵ which is also known to act as a non-radiative recombination site.^16–18 Once this boundary is formed, it extends throughout subsequent layers, unless countered by the introduction of a second BSF.

A correlation of an atomic structure with emissive properties in a semiconducting device is of critical significance in order to design routes to eliminate the main defects and improve efficiency. Such a correlation can be achieved by combining high-resolution imaging, for instance, scanning transmission electron microscopy (STEM), with spectroscopic imaging techniques, for instance, cathodoluminescence (CL).^19,20 However, STEM requires thin geometries where at least one dimension should be $≲$ 100 nm. Unless the sample is intrinsically thin, e.g., nanoparticles and nanowires, thinning of the structure is required. The thinning is commonly achieved using focused ion beam (FIB) milling and has to be performed carefully to achieve minimal thickness of obscuring material in line of projection to make it possible to image the feature of interest, while also limiting induced damage from the ions.

Previously, III-nitrides have been successfully prepared for STEM measurements following CL.²¹ However, the cutout geometry is most often in cross section, and in this geometry, only a limited number of defects will be observable. To capture growth related defects, such as SMBs and TDs that appear along the growth direction, plan-view observations are more useful.

Although CL can and has been successfully performed on already cut cross sections,²² the spectroscopic investigation should ideally be of the pristine device, prior to STEM sample preparation, to avoid inducing surface damage and altering the geometry of the photo-active regions.²³ Additionally, high-energy electrons from STEM analysis can have a detrimental effect on the luminescence efficiency.²⁴ This motivates a spectroscopy–thinning–imaging workflow when correlating the data.

Here, we correlate hyperspectral CL imaging with atomically resolved plan-view images of structural defects in InGaN platelet nano-LEDs, grown with a high-In-content InGaN QW for red emission. Though the platelets in this study lack the necessary doping to be fully functioning LEDs, the structure is similar to that of a full LED and will, therefore, be referred to as nano-LEDs in the text for simplicity. Specific defective platelets, identified by CL prior to thinning, are imaged, and the defects are characterized by plan-view STEM. We use a method of initial mechanical preparation, which allows for CL analysis of pristine platelets, followed by precise plan-view FIB thinning. The method allows for a correlative comparison of multiple platelets in plan-view STEM to their pristine CL emission. We find a clear correlation between non-radiating features in the QW emission and SMBs passing through the QW, which, hence, could greatly reduce the efficiency of the corresponding nano-LEDs.

The sample studied in this work (supplementary material Fig. S1) was prepared with the procedure described in Ref. 25, where there are three major steps: (1) InGaN pyramid growth; (2) c-plane formation by chemical mechanical polishing (CMP) of InGaN pyramids; (3) InGaN regrowth on the as-formed c-plane, including an InGaN QW. The InGaN pyramids were selectively grown on GaN/sapphire substrates with a patterned $Si 3 N 4$ mask by metal-organic vapor phase epitaxy (MOVPE) equipped with a 3 × 2 in. showerhead.²⁵ The array of InGaN pyramids was polished down from the top apex using CMP to obtain a top c-plane.²⁶ After CMP, the remaining structure was about 80–100 nm thick at the center. After HF cleaning, the as-polished samples were loaded back into the MOVPE reactor for InGaN regrowth on the top c-plane. The regrowth started with a lower InGaN barrier on the top c-plane, which is about 30 nm thick and has a similar indium content as the original InGaN pyramids (∼17% In). Then, the growth temperature was lowered by 15 °C to grow a 60 nm thick InGaN transition layer (TRL, ∼23%). This TRL can further enhance the indium incorporation into the subsequent growth of the InGaN QW and can also promote carrier injection to the QW.²⁷ After growth of the single InGaN QW (4 nm thick, ∼31%), an InGaN top barrier was grown with a thickness of about 30 nm and an indium content similar to the lower InGaN barrier.

The sample was mechanically prepared according to the plan-view preparation method described in detail by Palisaitis²⁸ and also shown in supplementary material Fig. S2, allowing for a wedge shaped part mounted to a TEM-grid. The wedge had pristine platelets on its surface close to the cleaved edge with a very thin substrate underneath. Instead of performing the FIB-cutting directly (as done by Palisaitis²⁸), the sample was thoroughly cleaned in acetone and IPA for 3 min each in an ultrasonic bath to get rid of glue residues and surface impurities. This was followed by a drying step in an oven for 20 min at 300 °C.

The CL study was performed in a conventional scanning electron microscope (SEM, Zeiss EVO MA 15). All data were recorded in the hyperspectral mode by Delmic's ODEMIS software and using Delmic's SPARC hardware.²⁹ To correlate the CL and STEM data, overview SEM images were recorded from the edge of the cleaved sample. Areas with features that could easily be identified in the FIB were chosen for hyperspectral CL imaging. Mapping was performed by integrating a range of wavelengths, and these are shown in the logaritmic scale, individually normalized. The CL was performed at room temperature and to avoid sample damage, the probe current was low (5–25 pA) with an acceleration voltage of 3 kV.

To protect the surface and to support the slices of the protruding platelets after plan-view cutting, a layer of amorphous carbon was deposited on the surface, in between and on top of the platelets. This was done by painting the edge of the sample using a permanent marker³⁰ (Centropen OHP permanent 2636 F, Blue color) shown in supplementary material Fig. S2(b). The pen was chosen since it provided good coverage of the structures up to the edge but still did not obscure the edge itself, which was needed to correlate exact locations in SEM-CL, FIB, and STEM [supplementary material Figs. S2(b) and S2(c) and S3]. The amorphous support is simpler to apply than conventional methods of covering protruding nanostructures for FIB plan-view lamella³¹ but with slightly less rigidity of the thinned region. After application, the ink from the pen was thoroughly dried by placing the sample on a hotplate to evaporate the alcohol solvents.

Subsequently, the sample was thinned in a FIB (Carl Zeiss crossbeam 1540 ESB). Since small amounts of material were to be removed, relatively gentle currents of Ga-ions were used: 50–200 pA (30 kV), both from top and bottom sides. The thinning was performed from one side until the wanted vertical position in the platelet was achieved. This was followed by thinning from the other side to a thickness less than 100 nm. In this case, the wanted position was close to the top, the location of the QW. Multiple windows were opened at the sites of interest, correlated by the previously identified locations of CL mapping (supplementary material Figs. S2 and S3). To retain the structural integrity of the amorphous lamella, the windows were kept narrow (∼2.5  $μ$m). The resulting openings contained electron transparent suspended slices of the platelets supported by the deposited amorphous carbon from the permanent marker. A reference conventional cross section liftout was also performed on the same system and sample.

The STEM analysis was performed using the Linköping double-corrected FEI Titan³ 60–300 microscope. Imaging was performed at 300 kV using collection angles ∼66–200 mrad and ∼21–200 mrad for high-angle annular dark-field (HAADF) and annular dark-field (ADF), respectively. Scans were acquired as either single scans or a series of scans reconstructed using the Smart Align³² plugin for Digital Micrograph (Gatan, Inc.), and geometric phase analysis (GPA) was performed using GPA for Digital Micrograph (HREM Research, Inc.).

The investigated nano-LEDs comprise an array of sub-micron sized platelets in the shape of truncated InGaN pyramids containing a single InGaN QW each. The platelet nano-LED, shown in the overview in Fig. 1(a), and supplementary material Fig. S1, as a schematic in Fig. 1(b), and in cross-sectional STEM in Fig. 1(c), contains: (1) a lower InGaN barrier; (2) a TRL with a higher In content; (3) a thin QW with an even higher In content; and (4) a top barrier with a similar In content as (1). Figure 1(c) reveals a dome-shaped interface inside the platelet, the shape after CMP, with subsequent layers of varying In forming of a flat top facet and a flat In-rich QW close to the top.

Multiple CL images (monochromatic maps) of pristine platelets in the region closest to the cleaved edge were acquired (see experimental details). For the platelets in Fig. 1(d), the CL reveals some local intensity variations in the extracted monochromatic images shown in Figs. 1(f) and 1(g), with wavelengths corresponding to the TRL and QW, respectively. The spectrum is shown in Fig. 1(e) and supplementary material Fig. S5, where the spectral windows for mapping are marked. Particularly, the QW emission reveals a pattern of dark lines, where the intensity is reduced to about 20%. Supplemental material Figures S2 and S3 and experimental details illustrate how specific platelets, chosen based on CL emission, were located and correlated, in SEM-CL, FIB, and STEM.

Figures 2(a) and 2(c) show low magnification STEM images of two platelets after thinning, while Figs. 2(b) and 2(d) show the corresponding emissive properties of the QW (∼590 nm). The platelet shown in the ADF image in Fig. 2(a) exhibits lines corresponding to structural defects, splitting the overall area of the platelet into three regions with a Y-shape. Its corresponding CL emission pattern of the QW-peak [Fig. 2(b)] shows the same Y-shape correlating the structural defects to non-radiative recombination. Similar observations can also be made for the platelet in Figs. 2(c) and 2(d). The images, hence, reveal a direct correlation between defects observed in the STEM images and the reduced emission in the CL images. In Fig. 2(e), showing a higher magnification of the correspondingly marked site in Fig. 2(c), a bright line is observed along the inclined facets using HAADF imaging. Since contrast in HAADF depends on the thickness and atomic number, this confirms that the In-rich QW is located within the thinned slice along with the observed defects.

Using high-resolution ADF-STEM, the two sides of the defective line are observed at the location marked in Fig. 2(c) and shown in Fig. 2(f). The lattices on the two sides are laterally shifted relative to each other, indicating that a partial BSF likely occurred below the QW during growth in either of the two domains, forming a SMB. Additionally, in order for the lattices to interact as observed at the SMB in Fig. 2(f), the two sides have to strain along the SMB as a BSF does not fully explain the exact difference between the two domains. As the SMB changes the direction, mainly between ${ 1 1 ¯ 00 }$ and ${ 11 2 ¯ 0 }$ planes, the regions around the direction change are additionally strained in order not to form dislocations (none observed in the slice where the SMBs change the direction). GPA of the region in Fig. 2(e) (shown as inset) reveals dilation of the lattice in the upper part of the figure, where the SMB changes the direction. Thereby, the observed defect boundaries are also associated with a locally strained lattice to allow for a dislocation-free lattice. Hence, the reduced QW emission in the form of lines can be identified as SMBs, as confirmed by HR-STEM.

However, as observed in Figs. 2(c) and 2(d) and in supplementary material Fig. S4, not all observed SMBs find correlation with dark lines in the CL map. In Fig. 2(c), multiple parallel lines are observed while only the centerline is observed to affect the intensity in the CL map. The reason for this discrepancy is proposed to be because not all defects intersect the optically active QW. The CL maps in Figs. 1(e) and 1(g) can be used to visualize the presence of defects by imaging the different layers through the different emission wavelengths. By analyzing a large number of platelets, it can be verified that most dark lines are present in both the QW and barrier emission. A few lines are only present in either layer [examples are marked in Figs. 1(f) and 1(g)], indicating that the SMBs can be generated as well as eliminated in any layer. A portion of the excitation takes place in the barriers and transfers into the QW. The presence of an SMB in the barrier near the QW may reduce the QW intensity. However, there are dark lines in the barriers that do not affect the QW emission. These may be located some distance from the QW, which in combination with a short diffusion length will not affect the transfer to the QW. Due to the low acceleration voltage, the excitation is centered at a depth around the QW. The much lower contrast in the barrier images indicates that the SMBs are only located in a part of the barrier and potentially only on one side.

In order to verify this by STEM imaging, the vertical position of the prepared window in relation to the specific platelet, hence, the location of the observed defects must be identified, similar to the QW observed in Fig. 2(e). In contrast to cross-sectional studies where the relative position of SMB and QW is easily determined, this is not straightforward for plan-view observations.³³ However, the inclined facets combined with the sample geometry enable the determination of the vertical position of layers within the platelet. In particular, the position of the QW can be determined as the higher In content reveals a brighter contrast at the edge of the platelet if present in the STEM window. This is the case observed in Fig. 2(e). Hence, SMBs reaching the edge of the platelet can be compared with the position of the QW observed in the same way. If the SMB passes through the QW, the intersection should cause the non-radiative recombination that is seen as a dark line in the CL images.

Figure 3 shows the structural properties of two locations at the edges of the same platelets as in Fig. 2. Both locations reveal defective boundaries; however, only the site shown in Figs. 3(a)–3(c) causes reduced QW emission. In Fig. 3(e), the QW can be observed as a bright line along the edge, shown at a higher magnification in Fig. 3(f). Figure 3(g) shows the region inward from where the QW intersects the inclined side facet and projects parts of the platelet both above and below the QW. Here, Fig. 3(g) shows a defect boundary, which is not present in Fig. 3(f), suggesting that the defect is located immediately above the QW. Since the defect is not present in the QW, the emission intensity from the QW at this location, Fig. 3(d), does not exhibit a dark line or a clear indent to the hexagonal shape, which is seen in Fig. 3(a). The QW is not observed in Fig. 3(b), which means that the platelet has been cut such that the QW ended up outside or at the surface of the STEM window. The window appears to have been made at an angle to the QW, and the location shown in Fig. 3(b) is presumably just below the QW based on the thickness of the window in relation to the thickness of the top barrier, as well as from following the QW on the inclined facet in the overview image shown in Fig. 2(c). Moreover, Fig. 3(c) reveals a defect boundary reaching all the way to the edge, showing its presence just below the QW. Following the reasoning of where the window is positioned, inferring to the propagation through and interaction with the QW, it explains the emission deficiency seen in Fig. 3(a).

To complement the images determining whether the SMBs interact with the QW or not, additional step-by-step CL mapping was performed on structures grown with the same growth recipe, shown in the supplementary material. This growth study compares separate samples where each consecutive sample has an additional layer of the full structure. Through CL mapping, it can be observed how defective lines are not seen on the bare polished templates. The conclusion is that the structure up to this point is mostly SMB-free. However, dark lines start to appear in the lower barrier. As the subsequent layers are added, most of the lines are found in all the layers; hence, most defects in the lower barrier propagate into the subsequent layers, while some are generated or eliminated. It appears that most defects are generated in the lower barrier layer grown on the CMP:ed surface, indicating that this surface may be the source of a majority of BSFs and subsequent SMBs, schematically illustrated in Fig. 4. Varying the In content, which is observed with the contrast in Fig. 1(c), strain or the general dome-shape with the six corners are likely reasons for the faults to occur.²⁶ Further details on this step-by-step CL analysis and stacking fault formation will be presented elsewhere.

In conclusion, plan-view observations by CL mapping and HR-STEM imaging allow for direct correlation between structural defects and emissive properties of nanoscale devices. Here, it was correlated that SMBs result in a local reduction in emission. It was further found that SMBs are associated with a strained lattice, avoiding lattice relaxation through dislocation formation, when the boundary changes the direction. Crucially, the growth of nano-LEDs similar to the ones presented here must avoid formation of defects such as stacking faults, which likely affect the emissive properties of QWs negatively. This is especially important for nano-LEDs due to their small active regions and the potentially multiple lines with relatively large area of emission deficiency.

See the supplementary material for additional data supporting the findings.

The authors thank the competence center for III-Nitride technology, C3NiT–Janzén, supported by the Swedish Governmental Agency for Innovation Systems (VINNOVA): Competence Center Program via Grant No. 2016-05190. The KAW Foundation is acknowledged for support of the Linköping Electron Microscopy Laboratory. The authors also thank the Swedish Government Strategic Research Area in Materials Science on Functional Materials at Linköping University, Faculty Grant SFO Mat LiU No. 2009-00971. The Swedish Research Council and the Swedish Foundation for Strategic Research are acknowledged for access to ARTEMI, the Swedish National Infrastructure in Advanced Electron Microscopy (Grant Nos. 2021-00171 and RIF21-0026). The microscopy, growth, and CL investigations were also supported by the Swedish Foundation for Strategic Research (Grant No. EM16-0024). Additional CL support came from the Crafoord Foundation, NanoLund, and the Swedish research council (Grant No. 2022-02832). Finally, NanoLund acknowledges support from the Swedish Research Council, and Reine Wallenberg, Mikael Björk, and Taiping Lu are acknowledged for valuable discussions.

The authors have no conflicts to disclose.

Axel R. Persson: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Visualization (equal); Writing – original draft (lead); Writing – review & editing (equal). Anders Gustafsson: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Resources (equal); Visualization (equal); Writing – original draft (supporting); Writing – review & editing (equal). Zhaoxia Bi: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (supporting); Writing – review & editing (equal). Lars Samuelson: Conceptualization (equal); Funding acquisition (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). Vanya Darakchieva: Funding acquisition (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). Per Persson: Conceptualization (equal); Funding acquisition (equal); Project administration (lead); Resources (equal); Supervision (equal); Writing – original draft (supporting); Writing – review & editing (equal).

The data that support the findings of this study are available within the article and its supplementary material.