Cathodoluminescence investigations of dark-line defects in platelet-based InGaN nano-LED structures

We have investigated the optical properties of heterostructured InGaN platelets aiming at red emission, intended for use as nano-scaled light-emitting diodes. The focus is on the presence of non-radiative emission in the form of dark line defects. We have performed the study using hyperspectral cathodoluminescence imaging. The platelets were grown on a template consisting of InGaN pyramids, flattened by chemical mechanical polishing. These templates are defect free, whereas the dark line defects are introduced in the lower barrier and tend to propagate through all the subsequent layers, as revealed by the imaging of different layers in the structure. We conclude that the dark line defects are caused by stacking mismatch boundaries introduced by multiple seeding and step bunching at the edges of the as-polished, dome shaped templates. To avoid these defects, we suggest that the starting material must be flat rather than dome shaped.


Introduction
There is currently significant interest in designing and fabricating micron-sized light-emitting diodes (μ-LEDs) for the entire visible spectral region [1][2][3][4][5][6][7][8][9].μ-LEDs typically have a width of below 100 μm, but can be as small as 1 μm [7,10,11].Below this size, they are referred to as nano-LEDs [7].One of the main applications is as individual pixels in mixed and augmented reality (MR/AR) displays, as well as high-resolution displays [11][12][13].The size of these pixels must be in the 1-10 μm range [3].The current technology is mainly based on a top-down approach, where the starting material is planar LED wafers where the individual μ-LEDs are etched out to the required size resulting in a significant loss of material.In larger-size LEDs, the blue and green emission is commonly based on GaN/InGaN quantum well (QW) or multi-QW (MQW) structures [14].For the red emission, the best LEDs are based on AlInGaP, with very high efficiency similar to that of the blue emission from the GaN/InGaN LEDs [12,14].However, the necessary reduction in size has a detrimental effect on the efficiency due to the etch damage to the side walls [2,8,15].GaN/InGaN structures do not seem to be affected as much, which is Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.usually attributed to a much shorter diffusion length of carriers in the nitrides, where a smaller region near the side walls is affected [16].Meanwhile, the light output power of red AlInGaP LEDs is more sensitive to environment temperature, and the efficiency decreases drastically at high current injections.Therefore, much interest has been focused on making red emitting LEDs based on GaN/InGaN.This is a technological challenge as the lattice mismatch between GaN and the InGaN with about 35%-40% indium needed for red emission is about 3.8%-4.3%[14].This mismatch increases the risk of introducing structural defects [17][18][19].As a comparison, the blue QW has a lattice mismatch of around 1.5%.As highly efficient blue LEDs can be made and have been widely used for solid state lighting, this is a manageable mismatch.Due to the lack of suitable substrates, it is difficult to reduce the mismatch between the red InGaN QW and the substrate.For growth on GaN substrates or GaN buffer layers, an intricate design of strain-compensating and defect reducing layers is needed [14].
A different approach is to grow the structures bottom up [13].This means that no material is wasted by etching away the layers in between the μ-LEDs.One such approach is to grow nanowires (NWs) using radial QWs as the emitting layer [20,21].Despite their small size, the NWs can be used when larger pixels are needed as the size of a pixel can be scaled from a single NW to an array of NWs.The NW approach has two major advantages over planar growth: (1) the small footprint on the substrate means that the NW can accommodate significant strain without introducing any structural defects at the interface.In addition, if the NWs are grown by selective area growth (SAG) on a masked substrate in a regular array, the mask will filter out most of the threading dislocations from the substrate [22].(2) The NW can expand/contract in the axial direction to accommodate the strain between NW and QW.In principle, the sidewall area could exceed that of the substrate area, leading to a larger QW area than the substrate area it is grown on.However, the a challenge is that radial QWs tend to vary in thickness and composition along the NW length [13].It is also difficult to grow the NW core with a constant ternary (InGaN) composition and it may not even be possible to grow NWs [21].

Fabrication
We have devised a scheme where we initiate the growth with an array of short SAG GaN NWs in a regular hexagonal array.On these NWs we grow InGaN, which forms pyramids with a hexagonal base [23].Just like the NWs described above, the pyramids are virtually free from structural defects propagating from the substrate as the mask blocks most of the threading dislocations [24].These pyramids could in principle be used for QW growth on the side facets.Unfortunately, this suffers from the same issues as the QW on the side walls of the NWs, namely a gradient in the composition and thickness.Instead of growing the QWs directly on the side facets, we have introduced a second step, where we truncate the pyramids down to thin platelets.This is done by chemical mechanical polishing (CMP) [1].After this procedure, the resulting platelets have a top c-plane area of 0.25-0.5 μm 2 and a thickness of 50-100 nm.The small thickness maximises the area of the top c-plane and the material that has grown above the mask can accommodate strain from the subsequent layers.By choosing a suitable InGaN composition in the pyramid for the red QW, we can keep the strain at a manageable level, similar to the strain in the planar blue-emitting LEDs grown on GaN substrates.For a blue-emitting LED, a QW with 15% indium is typically sandwiched between GaN barriers with a manageable level of strain.A similar level of strain can be achieved with a combination of InGaN barrier and InGaN QW of 10% and 25% indium for a green-emitting and 17% and 35% indium for a red-emitting QW.Using a relaxed platelet with InGaN barriers can result in virtually defect free planar QWs on the flat top-plane of the platelets.A similar approach has been presented in fabricating hexagonal platelets by SAG, but rather than using NWs to seed the platelets, the entire area opened in the mask defines size of the platelets [25,26].The advantage is that the CMP stage is not needed as the platelets form directly.The drawback is that the entire platelet is connected to the substrate, without the opportunity for strain relaxation using our approach.Furthermore, this approach does not have the advantage of filtering out the threading dislocations from the substrates.
Figure 1 illustrates the initial (a) and final (b) steps in the fabrication.A scanning transmission electron microscopy (STEM) cross-sectional image of the full structure is shown in figure 1(c) (sample type F).Due to the sub-micron size these structures can be referred to as nano-LED.The structures in this report consist of a regular array of platelets seeded by GaN NWs grown by metal-organic chemical vapour deposition from holes (diameter of 200 nm) in a SiN X mask using SAG [23], with holes arranged in a hexagonal pattern with a 1 μm spacing.The hetero-structured platelet comprises an InGaN template with a nearly flat top c-facet resulting from the CMP of the InGaN pyramid.On top of this, a lower barrier is grown, with nominally the same indium content as the template.Next, there is a transition layer, with a slightly higher indium content, incorporated in order to enhance the indium incorporation into the QW and to promote electron injection from the lower barrier.This is followed by a single QW (about 4 nm thick), and a top barrier which has the same composition as the lower barrier.As the growth progresses, the top facet of the platelets shrinks and the pyramidal side facets are re-established.This leads to a limited growth window before the pyramid shape is recovered.The indium content of the barrier is chosen so that the lattice mismatch between the QW and the barrier mimics that of blue QWs grown on GaN, which is manageable with high efficiencies in planar structures, as introduced above.The longer the intended wavelength beyond blue, the more indium is needed in the QW and barrier.The growth is described in detail in [1,27].

Hyperspectral cathodoluminescence characterisation
We performed optical characterisation using hyperspectral cathodoluminescence (CL) investigations in top view.Hyperspectral imaging records a full (or partial) spectrum in each pixel in the CL image [28,29], while simultaneous recording an SEM image.We used a scanning electron microscope (SEM) with an acceleration voltage of 3 kV and a probe current of 10-50 pA, at room temperature.The chosen acceleration voltage gives a penetration depth of about 50 nm, with a similar lateral spread.For the full structure, the top barrier, the QW and the TRL are excited directly.A higher acceleration voltage excites further layers, but with the penalty of a reduced spatial resolution.This is illustrated in the supplementary material.The CL set-up is described in detail in [29].It should be emphasises that the monochromatic images presented here are in linear scale where white is the highest intensity in the individual images and black means no intensity, and the images therefore represent the total intensity rather than the intensity variations.The structures investigated here are similar to a full p-i-n LED structure with two notable differences: There is no pn-junction as the barriers are undoped, and the top barrier is thinner than the corresponding barrier in the LED.The thinner top barrier enables us to image the QW emission at low acceleration voltages, as opposed to the reduced spatial resolution if a higher voltage is needed to excite the QW through a thicker top barrier, see supplementary material.
When we perform CL investigations on the platelets containing a full stack of layers, including a QW, we observe dark line defects in the QW emission, especially from the redemitting QWs.In a previous study, where we combined hyperspectral CL imaging with top-view imaging in highresolution STEM on the same platelets, we identified these defects as stacking mismatch boundaries (SMBs) [30,31].An SMB forms when two different stacking sequences, such as when KABABK and KACACK in wurtzite meet in plane where every second layer (B/C) is out of phase, and a mismatched boundary is created.As the growth progresses this boundary will continue in the growth direction.In planar cases, these boundaries can both be formed and terminated during growth [32].The significance of these SMBs is that they act as non-radiative defects [31,[33][34][35].
In the present study we have used a large number of samples of different types to elucidate where the SMBs are generated.The full structures have a single QW emitting either green (530 nm) or red (630 nm) light.We have also studied the optical properties of each layer, by ending the growth after each subsequent layer.All structures, except for the green-emitting structure have the same layer sequence with the same nominal composition and thickness.Table 1 lists the different types of samples used in the study.The different stages of growth are illustrated in figure S1 in the supplementary material.

Results and discussion
To complete the investigations into the dark line defects we reported in [30], we have studied other potential sources for them.Figure 2 consist of top view images (SEM and CL) of a randomly selected area with red-emitting platelets of a sample type F. It shows that many platelets exhibit dark line defects when imaging the QW emission.In the red type F sample, typically 50%-80% of the QWs exhibit one or more dark lines.It is worth pointing out that the pattern does not change with detection wavelength over the entire QW peak, eliminating a local variation in the strain or composition as origin of the dark lines.To optimize the efficiency of the red-emitting nano-LEDs, it is essential to understand the formation of these dark line defects and to eliminate them.In this study, we have investigated in which layer the dark lines originate.To do this we have studied the behaviour of the emission pattens in a series of samples with varied QW thickness to study the influence of lattice mismatch.In a second series, we have added layer by layer to observe in which layer the dark lines start to appear and we also follow the dark lines through the different layers.This is possible as different layers emit at different wavelengths due to the different compositions.As the emission is collected using hyperspectral imaging, there is no drift between the images extracted from different wavelengths.Different emission wavelengths were achieved by changing the indium content in the barriers and QW, as listed in table 1 (sample types F and G).The indium content was controlled by the growth temperature.Figure 3 shows a comparison between the red-and green-emitting platelets (types F and G).There is a significantly lower number of platelets with dark lines in the emission pattern of the greenemitting platelets of figure 3(b), typically 20%-30%.This seems to suggest that the dark lines may be introduced by the strain due to the difference in composition between the QW and barrier.This is however unlikely, as the strain between barrier and QW is similar in both cases.Another important observation is that the CL image of the green emission in (b) reflects the hexagonal shape of the c-plane where the QW is located.There are also hexagonal patterns in the red emitting platelets in figures 2 and 3.The hexagonal shape is a sign that the side facets are of high quality and do not affect the emission near the edges, as is the case for the etched μ-LED structures.There are variations in the density of dark lines in both cases, some areas of the sample have fewer and some have more, but the images presented in figure 3 are representative of the two samples.Similar observations can be made from different growth runs with similar indium compositions.We also observe dark lines where the blue QWs are grown on a layer of InGaN/GaN superlattices or a 30-40 nm thick InGaN layer [36].
A series of samples was grown with varying QW thickness, with the same nominal indium content in both QW and barriers, sample types F, H, I and J.This series was used to investigate if the mismatch between substrate/barrier/QW has introduced misfit dislocations due to a QW thickness above the critical thickness for dislocation generation [19].In a planar structure with the same barrier/QW combination, the critical thickness is expected to be >5 nm for the QW [37].The increase in intensity with reduced QW thickness in figure 4(a) is what could be expected from a reduction in dark line defect density.The emission is also blue-shifted due to increased quantization in the QW.Despite the increased intensity, the emission patterns from the QW still exhibit dark lines (figures 4(b) and (c)), with similar numbers of platelets affected.The increased intensity is most likely related to carrier separation in the QW due to a strain-induced electric field.The separation is reduced with decreasing QW thickness, resulting in a higher intensity for the thinner QW.The presence of dark lines in the thinnest QW rules out misfit dislocations caused by the mismatch between the QW and barrier as the origin of the dark lines.The spectra in figure 4(a) also reveal that the barriers are identical for all samples, as the peaks from the barrier layers at 440 and 510 nm are present at the same spectral position in all four samples.Since the dark lines do not appear to be caused simply by misfit dislocations at the QW interfaces, the cause must be found elsewhere.One hint comes when comparing the QW and barrier emission in figure 5 (sample type F).Most of the lines in the QW emission can also be found in the barrier emission.Though misfit dislocations at the interface between the barrier and QW can affect the barrier emission, the dark lines in the barrier indicates that the defects may have a different origin rather than in the QW growth.Under normal conditions, only a very thin layer of the barrier near the interface is affected.The main volume of the barrier is unaffected and therefore the contrast is usually quite low.The contrast in the images of the barrier emission is lower than in the QW.This indicates that the defects causing the reduced emission may only be partially present in the barrier.At the same time, the contrast is higher than expected from dislocations at the interface [38].It could also be that they are only present either above or below the QW.To explore the origin of the defects causing the reduced emission further, we have studied the emission as a function of adding each layer.It is worth pointing out that the dark centre of the platelets in figure 5(c) originates in the GaN seed.It either alters the strain or the indium incorporation in the lower barrier.When changing the detection window, the contrast is reversed, with a bright centre.
One potential source of the dark lines could be introduced during the CMP of the InGaN pyramids.To investigate the effect of the polishing, we performed CL studies on the templates after CMP (sample type A), as presented in figure 6.These platelets are somewhere between 50 and 100 nm thick.The individual templates are quite similar in emission patterns, where the InGaN shows a hexagonal pattern with slightly brighter corners and a dark centre.The variation in the intensity is most likely related to facet driven variations in the indium content during the pyramid growth.This was presented in more detail in [29].The dark centre is where the GaN seed NW penetrates the platelet surface, as observed in the high resolution SEM image in the inset in figure 6(a), and in the cross-sectional TEM image in figure 1(c).When recording images using either the GaN bandgap emission or yellow luminescence (YL), the centre is bright.The investigation does not reveal any dark lines in the template emission, leading us to conclude that the CMP does not introduce any structural defects and that the templates are free from defects causing dark lines.
We made another series of samples in order to determine in which layer the defects that cause the dark lines are created.This time each layer in the sequence was added.The series have samples with: (type A) as-polished template only, (type C) template plus lower barrier, (type D) template plus lower barrier plus transition layer, and (type E) template plus lower barrier plus transition layer plus a QW followed by a very thin top barrier.Images from this series is presented in figure 7. From this series, it is clear that already the lower barrier exhibits some the dark lines, sample (C).The contrast is weaker than in the QW, as they are not affecting the whole barrier.There is an overlap in the emission from the template and the lower barrier.Another observation is that the dark lines in the QW already exists in the transition layer in sample type (C).This indicates that the defects are extended along the growth direction.This supports the findings in [30], where the defect responsible for the dark lines in the QW emission was identified as SMBs.
The acceleration voltage used in figure 7 was too low to penetrate all the layers.3 keV excites about 50 nm into the sample.This is why the template and lower barrier appear as rings in some images.By increasing the acceleration voltage (see supplementary material), we have been able to follow the dark lines from the QW in sample (E) to the barrier and from the transition layer in sample (D) to the lower barrier.This confirms that many of the dark lines propagate through the layers.In most cases, the dark lines are in the same positions, but there are also examples of slight shifts between the layers.This was also observed in [30], where some of the SMBs in the STEM images did not match the dark lines in the QW emission perfectly with some minor displacements observed.At acceleration voltages above 6 keV, the spread of the electron beam inside the sample leads to a loss of resolution of the dark lines.This is shown in the supplementary material.
When imaged by high-resolution SEM, some surface features can be observed, including nano-trenches seen in figure 8.Many of the dark lines in the QW emission can be correlated with these micro-trenches in the surface of the platelets.This is consistent with published observation of SMBs in planar structures, where they are correlated with trenches in the surface [31][32][33][34][35].This is also fits with the observation in our platelets where the dark lines are present in several layers.We are therefore fairly confident that most of the dark lines we observe in the CL images are related to SMBs.It is worth pointing out that the dark lines are observed even at 10 K, though we have not made any systematic study at this temperature.
The origin of the stacking fault responsible for the SMBs can potentially be found in the shape of the template after the    CMP [27].Rather than a perfectly flat c-facet, it has a dome shape, as seen in figure 9(a).The AFM image of figure 9(b) shows steps on the surface in the middle of the lower barrier growth.A high density of steps can lead to a higher local growth rate, and the platelets will be flattened by growth from the edges.This can be seen in figure 1(c), where the original dome shape is visible as a variation in the contrast.The interface between the lower barrier and the transition layer is also visible in this STEM image.It is clear that the growth rate is significantly higher at the platelet edges as the structure is rapidly planarized.The concept of the growth was introduced in [30].
As the growth most likely is seeded at several locations (edges and/or corners of the platelet) simultaneously, the respective growth fronts will eventually meet in the middle of the platelet.As long as they have an identical stacking sequence, this poses no problem.If the stacking sequence at one location is out of phase, there will be a stacking mismatch when the growth fronts meet.As long as the growth continues to take place from the edges or corners, the boundary will propagate along the growth direction and continue until all the growth fronts are aligned to the same stacking sequence.We observe notably fewer platelets with dark line defects in the green-emitting QWs, as shown in figure 3.This could be due to the higher growth temperatures needed and lower indium contents for the green-emitting platelets, compared with the red ones [39,40].The higher growth temperature generally promotes a higher crystal quality.
We have previously observed that the emission intensity and lifetime increase with thickness of the lower barrier in samples with only this layer [27].The increase is significantly larger than the increase in volume, ruling out the increase in volume as the origin of the intensity increase.In full platelets with a red-emitting QW grown on a thicker lower barrier, there are fewer dark line defects in the QW emission.Both observations point to that the SMBs can be eliminated by a thicker lower barrier.Unfortunately, a thicker lower barrier will reduce the QW area as the growth tends to reshape the platelet into a pyramid.There may be an energy associated with the SMB, and it is therefore energetically favourable to eliminate the SMB as the growth proceeds.

Outlook and conclusions
The key to the success of our nano-LEDs is to significantly reduce the number of, or completely remove the SMBs with the optimization of the CMP step.A significant reduction of the height of the dome or even obtaining a totally flat surface to start with would most likely prevent the SMBs from forming.
Before the CMP step, the surface containing the pyramids is coated with SiO X , to protect the pyramids from being scraped away.As the SiO X is softer than the InGaN, it will be removed faster, resulting in a thinner layer between the pyramids.This is the most likely cause for the dome shape of the templates.The obvious way forward is to replace the SiO X with a coating material that has similar mechanical properties as InGaN.This should lead to a similar removal rate as the pyramids and flatter template surfaces.At the same time the coating needs to have other properties, like not introducing any defects and must be easily removed like SiO X .We are currently investigating potential candidates, and this will be the focus of a future report.
In conclusion, we have studied sub-micron-sized InGaN platelets with respect to dark line defects in the emission  pattern from individual platelets using hyperspectral CL imaging.We corroborate our previous findings that the dark lines are caused by SMBs in the platelets.By studying the heterostructures with each layer added separately, we observe that the original template with a nearly flat surface does not show any dark lines, only facet driven compositional variations.We conclude that some of the SMBs are introduced during the growth of the lower barrier.These propagate through the entire structure, in most cases all the way into the upper barrier.Some are generated higher up in the structure and some are terminated in subsequent layers.The most likely origin of the irregular stacking is the dome shape of the template, where the growth is seeded from the edges, containing many surface steps.It is likely that different seed areas have different stacking sequences, and when the growth fronts meet, the SMB is created.The remedy is to make sure that the original template is flat.This can be done by using a different protective/ coating layer in between the original pyramids for CMP.This layer must have a similar removal rate as the InGaN pyramids, rather than faster than the SiO X used in this study.

Figure 1 .
Figure 1.Illustration of the structures in this study.(a) is the asgrown pyramid and (b) is the final structure after polish and regrowth.The dashed line shows the interface between the polished platelet and the lower barrier, and the red line indicates the QW.(c) is a cross-sectional STEM image of a final structure.The template and lower barrier have the same nominal indium content, the TRL and the upper barrier have the same composition, with more indium and finally, the QW has the highest indium content.The STEM image is of a sample type F.

Figure 2 .
Figure 2. Overview of an area with red-emitting single QW platelets of type F. (a) SEM image (green) and (b) the corresponding CL emission (grey scale) from the QW, as indicated by the red detection window in (c).(c) An average spectrum from the entire area in (a).The emission at 360 nm corresponds to the GaN substrate, at 440 and 510 nm to the template and barriers and at 635 nm to the QW.The yellow circle in (b) indicates a platelet with perfect QW emission, whereas the green circle indicates a platelet with a dark line defect.

Figure 3 .
Figure 3.Comparison of the emission patterns from red and green emitting QWs.The CL pattern is overlaid on the corresponding SEM image.(a) is the red-and (b) is the green-emitting QWs.(sample types F and G).

Figure 4 .
Figure 4.A study of different QW thicknesses.(a) Spectra from a series of QWs with different thicknesses, where 4.0 nm corresponds to the thickness of the QWs in figures 2 and 3(a).CL images of (b) a slightly thicker (5.2 nm) and (c) a much thinner (1.6 nm) QW (sample types F, H, I and J).

Figure 5 .
Figure 5. Emission patterns from the QW and the barrier (Sample type VI).(a) SEM image.(b) is the QW and (c) is the barrier emission.The red ring points to a dark feature that is shared between the QW and barrier emission.

Figure 6 .
Figure 6.Emission pattern from the template after CMP.(a) SEM image and the inset is a high-resolution SEM image.(b)-(d) are CL images using the emission from the GaN (b), InGaN (c) and YL of GaN.Sample type A.

Figure 7 .
Figure 7. CL investigations of a series of samples where each layer has been added, as indicated in the rows.(Type A) starting from the template.The following samples have additional layers: (type C) lower barrier, (type D) lower barrier and transition layer and (type E) lower barrier, transition layer, QW and a very thin top barrier.For the CL images the emission corresponds to the template (purple), template/ barrier (blue), transition layer (green) and QW (red).The inset is the normalized spectra from the four different samples.The coloured rectangles in the spectra indicate the detection windows for the CL images in the different columns.The vertical scale bars are 1 μm and the 'X2' etc corresponds to how much the grey scale is enhanced.

Figure 8 .
Figure 8. High resolution SEM images (a and c) correlated with CL images (b and d) of a sample type F. The SEM image of (a) show some surface features, nano-cracks or nano-trenches, highlighted in (c) by red lines.These find a near perfect correlation with the dark lines in the CL images.

Figure 9 .
Figure 9. (a) SEM image of the template, sample type A. (b) Topview AFM image of a structure after growth of thin lower barrier, sample type B. The apparent step height is about 5 nm.Both images show that the surface is dome shaped, even after growth of a thin lower barrier.Reproduced from [27].CC BY 4.0.

Table 1 .
List of the types of samples used in this study.