Cathodoluminescence in a Scanning Electron Microscope Operated in Transmission Mode

Cathodoluminescence (CL) provides insights into the optical properties of matter beyond the diffraction limit, offering analyses similar to photoluminescence experiments, but with significantly improved spatial resolution. The spatial resolution in CL is primarily determined by the electron–hole generation volume, created by the inelastically scattered electrons of the primary beam. As for defect imaging, the contrast is predominantly given by the locally enhanced non-radiative recombination rate of the generated carriers at the defect site, enabling the resolution of structures with dimensions far below the charge carrier diffusion length.¹ Cathodoluminescence experiments can be carried out in scanning electron microscopes (SEMs) or in scanning transmission electron microscopes (STEMs). STEMs have the potential for achieving very small generation volumes and hence the highest spatial resolution in CL.^2,3 One criterion here is the availability of extremely small spot sizes of the focused beam, which can reach values of < 100 pm in probe-corrected systems at 300 keV. In combination with an electron-transparent sample geometry, this results in a reduction of the interaction between the primary beam and the sample. This limits the lateral expansion of the generation volume, which is increasingly determined by the impinging primary beam cone.

However, the actual spatial resolution in CL is typically impaired, as such small spot sizes can barely be achieved. This is because high accelerating voltages often exceed the destruction threshold of the material under investigation. Even for materials considered radiation-hard, such as GaN, the optical degradation already occurs for beam energies above 70 keV.⁴ Furthermore, it should be noted that the spot size is merely one of numerous parameters determining the spatial resolution of CL. The achievable spatial resolution may be contingent on a number of additional parameters, such as the sample geometry, charge carrier diffusion, and secondary photon emission. The last of these refers to the radiative recombination of photons at a specific location as a result of CL reabsorption. Another important consideration is the limitation on reducing sample thickness, as increased surface recombination can lead to a decrease in CL yield.

Modern scanning electron microscopes are capable of providing reasonably small spot sizes of approximately 1 nm, which is still below the typically achieved spatial resolution of CL.⁵ Despite that, only a few CL studies have been performed in transmission mode combined with electron transparent samples. ^6,7 The majority of CL investigations in SEMs were conducted on thick samples in planar geometry and low accelerating voltages in order to attain CL with high spatial resolution.⁸ In this approach, the rapid energy dissipation of the primary electrons results in small generation volumes, which allows us to visualize point defects in quantum wells.⁹ Disadvantages may arise from the fact that lower accelerating voltages increase the spot size, as a consequence of diffraction effects associated with the electron optics. Furthermore, this configuration enables limited chemical sensitivity through the utilization of the backscattered electron signal, given that these signals are partially obscured by the CL mirror.

In light of the fact that state-of-the art SEMs are equipped with segmented STEM detectors, this enables SEM-CL investigations in transmission mode in close analogy to a STEM. While the accelerating voltages are limited to 30 keV, SEM typically provides a higher degree of flexibility in adjusting beam currents.

Additionally, owing to the larger size of the parabolic mirror, CL experiments in a scanning electron microscope benefit from a larger homogeneous light extraction area, enabling quantitative comparisons of CL yield across larger fields of view.

However, while this approach is straightforward, it has rarely been applied. This article demonstrates the capabilities and particularities of CL performed in a SEM when operated in transmission mode in conjunction with a segmented STEM detector. This approach allows for the correlative recording of the STEM bright field (BF), dark field (DF), and high-angle annular dark field (HAADF) alongside the CL signal. The HAADF signal provides distinct advantages in terms of signal-to-noise ratio, as well as material contrast in comparison to in-lens or chamber detectors.

To illustrate the approach, we present a structural analysis of an AlN/(Al, Ga)N layer that has been grown by molecular organic vapor-phase epitaxy (MOVPE) on a high-temperature-annealed (HTA) AlN template structure on sapphire. It is demonstrated that spatial CL resolutions of < 100 nm can be readily achieved at room temperature. However, the limitation of accelerating voltages to 30 kV gives rise to particularities arising from the increased dynamic interaction between the electron beam and the sample, which must be taken into consideration. Firstly, the STEM-BF signal can be utilized to estimate sample thickness, as well as to analyze structural defects, such as dislocations. Correlative analysis of CL and the BF signal demonstrates that for sample thicknesses < 20 nm, the CL yield becomes negligible. The HAADF signal can be employed for analyzing chemical composition, which is limited due to dynamic electron scattering. For this reason, z-contrast-like contrast can only be observed for very low sample thicknesses.

Experiments were conducted in an Apreo-S electron microscope (Thermo Fisher) using 30 keV primary electrons. The system is equipped with a STEM detector consisting of a multi-annular photodiode with six concentric annuli providing a bright-field (BF) mode, four dark-field (DF1-4) modes, and a high-angle annular dark-field (HAADF) mode. Cathodoluminescence was collected using a Gatan Monarc system, equipped with a parabolic mirror for light collection and a photomultiplier (Hamamatsu R928) operated in photon counting mode. Monochromatic CL maps were recorded after light dispersion using a 1200 l/mm grating with a blaze angle of 300 nm. Figure 1a shows an image of the configuration of the components in the SEM chamber, while Fig. 1b displays a sketch of the setup. The pole piece is equipped with three in-lens detectors with the designations T1–T3, denoted as a trinity detection system.¹⁰ The parabolic mirror for collecting the CL signal is positioned just below the pole piece. The T1 detector is located at the lower end of the pole piece and is mainly designed for collecting backscattered electrons (BSE). T2 and T3 are located further above and are mostly designed to collect secondary electrons (SE) for morphological contrast. Beneath the mirror is the sample holder, and approximately 20 mm below that lies the segmented STEM detector. Additional low-temperature CL experiments were conducted using a SEM stage allowing for He cooling. This configuration precluded the detection of transmitted electrons, as the insertion of the STEM detector was impeded.

Bild vergrößern

For analysis, we used a sample consisting of a 2.2-µm-thick AlN, followed by a 1.8-µm-thick Al_xGa_1−xN layer with x = 0.65 grown by metal–organic vapor-phase epitaxy (MOVPE). Growth was conducted on a 200-nm-thick HTA AlN buffer on a sapphire substrate. The 200 nm AlN layer for HTA was deposited by sputtering and annealed at around 1700 °C for 2 h. Samples were prepared by mechanical polishing and Ar⁺ ion milling. The accelerating voltage of the Ar⁺ ions was gradually reduced from 4 kV to 0.2 kV until electron transparency was achieved.

With the CL mirror inserted, the in-lens detectors suffer significant signal degradation, which is typical for CL setups of such design. This is exacerbated by using 30 keV primary electrons in combination with an electron-transparent sample, since a substantial number of primary electrons are transmitted through the sample. This effect is demonstrated in Fig. 2a and b, showing an image of the (Al,Ga)N sample using the T1 and the HAADF detector, respectively. This cross-sectional specimen consists of two pieces glued together at their respective epitaxial surfaces, which makes them separated by a thin glue line (appearing dark). Although the T1 has the highest signal intensity among all available in-lens and chamber detectors, details of the sample structure are barely recognizable. The utilization of backscattered electrons in combination with a thick—i.e., electron-opaque—sample yields reduced chemical sensitivity in comparison to the transmitted HAADF signal, as shown in the supplementary Figure S1. This phenomenon is further exacerbated by the CL mirror, which partially obstructs the BSE signal. This makes it difficult to separate the (Al,Ga)N layer from the underlying AlN, and to correlate the image with the CL signal. Obtaining higher image quality with the in-lens detector would necessitate excessively long dwell times that may cause issues related to the sample drift. On the contrary, the HAADF signal recorded with the STEM detector using identical beam settings reveals a detailed picture of the sample structure. Compositional variations in the (Al,Ga)N along the growth direction are immediately evident, as well as presumably strain-related contrast due to threading dislocations.

Bild vergrößern

Interpreting such HAADF images in terms of “z-contrast” is however not straightforward and under these conditions only works for very thin sample thicknesses. Figure 3 illustrates this problem in a large-field-of-view HAADF image. As is commonly observed in cross-sectional samples thinned by double-sector ion milling, a wedge-shaped sample geometry with decreasing thickness towards the glue line near the focal point of the ion guns is created. Two intensity profiles (1, 2) were recorded across the AlN/(Al,Ga)N interface in regions with increasing sample thickness, shown as insets in Fig. 3. Only the thinnest area (1) exhibits the expected “z-contrast” behavior, with the HAADF signal in the (Al,Ga)N region appearing brighter than the AlN. In the thicker region (2), the contrast becomes inverted. This demonstrates that the HAADF contrast is highly dependent on the sample thickness. The reason for this is that for 30 keV electrons, kinematic imaging conditions only apply in very thin sample regions and quickly become dynamic as the sample thickness increases. This results in a higher electron absorption in the denser (Al,Ga)N and thus a lower HAADF intensity causing a contrast inversion with the AlN.

Bild vergrößern

The strong dynamic interactions between the sample and the primary electron beam can be used to estimate the sample thickness. This can be achieved by using the quantitative signal from the STEM detector in combination with Monte Carlo simulations as discussed, for example, in Ref. 11. For this approach, we restrict the detection to the BF segment of the STEM detector. This is necessary because each segment of the STEM detector has different gain settings that cannot be read out, making it difficult to achieve uniform sensitivity across the detector. The quantitative BF signal is obtained by subtracting the detector dark current and normalizing it to the full intensity of the incident primary beam. Figure 4a displays the corresponding BF STEM image of the sample in a region where the entire structure is preserved. To visualize more details, the image is displayed on a logarithmic scale. The vacuum region appears bright, as the primary beam hits the BF segment directly. Monte Carlo simulations were conducted using the CASINO code,^12,13 which handles inelastic scattering by means of the continuous slowing down approximation. For the sake of simplicity, we have neglected secondary electron generation, as they do not affect the number of transmitted electrons. Detailed simulation parameters can be found in the supplementary Table (S1). We have calculated the number of electrons transmitted through an Al_xGa_1−xN film with x = 0.5 of different thicknesses reaching a bright-field detector with acceptance angles as in the SEM. Based on the simulation data, an analytical fit in the form of a power function was derived, making it possible to convert the absolute BF signal into a sample thickness, as shown in the supplementary Figure S2. The resulting thickness map of the sample is shown in Fig. 4b. In order to minimize the influence of compositional variations or dislocations, which cause contrast in the BF image, the thickness map was Gauss-filtered. As commonly observed for specimens prepared in this fashion, the thinnest parts are located near the glue line. In this region, the thickness is about 100 nm and below. Moving away from the glue line towards the AlN interface, the sample thickness rapidly increases and reaches 500 nm near the interface. The thickness gradient is higher for the upper sample piece than the lower one, with wedge angles of approximately 17° and 6°, respectively. This may be related to the different orientations of the sample causing different ion etching sensitivities.

Bild vergrößern

Along with the HAADF and BF signal displayed in Figs. 2b and 4a, respectively, cathodoluminescence was acquired in the same region. Figure 5a shows the corresponding panchromatic CL image with the surfaces, identified from the BF image, indicated by a thin red line. Comparing the sample thickness against the total CL yield reveals practically no emission for sample thicknesses below 20 nm. This is related to the decreasing energy dissipation within the layer, as well as due to increased non-radiative surface recombination. The roughly vertical dark lines in the figure indicate the presence of threading dislocations (TD), which also cause non-radiative recombination. They are also visible in the BF image and allow for comparison with the CL image.

Bild vergrößern

The panchromatic CL image in Fig. 5a exhibits periodic intensity variations in the (Al,Ga)N layer along the growth direction which are also visible in the HAADF image in Fig. 2b. The latter suggests chemical composition fluctuations as the origin. For the upper sample piece, where the <11–20> direction is parallel to the beam, the resulting lines are approximately parallel to the AlN/(Al,Ga)N interface. For the lower sample piece, where the beam is parallel to the <1–100> direction, they are inclined at an angle of about 4°. To quantify the composition variations, a series of 2 nm bandpass images were recorded in the spectral range between 230 nm and 270 nm. To compensate for sample drift, the CL images were registered using the simultaneously acquired HAADF image series using the FIJI code.¹⁴ The image series was converted into a false-color RGB image for presentation purposes, with blue (red) components representing shorter (longer) wavelengths, as displayed in Fig. 5b. The emission peak is centered around 253 nm, which corresponds to a mean Al content of x = 0.65, using Vegard’s law and the parameters given in Ref. 15. The analysis of the shift of the center of mass of the emission along the growth direction, reconstructed from the image series, yields a variation of the emission peak of approximately ± 1 nm and a period length of around 100 nm. This corresponds to a composition variation in the range of about ±1%, assuming that the spectra are excited from the individual layers. Details about the compositional fluctuations are shown in the supplementary Figure S3. Apart from the periodic composition modulation, both sample pieces reveal a blueshifted emission directly at the (Al,Ga)N/AlN interface, indicating decreasing Ga content in the early stages of the growth.

The visibility of the composition fluctuations is more pronounced when CL is performed at low temperatures, as can be seen in Fig. 5c, which was carried out at 10 K using the liquid He cooling stage. Under these conditions, the STEM detector cannot be inserted, and thus image registration has to be performed using the in-lens detector signal. The much lower (B) SE yield makes registering the image more difficult, but still possible. As for the RT CL, the data were converted into a false-color RGB image as shown in Fig. 5c. The composition variations along the growth direction, as well as the Ga depletion at the early growth stage, become much more clearly visible. The prior effect is related to step-bunching during epitaxy, which may result from a too-high miscut angle,¹⁶ while the latter indicates an initial Ga deficiency in the gas phase of the MOVPE reactor. A quantitative comparison between the RT and lHe CL data is not possible, given that they were acquired in different regions of the sample and CL yield being thickness-dependent.

Finally, the interface region near the sapphire substrate and the HTA layer was investigated using a correlative BF and CL analysis. Given the absence of chemical contrast in this area, the focus is directed to extended defects, where the BF signal is particularly sensitive. Figure 6a shows the corresponding BF STEM image of the corresponding area, which was fast Fourier transform (FFT) bandpass-filtered to enhance the visibility of small features. The image clearly reveals structural defects at the interfaces between the sapphire/HTA AlN, as well as at the HTA AlN/epitaxial AlN interface, as indicated in the figure, which can be identified by means of dark horizontal lines. At the sapphire interface, these are related to misfit dislocations, while near the epitaxial AlN interface they stem from voids and stacking faults.¹⁷ Additionally, the BF image reveals the presence of threading dislocation emanating at the sapphire HTA AlN interface. These dislocations exhibit an inclination within the epitaxial AlN layer, which is due to the fact that it is compressively strained, resulting from the disparate thermal expansion coefficients of AlN and sapphire upon cooldown.

Bild vergrößern

In addition to the defects that can be attributed with a high degree of certainty, the BF image reveals two further features: Firstly, the epitaxial AlN displays short horizontal lines, which appear to be associated with dislocations. Secondly, a faint horizontal dark line is visible in the sapphire, approximately 300 nm below the AlN buffer interface. Structural defects, indicated with arrows, appear to emerge from this line and extend towards the AlN interface. The precise origin of these defects remains unclear and necessitate a thorough TEM analysis to provide a more detailed understanding, which is beyond the scope of this paper.

The corresponding panchromatic CL image is displayed in Fig. 6b as a color-coded image, which was recorded simultaneously. Some features identified in the BF image are also reflected in the CL image. Starting from the sapphire substrate, the dark line of unknown origin, located about 350 nm below the HTA AlN interface, is seen in CL as dark contrast. This finding serves to substantiate the hypothesis that this is indeed related to a defect, although it is conceivable that it was induced during the process of sample preparation. The HTA AlN layer exhibits a distinct, broad emission centered around 360 nm, as displayed in the inset of Fig. 6b. Similar emission has been reported to cause a parasitic emission in UV-C light-emitting diodes (LEDs) as a result of an increased oxygen content in the AlN.^18,19 As the CL yield of the epitaxial AlN is very low, structural defects, as seen in the BF image, cannot be resolved.

The results demonstrate the capabilities of performing CL in a conventional SEM operated in transmission mode, in combination with a segmented STEM detector. This allows parallel acquisition of the CL signal in combination with transmitted electronic signals, including BF, DF, and HAADF. Correlative analysis of the CL and the HAADF and BF signal facilitates identification of compositional variations and structural defects. Combined with Monte Carlo simulations, the BF signal can be used to quantify the sample thickness. For the specimen under investigation, this reveals that below 20 nm, CL is no longer observed for 30 keV primary electrons. While the HAADF is highly sensitive to chemical contrast, it strongly depends on sample thickness due to dynamic electron scattering effects, which only allows for composition quantification with the sample thickness being known. Combining the CL with the BF signal is advantageous for correlative studies of extended defects, such as dislocations.

Integrating a SEM cryo-stage with a STEM detector is highly desirable, as it enables low-temperature CL analysis in combination with HAADF or BF imaging. This allows for correlated investigations of structural and optical properties in electron-transparent samples. The larger dimensions of the SEM sample chamber, compared to a conventional STEM, make such modifications more feasible and easier to implement.