1Laboratory for Electron Microscopy, Karlsruhe Institute of Technology (KIT), Campus South, Engesserstr. 7, 76131 Karlsruhe, Germany
2Institute of Toxicology and Genetics, Karlsruhe Institute of Technology (KIT), Campus North, Hermann-von-Helmholtz-Platz 1, 76344 Eggenstein-Leopoldshafen, Germany
Copyright © 2017 Thomas Kowoll et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This study is concerned with backscattered electron scanning electron microscopy (BSE SEM) contrast of complex nanoscaled samples which consist of SiO2 nanoparticles (NPs) deposited on indium-tin-oxide covered bulk SiO2 and glassy carbon substrates. BSE SEM contrast of NPs is studied as function of the primary electron energy and working distance. Contrast inversions are observed which prevent intuitive interpretation of NP contrast in terms of material contrast. Experimental data is quantitatively compared with Monte-Carlo- (MC-) simulations. Quantitative agreement between experimental data and MC-simulations is obtained if the transmission characteristics of the annular semiconductor detector are taken into account. MC-simulations facilitate the understanding of NP contrast inversions and are helpful to derive conditions for optimum material and topography contrast.
Scanning electron microscopy (SEM) is a widely applied characterization technique to study the properties of nanoparticles (NPs). Secondary electron (SE) SEM images provide information on surface topography, size, and size distribution of NPs. The atomic number sensitivity of backscattered electron (BSE) SEM images can be exploited to distinguish NPs with different chemical composition to reveal, for example, contamination particles. BSEs are by definition electrons with a kinetic energy above 50 eV. However, a significant BSE fraction is elastically scattered and a maximum of the energy distribution of BSEs exists close to the primary electron (PE) energy . BSE SEM images depend on the backscattered electron coefficient η which is defined by the number of BSEs per primary electron. A considerable amount of work has been devoted to measurements and calculations of the atomic number dependence of the BSE coefficient. Early work on BSE SEM imaging was performed by Niedrig  and Joy [2, 3], and a summary is given by Reimer . There has been recently a strong tendency towards lowering which was initiated by the improvement of the electron optics and electron sources. Lower values reduce the size of the interaction volume and lead to a substantial improvement of the resolution. This applies in particular for images with BSEs with exit depths that are large compared to secondary electrons. Along this line, Cazaux  summarized measured and calculated values for a wide range of elements at electron energies below 5 keV. Overall considerable knowledge is available on and its dependence on the atomic number and orientation of the sample surface with respect to the incident beam and .
Despite that, BSE SEM imaging has been rarely exploited to quantify information from BSE images. This must be attributed to the fact that the properties of the detection system as well as brightness and contrast settings need to be properly controlled and taken into account for the image evaluation. One of the few examples was presented by Sánchez et al.  who determined the average atomic number of a set of polished metal and mineral samples based on the measured BSE intensity. Instrumental effects were taken into account by using reference samples with known , and samples with unknown were studied without changing the imaging parameters.
Topography effects can be excluded in BSE images if polished surfaces are analysed. This cannot a priori be assumed for nanoscaled objects with a pronounced topography like NPs. Moreover, the influence of the substrate on NP contrast cannot be neglected. A study of gold NPs imaged by BSEs was published by Hirsch et al. . They investigated BSE SEM contrast of Au NPs with sizes between 2 nm and 40 nm on a silicon substrate. Monte-Carlo- (MC-) simulations of BSE contrast revealed pronounced dependence of the NP contrast on . A contrast maximum occurs if the NP diameter corresponds to the electron range in gold, which requires the adaption of to the NP size. The NP contrast decreases and even vanishes if is significantly increased beyond the optimum . The BSE contrast of NPs on bulk substrates will be also affected by changes of with decreasing [5, 7]. Overall, unexpected effects can be foreseen for BSE SEM images of NPs on bulk substrates.
In our study we focus on the contrast of SiO2 NPs on glass and glassy carbon substrates with an indium-tin-oxide (ITO) coating. This type of substrate is interesting for correlative SEM and light microscopy studies of biological samples as demonstrated by Pluk et al. . We aim towards a quantitative understanding of BSE contrast as a function of the working distance and PE energy to optimize NP contrast. MC-simulations are employed to simulate the NP contrast on the ITO/glass and ITO/carbon substrates for comparison with experimental data to understand contrast formation. We will show that the properties of the used annular semiconductor detector need to be taken into account for quantitative comparisons of simulated and experimental data. This concerns the limited detection angle range and decreasing detector efficiency for low-energy electrons. Substantial deviations from the expected -contrast and even contrast inversions occur due to the small NP size, topography effects, and the complex substrate structure with a high surface coating in combination with small values. Finally, we will provide a generally applicable strategy on how BSE contrast of nanoscaled objects on complex bulk substrates can be optimized.
2. Materials and Methods
2.1. Sample Preparation
The investigated samples are a product of studies on SiO2 NP uptake in A549 cancerogenous human lung epithelial cells. In these experiments cells, cultured as previously described by Panas et al. , were seeded onto ITO-coated substrates and incubated with nonporous amorphous SiO2 NPs (Postnova Analytics, Landsberg am Lech, Germany) with a diameter of 90 ± 8 nm (nominal diameter 100 nm according to manufacturer information). SiO2 NPs were deposited directly on the ITO-coated substrates between the cells. These regions are studied in the present work. The substrates were contacted with conductive silver on aluminium sample holders (Plano GmbH, Wetzlar, Germany) with a diameter of 32 mm for the SEM investigations.
Two different substrate types were used. Type 1 consists of 160 ± 5 nm thick ITO layers on glass (amorphous SiO2) denoted by ITO160 in the following. For comparison, glassy carbon substrates covered by 22 ± 5 nm thin ITO layers were used denoted by ITO22. The second substrate type was chosen because glassy carbon is characterized by a low value compared to the SiO2 substrate. It will later become obvious that the thin ITO layer on the glassy carbon substrate will considerably contribute to the understanding of BSE image formation. The ITO160 substrates were purchased from PGO (Iserlohn, Germany), while the ITO22 ones were manufactured in-house by electron-beam deposition (Kurt J. Lesker Company, Hastings, UK) from ITO pieces on glassy carbon substrates (HTW, Thierhaupten, Germany).
2.2. Transmission Electron Microscopy
Transmission electron microscopy (TEM) and scanning transmission electron microscopy (STEM) combined with energy dispersive X-ray spectroscopy (EDXS) were performed to determine the thickness and composition of the ITO layers. (S)TEM was performed with a FEI Osiris ChemiSTEM operated at 200 kV and equipped with four Bruker silicon drift detectors. TEM cross-section specimens were prepared by focused ion-beam (FIB) milling. The measured compositions are included in Table 1. We note that substantial discrepancies between experimental and simulated BSE contrast of NPs result if the nominal instead of the real compositions are used in the MC-simulations.
Table 1: Material parameters for NP and substrates used in MC simulations.
Figure 1 shows high-angle annular dark-field (HAADF) STEM images of ITO160 (Figure 1(a)) and ITO22 (Figure 1(b)) in cross-section perspective. The ITO layers show bright contrast compared to the glassy carbon and SiO2 substrates. The samples are covered by Pt-layers for protection during FIB milling. The dark layer between ITO160 and the platinum layer in Figure 1(a) is an additional carbon layer, which was necessary to improve electrical conductivity for other experiments and has no further relevance for this work. The ITO layer in ITO160 appears dense with a homogeneous contrast and exhibits thickness variations of ±5 nm. The inhomogeneous contrast of the ITO layer in ITO22 suggests that some porosity is present and indicates that the average ITO density is smaller than the nominal one. MC-simulations performed with a reduced density of instead of 7.1 g/cm3 for ITO22 indeed agree well with the experimental results.
Figure 1: 200 keV cross-sectional HAADF STEM images of (a) ITO160 and (b) ITO22 covered by a protective Pt-layer.
2.3. Scanning Electron Microscopy
All samples were investigated in a FEI Quanta 650 ESEM, equipped with a Schottky field emission gun and an annular silicon solid-state BSE detector with an active detector area of approximately 200 mm2. The specimen stage was untilted and images were taken at normal incidence. SE images were acquired with an Everhart-Thornley detector. Three series of BSE images were acquired for ITO160 and ITO22 by varying between 3 and 17 keV while keeping the working distance (WD) constant at 4 mm, 6 mm, and 10 mm. Moreover, two series were acquired with different WD between 4 and 12 mm at constant of 5 keV and 10 keV. Every image was taken at different, but adjacent specimen areas to minimize contamination artefacts.
Images with 2048 × 1768 pixels and 1.46 nm pixel size were taken corresponding to a magnification of ×100.000. Spot size 3, 10 μs dwell time, and 16-bit greyscale resolution were chosen. Brightness and contrast values were adjusted to strictly avoid over- and undersaturation of the signal over a whole image series. For energies below 5 keV, the contrast had to be occasionally increased within one series to obtain reasonable signal intensities. The investigation of the influence of contrast and brightness variations on the measured NP contrast at = 5 keV and WD = 4 mm showed that only brightness alters the NP contrast considerably, whereas varying contrast setting only leads to minor changes. For each detector setup, that is, brightness and contrast settings, images with a blanked beam were acquired to obtain black level intensities (see data analysis section below).
2.4. Data Analysis
The NP contrast is given bywith the NP image intensity , the image intensity of the substrate , and the black level intensity of the particular detector setup . The images were analysed with the software ImageJ . To prevent errors in the subsequent averaging process, first potential bright and dark outliers (dark pixels and hotspots) were removed using the respective software feature. This filter replaces a pixel with the median intensity of the pixels in its surrounding if the pixel intensity differs by more than a certain threshold value from the median. The filter parameters were set as follows: pixels = 7 and threshold = 2000. Using the oval selection tool, a circle with a diameter between 40 nm and 50 nm was placed concentrically on a NP. The intensities of individual NPs were obtained by averaging the pixel intensities within the selected region. The substrate intensity was obtained by averaging the intensities of 10 large free areas at different positions. Finally, was measured for each detector setup using images acquired with blanked beam. The contrast of each NP was calculated on the basis of (1). Finally the average contrast was determined by averaging all . The resulting error represents the standard deviation of from an ensemble of 20 NPs. The evaluation of a relatively small NP ensemble is justified by the relatively small standard deviations of the NP contrast.
2.5. Monte-Carlo Simulations
MC-simulations were performed with the NISTMonte program  which was modified to take the detector properties into account. Two structural models were defined to calculate substrate and NP BSE intensities. Bulk substrates (amorphous SiO2 or glassy carbon) covered by an ITO layer with 160 or 22 nm thickness are assumed. For the calculation of the NP BSE intensity, a SiO2 NP with a diameter of 90 nm is placed on top of the substrate. Material parameters for NP and substrates are summarized in Table 1.
Simulation parameters were set as follows: 106 trajectories, Gaussian beam with full width at half maximum of 1 nm, and PE energies between 2 and 17 keV. We refrain from using PE energies below 2 keV, because the scattering cross-sections, especially Screened Rutherford cross-sections, fail to describe the backscattering coefficient η as surface barrier effects come into play . Screened Rutherford (ScR)  and Czyzewski (Cz) Mott  scattering cross-sections were used and compared with respect to their validity to describe the experimental data.
One modification of NISTMonte is related to the operation principle of a semiconductor BSE detector. The detected BSE intensity in a semiconductor detector is determined by the number of electron-hole pairs which are generated by the BSEs. BSEs with different kinetic energies correspondingly generate different numbers of electron-hole pairs. Hence, the measured BSE signal cannot be directly attributed to the number of BSEs but depends also on the energy of each BSE. The calculation of the energy loss by the continuous slowing down approximation by Joy and Luo  is already implemented in NISTMonte. However, the energy of the individual BSEs is discarded in further processing. For the MC-simulations in this work, we sort BSEs into bins according to their scattering angle and monitor the BSE energy in addition to the BSE number. The number of electron-hole pairs generated in the semiconductor detector can be calculated by summing up the BSE energies in the bins for the corresponding scattering angle range. Since the greyscale value in a BSE image is proportional to the number of electron-hole pairs generated in the detector, the NP contrast is calculated bywith the overall BSE energy for the NPs on the substrate and for the mere substrate .
Furthermore, a correction related to the efficiency of Si-detectors in converting electrons into electron-hole-pairs is included. The detector efficiency decreases with decreasing energy of the detected electrons due to the front metal coating of the detector. This can be described by a linearly decreasing transmission probability through the protective layer for BSEs with energies starting from the threshold energy , which denotes the electron energy at which 100% transmission is achieved. For higher BSE energies the transmission is 100% and does not depend on anymore.
Figure 2 shows the assumed transmission characteristic of the protective layer as a function of . was set to 3 keV according to the information provided by the microscope manufacturer. The value of the threshold energy is important for low-energy BSE imaging because it determines an additional energy loss of the BSEs before they reach the detector. Hence, the energy of each electron has to be corrected before it is assigned to a bin taking into account the reduced transmission probability and the additional energy loss. This can be achieved by integrating the transmission curve shown in Figure 2 from 0 to , which yields the transmitted energy if is below (3). Equation (4) describes the energy correction for BSEs with energies above .The distance between detector and specimen determines the detected angular BSE distributions which need to be exactly known for the MC-simulations. The nominal WD settings of the microscope, however, denote the distance between specimen surface and pole piece and must be reduced by the BSE detector thickness of . Hence, the nominal WDs were corrected in this work for the calculation of the minimum and maximum scattering angles. The smallest WD of 4 mm corresponds to an angular range of 1.78 rad–2.13 rad. It increases to 2.43 rad–2.85 rad for the largest WD = 12 mm. The scattering angle θ is defined as the angle with respect to the electron incidence direction. θ = 0 rad corresponds to forward scattering, while θ = 3.14 rad corresponds to backscattering perpendicular to the surface.
Figure 2: Scheme of BSE transmission for a semiconductor detector as a function of BSE energy.
The errors for the MC-data were calculated according to Gaussian error propagation on the basis of uncertainties with respect to the ITO density of ±0.5 g/cm3, SiO2 density of ±0.2 g/cm3, NP diameter of ±8 nm, and ITO layer thickness of ±5 nm. Statistical errors can be neglected due to the large number (106) of simulated electrons.
3. Experimental Results
The NP contrast was systematically investigated as a function of the working distance and primary electron energy for the two different substrates.
Figure 3 illustrates the dependence of the NP contrast on the WD at 5 keV on the ITO160 substrate. SE contrast in Figures 3(a)–3(c) is only weakly affected by the WD whereas the BSE contrast (Figures 3(d)–3(f)) changes considerably. The NP contrast is positive for 4 mm WD (Figure 3(d)) which is unexpected considering the average atomic number of SiO2 NP ( = 10) and the ITO layer ( ≈ 28.5). The BSE contrast decreases to a very low value at approximately 6 mm WD (Figure 3(e)) and is inverted for further increasing WDs, for example, at 10 mm (Figure 3(f)). The NPs show a diffuse dark contrast under these conditions. Surprisingly SE and BSE contrast is similar at 4 mm WD (Figures 3(a) and 3(d)). The topography of the ITO layer (tile-like structures) can be clearly resolved in both images. This is an indication for a superposition of BSE material contrast and topography contrast, which will be discussed in detail later.
Figure 3: 5 keV SE (a–c) and BSE (d–f) images of SiO2 NPs on ITO160 as a function of the WD given in the images. The image contrast was postprocessed for optimum visibility.
Figures 4(a)–4(c) show BSE images of SiO2 NPs on ITO160 obtained with different at 10 mm WD. The NP contrast is negative for 3 keV (Figure 4(a)). It is inverted if the PE energy exceeds 10 keV, where the contrast vanishes (Figure 4(b)). The contrast values remain negligible for further increasing and NPs can be hardly recognized. Yet, digital image analysis yields a measurable contrast of 0.04 ± 0.01 for = 17 keV (Figure 4(c)).
Figure 4: BSE images of NPs on ITO160 acquired at (a) = 3 keV, (b) = 10 keV, and (c) = 17 keV at a constant WD of 10 mm. The image contrast was postprocessed for optimum visibility.
A PE energy dependent contrast inversion is also observed for the second substrate ITO22. This is illustrated in Figures 5(a)–5(c) which show BSE images of SiO2 NPs obtained at 10 mm WD with 3 keV, 8 keV, and 17 keV electrons. NP contrast is negative at the lowest PE energy of 3 keV (Figure 5(a)) and already clearly inverted at 8 keV (Figure 5(b)) indicating that the contrast inversion takes place between 3 and 8 keV. The contrast decreases slightly for the highest energy (Figure 5(c)). It is noted that the topography of the ITO22 substrate shows smaller-scale features than ITO160, yet bright and dark areas can be distinguished indicating some roughness.
Figure 5: BSE images of NPs on ITO22 acquired at (a) = 3 keV, (b) = 8 keV, and (c) = 17 keV at a constant WD of 10 mm. The image contrast was postprocessed for optimum visibility.
Another contrast inversion can be observed if BSE images of NPs on both substrates are directly compared. Figures 6(a) and 6(b) show 6 keV BSE images of NPs on ITO160 (Figure 6(a)) and ITO22 (Figure 6(b)) at WD. Although identical imaging parameters were chosen, NP contrast is negative for ITO160 and positive for ITO22. At first glance this indicates dependence on the substrate material, but it will be shown in the following that it is the result of the different ITO thicknesses.
Figure 6: 6 keV BSE images of (a) SiO2 NP on ITO160 and (b) SiO2 NP on ITO22 taken at WD = 10 mm. The image contrast was postprocessed for optimum visibility.
A general characteristics of negative NP contrast are the diffuse NP appearance without any indication of the NP topography (Figures 3(e), 3(f), 4(a), 5(a), and 6(a)). Images of NPs with positive contrast show a substantially improved resolution.
4. Comparison of Measured NP Contrast with MC-Simulations
Contrast inversions observed in Figures 3–6 suggest that simple interpretation of BSE images in terms of material contrast is not adequate for complex sample structures. In the following we will elaborate a systematic approach to understand and optimize NP contrast. Particularly the latter goal is motivated by the tedious and time-consuming trial and error procedure for a particular scenario. To understand BSE contrast formation we compare MC-simulations with experimental data in Figures 7–9. Square symbols and solid lines represent measured contrast values. Two different scattering cross-sections are used in the MC-simulation because the optimum choice of the scattering cross-section depends on the PE energy and the atomic number of the specimen materials. Circular symbols with dashed lines represent MC-simulations performed with ScR scattering cross-sections and triangular symbols with dashed lines indicate MC-simulations performed with Cz Mott cross-sections.
Figure 7: BSE contrast of SiO2 NP on ITO160 and ITO22 as a function of the WD. Comparison of experimental data (square symbols/solid lines) and MC-simulations using Screened Rutherford (circular symbols/dashed lines) and Cz Mott (triangular symbols/dashed lines) scattering cross-sections for (a) = 5 keV and (b) = 10 keV. Note the contrast inversion for ITO160 in (a) at a WD of 6 mm.
Figure 8: SiO2 NP BSE contrast as a function of at 4 mm WD on (a) ITO160 and (b) ITO22. Experimental data are displayed by square symbols/solid lines, MC-simulations on the basis of Screened Rutherford are indicated by circular symbols/dashed lines and on the basis of Cz Mott cross-sections by triangular symbols/dashed lines.
Figure 9: SiO2 NP BSE contrast as a function of at 10 mm WD on (a) ITO160 and (b) ITO22. Experimental data are displayed by square symbols/solid lines; MC-simulations on the basis of Screened Rutherford are indicated by circular symbols/dashed lines and on the basis of Cz Mott cross-sections by triangular symbols/dashed lines.
The errors of the MC-simulations for NP on ITO22 are in general higher compared to NP on ITO160. This is related to the different thickness of the ITO layers, because all parameter variations have a much stronger impact on the contrast for ITO22 than for the thicker ITO layer, especially the thickness variation by ±5 nm. This illustrates clearly the necessity of precise knowledge of simulation parameters for MC-simulations of complex nanoscaled structures.