Cathodoluminescence Microscopy of nanostructures on glass substrates

Cathodoluminescence (CL) microscopy is an emerging analysis technique in the fields of biology and photonics, where it is used for the characterization of nanometer sized structures. For these applications, the use of transparent substrates might be highly preferred, but the detection of CL from nanostructures on glass is challenging because of the strong background generated in these substrates and the relatively weak CL signal from the nanostructures. We present an imaging system for highly efficient CL detection through the substrate using a high numerical aperture objective lens. This system allows for detection of individual nanophosphors down to thirty nanometer in size as well as the up to ninth order plasmon resonance modes of a gold nanowire on ITO coated glass. We analyze the CL signal-to-background dependence on the primary electron beam energy and discuss different approaches to minimize its influence on the measurement. ©2013 Optical Society of America OCIS codes: (180.0180) Microscopy; (170.5810) Cathodoluminescence; (310.6628) Subwavelength structures, nanostructures; (250.5403) Plasmonics; (170.6280) Spectroscopy, fluorescence and luminescence. References and links 1. J. Goetze and U. Kempe, “A comparison of optical microscopeand scanning electron microscope-based cathodoluminescence (CL) imaging and spectroscopy applied to geosciences,” Mineral. Mag. 72(4), 909–924 (2008). 2. E. J. R. Vesseur, R. de Waele, M. Kuttge, and A. Polman, “Direct observation of plasmonic modes in Au nanowires using high-resolution cathodoluminescence Spectroscopy,” Nano Lett. 7(9), 2843–2846 (2007). 3. T. Coenen, E. J. R. Vesseur, A. Polman, and A. F. Koenderink, “Directional Emission from Plasmonic Yagi-Uda Antennas Probed by Angle-Resolved Cathodoluminescence Spectroscopy,” Nano Lett. 11(9), 3779–3784 (2011). 4. A. I. Denisyuk, G. Adamo, K. F. MacDonald, J. Edgar, M. D. Arnold, V. Myroshnychenko, M. J. Ford, F. J. García de Abajo, and N. I. Zheludev, “Transmitting Hertzian Optical Nanoantenna with Free-Electron Feed,” Nano Lett. 10(9), 3250–3252 (2010). 5. R. Sapienza, T. Coenen, J. Renger, M. Kuttge, N. F. van Hulst, and A. Polman, “Deep-subwavelength imaging of the modal dispersion of light,” Nat. Mater. 11(9), 781–787 (2012). 6. H. Niioka, T. Furukawa, M. Ichimiya, M. Ashida, T. Araki, and M. Hashimoto, “Multicolor Cathodoluminescence Microscopy for Biological Imaging with Nanophosphors,” Appl. Phys. Express 4(11), 112402 (2011). 7. D. R. Glenn, H. Zhang, N. Kasthuri, R. Schalek, P. K. Lo, A. S. Trifonov, H. Park, J. W. Lichtman, and R. L. Walsworth, “Correlative light and electron microscopy using cathodoluminescence from nanoparticles with distinguishable colours,” Sci. Rep. 2, 865 (2012). 8. Y. Nawa, W. Inami, A. Chiba, A. Ono, A. Miyakawa, Y. Kawata, S. Lin, and S. Terakawa, “Dynamic and highresolution live cell imaging by direct electron beam excitation,” Opt. Express 20(5), 5629–5635 (2012). 9. R. Gomez-Medina, N. Yamamoto, M. Nakano, and F. J. García de Abajo, “Mapping plasmons in nanoantennas via cathodoluminescence,” New J. Phys. 10(10), 105009 (2008). 10. M. Kuttge, E. J. R. Vesseur, A. F. Koenderink, H. J. Lezec, H. A. Atwater, F. J. Garcia de Abajo, and A. Polman, “Local density of states, spectrum, and far-field interference of surface plasmon polaritons probed by cathodoluminescence,” Phys. Rev. B Condens. Matter 79(11), 113405 (2009). #194914 $15.00 USD Received 30 Jul 2013; revised 25 Oct 2013; accepted 31 Oct 2013; published 27 Nov 2013 (C) 2013 OSA 2 December 2013 | Vol. 21, No. 24 | DOI:10.1364/OE.21.029968 | OPTICS EXPRESS 29968 11. B. Dierre, X. Yuan, and T. Sekiguchi, “Low-energy cathodoluminescence microscopy for the characterization of nanostructures,” Sci. Technol. Adv. Mater. 11(4), 043001 (2010). 12. L. H. G. Tizei and M. Kociak, “Spatially Resolved Quantum Nano-Optics of Single Photons Using an Electron Microscope,” Phys. Rev. Lett. 110(15), 153604 (2013). 13. A. Ishikawa, F. Mizuno, Y. Uchikawa, and S. Maruse, “High resolution and spectroscopic cathodoluminescent images in Scanning Electron-Microscope,” Jpn. J. Appl. Phys. 12(2), 286–292 (1973). 14. A. C. Zonnevylle, R. F. C. Van Tol, N. Liv, A. C. Narvaez, A. P. J. Effting, P. Kruit, and J. P. Hoogenboom, “Integration of a high-NA light microscope in a Scanning Electron Microscope,” J. Microsc. 252(1), 58–70 (2013), doi: 10.1111/jmi12071. 15. N. Liv, A. C. Zonnevylle, A. C. Narvaez, A. P. J. Effting, P. W. Voorneveld, M. S. Lucas, J. C. Hardwick, R. A. Wepf, P. Kruit, and J. P. Hoogenboom, “Simultaneous Correlative Scanning Electron and High-Na Fluorescence Microscopy,” PLoS ONE 8(2), e55707 (2013). 16. T. H. Taminiau, S. Karaveli, N. F. van Hulst, and R. Zia, “Quantifying the magnetic nature of light emission,” Nat. Commun. 3, 979 (2012). 17. F. J. Garcia de Abajo, “Optical excitations in electron microscopy,” Rev. Mod. Phys. 82(1), 209–275 (2010). 18. L. Novotny, “Effective wavelength scaling for optical antennas,” Phys. Rev. Lett. 98(26), 266802 (2007). 19. D. Rossouw, M. Couillard, J. Vickery, E. Kumacheva, and G. A. Botton, “Multipolar Plasmonic Resonances in Silver Nanowire Antennas Imaged with a Subnanometer Electron Probe,” Nano Lett. 11(4), 1499–1504 (2011). 20. L. Reimer, Scanning Electron Microscopy: Physics of Image Formation and Microanalysis: Ludwig Reimer (Springer, 1985). 21. M. Kuttge, “Cathodoluminescence Plasmon Microscopy,” PhD Thesis at Utrecht University (2009). 22. C. W. Hagen and P. Kruit, “Optimization of focused ion beam performance,” J. Vac. Sci. Technol. B 27(6),


Introduction
Cathodoluminescence (CL) based techniques use the excitation of light by an electron beam to study the optical and electric properties of a material.As a characterization tool, CL has applications in many research fields, e.g. in geosciences, where it is used to identify minerals, crystalline phases and the presence of defects in solids [1].Here, the structures of interest typically have lateral dimensions in the order of micrometers.More recently, CL is being used to map the local density of states (LDOS) of photonic nanostructures: resonances of, e.g.optical antennas, dielectric cavities, and their effect on the properties of nearby emitters can be investigated at a resolution far below the wavelength [2][3][4][5].In cell biology, protein distributions could be visualized directly in electron microscopy (EM) using biofunctionalized CL generating nanolabels, in a process similar to immuno-labeling for fluorescence microscopy.Nano-phosphors, quantum dots (QD) and nano-diamonds with nitrogen-vacancy (NV) centers, are being investigated as potential bio-markers [6][7][8].For these emergent applications, it is important to optimize the CL detection efficiency, as the signal generated by nanoscaled structures can be very weak.
The most common CL systems use either a Scanning Electron Microscope (SEM) or a Transmission Electron Microscope (TEM).CL detection is in general performed from above the sample using a parabolic mirror that contains a small hole for the passage of the electron beam [9][10][11][12].However, this limits the amount of generated signal that can be detected, because most of the light will be emitted into the higher refractive index, usually the sample substrate.As an example, for a dipole emitter oriented perpendicular to a glass-vacuum interface 86% of the light goes into the glass (see Fig. 1(a)).Detection through a transparent substrate would then be beneficial to increase the collection efficiency.
Through-substrate CL detection could be performed with an objective lens (OL) placed below the sample, inside the vacuum chamber of a SEM.This configuration was already explored in 1973 by Ishikawa and associates.The OL collects the light, which is then directed by a mirror towards a photon detector on a side port of the chamber [13].In their work, the sample was mounted on an araldite substrate.However, to our knowledge, no further work has been reported on similar CL systems since then.A reason for this could be the requirement to mount the sample on araldite, which limits applicability.Other transparent substrates, like regular glass microscope slides, are known to generate a strong CL signal.Such CL as a background can easily overwhelm the signal from the structure of interest.
Another possible limitation the technique may have encountered could be the performance of the objective lenses available by the time, affecting the collection efficiency and therefore the success of the method.
Nevertheless, CL detection with an OL collecting light through the substrate would not only increase the signal detection efficiency, but it would bring additional advantages: (i) as the OL is placed below the sample, it does not block other detectors in the SEM vacuum chamber, and (ii) the symmetry of the OL system makes it easy to interpret the detection optics and correct for aberrations.Also, if substrates like glass could be used, it would extend the possibilities for CL microscopy, as in most of the CL characterizations reported so far the nanostructures are mounted on Si-based or other non-transparent conductive supports.Glass substrates could be particularly beneficial for the applications in biosciences and nanophotonics, because of (i) compatibility with other optical techniques that can complement the information retrieved from CL, (ii) bio-compatibility and the wide range of available surface functionalization techniques for glass or conductive oxide surfaces, (iii) applications in nanophotonics, e.g. in sensing and detection, may ultimately need a transparent substrate.Our aim is to accomplish efficient CL microscopy of potential biomarkers and plasmonic nanostructures supported on glass substrates.
We have recently built an integrated light and electron microscope in which an inverted high numerical aperture (NA) objective lens is placed in a SEM [14] (see Fig. 1(b)).In this system light microscopy and scanning electron microscopy can be performed simultaneously on the same area of a sample [15].The alignment of the light and electron axes guarantees a direct spatial correlation of the different imaging modes, which is particularly useful to perform CL detection through the substrate.Here, we will present the implementation of CL microscopy, including spectroscopic measurements, and evaluate the potential of the technique for CL characterization of nanostructures on transparent substrates.Considering the relatively weak signal from nanostructures and the background generated in the substrate, a main challenge is to optimize the CL signal to background ratio.

CL microscope
Figure 1(b) shows the system with an objective lens (CFI Plan Apochromat 40X 0.95 NA from Nikon) mounted inside the vacuum chamber of a commercial SEM (Quanta 200 FEG microscope from FEI).The integration has been performed by modifying the vacuum door of the microscope [14].The electron beam comes from the top where it interacts with the sample.In this process secondary electrons (SE), backscattered electrons (BSE) and CL are generated.To detect the CL, the objective lens collects the light and directs it outside the vacuum chamber as a collimated beam, as indicated in Fig. 2(a).A breadboard is attached on the outside of the door, where optical paths to different detectors are mounted.Flipping mirrors are used to switch the detection mode.In our setup, 3 optical channels are enabled: the first one focuses the light onto a CCD camera to form a wide field image (here only used for alignment of the optical path); a second one uses a photomultiplier tube (PMT) to detect CL intensity.The last path couples the light to a spectrometer (Princeton Instruments Acton SP2156 spectrograph in combination with a PyLoN:100BR-Excelon CCD camera) via a fiber (See Fig. 2(a)).
Alignment between the electron and the optical axes is done by scanning the electron beam over the sample while detecting the light intensity in the PMT for each position, so that a CL map of the area under exposure is obtained.For a uniform CL sample, a radially symmetric intensity distribution is observed, where the maximum value in the distribution corresponds to the center position of the OL and the full width at half maximum (FWHM) of the profile is determined by the smallest aperture along the optical system.The OL can be mechanically translated until the maximum of the optical signal matches the center of the area scanned with the electron beam (Fig. 2(d)).The markers on the substrate in Figs.For areas which are small compared to the variation in the detection efficiency of the OL, a CL map can be obtained by scanning the electron beam and recording the CL signal at each position.Alternatively, a CL image can be acquired by scanning the sample, while the electron beam is kept fixed in one position.The sample holder in our system is equipped with both long-range piezo translators (Vacuum-Compatible Miniature Translation Stages M662.v, from Physik Instrumente (PI) GmbH & Co) and a xyz-piezo scanner (TRITOR 100 from piezosystem jena, Inc).Wavelength information of the emission is obtained by enabling the spectrometer path during the scanning or when the electron beam is focused in a spot or region of interest (ROI).The CL acquisition (wide field, intensity or spectral) does not interfere with the standard SE, BSE and other available SEM detection modes.
The CL radiation can have a preferential direction depending on position and orientation of the emitter.A dipole on an interface or embedded in a thin film on a dielectric substrate has emission peaks near to angles corresponding to NA = 1 (see Fig. 1(a) and, e.g., also [16]).To collect light emitted near the total internal reflection angle, a high numerical aperture (NA) is required.We achieve a NA of 1.4 (CFI Plan Apochromat 100X from Nikon) using vacuum compatible immersion oil.For the perpendicular dipole example in Fig. 1(a), 99% of the downward emission would be collected with a 1.4 NA OL, compared to 88% for NA = 0.95.Additional losses due to elements along the optical path and the limited efficiency of the photodetectors will reduce the total detection efficiency to typical values of 5-10%.

Nano-phosphors cathodoluminescence
To illustrate CL microscopy on glass substrates, we examined a sample of nanophosphors ranging in size from 30nm to 100nm.Recently, such nanophosphors have gained interest because of their strong emission intensity and stability under electron-beam irradiation.Additionally, their potential for bio-functionalization make them ideal candidates for use as biomarkers in electron microscopy [7]. Figure 3 shows SEM and CL images of Ce:LuAG nanophosphors (from Boston Applied Technologies), diluted in ethanol and drop-cast on a ITO covered glass slide (170µm thick glass coverslip coated with ~70nm thick ITO layer, from SPI Supplies).The images were recorded with a 5kV electron beam and a probe current of approximately 0.5 nA.Clusters of particles of varying size can be clearly discerned in both images.By comparing the SEM and CL images (Figs.3(a) and 3(b)) it can be seen that light is excited in a region larger than the phosphors, which appears as a 'halo' around the structures in the CL images.This effect is mainly associated to backscattered electrons that emerge from the glass at a small angle and hit the phosphors, even when the primary electron beam is not directly exciting them.
The CL images show that the intensity drops with the particle size.Still, nanometer sized phosphors can easily be observed in CL, as it is shown in Fig. 3(b) for a 50nm particle.Also the smallest particle size in the nano-phosphor dispersion, 30 nanometer, can be observed in CL (Figs. 3(c) and 3(d)).Figure 3(d) shows the CL intensity profile along two of these particles, placed near to each other.Again, we attribute the slightly increased CL intensity in between the particles to excitation of both nano-phosphors via backscattered electrons originating from the underlying substrate.

Cathodoluminescence from plasmonic nanowires
The phosphor nanoparticles shown above are tailored for CL applications as the CL crosssection is relatively large.In the case of CL spectroscopy of radiative modes in plasmonic nanostructures, the excitation probability is much more reduced.For instance, the electron generation of surface plasmon polaritons (SPP), one of the main excitation sources of CL in metal nanostructures such as optical nanoantennas, is very low: around 3X10 −7 SPP per electron (per nm SPP wavelength) are generated with a 30kV e-beam impinging on a gold surface [10,17].We examine the capability of our system for plasmonic inspection using a sample of gold NWs, again on an ITO-coated glass substrate.The incident electron beam excites charge density oscillations confined to the metal-dielectric interface of the NW.These surface plasmons travel forth and back forming standing waves along the NW, which behaves as a resonator.Plasmons can out-couple as electromagnetic radiation, where the emission wavelength is determined by the plasmons excited, which in turn depend on the NW composition and dimensions, as well as on the position of the excitation probe.
Figure 4(b) shows a CL map obtained for a gold NW with a 2kV electron beam.Intensity maxima can be clearly observed at the NW endpoints and are related to a higher surface plasmon excitation probability.To characterize the plasmon mode profile of the NW we have acquired the CL spectral distribution at one of the endpoints.These are shown in Fig. 4(c).At least five wavelengths peaks can be identified at around 689nm, 730nm, 783nm, 852nm, and 951nm.To confirm that the modes observed correspond to the NW surface plasmon resonances, we calculated the expected modes assuming a one-dimensional Fabry-Pérot resonator for a gold NW 1.4µm long and with a diameter of 70nm.The resonant wavelengths were scaled to out-coupled radiation following the method described by Novotny [18].The calculated resonances were compared with the peaks from the measurement, where the best fit is shown in Fig. 4(d).From the plot, the peaks appear to correspond with the 5th up to the 9th order resonant modes.Here, the influence of the ITO/glass substrate has not been considered.
We can include this effect by modeling the NW behavior using finite-difference timedomain (FDTD) calculations.In these simulations, a dipole emitter is located 5nm away of one of the edges of the NW, placed on a vacuum/ITO interface.The dipole is oriented parallel to the NW long axis direction.The lower medium consist on a 70nm thick ITO layer on top of a semi-infinite glass medium.The calculations were performed using a commercial-grade simulator based on the FDTD method (Lumerical solutions, Inc.).
Figure 4(e) shows the fraction of the power dissipated towards the glass over the total power emitted by the dipole, together with the measured spectra.Again, there is a good agreement between both curves.We can obtain the spatial electric field distribution on a plane parallel to the interface (crossing the NW) to identify the resonant orders (see insets in Fig. 4(c)).The mode profiles indicate this time that the measured peaks correspond to the 6th up to the 10th resonant mode.Additional simulations (not included here) suggest that small variations (within 30nm) of the ITO thickness can induce a significant shift on the peak positions, adding uncertainty to the identification of the mode order.Further discussion about the effect of the ITO layer on the resonant peak position is outside the scope of this paper, but we would like to emphasize that both results match the measured peaks with high order modes (> 5th order).It is important to notice that for these high order resonances, the spectral shift of the peak positions induced by the surface plasmon phase change upon reflection at the ends of the NW can be neglected [19].
The images in Fig. 4(c) also show that the modes are almost uniformly distributed along the NW, so that we cannot observe a strong variation in the spectra by changing the e-beam position.Low order modes, which are more sensitive to the excitation probe position, are outside the detection range of the spectrometer (>1000nm, see also Fig. 4(d)).
Finally, both models also predict that the peak intensity (or excitation probability) decreases with the mode order.The wavelength dependence of the quantum efficiency of our spectrometer CCD accounts for the relatively smaller peak of the lowest order mode at 950nm.CL from the substrate is also observed as a wide background with a peak around 650nm-700nm (See Fig. 4(c)).Given the overall low plasmon CL excitation probability mentioned above, the fact that we are able to distinguish up to the 9th/10th order resonant mode clearly shows the detection efficiency of our system.

Discussion
The results above demonstrate successful CL imaging of nanostructures on glass based substrates.This can be partially attributed to the use of a high NA objective lens, which improves the detection efficiency and, for structures with a high angle emission profile, the signal to background ratio (s/b).However, a crucial factor is also the proper selection of the primary electron beam settings: in all situations described we were restricted to relatively low values for the acceleration voltage, and it was necessary to optimize the current for each case individually.This last fact was because the influence of the (substrate) background in relation to the (nanostructures) signal strength is sample specific.Especially for glass-based substrates, structural defects and impurities give rise to strong background signals.
To understand how the electron beam settings affect the (s/b) in a CL measurement, we consider the situation illustrated in Fig. 5(a).Below the focus of the e-beam, electrons scatter and further interact with the sample, defining an interaction volume.CL can be excited directly by the primary electron beam but also by scattered electrons in the interaction volume.In addition, CL can be indirectly excited by x-rays, or by charge carriers that diffuse and recombine [20].All these contributions build up the light signal on the detector.Most of the interaction volume emission will generate background, while only the small area where the e-beam is focused on the nanostructure will constitute the signal.
The depth of the interaction volume is proportional to the acceleration voltage.Additionally, the CL intensity increases with the acceleration voltage [17,21].Neglecting sample damage or saturation, the CL intensity is also proportional to the electron beam current.While the last two effects include both signal and background, changing the interaction volume depth only affects the background intensity.The acceleration voltage can drastically change this depth, from 40nm for a 2kV to 7µm for a 30 kV electron beam, as shown in Fig. 5(b).The images were obtained using Monte Carlo simulations (CASINO v2.4.8.1) considering a glass substrate with a typical ITO coating of 70nm.At 2kV, the entire interaction volume falls in this ITO coating.A general approach to reduce the background contribution is therefore to lower the acceleration voltage: for a sufficiently small energy the interaction range can be restricted to the structure of interest.However, reducing the acceleration voltage decreases the CL intensity: coherent CL emission, as the one coming from SPPs and transition radiation, increases with the acceleration voltage (above 1kV) [15].This is also valid for CL rising from inelastic electron-electron collisions, where the CL signal is proportional to the current and the inelastic cross section (σ): CL ∝ I × σ.While at high energies both I and σ increase with the acceleration voltage (V), at low energies (between 1keV and 5keV), the cross section approximately follows 1/σ ∝ V 0.5 [20].However, this behavior is dominated by the current dependence on the voltage (I ∝ V, and even I ∝ V 3 in a chromatic aberration limited system) [22].
Consequently, reducing the acceleration voltage can make the signal fall below a detectable level.This can be compensated by increasing the electron current, but using this approach at low energies (< 5keV), results in an electron probe size d p limited by chromatic aberration, which depends on the current as I ∝ d p 4 × V 3 /ΔU2 (provided that the electron beam aperture angle has been optimized to obtain the best resolution), where ΔU is the FW50 of the energy distribution of the electron source [22], and the voltage V has been fixed and can be treated as a constant.Then, an increment in the current considerably enlarges the probe size.This explains the reduced spatial resolution observed in the SEM and CL images in Fig. 4, which were obtained with a 2kV and ~1nA electron beam.
The substrate background is less problematic for structures with a high CL generation rate like the nano-phosphors.In this case the use of low primary e-beam energies does not need to be compensated with an increase on the current, so that a reduced interaction volume (and therefore a reduced background) can be achieved while the spatial resolution is kept in the nanometer range (Figs. 3 and 6).
Figure 6 shows a 50nm phosphor particle imaged at different acceleration voltages.For 2kV, the background CL is low but the spatial resolution is limited, whereas for 10kV the nano-phosphor is barely visible.Line profiles taken along the particle show the signal and background levels for each case.For this particular sample and experimental settings, the best compromise between contrast and resolution is obtained at 5kV and ~0.5nA current.In general, there is a trade-off between the CL resolution and the background level that needs to be considered to select the sample specific optimum acceleration.
Other options to reduce the CL contribution from the glass include background subtraction and optical filtering.Background subtraction assumes a uniform behavior over all the imaged area, which is hardly the case because: 1) defects and residues present in the substrate from fabrication or added during sample preparation create background fluctuations, even on 100nm length scales (see e.g.Fig. 6(e)); 2) the substrate under the structures will emit a weaker signal than a region directly exposed to the e-beam; 3) the background can be modified during e-beam exposure.Band-pass optical filters can be included in the detection path, and are a good choice when the emission wavelength is known and narrow compared to the background, and/or the interest is on a wavelength range where the background is not strong.
A promising alternative to minimize the background contribution would be to exploit the z sectioning property of confocal detection.This potentially allows unrestricted CL imaging with high acceleration voltages.Characterization of nanostructures in transparent substrates with a confocal CL detection system is work under progress in our research group.
Finally, it is worth mentioning that the approach to lower the acceleration voltage can be implemented in any CL system [5].However, an important increase in the signal intensity is obtained by collecting the light through the glass with a high NA objective lens, as shown in Fig. 1(a).While the NA will influence the signal collection, the resolution in a raster scanned CL image will depend on the primary electron beam properties (which affects the CL excitation probe size) as discussed above and not on the light detection approach.

Conclusions
We demonstrated successful CL imaging of nano-structures on transparent substrates, which are known to generate strong background signals.To this end, we used a highly efficient cathodoluminescence detection scheme where the light is collected through the substrate with a high numerical aperture objective lens.To minimize the background contribution, the acceleration voltage of the primary electron beam was decreased to typical values of 2-5 kV.In this way the electron penetration depth can be restricted to a few nanometers, but the signal intensity is also reduced.For strongly cathodoluminescent materials like nano-phosphors, individual 30nm particles can be easily resolved.On the other hand, for structures with a low CL cross lowering the acceleration voltage needs to be compensated by an increase in the current to maintain a detectable signal level, which for voltages below 5kV may mean that the spatial resolution gets compromised.We have shown that using this approach the weak higher order plasmon resonant modes of a gold nanowire on an ITO-coated glass slide can be detected up to at least the 9th order, highlighting the detection capabilities of our system.

Financial interest statement
A commercial product based on the integrated microscope used in this study is currently marketed by Delmic BV.A.C.Z., P.K., and J.P.H. are shareholder and advisor to Delmic.

Fig. 1 .
Fig. 1.(a) Emission profile of a dipole oriented perpendicular to a vacuum/glass (n = 1.55) interface.The red and yellow lines indicate the maximum collection angle for 1.4NA and 0.95NA, respectively.(b) Side view of our integrated microscope, with an inverted objective lens placed below the sample holder and the electron beam column, in the vacuum chamber of a SEM.
2(b)-2(d) serve to illustrate the alignment with the SEM image.

Fig. 2 .
Fig. 2. (a) CL detection system: an objective lens collects the light and sends it outside the SEM vacuum chamber, where flipping mirrors enable a CCD, a PMT or a spectrometer for detection.Top-left inset: optical paths mounted outside the SEM chamber.(b) SEM image and CL maps obtained by scanning with the electron beam when the optical and electron axes are (c) <10µm shifted and (d) when the axes are aligned within 1 µm.The maximum intensity corresponds to the objective lens center position, i.e. the optical axis, while the center of the image gives the electron axis.

Fig. 3 .
Fig. 3. Nano-phosphors on a glass substrate with an ITO layer: (a) SEM and (b) CL images.The CL intensity is proportional to the particle size and density.The red square highlights a 50nm particle, showed in the inset (horizontal field of view: 200nm).(c) CL map of two 30nm nano-phosphors next to each other.Inset: correlated SEM image (scale bar: 50nm).(d) CL intensity profile along the white dashed line indicated in c.

Fig. 4 .
Fig. 4. (a) SEM and (b) panchromatic CL images of a gold NW obtained with a 2kV electron beam.(c) Spectral distribution of the CL signal acquired by excitation on a spot over one of the edges of the NW.Top insets: FDTD simulation results for the electric field distribution (x component) along a NW for the different resonant mode orders.The emitter can be seen at the right end of the NW.(d) Measured resonance modes compared with the ones calculated using a Fabry-Pérot resonator model for a NW in air.(e) Red curve: FDTD calculation of the emission wavelength distribution for a (x oriented) dipole located at the edge of a 1440nm long, 70nm width gold NW on top of a 70nm thick ITO layer on glass.The black curve shows the spectra obtained in our measurements.

Fig. 5 .
Fig. 5. (a) Interaction volume of the electron beam with the sample: light generated from the substrate below the structure will also contribute to the CL signal, as a background.Low inset: typical spectral distribution of the CL generated in a ITO/glass substrate.(b) Monte-Carlo simulations of the electron interaction volume at 2kV (top) and 30kV (bottom) obtained for a SiO 2 substrate with a 70nm ITO layer.To illustrate the scale factor between them, the area surrounded by the green square in the 2kV image is projected into a yellow square (dot) in the 30kV plot.

Fig. 6 .
Fig. 6.Influence of acceleration voltage on the background signal.(a) SEM and (b) CL images of a 50nm phosphor nanoparticle.(c) Intensity profile obtained along the dashed line shown in b.Similar images obtained at (d-f): 5kV and (g-i): 10kV.