Cathodoluminescence for the 21st century: Learning more from light

On August 22, 1879, in a lecture to the British Association for the Advancement of Science,¹ Sir William Crookes announced that “radiant matter” (his term for the electrons emerging from his soon to be famous Crookes tubes) “exerts Powerful Phosphorogenic Action where it strikes.” He demonstrated for the group what today we would call the cathodoluminescence of calcium sulphide. This material, which was known to luminesce for an extended period following exposure to candlelight, produced even brighter light when exposed to electrons in vacuum (Fig. 1). A few years later, on February 23, 1896, with the 20th century on the horizon, The New York Times ran a feature article² entitled “The True Cathode Rays: Electricity in High Vacuum.” Their report placed Crookes' experiments in context with Röntgen's 1895 discovery of X-rays and hinted at the interplay of light and matter that would dominate the next century of physics.

With the development and subsequent commercialization of the scanning electron microscope in the period from ∼1930 to 1960, cathodoluminescence was positioned to move into the mainstream of science. Through the remainder of the century, it was applied primarily for characterization of the spectroscopy and spatial compositional variations of minerals, semiconductors and other luminescent solids, including biological specimens. A survey of some of the historically most highly cited CL papers, as well as more recent work, shows impacts in systems as varied as the shells of pre-historic brachiopods,³ green fluorescent protein,⁴ ZnO,⁵ protolithic zircon,⁶ diamond,⁷ InAs quantum dots,⁸ individual dopaminergic neurons,⁹ epitaxial GaN,¹⁰ live cells,^11,12 and failure modes in semiconductor lasers.¹³ A number of valuable cathodoluminescence reference books exist, with focus areas in inorganic materials, geo- and planetary sciences, and phosphor applications,^14–18 as well as a number of comprehensive review articles.^19–23 The role of this article is to highlight developments that have emerged after the turn of the 21st century, expanding and reinvigorating the field and its breadth of impact.

At the start of the 20th century, the basic characteristics of light, the “Phosphorgenic Action” that delighted Crookes' audience, were still open to debate. But fairly quickly the combined insights of Planck, Einstein, de Broglie, and Heisenberg would usher in the modern model of the wave/particle duality. Light would be characterized, like other particles, by its energy and intensity, but also its wavelength, momentum, rest mass and charge (or lack thereof), polarization, spin, parity, and quantum (boson) nature. Although early CL focused on measurements of the first of those two properties, the tremendous progress in photonics in recent decades has translated into progress in CL as well. Starting in ∼2005, 100 years after Einstein's famous paper on the photoelectric effect, new developments in approaches and instrumentation for CL have taken advantage of a wider range of photon properties beyond energy and intensity and enabled materials scientists to access and maintain this information—to “learn more from light” emitted under electron beam excitation. This review summarizes those recent developments, provides examples, and suggests a bright future for CL in contributing to materials science across a full range of materials and technologies—semiconductors, phosphors, plasmonic structures, nanostructures, photovoltaics, and biological and geological materials.

CL is light in the UV/VIS/NIR spectral regime emitted from a material under fast free electron irradiation (E_electron = 0.5 keV–300 keV). Although this definition is rather straightforward, there are many distinct physical processes that lead to light generation under an electron beam stimulus, all of which are usually referred to as CL. These processes can be divided into two main classes: incoherent and coherent CL emissions.

The incoherent form of emission [Fig. 2(a)] is spontaneous in nature and as such there is no fixed phase relation with the excitation. The resulting emission is typically incoherent, unpolarized, and isotropic when generated in a homogenous medium. The light emission process involves spectroscopic/quantum mechanical transitions, where the incoming electron beam promotes valence electrons of a material into an excited state which in some cases can radiatively decay with the emission of a photon. In semiconductors and dielectrics, the excited electrons enter the conduction band after which they can decay directly to the valence band (band edge emission) or through intermediate states in the band gap. In quantum confined structures, such as quantum dots, the bands themselves become quantized which can also be visible in the CL emission.

In incoherent CL, a distinction is often made between intrinsic and extrinsic CL. Intrinsic CL corresponds to emission intrinsic to a pure material/crystal such as band edge emission and emission from defects native to that material (e.g., vacancies, interstitials, dislocations) that form intermediate states in the bandgap. Extrinsic CL originates from foreign dopants in a crystal which can also create defect states within the bandgap of the host material and can have a significant CL response even when present in small concentrations. These in-bandgap defects can be complex emitter systems such as nitrogen-vacancy (N-V) centers in diamond or rare-earth ion complexes in Y₃Al₅O₁₂ (YAG), SiO₂, or ZrSiO₄ (Zircon) matrices, for example, leading to rich CL emission spectra with multiple peaks. The interplay between different defects can be a complex affair. Defects can act as CL emission centers, sensitizers (enhancing CL emission of other centers), or quenchers (diminishing emission from CL emission centers).²² 

Incoherent CL emission bears a strong resemblance to photoluminescence where light is used to induce the electronic excitation.²⁴ There are profound differences as well, however. Because electrons have a mass, they carry significantly more momentum than photons and hence they do not obey the same selection rules for excitation. Furthermore, excitation light and lasers, in particular, typically have a narrow energy distribution allowing selective interrogation of specific transitions. The electron, on the other hand, acts as a supercontinuum source over a very large energy range. As such, multiple transitions are excited simultaneously. Furthermore, because of the high energy of the electrons, multiple excitation paths are available involving multistep relaxations via wide bandgaps (>6 eV), bulk plasmons, or even core transitions.^25,26 Which excitation mechanism is dominant can be difficult to say based solely on the CL signal, although other theoretical and experimental methods can provide insight. For example, how efficiently the electron transfers energy to specific channels can be probed (partly) with electron energy loss spectroscopy (EELS).²⁵ 

The second class of CL emission is referred to as coherent CL. This categorization comes from the fact that in these processes light emission has a fixed phase relation with the incoming electron and is always strongly polarized.²⁷ These processes can be understood best from an electrodynamic perspective where the electron is a moving point charge which is accompanied by evanescent electromagnetic fields. In particular, the electron has a vertical and radial electric field and an azimuthal magnetic field consistent with this charge distribution. When the electron experiences a change in the dielectric environment close to its trajectory, e.g., a surface, these fields can polarize the material resulting in subsequent light emission. This form of radiation is known as transition radiation (TR). Besides coupling directly to radiation, other electromagnetic modes such as localized optical resonances, surface plasmon polaritons (SPPs), and waveguide modes can also be excited. It has been demonstrated that there exists at least a qualitative relation between the emission probability that is measured in CL and the radiative local density of optical states.²⁷ This makes this form of CL radiation highly appealing for studying optical phenomena on small length scales.

Other coherent excitation geometries also exist. For example, when an electron travels faster than the phase velocity of light, a coherent shockwave can be formed known as Cherenkov radiation. Special forms of this radiation exist. For example, when an electron is moving parallel to a waveguide, its velocity can be matched to the phase velocity of light in the waveguide leading to a Cherenkov-radiation like excitation.^28,29 This particular geometry provides control over the excitation probability and emission wavelength. A similar effect can be achieved in periodically structured surfaces such as gratings^30,31 (Smith-Purcell effect) or in photonic crystals/metamaterials.^32,33 For the latter, the optical dispersion can be tuned to exhibit a negative phase velocity in which case the Cherenkov emission cone can be inverted. These designs can potentially be applied to create coherent (single-photon) light sources with a small footprint.^17,34

As coherent CL emission always occurs when there is a difference in refractive index with vacuum, it is present for every material. However, the probability of generating photons through these processes is generally rather small (∼10⁻⁴ photons per incoming electron).²⁷ In particular cases, the emission probability can be significantly higher and approach or even exceed unity, by optimizing the electron velocity and/or the excitation geometry.²⁸ However, normally the coherent CL signal is obscured in materials such as direct bandgap semiconductors, phosphors, ceramics, and dielectrics as the incoherent CL emission is orders of magnitude stronger in these cases. In metals and to a lesser degree indirect band gap materials such as silicon, the incoherent form of CL is strongly damped through non-radiative processes and the coherent CL makes up a significant fraction of CL response.³⁵ The low emission probabilities make it more challenging to detect experimentally and typically currents on the order of 0.2–10 nA are required to achieve enough signal. Nevertheless, with recent advances in light collection efficiency and detector sensitivities, a wide range of experiments have been done utilizing both types of emitted radiation.

Cathodoluminescence as a technique has evolved significantly over the last decades, profiting greatly from new innovations and technological advances in a variety of fields. First of all, electron microscopes have improved significantly. There is more flexibility in electron detection, the imaging and beam current are more stable, and the spatial resolution is higher. The performance of SEMs has improved, particularly at lower voltages, which translates directly into higher spatial resolution in incoherent CL experiments.

For the light collection inside the electron microscope vacuum chamber, various techniques exist including reflective optics (parabolic mirrors and elliptical mirrors), refractive optics (microscope objectives) fibers/near-field probes, and more simple metallic light pipes. The reflective optics approach is most common because it is relatively simple to integrate in the electron microscope and provides good performance because of the high numerical aperture (NA can be as high as 0.95).³⁶ They are usually mounted above the sample and have a hole in them through which the electron beam can access the sample. As a result, this approach is compatible with thick/opaque samples. If the collection optic is made from aluminum, it is compatible with a large wavelength range from the deep-UV to the near-infrared. Advancements have been made in the alignment of these mirror optics by including (motorized) micro-positioning in the design which greatly boosts the detection efficiency. Various mirror designs exist such as thin mirrors compatible with short working distances and immersion lens electron imaging and mirrors adapted for parallel backscatter electron (BSE) and/or energy dispersive X-ray spectroscopy (EDS) acquisitions or probe access for electron beam induced current (EBIC).

In terms of light detection, there have been significant advances as well. For CL spectroscopy, Czerny-Turner spectrographs are normally used which are robust, simple to use, and modular. By choosing the appropriate diffraction grating, the spectrograph can be optimized in terms of spectral range and spectral resolution. By rotating the grating, the CL can be detected wavelength by wavelength using a single pixel detector such as a photomultiplier tube (PMT). This is a serialized approach, however, making it relatively slow for high-resolution spectroscopy. Single-pixel detectors such as PMTs are also used abundantly for fast intensity mapping where dwell times as short as 1 μs can be reached. This enables video-rate CL imaging over large areas. CL intensity mapping often can be panchromatic (integrated over all wavelengths) or wavelength-filtered using dichroic optics or band pass filters. Time-resolved detection requires ultrafast detectors (see Sec. II B).

In hyperspectral CL imaging, the grating in the spectrograph projects a large range of wavelengths on a pixelated detector allowing parallel spectral detection. Because of the advances in silicon based charge coupled device (CCD) and complementary metal–oxide–semiconductor (CMOS) array imaging technology, many CL systems can now perform this parallel hyperspectral imaging in a fast and sensitive manner. State of the art cameras have quantum efficiencies of 70%-95%, a sensitivity of ∼1 digital count per incoming photon, and can acquire more than 1000 spectra per second. The cameras can be optimized for specific wavelength regimes covering all the way from the deep UV (200 nm) to the NIR (1100 nm). The wavelength range can be further extended into the infrared (up to 2.2 μm) by using InGaAs array technology, although the sensitivity and noise characteristics are not as good as the silicon based detectors for now. Efforts have been made to go beyond 200 nm moving into the XUV/soft x-ray regime which connects the CL energy range with the hard X-ray energy regime.³⁷ Similarly, we can envision extending the spectral range more towards the MID-IR (>3 μm) where one potentially can unravel phonon physics at small length scales, although this would require major modifications to detectors and optics. Figure 3 shows the typical wavelength/energy range covered in CL and which detectors can be used for spectroscopy. Some typical applications for given wavelength regimes are shown underneath together with several characteristic CL spectra for a variety of dielectric and semiconductor materials.

In a typical Czerny-Turner spectrograph used for CL, spectral resolutions of <0.1 nm per pixel (∼0.5 meV in the visible wavelength regime) can be attained by using gratings with a large number of lines per mm (>1200 lines/mm), although for many applications, a resolution of 1 nm is enough. At room temperature, the typical thermal energy is on the order of kT = 25 meV. Thermal broadening in solid state systems is more complex than in atomic or gas-phase systems and is often mediated by phonon interaction. Nevertheless, thermal broadening generally has a significant effect on the line width of spectroscopic transitions (on the order of kT) which means that the temperature is in fact limiting the spectral resolving power. As such, it is very useful to combine CL spectroscopy with a (cryogenic) cooling stage inside of the electron microscope. This can be a simple Peltier-cooled system (T_min = 220 K and kT = 19 meV), nitrogen cooled stage (T_min = 77 K and kT = 6.6 meV), or even a helium cooled stage (4.2 K and kT = 0.36 meV). These can be integrated at increasing cost and system complexity. Sample cooling has the additional advantage that for many materials the CL emission becomes significantly brighter because thermal quenching is reduced. One drawback of cooling stages is additional vibrations and drift that are potentially introduced into the system. The stage assembly and sample contract when cooled, potentially leading to extra drift. Furthermore, in the case of nitrogen and helium, there are gases flowing close to the sample and any turbulence in the flow may induce vibrations. The first issue can be mitigated by letting the sample stabilize for a long enough time such that it is in thermal equilibrium; the second can be resolved by creating a lamellar flow of cooling liquid/gas.

When performing CL scanning, the field of view (FOV) that can be addressed by scanning the electron beam is always limited. For spectroscopy measurements, it is normally limited by the input slit/fiber input which rejects light that is out of focus. The typical FOV for spectroscopy ranges from 10 × 10 to 50 × 50 μm depending on the exact realization of the optics. For CL-intensity mapping, the FOV can be significantly larger (1 × 1 mm). To attain larger FOV, the stage of the electron microscope can be scanned, an approach which is also used in simpler optical CL systems in which the electron beam is static.³⁸ The precision of electron microscope stages normally is not as good as the electron beam scanning so, depending on the desired resolution of the imaging, some overlap has to be taken such that good stitching can be performed. This is particularly important for high resolution imaging over large areas which paves the way for more automated large scale imaging.

Although panchromatic and hyperspectral CL imaging are highly useful, there are more degrees of freedom in the cathodoluminescence emission that can be exploited. In particular, the momentum distribution, i.e., the direction in which light is emitted from a material, can also be used to obtain valuable information on the optical properties of nanostructures and the electron-matter interaction. Applications include local probing of the band structure of (quasi)periodic systems,^39–42 multipolar decomposition of nanoparticle scattering,^43–47 directionality,^48–50 plasmon outcoupling,^51–53 and separation of coherent and incoherent CL processes in bulk materials.³⁵ In conventional CL spectroscopy and panchromatic imaging, the CL emission is usually focused onto a spectrograph slit or single-pixel detector and the momentum information which is contained in the CL beam coming from the collection optic is lost. There are various approaches to gain access to this momentum information and perform angle-resolved CL imaging.

The first method, proposed and implemented by Yamamoto and colleagues,^39,40 makes use of a pinhole that can be scanned transversely through the beam coming from a paraboloid collection optic in the chamber. This pinhole filters momentum space, i.e., it selects a particular emission angle where the angular width is determined by the combination of the pinhole size and the paraboloid collection optic. After the pinhole, the light is send to a spectrometer for spectral analysis. The advantage of this technique is that the angular response can be determined with high spectral resolution. A disadvantage is that the acquisition of a full angular profile can take a long time because the spectral acquisition has to be repeated for every pinhole position. One can envision though that this method can also be performed in a parallel fashion by projecting the Fourier image onto the slit of a spectrograph with a 2D CCD or CMOS camera as is commonly done in optical studies.⁵⁴ The pinhole-scanning technique has been applied to study diffraction and dispersion in periodic structures and in the studies of nanoparticle scattering.^39,40

A second complementary approach was developed by Polman and coworkers.^48,55 In this case, the CL emission coming from the paraboloid shown in Fig. 4(a) is directly projected onto a 2D imaging array giving access to a large part of the momentum distribution in the upper hemisphere in a single camera acquisition, making it an ideal method for fast angle-resolved imaging. This is illustrated in Fig. 4(b).

In both approaches, the CL beam coming from the paraboloid is imaged but such a measurement does not directly represent the angular profile in the desired form yet. The angular profile is best expressed by displaying the emission intensity per steradian as a function of spherical coordinates θ (zenith) and ϕ (azimuth). To that end, every transverse position in the CL beam has to be linked to a combination of θ and ϕ. Furthermore, due to the curvature of a paraboloid, the solid angle projection per camera pixel varies depending on the position on the mirror, leading to a non-uniform pattern where the apex of the mirror is much brighter. Both corrections can be done based on the precisely known paraboloid geometry. Figures 4(c) and 4(d) show an example of a conversion from raw data to an angular profile.

Polarization is another degree of freedom in light and is a key in many if not all light/matter interactions. CL polarization imaging can be a highly valuable source of information. In particular, one can perform a more rigorous multipolar decomposition in nanoscatterers, distinguish optical transverse magnetic (TM) and transverse electric (TE) modes, study chirality and more. Furthermore, it provides a direct way to separate polarized and unpolarized components and can be used, for example, to enhance contrast between a polarized emitter and unpolarized background from a substrate.

There are various approaches to perform polarization filtered CL imaging. In one approach, a linear polarizer (LP) is introduced into the beam path after which the light can be focused into a spectrograph for hyperspectral imaging. For proper polarization contrast, it is important in this case to also filter momentum space because the paraboloid and elliptical mirrors that are used for CL collection perturb the emission polarization except for the central part where the polarization is preserved. Using a slit or pinhole, this part of the mirror can be selected.^56,57 This method has been applied to determine emission polarization in semiconductor structures^58,59 and to distinguish different resonant modes in plasmonic systems.⁶⁰ Furthermore, one can break the symmetry in azimuthally symmetric structures, visualizing the dipole orientation of a resonant mode for a specific excitation position.⁵⁶ 

Recently, polarization studies also have been performed in momentum space by combining single-shot angle-resolved imaging as described in Fig. 4 with a rotating plate polarimeter, consisting of a quarter wave plate (QWP) and linear polarizer (LP).⁶¹ By performing six independent measurements for different polarimeter configurations (different angles for QWP and LP), the Stokes vector, describing the full polarization state of the CL emission, can be retrieved at the detector. This approach is well-known in optics and is employed in nanophotonics, astrophysics, quantum optics, and ellipsometry. The setup for this technique is shown in Fig. 5(a).

In this case, the paraboloid used for collection converts the light originating from the localized CL source into a parallel beam. In this process, the original emission polarization is converted into polarization components transverse to the parallel beam. Furthermore, as the mirror is not a perfect metal, it can polarize unpolarized emission and change the ellipticity of the emission for certain angles. To correct for these perturbing effects, the Mueller formalism can be used in which the response of the mirror is cast into a matrix form that operates on the measured Stokes vector in the detector plane. This matrix is angle dependent and thus needs to be calculated for every emission angle. When this correction is in place, the full polarization state can be retrieved for every angle that falls within the collecting mirror optic.

This technique has been applied to understand the optical properties of plasmonics bull's eye gratings [Fig. 5(b)] and semiconductor nanowire structures.^62,63 The polarization filtering can also be combined with the hyperspectral imaging which has been employed to study photonic crystal modes⁶⁴ and chiral near-fields,⁶⁵ although care has to be taken in this case as the retrieved polarization is averaged over a selection of emission angles. Through interferometric measurements or wavefront sensing, it is potentially possible to access the relative phase between the different angles. This would give access to the orbital angular momentum of light allowing studies of vortex beam generation by nanostructures, for example. Furthermore, as the full far-field information (full amplitude and phase) could be measured, a direct connection could be made to the near-field and corresponding charge distribution allowing an absolute multipole decomposition.

The time dynamics and statistics of the light emission process hold valuable information on the quality, photonic environment, and quantum mechanical properties of a (nano)material. As such, it is also of great interest for CL microscopy where such dynamics can be studied on small length scales. There are various experimental approaches to access this information. To measure the emission statistics, a Hanbury Brown and Twiss (HBT) interferometer setup can be used. This provides access to the g⁽²⁾ correlation function which is key in describing the emission statistics and can be used to measure photon (anti)bunching.^66–70 Furthermore, the lifetime of emission, i.e., the time that it takes for the emission intensity to decay to 1/e times the initial intensity, can be extracted from the measured lineshapes in the g⁽²⁾ function.⁶⁸ How fast the emission decays depends on the material type, quality, and the photonic environment as described by the local density of optical states. The abovementioned g⁽²⁾ approach has the advantage that it can be performed in a regular electron microscope with a continuous beam preserving spatial resolution and ease of use. A drawback is that g⁽²⁾ acquisitions can be time-consuming and as a result the required local electron-dose can exceed the damage threshold of the material under investigation. A more generic approach to determine the lifetime of emission is to have a fast excitation at a well-defined time. This requires an ultrafast pulsed electron microscope. Several approaches exist to create an ultrafast electron source which can coarsely be divided into two categories: electrostatic blanking and photoelectron excitation with a pulsed laser system.

In electrostatic blanking, a fast electric field is applied that deflects the electron beam away from its normal trajectory such that it is blanked on an aperture or sidewall of the electron microscope column. Usually, this is achieved with a “fast” beam blanker in which two capacitor plates are used to apply the field.^71,72 The repetition rate and pulse duration can be controlled with the driving electronics that charge the capacitors. By adjusting the electron lenses in the microscope and the size of the apertures in the column, the time resolution can be greatly improved. Recently, a pulse duration as short as 90 ps has been achieved in CL experiments in this manner, while preserving a λ/10 spatial resolution by performing conjugate blanking.⁷³ Another approach to blank the beam is using a RF/microwave cavity in which the time-varying electromagnetic fields of the cavity mode are used to create a pulsed source, although this has not been combined with CL imaging as of yet.⁷⁴ Similarly, THz pulses have been employed to compress electron pulses using a THz resonator.⁷⁵ 

A better time resolution can be achieved by irradiating the electron gun with an ultrafast laser pulse. In this case, the gun is kept at a condition (lower temperature, reduced potential) where it is on the edge of emitting electrons and the laser pulse is used to overcome the final barrier. The photoelectron generation process is complex and different physical effects can play a role depending on the laser power, wavelength, and tip conditions.^76,77 This approach has already been employed in ultrafast electron microscopy, 4D photon-induced near-field electron microscopy (PINEM) experiments,^78,79 and quantum coherent phase modulation of an electron beam,⁸⁰ as well as in CL microscopy.^81–85 In addition to an ultrafast electron source, ultrafast light detectors are also required both for the g⁽²⁾ measurements and lifetime imaging. These can be single pixel detectors such as avalanche photodiodes and photomultipler tubes, or gated 1D/2D detectors such as intensified/streak cameras which can be used for hyperspectral time-resolved imaging.

The above-mentioned time-resolved approaches have been successfully applied to a variety of compound semiconductor,^{68,69,81–85} and rare-earth doped materials⁸⁶ amongst others. For example, the effects of nanostructuring,⁸¹ strain,⁸² and defects⁸³ on the carrier dynamics and recombination were investigated using time-resolved CL imaging.

The power of conventional CL has traditionally been associated with the highly localized nature of the excitation as well as to the ability of an electron beam to generate luminescence in a wide variety of materials. The challenges of electron beam excitation (e.g., the need for high vacuum, potential material degradation, and local charging) are counteracted by CL's ability to simultaneously image the particular material or device at sub-micron or even nanometer scale and provide a highly localized generation volume for the luminescence, depending on the density and atomic structure of the material and the energy of the incident beam. It is only in recent years that optical microscopy techniques have begun to push beyond far-field resolution limits, but they are still unable to compete with electron beam imaging for the very highest spatial resolution of the excitation source.

One thing that conventional CL has traditionally shared with optical luminescence techniques has been the use of far-field or wide area collection optics, meaning that one collects optical emission from a much larger area than the point of generation. The relevant dimensions are illustrated in Fig. 6, for the case of a ∼20 keV excitation in GaAs. In most CL systems that use a collecting mirror, the collection area for emitted light would have dimensions on the order of ∼30–50 μm, compared with a generation volume in a bulk sample that is order of magnitude 1 μm or less. This means that even if carriers were to diffuse from the point of generation with a diffusion length of ∼10 μm, the resulting signal from their recombination would still be within the collecting FOV. In standard CL, the intensity of the sum of the light from that full collection area is associated to the point of generation. While this has the benefits of maximizing the intensity of the CL signal, it also effectively discards critical information from the actual spatial distribution of the emitted light, information that can be used to directly study energy transport within the material.

In “transport imaging,” by contrast, the emitted CL is imaged in a way that maintains the spatial distribution of the emitted light. Starting in 2003, Haegel and colleagues demonstrated transport imaging for CL, first using far-field imaging and then near-field imaging inside the SEM to maintain the spatial information associated with carrier recombination from either a point source or a line source of excitation. The work as of 2013 has been reviewed.⁸⁷ Here, we provide an overview of the approach and highlight more recent results.

The earliest “dual probe” CL measurements in the SEM, using the electron beam as a generation source and an independent second scanning probe to enable more localized collection, were performed with the goal of increasing CL resolution in bulk materials. This approach, however, has not become widely applied because progress in STEM-CL, as described in Sec. II F, enabled that more direct technique to reach new limits of resolution for spatial CL probing. Dual probe measurements today focus instead on probing energy transport—using spatially resolved optical collection or high resolution near field detection and scanning to maintain the spatial distribution of the actual luminescence from a fixed generation source.

To obtain this information, the emitted light is collected either through an internal microscope that reimages the spatial distribution of the luminescence on a CCD camera or through scanning an independent near field probe and measuring the intensity at the end of a fiber optic with a PMT or photodiode [Fig. 7(a)]. Fitting the resultant luminescence distribution (intensity as a function of distance) enables a direct measurement of the carrier diffusion length (Ld) [Fig. 7(b)]. When Ld is spatially uniform (e.g., a uniform thin film in the low excitation limit, Δn _majority ≪ n _majority, where n is the carrier concentration), or is anisotropic over a large distance [e.g., due to ordering behavior, Fig. 7(c)], the diffusion length can be obtained quite accurately, fitting intensity variations over several orders of magnitude.⁸⁸ For cases that involve localized spatial variation in Ld (whether due to spatial materials variations or a high excitation condition that makes recombination statistics a function of distance), the data fitting can be more complex. The power of the experimental approach is that one obtains a direct image of spatially resolved carrier transport, whatever combination of factors (material properties, excitation level, localized defects, and surface recombination) determines the effective diffusion length in a given direction.

Integrating an optical microscope inside the SEM is a straightforward means of transport imaging in cases where thin film or bulk properties are of interest and the carrier diffusion lengths are in excess of several microns, allowing for transport distances that exceed the optical imaging resolution limits. Using near-field collection requires integrating a combined atomic force microscope/near-field scanning optical microscope (AFM/NSOM) inside the SEM chamber. Figure 7(d) illustrates the near-field approach for imaging carrier diffusion inside a core/shell GaN nanowire.⁸⁹ Note that one also observes waveguided light emerging from the end of the nanowire structure, indicating that it is possible to map energy transport in the form of both electronic carriers and photons.

In order to study carrier diffusion from a fixed generation point, the AFM/NSOM must be able to scan the collecting probe, while the sample remains fixed. This is different than the standard operation of many AFMs, in which the sample is scanned relative to a fixed near field probe. For the work shown here, the particular AFM/NSOM system was a Nanonics Multiview 2000, which allows for independent scanning of tip or sample, as well as direct e-beam access to the sample surface. Particular attention must be paid to grounding both the sample and the NSOM tip to avoid charging and relative motion of the excitation on the sample during the near-field scan.

Figure 8 shows three different near-field examples of imaging energy transport within semiconductor structures. In Fig. 8(a), transport imaging is used to observe carrier diffusion in a multijunction solar cell in cross section, showing diffusion throughout the entire thickness of the intermediate GaInAs layer (middle cell) in a triple junction device.⁹⁰ The high quality of the material enables diffusion of carriers created at a single point (green dot) to diffuse all the way to the collecting junction. In Fig. 8(b), dual probe imaging has been used to map optical transport, using the NSOM probe to show the waveguided light emission from the end of a ZnO nanowire in response to excitation at a point 20 μm further down the wire.⁹¹ Finally, Fig. 8(c) illustrates the ability to use a line scan excitation source (indicated by the green line) to image the effect on local transport of a localized defect (in this case, an etch pit structure in a GaAs thin film).⁹² The contour plot of the recombination luminescence in the neighborhood of the steady state line source is a direct map of the diffusion paths that carriers follow, illustrating the “sink” effect of the defective region on carrier transport.

Quantitative results for diffusion behavior can be obtained by fitting the distribution profiles with the proper form of the diffusion equation for the geometry of interest. For 1D diffusion (e.g., carrier diffusion down a nanowire or net diffusion perpendicular to a line source in a thin film), an exponential decay, e^(-x/L_d⁾, is the correct form. However, for a point source in a thin film (where L_d is greater than the film thickness), the diffusion profile is described by a second order Bessel function of the zeroth order $Ko( r L d$⁠). When the carriers can diffuse in 3 dimensions (e.g., bulk samples), the luminescence emitted from various depths within the sample must be integrated. The appropriate form of the diffusion equation depends on the combined geometry of the source and the sample. Unfortunately, there are multiple examples in the literature of incorrectly using an exponential decay in situations where it does not apply.

The ability to image diffusion from a fixed point or line source offers a unique way to both visualize and understand transport in complex systems. Transport imaging experiments make clear the futility of attempting to label anisotropic materials [as shown in Fig. 7(c)] or materials with spatial variations due to defects or dislocations [Fig. 8(c)] with a single diffusion length. The future impact of TI is both in the unique experiments it enables by separating highly localized generation and detection and also in developing the analysis tools to ultimately map and image diffusion lengths in a way that reflects their vector nature in non-homogeneous materials.

With the increase in correlative and multimodal approaches (Sec. III), CL in all its forms is increasingly combined with techniques such as EBIC (for the measurement of charge collection), electron backscatter diffraction (EBSD, for the determination of local crystalline structure), scanning Kelvin probe microscopy (SKPM, for the measurement of local potential fluctuations), in addition to the broad toolbox of energy dispersive spectroscopy, and atomic force, optical, and scanning transmission microscopies. While such intensive studies can reveal critical complementary information and correlations, they also raise the question of the representative nature of any one such highly detailed but also highly localized observation. A good example is the combined use of CL spectroscopy, EBIC, and EBSD to understand optical, electrical, and structural correlations in thin film solar cell materials, where charge recombination at grain boundaries remains one of the main limiting factors in achieving the fundamental physical limits for energy conversion.

Advances in imaging and analysis software are enabling analysis of correlation between, for example, variation in CL intensity across a grain boundary with the structural nature of that boundary obtained from CL and EBSD measurements in the same region. In a range of recent work that attempts to address the variability issue, the number of grain boundary correlations spans the range from 100 to 600.^93–96 In work comparing CL intensities at grain interiors and grain boundaries in polycrystalline CdTe material, Moseley et al. were able to reduce uncertainty in average intensity measurements to less than 10%, by measuring between 97 and 480 examples for six different locations or interfaces.⁹⁷ 

An example is shown in Fig. 9, providing both EBSD and CL analysis of a region in as-deposited CdTe. Grain boundary disorientation is identified through EBSD and then those individual boundaries are linked to their CL intensity. For this particular material, an increase in CL intensity was found with increasing grain boundary disorientation angle, possibly associated with increasing passivation in the more open structure. Correlative techniques applying these statistically robust analyses will potentially play a large role in answering foundational questions relating structure to performance at critical interfaces and boundary structures in a wide range of materials for electronic, energy and photonic applications.

To first order, CL microscopy is usually considered a 2D technique. However, there are multiple ways to extract 3D information. First of all, the degree of penetration for the electron for different electron energies can be used to gain depth information. The degree of penetration can be simulated with Monte Carlo programs such as CASINO if the material geometry and density are known.⁹⁸ This can be particularly useful in stratified devices where certain layers can be visible or not depending on whether the energy of the electron is high enough as is schematically illustrated in Fig. 10(a). This method generally provides useful qualitative insights⁹⁹ and can easily be performed in most electron microscopes as no additional hardware is required. This acceleration voltage scanning approach is also used in scanning electron microscopy imaging to reconstruct the 3D geometry in organic and inorganic materials,^100,101 where a numerical algorithm is used to interpret the data. For CL, a rigorous reconstruction would be a complex affair as carrier diffusion, light reabsorption by the overlying material, and light outcoupling can affect the collected CL signal. Nevertheless, it could be an interesting avenue to explore.

A second approach involves slicing of samples with an ion beam or microtome.^102,103 The microtome is mostly useful for soft matter (biological tissue), whereas the ion beam can also be used for harder inorganic materials. These can used to study a material in cross section as indicated in Fig. 10(b). Both techniques can be employed in-situ in an SEM which is preferable for 3D imaging, although the microtome systems are usually quite bulky and have not been combined with CL to our knowledge. In a dual beam system, sequential milling and CL imaging steps can be performed as has been shown on biological¹⁰⁴ and geological materials.^{103,105–107} This method could be (fully) automated, in principle. Special care has to be taken in this case, to characterize the effect of the ion-beam milling process on the material under investigation as the ion beam can alter or damage material during milling. This potentially leads to artifacts and subsequent misinterpretation of results.

The third approach to obtain 3D information is tomography, a technique well-known from X-ray imaging, seismic imaging, and TEM amongst others.¹⁰⁸ In tomography, a structure is typically imaged under different orientations giving a tilt series dataset from which the 3D structure can be reconstructed using numerical algorithms [see Fig. 10(c)]. Although this has been used in EELS experiments to obtain the 3D information on optical modes in several plasmonic geometries,^109,110 in CL, this has only been done in a simpler manner by imaging multiple particles with a similar geometry under different orientations.¹¹¹ By including a tilting stage, the tomography can be performed on a single structure in principle. Both for CL and EELS, the tomographic reconstruction methods and which physical quantities can be/are reconstructed are still topics of experimental and theoretical discussion¹¹² but it is clear that the techniques yield exciting insights that go beyond the conventional 2D experiments.

Although the majority of CL studies are performed in SEMs, some are also done in transmission electron microscopes (TEMs).^{39,40,113–115} When performing CL, STEM mode is used where the electron beam is focused into a nanoscale probe. Performing CL in a TEM has a number of advantages and disadvantages as compared with SEM CL. When performing experiments in the TEM, one can benefit from the small probe size/interaction volume. In modern STEM systems equipped with a field emission gun (FEG) and a spherical aberration corrector, a probe size of ≤ 1 nm can be obtained at significant beam currents of ∼1 nA.¹¹⁶ Additionally, CL can be combined with the other imaging modalities that are available. For example, high-angle annular dark field (HAADF) and EELS data can be collected simultaneously. While the HAADF data can be used to see the geometrical features and material density variations, the EELS can be used to study the composition in the core energy loss regime and optical properties in the low loss regime.²⁷ As the HAADF and EELS signals are collected underneath the sample, they are normally not affected by the CL collection if that is done above the sample. The combination of CL and EELS is particularly powerful in plasmonics as CL probes the radiative part of the plasmon excitation, whereas the EELS probes the full extinction including both radiation and absorption, providing detailed insight into the optical properties of plasmonic structures.^117–120 Besides these techniques, one also has access to the full arsenal of TEM imaging techniques such as bright-field TEM, electron-diffraction and more, allowing in-depth correlative studies where the internal material structure as visualized by (S)TEM is correlated to the observed CL signal.

To give an example, STEM/CL has been intensively utilized for the characterization of semiconductor quantum structures. Here, the charge carrier generation and resulting CL emission are confined to a nano-scale volume due to thin geometry of the materials and the high primary electron energy used. Recently, it was reported that quantum disks aligned in a semiconductor nanowire were imaged with a 4 nm spatial resolution using STEM CL.¹²¹ 

There are also some disadvantages to STEM CL compared with the more common SEM CL. TEMs are significantly more expensive and complex to operate. Additionally, the space in a TEM is limited and as such the collection optics in the vacuum chamber needs to be much smaller. Typically, this is either a retractable mirror system or a system that is integrated in a TEM holder. In both cases, the light is usually coupled in an optical fiber (bundle) which provides an efficient route for light outcoupling. Nevertheless, it is difficult to reach the collection efficiencies available in SEM. Furthermore, the samples have to be electron transparent (typical thickness < 100 nm) which requires extensive sample preparation in some cases, e.g., preparation of a lamella with FIB and transfer to a TEM membrane. The high acceleration voltages used in TEM make coherent excitations more efficient (plasmons, TR, etc.). However, at TEM-compatible electron energies, it is hard to have any depth resolution or surface sensitivity when looking at incoherent CL emission as is possible in SEM (see Sec. II E). Furthermore, the imaging can be more damaging compared with SEM as the highly focused nature of the beam together with the elevated energies can cause severe knock-on damage. Nevertheless, STEM CL is gaining significant interest as it brings the CL technique to the true nanoscale.

Modern scanning electron microscopes often have large vacuum chambers and many vacuum ports to accommodate a large variety of peripheral equipment and detectors. This allows the performance of hybrid correlative experiments in which a variety of complementary techniques including CL are used. Besides CL, SEMs can be used for secondary electron imaging (surface topography), BSE (density contrast), EDS/WDS (elemental analysis), nanoSIMS (elemental analysis), EBIC (electrical properties), and EBSD (crystal structure). Some work has also been done to combine CL with electron beam lithography (EBL) where the CL is used to localize specific features in a sample such as buried epitaxial quantum dots, after which lithography is performed to define a cavity or solid immersion lens in the right location.^122–124 Although not all techniques can be performed in parallel because some of the signals can be blocked by the collection optic, they can still be performed in sequence as most CL systems have a retractable optics system. This means that many of these techniques, summarized in Table I, can be used in-situ in a single experiment allowing in-depth multimodal analysis of a material or device. If a separate detection path or system is required, then the use of markers, produced by FIB or other approaches, is a standard means of identifying a common region.

As an example of a multimodal approach, several groups have integrated high numerical aperture microscopes into SEMs.^125,126 By placing an optical microscope below the sample, with the same optical axis and focal plane, the system allows for correlative light and electron microscopy (CLEM), which is of interest in life sciences as well as physical sciences. While CLEM has a long history, recent advances to integrate the capabilities have overcome difficulties associated with sample transfer or locations of specific regions. Work has been performed using these integrated systems on both organic cellular systems,¹²⁷ where correlating the optical fluorescence and physical structure is of interest, and hybrid perovskites,¹²⁸ where the electron beam was used to excite the material for in-situ degradation studies. In addition, inverted optical collection underneath a sample mounted on a glass substrate can enable greater optical collection, where a significant fraction of the cathodoluminescence can be coupled into the substrate. This is particularly true for very small particles. Background luminescence from the glass slide can be limited by using low energy (<5 kV) excitation. Recent examples have included nanophosphors and plasmonics nanowires. Depending on the carrier diffusion or drift lengths involved, this type of integrated setup could also be used for the far-field transport imaging described in Sec. II C, if the electron beam excitation was fixed as a point or line source and the optical microscope was used to image the resulting luminescence pattern.

Some commercial entities (e.g., TESCAN) have also begun exploring the integration of AFM capability into SEMs, to allow for nanometric height profiling to measure FIB modifications in-situ. In 2016, Bercu et al. published a description of a scanning shear-force near-field system that performs simultaneous topography and near-field CL, producing very high resolution images with the e-beam/near-field probe distance fixed for a given scan.¹²⁹ If this integrated capability develops, a variety of near-field measurements, including NSOM, but also thermal, magnetic, or electrical probes, could be more easily integrated with e-beam systems.

The light that emerges from a material in response to electron beam excitation carries with it multiple layers of important information—information about the energy states of electrons and holes, recombination dynamics, plasmon behavior, carrier diffusion and other forms of energy transport, localized materials variations, and signatures of local carrier confinement. Thanks to significant advances in electron microscopy, photon detectors, and imaging capabilities, CL can now take full advantage of all these layers. In addition to expansion of the wavelength and time domains, variations of CL now give access to the angular distribution, polarization, spatial distribution, and correlation to structural and electronic properties of the emitted light. As the wavelength coverage of CL expands into the infrared and toward the soft X-ray limit, both the range of materials and the range of physical phenomena that can be studied will continue to increase.

The number of companies providing CL systems for electron microscopes has also increased. Today, these include Attolight, Horiba Scientific, Delmic, Gatan, and JEOL (for microprobe systems). These companies are increasingly integrating the advances surveyed here to make these new tools more broadly and easily accessible to the research community.

The interplay of light and matter—the incoming electron coupling to the outgoing photon—is at the heart of CL and was the story that Crookes brought to his rapt audience 138 years ago. Today, it provides a view into the most fundamental processes within the material or device of interest. CL has a unique ability to provide this insight for a wide range of materials, due to its combined high spatial resolution and energy excitation range, while utilizing an excitation source that is accessible at the individual laboratory scale. The advances we have seen in the last decade are now allowing the community to “do more with light”—a trajectory one can expect to continue in the decades ahead.

We acknowledge Sophie Meuret for helpful discussions on time resolved spectroscopy and review of the manuscript, John Moseley and Mowafak Al-Jassim for helpful discussions on statistical and correlative analysis, and Chuanxiao Xiao for a careful reading of the manuscript. Lindsay Hornbecker at NREL contributed to the manuscript preparation and Al Hicks provided the drawing for Fig. 5 and the artist's rendition of Fig. 1. Some parts of this work by N.M.H. were supported by the U.S. Department of Energy under Contract No. DE-AC36-08GO28308 with the National Renewable Energy Laboratory.

T.C. is employee of Delmic BV, a company that develops and sells cathodoluminescence systems.