By its very nature, an electron beam is a broadband (white) excitation source. Therefore, the resultant cathodoluminescence (CL) emission can be made up of light with various wavelengths (colors). The dominance of wavelengths relies on the configuration, geometry, and composition of the local material.
In panchromatic imaging, the total intensity of the combination of all the wavelengths is measured. However, a lot of information is lost in this way since the wavelength distribution (spectrum) usually includes valuable information on the local optical and structural properties of the material.
Color filters can be used to obtain wavelength information; however, this can be tiresome because a scan has to be carried out for each wavelength. In hyperspectral imaging, a complete spectrum is obtained in a parallel manner, offering a high-resolution spectrum for every electron beam position. The CL emission is directed toward a Czerny-Turner spectrograph, which includes a pixelated CMOS, CCD, or photodiode array and a diffraction grating.
The diffraction grating spatially disperses the different emission wavelengths over the camera in a manner that each line of pixels corresponds to a particular wavelength. This is shown in Figure 1. The coupling into the spectrograph can be through an optical fiber (not shown) or through free space (shown here). Efficient hyperspectral imaging needs a perfect parallel beam from the probation, which can be achieved only if the mirror is properly aligned. The SPARC system is extremely good at this because it has a sophisticated micro-positioning system.
Figure 1. Schematic representation of the hyperspectral imaging mode
Hyperspectral imaging generates a 3D datacube in which the first two dimensions correspond to the spatial electron beam position (x,y) and the third corresponds to the wavelength. This datacube is similar to what is obtained in WDS or EDS except for UV/VIS/IR wavelengths rather than X-ray wavelengths. The CL datacube includes plenty of information that can be visualized in several different ways.
For instance, the spectrum corresponding to a particular excitation position can be represented in a way schematically illustrated in Figure 2(b). Figure 2(c) illustrates two such spectra obtained on a quartz sample at different excitation positions. Both spectra display characteristic quartz CL peaks but it is evident that the blue peak is not present in spectrum 1. Based on the requirements, spatial averaging can be used to enhance the signal-to-noise ratio in the spectra.
Instead of showing the spectrum for a specific point, spatial differences in the emission can also be thoroughly visualized for every excitation position. For instance, it is possible to extract a (false) color RGB image from the datacube where the emission spectrum is divided in three RGB channels in a specific spectral range.
Here, the spectral region from 380 to 700 nm is selected, which covers both peaks as illustrated in Figure 2(c). The corresponding spatial false-color map is shown in Figure 2(e). In agreement with Figure 2(c), region 1 is red because the peak at 650 nm is dominant, whereas region 2 is purple since both peaks are present, which puts considerable intensity in both the blue and the red color channel.
Another technique for visualizing the datacube is by taking a slice through the datacube at a definite wavelength, generating a wavelength-filtered grayscale image (schematically shown in Figure 2(b)). Figures 2(f) and 2(g) are examples of such images for the same region, which clearly portray the absence of the blue peak in the central region 1. It is possible to minimize noise in the image by averaging over a larger bandwidth. These examples demonstrate the diversity and power of hyperspectral imaging.
Figure 2. (a) Hyperspectral CL datacube containing the 2D spatial electron beam position and the emission wavelength λ. For each point, a full spectrum is collected. (b) From a datacube, it is also possible to extract the spatial distribution for a specific wavelength. (c) CL spectra measured on quartz sandstone. (d) SEM image of a region on a quartz sandstone. (e) False color RGB image taken from λ = 380–700 nm as indicated in (c). The positions from which spectra 1 and 2 are collected are also indicated. Wavelength cross cuts through the datacube (shown schematically in (b)) at (f) 425 and (g) 650 nm corresponding to the two main peaks in the quartz.
This information has been sourced, reviewed and adapted from materials provided by Delmic B.V.
For more information on this source, please visit Delmic B.V.