Spatial and spectral dynamics in STEM hyperspectral imaging using random scan patterns
Introduction
In the last twenty years, significant improvements on the spatial resolution of scanning transmission electron microscopes (STEM) have been achieved by the widespread implementation of spherical aberration correctors. Currently, atomically-resolved images are routinely obtained for a wide range of materials going from semiconductors, nanostructures, to functional oxides, etc. The structural information can be complemented by the synchronous acquisition of several spectroscopic signals accessible, such as electron energy loss, X-ray emission or cathodoluminescence (CL), which can then be bundled into hyperspectral images.
Besides the instrumental characteristics, the ultimate spatial resolution achievable in spectro-microscopy is strongly limited by the radiation sensitivity of the sample. Indeed, higher spatial resolution corresponds to a smaller illumination area and number of scattering centers, which imply higher electron doses for a given signal intensity collected. Irradiation effects have a much stronger impact in hyperspectral imaging since inelastic scattering processes have significantly lower cross sections than elastic scattering employed in STEM imaging, thus hindering the signal collection efficiency. Furthermore, within the finite time required for recording a hyperspectral image, the sample might drift, transform and/or the spectroscopic signals of interest evolve. Therefore, a hyperspectral image does not always portray the sample in a reliable state since different pixels are acquired at different times. These unavoidable drawbacks motivate the development of novel operating modes that could reduce sample dependent limitations in STEM spectro-microscopy.
The use of unconventional scan pathways is a promising route currently explored to improve acquisition speed, reduce electron dose [1] and, in certain cases, to avoid distortions associated with scan [2] and sample instabilities. In a STEM, the electron beam is typically scanned over the region of interest following a sawtooth-path to fill all pixels of a rectangular frame. Alternative spiral scans have also been recently proposed for reducing fly-back image distortions arising at high scanning rates [2], [3], [4]. In the last years, subsampling has been extensively discussed as a very effective strategy for dose-reduced image acquisition. Indeed, STEM images are often associated with a certain degree of over-redundancy, and a good approximation of the full image can already be obtained from an appropriate subset of pixels. The subsequent application of image reconstruction techniques can permit to fill the missing data within the image matrix [5], [6]. The direct acquisition of sparse images presents the important advantage of reducing radiation dose, acquisition time and data size [6], [7]. An easy and implementable choice for sparse sampling is a fully random distribution of the pixels, which permit high subsampling with minimal reconstruction distortions [8]. The applicability of different reconstruction algorithms for S(T)EM imaging has been demonstrated on simulated random scan images obtained by extracting a subset of randomly chosen pixels from full images [5], [7], [9]. The effectiveness of these methods has also been shown for multi-dimensional data structures such as tomography image series [10], [11], STEM-ptychography [12], energy-dispersive X-ray spectroscopy [13] and electron energy loss spectroscopy (EELS) hyperspectral images [14]. The direct acquisition of random sparse STEM images can be attained by scripting the acquisition control software [15], but this method suffers from a very slow response time. More efficiently, Béché and coworkers developed an experimental set up which integrates a fast electromagnetic shutter: the electron probe follows the conventional scan pattern but the beam can be turned off at randomly selected pixels [16]. A similar approach has been employed also for EELS hyperspectral imaging [17]. Whereas the method permits a reduction of the electron dose, the acquisition time remains invariant with respect to a standard scan. Other methods that generate sparse random or pseudo-random scans have been proposed by exploiting the position uncertainties due to the deflection coils dynamics [9], [18]. The interest of sparse sampling and compressive sensing has been stressed for TEM and STEM imaging but only minor attention has been dedicated to hyperspectral imaging, although irradiation effects, acquisition times and data sizes are significantly more important in this context.
In this work, we present an original implementation of an effective random scan acquisition mode in STEM obtained directly using the scanning control unit. The setup permits an extended control of the scanning parameters and therefore offers a large flexibility on the acquired image data structure. The complete image matrix is filled in a fully randomized fashion, each pixel being stamped with its acquisition time. A series of subsampled random sparse images can then be extracted at successive time frames and inpainting image reconstruction applied. This method allows to decouple the space and time information and to monitor the sample and spectral signals dynamics. After illustrating the operating principle (Section 2) and its practical implementation (Section 3), we provide examples of EELS and nano-cathodoluminescence (nano-CL) hyperspectral images acquired with the newly implemented random scan mode (Section 4). With respect to the conventional raster scan mode, we demonstrate that random scan permits to limit dose accumulation effects, but we also emphasize the possibility of tracking and correcting for sample drift and to monitor the spectroscopic signal dynamics over time in hyperspectral images.
Section snippets
Random scan operating principle
Fig. 1(a) represents the sawtooth-scan order typically employed in STEM hyperspectral imaging: the region of interest is scanned from left-to-right and top-to-bottom. In the proposed random-scan mode, each pixel is visited only once but in a random order. In the first step, the list of pixel coordinates of the scanning matrix is built. This list is then shuffled to define the order sequence for pixel illumination; the matrix representing the pixel scanning order can be stored for
Practical implementation in STEM spectro-microscopy
The evolution of the scanning modules for STEM now allows to generate arbitrary scanning patterns and to explore new illumination modes. The random-scan method described here has been implemented using a custom scan module based on a field-programmable gate array (FPGA) with a 50 MHz clock frequency controlling a scan engine operating at 25 MHz directly driving the scanning coils currents. The system is currently installed in two dedicated STEM microscopes: a VG-HB501 and a spherical
Proofs of concept
The random scan method can be very generally employed in hyperspectral imaging with the sole limitation of longer probe displacements with respect to a standard raster scan. As already discussed, its potential interest arises in the spectroscopy of systems sensitive to electron irradiation, and in monitoring the spectral signal and image dynamics. Here we present some proofs of concept of the flexibility of the random scan method illustrating its applications in different spectro-microscopy
Conclusions
In this work, we have presented the implementation of a random scan operating mode in STEM obtained at the hardware level via a custom scan control module, and provided examples of its use in EELS and nano-CL hyperspectral imaging. With respect to the conventional raster scanning mode, probe displacements are longer and thus random scan might not be as suitable for simple STEM imaging, but this is not problematic in STEM spectro-microscopy for which longer illumination times are required. The
Declaration of Conflict Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
We acknowledge support from the Agence Nationale de la Recherche (ANR), program of future investment TEMPOS-CHROMATEM and TEMPOS-NANOTEM (No. ANR-10-EQPX-50). The work has also received funding from the European Union in the Horizon 2020 Framework Programme (H2020-EU) under Grant Agreement No. 823717 (ESTEEM3). Authors would like to acknowledge Daniele Preziosi for the LAO–NNO thin film growth and Alexandre Gloter for the FIB lamella preparation.
References (44)
- et al.
How should a fixed budget of dwell time be spent in scanning electron microscopy to optimize image quality?
Ultramicroscopy
(2018) - et al.
Compressed sensing for STEM tomography
Ultramicroscopy
(2017) - et al.
High speed/low dose analytical electron microscopy with dynamic sampling
Micron
(2018) Radiation damage to organic and inorganic specimens in the TEM
Micron
(2019)- et al.
Image registration of low signal-to-noise cryo-STEM data
Ultramicroscopy
(2018) - et al.
Cathodoluminescence in the scanning transmission electron microscope
Ultramicroscopy
(2017) - et al.
Defect-induced blue luminescence of hexagonal boron nitride
Diamond Related Mater.
(2016) - et al.
A sub-sampled approach to extremely low-dose STEM
Appl. Phys. Lett.
(2018) - et al.
Dynamic scan control in STEM: spiral scans
Adv. Struct. Chem. Imaging
(2016) - et al.
Precision controlled atomic resolution scanning transmission electron microscopy using spiral scan pathways
Sci. Rep.
(2017)
Compressed sensing of scanning transmission electron microscopy (STEM) with nonrectangular scans
Microsc. Microanal.
The potential for Bayesian compressive sensing to significantly reduce electron dose in high-resolution stem images
Microscopy
Compressed sensing and electron microscopy
Modeling Nanoscale Imaging in Electron Microscopy
Compressed sensing MRI
IEEE Signal Process. Mag.
Sparse imaging for fast electron microscopy
Computational Imaging XI
Reduced-dose and high-speed acquisition strategies for multi-dimensional electron microscopy
Adv. Struct. Chem. Imaging
Subsampled stem-ptychography
Appl. Phys. Lett.
Reconstruction of partially sampled multiband images application to STEM-EELS imaging
IEEE Trans. Comput. Imaging
Towards the low-dose characterization of beam sensitive nanostructures via implementation of sparse image acquisition in scanning transmission electron microscopy
Meas. Sci. Technol.
Development of a fast electromagnetic beam blanker for compressed sensing in scanning transmission electron microscopy
Appl. Phys. Lett.
Practical implementation of compressive sensing for high resolution STEM
Microsc. Microanal.
Implementing an accurate and rapid sparse sampling approach for low-dose atomic resolution stem imaging
Appl. Phys. Lett.
Cited by (25)
Event-based hyperspectral EELS: towards nanosecond temporal resolution
2022, UltramicroscopyEvent driven 4D STEM acquisition with a Timepix3 detector: Microsecond dwell time and faster scans for high precision and low dose applications
2022, UltramicroscopyCitation Excerpt :The Timepix3 detector was mounted on a probe-corrected FEI Themis Z using a custom designed retractable mount interface shown in Fig. 1(d). In order to synchronize the scan coils with the detector, a custom scan engine is used [34,35]. The detector and scan engine are synchronized by using a 10 MHz reference clock output which is then multiplied with a custom designed phase locked loop circuit to act as a 40 MHz master clock for the Advapix in order to keep both scan engine and camera synchronized in time.
Probing molecular vibrations by monochromated electron microscopy
2022, Trends in ChemistryCitation Excerpt :The second issue is that the transmission configuration will invoke unavoidable radiation damage, despite the low current of monochromated electron beam (5–10 pA). This shortcoming can be partially diminished by (i) cooling the sample, (ii) limiting the total electron dose, and (iii) using new acquisition methods, such as random scan patterns [74]. Conducting experiments at cryogenic temperatures will significantly reduce the radiolysis damage [14,47–49].
Making the most of your electrons: Challenges and opportunities in characterizing hybrid interfaces with STEM
2021, Materials TodayCitation Excerpt :However, many algorithms have been developed to improve on this tactic by including learned information about the sample. One common method is a beta factor process analysis (BPFA) approach [186], where a dictionary learning Bayesian model fills in missing pixels probabilistically from known elements for image restoration [182,185,187–189]. Additionally, there exist inpainting methods for suppressing the noise associated with undersampled data [182].