Beyond Current Limits of Resolution,

Environment, and Data Analysis

Pre-Meeting Congress
Microscopy and Microanalysis 2019

Supported by Aberration-Corrected Microscopy FIG

August 4, 2019
7:30 am – 6:00 pm

Oregon Convention Center

Rooms F149-150
 Welcome
It is our pleasure to welcome you to the second annual Next-Generation
Transmission Electron Microscopy (NexTEM) meeting, held as a Pre-Meeting
Congress for Microscopy and Microanalysis 2019. The microscopy field
continues to evolve at a rapid pace, enabling transformative discoveries
across the scientific world. This year we focus on the advanced detectors, in
situ capabilities, and unique data science tools being developed to address
longstanding questions in materials science, chemistry, and physics.

NexTEM 2019 welcomes experts from across all domains of electron

microscopy. We showcase work spanning high-speed detectors, magnetic
field-free imaging, machine learning, and ultra-high-resolution spectroscopy.
Our goal is to connect microscopists, instrument manufacturers, and
scientists of all backgrounds to facilitate cross-disciplinary collaborations and
identify the developments needed to for breakthroughs in important research
areas. The workshop is again organized to allow for both the exchange of
research and the building of bridges between researchers. We are particularly
excited to host many students and early career scientists, whose contributions
in the coming years will shape the trajectory of the field.

As before, we hope that NexTEM will help frame a vision for the more
meaningful, efficient, and dynamic microscopy of tomorrow. We’re glad that
you can join us!

Warm regards,

Steven R. Spurgeon Mitra L. Taheri Demie Kepaptsoglou

Staff Scientist Professor Senior Lecturer
Energy and Environment Directorate Department of Materials Science SuperSTEM Laboratory
Pacific Northwest National Laboratory and Engineering and University of York
Johns Hopkins University
Agenda
7:30 – 8:30 a.m. REGISTRATION, COFFEE, AND BREAKFAST

8:30 – 8:45 a.m. Opening Remarks

Steven Spurgeon | Pacific Northwest National Laboratory
Mitra Taheri | Johns Hopkins University
Demie Kepaptsoglou | SuperSTEM Laboratory

8:45 – 10:30 a.m. Session 1: Advanced Detector Technologies for New Imaging Modalities

8:45 – 9:15 a.m. Introduction to Direct Electron Detector Technologies and Applications
David Muller | Cornell University

9:15 – 9:45 a.m. The Need for Speed: Opportunities in Materials Characterization Created by
Fast Detectors and High Speed Electron Beam Control
Paul Voyles | University of Wisconsin – Madison

9:45 – 10:15 a.m. Magnetic-field-free Atomic Resolution STEM

Naoya Shibata | University of Tokyo

10:15 – 10:30 a.m. Open Discussion

All Attendees

10:30 – 11:00 a.m. Morning Break: Coffee and Refreshments

All Attendees

11:00 – 12:15 a.m. Session 2: Frontiers of In Situ Imaging and Analysis

11:00 – 11:30 a.m. Tentative: Electron Microscopy Advances in Catalysis

Stig Helveg | Haldor Topsoe

11:30 – 12:00 p.m. In-Situ S/TEM and their Applications for Energy Materials: The Challenge and
Opportunities
Chongmin Wang | Pacific Northwest National Laboratory

12:00 – 12:15 p.m. Open Discussion

All Attendees

12:15 – 1:15 p.m. Lunch Break

All Attendees

1:15 – 2:30 p.m. Session 3: High-Resolution Electron Spectroscopies

1:15 – 1:45 p.m. Vibrational Spectroscopy in STEM-EELS at High Spatial Resolution

Quentin Ramasse | SuperSTEM Laboratory
1:45 – 2:15 p.m. Optical Spectroscopy with an Electron Microscope and a Light Collector /
Injector
Luiz Tizei | Université Paris Sud

2:15 – 2:30 p.m. Open Discussion

All Attendees

2:30 – 3:00 p.m. Afternoon Break: Coffee and Refreshments

All Attendees

3:00 – 5:00 p.m. Session 4: Emerging Machine Learning, Image Processing, and Simulation
Methods

3:00 – 3:30 p.m. Solving Electron Microscopy: Towards Reconstruction Methods for Thick,
Strongly Scattering Materials
Hamish Brown | Lawrence Berkeley National Laboratory

3:30 – 4:00 p.m. Theory-experiment Matching via Machine Learning Physics-Coupled

Frameworks
Rama Vasudevan | Oak Ridge National Laboratory

4:00 – 5:00 p.m. Discussion: Harnessing Emerging Techniques

All Attendees

5:00 – 6:00 p.m. Reception: Appetizers and Drinks

All Attendees
We appreciate the support of our sponsors – thank you!

Gold

Silver

Bronze
Introduction to Direct Electron Detector Technologies and Applications

David A. Muller1,4,*, Zhen Chen1, Yi Jiang2, Kayla X. Nguyen3, Yu-Tsun Shao1, Yimo Han1, Prafull
Purohit2, Mark W. Tate2, Sol M. Gruner2,4, Veit Elser2
1.
School of Applied and Engineering Physics, Cornell University, Ithaca, NY 14853, USA
2.
Department of Physics, Cornell University, Ithaca, NY 14853, USA
3.
Department of Materials Science, University of Illinois, Urbana, IL 61801, USA
4.
Kavli Institute at Cornell for Nanoscale Science, Ithaca, NY 14853, USA
* Corresponding author: [email protected]

The new generation of direct electron detectors are reshaping what can be measured with electron
microscopes. Monolithic active pixel sensors (MAPS) detectors, which are optimized for a large number
of low-dose-rate, small pixels, have revolutionized cryo-electron and in-situ TEM. These detectors are
poorly-suited to the large currents and concentrated intensities that characterize STEM signals. Here,
pixel array detectors (PADs) such as the EMPAD, where large pixels allow for more sophisticated charge
measurements, have demonstrated the necessary sensitivity, dynamic range and dose-rates to record the
full scattering distribution of every transmitted electron at every probe position [1], enabling quantitative
and dose-efficient imaging of strains with sub-pm precision, magnetic fields from layers only a few atoms
thick, ferroelectric textures and vortices, and ptychographic imaging well beyond the diffraction limit of
the microscope.

For all of these applications, the key performance metrics for STEM is the ability to record a low-noise
signal with a large dynamic range in as short a time as possible. Figure 1a shows the accuracy of recording
the angular deflections of the center of mass of a diffraction disk is dominated by the counting statistics,
requiring a large number of electrons for a precise measurement, and relatively few pixels are needed,
with diminishing returns beyond 10x10 pixels per diffraction disk. Fig 1b shows measurements of
magnetic Skyrmions in a sub-nm thick cobalt film, with a precision of 0.1µrad.

By combining our new design of electron microscope pixel array detector (EMPAD) [1] which has the
dynamic range to record the complete distribution of transmitted electrons at every beam position, and a
ptychographic phase retrieval algorithm to process the data, we have been able to increase the spatial
resolution well beyond the traditional lens limitations reaching a 0.39 Å resolution for MoS2, at 80 keV,
the same dose and imaging conditions where conventional imaging modes reach only 0.98 Å [2]. The
improved resolution, dose efficiency and robustness to environmental noise enabled by ptychography
make it easy to identify defects such as sulfur monovacancies, as well as subtle structural arrangements
and tilts on the sulfur sublattice that are undetectable by conventional imaging modes. For twisted bilayers
we are able to resolve the shear distortions and interactions between the layers (Figure 1). [3]

References:

[1] MW Tate, et al, Microscopy and Microanalysis 22 (2016), p. 237.

[2] Y. Jiang et al, Nature 559, (2018), p343.
[3] We thank Jiwoong Park, Saien Xie, Hui Gao, Ming-Hui Chiu, and Lain-Jong Li for samples. Research
supported by US National Science Foundation (grants DMR-1539918, DMR-1429155, DMR-1719875),
DARPA (D18AC00009) & US Department of Energy Basic Energy Sciences (DE-SC0005827).
Figure 1. Measuring the angular deflections needed for strain mapping and magnetic field imaging. (a)
The smallest measurable angular deflection depends strongly on the dose, and only weakly on the number
of pixels. Usually only 10 pixels/disk diameter are needed. However, for rapid mapping, a high dose rate
(>107 e/ms/disk) is essential. This translates to a few pA/pixel which is beyond the linear count rate of all
pulse counting detectors, but still achievable with charge integrating designs. (b) EMPAD measurements
at 0.1µrad precision of magnetic fields from Néel-type Skyrmions in a single 0.9 nm thick cobalt film,
sandwiched between heavy metal layers.

Figure 2. Real space resolution demonstration of full-field, in-focus ptychography using a twisted bilayer
MoS2. (a) Synthesized ADF image; (b) Phase image from ptychographic reconstruction. Data adapted
from [3]
The Need for Speed: Opportunities in Materials Characterization Created by Fast
Detectors and High Speed Electron Beam Control

Paul M. Voyles1*
1.
Department of Materials Science & Engineering, University of Wisconsin-Madison, Madison, WI USA
* Corresponding author: [email protected]

High speed electron scattering with pulsed sources has revealed a variety of phenomena at timescales from
µs to fs. However, pulsed sources add instrument complexity and expense and space charge effects in
pulses impose special requirements, such as single electron pulses limited to stroboscopic experiments
and megavolt voltages. High speed electron microscopy with continuous sources has the potential to create
new opportunities in materials characterization. The key enabling technologies are fast pixelated direct
electron detectors and high speed beam manipulation using electrostatic deflectors or RF cavities.
The current state of the art in electron cameras operate at frame rates up to ~5000 frames per second (fps),
but this is far from the limit of CMOS-based cameras, as light cameras can achieve operate at up to
1,000,000 fps [1]. Faster cameras are coming to electron microscopy. The Molecular Foundry at Lawrence
Berkeley lab recently tested a camera with 576×576 pixels operating at 87,000 frames per second [2]. The
UW-Madison Materials Research Science and Engineering Center and Direct Electron Inc. are developing
a camera that will operate at 2,000 to 120,000 fps as the pixel count is reduced from 1024×1024 to
128×128. The UW camera also will feature low, high, and high dynamic range gain modes, operation in
rolling or global shutter mode, and correlated double sampling readout for reduced noise.
These new cameras will enabling 4D STEM experiments with µs pixel dwell times similar to what is
commonly used with integrating STEM detectors currently. The high speed will reduce the influence of
instabilities, enable drift correction, and enable experiments including time-resolved 4D STEM, densely-
sampled strain maps over large areas, and diffraction tomography. High speed will also enable improved
detector dynamic range by rapid frame averaging.
However, µs time resolution experiments using TEM imaging with current continuous sources will be
strongly limited by low intensities. An alternative is take a statistical approach, in the electron equivalent
of x-ray photon correlation spectroscopy, electron correlation microscopy (ECM) [3-5]. ECM uses
intensity fluctuations in diffraction, measured either in dark-field TEM or STEM nanodiffraction, to
characterize fluctuations in a system from the time autocorrelation function of the intensity, I(t).
〈()* + ,〉〈()* + .*,〉
!" ($) = 〈((* + )〉/
,

The noise in g2(t) is largely governed by the total number of detected electrons, not the signal in particular
measurement I(t), so ECM can tolerate very low dose per image.
ECM has been used to study structural fluctuations in metallic glass forming liquids as a function of
temperature above the glass transition temperature, Tg using in situ heating [3-5]. Figure 1 shows spatial
maps of the structural relaxation time, t, derived from ECM data on Pt57.5Cu14.7Ni5.3P22.5 nanowires [5].
These data show that the motion of atoms in a liquid are spatially heterogeneous. At the nanometer scale,
t varies by as much as an order of magnitude. This spatial variation is central to theories of the glass
transition. Figure 1 also shows a ~1 nm thick surface layer with order of magnitude faster dynamics than
the bulk, which promotes homogenous nucleation of crystals without catalytic nucleation from the surface.
The structural relaxation time in a liquid
in equilibrium is related to the viscosity,
which describes the liquid response to
pressure, by the Debye-Stokes-Einstein
(DSE) relation. The DSE is one example
of a fluctuation dissipation relation,
which connect the behavior of a system
under external stimuli to its fluctuations
in equilibrium. Any fluctuations that
produce diffraction are amenable to
characterization with ECM. Examples of
include fluctuating domains in
ferroelectrics or ferromagnets at constant
field, which are connected to domain
switching behavior, and atomic hopping Figure 1: (a)-(e) ECM-derived maps of t for a Pt-based
without a concentration gradient, which metallic glass nanowire showing spatially heterogeneous
is related to diffusivity. Current cameras dynamics within the wire and a surface layer with fast
limit the time resolution of ECM to 10s dynamics. (f) Arrhenius temperature dependence of surface
of ms, but next-generation cameras will and bulk t.
enable probing of time scales <100 µs.
The second key technology is fast beam control, with electrostatic systems offering beam positioning at
nanosecond timescales and RF cavities offering switching at picosecond timescales. Kisielowski et al.
recently demonstrated that RF modulation of a continuous electron beam can dramatically reduce beam
damage in MgCl2 compared to the same dose in a continuous beam [6]. The RF modulation was so fast
that many of the “on” pulses did not contain even a single electron, but ensuring that the beam is off for
certain periods allowed the material to recover, rather than accumulating damage. Nanosecond beam
positioning will enable ultrafast STEM scanning and acquisition schemes based on compressed sensing.
Compressed sensing will enable multiplexing of signals on a pixelated detector for even faster readout,
and, while it does not reduce the total dose required to acquire an image at fixed signal to noise ratio [7],
it does allow that dose to be distributed amongst fewer spatial positions [8]. For samples with limited
ability to dissipate dose at short length scales, this may also limit accumulation of beam damage [9].
References
[1] S Okura et al., 2014 Symp VLSI Circuits Dig Tech Pap, 10.1109/VLSIC.2014.6858429(2014), p. 1.
[2] Successful Demonstration of Fastest Electron Detector Ever Made, (2019),
https://foundry.lbl.gov/2019/03/17/successful-demonstration-of-fastest-electron-detector-ever-made/
[3] L He et al., Microscopy and Microanalysis 21 (2015), p. 1026.
[4] P Zhang et al., Ultramicroscopy 178 (2017), p. 125.
[5] P Zhang et al., Nature Communications 9 (2018), p. 1129.
[6] C Kisielowski et al., Advanced Functional Materials 29 (2019), p. 1807818.
[7] W Van den Broek et al., IEEE Transactions on Computational Imaging, DOI:
10.1109/tci.2019.2894950(2019).
[8] A Béché et al., Applied Physics Letters 108 (2016), p. 093103.
[9] Development of ECM is supported by the NSF (DMR-1807241). Development of advanced detectors
and microscopy facilities are supported by the Wisconsin MRSEC (DMR-1728933).
Magnetic-field-free Atomic Resolution STEM

Naoya Shibata1,2
1.
Institute of Engineering Innovation, The University of Tokyo, Tokyo 113-8656 Japan
2.
Nanostructures Research Laboratory, Japan Fine Ceramic Center, Nagoya 456-8587, Japan
[email protected]

Atomic-scale structures of defects in materials and devices are essential to control their functional
properties. Aberration-corrected scanning transmission electron microscopy (STEM) is a powerful
technique to directly observe atomic-scale structures of individual defects, such as interfaces, inside
materials and devices. In state-of-the-art aberration-corrected STEM, a probe size of less than 0.5Å in
diameter has been realized [1], which is smaller than the Bohr radius of a hydrogen atom. The following
interesting question arises: beyond just atoms, what might become observable by using such fine
probes?
One answer to this question may come from exploring new possibilities in phase-contrast imaging in
STEM. By using more elaborate detectors, we can not only image atomic structures of materials, but can
also image local electromagnetic fields inside materials through differential phase contrast (DPC)
imaging techniques [2-4]. Applying DPC imaging in atomic resolution STEM [5], we can now directly
visualize the electric field distribution within single atoms [6,7], i.e. the electric field between the
positively charged atomic nucleus and the negatively charged electron cloud. Moreover, that the electric
field distribution inside an atom can be quantitatively observed should further mean that this image can
be converted into the total charge density distribution according to the relevant Maxwell equation. That
is to say, we may be able to directly image the charge density distribution inside an atom or in between
atoms in real space. Thus, atomic-resolution DPC STEM opens a new stage of microscopy that enables
directly visualization of the structure within individual atoms via electric field imaging.
It then becomes tempting to directly observe magnetic fields at atomic resolution, which should strongly
correlate with the properties of magnetic materials and devices. However, by using conventional
electron microscopes, atomic-resolution observation of magnetic materials is essentially very difficult,
because high magnetic fields (2-3T) are always exerted on samples inside the magnetic objective lens.
Very recently, we have developed a new magnetic objective lens system that realizes a magnetic field
free environment at the sample position [8]. Using this new objective lens system combined with the
state-of-the-art higher order aberration corrector [9], direct atom-resolved imaging in a magnetic field
free environment has finally been achieved by STEM [10]. Figure 1 shows an annular dark field STEM
image of Fe-3%Si alloy, one of the most important engineering soft magnetic materials. This novel
electron microscope (Magnetic-field-free Atomic Resolution STEM: MARS), shown in Figure 2, is
expected to be used for research and development of advanced magnetic materials and devices. The
details of this new STEM will be presented in the talk.

[1] S. Morishita et al., Microscopy, 67, 46 (2018).

[2] N.H. Dekkers and H. de Lang, Optik, 41, 452 (1974).
[3] H. Rose, Ultramicroscopy, 2, 251-267(1977).
[4] J.N. Chapman, I.R. McFadyen and S. McVitie, IEEE Trans. Mag., 26, 1506-1511(1990).
[5] N. Shibata et al., Nature Phys., 8, 611-615 (2012).
[6] N. Shibata et al., Nature Comm. 8, 15631 (2017).
[7] R. Ishikawa et al., Nature Comm., 9, 3878 (2018).
[8] Y. Kohno, S. Morishita and N. Shibata, Microsc. Microanal. 23 (Suppl. 1), 456-457 (2017).
[9] H. Sawada et al., J. Electron Microsc., 58, 341-347 (2009).
[10] N. Shibata et al., Nature Comm. 10, 2380 (2019).
[11] The author thank all the collaborators of this research, especially Y. Kohno, T. Seki, S. Morishita,
A. Nakamura, A. Kumamoto, H. Sawada, T. Matsumoto, S.D. Findlay and Y. Ikuhara for their
contribution to the works shown in this presentation. The author acknowledges support from the JST
SENTAN Grant Number JPMJSN14A and the JSPS KAKENHI Grant number JP17H01316.

Figure 1. Atomic-resolution annular dark field STEM

image of Fe-3%Si alloy observed from the [110]
direction. The magnetic field at the sample position is
kept well below 0.2mT.

Figure 2. The MARS microscope installed in the

University of Tokyo.
In-Situ S/TEM and their Applications for Energy Materials: The Challenge
and Opportunities
Chongmin Wang

Pacific Northwest National Laboratory, 902 Battelle Boulevard, Richland, WA 99352, USA

Direct observation of process under a device operating condition, as generically termed as in-situ
and operando, appears to be one of the most promising and efficient approaches for revealing
mystery that leads a system to deviate from the otherwise ideal operating condition, typically such
as answering the questions why and how a rechargeable battery fade. Often, in-situ and operando
observations are not straightforward, demanding appropriate unique deliberation, if not an
invention, of new capabilities to allow the system to be observed while operating at normal
conditions. Scanning/transmission electron microscopy (S/TEM), as a powerful analytical tool for
addressing many unique scientific problems in various disciplines, shows no exception to this
general expectation of coupling in in-situ and operando imaging for solving the challenging
science questions. This is especially true as considering tremendous progress has been made on
the development of aberration corrected S/TEM and scanning transmission electron microscopy.
The questions now come to how we extend the microscopy and spectroscopy methodologies to
analyze materials under operating condition, typically such as real time observation of structural
and chemical evolution of electrode materials in batteries, catalytic process, oxidation and
reduction, bio-tissue in a liquid cell, defects generation and interaction under deformation
conditions, and mass and charge transport process.

Approaches that are employed to provide in-situ observation of chemical reactions and
microstructural evolution include, but are not limited to, the use of differential pumping
technologies, the modification of sample holders to enable the application of an external stimulus
such as an electromagnetic field or stress, employing thin membrane closed cells for experiments
involving liquids or gases, and use of liquid with low vapor pressure. In view of these
developments, this work will focus on in-situ techniques that developed for probing into the energy
materials, highlighting direct observation of structural and chemical evolution, solid-gas interface,
nanoparticle interaction, phase transformation and their correlation with mass and charge transport,
providing insights as how active materials failure during the cyclic charging and discharging of a
battery. In perspective, challenges and possible direction for further development of the in-situ
S/TEM imaging and spectroscopic methods for both functional and structural materials and other
field will also be discussed.

As a specific example, nanoparticles of either as an active functioning component or as a reaction

product is of great importance for a range of science and engineering, such as catalysis, nucleation
and growth, metal air-battery, and oxidation and corrosion. A typical quesion is the deactivation
of catalyst particles. It has been commonly realized that deactivation of catalysts stems from
surface area loss due to particle coarsening, however, for which the mechanism remains largely
unclear. Direct in-situ TEM observation reveals unprecedented details at atomic scale with respect
to the behavior of Pt nanoparticles for fuel cell application, especially the coarsening dynamics of
Pt nanoparticles on bare CNTs surface and in Nafion covered Pt/CNTs under O2 and H2O gas
environment. We found that the coarsening of Pt nanoparticles proceeded through migration and
coalescence process. The migration rates of Pt nanoparticles on bare CNT surface is much faster
in O2 than that in H2O. The strong oxygen chemisorption on Pt nanoparticles weakens the
interaction between Pt and CNT surface, leading to a fast migration in O2. Incorporation of Nafion
electrolyte layer put on a mechanically confinement to Pt/CNTs which reduces the Pt migration
rate in O2. However, this mechanical confinement is largely relieved by introducing H2O, and a
lubricated interface is created, both leading to a faster migration rate of Pt in H2O than that in O2.
These results bridge the gap between the previous in situ TEM studies in idealized conditions and
the technological relevant conditions of Pt catalyst for PEMFC, thus provide insights to rational
design of Pt/carbon catalysts incorporating Nafion electrolyte to achieve a long-lasting PEM fuel
cells.

Figure 1. Migration and coalescence process of Pt/CNTs at 100 oC. (a) HRTEM image shows
two separated Pt nanoparticles in vacuum. (b) HRTEM image shows the distance between two Pt
nanoparticles decreased but still maintained their individual shape and orientation after 30 minutes
in H2O. (c) A series of time-lapsed HRTEM images show the coalescence process of two Pt
nanoparticles after switching gas from H2O to O2. (d) Schematic illustrates the coalescence of these
two Pt nanoparticles involves rotation of one NP and growth through orientation match.
References
[1] Yuyan Shao, Jiehe Sui, Geping Yin, and Yunzhi Gao, “Nitrogen-doped carbon
nanostructures and their composites as catalytic materials for proton exchange membrane
fuel cell”, Applied Catalysis B: Environmental. 2008, 79, 89-99.
[2] The work was conducted in the William R. Wiley Environmental Molecular Sciences
Laboratory (EMSL), a national scientific user facility sponsored by DOE’s Office of
Biological and Environmental Research and located at PNNL.
Vibrational Spectroscopy in STEM-EELS at High Spatial Resolution

Quentin M. Ramasse1,2, Fredrik S. Hage1, Demie Kepaptsoglou1,3, Rebecca J. Nicholls4, Jonathan R.

Yates4, Leslie J. Allen5,6
1.
SuperSTEM Laboratory, STFC Daresbury Campus, Keckwick Lane, Daresbury WA4 4AD, U.K.
2.
School of Physics and School of Chemical and Process Engineering, University of Leeds, Leeds, U.K.
3.
School of Physics, University of Melbourne, Parkville, Victoria 3010, Australia
4.
Department of Materials, University of Oxford, Oxford United Kingdom
5.
Ernst-Ruska-Centre for Microscopy and Spectroscopy with Electrons, Forschungzentrum Jülich,
Jülich, Germany
6.
School of Physics, University of Melbourne, Parkville, Victoria 3010, Australia

A new generation of electron beam monochromators has recently pushed the energy resolution of
(scanning) transmission electron microscopes deep into the sub 10meV range [1]. In addition to the
obvious increase in resolution which has made exploring the phonon region of the EELS spectrum
possible [1], the increased flexibility of these instruments is proving hugely advantageous for materials
science investigations. The energy resolution, beam current and electron optics can be adjusted
seamlessly and traded off each other as necessary within a greatly increased range.

In an early application to the field of two-dimensional materials, the cubic (c-BN) and hexagonal (h-BN)
phases of boron nitride were shown to give rise to markedly different phonon excitations in EELS [2]. A
better understanding of this phonon response can be achieved by observing the dependence of the
phonon peak under different optical conditions, mapping the energy and intensity of the peaks in
momentum space. Such momentum-resolved vibrational measurements had so far been limited to “bulk”
techniques (e.g. inelastic x-ray and triple-axis neutron scattering spectroscopies), the average surface
response (e.g. reflection EELS) or small momentum transfers (e.g. optical techniques). Carefully
balancing the intrinsic trade-off between simultaneously achievable resolutions in real and momentum
space, it is possible to map the vibrational response of boron nitride along different directions of the
Brillouin zone, using a ~1nm electron probe thus probing a sample volume ~1010-1020 times smaller
than with optical techniques [3,4].

A related technical approach can also be used to achieve atomic resolution in phonon spectroscopy at
atomic resolution. By displacing the EELS collection aperture away from the optic axis of the
instrument to angles similar to those used for Z-contrast image detectors, the contribution of electrons
that have undergone dipole scattering as well as elastic scattering is minimised and this optical geometry
favours instead the highly localised impact transitions in the recorded vibrational spectrum [5].

Simulations based on the quantum excitation of phonons (QEP) formalism show good agreement with
the experimental data and reproduce the atomic-resolution contrast obtained in the experimental images.
These demonstrate conclusively and directly that inelastic scattering associated with phonon excitations
plays a central role in generating the contrast widely used for atomic resolution high angle annular dark
field imaging [6].
References

[1] OL Krivanek et al., Phil. Trans. Roy. Soc. 367 (2009), pp. 3683-3697; T.Miyata et al., Microscopy 5
(2014), pp. 377-382 ; O.L. Krivanek et al., Nature 514 (2014), pp. 209-212.
[2] RJ Nicholls et al., Microsc. Microanal. 21 Suppl.3 (2015), pp. 1469-1470.
[3] FS Hage, RJ Nicholls, JR Yates et al., Science Advances 4 (2018), eaar7495.
[4] RJ Nicholls, FS Hage, DG McCulloch et al., PRB 99 (2019), 094105.
[5] FS Hage, DM Kepaptsoglou, QM Ramasse et al., PRL 122 (2019), 016103. (2019).
[6] SuperSTEM is the UK National Research Facility for Advanced Electron Microscopy, funded by the
Engineering and Physical Sciences Research Council (EPSRC).

Figure 1. (a) Selected experimental and modelled momentum resolved hBN EEL spectra. (b) “Aloof”
hBN EEL spectrum, acquired at a beam incidence of 74.5° to the crystallographic c-axis and at an
impact parameter of b = 20nm. In-plane and out-of-plane phonon mode contributions are indicated. (c)
Modelled and (d) experimental phonon dispersions, where q is the wave vector.

Figure 2. Left: Experimental geometry for atomic-resolution vibrational spectroscopy; Right:

Comparison between experimental phonon image for hexagonal boron nitride (integrating the phonon
sector of the spectrum) and simulation of the inelastic scattering using the QEP formalism [5].
Optical Spectroscopy with an Electron Microscope and a Light Collector / Injector

Luiz H. G. Tizei1*, Noémie Bonnet1, Pabitras Das1, Yves Auad1, Steffi Woo1, Alberto Zobelli1, Jean-
Denis Blazit, Marcel Tencé1, Odile Stéphan1, and Mathieu Kociak1
1.
Laboratoire de Physique des Solides, Université Paris-Sud, CNRS-UMR 8502, Orsay 91405, France
* Corresponding author: [email protected]

Traditionally, optical spectroscopy is performed with light as the excitation source, with a spatial
resolution (with few exceptions for specific techniques) limited by light diffraction. For this reason,
optical spectroscopy using electrons has been developed either in absoprtion (electron energy loss
spectroscopy, EELS) or in emission (cathodoluminescence, CL) leading to nanometer spacial resolution
[1-3]. The increased spectral resolution due to modern monochromators (sub 10 meV) [4,5] has
attracted more attention to EELS for nanooptics and stretched its applicable range into the infrared.

In this contribution, we will start by describing a home-made light collection/injection system [6]
designed for scanning transmission electron microscopes. Recent cathodoluminescence experiments at
cryogenic temperatures performed with this system will be described, specially the detection of single
point defects throught their light emission statistics [7, 8]. Specifically, non-sequential electron beam
scan patterns (i. e. random scan patterns) have allowed us to follow the slow scale (seconds) evolution of
defects’ emission (blinking and spectral diffusion) under electron irradiation [9].

Despite a clear interest due to the increased spacial resolution, there are many aspects accessible in
standard optical spectroscopy that are missed in traditional spectroscopies with electrons: excitation
energy, polarization, and temporal selection, to cite a few. Moreover, electron-matter and photon-matter
interaction are different, leading to different in CL and photoluminesnence (PL) signals (as recently
observed for nitrogen-vacancy centers in diamond [10]). Finally, electron-photon interactions mediated
by matter [11,12] give access to properties that cannot be probed by electrons or photons independently
at the nanometer scale [13].

So we will take a step back and show how we can use the same light collection/injection system to
perform ~µm scale PL mapping of NV centers in diamonds. For this, we have used the parabolic mirror
of the light collection/injection system to focus a 543 nm laser onto a few micron wide spot and to
collect the emitted light signal from diamonds. Using this setup, the emitted light under electron, laser or
combined excitation can be observed (Figure 1a). Spatial maps of the signal variation were measure by
scanning the electron microscope sample state (Figure 1b-d).

Laser excitation concomitant with electron spectroscopy has been used to perfom gain spectroscopy [11,
12, 14]. We have developped an electron gain setup which, contrary to existing ones, does not require a
pulsed gun and can be, therefore, a more easily democratized. We will discuss how the ability to form a
few micrometer wide laser spot has significantly advanced gain experiments in our setup, allowing the
detection of polarization- and energy-resolved signals [15].
References:

[1] N. Yamamoto, K Araya, and F. J. G. de Abajo, Phys. Rev. B 64 (2001), p. 205419.

[2] J. Nelayah, et al, Nat. Phys. 3 (2007), p. 348.
[3] L. F. Zagonel, et al, Nano Lett. 11 (2011), p. 568.
[4] O. Krivanek, et al, Nature, 514 (2014), p. 209.
[5] L. H. G. Tizei et al, Phys. Rev. Lett. 114 (2015), p. 107601.
[6] M. Kociak, L. F. Zagonel, Ultramicroscopy 176 (2017) p. 112.
[7] L. H. G. Tizei and M. Kociak, Phys. Rev. Lett. 110 (2013) p. 153604.
[8] R. Bourrellier, et al, Nano Lett. 16 (2016) p. 4317.
[9] A. Zobelli, et al in preparation (2019).
[10] Meuret, private communication.
[11] S. T. Park, M. Lin, and H. Zewail, New Journal of Physics 12 (2010) p. 123028.
[12] A. Feist, et al, Nature 521 (2015) p. 200.
[13] V. Di Giulio, M. Kociak, and F. J. G. de Abajo, arXiv:1905.06887 (2019).
[14] P. Das, et al, Ultramicroscopy 203 (2019) p. 44.
[15] Y. Auad and M. Kociak, unpublished results (2019).
[16] The authors acknowledge funding from the program of future investment TEMPOS-
CHROMATEM (No. ANR-10-EQPX-50).

Figure 1: a) Spectra measured with electron (CL, blue), 543 nm laser (PL, orange) and both excitations
(CL+PL, green). b-c) Spectrum images acquired by scanning the sample with electron, electron+laser
and laser excitations, respectively.
Solving Electron Microscopy: Towards Reconstruction Methods for Thick, Strongly
Scattering Materials

Hamish G. Brown1

1.
National Centre for Electron Microscopy, Molecular Foundry, Lawrence Berkeley National
Laboratory, Berkeley, USA

A relativistic electron is an excellent tool for imaging condensed matter, it has a short wavelength (~ 2
pm) so can be focused to a small spot with modern electron optics. Since electrons are charged particles
they interact strongly with atoms via the coulomb force, making possible studies of very small amounts
of matter such as individual grain boundaries in graphene or single atomic dopants within a material. A
side-effect of this strong interaction with matter is that electrons will often undergo complicated multiple
scattering in all but the thinnest samples. In this talk I will discuss how modern technologies in scanning
transmission electron microscopy (STEM), in particular fast readout electron cameras, give us data of
sufficient quantity and quality to solve the equations governing multiple scattering of the electron beam.
This suggests a future where we can “solve” the physics of electron scattering and render STEM results
more directly interpretable.

A fully quantum mechanical approach to the electron sample interaction is given by the Schrodinger
equation with a few approximations appropriate for a relativistic electron. The canonical solution to these
equations is the scattering (S) matrix solution [1], Fig. 1(a). However most reconstruction approaches,
such as ptychography and differential phase contrast (DPC) reconstruction, assume a very thin specimen,
the projection approximation, Fig. 1(b), which models specimen beam interaction by multiplication of the
illumination wave function with a transmission function [2]. Applications of ptychography and DPC are
wide-spread, for microscopy specimens that are indeed very thin, ptychography and DPC provide directly
interpretable maps of the electrostatic potential of the specimen [3,4]. Relative to conventional STEM
techniques, these techniques often provide additional benefits such as higher-resolution, better signal to
noise and computational correction of residual aberrations. For thicker specimens these techniques still
provide useful qualitative results, such as making weakly scattering atoms more visible [4].

In this talk I will discuss how with fast readout electron cameras in STEM we record four dimensional
datasets – a full 2D diffraction pattern for a 2D raster scan of the electron probe – sufficient information
to reconstruct the full scattering matrix, Fig 2 [5]. This provides an approach to tackling the reconstruction
of the specimen in the presence of full multiple scattering of the electron beam. For a crystalline specimen
the electrostatic potential can be solved directly from the scattering matrix [6,7], for the general case a
depth-section approach or inverse multislice approach is promising in some situations [8].

A final approach to overcoming the limits of the projection approximation is via the next generation of
electron lens aberration correctors. Recent developments make possible a probe with phase curvature that
is flat out to 50 mrad, or greater, in reciprocal space [9]. Such a probe would have a small depth of field
such that it would interact with a small volume of the specimen – thus enabling reconstruction via methods
that assume a weak interaction with the specimen [10].
References:

[1] Sturkey, L. "The calculation of electron diffraction intensities." Proceedings of the Physical Society
80 (1962), p. 321.
[2] Cowley, J. M., and Moodie, A. F., "The scattering of electrons by atoms and crystals. I. A new
theoretical approach." Acta Crystallographica 10 (1957) p. 609.
[3] Jiang, Y, et al. "Electron ptychography of 2D materials to deep sub-ångström resolution." Nature 559
(2018), p. 343.
[4] Shibata, N., et al. "Electric field imaging of single atoms." Nature communications 8 (2017) 15631.
[5] Findlay, S. D., “Quantitative structure retrieval using scanning transmission electron microscopy.”
Acta Crystallographica Section A 61 (2005), p. 397.
[6] Allen, L. J. et al., “Inversion of dynamical electron diffraction data including absorption.” Acta
Crystallographica Section A 56 (2000), p. 119.
[7] Brown, H. G. et al., Physical Review Letters 121, (2018) 266102.
[8] Gao S. et al., “Electron ptychographic microscopy for three-dimensional imaging”, Nature
Communications 8 (2017) p. 163.
[8] Hosokawa, F., et al. , Microscopy 62, (2013) p. 23.
[9] Brown, H. G., et al. "Large angle illumination enabling accurate structure reconstruction from thick
samples in scanning transmission electron microscopy." Ultramicroscopy 197 (2019) p. 112.

Figure 1. (a) In general the scattering of a fast electron (incident wave ψ0) must be modeled using a
scattering matrix (S). (b) The projection approximation assumes a very thin specimen, resulting in a
simpler model where the illumination is multiplied by a specimen transmission function T in real space .

Figure 2. Reconstruction of the scattering matrix for a 4D-STEM data set. (a) A through focal scan of
4D-STEM datasets is recorded for a Si110 crystal and images are synthesized by integrated circular
regions of the diffraction pattern in (b) to make focal series images of STEM bright-field images in (c).
An iterative through-focal series phase reconstruction method is used to reconstruct the phase in (d) and
the result is a representation of the scattering matrix.
Theory-experiment Matching via Machine Learning Physics-Coupled Frameworks

R. K. Vasudevan1, N. Borodinov1 L. Vlcek2, M. Oxley1, A. Lupini1, S. V. Kalinin1

1.
Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge TN 37830
2.
Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge TN 37830

Recent advances in imaging, and specifically in electron microscopy, have resulted in a wealth of data
on complex and disordered systems. Of special note is the ability to acquire convergent beam electron
diffraction (CBED) patterns across a sample with extremely high (sub unit-cell) precision, within so-
called 4D STEM, theoretically enabling much more structural information to be gathered on the sample
than using more conventional 2D imaging modes [1]. However, the tremendous increases in data
volume and complexity brings forth several basic questions, including (a) how can the 4D STEM data
be adequately utilized in a real structure determination setting, and (b) given the structural information
down to the picometer level, how can one utilize this data to gain information on the system. These data
can generally be simulated to a high degree of accuracy; yet, the methods of theory-experiment
matching for such data are still nascent.

To answer the first question, we created a workflow that can be followed for any structure determination
problem that marries deep learning, generative adversarial modelling, and simulations. The idea is
shown in Figure 1 (a). First, the CBED patterns are modelled based on a suite of N possible structures.
One caveat here is that the real structure (which is unknown) should nevertheless be in this dataset of N
modelled structures. For limited problems, this is not a major limitation – for example, if observing
intermixing across an epitaxial atomic interface, it is possible to model all possible levels of intermixing.
Second, based on the modelled structures, the resulting CBED patterns should be simulated. These can
then be used to train a classifier, which can be based on some machine learning model (e.g., a deep
convolutional neural network) that takes as input the simulated images, and outputs the structural
descriptors (for example, the degree of intermixing across the interface). Finally, this can then be applied
to experimental data to yield the structure, thus solving the inversion problem.

Naturally, there exist many somewhat mundane but still important issues when comparing the
experimental and simulated datasets for 4D STEM. Apart from obvious ones such as microscope drift
and associated registration difficulties, the fact remains that most simulations are still imperfect
realizations of the actual experimentally observed structure. Microscope corrections (e.g., spatial
decoherence) can be artificially added, but other imperfections, including those of unknown origin, may
exist. We can tackle this problem again using modern tool from computer science and generative
modelling, shown in Figure 1 (b). Here, we have data from two distributions – one from simulated
images, another from experimental images, and the goal is to learn a mapping from one distribution to
the other, essentially learning the transfer function F shown in the figure. This can be achieved using an
adversarial approach, where a discriminator is trained to distinguish between images from the
transformed simulated image data and the experimental data, whilst a generator is trained on minimizing
the ability of the discriminator to perform the task by learning the appropriate transfer function that,
when applied to the simulated images, makes them indistinguishable from the experimental images. To
avoid mode collapse, and since paired examples are not available, this should be done with cycle-
consistency loss within the cycleGAN framework [2] that ensures that a mapping between the two
domains forward and backward results in minimal deviation. We applied this framework to structural
determination for 4D STEM data across the oxide interface between LaAlO3 and SrTiO3 and show that
the lack of this learned transfer function severely limits the ability to distinguish between different
structures in the 4D STEM data.

The second question asks what can be done once the structural details are obtained. Here, we will
discuss our recent work on developing a framework that utilizes principles of statistical physics to derive
lattice Hamiltonians to describe systems based on atomically resolved images in combination with other
macroscopic data (if available). The key insight is that imaging of the system provides a window to the
microstates, and therefore, understanding this distribution enables predictive modelling. By combining
information on local ordering derived from machine learning of local atomic neighbourhoods, it
provides a method to study phase transitions and underlying driving forces of disordered materials.
[3,4].

References:

[1] C. Ophus et al., Microscopy and Microanalysis 20(S3), 62 (2014).

[2] J.Y. Zhu et al., Proc. IEEE Int. Conf. Comp. Vis. 2223 (2017).
[3] L. Vlcek et al., ACS Nano 11, 10313 (2017).
[4] L. Vlcek et al., ACS Nano 13, 718 (2019).
[5] This work was supported by the US Department of Energy (DOE), Office of Science, Materials
Sciences and Engineering Division (RKV, SVK, LV, AL, MO). This work was conducted at the Center
for Nanophase Materials Sciences, a US DOE Office of Science User Facility.

Figure 1. (a) Workflow for structural determination problems utilizing simulations, experiment and
machine learning. (b) Generative adversarial modeling to learn a transfer function that converts
simulated CBED patterns to be closer to those from the experimentally derived distribution.
ATMOSPHERE FOR CATALYSIS
BRINGING YOU CLOSER TO NANOSCALE OPERANDO EXPERIMENTS

VISIT US AT BOOTH #644 VENDOR TUTORIALS @ 5:45 P.M.

Sign up at the MSA MegaBooth
REALISTIC OPERATING ENVIRONMENTS FREE to attend
Accurately blend and flow custom gas mixtures to
re-create even the most complex catalyst operating Tuesday, 6 August
environments. Utilize anything from a single Location: Protochips Booth #644
component gas to a custom industrial exhaust and
In Situ TEM: Creating more Realistic Experimental
add up to 1000ºC of heat.
Environments for Materials Research by Precisely
Controlling the Introduction of Humidity
CONTROLLABLE VAPOR INTRODUCTION Kinga Unocic of ORNL
Software-controlled introduction of water and solvent
vapors enables a level of precision and reproducibility Wednesday, 7 August
otherwise unattainable. Introduce humidity, alcohols, Location: Protochips Booth #644

light alkanes, and other sources as pure vapors or mix

Operando Liquid-electrochemical TEM for
them with a carrier gas while preventing condensation. Monitoring the Charge/Discharge Processes
inside Li/Na-ion Batteries
INTEGRATED GAS ANALYSIS Arnaud Demortiere of CNRS
Integration of the residual gas analyzer (RGA) enables
superior image stability and rapid gas detection
during your experiments. Software-controlled
cleanliness checks ensure safe exposure of the gas
Protochips
uanti ably Better ™
manifold to the RGA for residual gas inspection prior
to and in between experiments.
Revolutionizing Science
Continually breaking boundaries in electron microscopy
Gatan proudly showcases the K3™ IS direct detection camera, the GIF
Continuum™ systems, and the Monarc™ cathodoluminescence (CL)
detector. With these best-in-class systems Gatan continues to set new
benchmarks in electron microscopy.

K3 IS
• Count 1,500 full fps – 3.75x the frame rate of K2
• K3 IS model (1027) – 24 megapixels (5,760 x
4,092)
• K3 Base IS model (1026) – 14 megapixels
(3,456 x 4,092)
• Store 75 fps at full sensor resolution to 1,200
fps with sub area, no binning required

Continuum
• The next generation of advanced systems for
EELS & EFTEM
• >8,000 spectra per second at >95% duty cycle
• >10x faster system tuning
• Streamlined, workflow-based user interface

Monarc
• Acquire CL data with unmatched spatial
(<10 nm), angular (1°), and
wavelength (0.1 nm) resolutions
• Simultaneously capture angle- and
wavelength-resolved CL data
• Collect hyperspectral data up to 30 times
faster than other CL detectors

www.gatan.com