S. Rudinsky,1 M. Gendron,2 N. Piché,2 M. Marsh,2 and R. Gauvin3
1
Steam instruments, 931 E. Main St., Ste. 3, Madison, WI, 53703-2955
Object Research Systems, 101-760 Rue Saint-Paul Ouest, Montreal, QC, H3C 1M4
3McGill University, 3610 University, Montreal, QC, H3A 0C5
2
sam.rudinsky@steaminstruments.com
Abstract: Monte Carlo simulations are commonly used in elemental quantification using energy-dispersive spectroscopy (EDS). Here, the Monte
Carlo program MC X-ray was incorporated into the image processing software Dragonfly by Object Research Systems (ORS) as a simulation library.
The simulation program has been transformed into a complete microscope simulator where the tools of Dragonfly allow complex voxel-based
geometries to be constructed, and the electron beam and detectors can
be freely placed inside the 3D space. Computation times of simulations
have been improved drastically through new data structures and parallelization. Simulations of backscattered electron imaging and EDS mapping
are presented here to demonstrate the capabilities of this new library.
Keywords: scanning electron microscopy, energy-dispersive spectroscopy, segmentation, Monte Carlo simulations, Dragonfly software
Introduction
Monte Carlo simulations are widely used in applications
related to both electron beam imaging and microanalysis [1].
They provide information such as the interaction volume of
incident electrons and X-ray distributions [2], and they are useful in cathodoluminescence emission measurements [3] and
mass thickness estimations [4,5]. They also compute backscattered and secondary electron yields, which are important for
image simulations used to estimate signal variations [6,7] and
secondary election emissions of electrode materials [8]. Finally,
a crucial use of Monte Carlo simulations is to perform accurate
elemental quantification from energy-dispersive spectroscopy
(EDS) data. Matrix correction factors can be computed using
Monte Carlo simulations and then used to calculate accurate
compositional information from X-ray intensities [9–11]. This
allows quantification to be done without the need for standards
or pre-measured experimental databases [10].
The most widely used programs available are PENEPMA,
CASINO, and DTSA-II [12–14]. While these programs have
been well adapted for common uses in electron microscopy and
microanalysis, there exist some limitations to their usability.
The most notable of these are the computation time and restrictions on sample and microscope geometries. For example, simulations of electron numbers on the order of 105 at a single point
may take approximately a minute on a desktop computer [15].
These long computing times can make simulations of backscattered electron (BSE) images or EDS maps unfeasible in a reasonable amount of time. The limitations in terms of geometry
impact the usability of these programs on complex materials.
Important obstacles arising from non-flat sample geometries
can strongly impact accurate elemental quantification [16,17].
To simulate such materials, freedom in terms of sample and
microscope geometry and placement must be possible.
Here, we present the incorporation of the existing Monte
Carlo software MC X-ray [18] into the image processing software
40
doi:10.1017/S1551929520001315
Dragonfly developed by Object Research Systems (ORS) Inc.
[19]. The class system of Dragonfly allows substantial freedom
in terms of sample geometry and placement. Furthermore,
additions have been made to turn MC X-ray into a full microscope simulation tool capable of reproducing some elements of
an electron microscope. Multiprocessing has also been implemented to decrease computing times substantially, allowing
more complex problems to be easily evaluated. The manuscript
is structured as follows. First, the details and usability of the
program are described with reference to previous Monte Carlo
software. Then applications of the library to BSE image simulations and elemental quantification are presented. Finally, some
concluding remarks are made about the method.
Materials and Methods
MC X-ray is a Monte Carlo-based simulation package [18]
that was devised as an extension of CASINO [13,20] and Win
X-ray [21], specific for X-ray microanalysis. Scattering events are
modeled based on a stochastic process where electrons are simulated using a forward scattering random walk. An electron is initiated, and uniform random numbers are generated and used in
cross-sectional models to determine the particle’s path through
the sample material. The physical models, methodology, and calculations used are described in other work [21]. The previously
stand-alone MC X-ray program was incorporated into the image
processing software Dragonfly as a feature plug-in. The base
code was written in C++, and the interface used to gather the
input and call the simulation functions was written in Python 3.
Sample generation is done using the Multi-region of interest (ROI) class inside Dragonfly. A phantom material is created
as a MultiROI where each subgeometry is its own ROI. The
structures are voxel-based where the dimensions of the voxels
are chosen by the user. Once a phantom has been defined, labels
may be assigned to each ROI. Each label contains a property
where, for the purpose of libMCXray, these properties represent
the constituent elements of each region. The labels are vectors
whose lengths are the number of regions and whose values are
the composition of the elements in the indexed region. Thus,
a material may be generated with a multitude of elements in a
variety of regions with varying compositions. MultiROIs can
be created by the user or generated from other image data such
as stacks of images, which can be compiled into a 3D structure.
The simulation environment comprises an electron beam,
a sample material, and a series of detectors. The electron beam
is above the sample material normal to its surface. The user
input parameters of the beam are the incident electron energy,
incident current, and working distance. The rasterizing across
the sample surface is done by displacing the beam across the
www.microscopy-today.com • 2020 September
Downloaded from https://www.cambridge.org/core. IP address: 3.80.250.168, on 27 Jun 2021 at 02:35:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1551929520001315
libMCXray: A Monte Carlo Simulator for Signal Analysis
inside Dragonfly Software
Table 1: EDS detector parameters modifiable by user in
libMCXray.
Table 2: Compositions of the simulated system assigned to
the MultiROI of Figure 1.
Input properties for simulated EDS detector
Phase number
Element
Color in Figure 1
Crystal type (Al or Si)
1
Ni
Transparent
Crystal elemental density
2
Fe
Red
Crystal dimensions (radius and thickness)
3
Cr
Blue
Dead layer thickness
Diffusion length
Detector efficiency
Take-off angle of detector
Position of detector in space
Number of energy channels
possible transmitted signal images are output as 2D arrays of
floating-point real numbers. The EDS maps are output as 3D
arrays where the first two dimensions are the point coordinates
and the third dimension is the energy spectrum. Finally, if
point analyses are chosen, the X-ray depth distribution curves,
ϕ(ρz), of specified transitions may be output.
Results and Discussion
sample surface at steps specified by the user input resolution.
Once the sample material is constructed, it is placed so that the
beam and the sample are along the same axis. Their vertical
distance is specified by the desired working distance.
Detectors may then be added to the simulator. The BSE
detector may be defined as a disc or circular annulus below
the electron gun. The input in this case is the inner and outer
radius of the annulus or the radius of the disc. BSEs are defined
as those which have exited the sample and intersected the
detector. The EDS detector is defined by the properties specified in Table 1 and are similar to the stand-alone version of
MC X-ray. Multiple detectors can be added to the simulator,
each with their own properties. Last, bright-field and darkfield detectors can be added below the sample. Each is defined
by their distance from the exit surface of the sample and their
inner and outer solid angles. Again, intensities of either signal
are constructed from the number of electrons intersecting the
respective detectors upon exit from the sample.
The program output is separated into two parts: electron
and X-ray data. From the rasterizing, the BSE image and any
BSE image simulation. A MultiROI of a three-phase alloy
compiled from a focused ion beam image dataset was used to
simulate backscattered images. A 3D perspective of the MultiROI is depicted in Figure 1. Elemental compositions were
assigned to the phases prior to performing the Monte Carlo
simulation. A Ni-Fe-Cr system was investigated where each
element was associated to a pure single phase. The phases to
which each element is associated are presented in Table 2. The
incident probe position was rasterized across the top surface of
the sample shown in Figure 1. The sample plane at the incident
direction is displayed in Figure 2.
Simulations were performed at 15 keV using 5,000 electrons. Because the simulations of each probe position are independent, the method is easily parallelized, allowing for an
acceleration of 80 to 100 times the computing time of the original program. The simulated BSE image corresponding to a subsection of the incident plane of Figure 2 is presented in Figure 3.
The image consists of 512 × 512 pixels situated about the
center of the plane. This area is indicated in Figure 2 by a dashed
rectangle. Here, the three pure element phases are clearly distinguished with the fine structures from the underlying geometry clearly visible. The differences in signal level clearly reflect
Figure 1: 3D representation of sample geometry used in simulation experiment. Phase 1 is the matrix, phase 2 is in red, and phase 3 is in blue. The matrix
phase is transparent.
2020 September • www.microscopy-today.com
Figure 2: View of simulated material from beam incident direction. The area
across which the beam was rasterized is indicated by a dashed rectangle.
41
Downloaded from https://www.cambridge.org/core. IP address: 3.80.250.168, on 27 Jun 2021 at 02:35:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1551929520001315
Monte Carlo Simulator
Figure 3: Simulated BSE image of three-component Ni-Fe-Cr system at 15
keV using libMCXray.
the differences in atomic number between each phase. The
matrix is composed of pure Ni, the heaviest of the three elements, whose signal level is the highest. In comparison, the Cr
phase, contained in the majority of the needle-like structures, is
seen as the phase containing the lowest level of signal, consisting of points attached to the Fe phase and the large rectangular
structure in the bottom right corner of the image.
Figure 4: Simulated BSE image of Al-Si alloy containing three phases: pure
Al, Al2Cu, and Mg2Si.
42
Figure 5: Three-dimensional depiction of the EDS dataset simulated using
libMCXray. The image represents spatial dimensions of 512 × 512 pixels and
2048 energy bins.
Another system was simulated consisting of the expected
phases of an Al-Si alloy to show the versatility of the method
in simulating alloys closer to reality. Figure 4 shows a simulated BSE image of the same geometric structure, however here
the phases consisted of pure Al as the matrix, phase 1, Al2Cu
as phase 2, and Mg2Si as phase 3. Simulations were performed
under the same conditions as the Ni-Fe-Cr system.
Because Cu has a much higher atomic number than the
other constituent elements, and hence a higher probability
of scattering, the signal level of the Al 2Cu phase is the highest in the image. In comparison, the Mg 2Si phase contains
the lowest signal level due to its lower probability of scattering. This demonstrates that even multi-component phases
may properly and easily be simulated using libMCXray, and
the attributes of the Dragonfly software aid in proper image
analysis.
EDS spectrum analysis and mapping. EDS maps were
simulated for the above system using the same MultiROI and
positions. Here, 2048 channels were used across the energy
scale. Figure 5 shows the entire 3D dataset acquired using
libMCXray, two spatial dimensions and the third in energy.
The EDS simulation was performed on the Amazon AWS highperformance computing (HPC) network and took 1 hour on
500 computers each containing 8 cores. On an extremely performant workstation, such a simulation would take approximately 200 hours. Computing times are rendered much more
reasonable by the use of cluster computing.
www.microscopy-today.com • 2020 September
Downloaded from https://www.cambridge.org/core. IP address: 3.80.250.168, on 27 Jun 2021 at 02:35:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1551929520001315
Monte Carlo Simulator
Figure 6: Simulated EDS maps for (a) Ni Kα, (b) Fe Kα, and (c) Cr Kα transitions. These correspond to 7.477, 6.403, and 5.414 keV respectively.
Figure 7: (a) BSE image showing the point at which an energy spectrum was taken and (b) corresponding energy spectrum showing peaks from Ni, Fe, and Cr. The
labels are placed above the Kα peaks.
Slices of the dataset were taken at the corresponding Kα
transition energies for each element. These slices are depicted
as 2D maps in Figure 6.
The simulated EDS maps correspond exactly to the BSE
image and simulated geometry. Areas of white imply areas of
high intensity, while black describes areas of low to no intensity. The simulated spectrum at a point is depicted in Figure
7 along with the BSE image showing the point at which the
spectrum was taken.
At this point, contributions from all three phases are present. Although these results are all simulated, the accuracy of
MC X-ray simulations for K-shell ionizations has been demonstrated in previous work [22–25].
Conclusions
The Monte Carlo simulation program MC X-ray was
incorporated into the Dragonfly software package by ORS
2020 September • www.microscopy-today.com
to simulate BSE imaging and EDS analysis inside an allencompassing image processing software. Using parallelization and good software practice, the calculation speed was
improved by ∼80–100 times the original code. Furthermore,
using the Dragonfly interface, complex geometries may be
created or even imported from datasets such as those generated in focused ion-beam experiments. Future work will see
the addition of secondary electron detection and the incorporation of higher-order ionization effects such as secondary
fluorescence yields.
References
[1]
[2]
[3]
[4]
[5]
CGH Walker et al., Scanning 38 (2016) 802–18.
A Morozov et al., Eur Phys J D 48 (2008) 383–88.
M Toth and MR Phillips, Scanning 20 (1998) 425–32.
D Moro et al., Measurement 129 (2018) 211–17.
A Pazzaglia et al., Mater Charact 153 (2019) 92–102.
43
Downloaded from https://www.cambridge.org/core. IP address: 3.80.250.168, on 27 Jun 2021 at 02:35:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1551929520001315
Monte Carlo Simulator
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
AH Mir et al., Micron 126 (2019) 102712.
MHN Yio et al., Cement Concrete Res 89 (2016) 320–31.
HY Chang et al., Nucl Instrum Meth B 454 (2019) 14–22.
J Donovan et al., Microsc Microanal 25(3) (2019) 735–42.
P Horny et al., Microsc Microanal 16(11) (2010) 821–30.
N Brodusch and R Gauvin, J Microsc 267(3) (2017) 288–98.
X Llovet and F Salvat, Microsc Microanal 23(3) (2017)
634–46.
P Hovington et al., Scanning 19(1) (1997) 1–14.
NWM Ritchie, Microscopy Today 19(1) (2011) 26–31.
NWM Ritchie, Microsc Microanal 15(5) (2009) 454–68.
DE Newbury and NWM Ritchie, Scanning 35(3) (2013)
141–68.
DE Newbury, Scanning 26(3) (2004) 103–14.
R Gauvin and P Michaud, Microsc Microanal 15(Suppl 2)
(2009) 488–89.
Object Research Systems (ORS) Inc., Dragonfly 3.6. Montreal, Canada, 2018. Software available at http://www.
theobjects.com/dragonfly.
D Drouin et al., Scanning 29(3) (2007) 92–101.
R Gauvin et al., Microsc Microanal 12(1) (2006) 49–64.
R Wuhrer and K Moran, IOP Conf Ser–Mat Sci 55 (2014)
012021.
MJ-F Guinel et al., J Microsc 255(3) (2014) 128–37.
N Brodusch et al., Microsc Microanal 23(Suppl 1) (2017)
236–37.
C Teng et al., Microsc Microanal 23(Suppl 1) (2017) 1046–47.
MINITOOL
Precision Micro-Tools
Efficient, precise and realistically proportioned
instruments are ideal for microscopists. Our
line of micro-tools includes needles, gravers,
chisels, knives, hooks and mirrors, probes,
spatulas, scribes and microrulers.
All 32 tools are offered singly or in sets of eight
tools with handles. Available in tip diameters
from .025mm to 1.00mm.
www.minitoolinc.com
info@minitoolinc.com
MEO Engineering Company, Inc.
MEO Engineering Company, Inc.
High Technology on a Small Scale since 2004
High Technology on a Small Scale since 2004
Consumables & Service
Deposition Gas Injector
FIB or SEM - any OEM
FIB or SEM - any OEM
GALEX Instruments® D-GIS
Precursor Gas Injection Solution
Reduce maintenance costs with
PBS&T™ components:
• Extractors
• Suppressors
• Apertures
• Precursor Refills
• Four materials on the same port
• Easy to install, align and operate
• User-exchangeable precursor cartridges
• Designed for safety and compliance
• Standard precursors: Pt, W, Mo, C, SiOx
• Custom chemistries available
• Gas assisted etching version is available
• UHV-compatible version upon request
• Application development support
• Etching (XeF2, Bromine, Iodine)
• Deposition (C, Pt, W, Mo, Cu, SiOx)
• Custom Chemistries
We can support you with:
• DIY maintenance training
• On-site service and repairs
• Installation and relocation
• High voltage power supplies
• Ion and electron optics
• Preventive maintenance
• Custom-made components
• Aftermarket upgrades
• Process development
• Remote support
www.fibsemproducts.com
info@partbeamsystech.com
44
Carbon deposition on porous
thin film substrate
FIB-deposited Pt
E-beam Pt deposition within
alignment and tracking notches,
and over sample
Pt FIB and SEM deposition
for 3D slice-and-view tomography.
ZEISS Crossbeam 340 with
FIBICS Atlas Software and
GALEX Instruments® gas injection
system (D-GIS)
www.fibsemproducts.com
info@partbeamsystech.com
www.microscopy-today.com • 2020 September
Downloaded from https://www.cambridge.org/core. IP address: 3.80.250.168, on 27 Jun 2021 at 02:35:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1551929520001315
Monte Carlo Simulator
PNDetector’s SDDs are at the forefront of radiation detector technology, as they combine both excellent energy
resolution and short processing times. We are fabricating in our own cleanroom facility dedicated to the high quality
fabrication of modern silicon radiation detectors and is optimized to permit a continuous and efficient production flow.
High Resolution Silicon Drift Detector with superior performance
9 Up to 1 Mcps for single SDDs and several Mcps with our Multi-Element SDDs
9 Best energy resolution down to 121 eV
9 Best light element detection
9 P/B ration up to 20.000
9 Ultralow leakage current < 100 pA/cm2
9 Radiation hardness > 1014 photons/cm2
9 Wide selection of chip sizes and detector housings
Backscattered Electron
Detector
pnCCD 4D (S)TEM Camera
is designed to fulfil any customers wish from regular
needs up to highly specific
demands like high speed
imaging or low noise signal
detection at a broad range
of electron energies.
is a pixelated, ultrafast (up
to 7.500 fps), and direct
detecting electron sensor for
applications in TEM and SEM.
Thanks to its outstanding
signal to noise ratio, single
primary electron detection from
20 keV to 300 keV is possible.
Contact Information
PNDetector
Otto-Hahn-Ring 6
81739 Munich Germany
Tel +49 89 309047-100
Mail sales@pndetector.de
www.pndetector.de
Downloaded from https://www.cambridge.org/core. IP address: 3.80.250.168, on 27 Jun 2021 at 02:35:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1551929520001315
From Wafer to Chip ... to Module ... to System