Content uploaded by Bert Freitag
Author content
All content in this area was uploaded by Bert Freitag on Oct 24, 2016
Content may be subject to copyright.
This proprietary EDX design is called the Super-
X
TM
system.
Figure 2 shows the results of an analysis time
benchmark test: The Tecnai Osiris 200 kV S/TEM
(with ChemiSTEM technology) [3] was com-
pared to a conventional 200 kV S/TEM using
the same sample. The conventional S/TEM was
equipped with a Schottky FEG and Si(Li) detec-
tor at 0.3 sr (steradian) X-ray collection angle.
The acquisition times were adjusted to achieve
comparable X-ray statistics for measurements
on both systems. For equivalent EDX map sizes
and statistics, the results took just under 2
hours on the conventional system and just
under 2 minutes on the ChemiSTEM system, in
other words from ‘hours to minutes’. This
demonstrates the magnitude of the new tech-
nology’s performance advantage, that can be
used as a raw speed advantage, but, perhaps
more significantly, can be used to obtain results
that were previously unobtainable, such as
detection of 0.02 wt.% elemental concentra-
tions, identification of a catalyst core-shell
structure in 4 minutes, and fully quantified
600⫻600 pixel elemental maps in about an
hour with a pixel resolution of 3 Å and yielding
never-before-seen levels of light element
detection such as C, O, and N. These examples
are shown later in this article, but first we give
a more detailed description of the benefits and
advantages of the new technology.
The Super-X Advantage in Tilt Response
A major advantage of the Super-X design
comes from the large solid angle for X-ray col-
lection provided by four SDD detectors sym-
metrically arranged around the specimen. But
there is also an important advantage related to
specimen tilting. Figure 3 compares measured
X-ray count rates over a tilt range from –25° to
+25°, for a Super-X system with 0.9 sr solid
angle (red curve) and for a system with single
Si(Li) detector with 0.3 sr solid angle (blue
curve). The count rates were measured at 200
B I O G R A P H Y
Peter Schlossmacher
studied physics at the
University of Cologne,
Germany, with an
emphasis on solid state
physics. In his PhD thesis
at the Research Center
Jülich he investigated crystal defects in
GaAs. Subsequently he moved to the Insti-
tute of Materials Research at the Research
Center in Karlsruhe, Germany, as staff scien-
tist where he headed the TEM laboratory.
For the last 8 years he has worked for elec-
tron microscopy manufacturers and is cur-
rently Product Marketing Manager in the
NanoResearch Market Division at FEI Com-
pany, responsible for TEM products with
special focus on the Tecnai platform.
A B S T R A C T
A new technology for energy dispersive X-
ray spectroscopy (EDX) in the scanning/
transmission electron microscope (S/TEM) is
described that allows orders-of-magnitude
improvement in key analytical metrics, such
as sensitivity and time-to-data. This technol-
ogy incorporates a four SDD detector system
highly optimized for light element detec-
tion, collection efficiency, and improved tilt
response. We believe this new technology
will enable the S/TEM to answer many scien-
tific and engineering questions that were
previously out of reach and will lead to a
renaissance of interest in using EDX as a
prime tool in S/TEM microscopy and analyt-
ics. Here we describe the characteristics of
the new EDX technology and illustrate its
performance for the chemical analysis of
several application challenges each requir-
ing sensitivity and speed previously unatt-
ainable.
K E Y W O R D S
scanning transmission electron microscopy,
energy-dispersive X-ray microanalysis,
ChemiSTEM, Super-X, silicon drift detector,
nanostructures
AUTHOR DETAILS
Peter Schlossmacher,
FEI Company,
Achtseweg Noord 5,
5600 KA Eindhoven, The Netherlands
Email: peter.schlossmacher@fei.com
Microscopy and Analysis 24(7):S5-S8 (EU), 2010
S/ T E M- E D X S y S T E M
I N T R O D U C T I O N
ChemiSTEM Technology: A Revolutionary
Advance in X-ray Analytics
Information about a material’s chemical com-
position on the nanoscale is pivotal for an
understanding of nanostructures and devices,
and X-ray spectroscopy is a well-established,
robust and easy-to-use technique to obtain this
information. Better yet, the scanning/transmis-
sion electron microscope (S/TEM) can pair
microstructural information obtained from
high-resolution imaging with accurate chemi-
cal composition information. But there has
been an historical limitation with X-ray systems
for the S/TEM: they collect only about 1% or
less of all X-rays generated by the electron
beam passing through the very thin sample.
Recently, as the desired spatial resolution of
chemical analysis has become progressively
greater, the X-ray signal has decreased due to
fewer atoms excited in smaller analytical vol-
umes (a consequence of smaller electron
beams and thinner samples). This results in low
signal strength, bringing about low sensitivity
and hence long analysis times – until now.
ChemiSTEM
TM
technology has been devel-
oped at FEI over a 5-year period with the
express purpose of removing the old barriers in
performance, bringing orders-of-magnitude
improvement in many key analytical metrics,
such as sensitivity and speed. This reduces
analysis times from hours to minutes, and sam-
ple features and elements that previously
remained hidden can now be detected easily.
With this new technology, we believe X-ray
analytics in the S/TEM is poised to experience a
renaissance based on this new capability to
answer scientific questions that have until
today remained unanswered. We begin with
an explanation of how this new proprietary
technology is able to provide such break-
through capabilities.
A key performance metric of energy-disper-
sive X-ray spectroscopy (EDX) is the net mea-
sured X-ray count rate which depends on the
count generation rate (set by the beam cur-
rent) and the collection efficiency (set by the
detector system). The ChemiSTEM system
design is shown in Figure 1; it addresses both
needs: more beam current and more collection
efficiency of X-rays. The higher beam current is
provided by the proprietary X-FEG Schottky
electron source. This FEI high-brightness elec-
tron source can generate up to 5 times more
beam current at a given spatial resolution com-
pared to a conventional Schottky FEG source.
The higher efficiency detection system [1,2] is a
radically new design concept: it integrates four
FEI-designed silicon drift detectors (SDDs) very
close to the sample area. These detectors are
windowless to further boost collection effi-
ciency and light element detection capability.
Nanoscale Chemical CompositionalAnalysis
with an Innovative S/TEM-EDX System
Peter Schlossmacher, Dmitri O. Klenov, Bert Freitag, Sebastian von Harrach, and Andy Steinbach
FEI Company, Eindhoven, The Netherlands
Figure 1:
A schematic of the ChemiSTEM™design, showing the X-FEG high-
brightness, Schottky electron source, and the Super-X™ geometry
including four silicon drift detectors arranged symmetrically around the
sample and the objective lens pole pieces. This schematic is not to scale.
MICROSCOPY AND ANALYSIS NANOTECHNOLOGY SUPPLEMENT NOV E M B E R 2 0 1 0 S5
kV using standard commercially available NiOx
test films [4] with nominal thicknesses of
50⫾10 nm and integrated over the entire
energy range. The count rate is normalized to
unity at the maximum response of the Super-X
system, occurring at zero tilt angle. The Super-
X response never drops below 80% of the max-
imum count rate over the entire tilt range. By
contrast, the single detector system only
achieves its maximum count rate when the
sample is tilted strongly to +20° towards the
single detector. At zero tilt, the count rate is
already reduced by 30% or more due to geo-
metrical shadowing of the single detector by
the sample holder, and at negative tilt angles,
complete shadowing occurs at -10° where the
response drops to zero. All one detector sys-
tems suffer from a similar undesirable tilt
response, and it occurs equally for SDD or Si(Li)
detectors.
The ability to achieve high X-ray signal over
the whole S/TEM tilt space is a key performance
improvement of the new system. Many studies
in material science or chemistry require tilting
to angles that cannot be easily controlled a pri-
ori due to the unknown orientation of
nanoparticles or grain boundaries in poly-crys-
talline material. In these cases, the sample can
be tilted to optimize imaging conditions and
then does not need to be readjusted for opti-
mum EDX analysis conditions. Conversely, for
very many samples, an important crystal orien-
tation to be observed is prepared to be very
close to zero degrees tilt. For such samples, it is
a key advantage to have the maximum X-ray
response at zero tilt, as opposed to the ~30%
reduction in response single detector systems
experience due to detector shadowing at zero
tilt. In these cases, it may not be feasible to tilt
the sample towards the single detector to max-
imize response, due to the need to keep the
beam parallel to a material interface (for an
edge-on view of the interface).
Super-X Advantage in Total Response (1+1+1+1>4)
By simple logic, one expects the Super-X sys-
tem with four detectors to be ‘4 times better’ in
response than a one detector system (i.e.,
1+1+1+1 = 4). However, this logic does not
hold, because each individual Super-X SDD
detector has significant design improvements
in and of itself, compared to conventional indi-
vidual SDD detector systems. So the response of
each Super-X detector is ‘> 1’. The question of
how much greater leads us to consider two
major design improvements of the individual
detectors. First, as described earlier, since the
individual Super-X SDD detectors were
designed to be optimized for the FEI objective
lens pole piece design (shape, size, geometry,
etc.), we were able to completely eliminate the
problem of detector shadowing at zero tilt
angle. This means that each detector sits at a
‘take-off’ angle and location so that the detec-
tor is fully illuminated at zero tilt, with no shad-
owing loss. Achieving this required the design
of a special sample holder for the Super-X sys-
tem, as well as an optimized detector design.
Second, the Super-X detectors are window-
less by design. The detectors instead have
mechanical shutters providing complete pro-
tection when they are not in live mode. This
choice eliminates the need for ultra-thin poly-
mer windows that are the source of two severe
response losses for systems that do employ
them. Figure 4 illustrates the effect of windows
on X-ray collection efficiency, and there are
two key effects to understand, which are illus-
trated by the diagram in Figure 4a. First, the
polymer film [5] is supported by a silicon sup-
port grid, which blocks all X-rays hitting it and
this results in a substantial X-ray response loss
for all energies. Second, the polymer windows
themselves, although transparent to X-rays >1
keV in energy, do have substantial absorption
loss below 1 keV , particularly due to the car-
bon absorption edge around 300 eV, as seen in
Figure 4b, which plots X-ray transmission effi-
ciency for an SDD detector both with and with-
out polymer windows. This figure shows that
the efficiency loss due to windows (lower blue
curve) is significant over the entire energy
range (due to the polymer support grid), com-
pared to a windowless system (upper red
curve). The remaining loss in the windowless
system below 1 keV is due to residual absorp-
tion by the thin Al electrode and Si dead layer
at the front of the SDD detector. The window-
less system has a factor of at least 2-3 greater
transmission efficiency at energies below 500
eV. This improvement in low energy X-ray
transmission has a huge impact on light ele-
ment detection, as we will discuss later.
How to Compare Collection Efficiencies of
Different EDX Systems
The traditional S/TEM metric for judging EDX
collection efficiency is to use the ‘nominal’ solid
angle. By this, we mean the pure geometrical
solid angle subtended by the detector, viewed
from the eucentric sample point, and based on
only the detector’s cross-sectional area and the
distance from the eucentric point. Problemati-
cally, this does not include signal loss from
effects such as detector shadowing or win-
dows, which make quite a difference, as we’ve
seen. Sometimes non-active detector areas
such as metallic guard rings will be included
when calculating nominal solid angle. True
efficiencies thus depend on loss factors based
on often unknown parameters such as sample
holder geometry or detector elevation angle,
etc., and these parameters are not realistically
obtainable by users. We propose that the most
accurate way to compare EDX system efficiency
is by a method that involves measuring actual
X-ray count rates per applied beam current
using a standard sample of known thickness.
We have applied this comparison methodol-
ogy and the results are shown in Figure 5. Input
counts per second (cps) are plotted as a func-
tion of applied beam current for two S/TEM sys-
tems: A Tecnai Osiris with ChemiSTEM technol-
ogy and a 200 kV S/TEM with Schottky FEG and
Si(Li) detector of 0.3 sr nominal solid angle. The
small inset shows the same data on a smaller
scale closer to the origin. For both measure-
Figure 2: (a) 100x100 pixel EDX maps of a
semiconductor device taken on a Tecnai TF20 XT
(0.3 sr Si(Li) system) with 500 ms pixel
-1
dwell
times, ~0.7 nm spot size, 0.4 nA beam current,
and 1h 54min total map acquisition time.
Sample courtesy of NXP Research; maps by D.
Klenov, A. Carlsson, FEI. (b) 100x100 pixel EDX
maps of an equivalent semiconductor taken on
a Tecnai Osiris with ChemiSTEM technology with
5 ms pixel
-1
dwell times, ~0.3 nm spot size, 1.0
nA beam current, and 115s mapping time.
MICROSCOPY AND ANALYSIS NANOTECHNOLOGY SUPPLEMENT NOVEMBER 2010
S6
Figure 3:
Comparison of relative EDX count rates
of the Super-X system (on a Tecnai Osiris
with 0.9 sr collection angle) and a single
Si(Li) detector system with 0.3 sr nominal
solid angle. Both S/TEMs were operated
at 200 kV with the same (constant) beam
current. NiOx films were used as samples
for both tilt series. Positive tilt angles rep-
resent specimen tilts towards the single
detector for the Si(Li) system. Diagrams
above the graph show the effects of
detector shadowing for the four Super-X
detectors, and diagrams below show
shadowing effects for the single detector
system.
ab
S/ T E M- E D X S ySTEM
ments, the same FIB-cut InP sample was mea-
sured and the X-ray counts were integrated
over the full energy range. In order to be sure
all measurements were taken on sites of the
same thickness, EELS was employed to control
sample thickness by measuring t/λ(thickness in
relation to average mean free path). The first
observation from this figure is that the ChemiS-
TEM system on the Tecnai Osiris achieves more
than 5 times the X-ray count rate per unit
applied beam current than the conventional
system, even though this system is an 0.3 sr
nominal solid angle EDX system, which is con-
sidered at the high end of collection perfor-
mance for conventional detectors. In real
experiments, we typically find a 5-10 times col-
lection advantage of the Super-X system com-
pared to single-detector systems. The second
observation from Figure 5 is that more than 10
times higher beam currents can be applied for
the ChemiSTEM system with the X-FEG electron
gun, without high dead-time % saturation of
the new Super-X detector system. At an
applied beam current of 10 nA the input X-ray
count rate is over 400,000 cps at approximately
a 50% dead time of the Super-X detector sys-
tem, meaning that the output X-ray count rate
is over 200,000 cps at this value of applied cur-
rent. The Super-X system has significantly
higher input bandwidth for high count rates
compared to systems with only one high-speed
SDD detector, and thus the Super-X system can
operate at 400,000 cps input count rate with-
out significant degradation of energy resolu-
tion, which is not the case for systems with only
one high-speed SDD detector. For example, in
benchmark tests performed at 200,000 output
cps the Super-X system achieved typical energy
resolution of 136 eV (@ Mn K␣), whereas other
single SDD systems showed typical energy res-
olutions of 160 eV (Mn K␣) or higher at the
same count rate.
Light Element Sensitivity
Poor detection of light elements (such as C, O,
and N) has always been considered one of the
big weaknesses of EDX as a technique, but
ChemiSTEM technology removes this barrier to
light element sensitivity. As shown in figure 4,
the relative sensitivity of the windowless sys-
tem for low energy X-rays in 500 eV or lower
range is several times higher than systems with
ultra-thin polymer windows. When this low-
energy advantage is combined with the
energy-independent five to ten times collec-
tion advantage of the Super-X system, it leads
to an order-of-magnitude boost in light ele-
ment detection performance.
In Figure 6 we show measured data to
demonstrate that the enhanced low energy
performance shown Figure 4 is realized in the
Super-X system. The figure shows a spectra
from a NiOx film measured on the windowless
Super-X system and a conventional EDX system
with ultra-thin polymer windows. The count
rates of both spectra have been normalized
(set equal) at the Ni-K line at 7.4 keV in order
to demonstrate only the relative differences in
low energy efficiency. One can then see the
advantage of the windowless system by noting
the higher counts in the Ni-L, O-K, and C-K lines
below 1 keV. The relative peak height is more
than two times greater for oxygen at 528 eV
and more than three times greater for carbon
Figure 4:
(a) Schematic showing loss due to holder shadowing and detector window. Loss due to the window includes total absorption by the Si support grid
bars (at all energies) and selective absorption by the polymer window (at energies below 1 keV). (b) X-ray transmission efficiency versus energy for a
windowless SDD detector (red curve) and an SDD detector with thin polymer window (blue curve). Loss due to the both the Si grid bars and the poly-
mer window contributes to the lower efficiency across all energies of the detector with window.
MICROSCOPY AND ANALYSIS NANOTECHNOLOGY SUPPLEMENT NOV E M B E R 2 0 1 0 S7
Figure 5 (left):
Input count rate for ChemiSTEM tech-
nology (red curve) compared to stan-
dard technology consisting of Schot-
tky FEG and 0.3 sr Si(Li) detector (blue
curve). For details see text.
Figure 6 (below):
Comparison of a single windowless
SDD with a standard Si(Li) detector
with ultra-thin polymer window. The
two spectra have been normalized to
each other at the Ni-K line so that rel-
ative differences in the energy range
below 1 keV are apparent.
at 282 eV than for the system with polymer
windows. We stress that this comparison does
not include the primary Super-X collection effi-
ciency increase (which is energy independent)
since this has been removed in Figure 6 by the
normalization. Dramatic examples of the
improved light element detection are found in
the 45 nm PMOS example in the next section
ChemiSTEM Application Examples: Fast
Mapping at High Sensitivities
Figure 7 shows a ChemiSTEM 600⫻600 pixel
EDX map of a 45 nm PMOS transistor structure.
Semiconductor devices represent a demanding
application for chemical analysis because they
contain both heavy elements like tungsten (W),
hafnium (Hf), and tantalum (Ta), as well as light
elements like nitrogen (N), oxygen (O), and
sometimes even carbon (C), and measuring the
accurate quantitative distribution of all these
elements is necessary for device design and for
monitoring process success and quality. With
the increased sensitivity for light element
detection, now virtually all element of the peri-
odic system can be detected in one EDX map
acquisition. The map spans a large field-of-
view of about 190 nm and a pixel resolution of
~0.3 nm pixel
-1
. This map data was acquired at
a spectrum imaging rate of 20,000 spectra s
-1
(50 µs pixel
-1
dwell time) and 1 nA beam cur-
rent with drift correction applied to acquire
multiple frames in 100 minutes. The short pixel
dwell times of 50 µs were necessary in this case
to avoid sample damage (such as changing the
local distribution of carbon in the sample). The
fast electronics of the Super-X system permits
pixel dwell times as low as 10 µs (allowing spec-
tral EDX rates of almost 100,000 spectra s
-1
).
The total acquisition time of 100 minutes was
chosen to obtain good X-ray count statistics
permitting full quantification of each pixel
(without any binning). This high field-of-view,
high spatial resolution, high statistics data set
contains a huge amount of information and is
only possible in a reasonable time due to the
high sensitivity and high-brightness source of
the ChemiSTEM system. We estimate that such
a map would require more than three full days
of mapping time on a conventional system,
which is not practically possible to due instru-
ment drift and practical throughput concerns.
The nitrogen, oxygen, and carbon maps seen in
Memo: Line spacing
ChemiSTEM technology. This figure shows the
EDX spectrum of a certified NIST steel standard
(Standard Reference Material NBS No. 461). In
bulk, this low-alloy steel has certified concen-
trations of the following elements: arsenic
(0.028 wt.%), vanadium (0.024 wt.%), and tin
(0.022 wt.%). Although this sample was pre-
pared for TEM analysis using an FIB, it is
expected that minimal changes have resulted
in the composition. The full spectrum in Figure
10 was acquired in 600 seconds using a beam
current of 1.7 nA while scanning a micron-sized
area in order to average the composition over
the microstructure of the steel.
The three smaller zoomed spectra at the bot-
tom of Figure 10 show clear peaks from these
3 elements (V, As, and Sn) all of which have sub-
stantially good signal-to-noise ratio (Ga and Pt
peaks visible are due to the FIB preparation
process). This clearly demonstrates that the
ChemiSTEM technology on Tecnai Osiris is
capable of detecting low concentrations such
as the ~0.02 wt.% concentrations in this certi-
fied NIST steel sample, obtained in a reason-
able 10 minutes total acquisition time.
C O N CLU S I O NS
Whether the application calls for using the ulti-
mate speed benefits of ChemiSTEM technology
(mapping in ‘hours to minutes’, or huge area
maps with unprecedented sensitivity in rea-
sonable times), or the pursuit of ultimate sen-
sitivity (pushing to detection of ultra-low con-
centrations or hidden sample features) we
believe ChemiSTEM will give the scientific com-
munity the capability to answer many ques-
tions that were until now unresolved and will
lead to a renaissance of interest in using EDX
technology as a prime tool in S/TEM microscopy
and analytics.
R E F E R E N C E S
1. von Harrach, H. S. et al. Microscopy and Microanalysis 15
(Suppl 2):208, 2009.
2. von Harrach, H. S. et al. Proceedings of EMAG 2009.
3. www.fei.com/products/transmission-electron-
microscopes/tecnai.aspx
4. Commercially available from Agar or Pelco based on R. F.
Egerten and S. C. Cheng. Ultramicroscopy 55:43, 1994.
5. www.moxtek.com/x-ray-windows/ap3-ultra-thin-polymer-
windows.html2.
©2010 John Wiley & Sons, Ltd
Figure 7 show that fast mapping of light ele-
ments is possible in reasonable times, which
really represents a breakthrough in capability
for the technique of analytical S/TEM-EDX.
Additional elemental maps from this same
sample are shown on the cover of this Nano-
technology Supplement.
In fuel cell technology Pt nanoparticles are
used as catalysts for large scale applications.
The catalytic efficiency and stability of Pt can
be increased by adding Au and Fe to the
nanoparticles. Such nanoparticles produced by
coating Au particles with Pt(Fe) were investi-
gated, and a key materials composition ques-
tion is: Do the particles form a core-shell struc-
ture or not? With conventional EDX technol-
ogy (upper Au and Pt maps in Figure 8) there is
only an indication that particles may consist of
a Au core surrounded by Pt. Tecnai Osiris (lower
maps) however clearly reveals the core-shell
structure with three times better pixel resolu-
tion and 60 times faster. Due to the high speed
and sensitivity of ChemiSTEM technology the
core-shell structure of many nanoparticles can
be observed in a single, fast EDX experiment, as
shown in Figure 9. The lower Pt concentration
in the center of all particles due to their core
shell structure can be clearly seen, as well as
the presence of Fe in low concentrations, also
distributed in a shell geometry (the blue color
in the composite map in Figure 9).
Ultimate Sensitivity: The Detection of Low
Concentrations
The improved performance of this new tech-
nology can be turned into a huge speed bene-
fit (‘hours to minutes’) as we have demon-
strated, but the improved performance can
also be used to achieve much higher sensitivi-
ties than were previously possible. This benefit
is ultimately perhaps more important than
speed, because it allows the detection of hid-
den features and elements that previously
could not be seen. The ultimate test of EDX raw
sensitivity is the ability to detect very low con-
centrations of elements, for example, well
below 1 wt.%.
In Figure 10 we show the results of an exper-
iment demonstrating the sensitivity limits of
MICROSCOPY AND ANALYSIS NANOTECHNOLOGY SUPPLEMENT NOVEMBER 2010
S8
Figure 7:
600x600 pixel maps of a 45 nm PMOS transistor structure recorded with
50 µs pixel
-1
dwell time and 1 nA beam current. Drift correction was
applied to acquire multiple frames in 100 minutes. The maps were fully
quantified to eliminate contributions from overlapping peaks. Data cour-
tesy of D. Klenov, FEI.
Figure 8:
Upper Au and Pt maps were recorded with a Tecnai TF20ST equipped with
a standard Si(Li) detector as 64x64 pixel maps in 3 hours. The lower maps
were acquired using ChemiSTEM technology with a Tecnai Osiris: 200x200
pixel maps in 3 minutes. Samples and TF20 data courtesy of C. Wang, V.
Stamenkovic, N. Markovic, N. Zaluzec (Argonne National Laboratory). Tec-
nai Osiris experiments by D. Klenov (FEI) and N. Zaluzec.
Figure 10:
Top: Super-X spectrum of NIST steel standard NBS No.461 (log scale). Bottom: Zoomed spectra showing minor elements of vanadium (0.024 wt.%),
arsenic (0.028 wt.%) and tin (0.022 wt.%). Ga and Pt peaks are due to FIB preparation.
Figure 9:
Elemental maps at 300 x 300 pixels were acquired using 100 µs dwell time
per pixel, a beam current of 0.9 nA, and a total acquisition time of less
than 4 minutes (226 sec). Samples courtesy of C. Wang, V. Stamenkovic,
N. Markovic (Argonne National Laboratory). Tecnai Osiris experiments by
D. Klenov (FEI) and N. J. Zaluzec (ANL).
