You are on page 1of 3

 COVER STORY

From 2D to 3D Chemical Analysis


on the µm- to cm-Scale
Combining µ‑XRF and SEM/EDS with Serial Sectioning

Fig. 1: Loss of information from a 3D volume to a 2D section.

For a long time, the world of microanalysis was 2-dimensional. Recently, atom probe tomography (APT) sorption of X-rays before they can leave
and focused ion beam (FIB) serial sectioning (combined with other techniques) have become available for the sample surface. Any non-destructive
the 3D chemical analysis of small structures (<100 nm for APT and <20 µm for FIB). The article below il- technique promising 3D element dis-
lustrates, that the modern silicon drift detector (SDD) technology used in energy dispersive spectrometry tribution data quickly reaches its lim-
(EDS) on scanning electron microscopes (SEM) and in spatially resolved micro X-ray fluorescence (µ‑XRF) its when it comes to sample size and
makes it quite simple to obtain 3D chemical data of larger samples. The total time spent on such 3D re- resolution.
constructions (including the time spent on 2D map acquisition, grinding/ polishing and 3D data presen- Recent developments in silicon drift
tation) is comparable to the time spent on a 3D-FIB dataset. Advances in data processing make “sample detector (SDD) technology as well as in
digitalization” possible. This means that the sample can be almost completely polished away, but all the signal and data processing have made
chemical data is still recorded and can be presented as a virtual 3D data cube. How EDS and µ‑XRF com- µ‑XRF and EDS much faster [1, 2]. This
plement each other in 3D chemical analysis is also discussed below. enables the 3D chemical analysis of
larger samples (with internal struc-
tures ranging from µm to cm in size) in a
short amount of time. In order to obtain
reliable and quantitative results with
3D-µ-XRF and 3D-EDS, the sample has to
be sectioned (for example by a stepwise
grinding and polishing procedure). Ro-
Fig. 2: Typical
bots have been developed for this kind of
detection limits of
µ‑XRF and SEM/ serial sectioning [3, 4], as they can easily
EDS. For µ‑XRF, a deliver a fairly constant sectioning depth.
Rh-tube source and 2D element distribution maps are ob-
an oxide matrix
tained at each sectioning step and then
were assumed.
For SEM/EDS, a a virtual 3D data cube is reconstructed
sample composi- with 3D visualization software. Bruker’s
tion with a mean HyperMap* option makes it possible to
atomic number of
access each 2D map later on for re-ex-
28 (copper plus
element of interest) amination and post-processing. There-
was assumed. fore the fact that this is a destructive
technique is compensated by what can
2D Versus 3D 3D Structural Information Versus 3D be called “sample digitalization”.
Chemical Information
Imagine living in a 2-dimensional world.
A significant amount of information, es- Displaying internal structures of com- The Benefit of Combining µ-XRF and
pecially regarding the true size of ob- plex materials in 3D has become easier SEM/EDS
jects, their actual shape and specific as technology has made significant ad-
arrangement, would be missing. Further- vances, for example in X-ray computed Whether it’s a ceramic, a metal alloy, a
more, the number of objects observed in tomography (CT). This technique is non- semiconductor or a rock sample, most
a 2D image, defined as a plane within a destructive, but it mainly displays differ- analysts know that they need to combine
3-dimensional space, would most likely ences in phase densities. Revealing the the results obtained with several analyti-
not be representative for the whole vol- actual 3D element distribution in a sam- cal methods in order to completely un-
ume. All of these drawbacks are illus- ple on the millimeter to centimeter scale derstand the chemistry and structure
trated in figure 1. is complicated because of the self-ab- of a sample. This is mainly the case, be-

G.I.T. Imaging & Microscopy 2/2013 • 19


COVER STORY

cause every analytical technique has ad-


vantages and disadvantages, but it also
helps reduce precious instrument analy-
sis time.
SEM/EDS and µ‑XRF are well-estab-
lished analytical methods that comple-
ment each other in many ways. When
it comes to spatial resolution, one can
clearly “see more” with SEM/EDS (20
nm – 0.8 µm) than with µ‑XRF (≤ 25
µm). On the other hand, large scan areas
can be covered much faster with µ‑XRF
than with SEM/EDS. Figure 2 shows typi-
cal curves with detection limits for both
techniques. For elements heavier than
calcium (Z = 20), µ‑XRF is more sensitive
than EDS.

Fig. 4: 3D reconstructions of chemical data from the Gujba sample shown in figure 3. a) 3D-µ‑XRF
dataset “Volume 1”, b) 3D-µ‑XRF dataset “Volume 2” (4.82 x 4.84 x 1.78 mm), c) 3D-EDS dataset (3.84
mm x 1.92 mm x 80 µm). The 2D-EDS data was acquired as a mosaic of 3x2 maps as indicated by the
white lines. See table 1 for analytical details of all three datasets.

time typically spent on 3D-FIB serial sec- found within the metal particle (green
tioning (combined with EDS or EBSD). arrows in fig. 4c). Next to the sulfide-
It has been suggested that the metal free edges of the metal particle a lot of
particles and silicate spherules (“chon- sulfides are evident within the surround-
drules”) of the Gujba meteorite were ing impact melt matrix (white arrows in
Fig. 3: Photograph of the Gujba meteorite. formed from a vapor-melt plume, gen- fig. 4c) indicating their mobilization. With
erated by a major impact event between the 3D datasets, we can not only visual-
two protoplanetary bodies about 4.563 ize the actual size and shape of the metal
3D Chemical Analysis of a Meteorite Ga*** ago [5]. If this hypothesis is correct, particles better than in 2D. We can also
Sample one would expect that all metal parti- examine how the particles stick together
cles in the Gujba meteorite show a quite and grasp the magnitude of interactions
In this study, a meteorite sample (“Gu- uniform chemical composition. However, between them and the surrounding im-
jba”, shown in fig. 3) was chosen to dem- the 3D-µ-XRF data in figure 4a and 4b pact melt matrix. Our 3D data shows that
onstrate the gain of information from 2D shows that the chemical composition of at least one more (so-called “secondary”)
to 3D chemical analysis, as well as the the metal particles is in fact variable, su- impact event had a significant effect on
potential of combining µ‑XRF and EDS. ggesting a more complex origin. At least the chemical and possibly also the iso-
The rock contains various objects with 5 different types of metal particles are topic properties of the Gujba meteorite.
a wide range of sizes and structures: present. They can be characterized by:
larger rounded metal particles, silicate
Conclusions
spherules and fine-grained materials – ▪▪ their Ni content: low: ~5 wt %
making it an ideal example. (dark blue), intermediate: ~6.5 wt % Recent advances in silicon drift detector
Table 1 lists analytical details for (medium blue) and high: ~8.2 wt % technology combined with progress in
three 3D datasets derived from the Gujba (light blue) the automation of serial sectioning have
meteorite. Two datasets were obtained and created a lot of potential for 3D chemi-
with µ‑XRF using a Bruker M4 Tornado ▪▪ by the abundance, size and shape cal analysis with µ‑XRF and SEM/EDS.
and one dataset was obtained with a of sulfide inclusions (red). The total time spent on obtaining the 3D
Quantax EDS system using a XFlash chemical data (including sample prep-
6|60 silicon drift detector. The 3D-µ‑XRF Many sulfide inclusions are smaller than aration, 2D data acquisition and time
dataset “Volume 1” almost covers the 6 µm in size – this is where EDS is help- spent on the 3D reconstruction) is com-
complete meteorite sample (fig. 4a). A ful. The better spatial resolution of the parable to the time typically spent on se-
selected volume of interest was exam- 3D-EDS dataset (fig. 4c) allows the vis- rial sectioning with FIB (in combination
ined with higher resolution using both, ualization of the large number of small with EDS or EBSD). However, the cov-
3D-µ‑XRF (“Vol. 2”, fig. 4b) and 3D-EDS sulfide grains present in some of the ered sample volumes are several orders
(“EDS”, fig. 4c). For all three datasets, metal particles. An example is the metal of magnitude larger than with FIB (about
2D element distribution maps were im- particle on the right side of the 3D-µ‑XRF 20 x 15 x 5 µm are typical for FIB). Vol-
ported into a software for 3D reconstruc- “Volume 2” (area indicated by the yellow umes of up to 20 x 15 x 5 cm can be cov-
tions**. Even though serial sectioning arrow in fig. 4b) and the 3D-EDS dataset ered with µ‑XRF. Therefore, analytical
was performed by manual grinding and (yellow arrow in fig. 4c). Furthermore, questions involving larger sample vol-
polishing, the total time spent to produce only with EDS, bowl-shaped structures umes can easily be addressed with the
the 3D-datasets is still comparable to the can be resolved for some of the sulfides method described here.

20 • G.I.T. Imaging & Microscopy 2/2013


 COVER STORY

Acknowledgements Tab. 1: Analytical details

We would like to thank Rhian H. Jones µ-XRF: Vol. 1 µ-XRF: Vol. 2 EDS
(University of New Mexico, Albuquerque) Volume ~2200 mm³ 41.5 mm³ 0.6 mm³
for providing a test sample at the begin-
Layers 36 38 21
ning and for all of her advice during this
study. Section depth ~148 µm ~48 µm ~4 µm
Total depth 5.18 mm 1.78 mm 80 µm
Voxel size 32 x 32 x 148 µm 6 x 6 x 48 µm 1.6 x 1.6 x 4 µm
References
Instruments M4 TORNADO M4 TORNADO QUANTAX with XFlash 6|60
[1] Fiorini C. et al.: Review of Scientific Instru-
ments 68(6):2461–2465 (1997)
HV 30 kV 30 kV 15 kV
[2] Ritchie N.W.M. et al.: Microscopy and Microa- Beam current 600 µA 600 µA 5 nA
nalysis 18(4):892–904 (2012) Input countrate ~150 kcps ~150 kcps ~130 kcps
[3] Alkemper J. and Voorhees P. W.: Journal of Mi-
Acquisition time per layer 80 min 60 min 90 min
croscopy 201:388–394 (2000)
[4] Spowart J.E.: Scripta Materialia 55:5–10
Grinding/sample prep time ~20 min ~15 min ~15 min
per layer
(2006)
[5] Krot A. N. et al.: Nature 436:989–992 (2005) Total time ~60 h ~50 h ~40 h

* HyperMap is Bruker’s version of position tagged


spectrometry (PTS), where a complete spectrum Authors Contact
is acquired and stored for each pixel of the map- Dr. Jana Bergholtz (maiden name: Berlin)
Dr. Jana Berlin,
ping, allowing offline analysis from the database Bruker Nano GmbH
Ulrich Waldschläger,
even after the sample is no longer available. Berlin, Germany
Dr. Roald Tagle
** Tel.: +49 30 670990748
VSG’s Amira was used here.
*** 
Fax: +49 30 67099030
Ga = gigaannum- a unit of time equal to
jana.bergholtz@bruker-nano.de
1,000,000,000 years. www.bruker.com
/FVSPOTOFUXPSLNBSLFEXJUIBÜVPSPQIPS GBMTFDPMPSSFOEFSJOH

Every image is a work of art!


The innovative pco.edge XJUIPVUTUBOEJOHsCMOS technology
TFUTOFXTUBOEBSETGPSTDJFOUJÛDDBNFSBTZTUFNT*UDPNCJOFT
BOFYDFQUJPOBMEZOBNJDSBOHF   EJHJUJ[FEJOCJU

NBYJNVNGSBNFSBUF JNBHFTT
BOEIJHISFTPMVUJPO
°QJYFM
XJUIFYUSFNFMZMPXSFBEPVUOPJTF med e-

JOUPPOFWFSTBUJMFPWFSBMMSFTVMUGPSEFNBOEJOHBQQMJDBUJPOT
www.pco.de
JO"NFSJDBXXXQDPUFDIDPN

G.I.T. Imaging & Microscopy 2/2013 • 21

You might also like