Measuring the Thickness of Aluminium Alloy Thin

Foils Using Electron Energy Loss Spectroscopy

A. Bardal* and K. Lie†
*SINTEF Materials Technology, N-7465 Trondheim, Norway and †Department of Physics,
Norwegian University of Science and Technology, N-7491 Trondheim, Norway

Combining electron energy loss spectroscopy and convergent beam electron diffraction
measurements, we have determined the mean-free-path for inelastic electron scattering for
four different aluminium alloys. Electron energy loss spectroscopy spectra were acquired
with the transmission electron microscope in image mode, without any objective aperture
inserted. The value for pure aluminium was determined as 119 ⫾ 5nm at an incident elec-
tron energy of 150keV. For the most common alloys with moderate amounts of alloying ele-
ments, no significant changes in the mean-free-path values are measured within the 5% ac-
curacy of the method. The-mean-free path values are required for accurate determination of
the thickness of thin foils for transmission electron microscopy using electron energy loss
spectroscopy. On the basis of our experimental results, we estimate a single measurement
of foil thickness to have an accuracy of better than 8% if the mean-free-path is known
within 5%. The outlined procedure for thickness measurements is very robust, with thick-
ness results being insensitive to experimental details such as inclination of the specimen or
diffracting conditions. © Elsevier Science Inc., 2000. All rights reseved.

INTRODUCTION chemical composition of the sample using

When using transmission electron micros-
copy for measuring the densities of micro-
structural features such as dislocations,
precipitates, or dispersoids in metallurgical
samples, information on the thin-foil thick-
ness is needed. It is important to obtain as
accurate measurements of the thickness as
possible. At the same time, to obtain statis-
tically reliable data, a large number of den-
sity measurements is often required. If a
thickness measurement is required from
each sample used for the density measure-
ments, the time consumption involved
with the thickness measurement becomes a
very important issue. Apart from the im-
portance for density measurements, an ac-
curate knowledge of sample thickness is
also in some cases important for absorption
corrections when determining the local
329
MATERIALS CHARACTERIZATION 44:329–343 (2000)
© Elsevier Science Inc., 2000. All rights reserved.
655 Avenue of the Americas, New York, NY 10010
energy-dispersive X-ray spectroscopy. A
number of methods have been used in the
past for measuring the thickness of thin
foils for transmission electron microscopy.
These include the contamination spot
method, the use of thickness fringes at an
inclined boundary, and the use of the X-ray
yield’s dependence on thickness (as re-
viewed, for example, in [1, 2]). These meth-
ods suffer from being time consuming, of
poor accuracy, or both. Convergent-beam
electron diffraction (CBED) yields good ac-
curacy, better than 2–5% relative error in
the simplest version where the periodicity
of extinction fringes under two-beam con-
ditions is utilized [3–5]. Even better accu-
racy can be obtained by quantitative many-
parameter fits to the intensity profiles of
the CBED disks [6]. The CBED methods
are, however, quite time consuming (sev-
1044-5803/00/$–see front matter
PII S1044-5803(99)00072-8
330
eral minutes per measurement at best), in-
volving controlled tilting of the sample and
fairly elaborate data evaluation.
The electron energy loss spectroscopy
(EELS) method for measuring sample
thickness is based on a simple relationship
between the intensity distribution in the
EELS spectrum and the ratio between the
sample thickness and a mean free path for
inelastic scattering [7]. With use of a paral-
lel EELS spectrometer, the method is very
fast, with a recording time of fractions of a
second per spectrum, and a data evaluation
time per spectrum of on the order of 20 s.
The method is, therefore, many times faster
than the alternatives. The real benefit from
this is gained when many thickness mea-
surements need to be made. For instance,
500 thickness measurements using EELS
would take about 3 h, whereas the same
number of measurements using CBED or
the contamination spot method would take
50 h or more. EELS, therefore, opens up the
possibility of sampling real distributions of
sizes or densities, where the alternative
would have been to obtain only mean
values based on a limited number of mea-
surements. Quite large matrices of materi-
als conditions can be studied, without
thickness measurements being a severe
time-limiting factor.
Some examples from recent use of the
method in projects at our laboratories may
illustrate the benefits of having a really fast
measurement technique at hand. In [8], we
studied deformation structures in a plane-
strain compressed aluminium alloy (3004)
to relate these to recrystallization behavior.
The distributions of dislocation density (␳),
subgrain boundary misorientation (␪), and
subgrain size (␦) were measured within the
different texture components of the mate-
rial. The work involved obtaining approxi-
mately 200 individual samples of disloca-
tion densities from different regions and
texture components of the transmission
electron microscopy specimens, each sam-
ple involving one thickness measurement.
In [9], the distributions of the same ␳, ␦, and
␪ microstructural parameters were mea-
sured for a series of materials that had been
A. Bardal and K. Lie
deformed over a wide range in the Zener-
Hollomon parameter (a dimensionless para-
meter for deformation kinetics that includes
strain rate, temperature, and activation en-
ergy for dislocation movement). Here, more
than 600 individual samples of ␳ were ob-
tained, each sample involving one thick-
ness measurement. Lodgaard and Ryum
(Norwegian University of Science and
Technology, Private Communication, 1998)
carried out a comprehensive study of how
the densities of ␣-Al(Fe,Mn)Si dispersoids
in a 6082 alloy varied with annealing time
in the temperature range of 520–550–580⬚C,
which involved approximately 400 thick-
ness measurements.
The EELS method has potential for being
very accurate, although not reaching the
accuracy of 2% or better obtained using
CBED. The accuracy of the EELS method
depends on the accuracy with which the
mean free path value is known for the mate-
rial in question, and on the accuracy with
which the intensity of the zero loss peak can
be extracted from the experimental spec-
trum. Malis et al. [10] established a parame-
terized expression for the mean-free-path
value based on experimental measure-
ments on 11 different materials. Using this
approach, thickness measurements are
expected to be accurate within ⫾20% for
most inorganic specimens, without any ex-
plicit calibration of the mean-free-path
value being necessary. Considerably better
accuracy is within reach, however, if it is
possible, and one takes the trouble, to ex-
plicitly calibrate the mean-free-path value
for the material in question. There are few
reports in the literature on the attainable
accuracy in the latter case. Working on an
AA3004 aluminium alloy, Botton et al. [11]
found a mean-free-path value of ␭ ⫽ 175 ⫾
5nm at an accelerating voltage of 300kV
and a spectrometer collection angle of
9.5mrad. That is, ␭ was determined with an
accuracy of 3% by means of combined
CBED and EELS measurements.
The motivation for the work reported in
this paper was to determine mean-free-
path values and accuracies for a selection
of aluminium alloys, and to establish a
Measuring Thickness of Aluminum Alloy Thin Foils with EELS
measurement procedure that takes full
advantage of the possible speed of EELS
thickness measurements, while at the same
time maintaining a highest possible accu-
racy. The major concern for such a proce-
dure is that one should avoid any time-con-
suming controlled tilting of the transmission
electron microscopy specimen. This means
that the crystal orientation and surface in-
clination might vary, and might be un-
known for the recorded EELS spectra. If the
specimen is in a strongly diffracting condi-
tion, the Bragg diffraction may cause an an-
gular redistribution of the ratio between in-
elastically and elastically scattered electrons.
If the EELS spectrum is recorded with an
angle-limiting aperture, the relative inten-
sity of the zero-loss electrons will be af-
fected by such elastic-inelastic coupling.
The coupled plural inelastic scattering and
elastic Bragg diffraction is not easily treated
theoretically, but experimental observations
show that the effect may be significant. Bot-
ton et al. [11] measured variations in elastic/
inelastic ratios of 10% across bend contours,
and attributed this to Bragg diffraction.
Our approach was to work with the
transmission electron micoscope (TEM) in
image mode while doing the EELS mea-
surements, as well as the imaging of the
microstructural feature of interest. The
EELS spectrometer can then collect all elec-
trons scattered up to large angles, and there
is no need to worry about angular redistri-
bution of the scattering, or to tilt the speci-
men to nondiffracting or well-defined dif-
fracting orientations. The specimen tilt and
crystallographic orientation can still, in
principle, affect the result by modifying the
probability for some of the inelastic scatter-
ing processes. The excitation of surface plas-
mons is enhanced if the sample surface is
inclined to the incident electron beam,
whereas excitation of inner-shell electrons
is affected by localization of the electron
Bloch waves under strongly diffracting
conditions. These effects were expected to
be small and probably negligible, but an
experimental verification of this view is in
order and is made here.
Using a combination of EELS and CBED
331
measurements, we determined the mean-
free-path value ␭ appropriate for the cho-
sen procedure and investigated how ␭ var-
ied with alloy composition and temper,
and with experimental condition such as
diffracting condition and specimen tilt. On
the basis of this work, mean-free-path
values and accuracies can also be estimated
for samples in which information on alloy
composition and temper is incomplete, and
for which no explicit calibration of ␭ has
been made.
THEORY
The total intensity of the electrons that are
inelastically scattered while being transmit-
ted through a thin foil of thickness t can be
expressed as:
i
I tot ( t ) = I surf + Σ n ,i P n ( t ) (1)
where Isurf is a thickness-independent sur-
face contribution, and Pni are probabilities
for the different inelastic scattering pro-
cesses at thickness t. These probabilities can
be expressed by a multiparameter Poisson
distribution [12]:
i n
P n ( t ) = 1 ⁄ n! ⋅ ( t ⁄ λ i ) ⋅ exp ( – t ⁄ λ ) (2)
where the index i denotes various types of
scattering processes, and the index n de-
notes the order of plural scattering. The
total mean-free-path value ␭, and the mean-
free-paths for the individual scattering pro-
cesses ␭i, are related by:
1 ⁄ λ = Σi 1 ⁄ λi (3)
as derived from the scattering cross-section:
σ = Σσ i (4)
because
σi = 1 ⁄ na ⋅ λi (5)
where na is the number of atoms per vol-
ume. Setting n ⫽ 0 in Eq. (2), yields:
t ⁄ λ = ln ( I tot ⁄ I 0 ) (6)
332
where I0 is the intensity of the zero-loss
peak, and Itot is the total intensity in the
spectrum.
The most important inelastic scattering
process is the excitation of volume plas-
mons, i.e., collective oscillations of conduc-
tion electrons [13]. With n ⫽ 1 and i ⫽ plas-
mon scattering in Eq. (2):
t ⁄ λ plasmon = I 1 ⁄ I 0 (7)
where I1 is the intensity contained in the
first plasmon peak. ␭ and ␭plasmon have been
shown experimentally to differ by approxi-
mately 30% (see, e.g., [14]), illustrating that
other processes than plasmon scattering
also contribute significantly. The other in-
elastic scattering processes that need to be
considered are coherent double plasmon
excitation, excitation of single conduction
electrons, excitation of single core elec-
trons, and excitation of surface plasmons.
In addition, the incident electrons undergo
phonon scattering. The latter process is
quasielastic (losses less than 100meV) and
the thermal-diffuse scattered electrons con-
tribute to the zero-loss peak under the ex-
perimental conditions given by a transmis-
sion electron microscope.
It is appropriate here to briefly discuss
the different scattering processes, the theo-
retically expected effect of microstructural
features of the aluminium alloy on the
cross sections, and the effects of experimen-
tal parameters such as sample inclination
and diffracting conditions. Because the
plasmon excitations are collective excita-
tions of a medium, the cross-sections of
these processes depend on the microstruc-
tural state of the alloy. In principle, it
makes a difference whether the medium
consists of a single phase with solute at-
oms, or if it is a composite medium with
several phases in the form of fine precipi-
tates, dispersoids, or primary particles in a
matrix. It is convenient to discuss and com-
pare the contributions from different scatter-
ing processes in terms of their cross sections,
because these are additive [see Eq. (4)].
Volume plasmons in aluminium can be
well described by the free-electron theory.
A. Bardal and K. Lie
The dependence of ␴p and ␭p on the energy
of the incident electrons and on the plasmon
energy Ep is given by Eqs. (8) and (5) [7], and
the dependence of Ep on the volume density
of conduction electrons (n) by Eq. (9).
2
σ p ( β ) ≈ Ep ⋅ ln ( β ⁄ θ E ) ⁄ n a a 0 m 0 v (8)
2 2 1⁄2
E p = [ ne h ⁄ ε 0 m ] (9)
Here, ␤ is the collection angle, ␪E is the
characteristic scattering angle, a0 is the Bohr
radius, m0 and m are the rest mass and rela-
tivistic mass of the electron, respectively,
and n is the volume density of conduction
electrons. Alloying elements in solution
give rise to a changed volume-density of
conduction electrons, Eq. (9), causing shifted
Ep, ␭p and ␴p values. The Ep value shows a
linear dependence on solute concentration
in certain cases, as reported for a number of
alloy systems such as Al–Mg, Al–Mg–Zn,
Al–Zn, Al–Ag, and Al–Li [15]. As an exam-
ple, we would expect 5% magnesium in
solid solution to yield a decrease in Ep of
approximately 0.25eV, corresponding to a
decrease in ␴p of about 2%, Eq. (8).
Apart from excitation of ordinary volume
plasmons, there is also the possibility for co-
herent double plasmon excitation, i.e., the
excitation of two plasmons by one electron
in a single scattering event [12]. The most
recent literature on this subject [16, 17] agrees
that the probability of this process is very
small, with P2/P1 being less than 1–3% (P2
and P1 are the probabilities for coherent
double plasmon excitation and ordinary
plasmon excitation, respectively). In [16], a
review is also given of the wide scatter in
reported experimental values for this ratio
as well as a discussion of possible sources
of error in the different experiments. In the
present context, this process can be com-
pletely ignored.
In addition to the volume plasmons that
propagate inside the solid, longitudinal
waves of charge density that travel along
the surface, i.e., surface plasmons, are ex-
cited by the incident electron. In the practi-
cal case when the metal is covered by a thin
layer of surface oxide, the characteristic en-
Measuring Thickness of Aluminum Alloy Thin Foils with EELS
ergy of the surface plasmon is Es ⫽ Ep/(1 ⫹
ε1)1/2, where ε1 is the real part of the dielec-
tric permitivity of the oxide. Calculations
and experiment confirm that a typical na-
tive oxide thickness of 4nm is sufficient to
yield a surface plasmon energy of alumin-
ium of Es ⬇ 7eV [18, 19]. For an Al speci-
men with native oxide at normal incidence,
the probability for surface plasmon excita-
tion is estimated as 2% [7]. The surface
plasmon peak is easily visible in spectra
from thin specimens, and will be stronger if
the specimen is tilted (i.e., the incident elec-
tron arrives at a non-normal incidence an-
gle). A general expression for the effect of
incidence angle on the probability for sur-
face plasmon excitation is given in [20]. The
total surface plasmon intensity for a given
angle of incidence, integrated over all scat-
tering angles, can be derived from this ex-
pression, but cumbersome integration is re-
quired.
A competing mechanism to excitation of
the conduction electrons as plasmons, is the
excitation of single conduction electrons
[7]. These single-electron (e–e) excitations
involve larger momentum transfer than
plasmon excitations and scattering of the
transmitted electrons to higher angles than
the cutoff angle for plasmon excitation. In a
theoretical treatment, Ritchie and Howie
[20] link the plasmon and high-angle sin-
gle-electron regimes. Their scattering cross-
section reduces to that of plasmon scatter-
ing in the limit of small angles and to that
of Rutherford scattering from a free elec-
tron in the limit of large angles. Their
mean-free-paths would suggest that about
10% of all electrons suffer e–e scattering to
a high angle in a thin sample. Eaglesham
and Berger [21] give experimental evidence
for single-electron excitations being a sig-
nificant contribution to the energy loss
spectrum. For aluminium they find the e–e
contribution to be comparable to that of
phonon scattering at room temperature.
Calculations by Egerton and Wong [22] are
in good accordance with this. Their calcula-
tions of e–e cross-sections, using the Lenz
model [23], show an increase with atomic
333
number roughly as ␴ ∝ Z1/3, which can be
used to estimate how the e–e cross-section
varies with alloy content.
For excitation of core electrons (ioniza-
tion), the scattering cross-sections can be
treated within an atomic framework to a
good approximation (although the fine
structure of ionization edges, of course, can
depend strongly on bonding and the chemi-
cal state of the compound). Hence, ioniza-
tion cross-sections for the alloy can be
found by simple averaging of cross-sections
for the different species, according to the
overall composition of the probed volume,
and are essentially independent of the
alloy’s microstructural state and temper.
Scattering cross-sections for the excitation
of core electrons can be calculated to good
accuracy using, for example, the hydro-
genic model [7] or the Hartree-Slater model
[24]. Both approaches are implemented in
the commercial EELS software [25]. The
core-loss scattering process of major rele-
vance in this context is the ionization of
aluminium L electrons with an ionization
edge onset of approximately 73eV. For al-
loys, the core-loss contribution to the total
scattering cross-section will be an average
based on the fractions (atomic percentage)
of the alloying elements.
Because core electrons are localized close
to the atomic nuclei, the excitation of these
exhibit channeling effects. That is, an ion-
ization edge will become more or less
prominent depending on the location in the
unit cell of the atoms being ionized, relative
to those that lie on the Bragg reflecting
planes [26]. Hence, the t/␭ ratio measured
by EELS will, in principle, be affected by
the diffracting conditions in the probed
region of the sample, as given by its crystal-
lographic orientation with respect to the
electron beam. Calculations of the localiza-
tion of inelastic scattering by Kohl and
Rose [27] show that ionization of alumin-
ium L electrons (onset at ⌬E ⫽ 73eV) is de-
localized over more than the unit-cell
dimensions. Hence, the channeling effects
on measured t/␭ ratios are believed to be
small and possibly negligible.
334
EXPERIMENTAL METHODS
The transmission electron microscopy work
was carried out using a Philips CM30
TEM/STEM microscope, equipped with a
Gatan model 666 PEELS system. A liquid–
nitrogen-cooled sample holder was used,
to reduce contamination and contribution
to the CBED patterns from thermal diffuse
scattering. The measurements were made
at an accelerating voltage of 150kV. For
measuring the densities of dislocations,
precipitates, or dispersoids in aluminium
alloys it is advisable to work at an acceler-
ating voltage below 170kV, which is the
threshold voltage for radiation damage by
the knock-on process [28, 29]. The small
dislocation loops resulting from this pro-
cess, particularly when working close to
low-index zone axes, can greatly obscure
the image contrast and lead to erroneous
results.
For each of the four alloys we investi-
gated, between 6 and 14 pairs of CBED/
EELS measurements were made from posi-
tions on the samples having thickness
values in the range 80–400nm.
The purpose of the CBED measurements
was to obtain reasonably accurate measure-
ments of sample thickness, which could be
use as reference values when extracting
mean-free-path values ␭ from the less accu-
rate t/␭ values obtained through the EELS
measurements. CBED patterns were re-
corded with the sample oriented in the
two-beam diffracting condition with a
(220)- or (311)-reflection excited. Care was
taken to avoid the well-known sources of
error that can affect thickness measure-
ments, as summarized by Ecob [5]. That is,
both intensity minima and maxima of the
extinction fringes were included in the
analysis, only g-vectors of sufficient magni-
tude were used, and measurement errors
were minimized by scanning the negatives
from the transmission electron microscope
and using intensity line profiles for deter-
mining positions of minima and maxima.
Data evaluation followed the well-estab-
lished procedure as described, for example,
in [3]. That is, maxima and minima of the
A. Bardal and K. Lie
220 (or 311) rocking curve intensity profiles
were plotted in a si2/(ni,xk)2 vs. 1/(ni,xk)2
plot. Here, si is the deviation parameter at
the i-th minimum or maximum, ni are inte-
gers corresponding to the order of minima,
and xk are numbers corresponding to the
order of maxima, as described in more de-
tail in [3–5]. The t- and ␰g-values were ex-
tracted by fitting a straight line to the data,
determining the line’s slope (b) and its in-
tercept with the ordinate axis (a), and using
the relationships: t ⫽ a⫺1/2 and ␰g ⫽
(⫺b)⫺1/2. As long as the extinction distance
of the operating Bragg reflection is known
approximately, there is no need to resort to
dynamical calculations to extract thickness
values, an approach taken in [30]. Instead,
the starting index n1 can be unambiguously
chosen as the one that yields a constant ex-
tinction distance for all patterns analysed
under the same diffracting conditions [5].
Our measured extinction distances were
compared to the calculated values of ␰220 ⫽
130nm and ␰311 ⫽ 164nm (at E ⫽ 150kV) ob-
tained using two-beam theory and Doyle-
Turner atomic scattering factors [31]. For
the CBED measurements, the standard er-
rors of the least-squares fit were taken as
the uncertainities of the single measure-
ments of t and ␰g.
EELS measurements were made with the
microscope in the image mode, and with
the objective aperture removed. Hence,
electrons that had been scattered through
large angles were admitted into the spec-
trometer entrance aperture. The collection
angle in this condition is larger than
100mrad, meaning that essentially all trans-
mitted electrons are collected. A spectrom-
eter entrance aperture having a diameter of
3mm or less, and an image magnification of
50,000⫻ was used. The probed areas thus
had diameters less than 60nm, and thick-
ness variations in the probed areas (which
cause non-Poissonian intensity distribu-
tions in the spectrum) were negligible.
The electronics of our spectrometer (pur-
chased 1989) originally had a defect caus-
ing a nonlinear detector response on the in-
cident signal as described by Egerton et al.
[32]. The problem was cured by making a
Measuring Thickness of Aluminum Alloy Thin Foils with EELS
modification to the electronics as suggested
in Egerton’s paper, and a linear response
was verified by plotting zero-loss peak
(ZLP) intensity (with the electron beam
through a hole in the sample) vs. incident
beam current. The background level of the
spectra caused by detector dark current,
and readout noise was subtracted using
standard procedures of the commercially
available software [25]. To correct for gain
variations across the photodiode array of
the EELS detector, we divided the spectra
by a gain correction curve. This was ob-
tained through the procedure described by
Boothroyd et al. [33], in which several spec-
tra are acquired at different positions on
the photodiode array and averaged to pro-
duce a gain-corrected spectrum. After this
correction, we found that sequential EELS
recordings at the same position of the sam-
ple, but onto different parts of the EELS
detector array, yielded identical intensity
distributions in the recorded spectrum. On
our spectrometer setup, this is not entirely
the case when using only the default, stan-
dard procedure [25] for correcting gain
variations across the detector array.
The ZLP and full spectrum intensities, I0
and Itot in Eq. (2) were determined using a
standard procedure of the commercially
available software [25]. Using this proce-
dure, the background level of the recorded
spectra (caused by detector dark current
and readout noise) was checked on the far
low-energy side of the ZLP and corrections
were made to the background if necessary.
The tails of the ZLP were modeled using a
Lorentzian extrapolated under the loss
data. Provided that proper care has been
taken to ensure a linear response of the de-
tector on the incident signal, the main
source of error in determining the t/␭
value lies in the extraction of the ZLP inten-
sity from the total intensity of the spec-
trum. The details of the procedures for ex-
tracting I0 affects the resulting t/␭ value
significantly. It is difficult to assess theoret-
ically what the accuracy of the generic ap-
proach of fitting a Lorentzian is, but the
empirical results of the present study give a
good indication. The chosen procedure can
335
be regarded as sufficiently correct to yield
thickness values with an accuracy of ap-
proximately 8%, as witnessed by our re-
sults and as discussed later. The procedure
yields a thickness independent ␭ value
over the thickness range 80–400nm at least,
as required.
We also tested inferior procedures for ex-
tracting Io (as implemented, for example, in
earlier versions of commercial software) in
which no check of the background level is
made, and in which the ZLP is extrapolated
using an exponential. These procedures
yielded apparent ␭ values that varied with
thickness when t/␭ was calibrated against
CBED thickness values. We observed a sys-
tematic apparent increase in ␭ with thick-
ness of 5–10% relative over the investigated
thickness range, when using the inferior
procedures.
MATERIALS
Four different aluminium alloys were in-
vestigated. A commercially pure alloy
(sample A) was included, along with alloys
that have a reasonably high density of fine
precipitates (sample B), or that have a rea-
sonably high concentration of solute atoms
(samples C and D). The effect of solute at-
oms on the plasmon mean-free-path de-
pends on the solute species, as different
species can modify the density of free (con-
duction) electrons (n) to a varying degree.
For all samples, it was verified with energy
dispersive X-ray spectrometer (EDS) in the
TEM that the composition of the probed
volumes in the thin foil was close to the
overall composition of the bulk material.
No elaborate EDS studies were carried out,
though, to determine very accurately the
exact composition; that is, only theoretical
k-factors were used. Also, absorption ef-
fects were ignored, because EDS recordings
were made in very thin parts of the TEM
specimen. The quoted compositions should,
therefore, only be regarded as approximate
values, possibly with up to about 20% rela-
tive error in the different elements. This ac-
curacy is sufficient for the present context.
336
Specimen A was aluminium of 99.5%
commercial purity, with Fe and Si present
only in significant amounts as large inter-
metallic particles, and with no significant
amount of alloying elements in solid solu-
tion or as fine scale precipitates. Specimen
B was a 7030 type alloy containing 5 wt.%
Zn and 1.2 wt.% Mg. The alloy was given a
solution heat treatment at 480⬚C for 30 min
followed by quenching, and aging at 150⬚C
for 2 h. In this peak-aged condition, the ma-
jor alloying elements are present as approx-
imately 1 vol.% ZnMg2 precipitates. The
reason for including this alloy was its fairly
high density of fine precipitates. Although
a volume percentage of around 1% may
sound quite modest, this is about as high a
value as one may find in the most impor-
tant 2xxx, 3xxx, 5xxx, 6xxx, and 7xxx alloy
series. Specimen C was a model alloy of
99.99% purity aluminium alloyed to con-
tain 1.6 wt.% of manganese, and given a so-
lution heat treatment at 640⬚C for 1 h, fol-
lowed by quenching, to bring all the
manganese into solid solution. Specimen D
was a model alloy of 99.99% purity alumin-
ium alloyed to contain 5 wt.% magnesium,
and given a solution heat treatment at
525⬚C for 3 h, followed by quenching, to
bring all the magnesium into solid solution.
All specimens for transmission electron
microscopy were prepared by mechanical
grinding and polishing, and standard elec-
tropolishing procedures using a Struers
Tenupol electropolishing unit, a methanol–
nitric acid electrolyte at approximately
⫺30⬚C, medium flow rate, and a potential
of 20V producing a current near 200mA
during thinning.
RESULTS AND DISCUSSION
Figure 1 shows results of the CBED mea-
surements, with extinction distance plot-
ted vs. thickness. The arithmetic mean of
the measured ␰220 values is 130 ⫾ 6nm,
which is equal to the theoretical two-beam
value of 130nm. The arithmetic mean of the
measured ␰311 values is 170 ⫾ 10nm, which
is close to the theoretical two-beam value of
A. Bardal and K. Lie
164nm. For all points, the measured ␰g
value lies within 2␴ of the mean value,
showing that the CBED results are very re-
liable. The standard errors of the single ␰g
values range from 23% relative to 3% rela-
tive, as indicated with vertical error bars.
As expected, the measured ␰g values do not
vary with thickness. The standard errors of
the single thickness values range from 0.8%
realtive to 5% relative, as indicated with
horizontal error bars. The average standard
error of the thickness values was 1.8%.
All the mean-free-path (␭) values, as ob-
tained by comparing pairs of CBED thick-
ness and EELS t/␭ results, are plotted vs.
sample thickness in Fig. 2. There is no vari-
ation of the ␭ values with thickness, show-
ing that that the procedure used for extract-
ing the values is working. In the plot, the
mean-free-path values for specimens A–C
appear to be similar, whereas the values for
specimen D, with 5% magnesium in solid
solution, appear to be slightly higher. Sta-
tistical treatment of the data, as summa-
rized in Table 1, shows that there are no
significant differences in the ␭ values for
specimens A, B, and C. The average ␭ value
for specimen D is 5% higher compared to
pure Al (specimen A), but this difference still
cannot be termed significant at the ␴ level,
given the accuracy of the measurements. The
accuracy with which the ␭ values are deter-
mined are 2–5%.
Apart from the measured ␭ values, we
have also included in Table 1 the measured
plasmon energies Ep, the total scattering
cross-section ␴total calculated from ␭ using
Eq. (5), and the scattering cross-section for
plasmon excitation [calculated using Eqs.
(4) and (7), and the measured Ep value].
Cross-section values were included for the
purpose of comparing the relative impor-
tance of different scattering processes. For
calculation of the plasmon mean-free-path,
the cutoff angle ␪c was set equal to ␤ in Eq.
(7), and a value of ␪c ⫽ 5mrad was used, on
the basis of the discussion in [7]. ␴p values
are found to be typically 9.3 ⫻ 10⫺5nm2,
which amounts to approximately 66% of
the total cross section ␴total (typically 14.0 ·
10⫺5nm2). The statistical relative errors in
Measuring Thickness of Aluminum Alloy Thin Foils with EELS 337

FIG. 1. Extinction distances as measured using convergent beam electron diffraction (CBED) plotted vs. thin-foil
thickness.

the calculated ␴p values are equal to those
of the corresponding Ep values, i.e., near
0.2%. The total uncertainty in the calculated
␴p value (related to the theoretical model) is
not known, but any error would yield the
same systematic shift for the different al-
loys. Hence, ␴p for specimen D is signifi-
cantly lower. It is probably the main contri-
bution to the lowered value of ␴total for this
alloy, but more precise statements cannot
be made with the given measurement accu-
racy of ␭ (and ␴total). The decrease of ap-
proximately 0.20eV in Ep and of approxi-
mately 1.5% in ␴p is in good agreement
with theory, Eqs. (8) and (9).
The cross-sections for core-loss excitation
were calculated using the Hartree-Slater
model implemented in [25], and the average
core-loss cross section for each alloy was
found from the alloy composition. The
edges which were considered were Al K,
Mg K, Mn L, and M, Zn L, and M. The ap-
proach yields a cross-section of 2.4 ⋅ 10⫺5nm2
for all alloys, and no significant differences
between the alloys. This process is seen to
contribute with approximately 17% to ␴total.
338
FIG. 2. Experimentally determined mean-free-path values plotted vs. thin-foil thickness.
The remaining approximately 17% of the to-
tal scattering cross-section must be due to
other scattering processes, essentially sin-
gle-electron scattering of conduction elec-
trons. The apparent relative contribution of
the single-electron process is in good accor-
dance with earlier work [20–22].
It is now interesting to discuss the accu-
racy, ␦tj/tj, of a single EELS measurement
(j) of thin foil thickness, once the ␭ value
has been calibrated, and has a known accu-
racy ␦␭. From Eq. (6) we obtain:
δt j ⁄ t j = δλ ⁄ λ + δL j ⁄ L j
where Lj ⫽ ln (Itot/I0)j. Apart from the sys-
tematic uncertainty in the calibrated value
of the mean-free-path , ␦tj/tj only depends
on ␦Lj/Lj, i.e., how accurately the ZLP in-
tensity can be extracted from the EELS
spectrum. With the generic approach to ex-
A. Bardal and K. Lie
traction of I0 by fitting of a Lorentzian to
the ZLP tail, there is no general answer to
what the magnitude of ␦Lj/Lj is. An esti-
mate of the typical, approximate ␦Lj/Lj ra-
tio can, however, be made on the basis of
our experimental measurements.
For this, we again look at the pairs of
EELS and CBED measurements. In Fig. 3,
for each pair (j) of measurements, we have
plotted the relative deviation between the
determined ␭j and the average value ⬍␭j⬎
found for the alloy in question. This de-
viation is caused by the uncertainty in
(10)
extraction of I0 from the EELS spectrum
and by uncertainties in the CBED thick-
ness determination (0.8–5%, illustrated
by the horizontal error bars). In all
cases, [␭j ⫺ ⬍␭j⬎]/⬍␭j⬎ is found to be
less than 10% relative. The larger devia-
tions tend to be found for thickness values,
Measuring Thickness of Aluminum Alloy Thin Foils with EELS 339

Table 1 Experimentally Measured Mean Free Paths (␭) and Plasmon Energies (Ep) and
Derived Scattering Cross-Sections
Specimen A B C D

Alloy composition Al (comm.pure) Al–5 Zn–2 Mg Al–1.6 Mn Al–5 Mg

and temper Peak aged— Sol. treated— Sol. treated—
1 vol.% ZnMg2 1.6 wt.% Mn 5 wt.% Mg
precipitates in solid sol. in solid sol.
␭ (nm) Experimental 119 ⫾ 5 121 ⫾ 2 118 ⫾ 5 125 ⫾ 4
result
Ep (eV) Experimental 15.18 ⫾ 0.02 15.18 ⫾ 0.02 15.24 ⫾ 0.03 14.98 ⫾ 0.01
result
␴total (10⫺6 nm2) 140 ⫾ 6 138 ⫾ 3 140 ⫾ 6 133 ⫾ 4
Calculated from exp. ␭
␴p (10⫺6 nm2) 95.0 95.0 95.4 93.7
Calculated from exp. Ep

t/␭ ⬍ 1.5, but do not correlate particularly
with the CBED errors. The standard devia-
tion of all the [␭j ⫺ ⬍␭j⬎]/⬍␭j⬎ values is
3.4%. It should then be a good approxima-
tion to state that the accuracy in ␦Lj/Lj is
better than 3%. This is then the precision
when making relative comparison between
single thickness measurements from the
same sample. From (10), the accuracy in a
single thickness measurement by EELS is
better than 8% if ␭ is known within 5%.
One advantage of working with the TEM
in image mode during the EELS recordings
is that one should presumably not have to
worry about how the crystallographic ori-
entation of the probed region of the foil
would affect the measured thickness value,
as discussed in the introductory part of this
paper. This presumption was checked ex-
perimentally by making EELS recordings at
some well-defined crystallographic orien-
tations of the thin-foil specimen. A pair of

FIG. 3. Relative error in single ␭j values, as found from CBED and EELS results. The errors have contributions
from CBED thickness determination (horizontal error bars) and from extraction of I0 from the EELS spectra.
340
EELS spectra were recorded from a fixed
position on the sample, first with the sam-
ple oriented at the diffracting condition
and then oriented at a nondiffracting condi-
tion 1–2 degrees off. The conditions checked
were the strongly diffracting (111) and
(220) two-beam conditions. Results are dis-
played in Fig. 4, which shows the ratios be-
tween the measured t/␭ value at the dif-
fracting condition and the measured t/␭
value at the nondiffracting condition, plot-
ted vs. specimen thickness. All measured
ratios were found to deviate with less than
approximately 6% from unity, and there
are no systematic shifts to above or below
unity. There are, furthermore, no systemtic
variations with specimen thickness.
Another measurement series was made
to investigate the effect of the surface orien-
tation on the mean-free-path. The sample
FIG. 4. Ratio between t/␭ with the specimen in a strongly diffracting condition and t/␭ with the specimen in a
nondiffracting condition.
A. Bardal and K. Lie
was gradually tilted more and more, and
EELS recordings were made at a fixed posi-
tion, while ensuring that no strong beams
were operating in the diffraction pattern.
After correcting for the influence of speci-
men tilt (␥) on the effective foil thickness
(t⬘ ⫽ t/cos␥), variations in measured t/␭
values could be studied. Results are shown
in Fig. 5, for a specimen thickness of ap-
proximately 0.7␭. No significant variations
with surface inclination are found, i.e., any
possibly increased probability for surface
plasmon excitation does not significantly
affect the results.
When working with aluminium alloys for
which no explicit calibration of the mean-
free-path value has been made, fairly accu-
rate estimates of ␭ can still usually be made
based on the present investigation. For most
1xxx, 2xxx, 3xxx, 5xxx, 6xxx, and 7xxx series
Measuring Thickness of Aluminum Alloy Thin Foils with EELS
FIG. 5. Thickness/mean-free-path ratio (t/␭) plotted vs. tilt of thin-foil specimen.
alloys, a value of ␭ ⫽ 120 ⫾ 10nm can be
used (at 150kV), irrespective of alloy tem-
per, as long as larger particles are avoided
in the probed volume. In the mentioned al-
loy series, only magnesium can (in special
alloys occasions) be suspected to be present
at sufficiently high solute concentrations to
motivate the use of a slightly higher ␭
value. Other solute atoms will have a more
negligible effect at the concentrations that
are possible in the mentioned alloys series.
Fine precipitates and dispersoids in the
mentioned series will be present at volume
densities that are comparable to or smaller
than in the Al–Mg–Zn sample studied in
this work. Accuracies and appropriate ␭
values for more strongly alloyed systems,
such as, for example, Al–Li cannot easily be
predicted by the present approach.
341
CONCLUSIONS
Pairs of EELS and CBED measurements
were compared to determine the mean-
free-paths ␭ for inelastic electron scattering.
EELS spectra were acquired with the TEM
in image mode and without any objective
aperture inserted. With the procedures
used, ␭ values could be determined with an
accuracy of better than 5% for four differ-
ent aluminium alloys. The value for pure
aluminium was determined as 119 ⫾ 5nm
at an incident electron energy of 150keV.
For the most common alloys with moderate
amounts of alloying elements, no significant
changes in the mean-free-path values are
measured within the attainable accuracy of
the method. An alloy with 5% Mg in solid
solution exhibited a ␭ value of 125 ⫾ 4nm.
342
With the given accuracy, the measured
shift cannot be termed significant at the ␴
level. Still, one expects in this alloy an in-
crease in ␭ of 2–3% compared to the pure
alloy, owing to a lower density of conduc-
tion electrons and a corresponding lower
probability for plasmon excitation. On the
basis of our experimental results, we esti-
mate a single EELS measurement of thin-foil
thickness to have an absolute accuracy of
better than 8% if ␭ is known within 5%, the
difference coming from inaccuracy in extrac-
tion of the zero-loss intensity from the EELS
spectrum. Using the outlined procedure, the
EELS method for thickness measurements is
very robust, with thickness results being in-
sensitive to experimental details such as in-
clination of the thin-foil specimen or crystal-
lographic orientation/diffracting conditions.
In cases where no explicit calibration of ␭ has
been made, for most 1xxx, 2xxx, 3xxx, 5xxx,
6xxx, and 7xxx series alloys a value of ␭ ⫽
120 ⫾ 10nm can be used (at 150kV), irrespec-
tive of alloy temper.
References
1. D. B. Williams: Practical Analytical Electron Micros-
copy in Materials Science. Philips /Verlag Chemie
International, Weinheim, Basel (1984).
2. D. B. Williams and C. B. Carter: Transmission Elec-
tron Microscopy. Plenum Press, New York (1996).
3. P. M. Kelly, A. Jostsons, R. G. Blake, and J. G.
Napier: The determination of foil thickness by
scanning transmission electron microscopy. Phys.
Stat. Sol. (a) 31:771–780 (1975).
4. S. M. Allen and E. L. Hall: Foil thickness measure-
ment from convergent-beam diffraction patterns.
An experimental assessment of errors. Philos. Mag.
A46:243–253 (1982).
5. R. C. Ecob: Comments on the measurement of foil
thickness by convergent beam electron diffraction.
Scripta Met. 20:1001–1006 (1986).
6. J. C. H. Spence and J. M. Zuo: Electron Microdiffrac-
tion. Plenum Press, New York (1992).
7. R. F. Egerton: Electron Energy Loss Spectroscopy in
the Transmission Electron Microscope. Plenum Press,
New York (1986).
8. A. Bardal, I. Lindseth, H. E. Vatne, and E. Nes:
Dislocation Densities, Subgrain Sizes and Aub-bound-
ary Misorientations Within the Different Texture
Components of Hot-Deformed AlMgMn. Proc. 16th
Risø Int. Symp. on Mater. Sci.: Microstructural and
A. Bardal and K. Lie
crystallographic aspects of recrystallization, N.
Hansen, D. Juul Jensen, Y. L. Liu, and B. Ralph,
eds., Risø national laboratory, Roskilde, pp. 261–
266 (1995).
9. T. Pettersen, A. Bardal, I. Lindseth, and E. Nes:
Characterization of Deformation Microstructure in
Hot Plane Strain Compressed Al–1Mg–1Mn. Proc.
ReX 96 - 3rd Intern. Conf. on Recrystallization and
Related Phenomena, T. R. McNelley, ed., ReX’96
International Advisory Board and Organizing
Committee, pp. 495–501 (1997).
10. T. Malis, S. C. Cheng, and R. F. Egerton: EELS log-
ratio technique for specimen-thickness measure-
ment in the TEM. J. Electron Microsc. Tech. 8:193–
200 (1988).
11. G. A. Botton, G. L’Esperance, C. E. Gallerneault,
and M. D. Ball: Volume fraction measurement of
dispersoids in a thin foil by parallel energy loss
spectroscopy. Development and assessment of the
technique. J. Microsc. 180:217–229 (1995).
12. J. C. Ashley and R. H. Ritchie: Double-plasmon ex-
citation in a free-electron gas. Phys. Stat. Sol. 38:
425–434 (1970).
13. H. Raether: Excitation of plasmons and interband
transitions by electrons. In: Springer Tracts in Mod-
ern Physics, vol. 88. Springer, Berlin (1980).
14. P. E. Batson and J. Silcox: Experimental energy-
loss function, Im[⫺1/ε(q,␻)], for aluminum. Phys.
Rev. B 27:5224–5239 (1983).
15. D. B. Williams and J. W. Edington: High resolu-
tion microanalysis in materials science using elec-
tron energy-loss measurements. J. Microsc. 108:
113–145 (1976).
16. P. Schattschneider and P. Pongratz: Coherence in
energy loss spectra of plasmons. Scanning Microsc.
2:1971–1978 (1988).
17. R. F. Egerton and Z. L. Wang: Plural-scattering de-
convolution of electron energy-loss spectra re-
corded with an angle-limiting aperture. Ultrami-
croscopy 32:137–148 (1990).
18. E. A. Stern and R. A. Ferrell: Surface plasma oscil-
lations of a degenerate electron gas. Phys. Rev. 120:
130–136 (1960).
19. C. J. Powell and J. B. Swan: Effect of oxidation on
the characteristic loss spectra of aluminum and
magnesium. Phys. Rev. 118:640–643 (1960).
20. R. H. Ritchie and A. Howie: Electron excitation
and the optical potential in electron microscopy.
Philos. Mag. 36:463–481 (1977).
21. D. J. Eaglesham and S. D. Berger: Energy filtering
the “thermal diffuse” background in electron dif-
fraction. Ultramicroscopy 53:319–324 (1994).
22. R. F. Egerton and K. Wong: Some practical conse-
quences of the Lorentzian angular distribution of in-
elastic scattering Ultramicroscopy 59:169–180 (1995).
23. F. Lenz: Zur streuung mittelschneller Electronen in
kleinste Winkel. Z. Naturforsch. 9a:185–204 (1954).
Measuring Thickness of Aluminum Alloy Thin Foils with EELS
24. R. D. Leapman, P. Rez, and D. F. Meyers: K, L, and
M shell generalized oscillator strengths and ion-
ization cross sections for fast electron collisions. J.
Chem. Phys. 72:1232–1243 (1980).
25. Gatan EL/P Software, version 3.0. Gatan Inc.,
Pleasanton, CA (1995).
26. J. Taftø and O. L. Krivanek: Characteristic energy-
loss from channeled 100 keV electrons. Nucl. In-
strum. Methods 194:153–158 (1982).
27. H. Kohl and H. Rose: Theory of image formation
by inelastically scattered electrons in the electron
microscope. Adv. Electron. Electron Phys. 65:175–
200 (1985).
28. H. M. Simpson and R. L. Chaplin: Damage and re-
covery of aluminum for low-energy electron irra-
diations. Phys. Rev. 185:958–961 (1969).
29. A. Wolfenden: Electron radiation damage near the
threshold energy in aluminum. Rad. Effects 14:225–
229 (1972).
343
30. F. R. Castro-Fernandez, C. M. Sellars, and J. A.
Whiteman: Measurement of foil thickness and ex-
tinction distance by convergent beam transmis-
sion electron microscopy. Philos. Mag. A 52:289–
303 (1985).
31. P. A. Doyle and P. S. Turner: Relativistic Hartree-
Fock x-ray and electron scattering factors. Acta
Crystallogr. A24:390–397 (1968).
32. R. F. Egerton, Y.-Y. Yang, and S. C. Cheng: Char-
acterization and use of the Gatan 666 parallel-
recording electron energy-loss specrometer. Ultra-
microscopy 48:239–250 (1993).
33. C. B. Boothroyd, K. Sato, and K. Yamada: The De-
tection of 0.5 at % Boron in Ni3Al Using Parallel
Energy Loss Spectroscopy. Proc. XIIth International
Congress for Electron Microscopy, L. D. Peachey
and D. B. Williams, eds., San Francisco Press, San
Francisco, pp. 80–81 (1990).
Received December 1998; accepted November 1999.