Section III

Analytical methodology
Nuclear Instruments and Methods in Physics Research B77 (1993) 95-109
North-Holland
NIIIMB
Beam Interactions
with Materials&Atoms

Quantitative PIXE microanalysis of thick specimens

J.L. Campbell, D. Higuchi, J.A. Maxwell and W.J. Teesdale
Guelph - Waterloo Program for Graduate Work in Physics, University of Guekh, Guelph, Ontario NlG 2W1, Canada

Methods of standardization in quantitative micro-PIXE analysis are reviewed and various issues that bear on analytical accuracy
are explored; pertinent recent work on Si(Li) X-ray detector response is included and some geochemical examples are drawn upon.
Extension of the GUPIX software to deal with multilayer targets, including secondary fluorescence within and between layers, is
reported, analytical examples include alloy foils and multilayer solar cell structures.

1. Introduction the specimens, and contain the particular elements

In an elegant review of standardization methods in
electron probe microanalysis, Newbury [ 11worked from
the general functional relationship between the inten-
sity I(Z) of a particular X-ray of an elemental con-
stituent Z of the specimen and its concentration C,:
I(Z) =f(C,, ME, IE);
here ME represents matrix effects, i.e. the influence of
constituents other than Z upon I(Z); IE represents
instrumental effects such as detection efficiency. Our
intent here is emulate for pPIXE the development
presented by Newbury for EPMA. Examples of both
major element and trace element analysis will provide
illustrations and will identify some residual problems in
PIXE quantitation. Extraction of a desired concentra-
tion C, via eq. (1) from a measured X-ray intensity
I(Z) is predicated upon the reliable extraction of I(Z)
from a Si(Li) X-ray spectrum that may be rendered
very complex by a multiplicity of overlapping peaks.
This necessitates attention to both intrinsic and extrin-
sic effects that cause Si(Li) line shapes to deviate from
the Gaussian ideal.
Although various references are cited, this article is
not intended as a review paper covering the entire
literature, but rather as a distillation of recent experi-
ence.
2. Standardization
2.1. General principles
Newbury [l] discussed four means of solving eq. (1).
At the empirical extreme, working curves relating I(Z)
to C, may be constructed if one can obtain or synthe-
size a suite of standards which share identical ME with
that are of interest in accurately known concentrations.
These standards are analyzed under precisely the same
conditions as the specimen and so the IE need not be
quantitatively known. Specimen concentrations are
read off the curves by interpolating at the observed
X-ray intensity values. This is practical in a few cases;
bone specimens, for example, can be mimicked by
(1) hydroxy-apatite [2] with appropriate trace elements
incorporated. But for a proton microprobe facility such
as ours, dealing with a wide range of geochemical and
materials science specimen types, the working curve
method is far from feasible. And it would be quite
impractical for some of the more complex applications,
e.g. multilayer specimens, with which p,-PIXE is now
(dealing.
Some flexibility is gained by working with the meas-
ured ratio SZ,R of “sensitivities” I(Z)/C, relative to
that measured for a reference constituent (I(R)/C,),
which is usually a high-concentration element. The
SZ,R values have to be determined by measurements
on known multielement standards and the sensitivities
obviously are subject to matrix effects. This approach
offers the possibility of some degree of cancellation of
matrix effects between the elements Z and R, but the
cancellation is only partial.
A major shift away from these empirical approaches
is possible if the physics of the beam-specimen inter-
action is well enough known (and an accurate data
base exists) that ME may be calculated. In EPMA the
so-called ZAP approach provides the ratio of ME
between a specimen and some standard, whose similar-
ity to the specimen need only be general rather than
precise. In PIXE, which actually better satisfies the
stated condition, ME may be calculated either in abso-
lute terms for the specimen or as a ratio between
specimen and standard. Now that standards need not
mimic the specimen precisely, flexibility is much in-

0168-583X/93/$06.00 0 1993 - Elsevier Science Publishers B.V. All rights reserved III. ANALYTICAL METHODOLOGY
96 J.L. Campbellet al. / PIXE microanalysisof thick specimens
creased; a wide range of specimen types may be han-
dled with a very small suite of well-chosen standards.
The greatest flexibility is achieved by reverting to a
“first principles” approach that abandons standards
entirely. This too is possible in p,-PIXE, given the
simple physics and the excellent data base. But it
demands absolute measurement of proton beam flux,
which is less straightforward than superficial examina-
tion might suggest. At an early stage in thick target
PIXE’s development, Clayton 131 demonstrated lo-
15% systematic errors in such analyses based on first
principles. The discrepancies were a function of speci-
men type, and Clayton corrected them by analyzing
reference materials of the appropriate type. By so
doing, he effectively demonstrated the efficacy of the
third of Newbury’s approaches, as outlined in the pre-
vious paragraph.
2.2. The matrix correction approach
The PIXE version of eq. (1) has appeared fre-
quently in the literature; it results from an integration
of the X-ray intensities recorded from each point on
the linear path, within the specimen, of the decelerat-
ing proton:
X
(2)
In this equation, for the K shell case: Z(Z) is the
intensity in a particular K X-ray line, usually but not
necessarily the dominant K, or Kal; Np is the number
of protons of energy E, incident at angle a! to the
normal upon the specimen, and uz(E) the ionization
cross-section for the K shell; wz is the corresponding
fluorescence yield and b, the fraction of K X-rays in
the selected line; N, is Avogadro’s number; A, is the
atomic mass of element Z; S(E) is the proton stopping
power; 6ro is the X-ray take-off angle; flci is the
absolute efficiency of detection, which involves both
Si(Li) detector solid angle and intrinsic efficiency of
the silicon crystal; tZ is the attenuation of the X-rays in
any absorber (filter) placed between specimen and
detector; p/p is the attenuation coefficient within the
specimen matrix for the X-ray line of interest.
The L and M-shell equivalents are essentially the
same, albeit with the additional complication of sub-
shells; they thus involve the Coster-Kronig probabili-
ties that transfer the proton-induced vacancies among
subshells of a major shell prior to X-ray emission.
The practical unit of integrated beam charge being
the microcoulomb (l.&), we define lr(Z, Cz,,) as the
computed or theoretical X-ray intensity per PC per
steradian and per unit concentration. The Cz,, are the
concentrations of the major elements which comprise
the specimen matrix and thus determine ME. Then the
measured intensity for Q ~J,Cof charge is
There are several important points to note. First, the
IE, viz. the terms 0, &. and tz, factor out; they will
therefore cancel in the comparison of specimen (SP)
and standard (ST) for X-rays of a particular element
z:
IS’(Z) c;p I,“‘(Z, Cz,,)
___=-
IS’(Z) c;= I,“(& Cz,,) . (4)
In addition, measurement of Q needs to be correct
only in relative terms between specimen and standard.
The ME are represented by the term I&Z, Cz,,),
which involves both the compound stopping power and
the compound attenuation coefficient, which are con-
centration-weighted sums’ of these quantities over the
major constituents. If the C,,, subset is known, e.g.
from prior EPMA analysis, then the I,(Z, C,,) are
immediately calculable, and eq. (4) offers a one-step
solution for the trace element concentration subset
CZ,t. The further advantage of this ratio approach for
each Z-value of interest is the additional cancellation
of fluorescence yields and relative X-ray line intensi-
ties, and the partial cancellation of ionization cross-
sections within the integrals. The degree of partial
cancellation of stopping power and attenuation coeffi-
cients will increase with the similarity of the specimen
and the standard. We have discussed the accuracy of
the data base in some detail elsewhere [4] and with one
exception will not repeat that here. We simply note
that the data base is sufficiently well known that a
generic, as opposed to a specific, similarity between
specimen and standard will suffice. The exception per-
tains to elements of very low atomic number, which are
of increasing interest to PIXE analysts, and will be
dealt with in section 3.4.
The situation is more complex, but entirely tractable,
when the major element concentrations Cz, are un-
known, a situation that we encounter with ‘increasing
frequency in our geochemical work. The approach here
is to begin by estimating a set of values for the Cz,,,
compute the corresponding I,(Z, C,,,), fit the meas-
ured X-ray spectrum by least squares to obtain I(Z),
and via eq. (4) deduce a new set of C,,,. This process
is iterated until a self-consistent set of Cz,, is ob-
tained. Computing overheads have long ceased to be a
deterrent to this inclusion of one iterative process
(spectrum fit) within the iterations of another (con-
centration determination).
 J.L. Campb~~ et at. / PIXE ~icroa~a~~isofthick specimens 91

Eq. (4) presents a choice as regards the standard.
We indicated above that, given an accurate data base
from which to calculate ME, only a general as opposed
to a close resemblance was necessary between speci-
men and standard. In this spirit we use synthetic sul-
fide minerals, containing known quantities of one or
two trace elements, to standardize measurements upon
a wide variety of oxide, silicate and other minerals.
However if the concentration of one major element in
the specimen is known (e.g. from EPMA) then the
specimen itself may serve as standard; an example is
the analysis by Rogers et al. [5] of the NIST standard
610 (trace metals in glass), effected relative to the
major constituent, calcium. This confers the further
advantage of cancelling any residual errors in proton
stopping powers, which are now identical in specimen
and standard.
The iterative solution discussed above for major
elements has been incorporated in the GUPIX soft-
ware package (GUPIX92 is the current version) for
fitting spectra and providing concentrations 161.A use-
ful concept, due originally to Lagarde et al. [7], is that
the matrix may include an “invisible element”, i.e. one
whose X-rays are too low in energy to appear in the
spectrum. The condition is then imposed that all con-
centrations sum to 100% and the invisible element
concentration is iterated with the others in the compu-
tation of concentrations. GUPIX92 allows there to be
more than one invisible element, with the necessary
proviso that their ~ncentrations remain in a specified
ratio; this would permit for example the presence of
SiO, as a defined species. EPMA practice suggests a
useful variant for the specific case of the wide range of
silicate minerals where the elements are all present in
oxide form. This is to deduce the oxygen contribution
for each oxide by applying the stoichiometric ratio to
each cation concentration; one then can compare the
total of all element concentrations to 100% as a fur-
ther check of the analysis.
The characteristic X-rays of major elements may
induce significant secondary fluorescence both of trace
elements and other major elements. We have incorpo-
rated a full treatment of secondary fluorescence in
GUPIX92. Its accuracy has been examined elsewhere
[S] and we have further comments later in this regard.
3. Major element anaiysis: practical aspects
Micro-PIXE has to date been concerned much more
frequently with trace elements than with major ele-
ments, the latter frequently being predetermined by
EPMA in a complementary use of the two methods.
However, geochemists (for example) who have to pur-
chase time outside their laboratory for both techniques
may well be accommodated in full by micro-PIXE, to
their obvious economic benefit.
The advantages and disadvantages of the standard-
ization approach embodied in eq. (4) i.e. of an ap-
proach based on computation of ME and cancellation
of IE are now explored. The examples used are mainly
from our own materials science and geochemistry expe-
rience.
3.1. Attenuation and secondary fluorescence uncertain-
ties
Pure single-element standards are convenient and
attractive in that they preferentially transmit their own
characteristic X-ray lines, thus minimizing the impact
of X-ray attenuation coefficient errors on the analysis.
In our work on silver and platinum-group elements as
traces in sulfide minerals, it is sometimes necessary to
examine major element variations. An example is iron
and nickel in pentlandite, where there is interest in a
smuggested [91 linear correlation between the trace Pd
content and the Ni/Fe ratio. The major and trace
analyses can be effected in a single measurement with

Table 1
Results of PIXE analyses of Fe/Ni alloys. Cases A use theoretical p/p for Fe; B uses measured I( /p
Standard Element Reference Concentration [%I 10
PIXE (A) PIXE (B)
NIST SRM 1159 (thick target) Si 0.5 0.24 0.24 0.01
Cr 0.069 0.07 0.07 0.003
Mn 0.3 0.29 0.29 0.01
Fe 51.0 51.7 51.2 0.1
Ni 48.2 47.3 48.4 0.1
lnvar FOBS a (10 pm thickness) Ti - 0.1 0.1 0.005
Cr 0.12 0.12 0.004
Mn 0.34 0.34 0.007
Fe 64.0 63.2 63.4 0.1
Ni 36.0 34.2 35.3 0.1
a Mn + Si + C < 1%. hence Fe and Ni slightly below nominal values.

III. ANALYTICAL METHODOLOGY

98 J.L. Campbell et al. / PIXE microanalysb of thick specimens

3 MeV protons, using a 0.35 mm Al absorber to prefer-
entially suppress the dominant Ni and Fe peaks. Due
to the absorber, no Si X-rays are detected and so Si is
treated as an “invisible element”, the total Csi + C,
+ C, being 100%. The standards are pure iron and
nickel targets.
This appears straightforward, but we felt that the
procedure warranted testing. An appropriate way to do
this is by analysis of the NIST nickel-iron alloy stand-
ard reference material (SRM 1159>, since it has 48% of
nickel and 51% of iron. We find consistently (see table
1) that we underestimate the nickel concentration and
overestimate that of iron. This is unlikely to arise from
inhomogeneity in the standard, since the microbeam is
rastered over an area 0.4 x 0.4 mm. We have observed
the effect before [8], in conventional broad-beam
thick-target PIXE with a different detector. And (see
section 5 below) we observe it also in thin Invar alloy
foils.
This discrepancy is not attributable to either ioniza-
tion cross-sections or stopping powers since we are
dealing with elements which are separated by only two
units of atomic number; any discrepancies would be
the same for both, and would tend to cancel. Suspicion
therefore falls on the attenuation coefficients p/p, and
specifically on the largest of these, that for the nickel
K, line in iron, since this quantity strongly influences
both the matrix attenuation for NiK, and the sec-
ondary fluorescence of iron. While the p/p theoretical
data base used here [lo] is felt to be accurate generally
to 2-3%, its authors warn of anomalies at X-ray ener-
gies above and within 1 keV of an absorption edge. A
detailed experimental study of this region for iron [II]
suggests a p/p value for NiK, that is 6% above
theory. Fig. 1 presents a comparison of the measured
and theoretical p/p values in the vicinity of the iron K
absorption edge. When the g/p values from ref. [ll]
are substituted, together with the corresponding photo-
electric cross-sections, in the GUPIX data base [6],
agreement between our data and the recommended
concentrations improves significantly. We do not have
similarly extensive experimental p/p values for attenu-
ation in nickel, and so the correction of the data base
is necessarily incomplete.
The point that we wish to demonstrate is simply
PIXE’s susceptibility to error in attenuation coeffi-
cients at X-ray energies near absorption edges. This
kind of error is likely to be encountered for other
element pairs spaced by one or two Z-values. GUPIX
has therefore been modified to provide a user-friendly
means by which a user may substitute preferred experi-
mental p/p values for those in the data base. In
addition, these observations argue against the use of
semiempirical formulae for attenuation coefficients,
since these cannot accommodate the oscillatory be-
haviour (see fig. 1) near an absorption edge.
3.2. Light major element amlysis at low proton energy
If the matrix is comprised mainly of low-Z elements,
then the presence of any absorber between specimen
and detector is precluded and the intense X-rays of the

Energy (keV)
Fig. I. Difference A between measured and XCOM data base values of X-ray attenuation coefficient in iron at energies near the K
absorption edge.
 J.L. Campbell et al. / PIXE microanalysis of thick specimens 99

major elements will dominate the throughput of the

electronic system to an extent that very few trace
element X-rays will be recorded. Simultaneous deter-
mination of C,,, and Cz,t is therefore not possible.
Further problems are the deleterious effects of scat-
tered protons on the Si(Li) detector response, together
with long-term damage to the crystal [12]. We there-
fore derive 0.75 MeV protons from a 1.5 MeV Hz
beam, which is transported under the same beam opti-
cal conditions as the 3 MeV H+ beam used in subse-
quent trace element analysis. At 0.75 MeV the scat-
tered protons do not penetrate the beryllium window
of our detector.
The resulting high counting rates necessitate in-
creasing the specimen-detector distance and inserting
a 0.3 mm collimator. GUPIX92 offers the facility of
storing the major element concentration files that are
obtained in a suite of 0.75 MeV measurements; these
are recalled by GUPIX92 in order to generate matrix
corrections when it is processing subsequent trace ele-
ment data taken in the same sequence at 3 MeV
proton energy.
Since pure single-element standards are not avail-
able for all light elements, recourse may be necessary
to compound standards. As an example, table 2 shows
results for analyses of the two minerals sanidine (spec-
-~~,“‘;““~““;,“1
1

trum in fig. 2) and bustamite where the mineral albite ENERGY (KeV)
was employed as a standard for the elements sodium, Fig. 2. PIXE spectrum of sanidine recorded using 0.75 MeV
aluminum and silicon; single-element standards were protons.
used for higher-Z elements. The oxygen concentration
was found by assigning the normal valence to each
cation species and deducing the oxygen corresponding to the content of sodium, aluminum, etc. Not only is
there good agreement between recommended and
Imeasured concentrations for the various elements, in-
cluding oxygen, but the total element content in each
Table 2 case is close to lOO%, which provides a check on
u-PIXE analyses of the minerals bustamite and sanidine using consistency.
an albite a standard; R = recommended, M = measured The advent of ultrathin window (UTW) detectors
Element Concentration [%I
permits recording of the oxygen lines and averts the
need for an invisible element in the computation. The
Bustamite Sanidine
two prices paid for this complete coverage of the X-ray
R M R M spectrum would be (i) a smaller detector (since only
0 38.39 38.5 kO.4 46.28 45.4kO.3 small-diameter windows are feasible) and (ii) entry of
Na 0.04 0.11+0.1 2.23 2.2 * 0.15 scattered protons. The first is of no matter for major
Mg 0.13 0.14kO.03 - 0 elements since count rates remain adequate, but the
AI 0.02 0 9.93 9.710.12 decreased efficiency would be a detriment in trace
Si 22.46 22.3 +O.l 30.23 29.7 i 0.2 analysis of the same specimen. Musket [12] has demon-
K 0.05 0 10.05 10.3 + 0.2 strated removal of the scattered protons in conven-
Ca 13.56 13.8 +0.3 0.14 0
tional PIXE by interposing an arrangement of compact
Mn 18.83 19.6 +0.9 - 0
Fe 6.32 6.1 +0.9 - 0
permanent magnets (see fig. 3) between specimen and
Zn 0.20 0 _ 0
detector. It remains to be determined if this solution is
Ba 0 0.98 1.0+0.2 compatible with a microbeam. If it proved to be com-
patible, one could conduct both major and trace ele-
Total 100 100.6 99.85 98.3 ment analysis with 3 MeV protons, which would be
a AIbite concentrations are: 48.76% 0; 8.60% Na; 10.34% Al; much more convenient than the mode described here,
32.03% Si; 0.18% K 0.09% Ca. which involves switching between two energies.

III. ANALYTICAL METHODOLOGY

100 J.L. Campbell et al. / PIXE microanalysis of thick specimens

3.3. IE for light elements

For elements of low atomic number the cancellation

of intrinsic detector efficiency implicit in eq. (4) is very
important. Various factors cause significant uncertain-
ties in the intrinsic efficiency.
The beryllium window thickness is unlikely to be
known precisely; a 10% error at 8 pm thickness causes
8% error in the recorded intensity of sodium K X-rays.
Any ice layer [13] will cause significant and time-de-
pendent attenuation which is difficult to estimate, ne-
cessitating regular checks via for example the intensity I I I I I I
2 4 8 10 12
ratio of copper L and K X-rays, and subsequent deic- EN& (keV)
ing, if necessary. Another and a more complex issue, is Fig. 4. Comparison of measured tail-to-Gaussian intensity
that of the widely misunderstood frontal silicon layer in ratios with the model of eq. (5) (upper curve), and a model
which charge collection is incomplete. The thickness based on electron escape only (lower curve).
tICC of this layer is conventionally determined by re-
garding the ICC layer as an inert silicon absorber of
photons, and measuring the ratio of intensities in the
Gaussian (A,) and the low-energy tailing portion (A,) region penetrate back into the ICC layer. In this model
of a peak. For K X-rays of elements like iron and the tail-to-peak ratio is
copper a significant portion of the overall tailing is due
to the intrinsic Lorentzian X-ray line shape [14], and if 4
-= [l - exp(-k%C)I +J
this fact is ignored an overestimate of the ICC layer AG exp(--CLtICd-J ’
thickness will result. This necessitates using photons
where
from a monochromator, as done by Krumrey et al. [15],
who showed that, rather than being well-defined, the
J=lj:_+“~O.5(1-~)Pe~(-~r)dr, (6)
tICC value deduced in this manner was a linearly
increasing function of photon energy. Subsequent to
and R is the photoelectron range. This expression
verifying this observation, we have developed [16] an
differs from the conventional one in the appearance of
extended model for the ICC layer. Every X-ray interac-
the term J; it provides an excellent fit to the photon
tion in the layer is taken to result in a degraded event
energy dependence of AT/A, (fig. 4) and hence a
due to imperfect collection of the electron-hole pairs
unique t,,, value.
liberated; further degraded events arise when photo-
If the ICC thickness value were determined by the
electrons from interactions in the subsequent “perfect”
conventional approach (J absent from eq. (5)) with
monoenergetic 5.9 keV K, X-rays of manganese, then
WI the rIcc value obtained would be a factor 1.7 times
larger than the true t,,,; use of the erroneous value
/Out (to Si(Li) ) would then cause an error of some 6.4% in efficiency
for the sulfur K X-ray line. If a radionuclide 55Fe
source were used to provide the Mn K X-rays, the error
would be even larger, due to the enhancement of the
low energy tailing by the Lorentzian contribution.
In sum, the rapidly changing efficiency below 5 keV,
subject to large errors, is best cancelled from the
analysis by using standards as per eq. (4). Where sin-
gle-element standards do not exist, e.g. Na, S, P, there
are convenient mineral standards containing these ele-
ments in the form of simple compounds.

3.4. Data base for light elements

Distance (inches)

Fig. 3. Compact magnetic deflection unit for removal of The GUPIX data base employs the most sophisti-
scattered beam particles from the axis of a UTW Si(Li) cated ECPSSR K ionization cross-sections available to
detector [El. date [17], together with Bambynek’s 1984 compilation
 J.L. Campbell et al. / PIXE microanalysis of thick specimens
. 1.74 keV and then drops by a factor of 10 for the K
. X-rays of Si, Al, etc., spectrum fitting has to contend
t
R . .
1.06
.
.
1.00
. .
Fig. 5. Ratio R of reference X-ray production cross-sections
[19] to those predicted by ECPSSR theory [17] and the K
fluorescence yields of Bambynek [18].
of K fluorescence yields as reported by Hubbell [18].
Paul’s useful set of reference cross-sections [19] is
based on critical assessment of all measured data,
interpreted via Krause’s fluorescence yield compilation
[20]. In fig. 5 we compare the X-ray production cross-
sections (~~0~ from GUPIX with the Paul/Krause
values i.e. the best measured values. Below 2 = 20
there is a widening discrepancy which merits investiga-
tion. In the PIXE analysis context, comparison of spec-
imen and standard for each element using eq. (5) will
ensure significant (although not complete) cancellation
of errors inherent in uK and wK.
GUPIX does not deal with elements lighter than
sodium. Recent very careful K X-ray production meas-
urements for 6 I Z I 12 by Yu et al. 1211using 0.75-4.5
MeV protons suggest that the ECPSSR cross-sections
somewhat underestimate measured values, the discrep-
ancy widening as Z decreases; carbon, however, de-
parts from this well-defined trend, an observation that
may suggest problems with the fluorescence yields [20]
used. These yields are of order 0.001 and directly
measured values are presumably subject to large uncer-
tainty. Willemsen and Kuiper [22] have reported meas-
urements of the K X-ray yields of N, 0, and F; they
compared these to predictions of the BEA theory using
the same fluorescence yields, and found large discrep-
ancies. When K-shell fluorescence yields deduced from
experimental measurements of the complementary
Auger yields were used, the agreement between meas-
ured and predicted X-ray intensities was much better,
although there is still uncertainty due to possible ef-
fects of chemical bonding upon the wK values. All of
this emphasises that it is better to use standards as
opposed to relying on the data base, with the proviso
that there could even be differences in fluorescence
yield between specimen and standard.
101
3.5. Detector line shapes for light elements
Since low-energy tailing is maximum at X-ray ener-
gies just above the silicon K absorption edge energy of
with rather variable tailing contributions in light ele-
ment work. Unless it is accounted for in systematic
fashion in a spectrum-fitting code, this tailing can
cause error in the intensities of weak peaks that are
superimposed on the tails of intense peaks of slightly
higher energy. A further problem noted by many au-
thors is a time dependence of the tailing intensity and
shape; we have observed this phenomenon and we
have noted that in our case it correlated with the
existence of an ice layer (observed via loss of efficiency
for low energy X-rays). When the ice layer was re-
moved using the heating device that is integral to Link
Analytical Si(Li) detectors, the tailing became mini-
mum and constant. This suggests that the variable
tailing is connected to electric field distortion caused
by the ice layer.
In addition, preferential absorption of the back-
ground continuum in the silicon ICC layer produces a
“notch” in the backgr?und that can detract from accu-
racy of Si concentrations in specimens; this phe-
nomenon is well-known in EPMA. Musket [13] has
drawn attention to a similar phenomenon in oxygen
determination caused by differential absorption of the
background continuum in the oxygen of the ice layer;
X-rays of energy just above the K edge (0.532 keV) are
strongly absorbed, while those having energies just
below the edge are preferentially transmitted. The
resulting “bump” in the background may be inter-
preted as an excess of oxygen in the specimen if the
specimen is fitted by least-squares programs which
assume monotonic background. This is only observed
for windowless or ultrathin window detectors. While
these detectors are advantageous in providing oxygen
data, there are some risks as regards the accuracy of
that data if the detector is not adequately deiced. The
alternative solution for light major elements in miner-
als is then to retain Be-window detectors and to adopt
the conventional EPMA procedure of complementing
aach observed element (Na, Mg, Al, . . . ) with oxygen as
determined by its oxide stoichiometry; this ensures
proper matrix corrections.
3.6. Specimens of finite thickness: use of backscattered
protons
The above discussion is predicated upon specimens
that are sufficiently thick to stop the proton beam or at
least to reduce its energy to a level at which X-ray
production is negligible; this is typically 50 pm in our
geochernical work. However, if the beam penetrates
III. ANALYTICAL, METHODOLOGY
102 J.L. Campbell et al. / PIXE microanalysis of thick specimens

_”
4 6 6 10 12

ENERGY (keV)
Fig. 6. PIXE spectrum of a Cd,Hg,_,Te film deposited on an indium-tin-oxide coated glass substrate.

400 jc j” jca
BS
303

,” 250
:
I

c
3 200

0
” 150

‘..
:_ ., ... ,.., ..‘, ‘. .‘. ..

50

I ,_. ..’
0
1500.3 L750.0 2000.0 2250.0 2500.0 2750.0 3ooo.a

energy/ keV

energy/ keV r-
Fig. 7. Backscatter and PIXE spectra of a fly-ash particle [23].
 J.L. Campbell et al. / PIXE microanalysis qf thick specimens 103

the specimen, then either the thickness or the emer-

gent energy is needed in order to compute ME. This
situation is best avoided, but use of a substrate that
emits an intense characteristic X-ray can provide a
solution. As an example, we have had to measure the
composition of multielement films deposited on glass
substrates and having unknown thickness. Fortunately
the high barium content of the glass results in the
presence of barium L X-rays in the spectrum, as shown
in fig. 6. The barium concentration is first determined
by analysis of the glass alone. Then a multilayer version
of GUPIX92 is used to process the spectrum from the
compound specimen. Different estimates of the film
thickness are made until one is obtained that repro-
duces the correct barium concentration in the sub- YI ,n
loo ““I ““““‘I “I ” ” ”
strate. The film element concentrations from that par- 10 15 20 25 30
ticular fit are then accepted. We develop this theme
further in section 5.
An alternative means of thickness measurement,
which simultaneously provides major element concen-
tration ratios is the spectroscopy of backscattered pro- 0
tons as demonstrated by Jaksic et al. [23]. While the
scattering is not pure Rutherfordian, there is a suffi- -10
1
cient data base of inelastic scattering cross-section to 10 15 20 25 30

enable the creation of a successful BS spectrum simu- ENERGY

lation program; this then provides quantitative meas- Fig. 8. Details of the fit to PIXE spectrum of pelletized
ures of thickness and of light element concentrations. CCU-1 reference material. The residuals are in units of one
The example we offer is that of a fly-ash particle shown standard deviation.
in fig. 7.

While some SRMs may be used directly (e.g. the

4. Trace element analysis: practical aspects NIST glass and the NlST Fe/Ni alloy mentioned ear-
lier), others exist in a form that is not immediately
We turn now to the more frequent application of amenable to PIXE use. An example is the CANMET
p-PIXE i.e. the determination of trace element con- reference copper concentrate CCU-1, which is a pow-
centrations in a known matrix. der mixture comprising 82% chalcopyrite, 9% pyrite,
9% sphalerite and a trace of pyrrhotite. Element con-
4.1. Trace element standards and homogeneity centrations are accurately known via an interlaboratory
analysis [24], in which atomic absorption of digested
The standardization approach embodied in eq. (4) samples was the main method. We took PIXE spectra
now demands a standard matrix of some general simi- from this material mixed with CFll cellulose (15%)
larity to the specimen, and containing each trace ele- and pressed under 25000 PSI to form pellets. To
ment of interest. compensate for expected inhomogeneity at the micron
These are various standard reference materials in level, the 10 ym wide beam was rastered over a 400 x
solid form which contain a suitably large number of 400 km area. The resulting PIXE spectrum, shown in
trace elements that one is able to interpolate with very fig. 8, shows the kind of problem that arises in such
little uncertainty for those missing. A degree of error standards. The fit, as judged by the residuals, is bad in
will be incurred in fitting the complex PIXE spectrum the pileup region of copper and zinc; as the microbeam
with many trace element contributions. The analysis of scans, it encounters “clumps” of zinc (i.e. sphalerite),
Rogers et al. 151 of the standard reference material causing a momentarily high count rate with excessive
NIST 610, a glass containing various rare earths, is a pileup. The “pileup element” method [25] used to fit
good example. Some authors will prefer to adopt the spectrum assumes constant count rate conditions;
standards in which there are fewer elements, and in it cannot cope with varying conditions. In a sequence
which the concentrations are at the minor as opposed of analyses of ten spots by this method, the error given
to the trace level, and to seek to supplement these with by the fitting program for major element concentra-
a reliable means of interpolation in Z. tions corresponded closely to the expected standard

III. ANALYTICAL METHODOLOGY

104 J.L. Campbell et al. / PIXE microanalysis of thick specimens

Table 3 Table 4
Variation in major element concentrations in pelletized X-ray intensity variation among ten replicate PIXE analyses
CCIJ-1 standard of synthetic pyrrhotites (Fess) containing trace elements at
0.1% concentration
Element Concentration [%I Standard deviation [%I
Internal External Beam spot diameter Standard deviation a [%I
[wnl Fe Cu Se Pd
Fe 30.80 0.36 2.4
cu 24.71 0.21 2.4 Standard A: Fe,S (Cu, Pd)
Zn 3.22 0.18 10.1 5x5 cTh 1.82 2.25 2.29
a, 0.29 1.84 2.10
50x50 uh 0.43 1.68 3.67
0.28 1.82 2.08
deviation from simple counting statistics. But the Standard B: Fe,S (Se, P:)
standard deviations for ten replicate measurements 5X5 "h 1.38 10.67 1.79
(see table 3) were considerably larger, and clearly sug- “e 0.27 1.27 2.48
gested a much greater heterogeneity for zinc than for 50X.50 uh 0.69 6.68 2.63
iron and copper. fie 0.27 1.22 2.44
Recognizing the need to homogenize geological a (TV= observed standard deviation, a, = expected standard
standard materials on the scale of the volume sampled deviation ( = square root of counts).
by the beam, Ryan et al. [26] fused US Geological
Survey rock standards to form glasses. In two cases
(BCR-1 and AGV-l), EPMA analysis demonstrated tion of heterogeneity. The GSP-1 material was
homogeneity in major and minor elements, but some markedly heterogeneous due to the distribution of
trace elements measured by p-PIXE at eight spots minute grains of zircon and quartz.
showed spreads in concentration more than twice the For these reasons we have preferred to conduct our
statistically expected range. The summed spectrum gave geochemical work with synthetic sulfide minerals pre-
concentrations that agreed to 1% on average with pared by CANMET [27]. But even these simple sys-
recommended values, lending credence to the observa- tems can be heterogeneous. Table 4 shows the varia-

(b)
105

ti
g lo4
iI
V
p: lo5
E

100 r IO0 t.
14 16 16 20 14 16 16 20

14 16 16 20 14 16 16 20

ENERGY ENERGY
Fig. 9. Details of fits to a zirconium PIXE spectrum: (a) no scattering step is included in the line shape, and yttrium is included in
the element list; (b) a scatter step is included.
 J.L. Campbellet al. / PIXE microanalysisof thickspecimens 105

tion in X-ray intensity from ten separate beam spots on

two synthetic pyrrhotites (Fe,S), one containing 0.1%
each of Cu and Pd, and the other 0.1% each of Se and
Pd. The observed standard deviation crh is compared
with the value qe expected from conventional square
root containing error. In each case the iron is inhomo-
geneous at the 5 p,rn scale and this persists even with a
50 pm spot. The trace elements Cu and Pd are homo-
geneous at both scales, but the Se is markedly inhomo-
geneous. This demonstrates the need to check even
simple standards for inhomogeneity.
The multielement standards prepared by Ryan et al.
[26] have the attraction of providing standardization
across a wide range of atomic numbers; it is straight-
forward to interpret the X-ray yields rS7.@) for absent
elements, facilitating use of eq. (4). On the other hand
standards such as our synthetic sulfides, with rather
high concentrations of just one or two elements, pro- 105

vide spectra very quickly and fitting these should incur

very little error; but it is then necessary to have a
IO’
reliable interpolation scheme, an aspect we take up in
section 4.4 below.
10*
4.2. Instrumental effects

Most trace element work involves higher-Z trace 10Z

elements in a lower-2 matrix, although the archaeome-
try area provides notable exceptions where clever use
has been made of compound critical filtering to im- 10’

prove the peg-to-background ratio of low energy lines

in the presence of intense higher energy lines [28]. One
is typically working with X-rays in the energy range
5-35 keV. The problem of intrinsic detector efficiency
becomes much more tractable than in the low energy
region (see section 3.3). The intrinsic efficiency is es-
sentially constant up to 15 keV and its falloff beyond
that is determined mainly by the silicon crystal thick-
ness, which may be accurately determined with conve-
nient radionuclide standard sources; depending on the
effort devoted to this task, & can be known with some
1% standard deviation.
Two further instrumental effects merit attention.
Both relate to the detector line shape. At the low
X-ray energies discussed in section 3.5, the departure
of line shape from the hypothetical Gaussian form is a
quasi-exponential low-energy tailing that arises from
loss of charge near the detector’s front surface. The
intrinsic effect decreases with increasing X-ray energy
and becomes negligible at 15 keV. At around 15 keV a
different form of tailing sets in, this being extrinsic in
origin; it arises from Compton scattering of X-rays in
the specimen and on nearby structures in the specimen
chamber. In our experience these tails are quite well
represented by truncated flat shelves, terminating at an
energy equal to the energy loss incurred in 180” scatter-
ing, provided the detector is collimated and thus does
LO@L. 7 ’ , I
20 25 30
ENERGY (kev)
Fig. 10. Fits to the PIXE spectrum of tin: (a) with no
Lorentzian contribution; (b) with a Lorentzian contribution
(r = 11.2 eV).
not “see” chamber structures responsible for multiple
scattering. In fig. 9 we illustrate the need to describe
this Compton tail in fitting a zirconium metal spec-
trum. If the tail is omitted and GUPIX is permitted to
include yttrium in its element list, it generates an
yttrium concentration around 450 ppm, which is spuri-
ous. This has necessitated careful attention to tail
description in our work on colour-banded zircons,
where yttrium may play a role in zone coloration.
The second issue is that of the intrinsic Lorentzian
shape of an X-ray line. In a special version of GUPIX
we have introduced the option of convoluting the
Gaussian plus tailing line shape with this Lorentzian
distribution to generate the Voigt function that truly
represents the line in question. It has been customary
in Si(Li) spectroscopy to neglect this effect entirely on
the grounds that ~rentzian widths are much smaller

III. ANALYTICAL METHODOLOGY

106 J.L. Campbellet al. / PIXE microanalysisof thick specimens
than typical detector resolution. The fits to a tin K
X-ray spectrum in fig. 10 show that this neglect is not
entirely justified even though the Lorentzian width of
11.2 eV is significantly less than the full width at half
maximum value of 300 eV. Fitting with Gaussian line
shapes causes a 2% underestimate of the I& line
intensity, as compared to using Voigtians. For zirco-
nium K,,, where the Lorentzian width is smaller, the
effect is 1%.
These Lorentzian effects probably do not justify in
routine analysis the computation time needed to prop-
erly represents the line shapes. But they should be
borne in mind as a source of small discrepancies.
4.3. The H due method
We and others have found the following method
rapid and convenient in performing large numbers of
analyses for trace elements in known or predetermined
matrices. Rather than standardizing with many ele-
ments and cancelling IE as in eq. (4) we determine the
properties of any Si(Li) detector very carefully, so that
the instrumental effects are accurately known, and
then we rely on a standard containing only one or two
trace elements.
Returning to eq. (3) we first recognize that if sec-
ondary electron suppression is not perfect then Q may
merely be proportional to collected charge rather than
equal to it; alternatively Q may be measured indirectly
e.g. by sampling beam particles scattered off a rotating
vane: in each case some correction factor fQ should
multiply the measured quantity Q so that Qf, is the
actual charge in FC. We combine the unknown fp and
the solid angle fl into an instrumental constant H that
effectively characterizes the geometry and charge cali-
bration of the PIXE system. Thus
I(Z) =~Q&,W,(-G Cz,,). (7)
A measurement of I(Z) for a standard containing
concentration C, of a single element then provides a
value for H, and this in principle standardizes the
instrument for all elements. It is interesting to note
that this is precisely the same as the approach taken by
Fei and Van Espen [29] in their recent work on quanti-
tative analysis by X-ray fluorescence.
Fig. 11 shows H-values obtained in this way using
pure single-element standards. Protons of 0.75 MeV
energy were used for K X-rays of nickel and titanium
and L X-rays of niobium, silver and tin, and 3 MeV
protons were used for K X-rays of nickel and higher-Z
elements. Particular attention was paid to determining
the efficiency and line shape of the Link Si(Li) detec-
tor used [16]. The result indicates that the data base
and the IE determination are reliable enough that a
single element is indeed capable of providing the pa-
rameter H. A similar demonstration of the constancy
.---/ / ,-._
i
Fig. 11. H values measured for a well-characterized Si(Li)
detector. The inverted triangles represent L X-ray data from
Nb, Ag, Sn using 0.75 MeV protons. The filled circles repre-
sent K X-ray data taken with proton energies of 0.75 MeV
(Ti, Ni) and 3 MeV (Ni and higher-Z elements).
of the instrumental constant in the X-ray fluorescence
analysis case can be found in ref. 1291.
There is in fig. 11 an indication of a slight decrease
in H for the K X-rays of elements of Z > 40. This is
pursued in fig. 12, where K X-ray data were taken for a
wide range of single-element standards using 3 MeV
protons and a different detector, whose 5 mm nominal
thickness should ensure constant efficiency over a wide
range of X-ray energies. These results demonstrate the
hazards of inadequate attention to IE. In fitting the K
X-ray spectra with the GUPIX code, Compton tails
were neglected for the elements in the atomic number
region 40 to 52, as were the effects of the finite
Lorentzian linewidths. The data in the lower portion of
the figure have had these effects incorporated, and
only now does H become a constant, as it should be.
Obviously the simplicity of the H value method can
result in error if exploited naively.
We observed that, at 3 MeV, L, X-rays give a
slightly but significantly different H value from that of
K, X-rays, which presumably reflects errors in the
ionization cross-section data base. We handle this in
GUPIX by the empirical manoeuvre of defining sepa-
rate parameters H, and H,. Finally in fig. 12 we
observe that the falloff in the H, value at low 2 arises
because we have not yet characterized this particular
detector’s line shape at low photon energies. We han-
dle this in GUPIX by allowing both H, and H, to be
functions of X-ray energy; they are measured for a few
elements, stored in files, and GUPIX interpolates for
other elements as needed.
The above has shorn us that for K X-rays, H is
indeed a well-defined quantity for a particular set of
targets, namely pure metal standards. With a much
more limited set of synthetic sulfide minerals contain-
 J.L. Campbell et al. / PIXE microanalysis of thick specimens 107

ing different trace elements we have also demonstrated Table 5

an acceptably well-defined H [27]. Our argument that H values measured a using metallic iron and iron sulfide

standards need not closely mimic specimens requires Hs Hi%

us to demonstrate that the same H value is obtained
Iron 0.0406+ 0.0002
for very different materials. Were we to find different Iron sulfide A 0.0410+ 0.0005 0.0403* + 0.0004
H values for metals and nonconducting targets, this - 0.0007
would suggest inadequate suppression of secondary Iron sulfide B 0.0404+ 0.0008 0.0395k 0.0008
electrons, and our arguments would fail on grounds of
a Each value is the mean of five measurements, with the error
imperfect charge measurement. To explore this we
representing the range of values.
measured H for metallic iron and for two carbon-
coated pyrrhotite (Fe,S) specimens using 3 MeV pro-
tons, with a 125 ym mylar X-ray absorber. As shown in
table 5, the H values were very consistent. This obser- a few standard elements suffice to define the slight
vation emphasises the university of the method. variation in H that results from an incomplete knowl-
In sum then, H, should be a constant for a well- edge of instrumental effects.
characterized system. Data base errors however de-
mand separate H, and HL values. If the detector
characterization is somewhat superficial, IE errors can 5. Multilayer specimens
be compensated for by permitting HK and H, to be
X-ray energy dependent. Taken to its extreme of With the standardization of homogeneous speci-
course, this argument would return us to the multiele- mens now rather well developed, there has recently
ment calibration implicit in eq. (4); but in practice only been increasing emphasis on more complex specimen
types and especially on those comprising several dis-
tinct layers each of which individually is homogeneous.
As a result k-PIXE has begun to supplement and
1.6
3 extend Rutherford backscattering of He ions, previ-
ously regarded as the ion beam technique of choice for
t
0 . J multilayer specimens.
8% l.*
z 1.4- . l * .
Rickards and Zironi [30] presented a detailed ana-
T+ . 0. lytic treatment of X-ray generation in a target compris-
x . .-
. .
. . . ing up to three layers each composed of one or ,more
defined elements. Given the layer thicknesses, which
are crucial in determining both the slowing down of
the proton and the attenuation of X-rays, they could
compute the intensity ratios of K, X-rays from ele-
ments in different layers. This necessitated computing
secondary fluorescence of an element in one layer by
10 15 20 25 X-rays emanating from a different layer, which was
Energy (keV) effected by an extension of the classical secondary
fluorescence formalism of Reuter et al. 1311.An appli-
cation that illustrated p-PIXE’s potential was determi-
nation of the thickness of a titanium film (0.02-2 rJ,m)
z
I . . deposited on a steel substrate. Fig. 13 demonstrates
e.*.., l ' . .. l*
," 1.4 .
'. . that at appropriate proton energies of a few hundred
:
x .
1 keV, a comparison of measured and predicted K,
X-ray intensity ratios provides the titanium thickness
with uncertainty of order 10%.
The sensitivity of this method derives from the
strong influence of the film thickness upon the sub-
strate X-ray intensity. Demortier [32] has also demon-
1.0 ~~~~‘~~~~I~~~,/~‘~‘I~~‘~I~~
0 5 10 15 20 25 strated this technique, his examples including films of
Fig. 12. H values for K, X-rays, measured using 3 MeV gold on copper, zinc on iron, and nickel on copper. In
protons. The upper data result from spectrum fits which the Zn/Fe case, for example a layer thickness range of
neglected both low-energy line shape distortion by Compton 2-35 pm corresponds to more than a hundredfold
scattering and the intrinsic Lorentiian linewidth. These ef- variation in the intensity ratio of Fe K, to Zn K, when
fects were included in computing the lower data set. 3 MeV protons are used, providing a sensitive thick-

III. ANALYTICAL METHODOLOGY

108 J.L. Campbell et al. / PIXE microanalysb of thick specimens
,011
” I I I
01 0.1 1.0
t, (mg /cm2)
Fig. 13. Calculated ratio Ti/Fe of K, X-ray intensities for a
Ti layer of thickness t1 on a steel substrate [30]. Proton
energies between 400 and 700 keV were used. The circles and
crosses are measured points for two samples (the dashed lines
represent a different choice of ionization cross-sections in the
data base).
ness measurement; RBS using the same protons can
only measure up to 15 Km.
The next degree of complexity is the elemental
analysis of a buried layer. One example is the treat-
ment of fluid inclusions in materials of Ryan et al. [33].
Here the major and trace element concentrations of
the host matrix are determined by conventional PIXE.
The depth and thickness of a buried layer i.e. a fluid
inclusion are then determined by optical methods.
When the beam is directed onto the inclusion, one has
an unknown layer at reasonably well-defined position
within a well-characterized host. Ryan et al. [33] used
the chlorine K,/K, ratio from aqueous inclusions
containing 30% NaCl as a check on the optically deter-
mined depth. This method provides an accurate trace
element analysis of the buried layer.
The variety of multilayer problems precludes writ-
ing a computer code that will deal with every case. But
if the code user is willing to do some repetitive work,
he can exploit p,-PIXE’s multilayer potential.
In extending GUPIX92 to cope with multiple layers
it was first necessary to modify our secondary fluores-
cence formalism from the thick target case to the case
of a single target of “intermediate thickness”. Our
published treatment [8] built upon the classical treat-
ment of Reuter et al. [31], avoided the hitherto used
approximations regarding jump ratios at absorption
edges, and employed a modern data base. It gave good
agreement with recommended concentrations when
used to analyze standard reference materials, and slight
differences in predictions relative to other treatments
supported our use of exact photoelectric cross-sections
as opposed to jump ratios. The modification used for a
keV
single “intermediate thickness” target was tested on
64% Fe/36% Ni foils of 10 km thickness. As table 1
shows, the results deviated slightly from expectation
until measured p/p values for NiKa in iron were
substituted into the GUPIX data base; then, as in the
thick target case, rather better agreement resulted.
The program was then extended to deal with secondary
fluorescence among multiple layers.
The resulting code incorporated in GUPIX92 then
provides (a> computation of X-ray yields from a multi-
layer structure; by comparison of measured intensity
+ ratios with these, layer thicknesses may be deduced; (b)
element concentration determination in one or more
layers which are distinct in their elements; this case
involves a least-squares fit of the spectrum in which the
multilayer ME are the determinants of the intensity
ratios of lines for any given trace element. However, as
suggested above, intelligent use of the code can render
tractable problems that are too complex to treat in a
single run of the code.
As an example, we are currently analyzing nonstoi-
chiometric films of cadmium-mercury-telluride, thick-
ness unknown, deposited on a glass substrate, with an
intervening layer of nonstoichiometric indium-tin-
oxide whose thickness is unknown [34]. We take k-
PIXE spectra of the L X-ray region using three targets
viz. glass, glass + ITO, glass + IT0 + CMT, a typical
spectrum of the complete structure, which is a solar
cell, is in fig. 6; the corresponding H values are
measured using single-element standards. As indicated
earlier, the barium content of the glass provides a
means of determining the successive IT0 and CMT
thicknesses. The ITO/glass spectrum is fitted using
different IT0 thicknesses until the correct Ba concen-
tration is produced; this provides the IT0 concentra-
tions. Then the process is repeated for the CMT/
ITO/glass spectrum to determine the CMT thickness
and concentrations.
These various examples suggest that we can expect
to see an increasing number of specimens. of complex
structure being analyzed quantitative by CL-PIXE,which
will also provide a check on homogeneity.
Acknowledgements
This work was supported by the Natural Sciences
and Engineering Research Council of Canada. We
thank Dr. Louis Cabri for the CANMET synthetic
sulfide standards and Dr. Gerald Czamanske for the
CCU-1 standards.
 J.L. Campbell et al. / PIXE microanalysis of thick specimens
References
[I] D.E. Newbury, J. Trace and Microprobe Tech. 32 (1986)
103.
[2] M. Hyvonen-Dabek, J. Raisanen and J.T. Dabek, J. Ra-
dioanal. Chem. 63 (1981) 163.
[3] E. Clayton, Nucl. Instr. and Meth. 191 (19811567.
[4] J.L. Campbell, W.J. Teesdale and J.-X. Wang, Nucl.
Instr. and Meth. B50 (1990) 189.
[5] P.S.Z. Rogers, C.J. Duffy and T.M. Benjamin, Nucl.
Instr. and Meth. B22 (1987) 133.
[6] J.A. Maxwell, J.L. Campbell and W.J. Teesdale, Nucl.
Instr. and Meth. B43 (1988) 218.
[7] G. Lagarde, J. Cailleret and C. Heitz, Nucl. Instr. and
Meth. 205 (1983) 545.
[8] J.L. Campbell, J.-X. Wang, J.A. Maxwell and W.J. Tees-
dale, Nucl. Instr. and Meth. B43 (1989) 539.
[9] G.K. Czamanske, V.E. Kunilov, L.J. Cabri, M.L. Zientek,
L.C. Calk and A.P.Likhachev, Can. Mineralogist 30 (1992)
249.
[lo] M.J. Berger and J.H. Hubbell, National Bureau of Stan-
dards Rep. 87-3597 (NIST, Gaithersburg, 1987).
[ll] N.K. Del Grande and A.J. Oliver, Bull. Am. Phys. Sot.
16 (1971) 545.
[12] R.G. Musket, Nucl. Instr. and Meth. B15 (1986) 735.
[13] R.G. Musket, National Bureau of Standards Spec. Pub.
604 (NIST, Gaithersburg, 1981).
[14] J.L. Campbell and J.-X. Wang, X-Ray Spectrom. 21
(1992) 223.
[U] M. Krumrey, E. Tegeler and G. Ulm, Rev. Sci. Instr. 60
(1989) 2287.
[16] J.L. Campbell and J.-X. Wang, X-Ray Spectrom. 20
(1991) 191.
[17] M.H. Chen and B. Crasemann, At. Data Nucl. Data
Tables 33 (1985) 217;
M.H. Chen, private communication.
[18] J.H. Hubbell, National Institute of Standards and Tech-
nology Rep. 89-4144 (NIST, Gaithersburg, 1989).
109
[19] H. Paul and J. Sacher, At. Data and Nucl. Data Tables
42 (1989) 105.
[20] M.O. Krause, J. Phys. Chem. Ref. Data 8 (1979) 307.
[21] Y.C. Yu, M.R. McNeir, D.L. Weathers, J.L. Duggan,
F.D. McDaniel and G. Lapicki, Phys. Rev. A9 (1991)
5702.
[22] M.F.C. Willemsen and A.E.T. Kuiper, Nucl. Instr. and
Meth. B61 (1991) 213.
[23] M. Jaksic, G. Grime, J. Henderson and F. Watt, Nucl.
Instr. and Meth. B54 (199i) 491.
[24] G.H. Faye, W.S. Bowman and R. Sutarno, CANMET
Rep. 79-16, Copper concentrate CCU-1, a certified refer-
ence material (Dept. of Energy, Mines and Resources,
Ottawa, Canada).
[25] G.I. Johansson, X-Ray Spectrom. 11 (1982) 194.
[26] C.G. Ryan, D.R. Cousens, S.H. Sie, W.L. Griffin, G.F.
Suter and E. Clayton, Nucl. Instr. and Meth. B47 (1990)
55.
[27] L.J. Cabri, J.L. Campbell, J.H.G. Laflamme, R.G. Leigh,
J.A. Maxwell and J.D. Scott, Can. Mineralogist 23 (1985)
133.
[28] C.P. Swann and S.J. Fleming, Nucl. Instr. and Meth. B49
(1990) 65.
[29] F. Hei and P. Van Espen, Anal. Chem. 63 (1991) 2237.
[30] J. Rickards and E.P. Zironi, Nucl. Instr. and Meth. B29
(1987) 527.
1311W. Reuter, A. Lurio, F. Cardone and J. Ziegler, J. Appl.
Phys. 46 (1975) 3194.
[32] G. Demortier, S. Mathot and B. Van Gystaeyen, Nucl.
Instr. and Meth. B49 (1990) 46.
[33] C.G. Ryan, D.R. Cousens, C.A. Heinrich, W.L. Griffin,
S.H. Sie and T.P. Mernagh, Nucl. Instr. and Meth. B54
(1991) 292.
[34] J.L. Campbell, J.A. Maxwell and W.J. Teesdale, to be
published.
III. ANALYTICAL METHODOLOGY