barriers to quantitative electron probe x-ray...

Volume 107, Number 6, November-December 2002

Journal of Research of the National Institute of Standards and Technology

[J. Res. Natl. Inst. Stand. Technol. 107, 605-619 (2002)]

Barriers to Quantitative Electron Probe X-RayMicroanalysis for Low Voltage Scanning

Electron Microscopy

Volume 107 Number 6 . November-December 2002

Dale E. Newbury Low voltage x-ray microanalysis, defined as frequent peak interference situations andbeing performed with an incident beam further exacerbates detection limits. Future

National Institute of Standards and energy -5 keV, can achieve spatial improvements to the x-ray spectrometryTechnology, resolution, laterally and in depth, of limitations are possible with x-rayGaithersburg, MD 20899-8371 100 nm or less, depending on the exact optics-augmented wavelength dispersive

selection of beam energy and the spectrometry and microcalorimeter energycomposition of the target. The shallow dispersive spectrometry.

[email protected] depth of beam penetration, with theconsequent short path length for x-rayabsorption, and the low overvoltage, theratio of beam energy to the critical Key words: electron probe x-ray micro-ionmzation energy, both contribute to analyzer; energy dispersive spectrometry;minimizing the matrix effects in low voltage microanalysis; scanningquantitative x-ray microanalysis when the electron microscope; wavelengthunknown is compared to pure element dispersive spectrometry. x-ray spectro-standards. The low beam energy restricts metry.the energy of the atomic shells that can beexcited, forcing the analyst to chooseunfamiliar shells/characteristic peaks. Thelow photon energy shells are subject to Accepted: August 22. 2002low fluorescence yield, so that thepeak-to-continuum background isreduced, severely limiting detectability. Thelimited resolution of semiconductorenergy dispersive spectrometry results in Available online: http://www.nist.gov/jres

1. Introduction: Advantages of LowVoltage X-Ray Microanalysis

Electron probe x-ray microanalysis, as performed inthe conventional beam energy range, generally definedas 10 keV ' beam energy (Eo) ' 30 keV, provides apowerful tool for spatially-resolved analysis [1]. Spatialresolution on a lateral and depth scale of approximately1 [am to 5 p,m can be achieved, depending on the exactchoice of beam energy and specimen composition. Bychoosing the beam energy in the conventional range, asatisfactory analytical x-ray peak, from the K, L, or Mshells, can be found for elements with Z 2 4 (Be). Forsome elements, more than one family of characteristicpeaks can be selected, e.g., for copper, both the K- and

L-family peaks can be excited, while for gold, the L- andM-family peaks are accessible. "Satisfactory peak"means that at least one characteristic peak for an elementcan be excited with sufficient peak-to-background(characteristic relative to the x-ray continuum) so thatelemental concentrations in the major (C> 0.1 massfraction) and minor (0.01 ' C ' 0.1) ranges can bemeasured in reasonable counting times, of the orderof 100 s. Trace levels (C < 0.01 mass fraction) as lowas a mass fraction of 104 [100 ag/pg, or 100 parts permillion (ppm))] can be measured providing the spec-trometry technique employed has sufficient spectral

605

Volume 107. Number 6, November-December 2002


resolution. For electron probe x-ray microanalysis(EPMA), this requirement has been traditionally metthrough the use of wavelength dispersive spectrometry(WIDS). Energy dispersive x-ray spectrometry, withits poorer spectral resolution, is usually restricted toconcentration levels above 0.01 mass fraction forquantitative measurements with a limit of detection of0.001 mass fraction, especially if peak interferenceoccurs.

For a particular specimen composition, the spatialresolution of electron-excited x-ray microanalysisimproves as the incident beam energy is reduced.Interest in quantitative x-ray microanalysis at low beamenergy (Eo • 5 keV) has been greatly stimulated by thedevelopment of the high resolution field emissiongun scanning electron microscope (FEG-SEM). Theprincipal rationale for this interest is the possibility ofachieving x-ray spatial resolution similar in scale to highperformance FEG-SEM electron imaging. Analyticalresolution depends upon the x-ray range, which followsa relationship of the form [2]:

R (nn) = [27.6 A I(p Z0.8 9)] (EOI 67- EC1 67

) (1)

where Eo is the incident beam energy (keV), EC is thecritical excitation energy for the characteristic x rays ofinterest (keV), A is the atomic weight (g/mol), Z is theatomic number, and p is the density (g/cm3 ). The spatialresolution advantage of low voltage operation is seen inTables I and 2. Based on Eq. (1), Table 1 lists the rangeof x-ray production for Cu K (Ec = 8.98 keV) in variousmatrices (i.e., Cu as a trace constituent in variousmatrices) in the conventional beam energy range (10 to30 keV). For operation in the conventional range with

Table 1. Range of x-ray excitation for CuK (K-edge. EC = 8.98 keV)in various matrices in the conventional voltage regime

Matrix 25 keV 20 keV 15 keV 10 keV

C 6.3 p.m 3.9 pm 1.9 Pm 270nmSi 5.7 p.m 3.5 lLm 1.7 p.m 250 nmFe 1.9 pm 1.2p.m 570nm 83nmAu 1.0 ,.m 630 nm 310 nm 44 nm

Table 2. Range of x-ray excitation for Cu L (L-edge, EC=0.933 keV)in various matrices in the low voltagve regime

Matrix 5 keV 2.5 keV

C 490 nm 130 nmSi 440 nm 120 mmFe 50 nm 40 nmAu 80 mn 21 mn

the requirement that the overvoltage (U = EoIE) mustbe 2 2; Eo > 19 keV), the Cu K-shell x-ray productionrange ranges from approximately 5 lim for a lightelement matrix like carbon. For intermediate atomicnumbers, the Cu K-range is approximately 1 pLm whilefor high atomic number elements the range decreasestoward 500 nm. By operating in the low voltage regimeand utilizing the Cu L shell instead of the Cu K shell,Table 2, the x-ray production range for an overvoltageof U 2.5 (Eo = 2.5 keV) drops to about 100 nm forcarbon and to 20 nm for gold, an improvement in spatialresolution of approximately a factor of 50 in lineardimensions, or about 105 in volume. This level of resolu-tion approaches that achieved by high voltage analyticalelectron microscopy of thin sections (-100 nm thick-ness) but with the added advantage that specimens canbe examined in the FEG-SEM without thinning. Oftenspecimens can be examined in the FEG-SEM withoutany special preparation at all, especially if the combi-nation of a FEG-SEM and an environmental/variablepressure SEM is utilized.

A second positive aspect of low voltage microanalysisis the dependence upon beam energy of the matrix orinter-element effects, atomic number (scattering andstopping power), absorption, and secondary x-rayfluorescence, which must be considered in quantitativeanalysis. When calculated as a correction factor between --

multi-element unknown and pure element standards, all,of the matrix correction factors tend toward unity as thebeam energy is lowered into the low voltage regime.X-ray absorption, which is often a major source ofexperimental error at conventional beam energies, ismuch less significant under low voltage analysis condi-tions because of the greatly reduced absorption pathlength that results from the decrease in the beam pene-tration range and the exponential dependence of absorp-tion on path length through the material.

2. What Are We Actually Measuring?

When operating in the conventional beam energyrange, the x-ray generation and sampling depth aregenerally of the order of 1 plmr. While such a rangeis extremely shallow compared to the millimeter tocentimeter thickness of bulk specimens, it is neverthe-less sufficiently deep that the excited volume predomi-nantly samples the bulk composition in most cases.Although most pristine materials, when exposed tomoist atmosphere, react to form native surface oxidelayers, such layers tend to have thicknesses in the rangeof only a few nanometers, e.g., for aluminum, about4 nm of aluminum oxide (A12 03 ) rapidly forms on abare metal surface. When the x-ray generation range is1000 nm to 2000 nm, a surface layer with a thickness of

606

Volume 107, Number 6, November-December 2002Journal of Research of the National Institute of Standards and Technology

a few nanometers of oxide causes a negligible deviation(less than a percent) in the measured intensity from thatproduced by a target with an ideal bare surface.

In low voltage microanalysis, the sampling situationis much more complicated. With the much shallowerranges listed in Table 2, a surface layer of severalnanometers in thickness constitutes a much largerfraction of the sample. The analytical strategy needed todeal with this situation to achieve quantitative analysismust be based upon treating every specimen as a layeredtarget. Measurement at a progressive sequence of beamenergies can provide the critical information necessaryto separately characterize a layer or multiple layers on asubstrate.

3. Limitations to Low VoltageMicroanalysis: X-Ray SpectrometryProblems

3.1 WDS or EDS?

For well over two decades, low beam energy x-raymicroanalysis has been an established capability of theelectron probe microanalyzer (EPMA) equipped withwavelength dispersive x-ray spectrometry (WDS). WDShas the spectral resolution performance necessary tosolve the numerous spectrometry problems that arise inthe low photon energy range, as discussed below. Un-fortunately, the overall efficiency of the WDS is low dueto its small solid angle of acceptance, losses in thediffraction process, and most significantly, the timeserial nature of its spectral collection. The EPMA iscapable of delivering high beam current, 100 nA to1000 nA, in the focused probe, which can compensatefor the low WDS efficiency. This enables usefulmeasurements, including wavelength scans, to be madewithin practical time periods. Ideally, the WDS could beadapted to the FEG-SEM to serve the same role in lowbeam energy x-ray microanalysis. However, the limitingbeam current in the high brightness, cold field emissiongun SEM is typically in the range 100 pA to 1 nA, alevel that increases the time penalty for WDS measure-ments by a factor of 100 to 10 000 compared to the highcurrent EPMA. An attractive alternative is a FEG-SEMequipped with a Schottky gun that can produce highercurrents (100 nA to 500 nA), a more compatible bearncurrent situation for the use of WVDS. However, theSchottky gun has inherently lower brightness than thecold field emission gun, and therefore poorer electron-optical performance (beam current into a specific probesize) at low beam energy.

With this severe limitation imposed on utilizing con-ventional WDS in low beam current instruments, the

usual choice for low beam energy x-ray spectrometry inhigh resolution SEMs necessarily becomes the semi-conductor energy dispersive spectrometer with a mono-lithic silicon crystal (Si-EDS). Si-EDS has sufficientgeometric collection efficiency and quantum efficiencyto operate with the picoampere to low nanoamperebeam currents of the FEG-SEM. Unfortunately, theoptimum resolution performance of approximately125 eV to 129 eV (full peak width at half the maximumpeak intensity, FWHM) at MnKcx (5890 eV) leadsto significant limitations in x-ray spectrometry of lowenergy photons. Moreover, when photons below 1 keVare to be measured, the design of the EDS must beoptimized to yield the best possible performance in thisrange. Low energy photons have a relatively short rangeand are therefore absorbed near the front face of thedetector where they are particularly vulnerable to"incomplete charge collection" artifacts. These artifactsinclude peak shift and resolution degradation due todistortion on the low energy side of the peak. Newdetector designs and improved detector processing toaddress the problem of incomplete charge collectionhave improved the resolution performance in the lowphoton energy region below 1 keV. Resolution ofapproximately 60 eV at C K (282 eV) can now beachieved, compared to 90 eV to 100 eV obtainablewith "old technology" detectors, despite equivalentresolution performance at MnKot, where the x rays areabsorbed deeper into the detector crystal [3].

3.2 Peak Interference Problems

The relatively broad resolution response of Si-EDScompared to the natural width of characteristic x-raypeaks leads to significant peak interference problems inthe low photon energy range. Examples of technologicalmaterials which give rise to severe interferences inSi-EDS spectra in the low voltage analysis regimeinclude tungsten silicon compounds, e.g., W5Si3, shownin Fig. 1 (SiKao at 1.740 keV with WMot at 1.775 keV);strontium orthosilicate, SrSiO3 in Fig. 2 (SiKoL at1.740 keV with Sr LoL at 1.806 keV); brass, 7OCu-3OZnin Fig. 3 (CuLox at 0.928 and ZnLox at 1.009 keV); andstainless steel, Fe-18Cr-iONi in Fig. 4 (CrLoL at0.571 keV; FeLoL at 0.711 keV; and NiLoL at 0.849 keV).While the individual spectral components of these inter-ferences possibly can be separated by mathematicalpeak fitting procedures, such as multiple linear leastsquares fitting, the stability of such procedures becomesquestionable when the relative peak height differencesbecome large or when limited counting statistics areavailable.

607



0

u,

WM;WMy

,.

Photon Energy (keV)

Fig. 1. Si-EDS (FWHM = 129 eV at MnKcx, 61 eV at C K) spectrum of WSi2 showing interferences (SiKcr at1.740 keV with WMps at 1.775 keV) revealed by W-peak fitting with multiple linear least squares. The thin-linetrace shows the residuals after removal of the W-peak structure.

The interference situation is especially severe whenwe must select combinations of elements that involve thefluorescence yields of the L and M shell peaks relativeto the K-shell peaks. Although the K-shell peaks of thelow atomic number elements have low fluorescenceyields compared to higher energy K-shell photons, theyields for the L and M-shell peaks in this energy regionare much lower still. This problem can be seen inFig. 5(a) for BaTiO3, where the Ba M-family peaks arebarely visible against the background despite the largefraction of Ba in the compound (0.589 mass fraction).The Ti L-peaks are unresolved and only form a shoulderon the 0 K peak, despite the fact that Ti is of similarmass fraction (0.205) as oxygen (0.206). Since manymaterials of interest contain oxygen, and carbon is oftenpresent as a contaminant, it is interesting to considerinterferences that occur in the C K to 0 K region of thespectrum. Figure 6 shows this region and the locations ofthe main peaks from various L- and M-family elementsthat would potentially suffer interference in this region.As can be seen in Fig. 5(a) and Fig. 6, a relatively smallmass fraction of carbon and oxygen could cause severe

interference with the L- and M-family peaks because ofthe large difference in relative fluorescence yields.

3.3 Consequences of Low Overvoltage: Limitationson Available Analytical Peaks

In EPMA analysis performed in the conventionalbeam energy range, good operating strategy involvesselecting an incident beam energy, Eo, that producesan overvoltage, U = EoIE. = 2, where E. is the criticalexcitation edge energy, for the highest energy shell ofinterest [1]. Such an overvoltage generally results in anadequate value of the characteristic-to-continuum ratio,the peak-to-background (P/B), to permit detection ofconcentration levels to trace levels (C < 0.01 massfraction) for most elements with Si-EDS and withreasonable counting times (t < 500 s). This strategyprovides at least one useful characteristic peak for allelements with Z Ž4 (Be), and for many elements, morethan one shell can be excited, e.g., K and L for Cu, orL and M for Au. Thus, the analytical strategy forshell selection for a particular beam energy

608



c.* aO rh ky: 0.6000 Uorh: 93

OK

CK A

Rsti5: snb.ns tink-: Sr 3B

Photon Energy (keV)

Fig. 2. Si-EDS (FWHM = 129 eV at MnKc-. 61 eV at C K) spectrum of strontium orthosilicate, SrSiO4 (SiKca at 1.740 keV with SrLX at1.806 keV),

e.g., EO = 20 keV, in the conventional range can bepresented in the form of an annotated Periodic Table, asshown in Fig. 7.

When the low voltage microanalysis regime is con-sidered at the upper limit beam energy of 5 keV, theovervoltage limits access to the more energetic shells ofthe intermediate and high atomic number elements, asshown in Fig. 8 [41. Changes must now be made in theselection of shells for some elements compared to theselection for conventional analysis, e.g., Ti-L instead ofTi-K, Ba-M instead of Ba-L. Moreover, overvoltages aslow as U> 1.1 must be accepted to obtain a peak formeasurement for some elements, but with an inevitabledecrease in the P/B, and consequently sensitivity, thatcan be obtained. As the beam energy is lowered furtherinto the low voltage microanalysis regime, significantportions of the Periodic Table become inaccessible juston the basis of poor excitation, as shown in Fig. 9 forED=2.5 keV and U = 1.1.

3.4 Restriction to Shells With LowFluorescence Yield

The poor efficiency of characteristic x-ray excitationdue to low overvoltage is not the only impedimentto low voltage microanalysis. As the beam energy isreduced below the conventional operational range intothe low voltage regime (Eo ' 5 keV), the energy of theshells that can be ionized necessarily falls, so thataccess to the deeper shells of the heavier elements isreduced or eliminated. The accessible L and M shellsthat can be efficiently ionized for the intermediate andheavy elements are often subject to low fluorescenceyield (the fraction of ionizations that result in photonemission). Low peak fluorescence yield translatesinto a low value of the generated characteristic-to-continuum (P/B) ratio. After convolution with the EDSresolution broadening function, the measured P/B ratiois even further reduced.

609

U, .

0

u

'u:i

ov4

1.55 0.255WIRK: -17 SRSIGLS.TXT:



Ch kV: 0.9300 4lork: 2434 Results: 253.4b H1arker: CU 29

CuLa,p

ZnLa,p

I H _ _ _ _ _ _ _ _ _ _ _ _ _

1.g 1.25

Photon energy (keV)1 .52

Fig. 3. Si-EDS (FWHM = 129 eV at MnKe, 61 eV at C K) spectrum of brass CuZn (CuLoa at0.928 and ZnLta at 1.009 keV)

P/B is a critical term in determining the limit ofdetection. The intensity of characteristic x-rays dependson the overvoltage, U, which is the ratio of the incidentbeam energy, Eo, to the excitation edge energy, Ec, forthe characteristic peak of interest: Experimentally, themeasured characteristic intensity (normalized for beamcurrent and detector solid angle angle) is found tofollow a relation of the form:

(2a)P - (U-i1)

U=EoIEc

When we consider the background directly under acharacteristic peak, then Ev = Ep. Although Ep < E. for agiven element, we can approximate U = EolEc - Eol E,.With this approximation, Eq. (3a) can be rewritten as:

(3b)

The peak-to-background ratio, P/B, can then bedescribed as:

PIB = (U - 1)'Z (U -1) = (I1/Z) (U -l)n-' (4)(2b)

where the exponent n is typically found in the range 1.3to 1.7 [1,5]. The background of the electron-excitedspectrum consists of (ideally) the x-ray bremsstrahlung,which forms a continuous distribution from the lowestphoton energy up to the energy of the incident beam, Eo,as it enters the target. The x-ray continuum intensitydepends upon the average atomic number, Z, and thecontinuum photon energy, E^, of interest:

B - Z (Eo - E^)IEv (3a)

The behaviors of the peak and peak-to-background inthe low overvoltage range predicted by Eqs. (2) and (4)are shown in Fig. 10. The plot of P/B vs U shows asharp decrease in P/B. which is an important factor inthe limit of detection, as the overvoltage decreasestoward unity.

The consequence of a low fluorescence yield isillustrated in Fig. 5(a) for the peaks that constitute theBa M-family, measured with a Si-EDS with resolutionperformance of 129 eV at MnKa and 60 eV at C K.

610

u:

B - Z (Eo-E�)IE� - Z (U-1)

Volume 107, Number 6. November-December 2002

Journal of Research of the National Tnstitute of Standards and Technology

h": 71 Ch VW:0.7?1OO W-k: 11

CrLa ICK

e565 Restts i43.t I ttcrtcer

FeLa

NiLa-

0.00t 0.21 SPOK: -12 47955it.TXT:

Photon energy (keV)

Fig. 4. Si-EDS (FWHIM = 129 eV at MnKa, 61 eV at C K) spectrum of stainlesssteel, CrFeNi (CrLce at 0.571 keV; FeLc at 0.711 keV. and NiLcx at 0.849 keV).

The L-shell is the usual choice for an analytical peak forbarium in the conventional beam energy range, but theL-shell is not accessible in the low beam energy range,since the BaL3 E, is 5.247 keV. The M-shell peaksillustrated in Fig. 5(a) show a low P/B despite thepresence of Ba in the compound at the level of 0.588mass fraction.

Low P/B has a deleterious impact on the limit ofdetection. The concentration limit of detection can beestimated from the spectrum of Fig. 5(a) using theZiebold (1967) expression (originally derived for a pureelement but modified to accommodate measurementson a compound) [6]:

CDL> (3.29 a)l[n 7-(P/Ci) (P/B)] (5)

where a is the constant in the Ziebold-Ogilvie hyper-bolic equation [7] linking the k-value (where k = peakintensity of unknown/peak intensity of standard) andthe concentration C, n is the number of measurements,'r is the measurement time, P is the peak counting rate,C, is the concentration (mass fraction) of the element ofinterest in the compound, and (P/B) is the peak-to-background on the standard. Based upon the measure-ment parameters for the spectrum in Fig. 5(a) (the peak

counting rate was taken at a deadtime of 20 %), andestimating a = 1, the values in Table 3 are predicted forconcentration limit of detection, CDL (mass fraction), ofan element similarly excited as barium. CDL is calculatedas a function of measurement time (for a singlemeasurement) for the most prominent M-peak observedfor Ba (the M3 N, transition).

Thus, under low voltage analysis conditions withconventional Si-EDS, the detection situation for anelement excited with the efficiency of the bariumM-shell can only be measured across the full majorconstituent range (C> 0.1 mass fraction) with a count-ing interval of 1000 s and 20 % deadtime. To detect aconstituent at the middle of the minor range, an accumu-lation time of 10 000 s or more would be necessary.Trace level measurements (C < 0.01) would be impracti-cal with such low excitation.

It should also be noted that the relative peakabundances indicated by the lines (derived from therelative peak intensities reported in the Bearden Tables)strongly disagree with the relative heights of the peaksobserved experimentally. For example, the strongestpeak appears to be associated with the M3NI transition,rather than the My peak (the transition M3N4,5) reportedin the references.

611

1-1rn_4"

10w

A

cn

9

r.



Cht 52 Ch eJ: 0.5200 More: 27760.62 2 .1. -

0IQ

Ti cz1

C Ka JTiLpf t

Ti Lq

H.- ,or a e 56

BaTiO3

Conventional Si EDS (129 eV)

E3= 3 keV

Ni

VZ41 M3PV2 M402

. 2 M5Ot

1 Y

M2N4

025 3.CC SC 0.53 0 P5 0 00 1.00 1.73 1.S5

a

ch"": 2704 Ch kU:0.4326 Work 4 - 0'

TiLaCR TiLl

-- 1: - ,-0.00 0.13 0.26

WORK: 119 BTO103 orystal from 91 multimount,

I MWrker: Eftl 56

OK

BaTiO3

NIST Microcalorimet4r EDS

Eo = 3 keV

13N1

MlM402

:1 M503Jr? AL'

MYt

INI, kS

M2N4

-,I-11F-.ve00

I~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

0.38 D.51 O.54 0.77 0.90 1 .02 1.1nput 500 output 285 real time 840 s I i'e time 500 s DTS = 406 soon mode nag 1650 with correction

612

137666

:).CC C. ̂ 31o05: "2 5f6103.T)T:

7:

bFig. 5. (a) Si-EDS (FWHEM = 129 eV at MnKac. 61 eV at C K) spectrum of BaTiO3 excited in the low voltage analysis region with Eo = 3 keV;(b) spectrum of BaTiO 3 as measured with the microcalorimeter EDS.

- -l

I


0

u

Photon Energy (keV)

Fig. 6. The C K and 0 K region of the x-ray spectrum showing interferences withL- and M-family peaks of heavy elements.

Semiconductor EDS (129 eV resolution at MnKuo)

Eo>20 kU >2(El

eV vn: K-shell UK&L<10 eV) [] L-shell EIL&M

Not detectable 5D M-shel_

Na

Be

Mg I.Al I Si I p I S ci Ar

K Ca Sc Ti V Ga_ As _ B Kr

Rh Sr Y Zr Nb| 1Mo RT RIu Rh Pd Ag Cd In Sn- Sb Te I Xe

Cs BIaItI'4NY90 WRt0L .. W AE -NM WYPfE tR

_*R AtCe I li W

f. LiU N Piz Aiis. C'nu BIk (It I f: t Mld Nf\ Li'

Fig. 7. Periodic Table showing typical choices of atomic shells for operation in the conventional beamenergy range, Eo 10 keV to 30 keV.

613

BIC IN I 01 F lNe

Volume 107. Number 6, November-December 2002


Semiconductor EDS (129 eV resolution at MnKa.)

Eo = 5 keV LI K-shell

Uo > 1.1 (Ec< 4.5 keV) LI L-shell

[ Not detectable ID M-shel

Be

Na Mg 1S si ci

IAr

K Ca Sc Ti V Cr MNIn Fe Ca Ni Cuzn Ga Ge AU Se v Kr

Rb f Sr Y; Zr m o Tc Ru Pt _d Ae Cd I n Sn Sn Te JO XLe

Cbis iB f -- tW -T Re Oif Ir 4 -Pi = Ph Pt: Rit

Fig. 8. Periodic Table showing choice of atomic shells available for operation in the low beam energyrange, Eo = 5 keV and U = 1.1.

Semiconductor EDS (129 eV resolution at MnKac)

Eo = 2.5 keV

Uo > 1.1 (Ec< 2.25 keV)

iL [Not deteMBe

|Na.|Mg

- C I

ctable

Sc ITI I V Cr I Mnj Fel Co I MI IZn

Rb Sr Y Zirr

0 Ba M i e' s V- 4 5W R o4j i ! ;Ai

E] K-shell

E L-shellE IM-shell

I Al I Si

Ga cGe

I

1iP 1

A, I Se

p N T w . , !T 5 N I ::N N N7 T == T!'cc t- Pr N PM& Snin ' 1qI ! Dj " " " 4 II~ ~ ~ ~ ~ ~ ~~~~~. Ho_ : Er 'j;g Jii ::1{=\g X'DL =:. , = I1-141

Fig. 9. Periodic Table showing choice of atomic shells available for operation in the low beam energyrange, Ea = 2.5 keV and U = 1.1. Note significant loss of elements that can be effectively measured.

614

C" -eP -- P:nsw In PrE -- Ho ULr lTn T in

MC P pP !AJdCrn': Bk cDr Er E ul

Br KrI

_ _ _

B I C I N I 0 1 F INe

-11- - - - - - I I __M

w_W i.trE,40

I

I

f

IIIII

II

711 M:i:

9

=':. s ._ g :;, .o

�Fri


0.8 _

0.6 -

0.2-/

1 1.2 IA 1.6 1.0 2

Overvoltage, U

Fig. 10. Behavior of the peak (P) and peak-background (P/B) ratioas a function of overvoltag

Table 3. Predicted linits of detection for a Ba-like element inBaTiO3 with Si-EDS (Eo = 3 keV and 20 % deadtime)

los IOOs 1000 s 10 000 s

CDL 0.98 0.31 0.098 0.031

4. Possible Solutions to X-RaySpectrometry Problems

Two emerging technologies may significantlyimprove the measurement situation for low energyphotons in the 100 eV to 5 keV range: (1) WDSaugmented with x-ray optics and (2) microcalorimetry.

4.1 WDS/X-Ray Optics

The WDS is constrained in two important waysrelative to the low intensity x-ray source available inFEG-SEM instruments operating in the low voltageregime: (1) low efficiency due to the small solid angleof collection and the inefficiency of the diffractionprocess for some diffractors and (2) the narrow energybandpass of a few eV, meaning that most of the excitedspectrum is lost at the diffractor and never reaches thedetector. Instrumentation advances have been proposedand/or made to at least partially overcome theselimitations.

1. Increasing the WEDS efficiency: The efficiency ofthe WDS has been improved through incorporation ofcapillary x-ray optics [8] (Agnello et al., 1997). In theirdesign, which is now available commercially, a capillary

optic bundle tapered on one end is used to collect x raysover a large solid angle, conduct the x rays alongthe inside surface of the capillaries by total externalreflection, and then present a parallel beam of x rays toa flat crystal diffractor. The flat diffractor scatters witha high degree of efficiency into a gas proportionalcounter. This system has realized a gain of a factor ashigh as 50 compared to a conventional curved crystalWDS for characteristic x rays in the energy range from100 eV to 1.8 keV.

2. Parallel detection in WDS: The greatest loss inavailable information occurs because of the time serialnature of the WIDS measurement. Moreover, peak topmeasurements alone are insufficient. For accurateintensity measurement, the background under the peakmust be estimated, usually by directly measuring thebackground on either side of the peak. Fiori et al.observed that a dispersed image of a WDS peak could beobtained by taking advantage of the focusing action ofa curved crystal wavelength dispersive spectrometer [9].By placing both the electron-excited source and asuitable multichannel detector inside the Rowland circlewhile maintaining the diffractor in its conventionallocation on the circle, a dispersed image of the x-raypeak and the nearby background could be obtained.Such a parallel measurement of the peak would improveefficiency by permitting detection of all characteristicphotons that comprise the peak, which would increasesensitivity, and additionally, it would reduce the signifi-cance of chemical effects that alter peak shape, since thefull peak shape could be recorded for every measure-ment. Moreover, for quantitative analysis, the criticalstep of background removal could be made moreaccurate by fitting a large number of channels of datacompared to conventional WvDS practice of on-peak,off-peak measurements.

4.2 Energy Dispersive X-Ray Spectrometry byMicrocalorimetry

The development of the microcalorimeter energydispersive x-ray spectrometer (plcal-EDS) is one of themost promising approaches for solving many of thespectrometry problems that limit low voltage micro-analysis [101. The pLcal-EDS operates on the principle ofabsorbing an x-ray photon in a metal target and measur-ing the resulting temperature rise. The ptcal-EDSoperates at extremely low temperature, 100 mK, and thedetector makes extensive use of superconductingcryoelectronics for the critical thermometry that consti-tutes the measurement of photon energy. The principalcharacteristics of the microcalorimeter-EDS as relatedto low voltage microanalysis can be summarized asfollows [11].

615

Volume 107. Number 6, November-December 2002Journal of Research of the National Institute of Standards and Technology

Resolution: The microcalorimeter-EDS providesresolution performance, 2 eV to 10 eV, similar to that ofWDS but with operation in an energy dispersive mode.Figure 11 shows a plot of resolution for a variety ofspectrometers: Si-EDS, WDS, and the microcalori-meter-EDS. The resolution of the microcalorimeter-EDS is similar to the WDS over most of the photonenergy range. For the very low energy photons below500 eV for which layered synthetic materials (LSM)are the leading choice as WDS diffractors, the micro-calorimeter-EDS actually has a significant resolutionadvantage, 2 eV to 3 eV compared to 10 eV to 14 eV forLSMs.

This resolution performance of the microcalorimeter-EDS is best appreciated by inspecting examples ofspectra for materials of interest. Consider the difficultexamples for conventional Si-EDS presented in Figs.1-4. Figures 12-15 show the performance of themicrocalorimeter-EDS on the same or very similarmaterials: WSi2, SrSiO3 , brass, and a mixture of thetransition metals. In all cases, the major and most minor

Is

0

c,c9

laA0 [

£

-0

LSMS.f

13 Miicrc- - - WiU) I

_ IS, i

peaks of the elements are well resolved, permittingunambiguous elemental identification for qualitativeanalysis, providing well separated peaks for accurate

121

10(

.0

4)

4C�

00z�41;=L

Energy (eV)

Fig. 12. Microcalorimeter energy dispersive x-ray spectrum of WSi2,showing separation of the SiK peak from the W-M family x-raypeaks.

- . .-

a 9mAc2l aiAmeter (analog prncesing)5#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

alcrir(diitp=wd~sng) U

f a

* Low khV walI

I .* . v j S . s W S. . t .

2000 400D 6000 800 to0o

Energ (eV)

Fig. 11. Comparison of resolution vs photon energy for various spectrometers.

616

i



Ch kV: 19065 Uork: 1032 Results: 0

SiKaSrSiOAE 0=5keV

Mce'ker: Sr 3B

SrLac

SrLp

Fig. 13. Microcalorimeter energy dispersive x-ray spectrum of SrSiO3 , showing separation of the SiK peak fromthe Sr-L family x-ray peaks.

617



Work: 215 ,I

BrassEC 5 keV

11arker: Zn 30

CuLca

CuLP3

ZnLa

HZLI

Fig. 14. Microcalorimeter energy dispersive x-ray spectrum of brass.showing separation of the CuL family peaks from the ZnL family peaks.

700 800

Energy (eV)

Fig. 15. Microcalorimeter energy dispersive x-ray spectrum of a complex alloy,showing separation of the L-family peaks for Mn, Fe, Co, Ni and Cu.

618

V0


peak intensity measurement necessary for quantitativeanalysis, and yielding high P/B ratios to achieve lowlevels of detection. The improvement in the spectro-metry situation is well illustrated by the case of BaTiO 3shown in Fig. 5(b) where the microcalorimeter EDSprovides complete resolution of the TiLl and TiLoXpeaks from 0 K, and separation of several of the numer-ous Ba M-famnily x-ray peaks. The difficult measure-ment situation imposed by the low fluorescence yield ofthe Ti L-family and the Ba M-family is obvious from therelative peak heights observed in Fig. 5(b). Whilemathematical deconvolution of the Si-EDS spectrumwould be possible, the direct resolution of the peaks ismuch more desirable to achieve a robust analysis.

The improvement in the limit of detection that canbe obtained with the microcalorimeter-EDS for thedetection of Ba-like elements in BaTiO 3 based on thespectrum shown in Fig. 5(b) is given in Table 4 [11]. Thegreat improvement in the detection of the various peaks,including the Ba M;, in the microcalorimeter EDSspectrum compared to the Si-EDS spectrum renders thequalitative analysis far more reliable. While there is animprovement of only about a factor of 5 in CDL dueto the greater P/B of the microcalorimeter EDS,the limiting maximum spectral counting rate (about800 c/s, full spectrum) of the microcalorimeter EDSreduces the improvement factor. The development ofarrays of microcalorimeter EDS detector chips, each ofwhich is capable of ~1 kHz counting rate, will be agreat aid to achieving useful low levels of detection forpoorly excited elements [12] (Nam et al. 2001).

Table 4. Predicted limnits of detection for a Ba-like element inBaTiO3 with the microcalorimeter-EDS (Eo = 3 keV and 20 %deadtime)

los 100 s 1000 s 10,000 sCDL 0.179 0.057 0.0179 0.0057

5. Summary

Low voltage microanalysis offers a great improve-ment in spatial resolution, both lateral and in depth, butat the expense of analytical flexibility. At an incidentbeam energy of Eo = 5 keV, useful characteristic peaksfor analysis can be found for all elements in the PeriodicTable, excepting H, He, and Li, providing a low over-voltage, U > 1.1 is acceptable. At lower beam energies,e.g., Eo = 2.5 keV, significant portions of the PeriodicTable are not accessible even with U as low as 1.1.The choice of elemental peaks is further complicatedby the low fluorescence yields of low photon energy

L-shell and M-shell peaks. The relatively poor energyresolution of the semiconductor EDS leads to furtherlimitations upon the spectral measurements, especiallyin situations of peak interference or of low concentrationlevels, where the low P/B of Si-EDS results in poordetection. Advances in x-ray spectrometry, includingoptics-augmented wavelength dispersive spectrometryand microcalorimeter energy dispersive spectrometry,hold considerable promise for improving the measure-ment situation to extend the practical applicability oflow voltage microanalysis.

6. References

[1] J. I. Goldstein, D. E. Newbury, D. C. Joy, C. Lyman, P. Echlin,E. Lifshin, L. Sawyer, and J. Michael, and Scanning ElectronMicroscopy and X-Ray Microanalysis, 3rd Ed., Kluwer-Plenum,New York (2002).

[2] K Kanaya. and S. Okayama, J. Phys. D.: Appi. Phys. 5.43(1972).

[3] J. J. McCarthy, The Effect of Detector Dead Layers on LightElement Detection, in X-Ray Spectrometry in Electron BeamInstruments, D. B. Williams, J. I. Goldstein, and D. E. Newbury,eds., Plenum, New York (1995) pp. 67-81.

[4] D. C. Joy and D. E. Newbury, Advanced SEM Imaging, inCharacterization and Metrology for ULSI Technology, D. G.Seller. A. C. Diebold, W. M. Bullis, T. J. Shaffner, R. McDonald,and E. J. Walters, eds. American Inst. Phys. ConferenceProceedings 449 (1998) pp. 653-666.

[5] E. Lifshin, M. F Ciccareli, and R. B. Bolon, in Proc. 8th Intl.Conf. on X-Ray Optics and Microanalysis, D. R. Beaman, R. E.Ogilvie, and D. B. Wittry, eds., Pendell, Midland, Michigan(1980) p. 141.

[6] T. 0. Ziebold, Anal. Chem. 39, 858 (1967).[7] T. 0. Ziebold and R. E. Ogilvie, Anal. Chem. 36, 322 (1964).[8] R. Agnello, J. Howard, J. McCarthy, and D. O'Hara, Micro.

Microanal. Suppl. 2 3, 889 (1997).[9] C. E. Fiori, S. W. Wight, and A. D. Romig, Jr., Microbeam

Analysis, D. G. Howitt, ed., San Francisco Press (1991) p. 327.[10] D. A. Wollman, K. D. Irwin, G. C. Hilton, L. L. Dulcie, D. E.

Newbury, and J. M. Martinis, J. Micros. 188, 196 (1997).[11] D. Newbury, D. Wollman. K. Irwin, G. Hilton, and J. Martinis,

Ultrarnicroscopy 78, 7 (1999).[12] S. W. Nam, D. A. Wollrman, D. E. Newbury, G. C. Hilton, K. D.

Irwin, D. A. Rudman, S. Deiker, N. F. Bergren, and J. M.Martinis Microsc. Microanal. Suppl. 2: Proceedings 7 1050(2001).

About the author: Dale E. Newbury is an NIST Fellowin the Surface and Microanalysis Science Division ofthe Chemical Science and Technology Laboratory ofNIST in Gaithersburg, MD. He was educated atLehigh University and the University of Oxford. Hisresearch interests include x-ray spectrometry, scanningelectron microscopy, and Monte Carlo modeling ofelectron beam-specimen interactions. The NationalInstitute of Standards and Technology is an agency ofthe Technology Administration, U.S. Department ofCommerce.

619

COPYRIGHT INFORMATION

TITLE: Barriers to Quantitative Electron Probe X-RayMicroanalysis for Low Voltage Scanning ElectronMicroscopy

SOURCE: J Res Natl Inst Stand Technol 107 no6 N/D 2002WN: 0230502167012

The magazine publisher is the copyright holder of this article and itis reproduced with permission. Further reproduction of this article inviolation of the copyright is prohibited. To contact the publisher:http://www.nist.gov/

Copyright 1982-2003 The H.W. Wilson Company. All rights reserved.

barriers to quantitative electron probe x-ray...

Documents