Monitoring the Performance of Energy Dispersive Spectrometer Detectorsat Low Energy

PIERRE HOVINGTON, GILLES L’ESPÉRANCE,* ERIC BARIL,* MICHEL RIGAUD†

Département de Génie Mécanique, Université de Sherbrooke, Sherbrooke; *Center for Characterization and Microscopy of Materials,École Polytechnique de Montréal, Département de Métallurgie et Génie des Matériaux, Montréal; † NSREC-Refractory Materials Chair ofÉcole Polytechnique (CIREP), École Polytechnique de Montréal, Montréal, Québec, Canada

SCANNING Vol. 17, 136–139 (1995) Received April 4, 1994© FAMS, Inc. Accepted with revision May 2, 1994

Summary: The performance of the energy dispersive spec-trometer (EDS) detectors has to be established in its originalstate and monitored periodically to ensure the stability andlong-term performance of the system. Because EDS detectorsare cooled by liquid nitrogen and because the environment isnot an ultra-high vacuum, ice and contaminants will build upin front of the detector crystal, decreasing the detection effi-ciency at low energy. Few performance tests are suitable for thespecifics problems found at low energy. The proposed test isbased on a new procedure for the modelling of the decrease ofdetection efficiency (DODE) with time (Hovington et al.1993). It was found that the deposition rate of both ice and car-bon increased significantly when porous uncoated sampleswere observed. Significant difference between spectra acquiredimmediately after different conditioning was also found. Theproposed procedure can be used to determine the optimumtime before a conditioning cycle, to detect an abnormal accu-mulation of contaminant both in front of the detector and at thesample surface, and to diagnose a broken window and a mal-function in the conditioning apparatus.

Key words: EDS, performance tests, windowless, low-energyx-ray, detection efficiency

Introduction

In the microanalysis of bulk samples, the spatial resolutionof x-ray analysis is greatly influenced by the accelerating volt-age (Eo) of the primary beam. Because of the reduction of thedepth distribution at low voltages (∝ Eo

5/3, Kanaya andOkayama 1972), quantitative analysis with a resolution of <100 nm can be achieved (Boyes et al. 1991). Furthermore, newcold cathode field emission scanning electron microscopes(FESEM) can provide a high-current density (≈109A/cm2) in a

very small probe diameter at low accelerating voltage (≈3nm@ 1.5 keV, Pawley and Joy 1992). Hence, the interest in micro-analysis and the low- energy end of the spectra ( E < 2 keV) hasincreased. However, in the low-energy region, several prob-lems are found (nonlinear background, high convolution prob-ability with the L and M lines of heavier elements). In addition,because the energy dispersive spectrometer (EDS) is cooled byliquid nitrogen, contaminants and ice can build up (Cohen1982, Foitzik et al. 1991, L’Espérance et al. 1990, Musket1981). These layers act as an additional window in front of thedetector, decreasing its detection efficiency at low energies ( E< 1.5 keV, Hovington et al. 1993).

In addition, the performance of an EDS detector has to becharacterized both as received to establish a baseline of perfor-mance and to ensure that it meets the specifications contractedwhen ordering the detector and to ensure periodically its sta-bility and its long- term performance. Several tests have beenproposed and briefly reported by Zemyan and Williams (1993).At low energy, the Ni L/Ni K ratio is often used to monitor therelative efficiency at low energy. However, this procedure can-not be used to monitor the decrease of detection efficiency(DODE) with time for the entire low-energy region. Becauseof the strong energy dependency of the absorption process, thefull computation of the deposited layer of both ice and carbonhas to made and used for a complete calculation of the DODEwith time at each recorded energy (Hovington et al. 1993).

A new standard procedure for the modelling of the DODEwith time for low-energy EDS spectra has recently beenreported by Hovington et al. (1993). In this new procedure a“monitoring” sample is acquired at each “microanalysis day.”The monitoring spectra are then used for accurate computationof the deposited thickness of both carbon and ice. Once thethicknesses of the deposited layers are computed, the standardselements and/or the compared spectra can be corrected for thedecrease in detection efficiency that has occurred. The accu-racy of the procedure is within 2% for x-ray lines not stronglyabsorbed into the contamination layers. Hence, this proceduremonitors the relative performance of an EDS detector by thecomputation of the deposited layer in front of the crystal.

This paper describes the use of this procedure to monitor theperformance of EDS detector at low energy, by which abnor-mal accumulation of carbon both in front of the crystal and onthe surface sample has been clearly determined.

Address for reprints:

Pierre HovingtonDépartement de Génie MécaniqueUniversité de SherbrookeSherbrooke (Québec) J1K 2R1Canada

Materials and Methods

The EDS spectra were acquired with a Link LZ-4 30 mm2 ×3 mm Pentafet detector with a take-off angle of 45° which canoperate with a Be window (≈20 µm thick), an ultra-thin win-dow (UTW) (Hovington and L’Espérance 1992), or in the win-dowless (WL) mode. The EDS is attached to a JEOL 840. Thedetector can undergo a conditioning cycle engaged by a man-ufacturer-controlled operation. This conditioning process takesabout 2 h to complete. All of the scanning electron microscope(SEM) samples used were pure metal standards mounted in thesame bakelite holder. To avoid extra contamination, both on thedetector crystal and on the samples, no SEM work was done ata specimen chamber pressure > 5 × 10-6 Pa (4 × 10-5 mtorr).

Results and Discussion

Monitoring of the Detection Efficiency afterConditioning

For efficient comparison of the low-energy region (E < 1.5keV) of spectra acquired at different times, standard spectrahave to be acquired to monitor the possible change in theresponse function of the detector. After a conditioning processit is important to ensure that (1) the heating process has notchanged the detection capability of the Si(Li) crystal [increasein noise peak and in incomplete charge collection effect(Nicholson et al. 1984) and loss of energy resolution], and that(2) the detector has recovered its original detection efficiency(all of the absorbing contaminants have been removed).

Figure 1 presents low-energy spectra of an Ni standard sam-ple taken under the same experimental conditions and imme-diately after two different conditionings. The detector was usedbetween the two conditionings and, consequently, contamina-tion by both carbon and ice occurred. In Figure 1, the condi-tioning cycle was able to remove 1103 ± 7 nm of ice and 77 ±3 nm of C in front of the crystal. A comparison of the two spec-tra also shows that the detector has reached the same level ofefficiency found after the two conditioning cycles.

Thus, it is important to note that, even if spectra are takenimmediately after a conditioning process, this does not ensurethat the detector has reached the same level of detection effi-ciency and a monitoring sample has to be used for verification.The new procedure for the DODE with time cannot only detecta relative change in the detection efficiency, but can also com-pute the thickness of the absorbing layer remaining in front ofthe crystal. In the case of a decrease of detector performance atlow energy (decrease of the energy resolution and increase innoise peak and in incomplete charge collection effect, ...), theresidual of the fit between the two compared spectra can beused efficiently.

In Figure 2 we present three spectra acquired immediatelyafter three different conditioning cycles. When we comparespectra a and b, it is clear that the detector has not reached thesame level of efficiency after the two conditioning cycles.Between the two conditionings, the detector was removed from

the column to replace the conditioning resistor. During the nec-essary repair, the vacuum inside the detector was broken. Bycomparing the two spectra (a and b), we found that a 250 nmthick carbon layer and a negligible amount of ice are still pre-sent in front of the crystal after the conditioning. Hence, itappears that this tenacious carbon layer was not depositeddirectly on ice since it was not removed with the ice. This extra250 nm layer of carbon was most probably deposited directlyin front of the crystal during the time that the vacuum inside thedetector was broken. For the next conditioning cycle required—because the deposited thicknesses of carbon and ice werevery large (over 2 µm of ice and 1 µm of carbon)—we chose tocarry out three consecutive conditioning cycles before anymeasurement was taken. Spectrum c was acquired immedi-ately after the third conditioning cycle. Comparing the spec-trum of conditioning a and c (cf. Fig. 2), we can see that thethree conditioning cycles were able to clean up the detector

P. Hovington et al.: EDS detectors performance at low energy 137

FIG. 1 Low-energy spectra of Ni acquired immediately after two differ-ent conditionings (A, B).

FIG. 2 Ni spectra acquired immediately after three different condition-ings (a,b, and c). Shown in insert is an enlarged view of the spectra in therange 0.25–0.60 keV.

0

C Kα

Noise

Ni L

Ni Kα

Ni Kβ

A10,000

1,000

100

15,000

10,000

5,000

0

1500

1000

500

00.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60

B

1 2 7 8 9 10Energy (keV)

Cou

nts/

chan

nel

Cou

nts/

chan

nel

0.25

Ni L

Conditioning day

a & b => 1 Cond. cyclesc => 3 Cond. cycles

0.50Energy (keV)

0.75 1.00 1.25

138 Scanning Vol. 17, 3 (1995)

crystal. The detector then reached the detection efficiencyrecorded after conditioning a.

Detection of Abnormal Accumulation of Contaminantsin Front of the Detector

We used the procedure for the modelling of the DODE withtime to detect abnormal accumulation of contaminants in frontof the detector. In the next example, porous uncoated iron sam-ples containing residual oil at the surface of the sample wereobserved in the first 10 days after a conditioning cycle in ourSEM. In Figure 3 we plot the computed thicknesses of ice andcarbon recorded at different times after conditioning. After thefirst 10 days, the porous sample was cleaned and coated withan Au/Pd conductive layer (approximately 100 nm thick). Forcomparison, we plot in Figure 3 the thicknesses computedwhen standard samples (i.e., nonporous samples) were used(normal operating cycle). We first note that the deposition rateis nonlinear for all the operating cycles and that a large increaseis recorded in the first days after the conditioning. When wenow compare the two operating cycles, we note an interestingdecrease of the computed carbon thickness after 10 days. Theremoval of carbon could be explained by the fact that, since ourdetector has a relatively high take-off angle (45o), the relativelythick (≈600 nm) viscous carbon layer can slide off the ice bygravity and form a drop at the end of detector. In fact, in anothersimilar SEM we worked with (JEOL 840 A with a Link LZ-4detector), equipped with a viewing window in the specimenstage, a viscous drop can often be observed at the end of thedetector when porous samples are frequently observed.

Detection of Abnormal Accumulation of Contaminantat the Sample Surface

The procedure for the DODE with time is used to monitorchanges at the detector level. Thus, the difference between aDODE-corrected spectrum compared with a related experi-mental spectrum can be used to show an abnormal x-ray inten-sity as a consequence of a carbon or oxidized layer on the sur-face of the sample. In Figure 4 we present the low-energyregion of three Cd spectra. Two spectra were accumulatedimmediately and 23 days after the conditioning of the detector,and the third spectrum is a fitted spectrum. The computedthicknesses of ice and C are 649 ± 8 nm and 89 ± 2 nm, respec-tively. Also presented in Figure 4 is the residual of the fit whichshows two Gaussian shape functions centered at the C Kα andO Kαpeak positions. These two residual peaks indicate that thespectrum on Day 23 was acquired in a region of the samplewhich contains more carbon (probably extra contamination)and more oxygen (probably oxidized Cd). Evidently, some ofthe counts from carbon and oxygen are the result of fluores-cence into the carbon and ice layers in front of the crystal. Thecomputed fluorescence intensities are 10 and 42 counts for car-bon and oxygen, respectively. The net intensity in the residualpeak is approximately 1790 and 633 counts for carbon andoxygen, respectively. With the carbon intensity we can calcu-late that the analyzed region on Day 23 had an extra carbon

layer, approximately 300 nm thick, compared with the regionanalyzed immediately after conditioning.

Hence, because the Day 23 Cd spectrum was acquired in acarbon-contaminated region, this spectrum could not be usedfor a reference spectrum in quantitative work. It is important tonote that, because the carbon peak is convoluted with the Cd Mlines, the presence of extra C intensity can only be determinedby such a comparison.

Other Applications

Determination of the Optimum Time before a Conditioning Cycle

A major advantage of the proposed procedure is that itallows us to compute the thickness of the contamination layersin front of the crystal. Thus, one can use this computed thick-ness to set a threshold value at which a conditioning must be

FIG. 3 Computed absorption thicknesses of carbon and ice when porous(filled symbols) and nonporous samples (empty symbols) are used.

FIG. 4 Cd spectra acquired immediately and 23 days after a condition-ing cycle. A fitted spectrum is also presented (dot). The fitted thicknessesfor carbon and ice are 649 ± 8 nm and 89 ± 2 nm, respectively.

C Kα0 Kα

10,000

1,000

100

10

1

Use of poroussamples

Ice, C2

Ice, C1

C, C2

C, C1

0 5 10 15 20 25 30Days after conditioning

Com

pute

d ab

sorp

tion

thic

knes

ses

(nm

)C

ount

s/ch

anne

l

35 40 45 50

2,000

1,500

1,000

500

00.00 0.25 0.50 0.75 1.00 1.25 1.50

Energy (keV)

Noise Cd M

Cd Day 0

Cd Day 23

Residual

Fitted spectrum

carried out. The critical thicknesses of ice and/or carbon beforea conditioning cycle is required must be set according to thedetection limit needed at low energy. It is important to note thata maximum delay between two conditioning cycles is desirablebecause the long-term effect of the conditioning on the perfor-mance of the Si(Li) crystal has not been established.

Detection of a Broken Window

A periodic accumulation of a standard spectrum can be usedto ensure that the UTW is still intact. The UTW in our Link LZ-4 system is a pure Al film, approximately 120 nm thick,deposited on an Ni grid (Hovington and L’Espérance 1994).The Al film is very fragile and can easily be blown off by a dif-ferential pressure between the specimen chamber and thedetector. Also, we experienced (Hovington et al. 1993) thegradual thinning of a UTW in a transmission electron micro-scope.

To verify the condition of the UTW periodically, we com-pared Ni standard spectra accumulated with a UTW at differ-ent times. Before comparison, the spectra were corrected forthe DODE with time. The WL spectra are used to compute thedeposited thicknesses of both carbon and ice. Hence, if a sta-tistical difference is observed between the different UTW-cor-rected spectra, testing procedures are engaged. Testing theUTW as an efficient barrier to light by specular reflection on ahighly reflected surface sample usually is a simple and efficienttest to detect a broken window.

Detection of a Malfunction in the ConditioningApparatus

The conditioning cycle of an EDS detector can be done eitherby a predefined manufacturer-controlled operation or by pro-gressively evaporating the liquid nitrogen of the Dewar andkeeping the detector (with the high tension previously removed)at room temperature for a given period of time. For both cases,however, one has to verify the efficiency of the conditioningcycle. By comparing spectra acquired immediately before andafter a conditioning, one can ensure that the conditioning cycleworked properly. When we now compare Ni spectra acquiredimmediately after two conditioning cycles, the efficiency of thetwo conditioning cycles can easily be established.

Conclusion

In conclusion, the performance of an EDS detector at lowenergy has to be established in its original state and monitoredperiodically to ensure the stability and long- term performanceof the system. Because EDS detectors are cooled by liquidnitrogen and the environment is not an ultra-high vacuum, iceand contaminants will build up in front of the detector crystal.It was found that the deposition rate of both ice and carbonincreased significantly when porous uncoated samples wereobserved. Hence, if such samples must be analyzed in an SEM,conditioning cycles must be engaged more frequently.

It is also crucial to verify that the detector has reached thesame level of detection efficiency after a conditioning cycle.Abnormal accumulation of a carbon layer onto the detectorcrystal, which was not removed by a conditioning cycle, wasfound following a repair of the detector.

With the proposed procedure for monitoring the perfor-mance of EDS detectors at low energy, it is possible to deter-mine regions of samples contaminated with oxygen and car-bon. As a result, such spectra must be discarded for anyquantitative work.

Finally, the procedure can be used to detect a broken win-dow, a malfunction in the conditioning apparatus, and also todetermine the optimum time before conditioning is necessary.

Acknowledgments

The authors are grateful to Gérard Guérin of École Poly-technique de Montréal for carrying out some of the experi-ments and to Dr. Raynald Gauvin of Université de Sherbrookefor helpful discussion. Financial support from CRSNG/NRSEC and the Centre for Characterization and Microscopyof Materials ((CM)2) of Ecole Polytechnique de Montréal isgratefully acknowledged.

References

Boyes ED, Hartmann IR, Gooding F, Sokola D, Hanna L, Smith DL: EDXChemical Microanalysis of Bulk Specimens in the SEM at Low BeamVoltage and High Spatial Resolution (<0.1µm). EMAG (1991)

Cohen DD: M-shell x-ray emission for determining the low energy effi-ciency of Si(Li) detectors. Nucl Instru Meth 193, 15–19 (1982)

Foitzik AH, Sears JS, Heuer AH, Zaluzec NI: Massive icing of A HP EDSdetector with an ultrathin window. Proc 49th Annual EMSA (Ed. Bai-ley GW). San Francisco Press (1990) 752–753

Hovington P, L’Espérance G, Baril E, Rigaud M: A standard procedure forthe modeling of the decrease in detection efficiency with time for lowenergy EDS spectra. Microbeam Analysis 2 277–288 (1993)

Hovington P, L’Espérance G: Determination of detection efficiency of aUTW detector and modeling of low energy EDS spectrum used forthe quantification of EDS analyses of light elements. In MicrobeamAnalysis-1992 (Eds. Bailey GW, Bentley J, Small JA). San FranciscoPress, San Francisco (1992) 1234–1235

Hovington P, L’Espérance G: In preparation (1994)Kanaya K, Okayama S: Penetration and energy-loss theory of electrons

in solids targets. J Appl Phys D 5, 43–58 (1972)L’Espérance G, Botton G, Caron M: Detection and quantification prob-

lems in the analysis of light elements with UTW detectors. InMicrobeam Analysis-1990 (Eds. Michael JR, Ingram P). San Fran-cisco Press, San Francisco (1990) 284–285

Musket RG: Properties and applications of windowless Si(Li) detectors.Energy Dispersive Spectrometry. NBS Special Publication 604(1981) 97–126

Nicholson WAP, Adam PF, Craven AJ, Steele JD: The shape of theBremsstrahlung background in thin-film x-ray spectra. In Analyti-cal Electron Microscopy (Eds. Williams DB, Joy DC). San FranciscoPress, San Francisco (1984) 258–262

Pawley J, Joy DC: High Resolution, Topographic SEM and RadiationDamage: The Rationale for Using Low-Beam Voltage. ElectronMicroscopy, Volume 3. EUREM 92, Granada Spain, (1992)743–747

Zemyan SM, Williams DB: Characterizing an energy dispersive spec-trometer on an analytical electron microscope. Microbeam Analysis,S182–183 (1993)

P. Hovington et al.: EDS detectors performance at low energy 139