A calibration procedure for the detection efficiency of energy dispersive X-ray spectrometers (EDS) used in combination with scanning electron microscopy (SEM) for standardless electron probe microanalysis (EPMA) is presented. The procedure is based on the comparison of X-ray spectra from a reference material (RM) measured with the EDS to be calibrated and a reference EDS. The RM is certified by the line intensities in the X-ray spectrum recorded with a reference EDS and by its composition. The calibration of the reference EDS is performed using synchrotron radiation at the radiometry laboratory of the Physikalisch-Technische Bundesanstalt. Measurement of RM spectra and comparison of the specified line intensities enables a rapid efficiency calibration on most SEMs. The article reports on studies to prepare such a RM and on EDS calibration and proposes a methodology that could be implemented in current spectrometer software to enable the calibration with a minimum of operator assistance.

Abstract. Analysis of thin film layers on bulk substrates is carried out using a technique based on the ~(pz) model of the depth distribution of X-ray emission. Both the composition and thickness of individual layers can be determined provided that the same element is not present in more than a single layer.The application of this method to the analysis of thin titanium-boron nitride bilayers on silicon or molybdenum substrates is discussed. X-ray intensities were measured by energy dispersive spectroscopy with a windowless or ultra thin window detector. The thickness of a 10 nm titanium layer could be estimated to within about +__ 1 nm, which is comparable with the depth resolution attainable by Auger sputter profiling. Key words: EDS, thin films, multilayers, light elements, ~(pz).The ~(pz) technique for quantitative X-ray microanalysis has been shown to provide significantly improved quantitation, particularly for the light elements [1], because it incorporates a more physically realistic model of the generation and absorption of X-rays with mass depth than the conventional ZAF correction methods. Due to this more accurate description of the depth distribution of X-ray emission, the O(pz) approach may be readily adapted to the determination of composition and thickness in thin films and multilayers on substrates. An early attempt to apply this concept to the analysis of single layers [2] employed a purely empirical form of the emission curve. Various improved expressions for the distribution function have since been proposed [3][4][5][6].Recently, computer software has been developed capable of analyzing multilayer structures containing light elements [7], using the established theoretical model of Pouchou and Pichoir [8,9]. The present paper will discuss the application of this procedure to the analysis of very thin (-~ 10 nm) boron nitride and titanium layers deposited on silicon and molybdenum substrates, employing data generated by

The procedures presently available for X-ray microanalysis (XRMA) of layered structures derive more or less directly from the semi-empirical ϕ(ρz) models proposed in the 80's. In the last years, the spreading of advanced computer programs such as Strata made more people aware than these methods were effective, and could be used in many cases as a complement or a substitute to near-surface analysis methods. New approaches for ϕ(ρz) reconstruction, based on the angular and energetic distributions of the electrons, are emerging (e.g. IntriX model). They should enable to describe more accurately structures with strong atomic number variations. Monte-Carlo simulations (M-C) are useful to assess ϕ(ρz) models, but for daily work their effectiveness seems presently restricted to the elaboration of calibration charts to be used for repetitive situations.

Quantitative X-ray mapping has always been one of the ultimate goals of many X-ray microanalysis application programs. However, only semi-quantitative X-ray mapping (peak intensities) is commonly used in WDS and EDS techniques. EDS maps are easier to produce, because the multichannel detection gives direct access to the intensities, and also because beam scanning may be used over wide areas without any danger of spectrometer “defocusing”. WDS mapping, because of the monochannel detection, has most of the time been limited to one map per spectrometer. A few attempts have been made in both techniques to produce a quantitative map by calibration, i.e. using a few quantitative analyses performed on some pixels to calculate the composition at each pixel of an image. What we propose here is to improve the WDS acquisition scheme and the automation of the quantitation to the point where the user can collect automatically all X-ray maps of the elements in the sample and then run a quantitative routine which automatically calculates the corresponding composition images, through the usual ϕ(ρz) procedures (PAP/XPP).

The emerging class of silicon drift detectors (SDD) will significantly improve elemental x-ray mapping as performed in the scanning electron microscope (SEM) [1,2]. Several attributes of the SDD make it particularly useful for mapping. The SDD can operate with very short time constants, with the limiting value at 1 µs or less, so that the output count rate can exceed 500 kHz, which is at least an order of magnitude faster than the conventional monolithic crystal Si-EDS for similar resolution performance [3]. Moreover, individual SDDs can be made with a large active area, e.g., as large as 100 mm 2 , and arrays of SDD detectors have been demonstrated, giving solid angles approaching π steradians. With this combination of large solid angle and high pulse processing speed, the SDD becomes extremely attractive for performing elemental mapping when the x-ray flux can be made high. This is possible with SEMs capable of delivering focused probes carrying high beam current (e.g., 10 nA -500 nA), given that the specimen can withstand the high electron dose necessary to produce high x-ray count rates. Moreover, with such high x-ray flux, it becomes attractive to perform elemental mapping in the spectrum imaging mode, whereby a complete EDS spectrum, e.g., 2048 channels of 10 eV width with at least a 16-bit intensity range, is stored for each map pixel. The resulting spectrum imaging datacube thus contains all of the information possible within the performance constraints of the SDD detector. Such a database enables the user to pose any analytical spectrometry question during subsequent data evaluation with data mining.Spectrum imaging at 500 kHz with 10 ms pixels would yield approximately 5000 counts in the spectrum, which should be sufficient to detect all major constituents (concentration, C > 0.1 mass fraction) as well as many minor constituents (0.01 < C < 0.1). At 1 ms per spectrum sample, there would still be 500 counts per pixel, which could detect major constituents and be useful for surveying. A 128x128 map at 10 ms per pixel could be recorded in 164 seconds, not considering the time overhead for storing the data. At such extremely high x-ray count rates, properly binning the data stream to implement the spectrum imaging mode raises significant challenges. Spectral distortions are often detected at the shortest time constants, and deadtime correction may fail. Spectrum imaging has been explored with a 50 mm 2 SDD developed by Radiant Detectors LLC that was controlled with SAMx spectrum collection software [4,5]. The resulting spectrum image datacubes were evaluated with the SAMx software as well as the NIST LISPIX image processing software engine, which was also used for data mining. Raney nickel (which contains Ni and Al as major constituents) was used as a test material for spectrum image mapping because of its complex chemical microstructure, the differences in relative excitation the presence of minor and trace constituents both as alloying agents and as contamination. Figure 1 shows maps of AlK, NiK, and FeK ...