FAST ELEMENTAL MAPPING WITH MICRO-XRF

Transcription

1 286 FAST ELEMENTAL MAPPING WITH MICRO-XRF Haschke, M.; Rossek, U.; Tagle, R.; Waldschläger, U. Bruker Nano GmbH, Berlin, Schwarzschildstr.12 ABSTRACT X-Ray optics are now in common use for concentrating X-rays in small sample areas. Utilizing past experience a second generation of benchtop instruments have been developed for µ-xrf applications. New components like X-ray tubes with high brilliance, optics with better transmission efficiency and detectors with high count rate capability offer possibilities for improvement in the analytical performance of these instruments. This paper discusses the requirements for a fast distribution analysis based on the different components of an instrument and investigates their influence on its analytical performance. But not only the optimization of hardware is important for a high performance; the data handling needs to be efficient and convenient for the user. In particular for distribution analysis, position tagged spectroscopy (PTS) is the current state of the art. In PTS, a complete spectrum is saved for every pixel in the image, offering a wide range of possibilities for data post-processing. These possibilities are demonstrated for the analysis of a meteorite. INTRODUCTION For more than 15 years X-ray capillary optics have been used in analytics to concentrate tube radiation on µm-size spots [Carpenter, 1989; Yamamoto and Hosokawa, 1988; Haschke et.al, 2002]. This allows the analysis of small sample areas and the determination of elemental distributions with high spatial resolution. New analytical possibilities are opened because elemental analysis with X-Ray fluorescence on small spots requires only simple sample preparation and allows easy sample handling. The analytical result can be obtained quickly and independent of the size and shape of the sample. An additional advantage of this method is the high sensitivity for trace elements [Janssens et al., 2000; Haschke et al., 2002]. µ-xrf is used in many applications working with multiple material types, including the examination of minerals to determine the element distribution in geological samples, to study pigments in art objects, in forensic investigations and many others. Over the last few years µ- XRF has been able to show clearly its potential as analytical technique. As with all analytical techniques, µ-xrf is always subject to the pressure of increasing analytical performance including better sensitivity and resolution, higher speed of the measurement and better usability. With that target a concept for a new instrument has to take in to account the following issues: Faster measurement with the same or even higher sensitivity can be achieved only with higher fluorescence intensity The time for analysis can be reduced if the time for sample positioning is short Data acquisition and handling should not limit the speed of the measurement Furthermore this instrument should be adapted for applications in industry which requires a robust and failure safe design. It will be shown how these requirements were realised for the development of a new instrument from Bruker Nano the M4 Tornado (see figure 1) [Bruker webpage]. The general concept of this instrument will be discussed and its analytical performance will be documented for the examination of a Meteorite from Campo del Cielo.

2 This document was presented at the Denver X-ray Conference (DXC) on Applications of X-ray Analysis. Sponsored by the International Centre for Diffraction Data (ICDD). This document is provided by ICDD in cooperation with the authors and presenters of the DXC for the express purpose of educating the scientific community. All copyrights for the document are retained by ICDD. Usage is restricted for the purposes of education and scientific research. DXC Website ICDD Website -

3 287 Fig. 1: M4 Tornado INSTRUMENT CONCEPT The requirements for high speed mapping are examined in [Haschke and Waldschläger]. There it was found that not only spot size determines spatial resolution but also step size and accumulated intensity per pixel. Step size can be reduced to a quarter or to a fifth of spot size to improve the spatial resolution [Unser 2000]. In the case of sufficient intensity per pixel spatial resolution then can be even smaller than spot size, but it has to be considered that both of these factors increase measurement time significantly. EXCITATION OF THE SAMPLE For the excitation of the sample a fine focus X-Ray tube in combination with a poly-capillary optic is used. The poly-capillary optic can capture a large solid angle of tube radiation but only from a small area of the tube target comparable to the spot size on the sample. Therefore the brilliance of the tube is an important parameter. It depends on the size of the illuminated area and on the tube power. Because most of the tube power is converted into heat on the target and the heat needs to be dissipated the relation of tube power to spot diameter is limited. Tube brilliance is proportional to tube power and to square of spot diameter. This means even for a reduced tube power the brilliance can be enhanced if the target spot size is sufficient reduced. In the M4 Tornado a Rhodium-side window tube with a spot size of < 50 µm is used with a maximum tube power of 30 W. In comparison to a tube with a spot size of 80 µm and 50 W this increases the brilliance by more than 50 %. The spot size on the sample and the excitation intensity depends upon the X-Ray optics used. For the analysis of a wide range of elements poly-capillary optics offer the best excitation conditions. Their poly-chromatic spectrum warranties the excitation of most elements with comparable efficiency and they can focus the excitation radiation to small areas. With poly-capillary optics the highest excitation intensity can be realized. This is because they capture a large solid angle of tube radiation and can concentrate this radiation to a small spot. The main parameters of poly-cap optics are transmission efficiency and spot size. The transmission efficiency of poly-cap optics has been improved in the last few years. Now they

4 288 are in the range of %, an improvement of a factor of 2 or 3. But these optics also change the energy distribution of the spectrum low and high energies are not so efficiently transmitted as are medium energies in the range of 3 15 kev. This also influences the sensitivity for detection of elements with excitation energies in these energy ranges and needs to be considered in case of quantification. The final spot size depends strongly on the working distance i.e. on the distance from the end of the capillary optic to the sample. This is because the radiation is propagated in the optic by multiple total reflections therefore the exit radiation has a maximum divergence according to the critical angle of total reflection. For an instrument to be used in a robust environment the working distance should be large enough that the possibility of touching the optics against the sample is minimised. In the M4 Tornado the working distance is 10 mm, limiting spot size to µm. In addition the excitation intensity is partially influenced by the working distance. For a shorter working distance the poly-capillary optic needs a shorter curve radius. This reduces slightly the reflectivity in the capillary optics and increases the loss of intensity by absorption. DETECTION OF FLUORESCENCE RADIATION If the sample will be excited with high efficiency than also the high flux of fluorescence radiation needs to be detected and processed. For that purpose Silicon-Drift-Detectors (SDD) are used now [Bruker webpage]. Previously Lithium-drifted Si-detectors (Si(Li)) were used mostly for energy dispersive X-ray spectroscopy. The SDD s are not only electrically cooled and do not need liquid Nitrogen like Si(Li) s, but they have also extremely short shaping times and can therefore handle very high count rates. Another important feature of SDD s is that the energy resolution is not degraded for high count rates. Nevertheless the basic energy resolution will be influenced slightly by the shaping time i.e. for an optimum throughput of 60 kcps the energy resolution is in the range of 130 ev and for an optimum throughput of 130 kcps in the range of 140 ev. The optimum throughput generates a dead time of approx. 30 %. These values are valid for 30 mm² sensitive detector area which is also important to capture a large solid angle of fluorescence radiation from the sample. With this arrangement count rates for pure elements can be reached in the range of 400 to 500 kcps (for elements like Mn, Fe, Co, Ni, Cu). SAMPLE POSITIONING Samples for µ-xrf can be relatively large, requiring a large sample chamber for their positioning. The M4 Tornado has a sample chamber of 600 x 350 mm². The size of a sample that can be examined without new positioning on the stage is 200 x 160 mm with a maximum height of 125 mm. The sample chamber has to be closed for the measurement both due to the requirements for radiation protection and for the evacuation of the chamber. Three camera systems are used to allow exact sample positioning. One camera is in the door and allows an overview into the sample chamber like a fish eye, two further cameras allow a view perpendicular to the sample surface with different magnifications the low magnification displays an area of approx. 15 x 11 mm², the high magnification 1.5 x 1,1 mm² i.e. with 10 times higher magnification. Additionally for fast orientation on large samples a mosaic image can be generated by movement of the stage to adjacent sample positions and stitching the single images together

5 289 for a large overview. The Graphic User Interface (GUI) allows the simultaneous display of two of these images (mosaic, low magnification, high magnification, overview) which facilitate the orientation on the sample and the final positioning. In a µ-xrf instrument the X-Ray beam cannot be deflected in different positions like an electron beam in an electron-microscope. Therefore the sample has to be moved with the stage into the measurement position. For that reason there is a motorised X-Y-Z-stage which can be controlled by joystick, mouse actions or automatic functions like autofocus. The stage needs to be both fast and precise. Either stepping motors or dc-motors with encoders can be used. The M4 Tornado works with dc-motors because they allow smoother and faster movement. The M4 Tornado stage has a maximum speed of 200 mm/s. The speed is adapted to the displayed sample image i.e. differs for high and low magnification by a factor of 10. For exact sample positioning the spot size has to be taken into account. The smallest step size should be at least 1/6 to 1/10 of the spot size. Smaller step sizes are not required for an exact single point positioning. Nevertheless the step size in the M4 Tornado is less than 1 µm. Also important for minimizing the time required for distribution analysis is the mode of stage movement. The M4 Tornado performs the measurement on-the-fly i.e. the stage is continuously moving and the detector is collecting all the time. In older instruments the stage moves to a position stops, accumulates fluorescence radiation and then moves to the next position. In that case not all time can be used for data acquisition. For the measurement onthe-fly the stage has to move with a very stable and controlled speed to guarantee high reproducibility of stage positions. Therefore the stage is accelerated to a fixed velocity before the data acquisition starts. After finishing a line scan the stage slows down and moves back for the next line. Due to the measurement on-the-fly the spots are not real circles but ellipses, but for a distribution analysis this is not important. These measurements give a qualitative overview of the elemental distribution in the sample. For correct quantification single points have to be measured. A further reduction of measurement time is possible in case of measurement of distributions in a serpentine i.e. the stage movement in both directions is used for data acquisition. This reduces the total time by a factor of approx. 2, but due to a backslash of the stage in the range of 20 µm this measurement mode is possible only for step sizes larger than 50 µm. DATA PROCESSING Data handling is needed for both quantification and data evaluation of distribution analysis. Quantification is performed mostly for single point measurements. In that case it has to be considered that with µ-xrf mostly small spots in in-homogeneous or irregular shaped samples are examined. This implies that the sample composition can change with the measured position. A highly accurate analysis in this case is very challenging because it would require different sets of reference samples for calibration depending on sample position. Therefore standardless Fundamental-Parameter models are preferred for quantification in µ- XRF even though they are not as accurate [Elam et al.; 2004]. One possibility is that the element intensities are calculated based on the Sherman-relation for assumed concentrations. This procedure is repeated iteratively until the predicted and measured spectra converge to each other within given limits. Typically the calculation is convergent after a few iterations. The exact procedure used in the M4 Tornado that also estimates uncertainty of the quantification result is described elsewhere [Malzer et al.).

6 290 For distribution analysis complete data saving is of importance because mapping acquisition times can be very long. Therefore Position Tagged Spectroscopy (PTS) [Mott and Friel, 1999, Kotula et al., 2003] which saves a complete spectrum from every pixel is state of the art. In the M4 Tornado this type of data handling is called a HyperMap. It generates a four dimensional data set two dimensions are the X-Y-coordinate; third dimension is the energy of the spectrum and fourth dimension the intensity. This data set opens a wide variety of post-processing. It is possible to look at the spectrum of every pixel or sum up the spectra of several pixels to improve the counting statistics to generate the distribution for every energy range, either for elements that are in the examined sample, for the complete energy range or also for special background energies to calculate the artificial maximum pixel spectrum which contains for every channel the highest intensity for that channel of the complete mapping. This allows the identification also of hotspots in the sample to evaluate the linear distribution of every line in the mapping etc. But if a complete spectrum is saved for every pixel fast data handling is required, in particular if the measurement time per pixel can go down to less than 1 ms. The size of a complete spectrum with 4000 channels for a short measurement time will be 4 kb. For a mapping with 500 x 500 pixels with the size of the dataset dependent on the counts per channel, it can go into the range of 1 GB. ANALYSIS OF GEOLOGICAL SAMPLES Despite all improvements in the analytical performance of an instrument due to hardware engineering, the development of appropriate data mining and data handling tools are indispensable. The possibilities of the M4 software for data processing based on the HyperMap position tagged spectroscopy will be illustrated on the measurement of a meteorite sample. The meteorite sample is a polished slab of Campo del Cielo, an IA Iron meteorite with a large chromite inclusion. Figure 2 shows a video image of the polished surface of the meteorite with the inclusion. Distribution data for a total analysed area of 13.5 x 10.8 mm² was collected. The measurement was performed with a step size of 15 µm which results in a pixel number of 900 x 720. The data acquisition was performed in two subsequent runs. The measurement time per pixel was 3 ms. This results in a total measurement time of approx. 1.8 h.

7 291 Fig. 2: Examined area of the meteorite The sum spectrum (red) of the complete mapped area together with the maximum pixel spectrum (blue) is shown in figure 3. The maximum pixel spectrum (MSP) is calculated for the highest content of every channel independent of the pixel. The two spectra are normalized to the Rh-Rayleigh scattered peak, which is independent of the sample composition for better comparison. The comparison allows the identification of elements that are not distributed within the entire measured area but are present as hotspots, for example as discrete mineral phases of the meteorite. In the examined sample this is valid for Ti, Cu and Zn which cannot be detected in the sum spectrum but have a relatively high intensity in the MSP. Without the MSP these elements could not be identified and therefore their distribution not examined. Also Cr has a significant higher intensity in the MSP than in the sum spectrum i.e. it is also concentrated in limited areas. But because these areas are large enough a Cr peak can be detected in the sum spectrum and the Cr-distribution can be calculated.

8 292 Fig 3: Sum spectrum (red) and maximum pixel spectrum (blue) of the examined area normailzed to the Rh-K-scatterer The distribution of the main components of the meteorite is displayed in figure 4 but that of the elements identified only with help of the maximum pixel spectrum i.e. Ti, Cu and Zn in figure 5. The distribution of Zn shows a signal only in two small hot spots. Similar is the case for the elements Ti and Cu. They are located also only in special regions of the sample. Fig. 4: Distribution of main components of the meteorite

9 293 Fig. 5: Distribution of Ti, Cu and Zn which are identified in the maximum pixel spectrum An additional way of extracting information from the data set is the display of the intensity distribution of larger energy ranges for example the distribution of total intensity or the distribution of the Compton scattered radiation. These intensity plots display a combination of concentration and excitation efficiency or of medium atomic number of the sample, respectively. The distribution of elements with high excitation efficiency and high concentration results in a high total intensity, elements with low atomic number have a high Compton scattering intensity. Similar types of intensity distribution (electron backscatter or secondary electron) are well known from electron microscopy and also widely used to illustrate the nature of the sample. The distribution in figure 6, based on the Compton scattering, shows for the center of the image a higher intensity. There is a Chromite inclusion in the meteorite which contains larger amounts of light elements e.g., Cr and S which has a higher scattering efficiency in comparison with the Fe and Ni from the meteorite metal. The boundary of the large Chromite inclusion has a higher S-content as can be seen in figure 4 and therefore the scatter intensity in this region is also higher.

10 294 Fig. 6: Distribution of Comton scattered radiation Both the sum spectrum and the MSP also contain a few peaks which are the result of Bragg reflection. These peaks are generated by energy dispersive diffraction. The polychromatic tube radiation will be scattered on crystallites of the sample. In dependence of lattice distance and orientation of the crystallite different energies will be reflected with higher intensity than the diffuse scattering. They generate peaks in the measured spectra. In particular in the MSP they can be easy identified as seen in figure 7. If the crystallites capture a larger area the Bragg peaks can be used to display the distribution of their orientation. This is displayed in figure 8. The colors of energy regions that are used to mark the Bragg-reflexes in figure 7 are displayed in figure 8. Fig. 7: Maximum Pixel Spectrum with identified Bragg reflexes

11 295 Fig. 8: Distribution of diffraction peaks that represent the distribution of crystallite orientation Another possibility to compress the information obtained from mapping is phase analysis. Phases are sample areas of similar composition which can be presented with the same colour in the distribution as demonstrated in figure 9. The phases can be defined in different ways: from the intensities of identified elements with chemometric tools like principle component analysis (PCA) or cluster analysis (CA). In this case it is possible to make adjustments of the sensitivity or for minimum phase size to distinguish between different phases or they can be defined by objects which are sample areas that have a homogeneous element distribution. They can be highlighted in the element mapping. Fig. 9: Distribution of phases which are defined by a prociple component analysis

12 296 The distribution of phases of the meteorite defined by PCA is displayed in figure 9. In particular for the center of the meteorite it can be concluded that it is made of a rim of Schreibersite (phase P2) surrounding a core of Daubréelite (phase P3) composed of Fe-Crsulfide, with some trace elements including Mn. The core is intergrowth with elongated and oriented crystals of Cr-rich Fe-Cr-sulphides (phase 4) in a matrix of low Cr Fe-sulphides. This core of the meteorite is covered of a FeNi (phase 1) alloy. For the given phases it is possible to calculate the number of pixels that are classified in a special phase. This gives a measure for the sample area of every phase. For the selected phases these areas are displayed both as number of pixels and as part of complete area. For a given phase the spectra of every pixel can be summed up for a subsequent quantification. These sum spectra are averaging the single pixel but also improve the statistics of the spectra. These results i.e. both the area of every phase within the complete mapping and their composition as result of the quantification are summarized in table 1. Table 1: Content of the different phases and their composition of the Meteorite Phase Pixels Area % P Wt% S Wt% Cr Wt% Mn Wt% Fe Wt% Co Wt% Ni Wt% P1 / Fe-Ni alloy , P2 / Schreibersite P3 / Daubréelite P4 / Cr-Fe-Sulfide SUMMARY µ-xrf has become an analytical method for the examination of irregular shaped samples or elemental distributions of in-homogeneous samples. In particular for distribution analysis the measurement time is an important parameter that determines the usability of instruments. For short examination times a high accumulated intensity is required to reduce the statistical error and to minimize measurement times per pixel. This is possible by high excitation intensity, requiring X-ray tubes with high brilliance and X-ray optics with high transmission efficiency. Recent developments improve the tube brilliance by smaller target spots of the tube that allow even for smaller tube power while increasing the brilliance by more than 50%. Improvements of the transmission of poly-cap optics by factors between 2 and 3 also enhance the excitation intensity for the sample. The fluorescence radiation generated by the higher excitation intensity has to be accumulated with energy dispersive detectors. This requires detectors with high count rate capability and good energy resolution even for highest count rates. Silicon- Drift-Detectors have countrate capabilities of several 100 kcps and in that wide range of countrate their energy resolution is practically not changing. All these improvements can only be used if the data handling is adapted to those countrates. Despite the high acquisition speed data accumulation is still the most extensive procedure for distribution analysis, therefore the saving of all accumulated data is very important. This is possible with position tagged spectroscopy which generates multi-dimensional data sets and allows comprehensive data post-processing with functions like calculation of spectra for different samples areas and their

13 297 subsequent quantification, display of distributions of all elements and also of selected energy ranges or display of intensity profiles. These wide-ranging functionalities were used for the examination of a polished slab of a meteorite from Campo del Cielo. After the data acquisition in a relatively short time for sample areas in the range of 150 mm² and with a step size of 15 µm it was not only possible to determine the distribution of the expected elements but also to identify elements that are concentrated in hotspots and are not visible in the sum spectrum. Also a phase analysis can be performed which shows the amounts of the different phases in the analyzed sample area and their averaged composition.

14 298 REFERENCES Bruker Webpage; Carpenter, D.A; X-ray Spectr. 18; 253; (1989); Improved laboratory X-ray source for Micro- Fluorescence Analysis Elam,W.T.; Shen,R.B.; Scruggs,B.; Nicolosi, J.; Advances in X-Ray analysis, 47; (2004); 104; Accuracy of standardless FP analysis of bulk and thin film samples using a new atomic data base Haschke, M.; Scholz, W, Theis, U.; Proc. of EDXRF-98, Bologna 1998; (1998); 157; X-Ray Fluorescence in the µm-range using capillary lenses Haschke, M; Scholz, W; Theis, U; Nicolosi, J; Scruggs, B; Herczeg, L; J. de Physique IV 12; 592; (2002); Description of a new Micro X-ray spectrometer Haschke, M,; Waldschläger, U.: Spatial resolution in Micro-XRF, submitted Janssens, K.H.A.; Rindby, A.; Adams, F.C.V.; Microscopic X-Ray Fluorescence analysis, Wiley, Chichester (2000) Kanngießer, B. Haschke, M; MicroX-Ray Fluorescence Spectroscopy in Handbook of Practical X-Ray Fluorescence Analysis, Ed. By Beckhoff, B; Kanngießer; B; Langhoff, N; Wedell; N; Wolff; H; Springer, (2006) Kotula, P.G; Keenan, M.R; Michael; J.R; Microsc. Microanal. 9, 1 17, 2003; Automated Analysis of SEM X-Ray Spectral Images: A Powerful New Microanalysis Tool Malzer, W.; Haschke, M.; Waldschläger, U.; Tagle, R.; Weirauch,D.; Quantification for µ- XRF; submitted Mott, R.B; Friel, J.J. J Microsc-Oxford 193, (1999); 2 14; Saving the photons: Mapping X- rays by position-tagged spectrometry Unser, M; Proceedings of the IEEE. Vol. 88, No. 4, 2000, ; Sampling 50 Years after Shannon Yamamoto, Y; Hosokawa, Y; Jap. J. of appl. Phys. 27; L2203; (1988); Development of an innovative 5 µm focused X-Ray beam energy dispersive spectrometer and its application