摘要:

Quantitative analysis with the electron microprobe. The first 50 years and beyond† Plenary Lecture Peter Duncumb University of Cambridge and 5AWoollards Lane, Great Shelford, Cambridge, UK CB2 5LZ Received 18th September 1998, Accepted 11th December 1998 The late Raymond Castaing is rightly regarded as the ‘father’ of microprobe analysis and there have been many thousands of publications on the subject in the 50 years since his work began. This paper attempts to summarise the way in which his work launched the subject of quantitative analysis and the major trends which have taken place since.The triennial series of ‘ICXOM’ meetings, which started in 1956, have followed these trends and are well complemented by the annual meetings of the Microbeam Analysis Societies in Europe and the USA. The future will see more emphasis on ease of use and reliability of the result. Driven by user demand, and helped by ever-increasing computer power, many instruments are now equipped with an intelligent interface to guide the user in setting up the instrument, running the experiment and processing the results.Soon this will incorporate a ‘knowledge base’ to draw together the experience of experts worldwide. In 50 years the subject has made some surprising leaps forward, and will continue to surprise us for many years to come. in existence were represented (with the sole exception of the Introduction one in Moscow). It was a stimulating environment to share.I It is 50 years since the late Raymond Castaing started building have a particularly vivid memory of the ‘Zeroth’ meeting of the first electron microprobe, working at the laboratories of the Electron Probe Analysis Society of America ( later the ONERA in Paris under the supervision of Professor Guinier. Microbeam Society of America), organised by T. D. McKinley, His work was to set oV worldwide interest in the subject, K. F. J. Heinrich and D. B. Wittry in 1964.With commercial which is expanding still; the most quoted reference in the instruments well into production, results were pouring in, and literature is undoubtedly his PhD thesis of 1951.1 This remark- new ideas were emerging for improved instruments and techable piece of work described the building of the first instrument niques, driven by the needs of the user as well as by natural and contained the origins of the basic theory which we still growth. It was a melting pot of new thoughts and debate, and use today.led to the annual series of Electron Microscopy Society of As a new research student myself in 1953, my first task was America/Microbeam Analysis Society (EPASA/MAS) meetto read that thesis, and I began to feel there was nothing left ings which has continued to this day. to be done in the subject. However, there was no mention of The ‘Zeroth’ MAS6 meeting, with the fourth ICXOM7 of scanning the electron probe electronically, to image the sample the following year, marked the end of the first phase of in terms of its X-ray emission, and my supervisor in microprobe development, which then diverged into several Cambridge, Dr.V. E. Cosslett, suggested that this would be parallel streams. Consequently, we shall look first at this a suitable subject for a PhD. My own thesis thus described an period as a whole, dealing with the early ‘ZAF’ theory, and ‘X-ray scanning microscope’ and contributed not at all to then pick up some individual threads in turn: Monte Carlo quantitative analysis.Indeed, the French led the field in the modelling, practical correction procedures for bulk samples, subject for several years, believing (with some justification) and the analysis of thin films in the analytical electron microthat accurate analysis was only possible if the probe was firmly scope. Finally, we shall look at some of the ongoing problems fixed at the focus of the spectrometer. Later the gap was to in standardless analysis and spectrum simulation, and speculate be bridged with ‘quantitative imaging’, but for this paper we on what an intelligent instrument of the future might oVer.shall concentrate on techniques of point analysis. Space does not permit a rigorous review of what is now a very large Early theory subject, and for this the reader is referred to some excellent books.2–5 Instead, I shall try to describe the flow of ideas Castaing’s first and second approximations which has taken place in the past 50 years in order to see One of the beauties of nature is the regular way that the where these might lead in the future.energy levels within an atom vary with atomic number, giving The first 15 years of microprobe analysis saw rapid growth; a predictable means of identifying elements by their character- the first ICXOM in Cambridge in 1956 had two papers on the istic X-ray emission, as Moseley’s work in 1912 well demon- subject, whereas the fourth in Paris in 1965 contained over 50.strated. In principle, the choice of using K, L or M lines gives At a workshop organised by L. S. Birks at the US Naval access to all elements in the Periodic Table from lithium (Z= Research Laboratory in 1958, all 12 instruments known to be 3) upwards, although in practice current detectors are unable to extend much below the K emission of boron (Z=5) at 185 eV. In the electron microprobe, emission is excited by the †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998.impact of a finely focused electron beam, of energy usually in J. Anal. At. Spectrom., 1999, 14, 357–366 357the range 5–30 keV, the latter being suYcient to excite adequately the L lines of uranium (Z=92) and the K lines of molybdenum (Z=42). The energy of the beam, and the scattering which it undergoes, determines the penetration of electrons within the sample and hence the spatial resolution of the analysis.Depending on the density of the sample and the beam energy, the depth analysed may range from about 0.1 to 10 mm, although the choice of kV is determined by many other factors, such as the critical excitation potential of the element to be analysed. The mass concentration cA of an element A in a multielement sample may be found by measuring the intensity of its characteristic emission, and to a first approximation there is a linear relation between the two.This leads to the simple relation given in Castaing’s thesis: Fig. 1 Primary X-rays are generated by incident electrons of energy E0 at various depths rz beneath the surface and are collected at angle kA#cA h by the spectrometer. The shaded area represents the volume within which the electron energy E exceeds the critical excitation energy Ec where kA is the ratio of the intensities measured from the for a given element, and hence the volume from which information sample and from a pure standard of element A.In general, about that element can be obtained. The curve on the right represents inter-element eVects are less important than in other forms of the distribution of characteristic emission with depth, plotted horizon- spectroscopy, but cannot be ignored. To express this departure tally, showing the maximum which occurs as the electrons are scattered from linearity, Castaing’s second approximation weighted the away from the incident direction before decaying exponentially to zero.contribution from each element with an ‘alpha’ factor, determined empirically, such that The absorption correction kA=aAcA/.i aici When Castaing started his work, these processes were not well known, and he expressed the primary excitation as a distri- To this point no physics is involved. The method was developed bution of ionisation w with mass depth rz beneath the surface. further by Ziebold and Ogilvie8 and is easy to apply once the If w(rz) is known, together with the appropriate mass absorpa factors have been established, but the results can be mislead- tion coeYcient m/r and take-oV angle h, the contribution from ing when the non-linearity is severe, and applies to only one set each layer beneath the surface can be summed to arrive at the of conditions.We therefore look for an accurate method for total emitted intensity. Thus the overall fraction of X-rays correcting the non-linearity which can be applied in all circum- leaving the surface is stances and yet requires only one calibration standard for each element. This requires a separate understanding of the physical f(x)= .w(rz) e-xrz d(rz) . w(rz) d(rz) (1) eVects occurring, so that they can be separately modelled. where x=m/r cosec h. Modelling the physics Unfortunately, w(rz) was not known at the time Castaing started his work, and he had to construct curves for f(x) from Fully accurate modelling is not possible, either because the experiments on known samples with varying degrees of tilt.physics is insuYciently well understood or the fundamental He recognised, however, that a much more satisfactory solu- physical data are simply not available. A tailor measures for tion would be obtained by measuring w(rz) directly, using a a suit, but knows he will have to trim the cloth when he comes thin layer of a second element buried at known depths beneath to test the actual fit because his measurements describe incomthe surface.Fig. 2 shows this principle. Together with pletely the complex shape of the human body. So it is in Descamps in 1955,9 he carried out a series of such measure- physical modelling; the art lies in specifying the main paramments, using evaporated tracers of zinc, copper and bismuth eters with an accuracy which is justified, but no more than is to obtain w(rz) curves for copper, aluminium and gold at justified, by knowledge of the physics, and ‘trimming’ each 29 kV.(Castaing later admitted that he had intended the parameter within these limits to give the best fit overall. In voltage to be 30 kV but had not allowed for the bias voltage microprobe analysis, this means comparing the measured with in the electron gun!) To extend the range of elements and the true composition of a large number of known samples, covering as wide a range of conditions as possible. What, then, are the physical processes taking place in microprobe analysis? Electrons focused on the sample are slowed down and scattered as they penetrate beneath the surface, some re-emerging as backscattered electrons (Fig. 1). Characteristic X-ray emission from the element to be analysed is generated within the volume irradiated, and some of this enters the spectrometer after undergoing absorption on its way out of the sample. The measured intensity may also be modified by secondary fluorescence generated deeper within the specimen.The resultant intensity is compared with that from a known standard and the ratio corrected to give the true concentration of the analysed element in the sample. Fig. 2 The tracer method used by Castaing and Descamps9 for Some complex processes are thus involved as the electrons determining the depth distribution w(rz). The Ka intensity from a slow down, are scattered and cause ionisation, resulting in thin layer of zinc, buried at depth rz in copper, is normalised to the radiation which undergoes absorption and may be enhanced intensity from an identical layer of zinc in free space. Similarly, tracers by fluorescence.These are the components which are common of magnesium and bismuth were used to determine w(rz) for Ka radiation from aluminium and La from gold. to most models. 358 J. Anal. At. Spectrom., 1999, 14, 357–366voltages covered, tracer experiments have since been carried to the eVect both from characteristic radiation and from the out by many others, summarised, for example, by Karduck continuum.A similar eVect may occur with subsurface precipiand Rehbach.10 tates too deeply buried for direct excitation by the electrons The first true model for the absorption correction was but within range of the primary emission. Correction is proposed by Philibert of IRSID in Paris. He was the first to normally impossible and the ‘fluorescence uncertainty’ can apply quantitative microanalysis to practical problems in the only loosely be quantified.steel industry, and was approaching the problem as a user. Fluorescence by the continuum, being excited by polychro- His model took the form of a quadratic equation for f(x) as matic radiation, is diYcult to model, and little progress was a function of atomic number and incident energy, which made until Henoc19 was able to bring the computer to bear in greatly speeded the correction process and made it more the late 1960s.Fortunately, a degree of compensation occurs versatile. His work was presented at the third ICXOM in when the intensity from the sample is compared with that 1962.11 At the same meeting, Green12 described a comprehen- from a standard so that the eVect may often be neglected, but sive set of f(x) curves obtained experimentally, showing the this is not the case in standardless analysis, as we shall eVect on absorption of critical excitation potential Ec.It was see below. natural to try to combine the two, and at the ‘Zeroth’ meeting in 1964, Duncumb and Shields,13 at Tube Investments near EVect of atomic number Cambridge, described a simple modification to Philibert’s Surprisingly, the actual generation of X-rays within the sample, expression to accommodate this eVect, agreeing well with and the way it depended on atomic number, was the last Green’s data. With hindsight, the work would have proceeded process to be well understood.The intensity Ig generated from much more quickly had computers been widely available; as element A in an alloy AB can be expressed as it was, we had to travel to IBM in London to run our Fortran programs, often to find that the previous day’s run had failed Ig=constant×cARAB(N0/AA).QA/SAB dE (2) from some trivial error. Later, Heinrich14 was able to improve where the integral term represents the intensity generated Philibert’s expression still further to a form which has survived along the track of an electron and the R factor allows for the to the present day.energy loss due to backscatter. The integral term follows Accurate modelling of the absorption correction would be directly from the definition of Q as the ionisation cross-section of no avail without good data for the mass absorption at a given point along the track, combined with the stopping coeYcients. In the mid-1960s, good experimental data were power S=-dE/d(rx) as the rate of energy loss with mass sparse, and there existed a strong need to create a coherent thickness at that point.N0 is Avogadro’s number and AA the set of tables covering the absorption of the major K, L and atomic mass of element A. The constant contains terms which M lines for all elements in the Periodic Table. Also at the cancel out in taking the ratio of Ig with the pure standard A. ‘Zeroth’ meeting was the reporting of such a set of tables Both Q and S were fairly well known in terms of electron created by Heinrich15 at the National Bureau of Standards energy and target atomic number, but not well enough to be (NBS, later NIST).These tables relied upon fitting the mass confident of an accurate analysis under all conditions, particu- absorption coeYcients between absorption edges with sets of larly for low voltages and low atomic number. Initially R smooth polynomials, allowing the coeYcient for any could only be estimated from the number of electrons back- wavelength in any element to be quickly computed.scattered, since little was known about their energy. Hence This represented a major step forward in practical there were three components to the expression, Q, S and R, microprobe analysis and set the scene for the development of each to be determined accurately and, as soon as numerical later correction programs. In addition to giving the coeYcients integration became practically possible using the computer, of the main characteristic lines, it also allowed the calculation the modelling process could begin.Earlier models by Archard of absorption over the range of wavelengths in the conand Mulvey20 and by Poole and Thomas21 provided valuable tinuum—important for simulation of the overall spectrum. stepping stones. Values for soft X-ray lines, where absorption is very strong, In Castaing’s thesis, the eVect of atomic number was treated have been separately studied by Henke et al.16 and later by as an empirical a correction, but by 1960 the way in which Bastin and Heijligers,17 but Heinrich’s work provided a good backscatter reduced the eVective intensity was becoming better framework over the majority of the Periodic Table.An understood. This was enthusiastically explained to me in that improved set of tables was published at the 11th ICXOM in year by Castaing himself as he was driving me across Paris— 198618 and is in wide use today. It is timely to acknowledge two activities which were not entirely compatible with one the role played by the NIST Group in quantitative microanalanother! It was clear that we needed to know the energy ysis over the years, both in the work which has come out of distribution of backscattered electrons in order to calculate the laboratory and in the workshops and meetings which have the ionisation which was lost to the sample.Some data existed brought together scientists at appropriate intervals.in the literature, but it was not until Bishop in Cambridge took up the problem as a PhD topic that accurate information Fluorescence eVects became available. This was published at the fourth ICXOM Curiously, the eVect of fluorescence excited by characteristic in Paris in 1965,22 and led Duncumb and Reed23 to derive a radiation was probably the first to be well understood, at least polynomial for calculating the backscatter parameter R as a for homogeneous samples, being well covered in an appendix function of incident energy and atomic number.Sewell et al.24 of Castaing’s thesis. He was able to derive the contribution later extended this approach to include the eVect of tilting the analytically for K excitation with suYcient accuracy for most incident electron beam. purposes, even where the eVect is strong—reaching, for example, 30% for a trace of chromium in steel. Since the early The ZAF correction procedure work, others have extended the theory to include K–L and By 1965, much of the ZAF correction procedure was in place, L–K excitation and these developments are well described and is in use to the present day.Correcting eqn. (2) for by Reed.4 absorption and fluorescence gives the measured intensity: Probably the greatest diYculty occurs at phase boundaries in non-homogeneous samples, where the interface may appear IAB=constant×cA RAB (N0/AA) to be blurred by fluorescence excitation across the boundary.A full discussion of this problem is given by Reed,4 referring ×.QA/SAB dE×f(x)AB(1+F)AB (3) J. Anal. At. Spectrom., 1999, 14, 357–366 359where F represents the combined eVect of fluorescence from the characteristic and from the continuum. Finally, if eqn. (3) is also applied to the pure standard, the ZAF correction for the measured intensity ratio may be stated as kA=cA RAB .QA/SAB dE RA .QA/SA dE × f(x)AB f(x)A × (1+F)AB (1+F)A or kA=cA{Z}{A}{F} (4) Since each of the three components is itself a function of concentration, the solution for cA, given kA, is obtained iteratively. Alternatively, if cA is approximately known, the correction factors {Z}, {A} and {F} can be calculated directly—sometimes a useful step in determining the best experimental conditions to use for accurate analysis.The main shortcoming in the theory as it stood in 1964 was the inability to give accurate results for the light elements, such as carbon, nitrogen and oxygen—elements of great interest to metallurgists and biologists alike.Two things then happened: it was demonstrated (again at the 1964 meeting) that carbon K radiation could be eYciently dispersed with the aid Fig. 3 Single and multiple scattering models for Monte Carlo of an artificial crystal of lead stearate, and the following year simulation, showing the relative number of events assumed and the the advent of the Monte Carlo technique made it possible to corresponding elastic and inelastic cross-sections.compute w(rz) and the corresponding absorption correction with greatly improved accuracy. These developments led to a although in practice most ‘single scattering’ models still flurry of work, both in light element analysis and in exploiting embody some elements of simplification from the multiple the other avenues unlocked by the invention of the Monte approach. Carlo method. Knowing the ionisation cross-section Q along the track of each electron, it is then possible to total up the X-ray emission Monte Carlo simulation techniques occurring in each layer of the sample, from all the electrons, to give the distribution in depth w(rz).The more trajectories Although known for several years in particle physics, the which are computed, the better is the statistics and the application of the Monte Carlo simulation technique to microsmoother the curve. From this point the absorption correction probe analysis was first explored by Green.25 His successor in may be calculated by numerical integration, as indicated Cosslett’s Group in Cambridge was Bishop,26 who carried the in eqn.(1). technique forward to practicality and tested it against his In collaboration with Bishop, and also reported at the 1965 experimental measurement of the energy distribution of backmeeting, Duncumb and Melford28 applied the Monte Carlo scattered electrons. His work was reported at the fourth method to the analysis of carbon in steel, showing that good ICXOM meeting in Paris at the same time as some similar results could be obtained if the w(rz) curve were accurately work by Shimizu et al.,27 who was close behind Green and known.As Fig. 4 shows, the shape of this curve is significantly Bishop. In essence, the principle is very simple and needed only a fast computer to put into eVect, although the term ‘fast’ then meant something diVerent to what it does now. The idea was to simulate the trajectories of a large number of electrons in an amorphous continuum of known density.Electrons are allowed to arrive randomly at the target atoms and to be scattered in accordance with the assumed laws of physics. Two processes were recognised: the near-continuous retardation which electrons undergo as they interact with loosely bound electrons and the elastic scattering resulting from interaction with the atomic nuclei. For the inelastic process Bishop made use of the 1931 Bethe relation, giving retardation S as a function of energy, and for the scattering events he used a screened Rutherford model from student physics.Simple though these assumptions were, they gave good agreement with his experimental measurements and gave confidence that the X-ray calculations would be accurate. Because of computer speed limitations, it was not at first practicable to calculate every scattering event, which run into many thousands for each electron trajectory. Consequently, the trajectory was divided into an arbitrary number of steps of fixed length, commonly 25–50, and the scattering crosssection increased appropriately. EVects of electron straggling Fig. 4 Depth distribution w(rz) for carbon Ka radiation in iron at were neglected and all electrons were assumed to have the 20 kV, calculated by Monte Carlo simulation by Bishop26 and tested same range. This is referred to as a multiple scattering experimentally by Duncumb and Melford28 (1965).Most of the approach, in contrast to the single scattering model describing radiation entering the detector comes from within 0.1 mg cm-2 (0.12 mm) of the sample surface. every event. The distinction is shown schematically in Fig. 3, 360 J. Anal. At. Spectrom., 1999, 14, 357–366diVerent to the curve for iron, owing to (with hindsight) the of known samples—preferably on a large number of varied and careful analyses. This comparison is expressed in the form high overvoltage (E0/Ec) for carbon K radiation and relatively low ionisation near the surface.The value at the surface of an error histogram, showing the relative diVerence between the measured and calculated k ratios for each sample, and becomes much more important, and indeed may be the major determinant of the intensity if the absorption is very high. expressing the overall result in terms of the mean error and standard deviation. Poole and Thomas initiated this approach Thus, any attempt to fit these curves must represent the surface ionisation w(0) accurately and follow the correct shape at least in their 1964 paper,21 showing that their correction procedure was superior to four others in common use at the time, and to the depth from which X-rays can escape.If this can be done for all conditions, it is possible to carry out a series of this type of contest has continued at intervals ever since. So it is that correction procedures have obeyed the rule of the Monte Carlo calculations once and for all and re-create w(rz) in parameterised form as an everyday correction procedure.survival of the fittest. In the early stages, the cynics would say that you could Whilst the first practical use of the Monte Carlo method was in light element analysis, it was quickly realised that it always prove that your own theory was the best if you could choose your own data set, but this is no longer true. As a could be usefully applied where the sample geometry was nonideal, for example, where the incident beam was inclined to further major contribution from NIST to the subject, Heinrich36 has collected a set of experimental data on 1826 the surface or the sample was layered. In the former case, the multiple scattering model can give useful information about well characterised binary alloys for exactly this purpose, and compared all the popular correction procedures on the same the way in which the w(rz) and f(x) curves are modified, but for the study of thin layers with or without a substrate the basis.By selecting appropriate sub-sets (e.g., diVerent beam voltages), the modeller can minimise the systematic errors in step length may be too large. For these reasons, and to avoid the approximations inherent in multiple scattering, the single each part of his theory separately (e.g., the absorption correction), whilst comparing the remaining random errors with the scattering model has received increasing attention.Some early results were reported by Reimer and Krefting29 at a workshop results of others. Overall, Heinrich found that the rms errors given by the best procedures fell in the region of 1.1–1.3% for organised by NIST in 1975, but it was some time before computers were fast enough to explore the method thoroughly. elements with atomic numbers >10. For the lighter elements, special considerations prevail and are discussed for example A more recent assessment of the Monte Carlo method was presented by Karduck and Rehbach30 at a further NIST by Bastin and Heijligers.17 So what have been the nature of these improvements? The workshop in 1988, which brought together advances in quantitative analysis in much the way that those in 1967 and 1975 point of departure is the ZAF procedure, which centres around early approximations made by Philibert for the depth distri- had done. They described some meticulous tracer experiments determining w(rz), compared with single scattering Monte bution w(rz).We have already seen how this can be improved in selected conditions by use of a Monte Carlo technique. Carlo calculations derived from Reimer’s model. This paid particular attention to the low energy X-ray emission and Hence a first task must be to model the true shape of the w(rz) curve for a range of target materials and beam energies, included the contribution from the production of fast secondary electrons (see Fig. 3).The variation of electron backscatter and a number of so-called ‘w(rz) methods’; sketched in Fig. 5, have attempted to do this. coeYcient with incident energy, not explained by the multiple scattering approach, was well described, and the results provided some definitive data on which to base some practical correction procedures. A further review of the whole subject was presented by Reimer at the European Microbeam Analysis Society (EMAS) workshop in 1995.31 There is still room for debate on how to make best use of the Monte Carlo technique in practical quantitative analysis.Is it a technique for the average user or does it belong to the underlying science? Joy and co-workers have attempted to answer this question, first establishing the basic cross-sections to be used,32,33 and leading to a practical guide to using the technique on a personal computer.34 The guide comes complete with computer programs for experimentation. Likewise, Reimer et al.35 have described the program Mocasim which uses their single scattering model to simulate complex layered structures and particle geometries.Hence much has been done to popularise the Monte Carlo method and establish it with the user, and it seems likely that more and more users will turn to it in unusual situations, particularly where the sample geometry is complex. Improved correction procedures Whilst a sound scientific basis is essential to any analytical technique, there comes a point where the scientist must make it as simple as possible to use.The average user does not wish to be concerned with the science, any more than the average car driver wishes to know exactly how the engine works. Unfortunately, the ‘driver interface’ of the microprobe analyser is not yet as well evolved as that of the motor car and will Fig. 5 Various models for representing depth distribution w(rz): (a) as provide an active topic of research for many years to come.measured by Castaing; (b)–(f ) mathematical models capable of calcu- The main criterion for judging whether an improvement is lation for a given radiation, target atomic number and incident beam energy. real in terms of accuracy is to test the theory by the analysis J. Anal. At. Spectrom., 1999, 14, 357–366 361The simplest of these is the quadrilateral model of Scott Pouchou and Pichoir,43 Amman and Karduck44 and Bastin and co-workers.45 Alternatively, as Karduck46 has shown the and Love,37 which allows a remarkable improvement in accuracy for the light elements because of the ability to set buried layers may be progressively uncovered by physical means, using sputter erosion or ‘dimple-grinding’.This allows independently the surface ionisation, as well as the position of the maximum and eVective penetration. As early as 1958, an improved depth resolution to be achieved through the use of a low energy beam.The analysis of multilayers is a growth Wittry38 had proposed that a Gaussian curve could be expected to fit well a diVusion process of this sort, although insuYcient area which will eventually become routine, but which at present requires great care to achieve meaningful results. data were available at that time to test the principle widely. It was not until 1981 with the work of Packwood and Brown39 that this was seriously attempted, and a Gaussian model was Thin film analysis used to form the basis for a sophisticated procedure PROZA, described by Bastin and Heijligers.17 Using this model, the The spatial resolution of analysis in bulk samples is always limited by the penetration of electrons into the sample, and it authors demonstrated remarkable success in analysing for the elements boron, carbon, nitrogen and oxygen, although their has long been known that this limit can be removed if the sample is prepared in the form of a thin film through which success was due equally to careful experimental technique.Recently they have improved PROZA still further40 by splitting the electrons can pass without being scattered sideways. An instrument combining the imaging facilities of a transmission the single Gaussian into two halves centred about the peak of the w(rz) curve, as proposed at the 13th ICXOM in 1992 by electron microscope with the analytical capability of a microprobe analyser (EMMA) was described by Duncumb47 at the Merlet.41 This has enabled the model to be accurately fitted to any w(rz) curve with only four independent parameters— ‘Zeroth’ meeting and became the predecessor of EMMA-4, the first commercial analytical electron microscope.The tech- the surface ionisation w(0), the peak position, the decay rate and the overall integral under the curve. Each of these nique relied upon focusing the electron probe to 100 nm or less and detecting the X-rays with a high-eYciency gas pro- parameters is then further expressed as a function of beam voltage, atomic number, etc.portional counter, although later a focusing crystal spectrometer was added. Nowadays the use of an Si(Li) or Ge In a similar way, Pouchou and Pichoir42 had shown that the w(rz) curve could be represented as a pair of parabolae detector, combined with a high brightness field emission gun, has reduced the resolution limit to <1 nm, as evidenced, for joining not at the peak, but at the point of inflection further down the curve. This ‘PAP’ model allowed an easy integration example, by Bando et al.48 in the imaging and detection of single atomic layers of Al–O and Al–N layers in an alumina of the area under the curve to obtain the total intensity and also f(x), the fraction of emission emerging.Later they modi- polytype. The technique has been used extensively both in biological and in materials applications. fied this in the ‘XPP’ model to an exponentially based expression which was claimed to be simpler and easier to Conversion of the X-ray data into chemical composition is simpler than with a bulk sample, since absorption in a thin apply. In all cases four or sometimes five parameters are needed to describe the shape of the w(rz) curve accurately and sample is usually negligible.On the other hand, the thickness of the film is often not well known and the best that can be the choice of the best expressions seems to be a numerical art in addition to being based on science.Nevertheless, they work done is to obtain the ratios of elements present rather than the absolute amounts. In principle this is straightforward since well, as error histograms show, and all the above procedures are in widespread use. the intensity from a given element is simply proportional to the number of ionisations multiplied by the fluorescence yield Given the means of obtaining accurate w(rz) curves for homogeneous bulk material, the way is open for the study of v, so that the ratio between two elements A and B is given by: layered samples containing one or more layers parallel to the surface in the region of interest.The simple case of a thin film IA/IB=[(vA/vB) (AB/AA) (QA/QB) (eA/eB)] cA/cB (5) on a bulk substrate was early studied to determine both film thickness and composition, and it was soon realised that the where AA and AB are their respective atomic masses. We are now comparing two diVerent wavelengths from the same technique could detect film thicknesses of less than one monolayer, say of copper on silver.sample, so we have to correct for the spectrometer eYciency e. This was initially determined by calibrating the spectrometer This stimulated interest, particularly in the semiconductor industry, in determining the existence of a buried layer, which from pure element standards of the bulk material—a proposal made simultaneously in 1967 by our own laboratory in Tube in principle is possible if the variation of intensity from an element within it is accurately measured as a function of Investments and by IRSID in Paris, with whom we were in close contact.49,50 incident beam energy.This technique can even be extended to the analysis of multiple layers (Fig. 6), as demonstrated by Practical applications of the analytical electron microscope (AEM) were made simultaneously in the biological and materials fields—the former by Hall in Cambridge and the latter by CliV and Lorimer in Manchester.Each field demanded a diVerent approach to quantification, Hall preferring to relate element composition to cell mass, determined by measuring the intensity of the continuum. CliV and Lorimer pursued the measurement of element ratios, but introduced a ‘k factor’, determined empirically for each pair of elements, to replace the term involving v, A, Q and e in eqn. (5). Each method was widely adopted by others in the respective fields as further instruments became available.Both are reviewed extensively by Nicholson.51 Fig. 6 Three methods for determining element distribution in a layered Progress at Manchester was matched in the USA by a group sample: (a) by varying incident beam energy, (b) by scanning a under Lyman, Goldstein and Williams at LeHigh University,52 bevelled sample at low kV and (c) by measuring a sample sputtered also in the materials field, and both laboratories have built in situ, also at low kV.The sputter technique allows thin layers to be special instruments to exploit the potential resolution of the progressively uncovered, giving a depth resolution which may be as low as 1 nm. AEM to the full. Progress over the past three decades was 362 J. Anal. At. Spectrom., 1999, 14, 357–366ignored. This arises from the eYciency with which X-rays are generated (element sensitive) and the proportion which are finally detected (energy sensitive). In full, the intensity IA measured for a given peak from element A in the pure standard may be written as IA={vA pA}{VeA} RA(N0/AA) .QA/SA dE×f(x)A(1+F) (6) where the new terms vA, pA, V and eA refer to the fluorescence yield, transition probability, detector solid angle and detector eYciency, respectively, terms which all cancel out in the ZAF eqn.(4). The detector eYciency eA is made up in turn of the transmission of the window and absorption of the detector itself. The fluorescence enhancement (1+F) refers to that from the continuum, which is no longer negligible.Hence uncertainties exist in standardless analysis which are not present with measured standards. To some extent these can be Fig. 7 Improvement in the minimum detectable level of an element overcome by calibration with a single known element, which by electron probe microanalysis, which was much reduced after 1970 eVectively calibrates the detector solid angle, but still a com- by the use of thin samples in the analytical electron microscope parison has to be made between radiations of diVerent wave- (AEM) and scanning transmission electron microscope (STEM) (CliV and Lorimer53).lengths. Alternatively, the element compositions may be normalised to add to 100%—a useful shortcut provided that no elements are omitted. summarised by CliV and Lorimer53 (Fig. 7), showing that the Pouchou56 discussed the factors governing the accuracy of minimum detectable limit has come down by six orders of standardless analysis, concluding that for the major K lines magnitude in that time to just a few atoms.Other techniques, an accuracy of 1% is practicable. For L lines the fluorescence such as electron energy loss spectrometry, are superior in the yields are less well known and, as La�ba�r57 has shown, are detection of single atoms, but the ease with which most complicated by radiationless Coster–Kronig transitions occurelements in the Periodic Table can be detected and quantified ring in the rare earth elements—an eVect particularly importstill leaves the AEM much in favour as a quantitative ant in the atomic number range 50–75.The transition instrument in many fields of science. probability pA is also poorly known and the best way forward may be to arrive at a set of semi-empirical values for the product of v with p, tested against measured intensities of La Standardless analysis and spectrum simulation and Ma lines.Indeed some empiricism seems essential. Using Energy or wavelength dispersion the NIST DTSA program, Newbury et al.58 find large errors in calculating pure element intensities from first principles and It is painful to recall now the many steps in making a advocate storing a suite of pure standard measurements for quantitative analysis in the early days, from preparing the use in the conventional k-ratio method. sample, obtaining a vacuum, aligning the column, setting the The main problem comes with the low energy radiations probe conditions and making surese did not vary as the below, say, 1 kV, where the window eYciency is less well spectrometer was set by hand on each peak in turn on sample known and the detector performance more problematical.For and standard. First to be automated was the vacuum system example, an error of 5 nm in the assumed thickness of the and column alignment, followed by servo control of the gold contact layer, or of 20 nm in the silicon dead layer, would stage and spectrometers. The real opportunity to simplify the each give rise to a 15% error in the intensity of carbon K process came, however, with the introduction of the Si(Li) radiation.A similar error could easily arise from uncertainty energy dispersive detector in 1968, which was faster, more in the fluorescence yield or transition probability for some stable and more eYcient than the crystal spectrometer and elements.Nevertheless, Pouchou is hopeful that an accuracy more easily controlled by computer. The big disadvantage, of of 5% may be achievable for K lines in this region, but this course, is that the energy resolution is worse by an order of will certainly require a better understanding of window and magnitude, which is why the crystal spectrometer is still detector performance and of the shape of the continuum at essential for quantitative work where peaks overlap, such as low energies.To add to these diYculties, detection limits in in rare earth samples. For low voltage work also, good this region are greatly reduced by the higher continuum and dispersion in the region below 1 kV is needed to improve the poorer resolution of the detector. In all these situations, peak-to-background ratio and separate K lines from L and M spectrum simulation can help us to understand what is going radiations of heavier elements. The pros and cons of ED versus on. WD spectrometry are well discussed by Reed54 and others in a recent volume on X-ray spectrometry in memory of the late C.Fiori of NIST. Spectrum simulation—how? In this same volume there is a good account of the growing Simulation of the entire spectrum is achieved by adding a interest in spectrum simulation, stimulated by the work of model for the continuum, inserting the other peaks in each Swyt and Fiori55 on a ‘desk-top spectrum analyser’ (DTSA). line series and applying a spread function appropriate to the This was a natural development from so-called ‘standardless’ resolution of the detector.The full technique is diYcult with analysis, in which the peak intensities from the pure standards a WD spectrometer as the eYciency of the crystal is less well were calculated, rather than measured, to obtain the known, but Reed59 has produced a useful program, Virtual appropriate k ratios. We return to spectrum simulation later. WDS, to simulate potential interferences from the main and higher order peaks from a range of crystals.This information Calculation of standard peak intensities is needed in order to choose the best line for analysis in complex cases such as the determination of rare earths in In calculating the intensity from the pure standard in isolation, the constant term originating from eqn. (3) can no longer be minerals, and also to indicate the best position for measuring J. Anal. At. Spectrom., 1999, 14, 357–366 363the background.In ED analysis, there are fewer peaks to contend with, but the level of the continuum is higher and must be known with greater accuracy. Calculation of the continuum is normally made from expressions not diVering greatly from Kramers’ law of 1923. This predicts a linear decrease in the energy per unit energy interval with increasing energy, falling to zero at the energy of the incident electron beam. The intensity emerging from the surface is modified by absorption and may be calculated using Heinrich’s absorption coeYcients by a method akin to the Philibert correction.The correction for backscatter is also straightforward; thus far so good. Some doubt remains as to how the Kramers expression should be adjusted (a) to maintain the spectral shape in agreement with experiment and (b) to give the observed variation of total intensity with target atomic number, but experimental studies by Small et al.60 and Trincavelli et al.61 have helped to resolve this problem.Fig. 8 Measured spectrum from carbon-coated gallium phosphide, The problem of determining the relative peak heights within showing C Ka, Ga La+b and P Ka+b peaks. The interpolated a series remains. All the above modelling applies to the a background starts from the first non-zero channel at 0.12 keV, which in this case is due to electronic noise, and misses an unsuspected peak peaks in the K, L or M series, and the others are obtained by from oxygen Ka where the background is sharply sloping.By contrast, ratio from tables—those contained in DTSA are reasonably the synthesised background follows the slope and shows also the comprehensive. No allowance is made for diVerential absorp- absorption edge under the Ga L peaks. tion, or for chemical eVects, so errors are to be expected particularly at low energies. Allied to this, the fluorescence yields are not well known in this region, as noted above. Despite these limitations, spectrum simulation has great value, purpose of the example is to show the eVect of the L3 edge on the continuum under the Ga L peaks—an eVect which is which is not yet fully developed.obscured in the normal process of interpolation. This can lead to major errors in background subtraction in this region, such Spectrum simulation—why? as missing the presence of the small oxygen peak altogether. The idea therefore grows that the unskilled user can usefully Given that ED spectra can be accurately simulated, what value will they have to the average user? There are probably four require the instrument to simulate what it thinks the spectrum should look like, based on the information it has acquired. main benefits to be obtained: (1) As an aid to setting up the experiment.If the composition Auto identification can go wrong: the window may be thicker than the instrument has been told; there may be line inter- of the sample is approximately known, the general shape of the spectrum can give a useful indication of peak overlaps, ferences which make an automatic analysis meaningless.In eVect, the instrument should say to the user, ‘This is what I noise and detectability which is helpful in determining the optimum beam voltage, probe current and counting time. think the spectrum should look like; have I missed anything?’. An admission of doubt coupled with the means to learn from (2) As a confirmation of peak identification.Auto identification is a facility found on many microprobes, which experience would give the instrument a useful degree of intelligence. embodies background subtraction and peak stripping. These processes can go badly wrong in some circumstances and a visual comparison of the measured spectrum with the simulated The intelligent microprobe spectrum provides a valuable assurance to the novice operator. (3) As a sensitivity check of window parameters. Modern We can define four stages on the way to an intelligent instrument: thin windows are complex in construction and diYcult to characterise, lending uncertainty to the calculation of detector (1) The automated repetition of routine tasks, such as pump-down procedure, accomplished by hard-wired control eYciency.It is therefore helpful to simulate a measured spectrum with a range of window thicknesses to see which in the 1960s and 1970s. (2) The programming of beam conditions, sample position best corresponds with reality. The same is true of the detector itself, e.g., the thickness of the gold contact or silicon dead and spectrometer settings for unattended analysis by computer control in the 1980s.layer; these may be distinguished to some extent by accurate measurements of peak intensity either side of their respective (3) The integrated design, execution and reporting of an experiment, allowing some elements of decision-making to the absorption edges.(4) As a check on background subtraction. This is often instrument during unattended or remote operation. This degree of intelligence now exists, and could well include the benefits carried out by interpolation across each peak—a method which obscures any absorption edges of the peak and which of spectrum synthesis and other techniques which are readily programmable. fails at low energy where the background is rapidly changing. Both of these problems can be overcome if the background is (4) The build-up of an ‘expert system’ within the instrument to make available to a newcomer the wisdom of an experienced synthesised from a first approximate analysis of the sample, which can later be refined when the composition is better user, and to add to the ‘knowledge base’ as further experience is acquired. A new style of programming is needed here to known.As an example of this last point, Fig. 8 shows a measured handle the complexity of the data, to test it for consistency and to include not only logical rules but also ‘fuzzy’ data spectrum from gallium phosphide, compared with a spectrum simulated under the same conditions.The beam voltage was embodying opinion rather than certainty. Fig. 9 illustrates a way of bringing these requirements 20 kV and the detector was ‘windowless’ but limited at low energies by the gold contact layer and an assumed dead layer together, identifying the essential steps in solving the overall problem, from the first symptom to the final cure.The of silicon. Minor peaks for carbon and oxygen are quickly identified by comparison with the simulated spectrum, but the microprobe itself is operated through an integrated control 364 J. Anal. At. Spectrom., 1999, 14, 357–366and ‘Solution’ beyond the analysis itself (see Fig. 9). Thus the ‘Problem’ for a steelworks manager might be ‘Why is my steel cracking?’ and the ‘Solution’ might be ‘Reduce the level of scrap input’.The intermediate steps in arriving at that solution are not of direct interest; the manager simply wants to know what to do with a reasoned degree of assurance. The ‘expert’, whether in the form of human or machine intelligence, can determine what sort of analysis should be performed, carry it out and present a recommendation. But all this is not yet; the centenary of Castaing’s work in 2048 will show whether or not this becomes a reality. Acknowledgements I acknowledge gratefully many recent discussions with P.J. Statham and I. Barkshire of Oxford Instruments on spectrum simulation and with K. N. Mason of Warwick University on expert systems. It has been a real pleasure over the years to have shared ideas on quantitative microprobe analysis with so many people. References Fig. 9 A structure for system intelligence, helping the user through 1 R. Castaing, PhD Thesis, University of Paris, 1951. the overall process of problem solving. This consists of a control 2 Electron Probe Quantitation, ed.K. F. J. Heinrich and system integrated with the microprobe and backed by an expert system D. E. Newbury, Plenum Press, New York, 1991. capable of learning from experience, either as factual data or as expert 3 J. I. Goldstein, D. E. Newbury, P. Echlin, D. C. Joy, A. C. Romig, opinion. The microprobe itself lies at the centre of the system, but is C. E. Lyman, C. Fiori and E. Lifshin, Scanning Electron only a small part of the total.Microscopy and X-Ray Microanalysis, Plenum Press, New York, 1992. 4 S. J. B. Reed, Electron Probe Analysis, Cambridge University system, which includes libraries of previous samples and Press, Cambridge, 1993. results, and behind this is an expert knowledge base. 5 V. D. Scott, G. Love and S. J. B. Reed, Quantitative Electron- Commercial systems already oVer an integrated system for Probe Microanalysis, Ellis Horwood, Chichester, 1995. 6 The Electron Microprobe, Proceedings of Symposium of the setting up the analysis, carrying it out and processing the Electrochemical Society, Washington, DC, 1964, ed.T. D. results, an example being the Inca system recently launched McKinley, K. F. J. Heinrich and D. B. Wittry, Wiley, New York, by Oxford Instruments. The system guides the user through 1966. the steps necessary to carry out point analysis or obtain 7 Optique des Rayons X et Microanalyse, Proceedings 4th ICXOM, images, and also allows the user to build a project library for Paris, 1965, ed.R. Castaing, P. Deschamps and J. Philibert, samples and results and even to access an interactive CD for Hermann, Paris, 1966. 8 T. O. Ziebold and R. E. Ogilvie, in The Electron Microprobe, more general queries about the technique itself. Doubtless Proceedings of the Symposium of the Electrochemical Society, other commercial systems oVer these facilities in some degree. Washington, DC, 1964, ed. T. D. McKirley, K. F. J. Heinrich and However, an expert system oVers another dimension to the D.B. Wittry, Wiley, New York, 1966, p. 378. degree of help a user could access, employing a knowledge 9 R. Castaing and J. Descamps, J. Phys. Rad., 1955, 16, 304. base which can be built up as experience develops, and which 10 P. Karduck and W. Rehbach, Electron Probe Quantitation, can draw on the experience of others. At one level there is a Plenum Press, New York, 1991, p. 191. 11 J. Philibert, in X-Ray Optics and Microanalysis, Proceedings of 3rd need to build up a set of rules to convert compositional data ICXOM, Stanford, CT, 1962, Academic Press, New York, 1963, into a form which is more meaningful in the context of the p. 379. problem itself. An example is particle classification, such as 12 M. Green, in X-Ray Optics and Microanalysis, Proceedings of 3rd the analysis of mineral grains or of gunshot residues. The ICXOM, Stanford, CT, 1962, Academic Press, New York, 1963, latter requirement is to classify particles into categories of p. 361. origin in order to determine whether a sample matches the 13 P. Duncumb and P. K. Shields, in The Electron Microprobe, Proceedings of Symposium of the Electrochemical Society, explosive fired in a certain gun. Washington, DC, 1964, ed. T. D. McKinley, K. F. J. Heinrich and Many particles are sampled and tested for the presence and D. B. Wittry, Wiley, New York, 1966, p. 284. level of certain elements by a set of rules which can become 14 K.F. J. Heinrich, Paper presented at 2nd National Conference on highly complex—perhaps running to several hundred or thou- Electron Microprobe Analysis, Boston, Electron Probe Analysis sand. Each rule is expressible in a logical form which may be Society of America, 1967. programmed in the normal way, but ‘expert system’ techniques 15 K. F. J. Heinrich, in The Electron Microprobe, Proceedings of Symposium of the Electrochemical Society, Washington, DC, 1964, oVer a way of detaching very large rule bases from the rest of ed.T. D. McKinley, K. F. J. Heinrich and D. B. Wittry, Wiley, the code and evolving these independently. Self-consistency New York, 1966, p. 296. can be checked as new rules are added, and the system can be 16 B. L. Henke, P. Lee, T. J. Tanaka, R. L. Shimabukuro and asked to explain its reasoning, i.e., which rules are fired, as B. K. Fujikawa, At. Data Nucl. Data Tables, 1982, 27, 1. diVerent questions are asked.Thus the possibility exists of 17 G. F. Bastin and H. J. M. Heijligers, Electron Probe Quantitation, giving to the unskilled user access not only to the experience Plenum Press, New York, 1991, p. 145. 18 K. F. J. Heinrich, in X-Ray Optics and Microanalysis, Proceedings of one expert, but also to the accumulated experience of of 11th ICXOM, London, Ontario, University of W. Ontario, 1987, several—indeed to a fund of knowledge beyond the capacity p. 67. of a single human mind. 19 J. Henoc, in Quantitative Electron Probe Microanalysis, At a deeper level, an expert system is also capable of ‘fuzzy’ Proceedings of Seminar at National Bureau of Standards, logic, in which the conditions are stated in terms of probabili- Gaithersburg, MD, 1967, NBS Special Publication, NBS, ties rather than logical certainties. In the microprobe context Gaithersburg, MD, 1968, p. 197. 20 G. D. Archard and T. Mulvey, in X-Ray Optics and Microanalysis, this gives the possibility of enlarging the concept of ‘Problem’ J.Anal. At. Spectrom., 1999, 14, 357–366 365Proceedings of 3rd ICXOM, Stanford, CT, 1962, Academic Press, 41 C. Merlet, in X-Ray Optics and Microanalysis, Proceedings of 13th ICXOM, Manchester, 1992, Institute of Physics Conference Series New York, 1963, p. 393. 21 D. M. Poole and P. M. Thomas, in The Electron Microprobe, No. 130, Institute of Physics, Bristol, 1992, p. 123. 42 J.-L. Pouchou and F. Pichoir, Electron Probe Quantitation, Proceedings of Symposium of the Electrochemical Society, Washington, DC, 1964, ed. T.D. McKinley, K. F. J. Heinrich and Plenum Press, New York, 1991, p. 31. 43 J.-L. Pouchou and F. Pichoir, Scanning, 1990, 12, 212. D. B. Wittry, Wiley, New York, 1966, p. 269. 22 H. E. Bishop, in Optique des Rayons X et Microanalyse, 44 N. Amman and P. Karduck, in Proceedings of 12th Intternational Congress for Electron Microscopy, San Francisco Press, San Proceedings of 4th ICXOM, Paris, 1965, Hermann, Paris, 1966, p. 153. Francisco, 1990, vol. 2, p. 214. 45 G. F. Bastin, H. J. M. Heijligers and J. M. Dijkstra, in Proceedings 23 P. Duncumb and S. J. B. Reed, in Quantitative Electron Probe Microanalysis, Proceedings of Seminar at National Bureau of of 12th International Congress for Electron Microscopy, San Francisco Press, San Francisco, 1990, vol. 2, p. 216. Standards, Gaithersburg, MD, 1967, NBS Special Publication, NBS, Gaithersburg, MD, 1968, p. 133. 46 P. Karduck, Mikrochim.Acta, 1998, 15, Suppl., 109. 47 P. Duncumb, in The Electron Microprobe, Proceedings of 24 D. A. Sewell, G. Love and V. D. Scott, J. Phys. D: Appl. Phys., 1987, 20, 1567. Symposium of the Electrochemical Society, Washington, DC, 1964, ed. T. D. McKinley, K. F. J. Heinrich and D. B. Wittry, Wiley, 25 M. Green, Proc. Phys. Soc., 1963, 82, 204. New York, 1966, p. 490. 26 H. E. Bishop, in Optique des Rayons X et Microanalyse, 48 Y. Bando, Y. Kitami, K. Kurashima, T. Tomita, T. Honda and Proceedings of 4th ICXOM, Paris, 1965, Hermann, Paris, 1966, Y. Ishida, Microbeam Anal., 1994, 3, 279. p. 112. 49 P. Duncumb, J. Microsc., 1968, 7, 581. 27 R. Shimizu, K. Murata and G. Shinoda, in Optique des Rayons X 50 J. Philibert and R. Tixier, Br. J. Appl. Phys., 1968, 1, 685. et Microanalyse, Proceedings of 4th ICXOM, Paris, 1965, 51 W. A. P. Nicholson, Mikrochim. Acta, 1994, 114/115, Suppl., 109. Hermann, Paris, 1966, p. 127. 52 C. E. Lyman, J. I. Goldstein, D. B. Williams, D. W. Ackland, 28 P. Duncumb and D. A. Melford, in Optique des Rayons X et S. Von Harrach, A. W. Nicholls and P. J. Statham, J. Microsc., Microanalyse, Proceedings of 4th ICXOM, Paris, 1965, Hermann, 1994, 176, 85. Paris, 1966, p. 240. 53 G. CliV and G. W. Lorimer, in Proceedings of EMSA/MAS 29 L. Reimer and E. R. Krefting, NBS Spec. Publ., 1974, No. 460, 45. Meeting, Boston, San Francisco Press, San Francisco, 1992, 30 P. Karduck and W. Rehbach, Electron Probe Quantitation, p. 1464. Plenum Press, New York, p. 191. 54 S. J. B. Reed, X-Ray Spectrometry in Electron Beam Instruments, 31 L. Reimer, Mikrochim. Acta, 1996, 13, Suppl., 1. Plenum Press, New York, 1995, p. 221. 32 D. C. Joy and S. Luo, Scanning, 1989, 11, 176. 55 C. R. Swyt and C. E. Fiori, in Proceedings of 12th International 33 S. Luo, D. C. Joy, J. R. Dunlap and X. Wang, in Proceedings of Congress for Electron Microscopy, San Francisco Press, San EMSA/MAS Meeting, Boston, San Francisco Press, 1992, p. 1642. Francisco, 1990, vol. 2, p. 108. 34 D. C. Joy, Monte Carlo Modeling for Electron Microscopy and 56 J.-L. Pouchou, Mikrochim. Acta, 1994, 114/115, Suppl., 33. Microanalysis, Oxford University Press, Oxford, 1995. 57 J. L. La�ba�r, Electron Probe Quantitation, Plenum Press, New 35 L. Reimer, M. Kassens and L. Wiese, Mikrochim. Acta, 1996, 13, York, 1991, p. 219. Suppl., 485. 58 D. E. Newbury, C. R. Swyt and R. L. Myklebust, Anal. Chem., 36 K. F. J. Heinrich, X-Ray Spectrometry in Electron Beam 1995, 67, 1866. Instruments, Plenum Press, New York, 1995, p. 305. 59 S. J. B. Reed, Mikrochim. Acta, 1998, 15, Suppl., 29. 37 V. D. Scott and G. Love, Electron Probe Quantitation, Plenum 60 J. A. Small, S. D. Leigh, D. E. Newbury and R. L. Myklebust, Press, New York, 1991, p. 19. J. Appl. Phys., 1987, 61, 459. 38 D. B. Wittry, J. Appl. Phys., 1958, 29, 1543. 61 J. Trincavelli, G. Castellano and J. A. Riveros, X-Ray Spectrom., 39 R. H. Packwood and J. D. Brown, X-Ray Spectrom., 1981, 10, 1998, 27, 81. 138. 40 G. F. Bastin, J. M. Dijkstra and H. J. M. Heijligers, X-Ray Spectrom., 1998, 27, 3. Paper 8/07315E 366 J. Anal. At. Spectrom., 1999, 14, 357&ndash

ISSN:0267-9477    

DOI:10.1039/a807315e

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

EPMA using low-energy (0.1–1.0 keV) X-rays—an historical perspective† Invited Lecture V. D. Scott* and G. Love Department of Materials Science and Engineering, University of Bath, Bath, UK BA2 7AY Received 28th August 1998, Accepted 19th October 1998 Studies with the electron-probe microanalyser using low-energy (<1 keV) X-rays are reviewed with particular reference to analysis of the ultra-light elements (atomic number Z<11). The paper begins by describing the historical development of methods for measuring low-energy X-ray spectra and goes on to show that minimum detection levels achieved for ultra-light elements are now comparable to those for heavier elements, in ideal circumstances about 100 ppm using wavelength-dispersive spectrometry and about 1000 ppm with energy-dispersive spectrometry.Regarding quantitative analysis, the correction models developed during the 1960s worked fairly well in most cases but, because they were based upon an X-ray depth distribution function of limited validity, were unsatisfactory for the ultra-light elements.Since then, advances made using empirical curve-fitting methods to derive this function have led to accuracies of a few per cent. relative, close to figures for heavier elements, any limitations being most likely associated with the accuracy of the mass absorption coeYcients used rather than the models themselves. The merit of utilising low-energy radiation for surface analysis is discussed and it is shown that, by employing a low-voltage electron probe to restrict X-ray excitation to surface layers and recording low-energy X-ray emission, surface films down to about 1 nm thickness may be investigated.Finally, examples are given to show the value of studying the shape of low-energy spectra as a means of identifying the phase of microstructural features. decade or so that a stage has been reached whereby high- Introduction quality EPMA data utilising low-energy X-rays have become Since its inception by Castaing1 more than 40 years ago, routinely available.electron-probe microanalysis (EPMA) has developed into one This paper provides an historical perspective to the subject of the most powerful techniques available for the chemical of low-energy X-ray studies in EPMA. It begins by describing analysis of microvolumes of material. The method involves the methods investigated for measuring low-energy X-rays and irradiating the specimen with a beam of high-energy electrons reviews the progress made in both wavelength-dispersive specto generate characteristic X-rays and then measuring the trometry (WDS) and energy-dispersive spectrometry (EDS).energy (or wavelength) of the emitted X-ray lines to identify The current status of the two methods is then compared. A the elements present. The concentration of the element may discussion follows of correction models proposed for providing be obtained by comparing X-ray intensities from the specimen quantitative EPMA data. Next, the advantages to be gained with those from reference standards and applying correction by using low-energy X-rays for surface analysis are considered factors, as originally proposed by Castaing1 and subsequently and the paper concludes with reference to phase identification developed by other workers in the field.The extensive use of of microvolumes via studies of the shape of low-energy EPMA in the fields of geology, chemistry, physics, electronics, X-ray spectra.engineering, biology, medicine and materials science bears testimony to its value in current scientific research. Throughout the history of EPMA, however, it has been Instrumental aspects evident that work using low-energy (<1 keV) X-rays, a region An advantage of X-ray emission analysis over other of the spectrum which contains K lines of ultra-light elements spectrochemical techniques lies in the relative simplicity of (atomic number Z<11), has made slower progress than X-ray spectra but, when analysing ultra-light elements, the investigations involving higher energy X-rays.The reasons for situation is complicated by the presence of L, M, etc., spectra this are not hard to find. First, low-energy X-rays suVer much from heavier elements. This increases the likelihood of X-ray greater absorption in the specimen and detection system than line overlap in the low-energy region, giving troublesome higher energy radiation; second, K lines from the ultra-light interferences when, say, carbon, nitrogen and oxygen are being elements may be overlapped by the large number of L, M, studied in the presence of transition metals; for example, etc., lines from heavier elements and thus create problems of vanadium La is separated by about 13 eV from oxygen Ka, identification; third, the shape of low-energy X-ray peaks is and scandium La and titanium Ll by only 2 eV from nitrogen aVected in a complex way by the chemical state of the emitting Ka.Further diYculties arise because of the eVect of chemical element. Indeed, it may be argued that it is only in the past bonds on the shape of X-ray peaks and the heavy absorption suVered by low-energy X-rays. Thus it follows that, to achieve †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998. good instrumental performance, developments must target J.Anal. At. Spectrom., 1999, 14, 367–376 367Fig. 1 Profiles of diVraction gratings. high-energy resolution in addition to high X-ray collection eYciency. In early EPMA studies, crystal spectrometry (referred to as Fig. 3 Deconvolution of pulse-height distribution curve into WDS) was the common way of analysing the X-ray spectra. components A, B and C. After Dolby.6 The instrumentation involved a Bragg-type arrangement, with the X-ray source (point of impact of the electron probe on the specimen), analysing crystal and X-ray detector positioned Dolby6 in the use of a pulse-height analysis technique for on a focusing circle such that X-rays of wavelength l were analysing low-energy X-ray spectra in the electron microprobe.diVracted at angle h by planes in the crystal according to nl= His method dispensed with the crystal for dispersing the X-rays 2d sinh, where d is the interplanar spacing and n is the order and relied upon the energy resolution of the X-ray detector— of diVraction.Initially, WDS of the ultra-light elements was a thin-window gas-filled proportional counter with a modest impeded by a lack of suitable diVracting crystals with the energy resolution of about 200 eV (full width at half maximum) appropriate d-spacing, >2 nm for the very low-energy X-rays. at the carbon Ka energy of 277 eV. The output signal of the However, an alternative diVracting element, the ruled grating counter was a distribution of pulse heights representative of as used for optical work, was being used in conjunction with the energies of the collected X-rays, with each peak broadened an X-ray probe for low-energy X-ray studies (see the review because of the statistical spread of the photon conversion by Fabian et al.2) and this was shown, by Nicholson and process in the counter.Hence X-ray peaks relating to neigh- Hasler,3 amongst others, to be viable in EPMA work.Several bouring elements in the Periodic Table overlapped, as shown grating profiles were evaluated for spectral dispersion and schematically in Fig. 3, curve T, a pulse-height distribution diVracted intensities, the grating being set up at a small angle curve which comprises three X-ray peaks. Deconvolution of of incidence to foreshorten the eVective grating spacing and the pulse-height distribution was achieved by electronic take advantage of strong specular reflection at low angles.methods and the signal relating to each component (element Holliday4 commented on the good performance of the blaze- A, B or C) obtained by positioning an electronic window at type grating (see Fig. 1) and some of his data are shown in the appropriate energy channel. Fig. 4, reproduced from Fig. 2 for carbon Ka spectra taken from a range of carbides. Dolby’s work, shows line traces across a specimen of beryllium Unfortunately, although providing high-energy resolution, the upon which had been mounted particles of carbon and lithium potential of the diVraction grating has never been fully realised hydroxide, a clear demonstration that the technique could in EPMA studies because of a low diVracted intensity and the separate eVectively the elements oxygen, carbon and beryllium. diYculty of integrating a grating spectrometer with the electron An advantage of the method, compared with the diVraction microprobe.Nonetheless, some interesting results were grating, was that its high collection eYciency meant it could obtained using dedicated instruments and Kozlenkov et al.,5 readily be used in conjunction with the scanning electron using a windowless electron multiplier and a crossed grating microprobe which had been recently introduced by Cosslett configuration, showed high element detection sensitivities were and Duncumb.7 An example taken with a four-channel apparpossible, quoting 0.04 wt.% for beryllium in copper and atus fitted to a commercial electron microprobe8 is shown in 0.02 wt.% for carbon in steel.Fig. 5, scanning pictures of beryllium carbides in beryllium. Concurrently, important developments were being made by Regarding the detection sensitivity of the method, this very much depended upon the type of specimen and values of about 1 wt.% of carbon in iron and about 2 wt.% of carbon in uranium have been quoted. Further progress with the Dolby Fig. 2 Carbon Ka spectra from carbides and graphite, taken using Fig. 4 Oxygen, carbon and beryllium concentrations across a beryllium specimen, mounted on which are particles of carbon and blazed diVraction grating spectrometer; normalised peak heights. After Holliday.4 lithium hydroxide, taken using pulse-height analysis. After Dolby.6 368 J. Anal. At. Spectrom., 1999, 14, 367–376Fig. 5 Scanning pictures; (a) electron, (b) carbon X-ray, (c) beryllium X-ray, showing beryllium carbides in beryllium, taken using pulse-height analysis method.After Ranzetta and Scott.8 method was limited by the relatively poor stability of the gas- filled proportional counter, although it must be said that the system produced beryllium X-ray pictures which were not only remarkable at the time, but also compare favourably with what can be achieved today. Indeed, the Dolby approach may be regarded as the forerunner of modern EDS, a method which has since come to the fore with the introduction of the solid-state detector.Meanwhile, the search for analysing crystals with a suitably large d-spacing was under way and among the materials examined were natural minerals such as penninite and clinochlore with d-spacings of about 1.43 nm.9 Acid phthalates were also being used on a routine basis, and still are today, but it was the advent of fatty acid pseudo-crystals, with d-spacings from 3 nm ( laurate) to 7 nm (melissate) and covering the X-ray wavelength range from about 1 to 6.5 nm, which made Fig. 7 Concentration versus depth profiles through surface layers on low-energy X-ray studies in the electron microprobe a practical carbo-nitrided steel, taken using a lead stearate analyser. After Whittle proposition. These crystals were made by building up layers and Scott.14 of long-chain molecules from a fatty acid solution using a Langmuir–Blodgett technique. A suitably heavy metal ion, voltages to improve the accuracy of quantitative data (reduced such as lead, was incorporated into the fatty acid to produce absorption factor) and to enhance spatial resolution.Fig. 7, planes of heavily scattering atoms which were separated by taken using a lead stearate crystal, shows element concen- two molecular chains of organic material, the planes in the tration profiles through the surface region of a carbo-nitrided pseudo-crystal then acting as the diVracting planes. Ong10 steel.14 The data clearly reveal an outer zone rich in oxygen, described the utilisation of stearate–decanoate11 crystals for a central layer of nitride and an inner zone rich in carbon— the study of the elements fluorine through boron and Fig. 6, crucial information for elucidating surface reactions and a typical example of work carried out at the time, shows diVusion processes during nitro-carburising. From studies such scanning pictures of carbonaceous inclusions in calcium– as these, minimum detection levels approaching 0.1 wt.% have magnesium carbonate.12 Andersen13 identified aluminium been reported for the ultra-light elements (Table 1).oxide inclusions in a nickel alloy using a stearate analyser and A characteristic of crystal analysers, especially stearates, is also anticipated the advantages of working with low probe that they produce higher order diVractions at spectrometer angles related to multiples of the X-ray wavelength, for example, at positions relating to 2l, 3l, etc. The intensity of higher order diVractions, which depends upon the structure factor for the particular crystal lattice, may pose problems of spectral interference, in which case they require suppression by using pulse-height analysis methods, particularly when searching for trace elements.However, the fact that higher order diVractions will occur at larger Bragg angles than the first-order peak means that the associated increase in spectral dispersion and resolution may be used to advantage for separating closely-spaced X-ray lines.The point is well illustrated in Fig. 8, taken from oxidised vanadium using a lead stearate crystal, which shows that oxygen Ka and vanadium La lines which are unresolved in first order are well separated in the third-order spectra. An alternative diVracting element which has been investigated for WDS work is the multilayer device (MLD), as originally proposed by Dinklage.15 It is made by depositing alternate layers of selected high and low atomic number elements, of prescribed thickness, on a suitable substrate so as to ‘tailor’ the d-spacing for a selected range of X-ray wave- Fig. 6 Scanning pictures revealing carbonaceous inclusions in Ca–Mg lengths. The MLD, whilst having a poorer spectral resolution carbonate, taken using lead stearate–decanoate analyser. After Baird and Zenger.12 than the fatty acid crystal, is less prone to high-order diVrac- J. Anal. At. Spectrom., 1999, 14, 367–376 369Table 1 Minimum detection levels (wt.%), with probe voltage 10 kV Radiation Sample TAP Pb laurate Pb stearate MLD 1a MLD 2a EDSa C K SiC — 0.04 0.025 0.03 0.009 0.2 N K Si3N4 — 0.07 0.075 0.03 0.095 0.5b O K Al2O3 0.019 0.015 0.017 0.005 0.008 0.12 aMLD 1 is a tungsten–silicon multilayer with 2d=6.09 nm; MLD 2 is a nickel–carbon multilayer with 2d=9.35 nm; EDS data refer to Si(Li) detector.bN K sensitivity is strongly dependent upon absorption in ED detector window. lution and stability to analyse X-radiations which include low energies. As with the Dolby apparatus, the detector received a fraction of the whole X-ray emission spectrum although, with this new development, the pulse-height distribution was processed using a multi-channel analyser, as used for nuclear physics studies.In order to minimise electronic noise, the Si(Li) detector was maintained at liquid nitrogen temperature and, for its protection, was isolated from the vacuum system of the electron microprobe by an approximately 10 mm thick beryllium window.Since in low-energy X-ray work the window absorbed the X-rays of interest, the detector was operated in the ‘windowless’ mode17,18 but then problems were revealed. One arose from excitation of the detector by unwanted electrons and very low-energy radiation (ultraviolet, optical and infrared ) and here the solution involved the introduction of an electron trap to deflect electrons and a thin window of aluminised plastic to absorb the unwanted electromagnetic radiation.Another arose from condensation of water vapour and oil (from the vacuum pumping system of the electron microprobe) on the cooled surface of the detector and this required improved vacuum systems and electronic conditioning circuits to warm the detector (with bias switched oV ) periodically and remove any ice. Additional developments of the Fig. 8 O Ka and V La spectra from oxidised vanadium taken using a lead stearate analyser; (a) first-order and (b) third-order diVractions.EDS detector and pulse-processing electronics were then found necessary in order to deal with (a) the presence of a ‘spur’ on the low-energy side of a peak relating to incomplete charge tions and can give higher intensity peaks. This aVords an collection, (b) the large electronic noise contribution and (c) increase in element detection sensitivity and Fig. 9(a), taken overlap of low-energy X-ray lines.19 An example showing the from a steel containing about 1 wt.% of carbon, shows the eVectiveness of the modifications carried out is illustrated in more readily distinguishable carbon peak obtained using a Fig. 10, a comparison of carbon spectra from silicon carbide tungsten–silicon MLD (d=3.05 nm) compared with that from obtained using early (1980s) apparatus and modern equipment. a lead stearate crystal (d=5 nm). Comparison of nitrogen The increased separation of the noise peak from the carbon spectra from a steel containing 0.4 wt.% of nitrogen, peak and the reduction in background level are clearly visible, [Fig. 9(b)], confirms the superiority of the MLD for these both improvements contributing to significantly higher detec- ultra-light elements.16 Some data concerning energy resolution tion sensitivities. As mentioned earlier, the detection sensitivity and minimum detection levels of two types of MLD are depends very much upon the type of specimen being analysed included in Tables 1 and 2.and, in favourable cases, figures of about 0.15 wt.% for, say, The prospects of using the EDS method in EPMA work carbon in iron have been achieved using modern EDS improved dramatically when, in the early 1970s, the lithiumequipment. 19 drifted silicon detector became available with suYcient reso- Recently, the performance of the high-purity germanium (HPGe) detector has improved suYciently for it to be used for EDS analysis of low-energy X-rays.20 The advantage of the HPGe detector is that it requires less energy to produce an electron–hole pair than silicon (2.96 compared with 3.8 eV), which means that it will have better energy resolution, all other things being equal.New problems had, however, to be overcome relating to its infrared sensitivity and the need for a thinner dead layer to compensate for higher X-ray absorption, but the improved dispersion available is evinced in Fig. 11, a comparison of Ti L spectra obtained using an Si(Li) detector (62 eV resolution at Ti L) and a modern HPGe detector (51.5 eV resolution).To conclude this section, some results are given in Table 2 for the energy resolution of modern WDS and EDS systems. The data indicate that the Si(Li) detector, with a resolution of 60–70 eV, would be incapable of distinguishing, satisfac- Fig. 9 Comparison of spectra taken using multilayer device (MLD) torily, ultra-light element spectra from the proliferation of L, and lead stearate (PBS) crystal: (a) carbon Ka from steel containing M, etc., lines from heavier elements since the separations 1 wt.% of carbon; (b) nitrogen Ka from steel containing 0.4 wt.% of nitrogen.After Love and Scott.16 involved are often 10 eV or less (the HPGe detector with its 370 J. Anal. At. Spectrom., 1999, 14, 367–376Table 2 Energy resolution of X-ray analysers (eV ) Radiation Energy/eV TAP Pb laurate Pb stearate MLD 1a MLD 2a EDSa C K 277 — 4.8 5.4 8.8 10.7 68 N K 392 — 7.6 9.6 11.3 15.5 67 O K 525 2.8 2.8 2.4 13.6 20.4 64 aSee Table 1.Fig. 10 ED spectra from silicon carbide using (a) early system and (b) later system, showing improved performance with modern pulse processing. After Bloomfield et al.19 Fig. 11 ED spectra from titanium using (a) an Si(Li) detector and (b) an HPGe detector, showing separation of Ti L lines. Fig. 12 Comparison of (a)WDand (b) ED spectra from oxidised iron. resolution of about 52 eV oVers little advantage in this respect).spectra, TAP may be preferred for high-resolution work but Hence it is no surprise to find that a high proportion of EPMA MLD for better detection sensitivity. work with low-energy X-rays has been carried out by WDS, using synthetic ‘crystals’ such as thallium acid phthalate (TAP, Quantitative analysis d#1.3 nm), lead stearate (d=5 nm) or multilayer devices (MLD) optimised to study a particular X-ray line. WDS In quantitative EPMA, corrections are needed to convert the systems give, in the best cases, energy resolutions of about measured X-ray intensity into a concentration of the analyte 5 eV or less.element. Factors to be considered in the computation may be Regarding element detection sensitivity, this is determined explained by reference to Fig. 13(a), a diagram showing the by the size of the X-ray peak compared with the background distribution of X-rays generated by a beam of electrons, with intensity and will, therefore, depend upon the eYciency of the emitted X-rays being collected at an angle y to the X-ray detection in addition to the energy resolution of the detector.For a peak to be visible, the X-ray count must exceed three standard deviations of the background intensity, the minimum detection level being given by (3/I0) ÓIB/t, where t is the counting time, I0 is the peak count for the pure element and IB is the background intensity. Hence it follows that the detection level depends also upon the type of matrix (X-ray absorption characteristics, for example), the extent of peak overlaps and the probe excitation conditions, which implies that there is no unique value for an element.As a guide, some comparative data relating to specific systems are given in Table 1 which show the superior sensitivity of WDS over EDS, a point which is clearly demonstrated by the WD and ED Fig. 13 Distribution of generated X-rays with depth: (a) section spectra from oxidised iron shown in Fig. 12(a) and (b). It is through specimen surface, where y is X-ray take-oV angle; (b) schematic representation of w(rz) versus rz curve. also apparent from Tables 1 and 2 that, in studies on oxygen J. Anal. At. Spectrom., 1999, 14, 367–376 371specimen surface. Based upon the idea that the generated intensity of characteristic X-rays was related to mass concentration, Castaing1 proposed that the intensity ratio of X-rays generated in the specimen and in a reference standard, obtained under identical analysis conditions, would give a measure of chemical composition.To allow for attenuation of the X-rays on their way out of the specimen, an absorption correction had to be applied which also took into account the fact that X-rays were generated at diVerent depths in the specimen. The variation of X-ray intensity, w(rz), with depth in the specimen is depicted in the w(rz) versus rz curve shown in Fig. 13(b), where rz is mass depth. Castaing dealt with absorption by introducing a factor, f (x), which he derived from w(rz) distributions obtained by experimental tracer measurements on a range of specimen materials.21 Corrections to allow for secondary X-ray generation (fluorescence) were also made and Castaing’s equation for characteristic fluorescence, although making the drastic assumption of a point X-ray source at the specimen surface, worked reasonably well and has formed the basis for subsequent treatments.22 Fortunately, in ultra-light element work, the size of both characteristic and continuous Fig. 14 Representations of w(rz) curves: (a) Philibert model with fluorescence corrections is small and may generally be ignored. experimental w(rz) curve (broken line); (b) modified Gaussian (Packwood and Brown; Bastin et al.); (c) parabolic (Pouchou and These early f (x) data were, however, graphical and Pichoir); (d) quadrilateral (Love and Scott). restrictive in their application and a more versatile absorption correction was needed.This was provided by Philibert,23 who derived an analytical expression for f (x) based upon the curve [Fig. 14(a)]. The figure shows that the maximum value of the function occurs too near the specimen surface, the tail physical interaction processes involved in X-ray generation. His use of a generalised mathematical description of the w(rz) of the distribution extends too far into the specimen and the value of w(rz) at the surface is not zero as was assumed.curve was made possible by the knowledge that, although w(rz) curves are a function of specimen and analysis con- Although attempts were made to improve the model, they met with limited success. ditions, their shapes are similar [this is the reason why much work has been devoted over the years to deriving an expression An alternative method of developing an expression for the X-ray depth distribution was then investigated which was which accurately models the w(rz) curve].Parameters were incorporated into his correction equation to deal with specimen based entirely on the utilisation of experimental w(rz) data rather than calculations using electron and X-ray physics. The characteristics such as atomic number, atomic mass and mass absorption coeYcient and with the conditions for analysis approach was made possible by the advent of computers which could, with the help of Monte Carlo methods, simulate large (probe voltage, X-ray take-oV angle, etc.).With the increasing availability of instrumentation for numbers of electron trajectories. In this way, w(rz) data were generated covering a wide range of experimental conditions. low-energy X-ray studies and the expanding range of investigations on specimens with elements widely separated in atomic The results were then used to formulate mathematical expressions which could be used to produce w(rz) curves for number, systems which often include the ultra-light elements,8 it was discovered that a further correction was needed.This, any set of conditions. Shape parameters were included in the equations to take account of specimen atomic number, X-ray the so-called ‘atomic number correction’,24 was required to take account of diVerences in electron penetration and retard- line energy and probe voltage and the emergent X-ray intensity was derived, as before, by applying absorption and ation behaviour in diVerent target materials.Several versions were proposed25–27 which treated electron behaviour in terms fluorescence factors. Three of the most popular w(rz) shape functions currently of stopping-power and backscatter factors and which were found to deal satisfactorily with most situations. Philibert used in correction programs are illustrated in Fig. 14(b), (c) and (d). These are (i) a surface-centred Gaussian modified at coined the term ‘ZAF’ to describe the application of the three corrections, i.e., atomic number (Z), absorption (A) and the surface by a transient function;31,32 essentially, the input parameters are the amplitude of the Gaussian, its half-width, fluorescence (F), an approach sometimes referred to as the ‘matrix’ method.the decay rate of the transient and the area under the modified w(rz) curve; (ii) a parabolic model33 constructed from a pair Over the years, the conventional ZAF method has been applied to a wide range of materials and, on average, has of intersecting parabolae; the shape function is defined by the position of the X-ray intensity maximum, the X-ray range, the given corrected results with root-mean-square (rms) errors of about 5% or less.It was apparent, however, that it still surface X-ray intensity and the area under the w(rz) curve; (iii) a quadrilateral model34 based upon two intersecting performed poorly in ultra-light element work, with rms errors of 10% or more in the corrected data.28 This was found to be straight lines and the axes, the shape parameters being the position and height of the maximum, the X-ray range and the the case even when peak areas rather than peak heights were used to derive X-ray intensities29 so as to allow for the eVect surface X-ray intensity.An atomic number correction factor was, initially, included implicitly in the Gaussian and parabolic of chemical bonds on X-ray peak shape. [The principle of converting peak intensities into the more relevant peak areas models but later it was expressed explicitly, in line with the quadrilateral model.As always, the fluorescence factor is was developed further by Bastin and Heijligers,30 who introduced ‘area-to-peak’ factors into their w(rz) correction rou- calculated separately. Thus, in retrospect, it could be said that the Castaing approach to quantitative analysis, that is, ratioing tine—see below.] The main weakness of the conventional ZAF method was to be found with the Philibert absorption correc- X-ray intensities using standards and applying a ZAF-type correction procedure, has remained essentially unchanged tion, that is, the X-ray depth distribution function implicit in the model, the reasons for its inadequacy being readily appar- today, although some ‘standardless’ methods are available for particular applications.35 ent when the function is compared with an experimental w(rz) 372 J.Anal. At. Spectrom., 1999, 14, 367–376Table 3 Comparative performance of correction models (% rms errors) Elements Model Oxides Nitrides Carbides Borides with Z>11 Gaussian 2.5 4.0 3.7 <6.0 2.0 Parabolic — — — 4.0 2.0 Quadrilateral 3.4 5.1 3.8 5.3 2.0 Data obtained when evaluating the three curve-fitting models for ultra-light element analysis are given in Table 3 and have been extracted from the authors’ own assessments.There has always been a question mark about the true value Fig. 16 Normal incidence probe: X-ray intensity ratio from oxidised of mass absorption coeYcients for ultra-light elements, and and unoxidised Fe–Cr alloy plotted against probe voltage (full lines) and X-ray intensity ratio from Cr2O3 standard and unoxidised Fe–Cr here the values provided by Bastin and Heijligers36–39 have alloy (broken line).The data reveal a Cr2O3 film on oxidised alloy. been used. It may be seen that the methods all give excellent After Cox et al.41 results, with an rms error of 5% or less, although the results would be substantially diVerent if alternative absorption coeYcients were employed.Included in Table 3 are perform- of electrons; for example, decreasing the probe voltage from ance figures for elements with Z>11, showing that accurate 15 to 5 kV reduces the X-ray range in a specimen of aluminium quantitative analysis for all elements with Z4 is now a oxide from about 1.7 to about 0.25 mm. Operating under these practicable proposition. conditions means that only X-rays with low critical excitation energy are generated, that is, K radiation from ultra-light elements or L and M radiation from some more heavier Depth resolution and surface films elements.The advantages to be gained in using L radiation The history of EPMA is full of reports which describe the are clearly illustrated in Fig. 16, showing plots of iron and analysis of coatings and surface layers, clear recognition of chromium La X-ray intensities against probe voltage from a the fact that a technique which employs a fine probe with a specimen of oxidised iron–chromium alloy.41 The data are restricted power of penetration (about 1 mm) can be used to expressed as ratios of oxidised to unoxidised alloy and it may advantage for characterising surface regions of a specimen.In be seen that at about 4 kV, the Fe La ratio is negligible, the much of the earlier work, the specimen was prepared for Cr La ratio has begun to increase and the Cr La emission analysis by sectioning through the surface regions, as was from the specimen matches that from a Cr2O3 standard.When demonstrated with the sample of carbo-nitrided steel illustrated taken together, the results indicate that the surface of the in Fig. 7. In this case the section was perpendicular to the oxidised alloy must be enriched with a film of Cr2O3 rather surface but taper sections may be employed as a means of than of mixed chromium–iron oxide. improving spatial resolution. Depth profiling may be carried There have been many publications which describe surface out by grinding a dimple on the surface using a steel ball studies with the electron beam at normal incidence and an covered with abrasive paste so as to expose, progressively, the early but simple method for estimating the thickness of surface underlying structure for examination, an approach found oxide films was developed for monitoring the oxidation of useful for the analysis of multilayer structures.40 In studies aluminium in air.The approach was based upon a correction such as these, spatial resolution is determined mainly by the model42 which assumed that, for thin surface layers, the X-ray X-ray volume as in bulk analysis, but spatial resolution and intensity could be considered constant over the entire X-ray detection sensitivity may be enhanced significantly if the range. Measurements were made of the oxygen Ka emission surface layer can be detached from the substrate prior to from the oxidised aluminium specimen and an aluminium examination.An example of the improvement achieved when oxide standard, and the oxide thickness was derived from an substrate interference is removed is illustrated in Fig. 15, equation which included the absorption factor, f (x), expressed obtained from an oxide film chemically stripped from oxidised as a function of X-ray range.43 Fig. 17 shows the plot of oxide steel. Here the scanning pictures show that fairly extensive mass thickness against time obtained when using a probe oxidation of metal grain boundaries has taken place while the voltage of 10 kV; the larger experimental variation observed remainder of the metal surface is still supporting a thin at greater thickness relates to the diVerent oxide growth rates (<50 nm) protective film of oxide.on diVerent grains of the aluminium substrate. It has been An alternative and more practical method for enhancing claimed that a detection sensitivity approaching 1 nm thickness detection sensitivity in surface analysis is to use a normal of oxide may be achieved by using this approach.incidence probe at a low voltage so as to limit the penetration As improved methods have been developed for carrying out the quantitative analysis of ultra-light elements in bulk speci- Fig. 15 Film stripped from oxidised steel, showing preferential oxidation of metal grain boundaries: (a) electron and (b) oxygen Fig. 17 Growth of oxide film in air on aluminium at 500 °C, 10 kV.X-ray pictures. J. Anal. At. Spectrom., 1999, 14, 367–376 373Fig. 20 Fig. 20. Results obtained from a series of specimens consisting of SnO and SnO2 layers on tin and lead substrates. Values for oxygen Ka ratios predicted by the model of Packwood et al.47 are shown as points and the Monte Carlo calculations of Cvikevich and Pihl48 as either continuous lines ( lead substrate) or broken lines (tin substrate).After Packwood et al.47 Fig. 18 Calibration curves for oxygen Ka k ratios versus oxide film thickness calculated using PAP thin-film model. The experimental points refer to specimens with known oxide thickness. After Willich consisting of SnO and SnO2 layers on tin and lead substrates. and SchiVman.44 Values for the oxygen Ka ratio predicted by the model are shown as points and are compared with the Monte Carlo mens, so corresponding advances have been made in the calculations of Cvikevich and Pihl,48 depicted as either continuanalysis of these elements in surface studies.Thus the recent ous lines ( lead substrate) or broken lines (tin substrate). The w(rz) curve-fitting methods described above have been used two sets of data appear to be in close agreement. Bastin et al.49 for surface layer analysis after suitably modifying their shape have also introduced modifications to their Gaussian-based parameters to take account of substrate and surface character- w(rz) model32 and shown it to work for single- and multilayer istics.Willich and SchiVman44 used the thin film correction structures. program [modified w(rz)] of Pouchou and Pichoir45 to deter- In discussing the advantages to be gained by using mine oxide film thicknesses ranging from 1–100 nm for a low-energy X-rays for surface analysis, Pouchou50 drew attennumber of systems. Using the correction model, calibration tion to the fact that higher peak counts and peak-tocurves were constructed showing the variation of oxygen Ka background ratios were available and that fluorescence eVects intensity ratio (referred to oxide standards) with oxide film would be minimised when working with L radiation.Of course, thickness and their validity was checked by reference to X-ray there may be problems and Pouchou also pointed to the eVects measurements on oxide films of known thickness (Fig. 18). caused by surface contamination, the influence of conducting The satisfactory performance of this correction program was coatings, the eVect of X-ray line interference, etc.Space further demonstrated by Willich46 when comparing calculated limitations preclude, however, detailed discussion, the topics and experimental values for carbon and fluorine X-ray emis- being well covered in the literature (see, for example, refs. 51 sion from a carbon–fluorine polymer film on silicon (Fig. 19). and 52).Packwood et al.47 considered that modification of their Gaussian-based w(rz) model32 to cope with surface analysis Phase identification was not as complicated as it might seem, arguing that certain shape parameters in the model were simplified because they As mentioned earlier, low-energy X-rays are produced by were not so critically dependent upon atomic number. Fig. 20 transitions involving outer valence electrons, that is, electrons shows some results obtained from a series of specimens which have energy states not unique to a particular atom but influenced by the state of the chemical bond.The changes in the soft X-ray emission spectra which result may be summarised as (a) a shift in peak position, (b) a change in intensity distribution and (c) the appearance of additional peaks. Generally the features are more pronounced, relative to pure element spectra, in insulators than in conducting compounds and alloys. There have been many investigations of valence band spectra and the correlation of X-ray peak shapes with electronic structure (see, for example, ref. 2) and a number of studies have involved work undertaken in the electron microprobe. One of the earliest examples was given by Holliday,4 who examined, using a diVraction grating spectrometer, the change in shape of the K emission spectra of the light elements with chemical combination. Fischer53 studied titanium Ll X-rays, using a stearate crystal analyser, and showed that changes in Fig. 19 Experimental and calculated k ratios for carbon Ka and fluorine Ka from C–F–O–H polymer film on silicon. After Willich.46 the spectra could be related with chemical composition of a 374 J. Anal. At. Spectrom., 1999, 14, 367–376range of titanium oxides (Fig. 21) (in these materials it is the carried out using acid phthalate or stearate crystal analysers of fairly high-energy resolution and the advantage to be gained L spectra which are associated with valence electrons and sensitive to the nature of the chemical bond).Baun and by utilising higher order diVractions, when available, to give increased spectral dispersion is self-evident in such work. For Solomon54 also studied the L spectra to diVerentiate between FeO, Fe3O4 and Fe2O3, whilst Shiraiwa et al.55 used the example, Bleay et al.,59 in an analysis of the commercial fibre Nicalon, demonstrated how much easier it was to distinguish change in the boron K X-ray band to distinguish boroncontaining compounds. More recently, Jasienska56 analysed L between carbides and forms of carbon by using the secondorder diVractions from stearate, and reference to these findings radiation to investigate changes in stoichiometry and oxidation state of doped wustite, and Kucha57 studied sulfur valency via is a convenient way of concluding this paper.Nicalon fibre, which is based upon silicon carbide, was the sulfur Kb spectra (Fig. 22). Remond et al.58 described the modelling of line profiles of low-energy spectra using various being considered as a possible reinforcement for aluminiumbased metal matrix composites but, surprisingly, the manufac- mathematical functions and referred to the eVect of selfabsorption and probe excitation conditions on peak shape.tured composite was found to contain deleterious particles of both aluminium carbide and aluminium oxide at the fibre– The implications of spectral deconvolution on quantitative analysis were considered and the findings were used for matrix interface.EPMA then showed that the fibre was not simply silicon carbide but also contained oxygen. When next deconvoluting rare earth overlaps at low element concentrations. The role played by crystallographic and chemical the shapes of the carbon Ka, oxygen Ka and silicon Kb spectra from the fibre were compared with X-ray peaks from environment in determining peak shape was also discussed. Most X-ray peak shape studies in the EPMA have been appropriate standards, it was deduced that the oxygen was present in the form of silicon oxycarbide.From the findings it could be concluded that the oxycarbide constituent of the fibre had reacted with aluminium while the composite was being made, forming aluminium oxide and aluminium carbide. The study was followed by an evaluation of Nicalon fibre as a reinforcement for magnesium alloys, and it was shown60 that the silicon oxycarbide was destabilised by attack from magnesium, forming magnesium oxide in the process.As may be seen from such studies, the ability of the electron microprobe to elucidate chemical reactions on a microscale is clearly evident, although it must be said that the method has not received the attention it deserves, probably because of the increasing use of EDS systems which lack the necessary energy resolution. Conclusion Although the use of low-energy (<1 keV) X-rays for EPMA has progressed more slowly than work using higher energy radiation, the stage has been reached where instrumentation is available which can give ultra-light element detection sensi- Fig. 21 Titanium L emission from titanium and its oxides. After tivities of about 100 ppm in many cases. Furthermore, accurac- Fischer.53 ies of a few per cent. relative may be achieved using current correction models, figures approaching those routinely obtained in heavier element analysis. A benefit of working with low-energy X-rays in the electron microprobe is that a low probe voltage may be used for X-ray excitation and thin (about 0.01 mm) surface films may then be detected.A further advantage is that, since the shape of lowenergy X-ray spectra is influenced by the state of chemical bonding, phase identification of microstructural features may be carried out in the EPMA. Acknowledgements JEOL (UK), EPSRC and DERA are thanked for support. References 1 R. Castaing, PhD Thesis, University of Paris, 1951. 2 D. J. Fabian, L. M. Watson and C. A. W. Marshall, Rep. Prog. Phys., 1971, 34, 601. 3 J. B. Nicholson and M. F. Hasler, Adv. X-ray Anal., 1966, 9, 420. 4 J. E. Holliday, in The Electron Microprobe, ed. T. D. McKinley, K. F. J. Heinrich and D. B. Wittry, Wiley, New York, 1966, pp. 3–22. 5 A. I. Kozlenkov, Yu I. Belov, V. G. Bogdanov and A. I. Shulgin, in Proceedings of the Fifth Conference Mikrosonde, ed. A. Roder, L. Kuchler and S. Dabritz, Physics Society of the DDR, Leipzig, 1981, pp. 47–50. Fig. 22 Sulfur Kb spectra from sulfur compounds. After Kucha.57 6 R. M. Dolby, Proc. Phys. Soc., 1959, 73, 81. J. Anal. At. Spectrom., 1999, 14, 367–376 3757 V. E. Cosslett and P. Duncumb, Nature (London), 1956, 177, K. F. J. Heinrich and D. E. Newbury, Plenum Press, New York, 1991, pp. 31–75. 1172. 34 V. D. Scott and G. Love, in Electron Probe Quantitation, ed. 8 G. V. T. Ranzetta and V. D. Scott, Br. J. Appl. Phys., 1964, K. F. J. Heinrich and D.E. Newbury, Plenum Press, New York, 15, 263. 1991, pp. 19–30. 9 W. L. Baun and E. W. White, Anal. Chem., 1969, 41, 831. 35 J-L. Pouchou and F. Pichoir, in Electron Microscopy and Analysis, 10 P. S. Ong, in The Electron Microprobe, ed. T. D. McKinley, ed. G. W. Bailey, J. Bentley and J. A. Small, San Francisco Press, K. F. J. Heinrich and D. B. Wittry, Wiley, New York, 1966, San Francisco, 1992, pp. 1650–1651. pp. 43–57. 36 G. F. Bastin and H. J. M. Heijligers, Quantitative Electron Probe 11 B.L. Henke, in X-Ray Optics and Microanalysis, ed. H. H. Pattee, Microanalysis of Carbon in Binary Carbides, University of V. E. Cosslett and A. Engstrom, Academic Press, New York, 1963, Technology, Eindhoven, 1985. pp. 157–172. 37 G. F. Bastin and H. J. M. Heijligers, Quantitative Electron Probe 12 A. K. Baird and D. H. Zenger, Adv. X-Ray Anal., 1965, 9, 487. Microanalysis of Boron in Binary Borides, University of 13 C. A. Andersen, in The Electron Microprobe, ed.T. D. McKinley, Technology, Eindhoven, 1986. K. F. J. Heinrich and D. B. Wittry, Wiley, New York, 1966, 38 G. F. Bastin and H. J. M. Heijligers, Quantitative Electron Probe pp. 58–74. Microanalysis of Nitrogen, University of Technology, Eindhoven, 14 R. D. T. Whittle and V. D. Scott, Met. Technol., 1984, 11, 292. 1988. 15 J. B. Dinklage, J. Appl. Phys., 1967, 38, 3781. 39 G. F. Bastin and H. J. M. Heijligers, Quantitative Electron Probe 16 G. Love and V. D.Scott, in EMAG 87, ed. L. M. Brown, Institute Microanalysis of Oxygen, University of Technology, Eindhoven, of Physics Conference Series, No. 90, Institute of Physics, Bristol, 1989. 1987, pp. 349–352. 40 P. Willich and R. Bethke, Mikrochim. Acta, Suppl., 1996, 13, 631. 41 M. G. C. Cox, B. McEnaney and V. D. Scott, Philos. Mag., 1974, 17 N. C. Barbi, A. Sandborg, J. Russ and C. Soderquist, in IITRI/ 29, 585. SEM/74, ed. O. Johari, Illinois Technology Research Institute, 42 H.E. Bishop, J. Phys. D: Appl. Phys., 1974, 7, 2009. Illinois, 1974, pp. 151–158. 43 G. Love and V. D. Scott, J. Phys. D: Appl. Phys., 1978, 11, 1369. 18 J. C. Russ, G. C. Baerwalt and W. R. McMillan, X-ray Spectrom., 44 P. Willich and K. SchiVman, Mikrochim. Acta, Suppl., 1992, 12, 1976, 5, 212. 221. 19 D. J. Bloomfield, G. Love and V. D. Scott, X-ray Spectrom., 1985, 45 J-L. Pouchou and F. Pichoir, in X-Ray Optics and Microanalysis, 14, 139. ed. J. D. Brown and R. H. Packwood, University of Western 20 R.A. Sareen, in X-Ray Spectrometry in Electron Beam Ontario Press, London, Ontario, 1987, pp. 249–253. Instruments, ed. D. B. Williams, J. I. Goldstein and 46 P. Willich, Mikrochim. Acta, Suppl., 1992, 12, 1–18. D. E. Newbury, San Francisco Press, San Francisco, 1995, 47 R. H. Packwood, G. Remond and J. D. Brown, in X-Ray Optics pp. 33–51. and Microanalysis, ed. J. D. Brown and R. H. Packwood, 21 R. Castaing and J. Descamps, J. Phys. Radium, 1955, 16, 304.University of Western Ontario Press, London, Ontario, 1987, 22 S. J. B. Reed, Br. J. Appl. Phys., 1965, 16, 913. pp. 274–280. 23 J. Philibert, in X-Ray Optics and X-Ray Microanalysis, ed. 48 S. Cvikevich and C. Pihl, in Microbeam Analysis—1979, ed. H. H. Patee, V. E. Cosslett and A. Engstrom, Academic Press, D. E. Newbury, San Francisco Press, San Francisco, 1979, New York, 1963, pp. 379–392. pp. 39–42. 24 D. M. Poole, in Quantitative Electron Probe Microanalysis, ed. 49 G. F. Bastin, J. M. Dijkstra, H. J. M. Heijligers and D. Klepper, K. F. J. Heinrich, National Bureau of Standards Publication Mikrochim. Acta, Suppl., 1992, 12, 93. No. 298, US Department of Commerce, Washington, DC, 1968, 50 J-L. Pouchou, Mikrochim. Acta, Suppl., 1996, 13, 39. pp. 93–131. 51 A. J. Campbell and R. Gibbons, in The Electron Microprobe, ed. T. D. McKinley, K. F. J. Heinrich and D. B. Wittry, Wiley, New 25 P. Duncumb and S. J. B. Reed, in Quantitative Electron Probe York, 1966, pp. 75–82. Microanalysis, ed. K. F. J. Heinrich, National Bureau of 52 G. Love, V. D. Scott, N. M. T. Dennis and L. Laurenson, Standards Publication No. 298, US Department of Commerce, Scanning, 1981, 4, 32. Washington, DC, 1968, pp. 133–154. 53 D. W. Fischer, Adv. X-Ray Anal., 1970, 13, 159. 26 J. Philibert and R. Tixier, in Quantitative Electron Probe 54 W. L. Baun and J. S. Solomon, Vacuum, 1971, 21, 165. Microanalysis, ed. K. F. J. Heinrich, National Bureau of 55 T. Shiraiwa, N. Fujino and J. Murayama, in X-Ray Optics and Standards Publication No. 298, US Department of Commerce, Microanalysis, ed. G. Shinoda, K. Kohra and T. Ishinokawa, Washington, DC, 1968, pp. 13–33. Tokyo University Press, Tokyo, 1972, pp. 213–218. 27 G. Love, M. G. C. Cox and V. D. Scott, J. Phys. D: Appl. Phys., 56 S. Jasienska, in X-Ray Optics and Microanalysis, ed. S. Jasienska 1978, 11, 7. and L. J. Maksymowicz, Academy of Mining and Metallurgy, 28 G. Love, M. G. C. Cox and V. D. Scott, J. Phys. D: Appl. Phys., Krakow, 1989, pp. 635–640. 1975, 8, 1686. 57 H. Kucha, in X-Ray Optics and Microanalysis, ed. S. Jasienska 29 G. Love, M. G. C. Cox and V. D. Scott, J. Phys. D: Appl. Phys., and L. J. Maksymowicz, Academy of Mining and Metallurgy, 1974, 7, 2131. Krakow, 1989, pp. 652–655. 30 G. F. Bastin and H. J. M. Heijligers, X-Ray Spectrom., 1986, 58 G. Remond, C. Gilles, M. Fialin, R. Marinenko, R. Myklebust 15, 135. and D. Newbury, Mikrochim. Acta, Suppl., 1996, 13, 61. 31 R. H. Packwood and J. D. Brown, X-Ray Spectrom., 1981, 10, 59 S. M. Bleay, A. R. Chapman, G. Love and V. D. Scott, J. Mater. 138. Sci., 1992, 27, 5389. 60 V. D. Scott and G. Love, X-Ray Spectrom., 1996, 25, 286. 32 G. F. Bastin, F. J. J. van Loo and H. J. M. Heijligers, Scanning, 1986, 8, 45. 33 J-L. Pouchou and F. Pichoir, in Electron Probe Quantitation, ed. Paper 8/06766J 376 J. Anal. At. Spectrom., 1999, 14, 367–376

ISSN:0267-9477    

DOI:10.1039/a806766j

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Quantitative X-ray microanalysis of beryllium using a multilayer diVracting device† Heiko Kleykamp Forschungszentrum Karlsruhe, Institut fu�r Materialforschung I, Postfach 3640, 76021 Karlsruhe, Germany Received 28th August 1998, Accepted 13th November 1998 A commercial synthetic Mo–B4C multilayer X-ray diVracting device with 2d=22.2 nm periodicity was used to extend X-ray microanalysis to the ultra-light elements Be and B in an existing instrument. The spectrometer covers a wavelength range between 5.2 and 13 nm.The wavelength of the Be Ka emission line from elemental Be is l= 11.35 nm and the full width at half maximum is DE=7.2 eV. The optimum working voltage U0 is 10 kV for moderate X-ray mass absorption of the targets. The determination of Be in oxides is less favourable owing to the high mass absorption. U0 has to be reduced to 5 kV. The chemical shift of the Be Ka line in BeO is Dl=+0.3 nm relative to pure Be. The successful determination of Be is demonstrated for Be-containing precipitates. The maximum solubility of impurities in annealed industrial Be heats is reported. Further, the analyses of the reaction products of Be with Li4SiO4 and Li2SiO3 are presented, where Li is a further ultra-light element.mMg=127 000 cm 2 g-1, whereas the Be self-absorption is 1. Introduction reasonable, mBe=6500 cm2 g-1.3 Wavelength dispersive X-ray microanalysis is even today the The inter-planar spacing of the X-ray diVraction device was most eYcient method for the determination of light elements selected so that at least two ultra-light elements with one in the mm range with respect to accuracy, precision and crystal could be measured, e.g.the Ka lines of Be and B. The detection limit, although instruments of comparable standards wavelength of the Ka line of Be was reported as l=11.4 nm are currently being developed which are based on element and that of B as l=6.76 nm.4 The synthetic Mo–B4C multispecific signals other than X-rays.Further improvements in layer device was manufactured by OSMIC (Troy, MI, USA), the eYciency of light element determinations (boron, carbon, with amorphous B4C acting as the spacer between the Mo nitrogen and oxygen, atomic number Z=5–8) were made monolayers. The number of pairs of layers required is 150, possible by the introduction of synthetic inorganic multilayer the crystal surface is 9×21 mm and the crystal arrangement X-ray diVraction devices which are now commercially avail- is nearly in Johann geometry.A periodicity of 2d=22.2 m was able. For this purpose, a multitude of heavy element films chosen. The crystal was installed in an existing Jeol consisting of a few monolayers and low atomic number layers JRXA-50/JSM-6400 instrument used for the analysis of radioacting as the spacer are sputtered alternately on a support. active materials. The Rowland radius of the relevant spec- The distance of every two monolayers corresponds to the trometer A is R=280 mm and the spectrometer ranges between interplanar spacing d of a conventional diVraction crystal.A b=126 and 320 mm (distance between sample and crystal ), nickel–carbon multilayer device of spacing 2d=9.5 nm was hence the available wavelength range is between l=5.2 and previously installed in an existing Camebax microbeam X-ray 13 nm. analyser. The spacing was optimised for oxygen measurement. The peak (P) and background (B) counts (the latter on the The net count rate, peak-to-background ratio and detection long range and the short range side of the Be Ka line) were limit could be improved compared with an STE crystal.1 measured on elemental Be and on BeO in the diVerential mode with a beam current of 150 nA and a 30 s counting time as a function of the working voltage U0.The results are illustrated 2. Beryllium X-ray spectrometry in Fig. 1 and 2. In elemental Be and Be alloys with low Considerations on X-ray yield and eYciency led to the idea of absorption of Be Ka radiation, peak P counts and net counts the determination of beryllium using a diVraction device with P-B pass through a maximum at U0=10 kV.The calculated an even higher lattice spacing. The X-ray count rate of Be Ka detection limit cmin=2c0(2B)1/2/(P-B) is about 0.2%, where radiation should be suYcient because the X-ray generation is c0 is the concentration of Be in the analysed target. the product of the primary electron excitation Q~exp(-Z) The situation is diVerent for the determination of Be in and the fluorescence yield vK which is proportional to Z4 for compounds with a high absorption of Be Ka radiation in the low Z.The product exp(-Z)Z4 as a function of Z passes a second components. The peak P and background B counts in maximum at Z=8 (oxygen). The yields of the X-ray generation BeO increase monotonously with increasing working voltage of Be (Z=4) and B (Z=5) are comparable.2 However, X-ray U0 between 4 and 10 kV.The net counts are low and pass detection could be problematic because the mass absorption through a flat maximum at about 5 kV, which is the optimum coeYcients of Be Ka radiation become very high, e.g. oxygen working voltage. However, the calculated detection limit of absorption mO=54 000 cm2 g-1 and magnesium absorption Be in BeO is only 3%. The spectrometer data for Be and B determination are given in Table 1. The spectral resolution Dl/l is 0.066, which corresponds to †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998.the full width at half maximum of the Be Ka line of DE= J. Anal. At. Spectrom., 1999, 14, 377–380 377Table 1 X-ray microanalysis of ultra-light elements using an Mo–B4C multilayer device with lattice spacing 2d=22.2 nm Parameter Beryllium (Z=4) Boron (Z=5) Ka emission line, l/nm 11.35 6.76 Spectral resolution, Dl/l 0.066 0.093 Full line width at half maximum, 7.2 17 DE/eV Target Be BeO LaB6 Net intensity, P-B/counts s-1 921a 38b 24400a P/B 16.9a 2.1b 12.6a Detection limit (%)c 0.2 3 0.03 aU0=10 kV, Ibeam=150 nA.bU0=5 kV, Ibeam=150 nA. cCounting time 30 s. Be are more strongly bound to the nucleus and the bonding energy is increased. On the other hand, the energy of the emitted X-rays is lowered. 3. Applications Li4SiO4 is a candidate as breeding material of the blanket in a fusion reactor.The neutrons of the reactor plasma transmute preferably the 6Li isotope by the formation of tritium and helium according to the nuclear reaction 63 Li+1 0n=4 2He+3 1T Fig. 1 Peak, background and net counts of the Be Ka radiation in Be The chemical reaction in a thermodynamically closed system metal as a function of the working voltage. is represented after 10% Li burn-up, i.e., 10% of Li atoms were transmuted, according to the following equation: 10Li4SiO4A8Li4SiO4+2Li2SiO3+4He+2T2O The situation is diVerent in the presence of metallic beryllium, which acts as a neutron multiplier. As oxygen has a higher aYnity to beryllium than to hydrogen, the following reaction Fig. 2 Peak, background and net counts of the Be Ka radiation in BeO as a function of the working voltage. 7.2 eV. The wavelength of Be Ka radiation was measured in elemental Be as l=11.35 nm whereas that in BeO was l= 11.65 nm. The relative chemical shift of the Be Ka radiation in BeO is Dl=+0.30 nm relative to pure Be (Fig. 3), which corresponds to an energy shift DE=-2.8 eV. The energy shift of the Be KLL peak in the Auger spectrum emitted from Be and BeO was reported as DE=-14 eV.5 The two L electrons Fig. 3 Be Ka line in (a) Be metal and (b) BeO as a function of the of Be are moved to O in BeO in order to fill completely the L sample-to-crystal distance b in the X-ray spectrometer with Rowland radius R=280 mm. shell of O. Consequently, the remaining two K electrons of 378 J.Anal. At. Spectrom., 1999, 14, 377–380Table 3 ZAF corrected quantitative analysis of (Mg, Zr,U)Be13 and MgBe13 precipitates in an intermediate beryllium product; working voltage 15 kV, take-oV angle 35°, beryllium determination by balance Parameter Mg Zr U Be Phase Radiation Ka La Ma Ka Standard Mg Zr UO2 — I/I0 0.098 0.166 0.039 0.700a Mass fraction 0.117 0.180 0.050 0.653 Mole fractn 0.060 0.025 0.003 0.912 Theoretical 0.071 0.929 (Mg,Zr,U )Be13 I/I0 0.181 — — 0.800a Mass fraction 0.170 — — 0.830 Mole fraction 0.070 — — 0.930 Theoretical 0.071 — — 0.929 MgBe13 aArbitrary values.Table 4 Specified impurities in melt-atomised beryllium pebbles and maximum solubility in the beryllium matrix after annealing at ~800 °C for 48 h under helium Fig. 4 Light-optical microstructure (a) and element distribution images Composition Max. solubility Max. solubilitya of Be (c) O (b) and Si (d) in the Li4SiO4–Li2SiO3–Be interaction Element (mass-%) (mass-%) (mol-%) zone of a fusion blanket after irradiation in the HFR reactor to 18% Li burn-up.Be 99.4 balance 99.969 O 0.22 — — takes place, again at 10% Li burn-up: Fe 0.09 0.06 0.010 C 0.07 — — 10Li4SiO4+2Be=8Li4SiO4+2Li2SiO3+2T2+2BeO Al 0.04 0.029 0.010 Si 0.03 0.019 0.006 Fig. 4 represents a light-optical microstructure of an Mg 0.01 <0.01 <0.004 Li4SiO4–Be mixed pebble bed after irradiation in the Petten Cl 0.01 — — High Flux Reactor (HFR) up to 18% Li burn-up.The X-ray F 0.008 — — distribution images of Be, O and Si at 15 kV working voltage U 0.0023 — — and the results of the quantitative point analysis elucidate the Cr — ~0.01 ~0.002 surface oxidation of the Be spheres and the solid state reaction aNormalised to 100%. between Li2SiO3 in the Li4SiO4 spheres and BeO by the formation of Li2BeSiO4. On the other hand, Li4SiO4 and BeO are compatible.6 Fig. 4 shows further that the Be distribution Further material was manufactured by atomising purified in BeO cannot be made visible because of the high mass liquid Be in an argon stream, resulting in the formation of absorption coeYcient of Be Ka radiation in oxygen.3 However, Mg-poor Be pebbles of diVerent diameters <0.2 mm, which the determination of Be is possible by point measurements at were thereafter annealed between 870 and 690 °C for 48 h.The 5 kV. Further, Li2Be2O3 that forms by the reaction measurements performed included Be determination. Some Be+Li4SiO4ALi2Si+BeO+Li2Be2O3 precipitates were detected, e.g., Al5Fe2 and Cr–Fe–Si phases.The matrix was quantitatively analysed. The concentrations in the quaternary Li–Be–O–Si system was detected.6 The given in Table 4 are the maximum solubilities of selected results for the quantitative analysis of Li2Be2O3 are reasonable (Table 2). Beryllium pebbles are foreseen as neutron multipliers in future fusion reactor blanket concepts. Industrial intermediate Be products which had been produced by a modified Kroll process and a subsequent reduction of BeF2 using Mg were investigated by X-ray microanalysis.The following precipitates in the Be matrix of 2 mm pebbles partially annealed up to 790 °C could be detected: (Mg,Zr,U )Be13, MgBe13, Mg2 Si, Al2Mg3 and (Fe,Cr) alloys. Examples of the quantitative X-ray microanalysis of (Mg,Zr,U )Be13 and MgBe13, Be calculated by balance are given in Table 3. The ZAF correction method used illustrates the successful analysis.Table 2 ZAF corrected quantitative analysis of Li2Be2O3; working voltage 6 kV, take-oV angle 35°, Ka emission lines, lithium determination by balance Parameter O Be Li Phase Standard YAG Be — I/I0 0.798 0.039 0.163a Mass fraction 0.601 0.224 0.175 Mole fraction 0.428 0.283 0.289 Fig. 5 Be apex of the isothermal section of the Al–Be–Fe system at Theoretical 0.429 0.286 0.286 Li2Be2O3 800 °C (from Myers and Smugeresky7 obtained by ion backscattering) aArbitrary value.and solubility limit ($) in this work using X-ray microanalysis. J. Anal. At. Spectrom., 1999, 14, 377–380 379metals in Be at about 800 °C. It is interesting that Al5Fe2 Acknowledgement precipitates were observed; however, the phase AlFeBe4 that The author acknowledges the metallographic preparation of would have been expected according to the phase diagram of the samples by Mr. E. Kaiser, HVT/HZ, and the X-ray the ternary Al–Be–Fe system7 was not found.The results are microanalysis by Mr. H. D. Gottschalg, IMF I. illustrated in Fig. 5. Probably the Fe/Al ratio is too low for AlFeBe4 formation. The high annealing temperature starting References at 870 °C could also be a reason for the disappearance of AlFeBe4 because this phase decomposes above 850 °C.8 1 H. Kleykamp, Beitr. Elektronenmikroskop. Direktabb. Oberfl., 1998, 22, 133. 2 P. M. Budd and P. J. Goodhew, Light-Element Analysis in the Transmission Electron Microscope: WEDX and EELS, Oxford University Press, Oxford, 1988. 4. Conclusions 3 B. L. Henke and E. S. Ebisu, At. Data Nucl. Data Tables, 1982, 27, 1. The determination of the ultra-light element Be was made 4 J. A. Bearden, Rev. Mod. Phys., 1967, 39, 78. possible by installation of a synthetic multilayer diVraction 5 H. J. Dudek, in Angewandte Oberfla�chenanalyse, ed. M. Grasser- device in one of the spectrometers of an existing X-ray bauer, H. J. Dudek and M. F. Ebel, Springer, Berlin, 1985, microanalyser. The count rates, detection limits and spectral pp. 127 V. 6 H. Migge, in Proceedings of the 14th Symposium on Fusion resolution are reasonable. The chemical shift of the Be Ka Technology, Avignon, 1986, Pergamon Press, Oxford, 1986, p. 1209. radiation in BeO was demonstrated. Successful quantitative 7 S. M. Myers and J. E. Smugeresky, Metal Trans. A, 1976, 7, 795; analysis was demonstrated by measurements of Be-containing 1978, 9, 1798. impurity phases and the maximum solubility of their compo- 8 H. P. Rooksby, J. Nucl. Mater., 1962, 7, 205. nents in the Be matrix and also by quantifying the reaction products in the quaternary Li4SiO4–Li2SiO3–Be–BeO system. Paper 8/06747C 380 J. Anal. At. Spectrom., 1999, 14, 377–3

ISSN:0267-9477    

DOI:10.1039/a806747c

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Standardless semi-quantitative analysis with WDS-EPMA† Ce�cile Fournier,*a Claude Merlet,a Olivier Dugneb and Michel Fialinc aISTEEM, CNRS, Universite� de Montpellier II, Sciences et Techniques du Languedoc, Place E. Bataillon, 34095 Montpellier Cedex 5, France bCEA VALRHO, DCC/DTE/SIM, BP 111, 26702 Pierrelatte, France cCentre de Microanalyse Camparis, URA 736-CNRS, Universite� P. et M. Curie, 75252 Paris Cedex 5, France Received 23rd September 1998, Accepted 11th January 1999 A standardless semi-quantitative method for wavelength dispersive spectrometry (WDS) electron probe microbeam analysis was developed with a view to simplifying the analytical procedure required with this apparatus.Based on spectrum acquisition, this method is a way to obtain the sample composition in a short time with the advantages of the WDS system and maintaining reasonable accuracy. To this end, three specific algorithms were designed. The first algorithm was constructed to find automatically all the elements present in the sample and to select the appropriate X-ray line for each element.It needs to index automatically, with a good confidence level, all the peaks which are detected. The second was constructed to convert the X-ray peak area from the measured spectrum into X-ray peak intensity (normally used inWDS procedures) with a Gaussian function which is governed both by the characteristics of the crystal and by the X-ray line itself. This is essential to prevent any underestimation of the concentration due to the measurement of truncated peaks related to low sampling frequencies, and to improve the counting statistics by using all the information given by the spectrum.Finally, the sample composition was obtained with the calculation of the absolute intensity of the selected line which takes into account the spectrometer eYciency. Quantitative analysis in electron probe microanalysis (EPMA) by2 is usually based on a comparison of the intensities of a characteristic line emitted from a sample and from a standard I=CAne- V 4p edvjpjl (1+f )(1+gck) N0 A Qj (E0) of known composition.This process consists first of selecting specific lines of the elements, with the hypothesis that these elements are present in the sample, and second of measuring PRmax 0 w(rz) exp(-xrz)drz (1) the X-ray intensities emitted from the unknown and the standards. The last step requires a lot of time and sometimes poses problems if proper standards are not available.Hence where ne-=number of incident electrons, V=solid angle of standardless analysis has been increasingly studied for quanti- detection, ed=eYciency of the detector (including the yield of fication by EPMA and it is often used in energy dispersive the detector crystal itself and the transmission through the spectrometry (EDS) and in wavelength dispersive X-ray window and front contact), vj=fluorescence yield of level j, fluorescence (WDXRF) spectrometry.1 pjl=transition probability equivalent to the relative line weight, In the proposed standardless semi-quantitative analysis f=fraction of intensity due to the secondary fluorescence procedure, two main points are developed: first, standard excited by the continuum and the characteristic X-ray lines, intensities are calculated when they have not already been gck=contributions of the Coster Kronig radiationless transacquired, and second, the X-ray intensities from the sample itions between subshells to the ionization on level j (equal to are computed from a qualitative wavelength dispersive spec- zero for the K lines), x=absorption factor which is the product trum.Nevertheless, this last point needs special adjustments of the mass absorption coeYcient m/r by the cosecant of the to be certain that the measured intensity in the spectrum is take-oV angle h, N0=Avogadro’s number, A=atomic mass, really the true emitted intensity.In this way, the use of the E0=energy of the incident electrons, R=backscattering factor, semi-quantitative WDS process is similar to the simple use of Qj(E0)=ionization cross-section of level j by an electron with EDS but with all the advantages of a WDS system [resolution, energy E0, w(rz)=distribution in depth of the primary ioniz- limit of detection, analysis of light elements (B, C, N, O)]. All ation on level j and Rmax=upper limit of integration of w(rz), these improvements are possible with a microprobe configurequal to infinity in the case of the w(rz) model.ation of a minimum of three spectrometers and with three In eqn. (1) the parameters can be separated into three diVerent diVracting crystals. Under these conditions, the procategories. First, ne-, N0, A, Pj and x are parameters that are posed procedure is a simple method to compute the composieasy to obtain. They are given by the experimental conditions tion of an unknown sample with satisfactory accuracy and to imposed on the sample or extracted from tabulated data.3 prepare for high accuracy measurements by using the usual Second, w(rz) and f are parameters given by the quantification quantitative analysis procedure.models used in EPMA. Finally, Qj, vj and gck are sample parameters and (V/4p)ed is a microprobe parameter. These 1 Theory parameters are very diYcult to estimate and they are the major 1.1 Computation of the X-ray intensity source of errors in the X-ray intensity calculation.The emitted intensity from an element of concentration CA under an electron beam with an energy E0 can be expressed (a) Ionization cross-section. The ionization cross-section varies with the overvoltage U according to lnU/Um (with †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998. m<1). J. Anal. At. Spectrom., 1999, 14, 381–386 381An expression for the ionization cross-section is, according to ence samples but calculated by eqn.(1). Consequently, in this case the factors Qj, vj, gck and (V/4p)ed must be calculated. the Bethe equation,4 Then, as shown previously, approximations need to be made in order to achieve the calculated standard intensity, and those Qj=Bjzj 1 Ej2 lnU Um (2) approximations must be chosen with the aim of producing the smallest uncertainties. where Qj=ionization cross-section, U=E/Ej, rate of overvoltage, ratio between the energy of the analysed X-ray line and the electron energy, Bj=a coeYcient that is not very 2 Experimental well known,5 zj=number of electrons on the ionized level The first step in most analyses by EPMA is the acquisition of and m<1.a qualitative spectrum to have an overview of an unknown The diYculty in the computation of the ionization sample. This spectrum is used to identify the elements present cross-section is to find the coeYcient Bj.It is generally admitted in the sample. Further, the information on the element concen- that for a given level, Bj keeps the same value for all the tration is present in the height of the peaks (at the limit of the elements. counting statistics). Therefore, it is often interesting to quantify (b) Fluorescence yield. The fluorescence yield describes the those peaks to have a first and quick approach to the sample probability of generating specified characteristic X-rays follow- composition.ing the ionization of subshells (K ionization for K radiation, The proposed WDS standardless semi-quantitative L3 ionization for La radiation) either by electrons or by procedure is based on two points: first, all, the X-ray line X-rays. Hence vj represents the ratio between the number of intensity is selected and computed from a qualitative wave- X-rays emitted by transitions and the number of primary length dispersive spectrum, and second, the quantification ionizations.method does not need measurements on standard samples. The approximations of Bambyneck,6 for the K line, and For quantification, the procedure uses reference values either Hubble et al.7 for the L and M lines, are used to describe the calculated or extracted from a measured database. fluorescence yield, vj ( j=K, L, M): Consequently, this procedure simultaneously and automatically identifies and quantifies the elements present in the vj=.anZn (3) sample. All operations are done very quickly (during the The fluorescence yield vj is known to vary continuously with spectrum acquisition).the atomic number of the element. 2.1 Conditions of the acquisition of a wavelength dispersive (c) Spectrometer eYciency. Standardless analysis is not spectrum possible without knowing the spectrometer eYciency. The The semi-quantification process is based on the intensities spectrometer eYciency is defined as the fraction of the X-rays measured from a qualitative spectrum.As a result, the micro- emitted from the source that is actually registered in the probe parameters are optimized, considering two aspects: (i) proportional counter. This parameter can be divided into two the acquisition time must not be too long, so as to maintain parts: the intrinsic eYciency ed and the geometric eYciency. the advantage of time saving of the procedure, and (ii) the The intrinsic eYciency is the fraction of the X-rays actually accuracy of the spectrum has to be good enough in order to hitting the proportional counter which is registered, while the be able to estimate the concentration.It must be emphasized geometric eYciency is the eVective solid angle V subtended by that the microprobe requires a minimum configuration of the spectrometer. The intrinsic eYciency is the result of the three spectrometers with three diVerent crystals so as to have eYciency of the crystal reflection and the eYciency of the a complete qualitative spectrum.Indeed, this spectrum must proportional counter. This last point depends on the detector contain all the information relative to the elements and to window and the fraction of the incident X-rays absorbed in their concentrations in the sample. the counter itself, and the crystal eYciency depends on both After several acquisition tests on an SX100 microprobe the geometry and the properties of the crystal. from Cameca, the main parameters of the microprobe were Unfortunately, expressions or general approximations of chosen as given in Table 1.The accuracy and the speed of the the eYciency for a wavelength dispersive spectrometer system acquisition are obtained by using a voltage high enough to do not exist. Moreover, owing to the mechanical nature of the excite the X-ray lines of all the elements in the Periodic Table. spectrometer and to the fluctuations in the density of the Moreover, they are also obtained with the use of a probe counter gas, the eYciency of a WDS system may change over current high enough to make the X-ray lines visible, even if a long period and sometimes requires recalibration for quantithe elements are present in low concentration.The lowest tative analysis. However, the spectrometer eYciency can be concentration which can be detected under these conditions obtained by measuring the intensity from a pure element with can be calculated.12,13 The compromise between the acquisition the microprobe and by comparing it with the calculated count rate for the same pure element.8,9 Table 1 Optimized parameters of the microprobe in order to acquire 1.2 Quantification in EPMA rapidly an accurate full spectrum.PC1 and PC2 are pseudo-crystals, The normal procedure in quantitative analysis consists first in i.e., multilayers of carbon and tungsten, and their 2d values are 6.1 and 9.5 nm, respectively selecting specific X-ray lines of the elements and second of measuring the X-ray intensities emitted from the unknown (I ) High voltage 20 kV and from a standard (IA).By using the relative intensity (K Probe current 40–100 nA ratio=I/IA), the composition of the sample is calculated by Minimum configuration 3 spectrometers the usual correction matrix or w(rz) model.10 The use of the Crystals PET, TAP, PC2. relative intensity under the same conditions of acquisition LiF, PC1 if possible Acquisition step 60×10-5 sinh suppresses the factors Qj, vj, gck and (V/4p)ed and simplifies Counting time 100 ms the calculation of the concentration.11 per acquisition step The general scheme of WDS standardless quantification PHA: counting mode Integral follows the same process as previously, the only diVerence Total acquisition time 2 min being that the standard intensities are not measured on refer- 382 J.Anal. At. Spectrom., 1999, 14, 381–3862.3 Computation of the intensity of the analysed X-ray The semi-quantification process to quantify the sample requires the intensity value emitted by the selected line of the analysed element, i.e., the peak height on the spectrum.Hence the second point of the semi-quantitative method aims to calculate the peak height from the peak area of the qualitative spectrum. This step is necessary because the peak height in the spectrum does not correspond to the true emitted intensity owing to the finite size of the acquisition step, which degrades the true peak height.Fortunately, the peak area remains constant and can always give a good measure of the concentration. (a) Mathematical approximation of a characteristic X-ray. The calculation of the peak height from the peak area is possible if the experimental peak is fitted with a mathematical function. Usually, the observed profile of a characteristic X-ray line can be described by a pseudo-Voigt function,14,15 which is a linear combination of a Gaussian function and a Lorentzian function, both added to the background B(E): Fig. 1 Ca Ka peak acquired for an andradite sample, on a PET crystal, with the optimized parameters given in Table 1. The acquisition P(E)=kG(E)+(1-k)L(E)+B(E) (4) step, varying from 1×10-5 to 123×10-5 sinh, illustrates the variations of the peak shape. The peak height decreases when the step increases where G(E), the Gaussian function, and L(E) the Lorentzian but the peak area remains constant. function, are defined as a function of the energy E as G(E)=Imax exp C-ln 2A E-Ej HWHMB2D (5) time and the spectrum accuracy is managed by setting the value of the acquisition step.Spectra acquired with several L(E)= Imax 1+[(E-Ej)/HWHM]2 (6) acquisition steps (Fig. 1) demonstrate the eVect on the peak shape: the peak height is increasingly underestimated when where Ej, Imax and HWHM are the energy of the X-ray line, the acquisition step size is increased. It also underlines the fact the intensity and the half-width at the half maximum intensity that, to some extent, the peak area is independent of the of the peak, respectively.selected step. Actually, the counting mode with the selected The Gaussian fits the top of the peak and the Lorentzian detectors continues to accumulate the counts during all move- fits the tail of the peak. Table 2 shows the comparison between ments of the crystal. For each step, the counts accumulated these functions (Gaussian, Lorentzian and pseudo-Voigt) and between two steps are assigned in one point on the spectrum.an experimental reference peak (acquired with a step of Each point of the spectrum is an average of the count rate 5×10-5 sinh). To avoid the background eVects, the top part between two acquisitions. This explains why the top of the of the peak is the only part considered, which means that only peak is truncated while the peak area remains constant. the counts higher than 20% of the maximum intensity measured on the spectrum are taken into account. The figures in Table 2 2.2 Spectrum process were obtained by comparing the partial areas of the mathematical functions with an accurately measured peak. Two important inputs are extracted from the qualitative From an evaluation of these data, it is clear that both the spectrum, the elements present in the sample and the analysed Gaussian and the pseudo-Voigt functions agree with the X-ray line that will be used for quantification.measurements when restricted to intensities greater than 20% of the maximum intensity. The Lorentzian function alone is (a) Search for elements. The search of all the peaks present not representative of an X-ray line; its maximum is lower in the spectrum is performed according to the peak-to-backand its width is higher so it has a greater area than an ground ratio. A point from the spectrum is considered to be experimental peak. a peak if the diVerence between the point intensity and the Hence the Gaussian function is good enough only to intensity of the background is at least six times higher than compute the top part of a characteristic X-ray.16 Indeed, the the standard deviation of the background.13 The detected peaks are then automatically indexed.The main criterion of Table 2 Comparison between the mathematical functions (G+L: identification is the position of the line in the spectrum. To pseudo-Voigt with k=0.57;8 L=Lorentzian, G=Gaussian) and an make sure that there is no ambiguity, the identification of the accurate experimental peak acquired with step 5×10-5 sinh.The peak is also checked by calculating the intensity of the indexed mathematical functions are computed from characteristics of the line relative to other X-ray lines from that element detected in experimental peak, i.e., the peak height and the half-width at half the spectrum. maximum intensity. The figures in the table represent the ratio between the partial area of the mathematical functions and the same partial area of the experimental peak.The partial area is made of all the (b) Selection of the quantification X-ray. Once all the lines counts greater than 20% of the maximum intensity of the experimental from the spectrum have been indexed, and all the elements peak have been identified, the analysis program can select, from each element, one X-ray line for the semi-quantification pro- Crystal cess. That X-ray line is extracted from the list of the identified LiF PET TAP lines in the spectrum.The first criterion of selection is to choose the most intense line of the available lines. Then the G+L (k=0.57) 1.06 0.99 1.02 system makes sure that there is no overlapping peak over the L 1.19 1.06 1.12 chosen line. If it is, the system starts the procedure of selection G 0.99 0.96 0.99 again with another available line. J. Anal. At. Spectrom., 1999, 14, 381–386 383where HWHM, the half-width at the half maximum intensity, depends on the crystal characteristics and the sinh position of the peak, as described by Reed,17 erf is the error function, X is the fraction of the maximum intensity which determines the limits of the partial area and partial area is measured on the spectrum.This last parameter corresponds to the sum of the counts higher than X% of the intensity measured at the top of the peak multiplied by the acquisition step. All these parameters are well known or can be calculated and consequently Imax, the Gaussian height equivalent to the true emitted intensity, can be calculated.(c) Case of the Ka lines in the LiF spectrum. The calculations have to be adjusted when Ka lines diVracted on LiF crystals Fig. 2 Gaussian curve and its characteristics, E0, HWHM and Imax. are used for the semi-quantification process. The Ka line The white zone contains the low counts not taken into account in the partial area (the shaded zone).observed in a qualitative spectrum from an LiF crystal is actually constituted of two peaks which overlap each other, the Ka1 and Ka2 lines. The Ka2 line intensity is half of the majority of the peak counts, and thus the information on the Ka1 line intensity. In that case, the resulting peak can be fitted concentration, is present in the top part of the peak, which with two Gaussians, one being the Ka1 line approximation leads to the conclusion that the Gaussian approximation leads and the other the Ka2 line approximation (Fig. 3). to the same result as the pseudo-Voigt function, but with The maximum intensity of the peak is always in this case simplified calculations. This approximation is correct for a determined from the peak area, which is measured on the significant peak, i.e., if the peak intensity is much higher than qualitative spectrum. The area taken into account is the peak the detection limit. In addition, working only with the top area without counts smaller than X% of the peak height, and part of the peak minimizes the background eVects and the partial area is calculated according to the following equasimplifies the fitting.tion: (b) Calculation of the intensity. The aim of this calculation partial area=Imaxa1 HWHM Óp 2Óln 2 is to compute the height value from the measured partial area of a characteristic X-ray line. The Gaussian approximation and the Gaussian tools provide this result. The partial area of C3 2 +erf(Ó-lnX)+ 1 2 erf(Ó-ln 2X)D (11) a Gaussian without low counts, as shown in Fig. 2, is described by the equation The maximum intensity of the resulting peak is the sum, at the diVraction position of the Ka1 line, of the contributions partial area=Imax S p ln 2 HWHM erf(Ó-1n(X)) (7) of the Ka1 and Ka2 lines: Imax=Imaxa1G1+ 1 2 expC-(Ea1-Ea2)2 2s2 DH (12) (d) Case of the pseudo-crystals. The width of measured peaks with pseudo-crystals is very broad and, even with an acquisition step of 60×10-5 sinh), the peak height always matches the true emitted intensity (Fig. 4). Nevertheless, the intensity is computed with the Gaussian approximation. The calculation of the intensity, i.e., the peak height, from the peak area allows the counting statistics to be improved and it provides an accurate result for the intensity value. Fig. 3 Gaussian fitting in the case of the Ka lines acquired on an LiF crystal. The Ka1 and Ka2 lines are fitted with two Gaussians, their equations being GKa1(E)=Imaxa1 .-ln 2 AE-EKa1 HWHMB2 (8) and GKa2(E)=Imaxa2 .-ln 2 AE-EKa2 HWHMB2 (9) and the partial area of such a Ka line is partial area=P+2 EKa2-Xa2 Gauss(Ka2)+PEKa2-Xa2 -2 Gauss(Ka1) Fig. 4 Example of a Gaussian approximation of an O Ka line on a pseudo-crystal of 2d=9.5 nm (PC2). (10) 384 J. Anal. At. Spectrom., 1999, 14, 381–386Fig. 6 Example of calculation of the maximum intensity for an Si Ka line acquired with the optimized parameters in Table 1 on a TAP crystal.The experimental peak (step 60 curve) is the basis of the Gaussian computation (Gauss curve). A reference peak (step 5 curve) is plotted to demonstrate the agreement between the maximum Fig. 5 Relationship between lnE with lne for a wavelength dispersive intensity calculation and the true emitted intensity. system (e is the crystal eYciency). The grey points are obtained from experimental results evaluated by measurement of the background. The black full lines are a polynomial approximation of degree 4.The so-called C Ka peak on the PC2 curve is the result of the anomalous Results and discussion reflectivity.18 The crystal reflectivity increases at the position of the As an illustration, the semi-quantification method was tested absorption edge of the C Ka line because of the presence of carbon on a garnet sample (andradite). First, we considered the in the PC2 crystal. computation of the peak intensities from a qualitative spectrum acquired with the optimized parameters presented in Table 1. 2.4 EYciency curves The data for that spectrum (acquired with an acquisition step of 60×10-5 sinh) were used to calculate the true intensity (a) Methods of acquisition. Two methods are available to emitted by the elements with the Gaussian approximation. An obtain these functions. If the intensities are measured from experimental spectrum was also acquired, with a small acqui- pure elements, these elements are chosen so that one of their sition step (5×10-5 sinh).This spectrum is accurate enough X-ray lines can diVract at a specific position of the crystal. to provide a reference intensity to assess the accuracy of the The ratio of these measurements to the calculated intensities Gaussian approximation. of the same elements provides the measurement of eYciency Fig. 6 represents the part of the spectrum which corresponds at a specific position of the crystal. These measurements are to the Si Ka line diVracted on a TAP crystal.The measured made with diVerent elements so that the eYciency can be peak with step 60 significantly underestimates the maximum measured at several positions of the crystal. intensity, compared with the reference peak. The loss of Alternatively, the eYciency may be obtained from the intensity in this case is 13%. The Gaussian approximation background. For each crystal, a pure element is chosen with calculated from the spectrum acquired with step 60 provides X-ray lines non-observable in its range and are excited using a good correction for the intensity, since the loss is only 1%.a high beam current. The measurement of the emitted con- Fig. 7 deals with the Fe Ka line diVracted on an LiF crystal. tinuum as a function of the range of the wavelength which The shape of the peak acquired with step 60 conceals the has been diVracted by the crystal is corrected for the variation superposition of the Fe Ka1 and Ka2 lines.As can be seen, of the continuum with the energy. The resultant curves now give the shape of the eYciency function (Fig. 5). Points obtained by one of the two methods are then approximated in the simplest case by a polynomial function of order 3 or 4. In some cases it is necessary to take into account the absorption edge in the proportional counter or in the crystal. These functions give the eYciency relative to the microprobe where the intensities are measured and where the quantification calculations will be made.(b) Stability of the eYciency. The experimental data for the peak intensities recorded by the microprobe can be used to estimate the spectrometer eYciency. These data are not perfectly stable over a long period of time and some deviations can be observed because of the instability of the counter gas pressure and also when a crystal shift occurs. To prevent such deviations, the eYciency functions can be adjusted at each calibration of the crystal.The position of the crystals is calibrated by measuring an appropriate X-ray line on a reference sample. Fig. 7 Example of calculation of the maximum intensity for an Fe Ka At the time of each operation of crystal calibration, the line acquired with the optimized parameters in Table 1 on an LiF intensity obtained on the reference peak is compared with that crystal. The experimental peak (step 60 curve) is the basis of the recorded in the file containing the eYciency data.If those Gaussian computation (GaussKa1 and GaussKa2 curves). A reference values are diVerent, the eYciency values are corrected accord- peak (step 5 curve) is plotted to demonstrate the agreement between ing to the diVerence between the values of verification and the the resulting curve of the two Gaussians (Gauss Ka1+Ka2 curve) with the true emitted intensity. values which are recorded. J. Anal. At. Spectrom., 1999, 14, 381–386 385Table 3 Example of application of the standardless semi-quantitative method in the analysis of an andradite sample.The quality of the sample quantification is evaluated by comparing the results of the semi-quantitative process with the true emitted intensities of each element present in the sample Concentration in True measured Calculated intensities DiVerence between Detection the andradite intensities/ in semi-quantitative the measured and limit Crystal Element sample (% m/m) counts s-1 conditions/counts s-1 calculated intensity (%) (% m/m) LiF Fe Ka 21.90 6350 5780 -9.0 0.16 TAP Si Ka 16.37 40840 40850 0.02 0.048 TAP Mg Ka 0.52 2850 2670 -6.3 0.011 PET Ca Ka 23.85 27500 25620 -6.8 0.039 PC2 O Ka 37.36 6800 6800 0 0.032 4 H.Bethe, Ann.Phys. (Leipzig), 1930, 5, 325. the resulting Gaussian curve, calculated from the spectrum 5 C. J. Powell, in Microbeam Analysis, ed. J. R. Michael and acquired with step 60, correctly fits the reference peak. Also, P. Ingram, San Francisco Press, San Francisco, USA, 1990, p. 13. the loss of intensity decreases from 28 to 9% with the Gaussian 6 W. Bambynek, in X-84 Proceedings on X-Ray and Innerapproximation compared with the reference measured Shell Processes in Atoms, Molecules and Solids, Liepzig, ed. intensity. A. Neisel and T. Mu� nzer, 1984, p. 1. 7 J. H. Hubbell, P. N. Trehan, N. Singh, B. Chand, D. Mehta, M. Finally, the results in Table 3 provide a comparison between L. Garg, R. R. Garg, S. Singh and S. Puri, J. Phys. Chem. the computation of the peak intensity with the usual method Ref. Data, 1994, 23, 339. and with the standardless semi-quantitative procedure. It is 8 X. Llovet, in Proceedings of 3rd Regional Workshop EMAS ‘98, interesting that the uncertainties in the results are always Barcelona, Spain, 13–16 May 1998, ed. X. Llovet, C. Merlet and <10% for a global time of 2 min. These uncertainties are F. Salvat, Edicions de la Universitat de Barcelona, Spain, 1998, mainly governed by the acquisition time. It is obvious that the p. 163. 9 J.Wernisch, X-Ray Spectrom., 1985, 14, 109. increase in the counting time per acquisition step or the 10 C. Merlet, Mikrochim. Acta, 1994, 114/115, 363. decrease in the acquisition step improve the result significantly. 11 J. Wernisch and K. Ro� hrbacher, Mickrochim. Acta, 1998, 15, 307. 12 S. J. B. Reed, Electron Microprobe Analysis, Cambridge Acknowledgements University Press, Cambridge, 2nd edn., 1993. The authors thank P. F. Staub and J. Phan of Cameca for 13 C. Merlet and J. L. Bodinier, Chem. Geol., 1990, 83, 55. 14 T. C. Huang and G. Lim, X-Ray Spectrom., 1986, 15, 245. their technical support and their contribution to this work. 15 G. Remond, P. Couture, G. Gilles and D. Massiot, Scanning Microsc., 1989, 3, 1059. References 16 D. Ku� chler, U. Lehnert and G. Zschornack, X-Ray Spectrom., 1998, 27, 177. 1 R. A. Barrea and R. T. Mainardi, X-Ray Spectrom., 1998, 27, 111. 17 S. J. B. Reed, Mickrochim. Acta., 1998, 15, 29. 2 V. D. Scott, G. Love and S. J. B. Reed, Quantitative Electron 18 R. Marmoret and J. M. Andre, Appl. Opt., 1983, 22, 17. Probe Microanalysis, Ellis Horwood, Chichester, 2nd edn., 1995. 3 R. Weast (Ed.), CRC Handbook of Chemistry and Physics, CRC Press, Boca Raton, FL, 76th edn., 1996. Paper 8/07433J 386 J. Anal. At. Spectrom., 1999, 14, 381&nda

ISSN:0267-9477    

DOI:10.1039/a807433j

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Development of low-energy X-ray fluorescence micro-distribution analysis using a laser plasma X-ray source and multilayer optics† Remko Stuik,*a Leonid A. Shmaenok,a Henri Fledderus,a Sergei S. Andreev,b Eugeny A. Shamov,b Sergei Yu. Zuev,b Nikolay N. Salashchenkob and Fred Bijkerka aFOM—Institute for Plasma Physics Rijnhuizen, P.O. Box 1207, 3430 BE Nieuwegein, The Netherlands. E-mail: stuik@rijnh.nl bInstitute of Physics of Microstructures, Ulyanov 46, Nizhny Novgorod 603600, Russia Received 30th September 1998, Accepted 17th December 1998 A new technique is presented for low-energy X-ray fluorescence micro-distribution analysis of low-Z elements at micrometer spatial resolutions.The technique is based on the use of a laser plasma X-ray source and spherically curved multilayer optics. A large collimator is used to focus the light from the laser plasma on the sample and a Schwarzschild mirror set is used to image the fluorescent radiation on a 2D CCD unit.A first system, now under development, is designed for detection of the carbon Ka-line. The system consists of a Cr/Sc collimator of 260 mm diameter, focusing 0.7 sr of the light from the plasma on the sample, and an Fe/C Schwarzschild mirror set with 20× magnification for detection of the carbon in the sample. A resolution in the micrometer range is expected to be achievable, with a detection limit of a few per cent. Upgrading of this system is expected to result in sub-micrometer resolution and a detection limit in the ppm range.Many non-destructive techniques exist for the surface analysis A number of practical applications require a knowledge of the spatial distribution of low-Z elements. This is the case in of the composition of materials, including EPMA (electron probe micro-analysis), PIXE (proton induced X-ray emission) the semiconductor industry, where most of the elements used are within the low Z regime (Z<16). For example, migration and XRF (X-ray fluorescence) analysis.However, most of these techniques are primarily appropriate for the detection of of a doping material through the substrate material and the distribution of polluting elements1 drastically influence the fluorescence lines at energies of several keV, thereby addressing the medium- to high-Z elements. Although high resolution, properties of the electronic device. Another application is the study of catalysts for use in oil refineries.During the refining down to the micrometer range, has been obtained by scanning XRF systems using a well focused (pencil ) beam of synchro- process, certain parts of the catalyst are polluted by carbon deposition, rendering it useless. An understanding of the tron radiation, this method involves a centralized, large-scale measurement facility. position of carbon deposition will allow the construction of improved catalysts. Other applications of micro-distribution Recent advances in the development of laser plasma X-ray sources and multilayer optics have allowed a diVerent method analysis (MDA) of the low-Z elements include the investigation of structures of boron on silicon surfaces2 and studies of XRF analysis suitable for the detection of sub-keV fluorescence lines of low-Z elements.The principle of the technique on BeF2 in glasses for ultra-low-loss fiber optics.3 Partly stimulated by other practical applications of laser consists of excitation of an element with a photon energy a few eV higher than the photon energy of the fluorescence line, plasma sources, such as extreme ultraviolet lithography (EUVL), the average laser power of application-specific lasers i.e., just above the corresponding absorption edge, where the cross-section for absorption is maximal.Although the fluores- has increased by several orders of magnitude, up to the kilowatt level.4 In some large-scale research facilities, peak power densi- cence yields are generally lower for the lower fluorescence energies, compared with the fluorescence yield for lines at ties up to 1020 W cm-2 (e.g., Titania, Central Laser Facility, Rutherford Appleton Laboratory) have been reached, resulting higher energies, the cross-sections for soft X-ray photon excitation are much larger.For carbon, for example, the cross- in the generation of X-rays with energies up to several MeV. Laser plasma X-ray sources are now becoming easier to main- section for photo-absorption is 104 times higher at 300 eV than at 8 keV.By using normal incidence multilayer optics, a tain and have increased eYciency for the conversion of laser photons into soft X-rays, while the inherent pollution by the high throughput can be obtained while at the same time high magnifications (typically 10–30×) can be achieved at a reason- plasma is being reduced to levels acceptable even for demanding applications. With these advances, laser plasma sources with able resolution.The use of a back-illuminated CCD detector gives both imaging over a large field on the sample at a high high average (soft) X-ray power have come within the reach of small laboratories, both in price and in size. Also, the develop- resolution and high sensitivity enabling short exposure times. The lines of interest for this technique are predominantly the ment of optics for the soft X-ray region has been boosted, e.g., K-shell fluorescence lines of the lighter elements (Li–Mg) and by EUVL research and X-ray microscopy.Multilayer mirrors the L-shell fluorescence of heavier elements (Al–Ga). Low- can now be made to reflect up to 67–68%5,6 near the Si LII, III energy (LE) XRF is generally a surface sensitive technique, line, while reflectivities of more than 10% at the C K line have probing only the top layer of a material. In the range from been obtained,7 all for normal incidence optics. A combination Be up to O, the penetration depth varies between 50 and of these technologies permits the design of an LE-XRF system 250 nm (Table 1).with a sensitivity comparable to those of other elemental analysis techniques, with the use of a laser plasma source keeping the set-up compact and relatively inexpensive. †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998. In the pilot experiment described in this paper, the imaging J.Anal. At. Spectrom., 1999, 14, 387–390 387Fig. 1 Three simulated mirror reflectivity curves for the detection of the C Ka fluorescence line (277 eV). From left to right: monitor mirror (274 eV), analyzer mirror (277 eV) and collimator mirror Fig. 2 Experimental layout, consisting of an illumination section ( laser (292 eV). Dotted line: photo-absorption cross-section of C, neglecting plasma, foil trap and collimator) and a detection section near-edge fine structure.(Schwarzschild set and CCD camera). of the fluorescence of carbon surface structures of micrometer clusters and larger fragments (‘debris’). The techniques used resolution was set as a goal. The excitation of carbon is for suppression of migration of all debris components include eVected at 292 eV, just above the K-shell absorption edge the following: (284 eV), after which an X-ray microscope, sensitive for the 1. Fast rotation of the target disk to redirect the relatively C Ka fluorescence line (277 eV), images the distribution of slow and predominantly larger fragments, by giving a directed emission across the sample surface.A monitor mirror next initial velocity to these particles.10 In our LE-XRF system, a (274 eV) to the fluorescence line is used for background target disk with a diameter of 5 cm was used with a velocity subtraction (Fig. 1). From the distribution of intensities, the of 600 rotations per second, i.e., an edge velocity close to distribution of the carbon concentration across the surface 100 m s-1.By optimizing the disk orientation, a spatial angle can be calculated. can thus be created free of the larger, slower debris fragments. 2. A ‘foil trap’—an assembly of foils positioned near the Experimental source along radial directions in a buVer gas (Fig. 2). This eVectively eliminates atoms and small clusters of atoms.11 The The experimental set-up can be divided in two parts (Fig. 2). trapping eVect is achieved due to collisional retardation (for The first part is intended for high intensity illumination of the atoms even thermalization) and scattering of particles in the sample by focusing radiation from a laser plasma X-ray source gas with subsequent deposition of these particles on the foils.with a curved multilayer mirror collimator. The second part Nitrogen was selected as a buVer gas owing to its high serves to detect the fluorescent radiation emitted from the transparency for 300 eV photons.Given the optical pathlength illuminated area, with spatial resolution. It is noted that other in our system (about 1 m), a pressure of several millibars can geometries with larger spatial separation of the two parts are be allowed, suYcient for full thermalization of fast debris possible, allowing further suppression of stray light and particle atoms in a gas layer of a few centimeters. The foil trap consists contamination from the laser plasma source.of 40 stainless steel foils of 40×200 mm, with a thickness of 120 mm, kept in a fan-like geometry by rigid holders. With a Illumination system foil separation of 1 mm at the plasma facing side, the trap system spans a solid angle of 1 sr, at a double-pass (to and The laser plasma source is generated by a KrF excimer laser (LPX-350, 248 nm, 1 J, 25 ns, 50 Hz), also used in experiments from the collimator) optical transparency of 80%. The collimator is a multilayer coated spherical mirror of on EUVL near 13.5 nm.8 The laser beam is focused into a 30 mm spot on a tantalum disk, providing a power density of 260 mm diameter and a radius of curvature of 270 mm.The mirror was made by magnetron sputtering deposition of 120 up to 3×1012 W cm-2. These conditions are known to result in a plasma with a temperature of 100 eV and an intense bi-layers of Cr/Sc, with a period of 2.15 nm on a fused silica quartz substrate. The measured reflectivity of this coating quasi-continuous emission spectrum reaching up to several hundred eV.9 The source is designed to maximally reduce amounted to 4.4% in a 0.033 nm band (FWHM) at 4.28 nm, which corresponds to an interface roughness of the coating of contamination of the optics and sample by target material ablated by the laser pulse, generally consisting of atoms, about 0.4 nm.The collimator is positioned at a distance of Table 1 Experimental data on a number of low-Z Ka fluorescence lines under comparable conditions Energy of Conversion eYciency Relative fluorescence plasma (at 1013 W cm-2)a/ Penetration fluorescence Reflectivity of Relative intensity Material line/eV Jx Jlaser-1 sr-1 eV-1 shot depthb/nm yield(C=1) 3 mirrors (%) on detector (C=1) Be 109 3.6×10-3 26 0.25 21 5.4×103 B 183 6.0×10-4 30 0.38 0.3 19 C 277 5.9×10-5 60 1 0.06 1 N 392 6.8×10-6 130 2.5 0.07 0.34 O 525 5.5×10-6 260 6.5 2.0×10-4 2.0×10-3 aData obtained by L.A. Schmaenok (unpublished).bOn pure gold at normal incidence (calculated ). 388 J. Anal. At. Spectrom., 1999, 14, 387–390current set-up is reduced to 9×10-5 nm per shot, which would reduce the reflectivity by 50% after 2×105 shots. This was measured using an angular Cu Ka scan of a test sample positioned after the foil trap system, giving about 15 nm of material deposited after 105 shots. Deposition on the sample could not be measured and is therefore at least a factor of two lower than deposition on the mirror, leading to a deposition of less than 1 nm for an illumination within 104 shots.In this case, errors due to deposition of material from the laser plasma (Ta) can be neglected as absorption due to Fig. 3 First cross-section of the X-ray beam after the collimator. After deposition will give only a small error in the sampling depth final alignment of the foil trap system, its shadows are expected and no contribution to the fluorescence signal. The expected to disappear.spatial resolution based on the current choice of the CCD unit and magnification of the Schwarzschild set is about 1 mm. Taking the current source conditions, which are by far not about 270 mm from the laser plasma source, resulting in a optimal for the generation of X-rays near 300 eV, an exposure collection angle of 0.7 sr or about 10% of the radiation from time of about 600 s is needed to detect 5% carbon at a the plasma. Hence 0.5% of all plasma radiation near the resolution of 5 mm.central wavelength of 4.28 nm is focused on to the sample, assuming an isotropic angular distribution of the source intensity. Outlook Although the current results already give good prospects for Fluorescence detection and analysis system a feasible XRF-MDA set-up for the detection of carbon, a number of aspects, especially regarding the X-ray illumination The detection and analysis system currently under construction consists of a 20× Schwarzschild multilayer mirror microscope intensity and pollution suppression, need to be further improved.In addition to the present results on the debris with geometrical parameters equal to the Schwarzschild set described by Artioukov et al.12 and a filtered CCD unit. The reduction techniques ( lifetime >2×104 shots), a lifetime of 109 shots has been achieved in similar experiments with a laser mirrors of the Schwarzschild set are coated for the C Ka line by applying 120 bi-layers of Fe/C with a d-spacing of 2.25 nm.plasma X-ray source for lithography.11 Also, non-solid laser targets, such as gas jet targets13 and droplet targets,14 are The peak reflectivity of a single mirror of the microscope is about 10%, resulting in a 1% transmission of the two-mirror being investigated. However, the freedom of choice of solid target materials, most eYcient in the wavelength range of system. It is noted that the choice of an Fe/C coating will give some extra background due to fluorescence of the carbon in interest, remains an advantage of the present approach. The power density of 3×1012W cm-2 achieved on target, suYcient the mirror, but no errors in imaging result as the radiation from the mirror is not focused on the detector.The increase for the generation of lower energy photons (about 100 eV), needs to be increased for the eYcient generation of higher in background by this eVect is expected to be at least 109 times lower than the fluorescence signal, for a pure carbon sample.energy photons. The current system achieved a maximum yield of 5×1010 photons s-1. Further laser pulse upgrades, leading However, better suppression of radiation with an energy higher than the carbon K-edge is achieved, owing to higher absorp- to a power density >1013 W cm-2, are expected to increase the X-ray output at 300 eV by at least an order of magnitude, tion, increasing the contrast between the part of the incident radiation scattered on the sample surface and the fluorescence as measured using another laser system (Table 1).A considerable capacity of the performance of the multilayer radiation from the sample itself. The microscope collection angle is about 0.1 sr, allowing for multiple microscopes with optics is connected with the quality of mirror substrates, in particular the surface roughness. The latter parameter strongly diVerent central wavelengths to be used simultaneously for the study of multi-element fluorescence analysis.A second mirror aVects the mirror reflectivity: decreasing the interface roughnesses of the collimator and Schwarzschild set to 0.2 instead with a slightly diVerent d-spacing, just next to the fluorescence line, is used for the determination of background levels of of 0.4 nm will increase the throughput of the three-mirror system 10-fold. A lower roughness also reduces the reflectivity scattered radiation. The expected reflectivity curves of both the analyzer and the background monitor are plotted in Fig. 1, of the peak side bands of the mirror reflectivity and therefore reduces the background levels. With an increased yield of the taking into account the roughness of the substrates used. Currently, the illumination system is being characterized plasma, together with improved optical components with higher reflectivity, a gain of about 100 could be achieved, and first measurements have been made on the beam profile at a position behind the foil trap.Fig. 3 shows the image of leading to the detection of 5% carbon with a resolution of 5 mm in 6 s. an X-ray beam cross-section close to the focus on the sample. This image is used for alignment of the foil trap system by Another point to be addressed is the homogeneity of the beam on the sample. Although the system was not fully aligned X-rays. When aligned properly, the actual focus is more homogeneous and the foils are no longer imaged in the focal during the taking of the first X-ray image of the beam crosssection of the illumination system, it shows some irregularities plane.From Fig. 3, we conclude that the maximum illuminated area on the sample is <1 mm2. By placing a calibrated and in the beam profile, partly caused by the misalignment and partly by imperfections in the collimator. These issues are properly filtered photodiode in the focus of the collimator, the illumination system is measured to deliver 109 photons per likely to be resolved by improved alignment and improved homogeneity of the collimator coating.shot on the sample at an energy of 292 eV in a bandwidth of 2.5 eV. This value was obtained at a laser power density on In first experiments, carbon structures deposited in a controlled environment will be used to characterize the system, the target of 3×1012 W cm-2 on target at an energy of 1 J per pulse. The measured intensity corresponds to a conversion but at a later stage also carbon structures will be studied which are found in more industrial situations, such as the eYciency (Elaser/E292 eV) of the laser plasma of about 2×10-6 in a 1 eV bandwidth with the current plasma heating con- carbon deposition in catalysts.In these circumstances the fine structure of the carbon edge becomes important and needs to ditions. Using the debris suppression techniques mentioned above, the deposition rate on the collimator mirror in the be studied in detail.J. Anal. At. Spectrom., 1999, 14, 387–390 3892 R. Cao, X. Yang and P. Pianetta, J. Vac. Sci. Technol., B, 1993, Conclusions 11, 1455. 3 A. Sarhangi and J. M. Power, J. Vac. Sci. Technol., A, 1992, The design and the first phases of the construction and 10, 1514. characterization of an experimental facility for LE-XRF- 4 F. A. van Goor, W. J. Witteman, J. C. M. Timmermans, J. van MDA, being developed for the determination of carbon have Spijker and J.Couperus, Proc. SPIE, 1994, 2206, 30. been described. The new technique includes illumination of 5 D. G. Stearns, R. S. Rosen and S. P. Vernon, Appl. Opt., 1993, the sample with X-radiation from a laser plasma source using 32, 6952. 6 E. Louis, A. Yakshin P. C. Go� rts, E. L. G. Maas and F. Bijkerk, in a large multilayer mirror collimator and analysis of the distri- Proceedings of the 43rd International Conference on Electron, Ion bution of fluorescence radiation across the sample surface and Photo Beam Technology and Nanofabrication, June 1999, using multilayer Schwarzschild optics.The lifetime of the submitted for publication. collimator, closest to the laser plasma source, was determined 7 N. N. Salashchenko, Yu. Ya Platonov and S. Yu Zuev, Nucl. to be 2×105 shots. The intensity on the sample (109 photons Instrum. Methods Phys. Res. A, 1995, 359, 114. per shot at a photon energy near the C K-edge) was measured 8 F. Bijkerk, L. A. Shmaenok, A.van Honk, R. Bastiaensen, Yu. Ya. Platonov, A. P. Shevelko, A. V. Mitrofanov, F. Voß, R. and first images of the alignment of the foil trap system have De�sor, H. Frowein and B. Nikolaus, J. Phys. III (Paris), 1994, been presented. Improvements of various components of the 4, 1669. current set-up would permit a decrease in exposure time, 9 A. P. Shevelko, L. A. Shmaenok, S. S. Churilov, adequate for the detection of 5% carbon at a resolution of R. K. F. J. Bastiaensen and F. Bijkerk, Phys. Scri., 1998, 57, 276. 5 mm, from 600 down to 6 s. 10 L. A. Shmaenok, C. C. de Bruijn, H. Fledderus, R. Stuik, A. A. Schmidt, D. M. Simanovskii, A. A. Sorokin, T. A. Andreeva and F. Bijkerk, Proc. SPIE, 1998, 3331, 87. Acknowledgements 11 L. A. Shmaenok, F. Bijkerk, C. Bruineman, R. K. F. Bastiaensen, A. P. Shevelko, D. M. Simanovskii, A. N. Gladskikh and S. V. This work was financially supported by FOM (the Foundation Bobashev, Proc. SPIE, 1995, 2523, 113. for Fundamental Research on Matter), STW (the Netherlands 12 I. A. Artioukov, A. V. Vinogradov, V. E. Asadchikov, Yu. S. Technology Foundation) and the EC program INCO- Kas’yanov, R. V. Serov, A. I. Fedorenko, V. V. Kondratenko and Copernicus (IC15 CT97 707). S. A. Yulin, Opt. Lett., 1995, 20, 2451. 13 H. Fiedorowicz, A. Bartnik, P. Paris and Z. Patron, in Proceedings of the 13th International Congress on X-ray Optics and References Microanalysis, Inst. Phys. Cont. Ser. 130, IOP Publishing, Bristol, 1992, p. 515. 1 P. Pianetta, N. Takaura, S. Brennan, W. Tompkins, S. S. 14 L. Rymell and H. M. Hertz, Rev. Sci. Instrum., 1995, 66, 4916. Laderman, A. Fischer-Colbrie, A. Shimazaki, K. Miyazaki, M. Madden, D. C. Wherry and J. B. Kortright, Rev. Sci. Instrum., 1995, 66, 1293. Paper 8/07614F 390 J. Anal. At. Spectrom., 1999, 14, 387&ndash

ISSN:0267-9477    

DOI:10.1039/a807614f

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

The monolithic X-ray polycapillary lens and its application in microbeam X-ray fluorescence† Xie Jindong,* He Yejun, Ding Xunliang, Pan Qiuli and Yan Yiming Institute of Low Energy Nuclear Physics, Beijing Normal University, Beijing 100875, China Received 2nd September 1998, Accepted 20th November 1998 A monolithic polycapillary X-ray lens is a single piece glass optic made in a furnace. It is composed of hundreds of thousands of individual capillaries. The X-ray lens can be used to collect divergent X-rays emitted from an X-ray source in a large solid angle and to transmit them with high eYciency by multiple total reflections in the individual capillaries, thus forming an intense focused beam. Recently, new monolithic focusing lenses have been manufactured, tested and used for microbeam X-ray fluorescence (MXRF) analysis in the authors’ laboratory. For the Mo Ka line, the measured focal spot diameter was 30 mm.A gain factor of more than 2000 compared with when an aperture was used instead of the lens was obtained.The application of the lens in MXRF reduced the detection limits for transition elements down to the sub-pg level when a 1 W X-ray source was used. In the last 20 years, microbeam XRF (MXRF) spectrometry thousands of individual capillaries in a single bundle.10 It can be made by single or multiple composition processing of has become an increasingly popular and promising analytical method. This is partly due to the increasing importance of the drawing a bundle of glass capillaries in a drawing tower.In a single processing step a bundle of thousands of monocapillaries detailed and comprehensive analysis of micromaterials and of extremely small samples of bulk materials.1,2 On the other is simply drawn through an oven to form a monolithic lens. Such lenses are designed for application at low X-ray energies, hand, the use of high flux synchrotron sources and the application of capillary X-ray focusing optics oVer very high where large diameters of the channels and a relatively small number of capillaries per lens are needed.For hard X-ray spatial resolution with powerful X-ray microbeams for such analyses. lenses, a much larger number of capillaries (hundreds of thousands) and smaller diameters of the channels (a few There are two types of capillary optics being used today: monocapillaries3–5 and polycapillary focusing systems6,7 microns) are required, so in a first step one has to make the polycapillary bundles by drawing in a drawing tower, followed (Kumakhov lenses).Both are based on multiple total reflection of X-rays on the inner surface of capillaries. The polycapillary by drawing a bundle of polycapillaries to form a hard X-ray lens. X-ray lens is composed of a large number of capillaries and can collect divergent X-rays emitted from an X-ray source in The monolithic X-ray lens is compact, flexible and can easily be installed in available devices.The important characteristic a large solid angle and transmit them in capillaries with high eYciency, forming a very small focal spot (focusing lens, for of a monolithic lens in comparion with an assembled lens is that the cross-section of each channel changes along its axis, use in MXRF) or a quasi-parallel X-ray beam (for X-ray diVraction studies). so that all channels of the optic are independently oriented to the focal point of the lens. In principle, the focal spot size is Among the diVerent types of X-ray lenses, the monolithic X-ray lens has unique advantages in application.The mono- determined only by the divergence of the X-ray beam as it emerges from each individual channel. In practice, of course, lithic X-ray lens is a single piece glass optic made of capillaries in a furnace. Such lenses are very compact, have high trans- it is very diYcult to realize this situation at the current level of technology. mission eYciency and are convenient for utilization. The manufacture and evaluation of monolithic lenses and their The design of an X-ray lens is based on modelling of the multiple total external reflection of X-rays on the inner surface application in MXRF in the X-Ray Optics Laboratory at Beijing Normal University was started 4 years ago, and its of the X-ray capillary channels with simple roughness correction.A polycapillary lens uses curved channels (radius of development is still in progress.In this paper a brief review of the manufacure, measurement curvature R) to cut oV X-rays of high energies which have grazing angles larger than the critical angle of total reflection of basic characteristics and application of the monolithic lens is presented. The possible combination of a polycapillary lens for a given material. A simple geometric calculation yields the following inequality for succesful transmission: with a monocapillary is also briefly discussed.R>2d/hc2 (1) The manufacture and design of the monolithic X-ray where d is the diameter of the channel, R the radius of lens curvature and hc the critical angle of total reflection, which is The development of the X-ray lenses in our laboratory involved linearly proportional to the wavelength of X-ray. X-rays will three steps. The first generation of X-ray lens is the optics not be transmitted when the geometry of the channel does not assembled from monocapillaries,8 the second generation is the satisfy inequality (1).For the monolithic X-ray lens, the optics assembled from polycapillaries9 and the third generation transmission eYciency depends on the photon energy and is the monolithic lens, which is made up of hundreds to geometric parameters of the lens such as the sizes of the channels (diameters and lengths) and the shape and size of the lens. In designing the monolithic lens, first one should deter- †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998.mine the type of lens and the desired energy range of X-rays J. Anal. At. Spectrom., 1999, 14, 391–394 391according to the requirements of the application. Then one determines the geometric parameters of the lens by a simulation calculation optimizing the performance of the lens. A computer code for designing X-ray monolithic lenses based on simulations of X-ray trajectories by Fresnel reflection and a refraction equation with wall roughness correction and diVerential geometry was developed.11 Characterization and measurement of monolithic lens A monolithic lens can be characterized by its physical parameters, namely the transmission eYciency, gain in power density, focal spot size and equivalent distance.10 The physical properties of the lens are to some extent also dependent on the characteristics of the X-ray source used in the experiments.Here, an REIS-I (Svetlane-Rentgen, St.Petersburg, Russia) X-ray source was employed for the evaluation lenses. The power of the X-ray generator was 5 W and an Mo anode X-ray tube with an 80 mm spot size of X-rays was used. A detailed description of the measurement system was publised previously.8 The whole experimental apparatus was placed along the optical track on an optical table. Several multi- Fig. 2 (a) Knife edge scan at the focal spot of lens F4 using the Mo dimensional holders were placed on the optical track for Ka line.(b) DiVerentiated knife edge scan at the focal spot of lens mounting and aligning the X-ray source, X-ray lens, knife F4 using the Mo Ka line. edge, aperture, X-ray detector, etc. A schematic diagram of the measuring set-up is shown in Fig. 1. Determination of the focal distance and of the focal spot size The lens must be aligned at the beginning of the measurement. The lens alignment can be achieved by changing the distance between the lens and X-ray source and the source position in a transverse direction relative to the axis of lens to maximize the X-ray intensity through the lens.At the maximum throughput, the distance between the source and the input of the lens is defined as the input focal distance. The distance between the output surface of the lens and the focal plane is defined as the output focal distance. The focal spot size can be obtained by measuring the integrated curves of counts of Fig. 3 FWHM of the output beam as a function of the distance from X-rays with a 0.5 mm thick Mo knife edge scanning at diVerent the exit surface of lens F4. distances from the lens output surface [see Fig. 2(a)]. These curves are then diVerentiated to give the space distribution of Table 1 Geometrical parameters and measured results for monolithic lens F4 the power density of the X-ray beam at that particular distance from ouptput surface of the lens. The focal spot size is defined Lens length 52 mm as the minimum full width at half maximum (FWHM) of the Entrance diameter 6.5 mm space distribution of the measured power density.The result Exit diameter 5.0 mm of such a procedure for lens F4 is shown in Fig. 2 and 3. The Input focal distance 45 mm main geometric parameters and physical characteristics of lens Output focal distance 15.65 mm Focal spot size 30 mm F4 are given in Table 1. Transmission eYciency at 17.4 keV 0.8% Gain of power density 2317 Transmission eYciency Equivalent distance 2.34 mm The transmission eYciency of a lens is defined as the ratio of the intensity of X-rays emerging from the lens on the output measuring the direct beam flux (without the lens) at a given side to the intensity of the X-rays incident upon the lens distance and making a correction for solid angle and distance.entrance. The former term can be measured directly at the The transmission eYciency of a lens is a function of the exit of the aligned lens.The latter term can be determined by X-ray energy, since the value of the critical angle of total reflection is dependent upon the X-ray energy. The transmission eYciency of lens F4 was measured to be 0.8% for the Mo Ka line. There is still the possibility of increasing the transmission eYciency by improving the technology and optimization of the lens parameters. It is should be pointed out that misalignment of the X-ray source–lens system in the direction perpendicular to the lens axis can significantly influence the transmission eYciency, as shown in Fig. 4.The experimental data can be fitted by a Fig. 1 Schematic diagram of the measuring set-up. (1) X-ray source; Gaussian curve. The FWHM of this curve depends strongly (2) lens; (3) aperture or knife edge; (4) HpGe detector; (5) five axis holder; (6) three axis holder; (7) optical track. on the critical angle of total reflection of X-rays and the spot 392 J. Anal. At.Spectrom., 1999, 14, 391–394Fig. 5 Spectrum of sputting alloy on Mylar film. Fig. 4 Transmission of the Mo Ka line as a function of source ground, and therefore the signal-to-background ratio can be misalignment for lens F4. higher and the minimum detection limit (MDL) of MXRF can be improved; and (c) compared with monocapillaries, the size of the X-ray source. The transmission curve implies that working space on both the source side and the sample side the X-ray source has to be positioned exactly at an appropriate can be increased, because the monolithic lens usually has an position for maximizing the transmission eYciency, because input focal distance of tens of millimetres and an output focal only the X-rays which can reach all channels under a grazing distance of more than 10 mm.angle less than the critical angle of total reflection will As an example, experimentally determined MDL values contribute substantially to the transmission eYciency.obtained in an MXRF analysis set-up which was described previously8 is discussed. A standard sample was prepared by Gain and equivalent distance of the lens sputtering a steel alloy on to a thin Mylar film in the Synchrotron Laboratory of the Institute of High Energy The gain in power density obtained by the lens is defined as Physics of the Chinese Academy of Sciences. A spectrum taken the ratio of the X-ray intensities at the focal spot of the lens from this alloy sample is shown in Fig. 5. Based on this with and without the lens. It is determined by directly measurspectrum and the nominal contents of the elements in the ing the focused beam intensity and the X-ray intensity with alloy, the MDL values were obtained using the commonly an aperture at the position of the focal spot without the lens. used equation For an isotropic X-ray source the gain (K) can be calculated by means of the following equation: MDL=3.29CNb1/2/Na (4) where C is the mass of each element in the sample, Nb the K(E)=AL f1B2 Sin Sspot g(E) (2) background counts and Na the net counts of peak area.The results are presented in Table 2. where g is the transmission eYciency of the lens for X-rays of All previously measured MDL values which were obtained energy E, L the distance from the X-ray source to the focal during the last 4 years are listed in Table 3 in order to illustrate spot of lens, Sin the entrance area of the lens, Sspot the area of the advances made in MXRF sensitivity using monolithic the focal spot and f1 the input focal distance.focusing X-ray lenses. It can be seen that the MDL for Fe is The gain of the lens directly reflects the amplification of the about 0.5 pg. There is no doubt that the MDL can be improved X-ray beam power density by the lens. Another useful magni- signficantly if a much more intense X-ray source is used. tude is the equivalent distance (Leq). The equivalent distance can be calculated with the equation Table 2 MDLs obtained by MXRF analysis using the lens F4.Conditions: Mo anode, 27 keV, 36 mA, 1000 s Leq=L/K1/2 (3) where Leq is the distance between the X-ray source and the Content Net peak Background MDL/ Element C/pg (counts) (counts) pg point where the power density of X-rays emitted directly by the X-ray source is equal to that in the focal spot of lens when Cr 15.7 4459 1986 0.526 the lens is used. An equivalent distance of 2–3 mm was Fe 52.9 14650 1756 0.498 achieved for lens F4 and the X-ray source REIS-I.This means Ni 12.8 2677 3080 0.873 that one can perform experiments at a point 2–3 mm from the Ga 179.9 16753 2386 1.896 X-ray source when the focusing lens is applied. As 169.1 7113 2471 3.888 Se 302.1 10543 1699 3.886 Application of monolithic focusing lens in MXRF Table 3 Improvement of MDL (pg) for elements obtained by MXRF MXRF analysis is used in many fields of pure and applied using a monolithic focusing X-ray lens science and industry,12–16 such as physics, chemistry, biology, materials science, earth science, life science, enviromental Cr Fe Ni Spot size and sample used Ref.science, medical science, microelectronics industry, metallurgy and even archaeology and forensic medicine for analysis and 84.2 56.6 235 157 mm, mix 05a 17 15.2 7.72 38.5 50 mm ( lens+aperture), 18 mapping of element contents. In recent years, monocapillaries mix 05a (straight, tapered and ellipsoidal ) have been applied in 0.865 0.722 50 mm, sput 02b 19 MXRF.4,5 The application of the monolithic X-ray focusing 0.526 0.498 0.873 30 mm, sput 01b This work lens has several advantages: (a) the lens can produce a very aSample prepared by pipettting a mixed standard solution on to a intense X-ray microbeam (tens of microns) from a relatively thin Mylar film.bSample was prepared by sputtering a steel alloy on weak X-ray source; (b) the lens has a definite bandwith of to a thin Mylar film.transmission, so it can cut oV the high energy photon back- J. Anal. At. Spectrom., 1999, 14, 391–394 3933 D. J. Thiel, D. H. Bilderback, A. Lewis and E. A. Stern, Nucl. Instrum. Methods, Sect. A, 1992, 317, 547. 4 A. Attaelmanan, S. Larsson, A. Rindby, P. Voglis and A. Kuczumow, Rev. Sci. Instrum., 1994, 65, 7. 5 K. Janssens, L. Vincze, B. Vekemans, F. Adams, M. Haller and A. Kno�chel, J. Anal. At. Spectrom., 1998, 13, 339. Fig. 6 Schematic diagram of the experimental set-up. 6 M. A. Kumakhov and F. F. Komarov, Phys. Rep., 1990, 191, 289. 7 Q. F. Xiao and N. Gao, in Microbeam Analysis—1995, ed. E. S. Etz, VCH, New York, 1995, p. 157. The combined system of lens and monocapillary 8 Y.-M. Yan and X.-L. Ding, Nucl. Instrum.Methods, Sect. B, 1993, 82, 121. Although the monolithic lens has the advantage of collecting 9 X.-L. Ding, W. Liang and Y.-M. Yan, J. Beijing Normal Univ. X-rays in a large solid angle, the focal sposize cannot be as (Nat.Sci.), 1995, 31 (Suppl.), 40. small as that obtained by means of a monocapillary. Hence it 10 Y.-M. Yan, ‘Proceedings of the 45th Annual Conference, August 3–8, 1996, Denver, USA’. Adv. X-Ray Anal., 40, CD version. is desirable to create a combination of a monolithic lens with 11 B.-Z. Chen and Y.-M. Yan, J. Beijing Normal Univ. (Nat. Sci.), a tapered monocapillary to form small spots. A schematic 1995, 31, (Suppl.), 30. diagram of the experimental set-up is shown in Fig. 6. The 12 D. C. Wherry, B. J. Cross and T. H. Briggs, Adv. X-Ray Anal., system of diVerent monolithic focusing lenses and a tapered 1988, 31, 93. monocapillary was measured with a Cu anode. The tapered 13 K. Janssens, L. Vincze, F. Adams and K. W. Jones, Anal. Chim. monocapillary had an entrance diameter of 290 mm, an exit Acta, 1993, 283, 98. 14 A. Rindby, P. Voglis and A. Attaelmanan, X-Ray Spectrom., 1996, diameter of 48 mm and a length of 102 mm. The beam size of 25, 39. the tapered capillary exit end at 2 mm was measured to be 15 D. A. Carpenter and M. A. Taylor, Adv. X-Ray Anal., 1991, 34, 36 mm. The intensity of the combination of a monolithic lens 217. and a tapered monocapillary is 0.1–1.5 times that of a tapered 16 G. J. Havrilla, X-Ray Spectrom., 1997, 26, 364. monocapillary. Ideal experimental results were not obtained 17 X.-L. Ding, Y.-J. He and Y.-M. Yan, J. Beijing Normal Univ. because the combined system of lens and monocapillary was (Nat. Sci.), 1995, 31 (Suppl.), 75. 18 X.-L. Ding, Y.-J. He and Y.-M. Yan, X-Ray Spectrom., 1997, not optimized. The optimization and measurement of the 26, 374. combined system of a monolithic focusing lens with a tapered 19 X.-L. Ding, Y.-J. He, F.-Z. Wei, J.-D. Xie, D.-C.Wang, Y.-D. Li, monocapillary which has exit diameter of 10 mm are in J. Chen and Y.-M. Yan, ‘Proceedings of the 46th Annual progress. Conference, August 4–8, 1997, Denver, USA’, to be published in Adv. X-Ray Anal., 41. References Paper 8/06836D 1 A. Rindby, X-Ray Spectrom., 1993, 22, 187. 2 K. Janssens, L. Vincze, J. Rubio, F. Adams and G. Bernasconi, J. Anal. At. Spectrom., 1994, 9, 151. 394 J. Anal. At. Spectrom., 1999, 14, 391–394

ISSN:0267-9477    

DOI:10.1039/a806836d

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Characterisation of surface layers formed under natural environmental conditions on medieval stained glass and ancient copper alloys using SEM, SIMS and atomic force microscopy† M. Schreiner,*ab G. Woisetschla�ger,a I. Schmitza and M.Wadsaka aInstitute of Analytical Chemistry, Vienna University of Technology, Getreidemarkt 9/151, A-1060 Vienna, Austria. E-mail: mschreiner@fch.akbild.ac.at bInstitute of Chemistry, Academy of Fine Arts, Schillerplatz 3, A-1010 Vienna, Austria Received 18th September 1998, Accepted 26th November 1998 Atmospheric corrosion is the result of interactions between a material and the surrounding environment.It involves a number of physical, chemical and electrochemical processes in the interfacial region ranging from the contacting atmosphere over the aqueous adlayer to the material itself. In this contribution surface analytical techniques such as scanning electron microscopy, in combination with energy dispersive X-ray microanalysis (SEM-EDX) and secondary ion mass spectrometry (SIMS) were applied to characterise the corrosion phenomena occurring on medieval stained glass and ancient bronze artefacts.A so-called leached layer has been formed on the glass surfaces due to an ion exchange process, where the potassium and calcium of the glass are replaced by hydrogen bearing species from the moist air. Subsequently, chemical reactions of the leached glass constituents K and Ca with acidifying gases in the ambient atmosphere has led to the formation of a weathering crust.On the surfaces of bronze artefacts a cuprous oxide (mainly) has been built up. Further chemical reactions are leading to crystalline weathering products such as brochantite or malachite depending on the environmental conditions. Additionally, tapping mode atomic force microscopy (TM-AFM) has been applied to study the initial stages of the weathering processes on glass with medieval composition and on pure copper.The task of the present work was not only to develop new analytical strategies and methods, but also to gain additional information of the surface processes involved. Protection against atmospheric corrosion requires such detailed understanding of the role of diVerent corrosion stimulating constituents in the environment, such as humidity, gaseous pollutants and aerosol particulates. or in the contacting ambient atmosphere or soil, which can Introduction occur in some cases, can be detrimental.The corrosion mechan- When exposed to the ambient atmosphere or to soil most isms leading to such surface layers are, in principle, well known inorganic materials such as glass and copper, form a thin layer and are related to trace species found in the corrosion medium. of corrosion on their surfaces. In the case of archaeological The surface layers formed do not reflect the atmospheric glass from the ancient periods this corrosion layer can have a composition directly, however, but clearly favour those atmosthickness of several tens of micrometres up to 100 mm and pheric constituents whose products have a certain solubility, yield an iridescent eVect due to the diVerent optical properties chemical reactivity and rate of formation.of the surface layer and the bulk glass. In the case of copper In the present paper, information on the corrosion layers alloys generally a brownish-green or greenish-blue layer is formed under natural conditions on medieval stained glass formed, designated the patina, a name derived from the green and ancient copper alloys is presented.This requires analytical crust often found on ancient Roman dishes, or patens. Both techniques adequate for the morphological and chemical iridescent glass artefacts and copper patinas are generally characterisation of the surfaces and the interfaces. Therefore, regarded as aesthetically pleasing and in the 19th century, the common tool of scanning electron microscopy in combieven, procedures were developed and patented in order to nation with energy dispersive analysis (SEM-EDS) and secproduce the splendid rainbow coloured gleam of glass surfaces.ondary ion mass spectrometry (SIMS) was used for the The famous iridescent Art Nouveau glass artefacts of Louis analysis of the corrosion products and the surface layers Comfort TiVany in New York or Johann Loetz in Vienna are formed over the centuries or even a millennium.Additionally, typical examples. Also, the use of copper alloys in art as well tapping mode atomic force microscopy (TM-AFM), a special as in modern architecture is based on the artificial formation technique of scanning probe microscopy (SPM), could be of patina. applied to studying the initial stages of corrosion processes. In general, once the corrosion layers on the glass surface, Sample glass with a chemical composition similar to medieval as well as on copper alloys, are established they tend to be stained glass and pure copper were investigated in situ and extremely stable.However, any changes in the surface layers exposed to a defined atmosphere. Due to the potential of AFM for revealing topographic changes down to the atomic scale under ambient conditions, a large amount of information †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998. could be provided. J.Anal. At. Spectrom., 1999, 14, 395–403 395[Cu3(SO4)(OH)4], atacamite [Cu2Cl(OH)3], gerhardite Corrosion of glass and copper [Cu2NO3(OH)3] and malachite [Cu(CO3)(OH)2]. In some Medieval stained glass used as window panes in glass paintings cases formate, acetate or oxalate also could be identified20 as of Romanesque and Gothic cathedrals, chapels or minsters is well as atmospheric particles and black spots, which are characterised by a high content of modifier ions.1,2 Systematic typically a few millimetres in diameter, very hard and very investigations carried out on artefacts all over Europe have adherent, containing some phosphorus in addition to copper.revealed that the majority of the samples consist of silica, lime, An extensive series of experimental and theoretical analyses magnesia and potash, instead of soda, which had been usually was carried out by Graedel20 concerning the formation of used for ancient as well as common modern glasses.3–7 Their copper patinas in the atmosphere. The most important result relative ratios vary from one glass to another, but their sum is that brochantite is the most stable of the major patina remains high (approximately 90% m/m).The remainder of the components in moderately acid solutions evolved from the less composition includes the oxides of phosphorus, aluminium, stable precursor mineral posnjakite. Additionally, a three or manganese, iron and sodium, traces of other elements such even four stage process could be determined in the evolution as barium and strontium3–6 and metal oxides added as of the visual appearance of exposed copper from its initial colouring agents.1,3 state to that of final patination: new, freshly cleaned, copper Studies on glass deterioration caused by atmospheric attack shows its salmon-pink colour; after a few weeks of exposure are very complex due to the numerous factors involved.2,3,8–10 to outdoor atmosphere copper turns to a dull brown shade, These factors depend on the characteristics of the glass (e.g., which gradually deepens to become black; finally, the greenishcomposition, colour, thermal history, surface roughness) as blue layer of terminal copper patina is formed.The time scales well as on the external conditions such as micro-climate, for such a patina formation vary substantially with geographic exposure and atmospheric pollutants.4–8 For glass in contact location, and there is some suggestion that the rate of with water or aqueous solutions two corrosion mechanisms patination has increased over the past half century.are described in the literature. The first is a leaching process, The results concerning the corrosion of glass and copper where the glass constituents such as Na, K and Ca are leached alloys represent later stages of the weathering process, where out of the glass, and incorporation of hydrogenr hydrogen the corrosion has already altered the surfaces drastically.bearing species into the silicate structure takes place according Hardly any information on the initial stages of the corrosion to eqn. (1). processes can be found in the literature because of the limited lateral resolution of the techniques used.21,22 KSiKO- M++H(sol)+�KSiKOH(sol)+M+ (1) This ion exchange process leading to the formation of a Analytical approach so-called leached layer3,11 on the glass surface occurs mainly in aqueous solutions with a pH <8, whereas in corrosion In corrosion science a set of analytical methods has been media with higher pH values a dissolution of the silicate applied in combination since each technique exhibits specific network, the second corrosion mechanism of glass, according capabilities and limitations.In general, analytical techniques to eqn. (2), dominates. for the morphological and chemical changes of the surface are applied such as SEM, EPMA, analytical electron microscopy, KSiKOKSiK+OH-�KSiKOH+KSiKO- (2) AES, XPS, XRD, low energy electron diVraction (LEED), Rutherford backscattering spectroscopy (RBS), SIMS and, Studies carried out on naturally weathered medieval glass have shown that both degradation processes also occur on medieval for special cases, bulk elemental techniques like XRF, AAS, NAA and ICP-AES in combination with selective chemical stained glass.7,12 The acidifying gases of the ambient atmosphere lead to an increase of the H3O+ concentration in the etching of surface and interface layers.23,24 In the present study SEM was used for the morphological associated precipitation (e.g., rain, fog) as well as in the water adlayer always present on the glass surface due to adsorption characterisation and in combination with the energy dispersive X-ray microanalysis for the chemical identification of the and condensation phenomena.12–15 On the other hand, the leaching process of K and Ca itself gives an increase in the corrosion products formed.For these a JEOL (1418 Nakagami Akishima, Tokyo, Japan) instrument, Type SEM 6400, and pH in the adsorbed water film, which can initiate the breakdown of the silicate network. As a last stage of these processes an energy dispersive system of Link (High Wycombe, Bucks., UK), Type eXL, was used. The glass samples, as well as the weathering products like gypsum (CaSO4·2H2O), syngenite (K2SO4·CaSO4·H2O), arcanite (K2SO4) or schoenite samples of the naturally weathered ancient bronze objects, had to be coated with a thin layer of carbon prior to analysis.(K2SO4·MgSO4·6H2O), and also carbonates such as calcite (CaCO3) or organic compounds such as Ca oxalate, can be In special cases small splinters of the glass and bronze objects were taken, embedded in resin and cross-sectioned perpendicu- formed as crystalline corrosion products depending on the chemical composition of the original medieval glass and the lar to the surface in order to study the elemental distribution in the near surface region and get an overview of the compo- environmental conditions.3,4 The formation of the familiar green patina on pure copper nents leached from the materials.The backscattered electron signal (BE images) and X-ray mapping (element distribution and copper alloys upon exposure to the atmosphere is perhaps the most striking and most common evidence for the inter- images) were used.Secondary ion mass spectrometry (SIMS) plays a unique action of atmospheric trace substances with metal surfaces. Similar to the weathering of medieval glass, the principle role in corrosion science and surface analysis due to its high detection power for most elements, the facilitation of in situ mineral components of the patina have been known for a long time16–18 but little mechanistic work took place until Mattson19 trace analysis with high spatial resolution and its use in depth profiling.However, a strong matrix eVect must be considered, constructed stability diagrams for copper in aqueous solutions relevant to those encountered during atmospheric chemical when interpreting the depth profiles quantitatively in some cases, and high beam energy can obscure depth profile details exposure. The work indicated that the minerals found in natural copper patinas appeared to be those which would be by disrupting the matrix significantly during sputtering.Additionally, applying SIMS to the analysis of glasses as well thermodynamically stable under atmospheric exposure. A number of insoluble or poorly soluble species are described, as metals covered with minerals as corrosion products is a complex procedure because of sample charging occurring such as copper oxides (Cu2O, CuO), copper sulfides (Cu2S, CuS), brochantite [Cu(SO4)(OH)6], antlerite during the analysis of insulators under energetic ion bombard- 396 J. Anal.At. Spectrom., 1999, 14, 395–403ment. Using a Cameca (Courbevoie, France) IMS-3f ion mass spectrometer the specimens, first coated with gold, were bombarded with primary mass filtered 16O- ions at 14.5 keV. A beam current of 100 nA, a beam diameter of approximately 60 mm, a raster size of 250×250 mm, and an analysed area with a diameter of 10 mm were selected for the analysis. The secondary ions were accelerated into the double-focusing mass spectrometer by an actual accelerating voltage of +4500 V, which was controlled and corrected at each measurement cycle by a high-voltage autocontrol computer routine for further minimisation of the residual charging eVects.This routine, already described in the literature,25 checks the position of the energy distribution of a reference mass (for the glass analyses 29Si+) and normalises the position of the energy distribution of all other masses to that reference element. Due to the large number of elements present in original medieval glass, a Fig. 1 Schematic of the optical head of the AFM used for the computer routine was used which enabled the simultaneous weathering of glass. analysis of up to 29 masses. The counting time for each mass was 1 s but almost no depth information was lost during the analysis because of the rather low erosion rates used. The erosion rates of the various layers were determined from depth measurements using a DEKTAK profilometer (Sloan Technology Division, Veeco Instruments, Inc., Santa Barbara, CA, USA) of the SIMS craters obtained by shorter runs.The introduction of AFM (atomic force microscopy) in the 1980s by Gerd Binning, Calvin Quate and Christopher Gerber26,27 opened wide the field of imaging surface topographies in real space down to the atomic scale. The new technique enables imaging of electrically conductive as well as nonconductive specimens. The instrument is based on the scanning tunnelling microscope (STM), but instead of measuring the tunnelling current between the metal tip and the sample surface the AFM measures the interatomic forces between a sharp tip mounted on a soft spring (the cantilever) and the sample surface. The probe is scanned relative to the sample surface by means of piezoelectric transducers and the vertical position of the tip is monitored by a very sensitive position detector.Fig. 2 SEM micrograph of the surface of a naturally weathered There are several sensitive modes of operation and diVerent medieval glass showing the thick weathering crust formed.types of microscopes available. In static mode AFM a force acting on the probing tip bends the cantilever until a static equilibrium is reached. The tip is in permanent contact with the sample surface and the repulsive forces are sensed. In the dynamic (tapping) mode AFM the cantilever holding the tip is vibrating close to its resonant frequency. The tip is only in intermittent contact or not in contact with the sample surface.Depending on the tip touching the surface in each lower cycle of oscillation or oscillating in the vicinity of the sample surface with a spacing of 10–100 nm diVerent force gradients shift the cantilever’s resonance curve. Generally, the advantages of dynamic mode AFM are its higher sensitivity to shallow long range force gradients and the lower overall influences of tample surface, because typically the load forces are lower compared with those of contact mode AFM.This is important, especially for soft samples such as biological materials or swollen glass, as has been shown in previous publications.28,29 Fig. 3 SEM micrograph of gypsum crystals formed on the glass Since AFM is not restricted to UHV conditions, it has surface during weathering under natural conditions. become a valuable tool for the investigation of processes at the solid–liquid as well as the solid–gas interface. In situ observations allow the time-resolved visualisation of dynamic of 4×4 mm and a height of 10 mm, were used.Cut glass samples with a chemical composition similar to medieval processes with spatial resolutions that could not be achieved up to now. In the present work a NanoScope III system stained glass artefacts and polished copper were fixed in a home-built sample holder which consists of an aluminium ring (Digital Instruments, Santa Barbara, CA, USA) was used in diVerent configurations. The detection system of this instru- with two facing fixation screws glued to a steel disc.The steel disc was magnetically fixed to the head of the scanner. ment is based on laser beam deflection oV a microfabricated cantilever. The sample is moved in the three spatial directions The in situ investigations concerning the weathering of glass with medieval composition as well as pure copper were carried by means of a piezo tube scanner. Commercially available Si cantilevers, with a length of 125 mm, resonance frequencies of out in nitrogen and synthetic air with diVerent humidity levels and diVerent amounts of corrosive gases such as SO2.The 265–378 kHz and an integrated pyramidal tip with a base area J. Anal. At. Spectrom., 1999, 14, 395–403 397relative humidity was achieved by humidifying one part of the gas stream. The nitrogen gas was purged through a bottle filled with bidistilled water. The mixing rate was adjusted by means of flow meters.SO2 was introduced by adding a gas stream of 10 ppm (v/v) SO2 in N2 in synthetic air to the main gas stream. A schematic of the experimental set-up is shown in Fig. 1. Results and discussion The atmospheric attack at medieval glass paintings A medieval glass painting consists of numerous pieces of glass coloured with metal oxides, which are held together by a firm but elastic network of thin lead stripes following the lines of the composition. An accelerated degradation of the glass material must be observed predominantly at the exterior Fig. 4 SEM micrograph of the glass surface beneath the weathering surfaces, which has led to the formation of a so-called weathercrust. ing crust. As already discussed, this layer consists mainly of gypsum or syngenite as crystalline corrosion products (Fig. 2 and 3) and reduces the transparency of the glass so that in of K and Ca in the near surface region can be demonstrated also at cross sections. In Fig. 5 the broad leached layer is some cases the total composition is barely recognisable. Usually, this weathering crust is extremely hard and can hardly clearly shown, which is depleted on potassium and calcium compared with the bulk, whereas the content of Si in the be separated from the glass material. The surface of the glass underneath is shown in Fig. 4 after removing the crust surface layer seems to be the same as in the bulk glass. In contrast to such tremendously corroded glass surfaces, mechanically.Using energy dispersive X-ray microanalysis in the SEM, a panels and samples of medieval stained glass objects with an apparently unweathered surface also were analysed. A smooth more or less marked depletion of the network modifiers such as potassium but also calcium can be detected. The depletion and generally unaltered glass with just a little iridescence could Fig. 5 SEM micrograph of a cross-sectioned medieval glass (a) with corresponding elemental distribution of Si (b), K (c) and Ca (d). 398 J. Anal. At. Spectrom., 1999, 14, 395–403be observed in such cases. In the literature3–6 it is stated that glass with a total amount of all network formers above 66% (m/m) and potassium and calcium as network modifiers is characterised by a relatively high weathering durability. It should be mentioned here that the chemical composition of a glass is important, but only one factor influencing the weathering durability and consequently the present condition of a stained glass. Other important parameters are the environmental conditions and the process of glass manufacturing.However, analytical investigations carried out by SEM-EDS reveal a marked depletion of K and Ca on the glass surface compared to the bulk material (Fig. 6). Fig. 7 summarises the results obtained from SIMS measurements on such scarcely weathered medieval glass samples without any crystalline corrosion products. The raw depth profile shown in Fig. 7(a) was obtained by recording ion intensities measured by cyclic switching between the masses. Already, from the raw profile [Fig. 7(a)] a lower intensity for the main glass constituents K and Ca and the minor components Na and Ba in the near surface region [layer 3 in Fig. 7(a)] can be deduced, whereas the intensities, for example, at the masses 1 for hydrogen and 29 for silicon, are increased compared to the bulk. The conversion of the measured secondary ion currents versus analysing time into a concentration–depth profile in Fig. 7(b) was carried out by applying relative sensitivity factors (RFS) determined at cross sections of intact medieval glass using Si as reference element. This quantification procedure is already described in the literature30 and its applicability to glass analysis was already confirmed in an earlier study.31 The depth scale was obtained by measuring the SIMS craters and considering the varying erosion rates in the diVerent layers.The DEKTAK measurements revealed that the higher intensity of silicon (mass 29) in the leached layer in Fig. 7(a) is caused predominantly by the higher erosion rates of the negative primary ions in these surface domains. It is obvious from Fig. 7(b) that the network Fig. 7 SIMS profiles of the elements present in the near surface region modifiers K, Ca, Na, Ba and also Mg are strongly depleted in of an apparently unweathered medieval glass. (a) Raw profile, (b) the leached layer compared to the bulk.Due to the natural concentration/depth profile obtained by quantification of the raw weathering process all these elements, except Mg, are depleted profile by using relative sensitivity factors and considering the erosion rates in the diVerent layers: 1=Au layer, 2=outermost region, 3= leached layer, 4=bulk glass. by a factor of approximately 10. For Mg a content of approximately 30% of the bulk concentration was determined in the leached layer.Although hydrogen can be detected by SIMS, quantification of raw profiles is rather diYcult due to the hydrogen-containing residual gas in the sample chamber. Additional comparative measurements with nuclear reaction analysis (NRA) have revealed that up to 30 atom% H is present in the leached layer.32 In situ studies of the weathering of glass with medieval composition In order to study the initial stages of the weathering of medieval potash-lime-silica glass with atomic force microscopy (AFM), a model glass with a chemical composition similar to medieval stained glass was used: 48.0% (m/m) SiO2, 25.5% (m/m) K2O, 15.0% (m/m) CaO, 4.0% (m/m) P2O5, 3.0% (m/m) MgO, 3.0% (m/m) Na2O and 1.5% (m/m) Al2O3.The glass was prepared by the Fraunhofer-Institut fuer Silicatforschung in Wuerzburg, Germany, from reagent grade oxides and carbonates, as already described in the literature.33 In order to obtain a fresh surface glass plates of approximately Fig. 6 Results of the energy dispersive X-ray microanalysis obtained 10×10×4 mm were cut under dry nitrogen and transferred from an apparently unweathered medieval glass. The composition of to the AFM immediately before measurement was started in the bulk material was measured at a cross section of the sample. A the set-up shown in Fig. 1. remarkable depletion of the network modifiers K and Ca will be determined even on such glass surfaces. As a first step a dry nitrogen atmosphere was applied in J.Anal. At. Spectrom., 1999, 14, 395–403 399Fig. 8 Cut glass seen with TM-AFM under nitrogen with 70% relative humidity at room temperature. The scan size is 5×5 mm2 and the height range is 200 nm from black to white. (a) After cleavage the surface shows no features. (b) After 4 min round features with an average diameter of 150 nm appear, which are growing with time. (c) After 104 min a fraction of smaller round features with average diameters of 50–100 nm occurs in addition to the large features.(d) After 12 h the smaller round features have arranged around the larger ones. (e) After 34 h the large features appear irregularly shaped. (f ) After 72 h both the small features show the irregular shape of swollen glass and the large features start to show crystalline shapes. order to perform the sample preparation and the necessary tinued with time and after 72 h still distinct types of features were found on the surface [Fig. 8(f )]. adjustments to the microscope. No visible topographic changes occurred within 1 h under dry nitrogen [Fig. 8(a)]. After As already discussed, mainly sulfates of calcium and potassium were identified as weathering products on the switching to humid nitrogen with 25% relative humidity at room temperature, the first round features could already be surfaces of medieval stained glass objects. Therefore, it is believed that SO2 is one of the main reasons for accelerated observed after 4 min.This fact clearly indicates that the sensitive glass surface is suYciently protected by the dry corrosion and degradation processes. As an example of the interaction of glass in contact with humidified nitrogen with nitrogen and hence the set-up is able to serve as a simple weathering chamber. The features formed grew with time and varying amounts of SO2, the images obtained at the glass surface exposed to N2 with 1 ppm (v/v) SO2 and 70% relative their average diameters ranged from 10 to 100 nm.After 90 min the entire surface was covered and a surface layer humidity are summarised in Fig. 9. One minute after cleavage features which can be divided into two types could be found was formed. Similar results were obtained during the weathering on the surface [Fig. 9(a)]. Besides the already known round features, indicating a swelling of the glass material, crystals experiments at 50% as well as at 70% RH. In Fig. 8 the images obtained by AFM are summarised.After 4 min again round with shapes like flowers are found [Fig. 9(b)]. These crystals grow with time and after 1 h one half of the surface features with an average diameter of 200 nm and an average height of 50 nm were formed [Fig. 8(b)]. The growing of these is covered by them [Fig. 9(c)]. Continuing the weathering it becomes evident that the large crystals further grow with time, continued and besides the round features a new type of smaller feature with average diameter of 150 nm and an average height whereas the smaller round features do not change any more with time up to 6 h of exposure [Fig. 9(d)]. The merging of of 20 nm could be observed after 104 min [Fig. 8(c)]. This indicates that already two distinguishable processes can be swollen areas of the glass surface is observed 29 h after starting the exposure in the lower left corner of Fig. 9(e). This process detected and the resulting topographic changes can be divided into two separate classes.Both kinds of features were growing continues and after 74 h approximately half of the former glass surface is covered by the swollen glass material. At the with time and after 12 h the distribution of the smaller features indicated that in the vicinity of the larger features, which start same time the crystals formed continued to grow [Fig. 9(f )]. These results clearly reveal that scanning force microscopy to show crystalline shapes, the formation of the smaller features was suppressed [Fig. 8(d)]. The number of the smaller can be considered as a useful tool in surface analytical chemistry when carrying out in situ investigations concerning the features significantly increased with time, whereas the number of the large features remained constant but their diameters as weathering of glass. The combination with other traditional methods on the one hand and the modification of known well as their heights increased [Fig. 8(e)]. This process con- 400 J.Anal. At. Spectrom., 1999, 14, 395–403Fig. 9 Cleaved glass seen with TM-AFM under nitrogen with 1.0 ppm (v/v) SO2 and 70% relative humidity. The scan size is 10×10 mm and the height range is 200 nm from black to white for the first 5 images and 500 nm from black to white for the last image. (a) One minute after cleavage the surface shows small features. (b) After 4 min round features and corrosion products with diameters from 50–400 nm are visible. (c) After 1 h the corrosion products are grown in diameter.Their shape is irregular and compares to flowers. (d) After 6 h the corrosion products are further grown in height. The small round features are still visible. (e) After 29 h a helix structure of the corrosion products is obtained. (f ) After 74 h the large corrosion products further grow and the small round features start to merge together. AFM techniques, as well as development of new techniques malachite can be determined as main weathering products in on the other hand opens fascinating possibilities for the in situ addition to cuprite, which is usually always present as a layer characterisation of surfaces and their changes due to chemical between the surface of the copper alloy and the corrosion reactions in selected atmospheres.products. In many cases these large crystals are an indication of weathering under natural conditions, whereas fine grained The corrosion of ancient copper alloys and pure copper products must be observed after an artificial patination process.Cross sections of such ancient copper alloys show clearly the Similar to the surface of medieval stained glass large crystals corroded surface in the backscattered electron image, as shown are formed on the surfaces of ancient bronze objects during in Fig. 11(a), with an inhomogeneous distribution of the main exposure to the ambient atmosphere (Fig. 10). Using X-ray elements Cu and Sn [Fig. 11(b)]. In addition to these compo- diVraction analysis mainly brochantite and in some cases nents a high amount of chlorine could be detected in the corrosion product as well as in the corrosion layer [Fig. 11(b)]. The sulfur K-lines, which have a coincidence with the Pb M-lines in the energy dispersive system, are also present in the corrosion products. Based on such results, as well as on the systematic investigations mentioned above,20–22 a reaction sequence has been suggested for the interaction of SO2 with water adlayers covering copper.After the deposition of SO2 molecules and absorption in the water adlayer, hydrolysis and formation of bisulfate ions in the adlayer occurs followed by a fast exchange between sulfite ions and the hydroxylated metal surface. Hence, the formation of inner sphere sulfite complexes at the water–oxide interface, subsequent slow detachment of metal ions from the surface and, finally, precipitation of sulfite containing corrosion products occurs; these last can be oxidized to sulfates.21,22 In order to prove this model and study the initial stages of the weathering process of pure copper, AFM investigations Fig. 10 SEM micrograph of corrosion products formed on the surface of an ancient bronze object. were carried out using the set-up in Fig. 1. Therefore, samples J. Anal. At. Spectrom., 1999, 14, 395–403 401Fig. 12 Polished surface of pure copper seen with TM-AFM in synthetic air with 80% relative humidity.The scan size is 1 x 1 mm, the black to white height range is 25 nm and the phase range is 25 degrees from black to white. (a) Topographic and phase image of the freshly polished surface. (b) and (c) The images, especially the phase images after 80 and 1700 min, respectively, show the formation of round features completely covering the surface. the tip and the sample can be mapped simultaneously to the Fig. 11 Backscattered electron image (a) of a cross-sectioned sample topographic data with the same lateral resolution.Fig. 12(a) of an ancient bronze object with the corresponding elemental distridepicts a freshly polished copper sample with some scratches bution of Cu, Sn, Cl, K, S+Pb M and Pb L (b). 1=Soil material, 2=corrosion products, 3=surface layer, 4=bulk. and contamination originating from the polishing treatment. Exposure to nitrogen with diVerent humidities did not yield to any corrosion phenomena. Therefore, the experiments were carried out under synthetic air with 60 and 80% relative of 10×10×2 mm from copper sheet (maximum 25 ppm trace metal impurities) were mechanically abraded with SiC paper humidity as well as in humidified synthetic air and 1 ppm (v/v) SO2.At 60% relative humidity a very slow alteration of the down to 4000 mesh with ultrasonic cleaning in ethanol between the grinding steps. This procedure was followed by two surface topography could be observed, whereas at 80% relative humidity the weathering process was much faster.After 80 min diamond polishing steps using 1 and 0.25 mm diamond pastes. After a final cleaning with ethanol the sample was dried with the sample surface was totally covered with round features, 20 nm in diameter and 1 nm in height [Fig. 12(b)]. These a lint-free tissue and immediately transferred to the exposure chamber. The copper sheet was exposed for less than 15 s to corrosion products grow with exposure time to 50 nm in diameter and 4–6 nm in height after 1700 min [Fig. 12(c)]. the ambient atmosphere. The results of these investigations are summarised in the The experiments carried out in synthetic air and 1 ppm of SO2 yielded similar results. A very slow corrosion rate occurred Fig. 12 and 13, where the left image shows the topography of the Cu surface and the right image is the so-called phase at 60% relative humidity compared to the changes of the surface observed at 80% humidity. As can be seen in the image.Therefore, the phase diVerence between the driving signal of the piezo crystal oscillating the cantilever in the images of Fig. 13, the metal surface is already completely covered with round features after 60 min of exposure TM-AFM and the resulting motion of the tip is exploited in order to access additional information about the interaction [Fig. 13(b)]. In addition to the exposure without SO2 the formation of a second type of feature can be detected. These of the tip and the sample material.Adhesive forces between 402 J. Anal. At. Spectrom., 1999, 14, 395–403depending on the relative humidity and the amount of pollutants in the ambient atmosphere. The corrosion rate at 80% is significantly higher than at 60% relative humidity. Presumably these features consist of cuprous oxide. In SO2- containing atmospheres the formation of additional weathering products can be determined with AFM but their chemical nature will be studied in the near future in combination with other modern surface analytical techniques.Acknowledgement The authors greatfully acknowledge Professor Dr. M. Grasserbauer, Professor Dr. G. Stingeder, Professor Dr. G. Friedbacher and Mr. K. Piplits for enabling this investigation and for their support during the SIMS and AFM measurements. The Austrian Science Foundation is greatfully acknowledged for financing the project no. 9220-TEC. References 1 G. Frenzel, Sci. Am., 1985, 252, 100. 2 R. G. Newton, Glass Technol., 1985, 26, 21. 3 R. G. Newton, The Deterioration and Conservation of Painted Glass—a Critical Bibliography, Oxford University Press, Oxford, UK, 2nd edn., 1982. 4 R. Collongues, M. Perez y Jorba and G. Tilloca, Verres Refract., 1976, 30, 43. 5 G. A. Cox, O. S. Heavens and R. G. Newton, J. Glass Stud., 1979, 21, 54. 6 M. Schreiner, Glastechn. Ber., 1988, 61, 197. 7 M. Schreiner, Glastechn. Ber., 1988, 61, 223. 8 J. E. Shelby, J. Vitko and C. G. Pantano, Solar Energy Mater., 1980, 3, 97. 9 S. Railakshmi, M. Chakraborty and S. Basu, Trans. Indian Ceram. Soc., 1981, 40, 166. 10 J. O. Isard and A. R. Patel, Glass Technol., 1981, 22, 247. 11 H. Scholze, J. Non-Cryst. Solids, 1982, 52, 91. 12 M. Schreiner, G. Stingeder and M. Grasserbauer, Fresenius’ J. Anal. Chem., 1984, 319, 600. 13 M. Schreiner, M. Grasserbauer and P. March, Fresenius’ J. Anal. Chem., 1988, 331, 428. 14 F. J. Briggs, The Chemical Durability of Medieval Glass. Technical Report, Department of Ceramics, Glass and Polymers, University of SheYeld, UK, 1978. 15 J.Alderborn, Investigation of Weathered Glass Surfaces with the Scanning Microscope, OECD-Report, DAS/SPR/71–35, 1971, 244. 16 W. H. J. Vernon and L. Whitby, J. Inst. Metals, 1930, 44, 389. 17 W. H. J. Vernon, Trans. Faraday Soc., 1931, 27, 255. 18 W. H. J. Vernon, J. Inst. Metals, 1932, 49, 153. 19 E. Mattsson, Mater. Perf., 1982, 21, 9. 20 T. E. Graedel, Corrosion Sci., 1987, 27, 639, 721 and 741. 21 D. Persson and C. Leygraf, J. Electrochem. Soc., 1995, 142, 1459. 22 T. Aastrup, J. Tidblad, C. Leygraf, M. Wadsak and M. Schreiner, J. Electrochem. Soc., submitted for publication. Fig. 13 Polished surface of pure copper seen with TM-AFM in 23 H. W.Werner and G. Garten, Rep. Progr. Phys., 1984, 47, 221. synthetic air with 80% relative humidity and 1 ppm (v/v) SO2. The 24 R. E. Whan et al., American Society forMetals, Metals Park, OH, scan size is 1×1 mm, the black to white height range is 25 nm and the USA, 9th edn., Metals Handbook, vol. 10. phase range is 25 degrees from black to white. (a) Freshly polished 25 G. Stingeder, Anal. Chem., 1988, 60, 1524. surface. (b) After 60 min the round features, as seen in Fig. 12(b), 26 G. Binnig and H. Rohrer, Surf. Sci., 1983, 126, 236. are covering the surface. Additional features with a diameter of 60 27 G. Binnig, C. Quate and C. Gerber, Phys. Ref. Lett., 1986, 56, 930. and height of 10 nm were formed. (c) and (d) In the topographic and 28 I. Schmitz, M. Schreiner, G. Friedbacher and M. Grasserbauer, phase images after 180 and 2.800 min, respectively, the compact Anal. Chem., 1997, 69, 1012. surface layer formed and single corrosion products of a diameter of 29 I. Schmitz, M. Schreiner, G. Friedbacher and M. Grasserbauer, 70–150 nm and a height of 18–35 nm can be seen. Appl. Surf. Sci., 1997, 115, 190. 30 D. E. Newbury, Scanning, 1979, 3, 110. 31 M. Schreiner, G. Stingeder and M. Grasserbauer, Fresenius’ Z. features show a round shape [Fig. 13(c)] and are growing with Anal. Chem., 1984, 319, 600. exposure time [Fig. 13(d)]. 32 M. Schreiner, M. Grasserbauer and P. March, Fresenius’ Z. Anal. Concluding these investigations it can be stated that the Chem., 1988, 331, 428. 33 M. Schreiner, J. Am. Ceram. Soc., 1989, 72, 1713. weathering process of pure copper and probably also of copper alloys leads to the formation of a surface layer, which covers Paper 8/07305H the entire material within the first 60–90 min of exposure, J. Anal. At. Spectrom., 1999, 14, 395–403 403

ISSN:0267-9477    

DOI:10.1039/a807305h

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Behaviors of small molten metal islands on several substrates† T. Ichinokawa,a H. Itoha and Y. Sakaib aDepartment of Applied Physics, Waseda University, 3–4–1, Ohkubo, Shinjuku, Tokyo 169, Japan bJEOL Ltd., 3–1–2 Musashino, Akishima, Tokyo 196, Japan Received 28th August 1998, Accepted 8th December 1998 Electro- and thermomigration of metallic islands of mm size produced by vapor deposition on the Si(100)2×1 surface were investigated by ultra-high-vacuum scanning electron microscopy at substrate temperatures higher than the melting-points of islands.The direction of electromigration due to electric current passing through the Si substrate depends on the type of metal, whereas the direction of thermomigration caused by a temperature gradient on the substrate surface is always from low to high temperature, independent of the type of metal. The speeds are 0.1–10 mm s-1 and are approximately proportional to the island radius and increase exponentially with temperature in both cases.The driving forces of the island migrations are explained by the diVusion theory of metals in Si due to the electric field or the thermal gradient. Furthermore, it was found that the melting-points of metal islands on the carbon substrate are lower than those of the bulk and, moreover, the wettability (the contact angle) of the molten Cu or Ag islands changes in an oscillatory manner on the SiO2 substrate with periods of 100–0.1 s depending on the island diameter and substrate temperature. The origin of the contact angle oscillation is explained by the periodic change of the interface profile between island and substrate and by the failure of Young’s equation due to the change of orientations of surface and interface tensions.sequently annealed at 600 °C in UHV. The substrate tempera- 1. Introduction ture was measured by the resistance of the substrate, which The interface properties of metal films deposited on Si or SiO2 had been calibrated using an infrared pyrometer.Various substrates with increasing temperature are important not only metal films of Au, Ni, Pd, Ag, In, Cu and Al were deposited for the fabrication of electronic devices in the semiconductor on the substrates at room temperature from a heated tungsten industry, but also for the investigation of metal–ceramic wire basket and the thickness was measured with a quartz interfaces. Moreover, the epitaxial growth of metal films on thickness monitor.The deposited specimens were transferred Si or SiO2 crystals is very complicated, depending on the type from a UHV specimen preparation chamber to the UHVof metal. A number of studies have been carried out for SEM system through a transfer tube. SEM observations were systems of metals on Si and SiO2. carried out at a primary electron energy of 10 keV. Dynamic During our investigation of metallic films on Si or SiO2 motions of islands on the substrates were observed by TV crystals by ultra-high-vacuum scanning electron microscopy scanning at temperatures around the melting-points of islands (UHV-SEM) as a function of temperature, we found several with resistive heating and were stored on a video tape.interesting phenomena for liquid and quasi-liquid metal islands. In the present experiments, several phenomena, e.g., 3. Results (1) electro- and thermomigration of liquid metal islands on Si substrates,1,2 (2) rotation of facets of Au quasi-liquid islands 3.1.Electro- and thermomigration1,2 on graphite at temperatures lower than the bulk melting-point Fig. 1(a) and (b) are UHV-SEM images showing the electro- and (3) contact angle oscillation of the small molten metal and thermomigration of Au islands observed by passing direct islands on SiO2,3 have been observed in real time. The results current through the Si substrate. The larger the island, the and preliminary discussions on these phenomena are presented higher is the speed.For Au islands, the direction of the in this paper. electromigration is opposite to that of the electric field. The velocity increases exponentially with temperature and approxi- 2. Experimental mately proportionally to the island radius. Palladium islands also migrate in the same direction as the Au islands, but Al The experiments were carried out using a JAMP-30 UHVislands migrate in the opposite direction to the Au islands. SEM system. A p-type silicon wafer with a resistivity of Ag, Cu and Ni islands do not migrate.On heating with 4–6 V cm and a thermally oxidized Si (100) wafer of size alternating current, no electromigration was observed. The 20×5×0.4 mm were used as substrates. The substrate crystals eVect of the gravity is negligible and the level of the trace left were held between tantalum electrodes and heated by passing behind after migration is lower than that of the surrounding direct current through the substrate.The surface of the Si substrate surface. The direction of the electromigration is crystals was cleaned by flashing above 1000 °C for a few independent of the type of Si substrate (p- or n-type). minutes at a vacuum pressure of <5×10-10 Torr. The oxide The islands also migrate owing to the temperature gradient surface was cleaned by Ar+ ion bombardment and subon the substrate surface from low to high temperature, independent of the type of metal. The velocity of the thermomigration for Au islands is 0.1–1.2 mm s-1 depending on the †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998.island radius and substrate temperature. For thermomigration, J. Anal. At. Spectrom., 1999, 14, 405–408 405migrations of metallic islands of mm size, we consider that electro- and thermomigrations are caused by diVusion of metal atoms into the Si substrate across a solid–liquid interface caused by an electric field or temperature gradient.The diVusion velocity v of the migrating atoms depends on the concentration gradient, the temperature gradient and the electric field, as shown by the following equation using Fick’s diVusion equation: v=-D d ln c dx - DQ*dT kT dx +BeZ*E (1) where D is the diVusion coeYcient, c is the concentration of metal atoms, Q* is an ‘apparent heat of transport’ for a Fig. 1 (a) Electromigration and (b) thermomigration of Au islands migrating atom as given by Davies,4 k is Boltzmann’s constant, on Si(100).B is mobility, Z* is the ‘eVective charge’ of the migrating atom as given by Huntington5 and E is the electric field. From our experiments, it can be seen that Q* is negative for every type of metal used because all islands migrate from the cold to the hot side. The interpretation of thermomigration was described by JaVe and Shewmon6 for several impurity atoms in Cu, Au and Ag metals, and it was reported that almost all impurity atoms migrate from the cold to the hot side at lower speeds than in Si.For electromigration, there are two sources as the driving force: the first arises from the direct Fig. 2 Low temperature melting of a Cu film deposited on graphite (at 1000 °C). it was also found that the larger the island, the faster is the speed. Furthermore, we observed island migration due to an electron beam. If an electron beam of 0.1 mm diameter with a current density of 106 A cm-2 is scanned slowly near an island, the island moves, following the electron beam.Hence, we can write a letter of mm size with an electron beam. Such an eVect is regarded as an island migration due to the temperature gradient, because all islands migrate towards the electron beam Fig. 3 Rotation of the facets of an Au island on a carbon substrate independently of the type of metal. (side view) at temperatures lower than the melting-point of the bulk Au.Although the phenomena observed in this experiment are Fig. 4 Contact angle change in a period of the oscillation for a Cu island of 20 mm diameter on an SiO2 surface. The maximum contact angle is 120° and the minimum is 30°. 406 J. Anal. At. Spectrom., 1999, 14, 405–408action of the external field on the charge of the migrating ion 20 mm diameter gradually spreads and then suddenly contracts in an oscillatory manner in a period of 200 s at 1100 °C. The (‘direct force’)7 and the second from the momentum transfer due to scattering of conduction electrons by migrating atoms oscillation period changes from 200 to 0.1 s depending on the island diameter and substrate temperature.The higher the (‘wind force’).5 For impurity atoms in metals, the theoretical estimation of the eVective force for electromigration is diYcult. temperature, the shorter is the oscillation period, and the larger the island, the longer is the oscillation period. The In semiconductors, however, the ‘direct force’ is of great importance, because the electric field in a semiconductor is oscillation phenomena were investigated experimentally as a function of the type of metal and substrate material in order greater than that in a metal.Therefore, the charge transfer from Si to migrating atoms, which is probably deduced from to reveal the mechanism of the contact angle oscillation. Similar phenomena were observed in Al–Al2O3 and Ag–SiO2 the relative value of the electronegativity between metal and Si atoms, is an important factor for estimating the direction systems.During the contact angle oscillation, concentric depression of electromigration. In fact, the opposite directions of electromigration of Au and Al are interpreted in terms of electronega- rings of 50 nm depth around an island were formed on the substrate surface, because the island diameter just before the tivity, because the electronegativity of Au is larger than that of Si and that of Al is smaller than that of Si.Examples of contraction in each period decreases by thermal evaporation and the interface between the molten island and SiO2 substrate the electronegativities of relevant elements decrease in the following order:8 Au (2.4)>Pd (2.2)>Ag (1.9)>Cu moves downwards during spreading, as shown in Fig. 6. The geometric change of the interface leads to a change in the (1.9)>Ni (1.8)>Si (1.8)>Al(1.5)>Mg (1.2).This order agrees well with the results of the present experiments. The orientations of the liquid and interface tensions and the failure of Young’s equation for the force components between out- driving force acting to an island is a sum of ‘direct forces’ acting on an individual ion and is proportional to r2, because ward and inward tensions induces the spread and contraction the eVective volume relating to the island migration is probably underneath the substrate surface, whereas a reaction force acting on the island is a surface tension proportional to r.Thus, the size eVect on the island migration has been explained. The work function is one of the factors that can explain the electromigration of islands. However, the values of work functions are irregular and cannot explain the directions of electromigration reasonably. 3.2. Rotation of facets of metallic islands at temperature lower than the bulk melting-point Fig. 2 shows a UHV-SEM image of island formation for a Cu deposited film several hundred nanometers thick on the carbon substrate at 1000 °C (the bulk melting-point of Cu is 1083 °C).The Cu film becomes quasi-liquid at temperatures lower than 1000 °C and forms islands caused by surface energy minimization through the liquid-like flow. For Cu on a graphite substrate, the island shape is almost hemispherical and the contact angle is larger than 120°. The Au islands have facets of low crystallographic indices, as shown in Fig. 3, and the facets move at temperatures several degrees lower than the bulk melting-point (1064 °C). The facets disappear at the bulk melting-point. It should be noted that the quasi-melting-points of deposited metal films are several tens of degrees lower than that of the bulk and the crystallographic orientation of liquidlike islands moves at temperatures lower than the bulk melting-point. 3.3. Oscillation of wettability of liquid Cu islands on SiO2 3 The oscillation of the wettability of molten Cu islands of several mm diameter on amorphous or single-crystal SiO2 was observed by UHV-SEM. Fig. 4 shows the change in the island Fig. 5 Scanning electron microscope images of a Cu island on an SiO2 crystal, (a) before and (b) after milling by a focused ion beam. shape with the period of oscillation. A Cu molten island of rSV = rLS + rLV cos q rSV = rLS cos a¢¢ + rLV cos q¢¢ rSV > rLS cos a¢ + rLV cos q¢ rSV = rSL + rLV cos q (a) (b) (c) (d) a¢ Fig. 6 Interpretation of the contact angle oscillation due to a failure of Young’s equation for horizontal components of outward and inward tensions. J. Anal. At. Spectrom., 1999, 14, 405–408 407motions of the liquid metal island. The cross-section of the substrates, were observed by UHV-SEM. To analyze the interface properties of a micro-area as a function of tempera- interface profile between island and substrate was obtained ture, it is suggested that three-dimensional analysis by using a with a focused Ga+ ion beam as shown in Fig. 5 and it was combination system of a focused ion beam and UHV-SEM is proved that the interface level is lower than that of the promising. substrate surface. Fig. 6 provides an explanation of the contact angle oscillation for a molten island taking into account the surface and interface tensions according to Young’s equation. References The shift of the interface level is probably caused by diVusion. 1 T. Ichinokawa, H. Izumi, C. Haginoya and H. Itoh, Phys. Rev. B, From the experimental fact that the contact angle returns to 1993, 47, 9654. the initial value after the contraction, we can see that the 2 T. Ichinokawa, C. Haginoya, D. Inoue and J. Kirschner, Jpn. substrate surface is not contaminated by Cu. Further analysis J. Appl. Phys., 1993, 32, 1379. of the interface property, however, should be performed to 3 M. Ohya, D. Inoue, H. Itoh and T. Ichinokawa, Surf. Sci., 1996, 369, 169. clarify the mechanism on the contact angle oscillation. 4 R. O. Davies, Rep. Prog. Phys., 1956, 19, 327. 5 H. B. Huntington, in DiVusion in Solids—Recent Development, ed. A. S. Nowick and J. J. Burton, Academic Press, New York, 1975, Conclusion ch. 6. 6 D.JaVe and P. G. Shewmon, Acta Metall., 1964, 12, 515. Several interesting phenomena of liquid metal islands on Si or 7 A. H. Verbruggen, IBM Res. Dev., 1966, 32, 93. SiO2 substrates, e.g., (1) electro- and thermomigration of 8 L. Pauling The Nature of the Chemical Bond, Cornell University liquid metal islands on an Si substrate, (2) rotation of facets Press, Ithaca, NY, 3rd edn., 1960. of quasi-liquid metal islands at temperatures lower the bulk, and (3) contact angle oscillation of liquid Cu islands on SiO2 Paper 8/06746E 408 J. Anal. At. Spectrom., 1999, 14, 405–408

ISSN:0267-9477    

DOI:10.1039/a806746e

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

New observation method for divergent beam X-ray diVraction patterns† Siegfried Da�britz, Enrico Langer and Wolfgang HauVe Institut fu�r Oberfla�chenphysik und Mikrostrukturphysik, Technische Universita�t Dresden, D-01062 Dresden, Germany Received 4th September 1998, Accepted 4th January 1999 The simultaneous observation of X-ray reflections by a high-quality charge coupled device (CCD) camera in a scanning electron microscope is presented. The possibility of immediate further processing and evaluation of the images by computer, avoiding the extensive photographic X-ray film procedure, is discussed.The divergent beam X-ray method has considerable importance for investigations in materials research. The experimental set-up is described and the advantageous application of the camera is demonstrated for diVerent examples. By means of divergent beam X-ray diVraction (pseudo-Kossel and computer techniques, there is the possibility of recording these faint reflections with a high resolution and time integrated technique), it is possible to obtain information on the crystal microvolume in which the interferences are generated.The CCD video camera. The main advantage is that the diagrams can be recorded and then processed immediately by a com- most important parameters which one can obtain with this method are crystal orientations, the lattice constants and the puter. With this technique, the tedious and time consuming work with the photographic film technique in a darkroom and real structure parameters of the crystal.Although the reflection/ background ratio is between 60 and 90%, one uses film the subsequent reading of the data on film information into the computer for the determination of physical parameters material for the detection of the X-ray interferences because of the relatively long exposure time (5–20 min) for taking is eliminated and, furthermore, it is possible to compare the reflections by means of a computer simulation program X-ray reflections.Only a few reports have dealt with the topic of using solid-state detectors instead of X-ray films for the developed by the authors. In this paper, the advantage of the CCD camera over the detection of Kossel lines excited by sychrotron radiation and as far as we know there are no reports on the detection of X-ray film method is demonstrated for diVerent materials. electron-induced Kossel lines and pseudo-Kossel lines.The Kossel line profiles have been investigated by means of solid- Experimental state detectors.1,2 Image plates proved very useful only for Experiment assembly alignment purposes, since the quality of the images was not suYcient in comparison with X-ray film.3 The X-ray interferences are produced in a scanning electron With the development of new high-resolution CCD sensors microscope (CamScan CS44) with a supplementary device that we have developed. Fig. 1 and 2 show the experimental assembly for the generation of the lattice source and the divergent beam X-ray interferences (Kossel and pseudo-Kossel †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998.technique) with the registration of the reflections in the X-ray Fig. 1 Schematic representation of X-ray interference methods using the CamScan CS44 scanning electron microscope. EP, electron probe; T, target; TH, target holder; S, specimen; SH, specimen holder; and FD, film detector.J. Anal. At. Spectrom., 1999, 14, 409–412 409film detector and of the same reflections with the CCD camera structures; 2/3 in sensor, 1280 (H)×1024 (V) pixels, pixel size 6.7×6.7 mm; fast shutter, 100 ns–1 ms; long exposure, for the divergent beam X-ray interferences (DBI). The X-ray interferences originating from the crystal generate light emis- 1 ms–1000 s; maximum transfer rates of the data, 132 Mbyte s-1; ‘serial high speed’ data transfer with PCI interface sion on a fluorescent phosphor screen, which has a very low intensity and must therefore be viewed by an extremly sensitive board and plug-in board (PCI local bus), utility software and cache memory chips and FOL-10 fibre-glass-cable (Fiber Optic camera with the possibility of long integration times. We use a SensiCam CCD camera, which allows exposure times Link) and objective, TEC-M55 telecentric lens, focal length: 55 mm.between 100 ns and 1000 s and shows maximum spectral response between wavelengths of 350 and 550 nm. The camera can be moved to the optimum distance from the fluorescent Computer. A Pentium II PC, 266 MHz, 64 Mbyte RAM screen, which was produced and optimized by ourselves, by was used with a 15 in color TFT active matrix liquid crystal means of diaphragm bellows. Subsequently the signals are display. passed simultaneously to the computer for evaluation. Results and discussion Divergent beam X-ray interferences The eYciency and the advantages of recording the DBI with The principle of divergent beam X-ray interferences, i.e.interthe CCD camera are demonstated on diVerent examples. ferences by the pseudo-Kossel technique, is that the excitation Fig. 3 shows the X-ray film pattern of Si [100] and the Fe source is outside the crystal lattice to be investigated. An [110] single crystals and of a BaTiO3 ceramic crystal on the electron beam of energy E0=30 keV is focused on a target left and the corresponding pattern from the CCD camera on (foil ) and excites the atoms of a metal foil (in this case Fe) to the right.The application of the DBI technique for the emit their characteristic X-rays. This radiation is diVracted on investigation of ceramics (in this case a polycrystalline the lattice planes of the crystal lattice of the specimen under specimen) was reported recently for the first time.8 The high the foil, as illustrated in Fig. 1, and interferes according quality of the CCD images on the right in Fig. 3 was achieved to the Bragg equation. Following the Laue conditions, the through background subtraction. interferences generate cone surfaces.4–8 Distances between In this comparison it is seen that the quality of the film the target and the specimen down to 0.01 mm can be attained method is unsurpassed. Because of the very high resolution and was 0.3 mm in this case. of the CCD camera, it is nevertheless possible to visualize most of the details of the reflection.This concerns not only Device data the Ka1,2 doublet splitting and the b-reflections of the X-rays [(202) and (011) reflections, respectively, of the Fe single SensiCam CCD camera. The specifications are as follows: type, Super-VGA, long exposure, black/white, 370XL; high crystal pattern in Fig. 3], but also details of the real structure of the crystals. From the pattern, the orientations and lattice resolution two-stage Peltier cooled, digital 12 bit CCD camera for the detection of extremely faint signals and extremely fine constants of the crystals and statements about the reflection Fig. 2 Schematic diagram of the new principle for the detection of divergent beam X-ray interferences with the high resolution CCD camera.Table 1 Crystal parameters Single crystal Orientation Lattice constants/nm Dislocation density/cm-2 Si cubic [100] 0.5431 ~106 Fe cubic [110] 0.2866 ~107 BaTiO3 tetragonal [001] 0.3994, 0.3994, 0.4038 ~109 410 J.Anal. At. Spectrom., 1999, 14, 409–412Fig. 3 Comparison of divergent beam X-ray patterns observed by means of ordinary X-ray film on the left and a high resolution CCD camera on the right for diVerent single crystals (the corresponding exposure times are given beside the pattern). The CCD camera pattern was modified by image processing and the high quality of these images were obtained through specific background subtraction. sections and the real structure (Table 1) are readily determin- powerful analytical tool that will be useful for many types of studies in the field of physics and materials science.able, e.g., with the computer simulation program KOPSKO.9 The precision for the lattice constants is Da/a=10-4–10-5 and for the crystallographic directions DQ=0.5°. Acknowledgement Conclusions We are grateful to thhe Forschungsgemeinschaft For silicon, iron and barium titanate it has been shown that (DFG) for support (contract No.Da 407/1–1). the divergent beam X-ray interferences can be recorded with a CCD camera and the information processed immediately by a computer for evaluation. The CCD camera is very sensitive References to the structure of the reflections and was advantageously 1 T. Cog, D. Bahr and G. Materlik, Phys. Rev. B, 1995, 51, 6761. established for the rapid recording of the reflections, e.g., the 2 T. Takahashi and M. Takahasi, Jpn. J. Appl. Phys., 1993, 32, 5159. exposure time was only 2 min for the Fe single crystal. The 3 Ch. Schetlich, S. Brenner and V. Geist, J. Synchrotron Radiat., precisions of the lattice constants and crystal orientations 1998, 5, 102. achieved by means of the CCD camera are comparable with 4 T. Ellis, F. Nanni, A. Shrier, S. Weissmann, G. E. Padawer and N. results obtained using X-ray film detectors. Hence the new Hosokawa, J. Appl. Phys., 1964, 35, 11. 5 D. J. Dingley and N. Razavizadeh, in Scanning Electron observation method for DBI changes the technique to a J. Anal. At. Spectrom., 1999, 14, 409–412 411Microscopy, ed. O. Johari, AME O. I. L. Hare, Illinois, 1981, 8 E. Langer, S. Da� britz, A. Ro�der and W. HauVe, Fresenius’ J. Anal. Chem, 1999, in the press. vol. IV, p. 287. 6 S. Da� britz, H. Horn, K. Kleinstu� ck, H. Waltinger and V. 9 E. Langer, R. Kurt and S. Da�britz, unpublished work. HoVmann, Cryst. Res. Technol., 1986, 12, 1531. 7 S.Da� britz, E. Langer and W. HauVe, Fresenius’ J. Anal. Chem., Paper 8/06922K 1997, 358, 148. 412 J. Anal. At. Spectrom., 1999, 14, 4

ISSN:0267-9477    

DOI:10.1039/a806922k

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Thin film X-ray microanalysis with the analytical electron microscope† Aldo Armigliato CNR-Istituto LAMEL , Via P.Gobetti, 101, I-40129 Bologna, Italy Received 28th August 1998, Accepted 29th September 1998 X-ray microanalysis, when performed in an analytical electron microscope (AEM) equipped with an energy dispersive spectrometer (EDS), is a powerful technique for determining chemical composition at high spatial resolution.The basic aspects of the generation and detection of X-rays will be reviewed and the problem of quantification of thin films (composition and thickness) will be discussed. 1. X-ray generation 2. Energy dispersive spectrometers The energy dispersive spectrometer (EDS) is practically the The physical factors which diVerentiate the generation of X-rays by electrons from that obtained by using X-rays as a only one employed in modern AEMs. Wavelength dispersive spectrometers (WDS), which were used to some extent in the primary source are as follows.Firstly, ionisation cross section, which gives the number of ionisations per atom which can be past, have been abandoned, despite their inherent better resolution and ability to detect light elements; this is due to their generated by an electron of energy E0:1 lower collection eYciency and stringent focusing requirements, Qi (cm2)=6.51×10-20nsbs[ ln(csT/Ec)-ln(1-b2)-b2]/EcT which implies a higher mechanical complexity. (E0>50 keV) (1) In general, the detector consists of a 30 mm2 Si(Li) device, with an entrance window consisting of a thick (7–12 mm) Be where ns is the number of electrons in the shell s (K,L,M...), film, which absorbs photons lighter than the Na K, or of a bs and cs are constants of the shell, b=v/c, T is the kinetic UTW (ultra thin window), which allows ultrasoft X-rays like energy of the electron of velocity v and Ec is the critical energy B K to be detected. These windows are generally made from of ionisation (also called absorption edge).Secondly, thin Al-coated polymer films or from diamond (or BN) thin bremsstrahlung cross section. The distribution is anisotropic, layers; all of them can withstand the atmospheric pressure and i.e., the emission depends on the angle h between the direction for this reason are also called ATW (atmospheric thin win- of the emitted photon and the incident direction x. The dows). If the sample is not cathodoluminescent, the window components Qx, Qy and Qz are given by:2,3 can be even removed, but an ultra-high-vacuum AEM should Qx=4.5×10-25 Z2{0.252+a[(Eb/E0)-0.135]-b[(Eb/E0) be used.The best energy resolution so far reported is 127 eV -0.135]2}/E0 (2a) (Mn Ka peak). With Si(Li) detectors, X-rays with an energy higher than a=1.47B-0.507A-0.833, b=1.70B-1.09A-0.627, 20 keV are not eYciently detected. This limitation can be A= exp(-0.223E0/0.2998Z2)-exp(-57E0/0.2998Z2), overcome by the high-purity Ge detectors (HPGe), which can B=exp(-0.828E0/0.2998Z2)-exp(-84.9E0/0.2998Z2) detect K lines of heavy elements [energies higher than E(Mo Ka)=17.5 keV ].In this way it is possible to analyse, with Qy=Qz=4.5×10-25 Z2{-j+k/[(Eb/E0)+h]}/E0 (2b) much higher accuracy, elements whose L lines strongly interh=(- 0.214y1+1.21y2-y3)/(1.43y1-2.43y2+y3); fere with the lines generated from other elements. Intermediate j=(1+2h)y2-2(1+h)y3; k=(1+h)( y3+j); voltage TEMs (300–400 keV) are particularly suitable for y1=0.220[1–0.390 exp(-26.9E0/0.2998Z2)] these applications.Moreover, the HPGe detector has a better y2=0.067+0.023/[(E0/0.2998Z2)+0.75] energy resolution than the Si(Li) one (an ultimate value of y3=-0.00259+0.00776/[(E0/0.2998Z2)+0.116 ] 114 eV for the Mn Ka peak, generated from a 55Fe source, has been obtained4). where Eb is the energy of the generated photon and Qy=Qz The measurement of X-ray spectra using EDS in a TEM due to radial symmetry. The units of Q are (keV photon poses the problem of the control of spurious radiations, which energy/keV energy interval per steradian per at cm-2).Thirdly, are mainly related to the illumination system and to the electron backscattering. Not all electrons of the incident beam specimen itself. An example of the latter eVect is the specimencomplete their path inside the sample (and eventually stop in generated bremsstrahlung, which is strongly anisotropic (see it). A fraction of them exits the specimen surface after a section 1); its intensity is given by:5 number of elastic or inelastic events (or even is reflected by a single backscattering process), becoming unavailable to ionise IB=GQx(E) sin2(h) (1-bcosh)4 the atoms of the sample.However, this phenomenon is signifi- cant in bulk specimens only, being generally negligible in thin films. +Qy(E) C1+ cos2(h) (1-bcosh)4DHNrtDEV/AE (3) where Qx and Qy (=Qz) are given by eqn. 2, h is the angle with respect to the incident beam direction, Nrt/A is the atom †Presented at the Fifteenth International Congress on X-ray Optics and Microanalysis (ICXOM), Antwerp, Belgium, August 24–27, 1998.concentration (in at cm-2) and E±DE is the energy interval J. Anal. At. Spectrom., 1999, 14, 413–418 413of the continuum photons considered. Eqn. 3 shows that the a factor, generally called K-factor, which does not depend on interaction of the sample with its own bremsstrahlung increases A and B only, but also on the specific AEM/EDS system: with the tilting angle, so self fluorescence may occur at points IA/IB=[(Qvp)A/(Qvp)B](eA/eB)CA/CB=(1/KAB)CA/CB well away from the region of interest; moreover, the intensity (6) increases also with sample thickness t and atomic number Z.For this reason, operation at zero tilt is generally rec- where KAB=[(Qvp)B/(Qvp)A](eB/eA) and is often termed the ommended;6 however, we will show in section 4.3.2 that ‘CliV–Lorimer factor’, as these authors were the first to deterquantitative analysis of thin cross sections of silicides on mine it experimentally.In this equation, v is the fluorescence silicon can be performed by a ‘2-tilt angle method’ with enough yield, p is the weight of the line and e is the detector eYciency. accuracy. The term e depends on the particular AEM/EDS combination In any case, experimental procedures and test specimens used. For this reason, although many compilations of KAB can be used to identify the source of the problem of spurious factors are available in the literature for practically all the X-rays, through the measurement of the peak5background elements that can be analysed by EDS, it should be checked ratio of specific X-ray peaks (‘P/B criterion’); this criterion carefully if the employed detection system was the same as in can be used to check the validity of the spectra obtained from one’s own experiment.12 To determine the K factors, three a thin specimen in a TEM/EDS combination.These test approaches are generally used, i.e., (i ) using a number of specimens can be Co films7 which, using an energy window of binary alloy standards (experimental ),11 (ii ) using the formulae DE=360 eV on the Co Ka peak and DE=20 eV on the for Q, v, p and e (computational ), or (iii ) by an extrapolation background, should yield a P/B value of 2050 at 100 kV, or based on pure elements.13 Cr films8 with windows of DE=700 eV and 10 eV for the Cr If absorption (and/or secondary fluorescence) cannot be Ka peak and the background, respectively; values of P/B of neglected, the local sample thickness must be known or 4000 and 6000 at 100 and 300 kV, respectively, have been measured.It can be assumed that X-ray absorption is negligible reported. if xrt<0.1,14 where x=(m/r)sLcosecy and (m/r)sL is the mass absorption coeYcient for X-rays of the lightest element L in 3. Quantitative analysis the sample s. If the thin film approximation does not hold, the absorption correction can be calculated by modifying the After subtraction of the background, the net peak intensities K-factor (eqn. 6) according to Goldstein et al.15 of the X-ray lines present in a spectrum from a thin film can be used for quantitative microanalysis. The background can K*AB= be mathematically modelled starting from the Kramers law:9 KAB[(m/r)sA/(m/r)sB][1-exp(-xBrt)]/[1- exp(-xArt)] (7) NE=KZav(E0-E)/E (4) where NE is the number of continuum X-rays generated in the where a normal incidence of the electron beam is assumed. energy range E to (E+DE) by an electron of energy E0, The uncertainty in the values of the mass absorption impinging onto a sample of average atomic number Zav.The coeYcients is still surprisingly large for X-rays with energies key step of the model consists in modifying the constant below 1 keV, despite the eVorts spent for many decades of this (actually not a constant) K by including terms related to century to find the best coeYcients in a wide range of photon absorption both in the specimen and in the detector.This has energies. Hubbell16 has recently shown that the estimated been performed by Ware and Reed,10 who proposed the uncertainties (within 1 s) of the photoionisation cross sections formula in a solid sample are of 100–200% for photon energies 100– 500 eV, and of 10–20% for energies in the range 0.5–1 keV. B(E)=K{(E0-E)(1/E)f(x) exp[-Si(m/r)irixi]+F(E)} Therefore, if light elements are present in the sample, the (5) absorption correction should be kept small by choosing a thin where f(x) is the absorption in the sample [x=(m/r)cosecy, y area of the sample (provided, of course, enough counts can being the angle between the direction of detection and the be accumulated in the spectrum for good statistics before sample surface], the exponential term corrects for absorption instability problems arise); on the other hand, K lines should of X-rays in the window (if any), gold contact and Si dead be preferred to L lines when possible.layer of the detector, and F(E) is an empirical correction term. Secondary fluorescence is usually a minor eVect; however, Another process for background subtraction is digital filtering, in case it limits the accuracy of quantification, to correct for a merely mathematical approach that does not involve the secondary fluorescence of an element A, the ratio fA=I*A/IA physics of X-ray production and detection.It is based on the of the X-ray intensities due to secondary and primary genersmoothness and slow variation of bremsstrahlung with energy, ation must be evaluated. The most popular treatment is the in contrast with characteristic peaks which vary much faster. one by Nockolds et al.17 The method filters the experimental spectrum employing a ‘top hat’ digital filter, which, through its convolution with the fA=CBvB[(rA-1)/rA](AA/AB) (m/r)BA(UBlnUB/UAlnUA) X-ray spectrum, produces a second-diVerence spectrum ×(rt/2)|0.923-ln(m/r)ABBrtseca| (8) (d2I/dE2); in this way it transforms the background into a where r is the jump ratio and a is the tilt angle.Finally, the horizontal line of value zero (small dI/dE), yet leaving the concentration of the two elements A, B in the binary alloy is: characteristic peak structure ( large dI/dE). After having subtracted the background, the peak intensities CB/CA=K*BA(IB/IA)(1+fA) (9) can be obtained by fitting either to a Gaussian profile or to where K*BA(=1/K*AB) is the CliV–Lorimer factor corrected library standard peaks, stored on disk or tape.A multiple for absorption. least-squares fitting procedure allows the determination of the From the above discussion it emerges that, if the thin film intensity values through calculation of the parameters that fit approximation is not fulfilled, it is impossible to determine the the library peaks to the experimental ones.sample composition without a knowledge of the sample thick- The most popular method of quantification was proposed ness. However, it is possible to determine simultaneously the by CliV and Lorimer;11 it states that, if the X-ray absorption film composition and thickness from the X-ray intensities. in the sample is negligible (thin film approximation), the ratio Among the proposed methods, the following have been more of the peak intensities IA/IB in a binary alloy is related to the ratio of the concentrations of the two elements CA/CB through extensively used. 414 J. Anal. At. Spectrom., 1999, 14, 413–418silicon substrate and thin cross sections for the AEM were 4. Methods for the simultaneous determination of prepared. In the microanalysis experiments, the X-ray intensit- film composition and thickness ies of Ti Ka (or Ge Ka) and Si Ka lines have been measured and ratioed, giving rise to the values Rm=I(Si Ka)/I(Ti Ka) 4.1. The extrapolation (or parameterless) method18 (or I(Si Ka)/I(Ge Ka)).A Monte Carlo code (CARLONE) In a binary alloy AB the ratio of the concentrations CA/CB generates two sets of computed corresponding ratios of X-ray (not corrected for absorption), as obtained from the CliV– intensities Rc, for a=0° and 20°, as a function of Ti (or Ge) Lorimer formula, is plotted versus the intensity IA (or IB or concentration C and mass thickness rt. A procedure similar IA+IB), measured in areas of diVerent thickness of the sample.to the one by Kyser and Murata21 enables one to determine The extrapolation to zero intensity gives the corrected CA/CB those C and rt values which minimise the diVerence value. To apply this method, the beam current and the |Rci(C, rt)-Rmi| for i=0°, 20°. This method is codified into a spectrum acquisition time must be kept constant. second program (ROSIN) and needs only the knowledge of Q, v, p, e, m/r for the two elements. 4.2. The diVerential X-ray absorption method19,20 Examples of applications of the method are given in Figs. 1–5. A TiSi2 sample of known composition has been This method makes use of the diVerence in absorption between analysed in cross section in two points of diVerent thickness. K and L (or L and M) lines, emitted from the same element. In Fig. 1 the plot of rt versus Ti concentration, relative to a In this way, assuming a binary alloy AB, one can measure thin area of the sample, for the two tilt angles (0° and 20°) is three X-ray intensities, e.g., A Ka, A La and B Ka, and shown; the two curves intersect at the point of C(Ti)= compute the expressions 33.6 at% and rt=0.20 mg cm-2.The plot of Fig. 2 refers to CA/CB=KKAB[I(A Ka)/I(B Ka)]F(rt)K (10a) the thicker region of the silicide stripe, which yields C(Ti)= 33.9 at% and rt=0.48 mg cm-2. The Ti concentrations are in CA/CB=KLAB[I(A La)/I(B Ka)]F(rt)L (10b) very good agreement with that one expected for a stoichio- CA/CA=KLKAA[I(A Ka)/I(A La)]F(rt)LK=1 (10c) metric TiSi2 phase (33.3 at%).Of course, the rt values refer to the local thickness of the AEM cross section, a parameter where the usual K-factors are included and F(rt) represents the absorption correction. In this treatment we have two unknowns, i.e., one concentration and rt, and two equations (among the 3 in the above expressions) so, in principle, the problem is soluble. Actually, due to the presence of exponentials in F(rt), an iterative procedure is necessary.Of course this method can be applied only to alloys where at least one element has an L (or M) line with energy higher than 1 keV (Z30 or Z60) for detectors with Be windows, and 0.3 keV (Z20) for windowless detectors. In this latter case, the uncertainty in mass absorption coeYcients16 may negatively aVect the accuracy in the quantitative microanalysis. However, unlike the multigeometry method in section 4.3, the analysis can be performed by acquiring a single spectrum at zero tilt angle, which reduces the spurious X-rays.6 4.3.The multigeometry method This method can be based on an analytical or a Monte Carlo Fig. 1 Plot of mass thickness rt versus Ti concentration, as computed approach. by the ROSIN program, on the basis of the Monte Carlo generated intensity ratios for the two tilt angles (0° and 20°). The corresponding 4.3.1. Analytical. The X-ray intensities of the lines chosen X-ray spectra were taken in a thin area of the TiSi2/Si cross section.for each element are recorded at diVerent tilt angles on the The two curves intersect at a point where CTi=33.6 at% and rt= 0.20 mg cm-2 (adapted from ref. 23). same area of the sample to be analysed. In the case of a binary alloy AB, we have to solve the two equations: CA/CB=KAB(rt)(IA/IB)1 (11a) CA/CB=KAB(rt)(IA/IB)2 (11b) where the suYxes 1 and 2 refer to the two tilt angles. The unknowns are two, i.e., rt and one concentration, so the problem is soluble.In practice, a value of the mass thickness rt at zero tilt is chosen and the KAB(rt) factor is calculated; then an iteration procedure adjusts rt until the CA/CB values at each tilt converge to the same value. The use of three or more tilt angles can be used to improve the fit.19 4.3.2. Monte Carlo method. The simulation model generally adopts the single scattering approach, whereas the continuous slowing down approximation by Bethe gives the energy loss between scattering points.21,22 The computer code determines the intensity emitted from a specimen, whose composition and thickness are assumed to be known.Fig. 2 As Fig. 1 for a thicker (rt=0.48 mg cm-2) region of the sample. This method has been applied to the analysis of TiSi2 and The Ti concentration is close to that found in Fig. 1 (33.9 rather than Si1-xGex films at 300 keV at 2 tilt angles (0° and 20°, respect- 33.6 at%) and in good agreement with that in a stoichiometric TiSi2 compound (CTi=33.3 at%) (adapted from ref. 23). ively).23–25 In both cases, the samples were deposited onto a J. Anal. At. Spectrom., 1999, 14, 413–418 415of no practical significance (the thickness of the TiSi2 film can readily be deduced from any micrograph taken in the sample). In Figs. 3 and 4 is reported the case of two Si1-xGex/Si heterostructures, cross sectioned for the AEM. The Monte Carlo method yields C(Ge)=7.5 at% for Fig. 3 and C(Ge)= 13.4 at% for the case of Fig. 4. The same structures have been analysed also by other techniques, like Rutherford backscattering spectrometry and double crystal X-ray diVractometry. The values obtained are 7.7 at% for SIGE 1 and 13.1 at% for SIGE 2, again in good agreement with the X-ray microanalysis. The method can also be applied to the case of L lines. Referring again to the case of Si–Ge alloys, the X-ray spectrum contains the Ge L line, in addition to the Si K and the Ge K ones. It is possible to apply the ‘2-tilt angle method’ to both the intensity ratios I(Si K)5I(Ge K) and I(Si K)5I(Ge L); moreover, one can keep the angle of incidence constant (e.g., at 0°) and exploit all the three intensity values I(Ge L), I(Si K) and I(Ge K).In this case, the program ROSIN is used to determine the C and rt values that minimise the diVerences Fig. 5 Example of application of the 2-tilt angle method by including |RcK-RmK| and |RcL-RmL| (‘2-ratio method’26). In Fig. 5 the the Ge L line in the intensity ratio [i.e., I(Si K)/I(Ge L)].The sample is SIGE 1 (CGe=7.5 at%, see Fig. 3). In this case the method does not plot rt versus C(Ge) relative to sample SIGE 1 is reported; work, the two curves being independent of the local specimen thick- the ‘2-tilt angle method’, applied to the K:L ratio [i.e., I(Si ness; however, at normal incidence, the Ge concentration is the same K)5I(Ge L)] does not work, as the two curves, at 0° and 20° as the one deduced by using the K5K ratios.In fact, the intersection of tilt, do not intersect. Surprisingly, however, both these with the K5K curve (a=0°) occurs for the same values of CGe and rt curves are actually straight lines, with constant C(Ge)=7.49 found in Fig. 3. This is an application of the 2-ratio method (see text and 7.92 at%, respectively; this is due to the fact that the for details). The diVerence in Ge concentration between the two vertical straight lines is only 5.6% (relative) (adapted from ref. 26). absorption correction is negligible, as (m/r)Sis$(m/r)Ges for these Ge concentrations. In other words, in this specific case the simple CliV–Lorimer thin film equation (eqn. 6) can be applied. Moreover, the ‘2-ratio method’ for normal incidence, which includes the curve of the K:K ratio (i.e., the same as plotted with closed circles in Fig. 3), yields C(Ge)=7.5 at%, the same value deduced from the ‘2-tilt angle method’ applied to the K:K ratios. It is also in very good agreement with the corresponding Ge concentration deduced from the DXA method by Horita et al.20 at normal incidence; for a 20° tilt, also the DXA method does not converge.26 It is not quite clear why the two straight lines give diVerent Ge concentrations; on the other hand, DC/C=5.6% (relative), which compares favourably with the accuracy attainable by the thin Fig. 3 Plot of mass thickness versus Ge concentration in a cross film microanalysis technique. Surely, this discrepancy cannot sectioned Si1-xGex/Si heterostructure (SIGE 1).The resulting combe attributed to the self-background eVect, which should excite position CGe=7.5 at% (or x=0.075) agrees with the values found by other techniques. Si K photons from the Si regions surrounding the Si–Ge stripe, to an extent increasing with tilt angle; this would result in an apparent smaller Ge concentration, unlike the situation in Fig. 5. A final consideration on the X-ray microanalysis of Si–Ge alloys is the choice of the m/r values, among the many proposed in the literature. The tabulations by Veigele27 and Heinrich28 have been considered and it is assumed the values that yielded Ge concentrations in closest agreement with the ones obtained by other techniques are correct.26 The tables by Heinrich are also recommended by Williams and Carter6 (p. 613). 5. Spatial resolution and analytical sensitivity The beam diameter d is usually defined as the FWTM (Full Width at Tenth Maximum) of the Gaussian electron intensity, which corresponds to 90% of the incident beam.As the electron beam travels through the specimen, it undergoes a spreading, due to elastic scattering by the atom nuclei. The spatial resolution is thus a function of the sample, and the Fig. 4 Plot of mass thickness versus Ge concentration in sample SIGE most accurate way of determining it is again Monte Carlo 2, which has a higher CGe with respect to SIGE 1. The value calculation.However, an analytical expression has been pro- corresponding to the intersecting point (13.4 at%) agrees with the one deduced from other techniques (adapted from ref. 25). posed by Goldstein et al.15 for the beam broadening, b, which 416 J. Anal. At. Spectrom., 1999, 14, 413–418is still in widespread use: about 100 cps with a resolution of 13 eV at 4.5 keV have been obtained with a spectrometer mounted on a conventional b=625(Z/E0)(r/A)1/2t3/2 (12) SEM.35 Thus, the resolution typical of WDS spectrometers where b and t are in cm, r in g cm-3 and E0 in keV.The value seems to be achievable in the near future, but for T<4 K; of the constant has been subsequently modified by Reed29 to however the smaller size of these detectors, as compared to 721. The width of the broadened beam at the exit surface of the semiconductor ones, imposes a marked increase in the the specimen is Rex=(b2+d2)1/2. This is, however, the worst solid angle of detection, so allowing count rates much larger case, the width R at t/2 being more appropriate.Accordingly,30 than 100 cps in the AEM. R is assumed as the X-ray spatial resolution and is given by: The combination of microcalorimeters (or superconductorbased detectors) with aberration-free lenses in the AEM (if and when they become available) will improve the spatial R= d+Rex 2 = d+(b2+d2)1/2 2 (13) resolution of X-ray microanalysis, so providing single atom sensitivity.36 From eqn. 12 it is evident that an increase in beam energy E0 from 100 keV to 300 or 400 keV reduces the beam broadening. Together with a parallel increase in the gun brightness, this is References a clear advantage of the so-called intermediate voltage AEMs. The minimum detectability limit of thin film X-ray micro- 1 See, for example, R. F. Egerton, Electron Energy-Loss analysis can be expressed by means of two parameters: MMF, Spectroscopy in the Electron Microscope, Plenum Press, New York which is the Minimum Mass Fraction (in wt%) of an element and London, 2nd edn., 1996, p. 405. in a matrix, detectable in the presence of the background 2 P. J. Statham, X-ray Spectrom., 1976, 5, 154. 3 P. Kirkpatrick and L.Wiedmann, Phys. Rev., 1945, 67, 321. produced by the specimen and the system; and MDM, which 4 R. A. Sareen, in X-ray Spectrometry in Electron Beam Instruments, is the Minimum Detectable Mass (g) of the element in the ed. D. B. Williams, J. I. Goldstein and D. E.Newbury, Plenum analysed volume. The MMF value can be calculated starting Press, New York, 1995, ch. 4. from the criterion that a peak containing P counts is statisti- 5 N. J. Zaluzec, Proceedings of the 9th International Congress on cally significant if it is three times higher than the standard Electron Microscopy, Toronto, 1978, vol. I, p. 548. deviation of the corresponding background counts B: 6 D. B. Williams and C. B. Carter, in Transmission Electron Microscopy: A Textbook for Materials Science, Plenum Press, P3(2B)1/2 (14) New York and London, 1996, vol. 4, p. 580. 7 W. A. P. Nicholson, C. C. Gray, J. N. Chapman and B. W. In the case of a binary alloy AB, applying the CliV–Lorimer Robertson, J. Microsc., 1982, 125, 25. formula PA=PBKBACA/CB (eqn. 6), one gets: 8 S. M. Zemyan and D. B. Williams, J. Microsc., 1994, 174. 9 H. A. Kramers, Phil. Mag., 1923, 46, 836. (MMF)A=3CB(2BA)1/2/KBAPB (KBA=1/KAB) (15) 10 N. Ware and S. J. B. Reed, J. Phys. E, 1973, 6, 286.Assuming a current density of 100 A cm-2 in a spot of 10 nm, 11 G. CliV and G. W. Lorimer, J. Microsc., 1975, 103, 203. 12 A good example of K-factors compilation can be found in the a counting time of 300 s, an MMF~0.3 wt% has been reported paper by P. A. Sheridan, J. Electron Microsc. Technique, 1989, for elements of Z ranging from 15 to 40 in a silicon matrix 11, 41. 100 nm thick.31 13 K. Rajan, J. McCaVrey, P. B. Sewell, C. R. Leavens and G. The minimum detectable mass (MDM) can be expressed L’Esperance, in Intermediate Voltage Microscopy and its by:31 Application to Materials Science, ed.K. Rajan, Electron Optics Publishing Group, Philips Electronic Instruments, Inc., Mahwah, NJ, USA, 1987, p. 11. MDM= Imin JQvpteV /4p (16) 14 R. Tixier and J. Philibert, Proceedings of the 5th International Congress on X-ray Optics and Microanalysis, ed. G. Mo� llenstedt where, apart from the known symbols, Q is now expressed and K. H. Gaukler, Springer Verlag, Berlin, Germany, 1969, in cm2 g-1, J is the current density and t is the counting time.p. 180. In the above experimental conditions, values of the order of 15 J. I. Goldstein, J. L. Costley, G. W. Lorimer and S. J. B. Reed, in Scanning Electron Microscopy 1977, ed. O. Johari, IITRI 10-20 g can be achieved. More recently, Williams and Carter6 Publishers, Chicago, IL, USA, 1977, vol. I, p. 315. (p. 634), referring to the analysis of Cr in a stainless steel 16 J.H. Hubbell, Compilation of Photon Cross Sections: Some performed at 100 kV with a FEG-STEM by Lyman and Historical Remarks and Current Status, paper presented at Michael,32 have hypothesized that, if the experiment would be EDXRS98, Bologna, June, 1998, to be published in X-ray repeated at 300 kV, in a region of the sample 16 nm thick with Spectrom. J=5×104 A cm-2 and t=2000 s, the MMF would be 17 C. Nockolds, M. J. Nasir, G. CliV and G. W. Lorimer, in Electron Microscopy and Analysis 1979, ed.T. Mulvey, The Institute of ~0.03 wt%, but the MDM~10-22 g (i.e., 20 atoms). Physics, Bristol and London, 1980, p. 417. Preliminary data demonstrating the feasibility of analysing 18 E. Van Cappellen, Microsc. Microanal. Microstruct., 1990, 1, 1. volumes containing <10 atoms have been recently reported.33 19 P. L. Morris, M. D. Ball and P. J. Statham, in Electron Microscopy and Analysis, 1979, Inst. Phys. Conf. Ser., 1980, 52, 413. 20 Z. Horita, K. Ichitani, T.Sano and M. Nemoto, Phil. Mag., 1989, 6. Future prospects A59, 939. 21 D. F. Kyser and K. Murata, IBM J. Res. Dev., 1974, 18, 352. Nowadays, X-ray microanalysis is surely the most widely used 22 A. Armigliato, A. Desalvo, R. Rinaldi and R. Rosa, J. Phys. D, attachment of the AEM. It has not suVered from the expected 1979, 12, 1299. competition of (parallel ) energy loss spectrometry (PEELS), 23 A. Armigliato and R. Rosa, Ultramicroscopy, 1990, 32, 127. apart from the analysis of light elements, due to its much 24 A.Armigliato, D. Govoni, R. Balboni, S. Frabboni, M. Berti, F. worse P/B ratio and the more stringent experimental require- Romanato and A. V. Drigo, Mikrochim. Acta, 1994, 114/115, 175. ments of this latter technique (thinner specimens, lower con- 25 A. Armigliato, R. Balboni, F. Corticelli, S. Frabboni, F. Malvezzi and J. Vanhellemont, Mater. Sci. Technol., 1995, 11, 400. tamination rates). 26 A. Armigliato, T. Lewis and R. Rosa, Mikrochim. Acta [Suppl.], Little progress is to be expected on the resolution of Si(Li) 1996, 13, 241. and HPGe detectors for physical reasons. The real improve- 27 W. M. J. Veigele, At. Data, 1973, 5, 51. ment will come from the completely new detectors, such as 28 K. F. J. Heinrich, in Proceedings of the 11th Conference on X-ray the microcalorimeters and superconducting tunnel junctions; Optics and Microanalysis (ICXOM-11), ed. J. Brown and R. recently, many eVorts have been made to make these devices Packwood, University ofWestern Ontario, Canada, 1986, p. 67. 29 S. J. B. Reed, Ultramicroscopy, 1982, 7, 405. suitable for X-ray microanalysis in the SEM.34 Spectra at J. Anal. At. Spectrom., 1999, 14, 413–418 41730 J. R. Michael, D. B. Williams, D. B. Klein and R. Ayer, 34 J. Martinis, in Microbeam Analysis-1995, ed. E. Etz, VCH, New J. Microsc., 1990, 160, 41. York, USA, 1996, p. 3. 31 D. C. Joy and D. M. Maher, in Scanning Electron Microscopy 35 D. A. Wollman, G. C. Hilton, K. D. Irwin and J. M. Martinis, in 1977, ed. O. Johari, IITRI Publishers, Chicago, IL, USA, 1977, Proceedings of Microscopy and Microanalysis 1996, San Francisco vol. I, p. 325. Press, San Francisco, CA, USA, 1996, p. 488. 32 C. E. Lyman and J. R. Michael, in Analytical Electron Microscopy 36 J. M. Titchmarsh, Mikrochim. Acta (Suppl.), 1998, 15, 37. 1987, ed. D. C. Joy, San Francisco Press, San Francisco, CA, USA, 1987, p. 231. 33 C. E. Lyman, J. L. Goldstein, D. B. Williams, D. W. Ackland, S. Paper 8/06757K von Harrach, A. W. Nicholls and P. J. Statham, J. Microsc., 1994, 176, 85. 418 J. Anal. At. Spectrom., 1999, 14, 413–4

ISSN:0267-9477    

DOI:10.1039/a806757k

出版商:RSC    

年代:1999

数据来源: RSC