Tobacco Documents Online

National Bureau of Standards Special Publication 506. Proceedings of the Workshop on Asbestos: Definitions and Measurement Methods held at NBS, Gaithersburg, MD, July 18-20, 1977. (Issued November 1978) MINERAL FIBER IDENTIFICATION USING THE ANALYTICAL TRANSMISSION ELECTRON MICROSCOPE D. R. Beaman and H. J. Walker The Dow Chemical Company Midland, Michigan 48640 Abstract In a transmission electron microscope equipped with an energy dispersive spectrometer (EDS), It is possible to obtain the high resolu- tion morphology, crystal structure, and elemental composition of sub- micron mineral fibers, particulate, and thin films. The reliability of fiber analysis is enhanced when fiber identification is based on the nearly simultaneous determination of these three characteristics because each of the individual modes can yield ambiguous information. Energy dispersive spectrometer data can be converted to elemental fiber compositions using known standard spectra or relative sensitivity factors which can be calculated or experimentally determined for a given , instrumental configuration. Calculated and experimental sensitivity : factors are found to agree within 15 percent for photon energies above . 1.5 keV. The relative error in composition calculated from EDS spectra ; will generally be better than 10 percent, but only if the TEM column and ' components have been properly modified to reduce the effects of extraneous x-ray generation and electron scattering. The sources of these problems are described and a procedure for minimizing the effects outlined. Proper aperturing, collimation, selection of materials of construction, and operating conditions can provide useful mineral spectra. It is often necessary to correct for x-ray absorption even in fine mineral fibers, and this may be done using reference standards or sensitivity factors corrected for absorption. The effect of absorption increases rapidly as the difference between the mass-absorption coefficients of the elemental constituents of the mineral increases. Carbon contamination which degrades both EDS spectra and electron diffraction patterns can be minimized by using low current density and short analysis times. Less than 15 percent of the chrysotile fibrils in a standard provided positive selected area electron diffraction patterns (SAED), but up to 50 percent did have the correct layer line spacing. The fraction of fibers providing good diffraction increases rapidly as the number of fibrils in a fiber increases. The reported differences in SAED quality arise primarily because investigators use differing criterion for defining a positive SAED pattern and the fiber size distribution examined varies. Sample preparation methods were reviewed and it was found that condensation washing is only reliable if loss corrections are applied, particularly in the case of amphibole fibers. In spite of the many problems, inter-laboratory end multiple sample reproducibility in the measurement of fiber concentrations can be t30 percent when using good procedures. Key Words: Carbon contamination; electron diffraction; mineral fibers; transmission electron microscope; x-ray spectroscopy. 249

Introduction The need to identify and determine the concentration of small mineral fibers in environmental samples provided motivation for the development of the analytical trans- mission electron microscope (ATEM) which consists of a conventional transmission electron microscope (CTEM) equipped with energy dispersive spectroscopy (EDS) and possibly scanning transmission electron microscopy (STEM) capabilities. In such an instrument it is possible to obtain from very small volumes of material high resolution morphology in the TEM or STEM mode, elemental data using the EDS, and structural information for crystalline materials in the selected area electron diffraction (SAED) mode. When identification is based on the nearly simultaneous determination of three quantities-morphology, elemental composition, and crystal structure-the reliability of the analysis is significantly improved because the individual modes sometimes yield ambiguous information. The limita- tions of each mode have been discussed previously [1,2]1. All modes are adversely affected by the presence of adjacent non-fibrous debris and overlaying films. Fibers that are too thin or too thick do not provide sufficiently good SAED patterns for positive identification by comparison with standards. Less than 15 percent of the chrysotile fibrils in a particular standard gave positive SAED patterns. Chrysotile diffraction is further degraded by electron beam bombardment and instrumental contamination. Energy dispersive spectrometry is not a panacea because there are different minerals with similar compositions and elemental substitution is common. Morphology is often compromised by the environment and interfering solids. The hollow-core or tubular appearance of chrysotile is distinctive but often absent and degraded during analysis. It is difficult to establish a protocol for basing identification on three criteria, but when this is done the quality of the analysis is significantly improved. This paper describes some of the difficulties associated with fiber counting in the ATEM with the goal of circumventing the problems. The data from an energy dispersive spectrometer can be converted to chemical concentrations but there is a need to calibrate the instrument and correct for x-ray absorption even in very fine fibers. There are instrumental limitations which degrade EDS spectra but can, to some extent, be avoided. Contamination seriously affects both the EDS spectra and SAED patterns, but there is little that can be done to avoid it in existing instruments other than to understand the problem. The reasons for the controversy concerning the quality of SAED patterns from mineral fibers are examined and criteria suggested for classifying chrysotile SAED patterns. Sample preparation methods are reviewed and some results of inter-laboratory reproducibility are presented. Sample Preparation The three methods of water sample preparation that are commonly used are summarized in table 1 and references 1-6. Water is vacuum filtered through 0.22 pm Millipore or 0.1 pm Nuclepore filters. Nuclepore has the advantage of being smooth and therefore not generating a replicated structure when carbon coated; it has the disadvantages of being prone to fiber loss during handling and sporadic occurrences of non-uniform solids deposition during filtration. Millipore retains fibers well but generates a structured background if carbon coated prior to destruction of the filter structure. In the method of condensation washing [1,2,6], TEM grids with carbon-coated Formvar films are positioned on the Ni support screen of the cold finger in a condensation washer. A piece of Whatman filter paper placed between the TEM grid and the Ni support screen has been shown to reduce fiber loss during solvent extraction [7]. The grids are preconditioned by the application of a few drops of acetone beneath the Ni support screen to prevent warping of the filter section. The filter sections are placed, sample side down, on the TEM grid immediately following pre-conditioning. The Millipore is removed in 10-50 minutes of acetone vapor extraction. The complete procedure and sources of errors are described elsewhere [1,2]. 1Figures in brackets indicate the literature references at the end of this paper. 250 E F.n

Table 1. Method of preparing liquids for ATEM analysis. Method + reference filter medium pre-treatment fiber fixation by vacuum evaporation of carbon Jaffe-fusion 3,4 0.22 71m Millipore fused in acetone vapor for 5-10 minutes yes pre-conditioning none extraction filter section on configuration grid on polyurethane in enclosed petri dish Jaffe-wick 5,6 0.1 um Nuclepore none yes 10 uL droplet of solvent onto sample positioned on grid filter section on grid on wire mesh on several layers of filter paper in enclosed petri dish Condensation washing 1,2,6 0.22 pm Millipore none no acetone wetting of grid without filter filter section on grid on cold finger in reflux column solvent acetone chloroform acetone duration of 12 hours 10-24 hours 10-50 minutes extraction In the Jaffe-wick method [5,6], the Nuclepore filter is carbon coated after filtration to fix the solids in place prior to filter extraction. The TEM grid is positioned on a wire mesh placed on several layers of filter paper in a,:tri dish. The carbon coated filter section is positioned on a grid and a 10 p1 droplet of chloroform is added to prevent warping. The layers of filter paper are,saturated with chloroform and the Nuclepore extracted slowly (10-24 hours) in the covered petri dish. In the Jaffe-fusion method [3,4], a portion of the Millipore filter is attached to a glass slide and placed for 5-10 minutes in acetone vapor. This short pre-treatment in acetone destroys the structure of the Millipore and therein avoids the formation of a replicated network structure during carbon coating which would interfere with fiber counting. The fused Millipore on glass is carbon coated and then extracted using acetone in the same manner as in the case of the Jaffe-wick method. One of the prime sources of error in the analysis is the fiber loss which occurs during sample preparation. Condensation washing is a popular method of preparation, but it introduces variability in the results and yields higher fiber losses than Jaffe-type methods [1]. While some investigators have obtained good results with condensation washing [8,9], there are a sufficient number of technique problems [1,2] so that serious differences occur in inter-laboratory comparisons. It is possible to correct for the losses associated with condensation washing using partially-extracted Jaffe samples to determine the total fiber concentration [1]. This requires additional preparation time and TEM analysis. Fortunately the chrysotile losses associated with condensation washing are usually below 20 percent [1] and can be considered insignificant if the duration of wash is less than an hour in a properly controlled washer. We have obtained reproducible results using Jaffe extraction of carbon-coated Nuclepore [2] and loss corrections in conjunction with condensation washing. 251

All of the above discussion refers to water samples. In preparing air samples it is preferable to low-temperature ash the filter because of the heavy filter loading associated with air sampling. The ash is then suspended in water and processed as a water sample. Because the ash tends to be clumped, it is necessary to subject the suspended ash to ultrasonic treatment. Instrumental Limitations Instrumental problems arise when using energy dispersive spectrometers, because TEMs were never intended to be used in quantitative chemical analysis and ATEMs have been constructed by retrofitting EDS and STEM capabilities to existing systems. There are two prime sources of the instrumental problem: 1) the EDS is not a focusing spectrometer and is Insensitive to the location of the x-ray source and, thus, will detect all x-rays with a line-of-sight path to the detector [3]; 2) in a typical CTEM column there is, in a confined volume, a high density of hardware such as pole pieces, apertures, anti-contamination surfaces, sample grids, samples holders and associated clips. These two features combine to yield remote x-ray generation, i.e., x-radiation originating from regions outside of the volume excited by the primary electron beam. This causes: 1) spectral peaks unrelated to the sample to appear in the EDS spectrum leading to quantitative inaccuracy and errors in identification; 2) increased background radiation which raises the detectability limits; and 3) a loss in spatial resolution. The sources of the problem are secondary fluorescence by characteristic and continuous radiation generated in the column apertures, backscattered electrons from the sample and its support, and scattered primary electrons. The use of high voltages to penetrate thin samples and retain good spatial resolution leads to the generation of characteristic and continuous radiation in column apertures. The second condenser (C2) variable aperture, which is the last aperture above the sample, poses the most serious problem. The maximum in the generated continuum at a bevrt energy of 100 keV and PtKa characteristic radiation both have wavelengths of abou~ 0.2A and are readily transmitted by thin Pt apertures, e.g., over 40 percent of the 0.2 Pt radiation is transmitted by an 100 pm thick Pt aperture. Most of this radiation will be dissipated by absorption in the column but any that does reach the sample area can generate secondary fluorescence at and near the sample which is unrelated to primary electron beam excitation. Because almost all primary electrons are transmitted by thin films and small particles, the backscattered electron fraction is small as indicated for Au films in figure 1 [11]. If the beam voltage is high and the sample thin, less than 5 percent of the incident electrons will be backscattered. Any electrons that are backscattered toward the detector can penetrate the 7.5 pm Be window of the EDS because they will, for the most part, have energies close to the incident beam energy. Eighty percent of the 100 keV electrons can penetrate 7.5 pm of Be and in so doing lose less than 5 percent of their energy. Most backscattered electrons do not reach the detector because they are confined by the strong objective lens field. They can, however, excite remote particulate matter and the support grid. 252

Acceleration Potential In keV Figure 1. The percentage of backscattered electrons as a function pf incident electron energy for two different thicknesses of Au. The data are from Philibert and Tixier [11]. Scattered electrons in the column cause electron beam tailing [12] which leads to excitation of areas in the sample immediately adjacent to the region of primary beam excitation. This effect is due to improper alignment and scattering by column components and increases in severity as the beam voltage is lowered. The following list indicates some steps that may be taken to alleviate these instrumental problems. The magnitude of the problem and, therefore, the effectiveness of these alterations will vary appreciably from one instrument to another because of differences in electron optical configurations, alignment procedures, column cleanliness, aperturing (sizes, materials, thicknesses, and location), and operating mode (TEM vs. STEM). I. Reduce the generation in and transmission of radiation by column apertures. a) Use thick apertures [13] b) Use Pt apertures rather than Mo or Ta [12,14] c) Use column inserts somewhere between C2 and the sample [15] d) The use of low acceleration potential reduces this problem, but promotes beam tailing, backscattering, and absorption effects e) Determine if performance depends upon the emission current for the instrument being used and the type of sample being studied II. Reduce the excitation of material remote to the sample. a) Specimen holders, specimens clamps, and support grids should be made of low atomic number materials (Be, graphite, or polymer) or coated with such materials [1,13,16] b) Use support grids with maximum open area [13] 253 2063105048

c) Coat components near the specimen such as anticontamination devices and sample support rods with low atomic number materials (Aquadage) d) The objective aperture must be removed during EDS data acquisition e) The sample support film should be as thin and have as low an atomic number as possible f) Operate at as low a tilt angle as will provide adequate EDS intensities (less area of grid exposed to excitation) III. Optimize the EDS detector configuration. a) Use the greatest Si(Li) crystal-to-sample distance that will provide adequate count rates [17] b) Collimate the detector with a low atomic number material c) The collimator should be thick enough or shielded with sufficient material (high z) to absorb any stray radiation [18] IV. Minimize electron scattering a) b) c) Use a smail (100 pm) condenser aperture [14] Operate at high acceleration potential Have the column clean and properly aligned These effects of extraneous radiation can best be examined by comparing spectra obtained on and off the edge of a thin film or fiber or by comparing the spectra obtained with the beam positioned in a hole (hole-count) [12] with spectra obtained on the sample. In performing on- and off-film measurements on a Sn-Cu-Cr film, 3 percent of the Cr intensity was attributable to Cr plating on the sample hold-down clip while the Cu TEM grid was responsible for 15 percent of the Cu signal. Insertion of an aperture just beneath the variable C2 aperture on a Philips EM300 operated in the TEM mode increased the Cu peak-to-background ratio and reduced the off-film Cu by 35 percent. The maximum peak- to-background ratios have been achieved using a column insert (1 mm ID x 2.57 ms OD x 3mm thick) in the lower end of the vacuum tube through which the variable C2 aperture passes. Kyser and Geiss [18] have found that operation in the STEM mode reduces the extraneous background by about a factor of two. Even after these precautions have been taken, it is still advisable to subtract the off-fiber spectrum from the fiber spectrum and to use as dilute a sample as feasible. A high density of solids on the grid may reduce the analysis time required to find fibers, but it seriously degrades the quality of SAEO patterns and.EDS spectra. Quantitative Analysis There are two aspects to quantitative fiber analysis of environmental samples in the ATEM, namely, the proper identification of the fibers coupled with the accurate determina- tion of the number of fibers per unit area. When the concentration of a specific mineral is sought the best procedure is to compare unknown spectra and diffraction patterns with those obtained from well-characterized standards in the same instrument using constant operating conditions. When unknown samples are encountered, it is advisable to compare ATEM data with the results of x-ray diffraction, infrared spectroscopy, and x-ray fluores- cence in conjunction with a careful consideration of the mineralogy of the problem. When the fibers, particles, or films of interest are thin, the following expression, originally proposed by Duncumb [19] and pursued by Cliff and Lorimer [20] and Russ [21], can provide good results; 254

CA IA _ (p-B)A ZB SAS r SAS T"F B (1) where I is the net peak intensity corrected for background and peak overlap and SAB is a relative sensitivity factor, i.e., the ratio of the detected intensities (IB /IA ) for two pure thin standards of the same mass thickness. Absorption, secondary fluorescence, and backscattering effects must be negligible for eq. (1) to be applicable. SAB is most easily measured on multi-element thin standards of known composition. There are not many experimental data and the bulk of what is available has been pub- lished by Cliff and Lorimer [20] and Sprys and Short [22]. SAB can be calculated from the following expression which is fully discussed elsewhere [21,23,24]: SAS 4 AA C10 +ZT) GB 1 n CE~ / EC,A exp' o~ 8e 13.9x10 4, AB C10-~ GA 1 n CE~ ) EC,B exp `- p I Be 13.9x10 4 (2) The subscripts A and B refer to the elements A and B. A is the atomic weight, z is the atomic number, G is the fractional emission in the line of interest, e.g., G(Ka12) °Ko.2 intensity/(Ka12 intensity + KS intensity), Eo is the acceleration energy in keV, Ec ie the excitation energy in keV, and p/plBe is the mass absorption coefficient for A or B radiation by the 7.5 pm Be window on the EDS detector. Note that this expression shows no dependence on the instrumental configuration. However, SAS values determined in different instruments may differ from each other and from theoretical values because: 1) the contribution of secondary fluorescence, back- scattering, and beam tailing may be vastly different in different instruments; 2) the Be window thickness and detector efficiencies may be different and, in some instances, the Si dead layer and Si crystal thickness may be significant; and 3) the samples used to measure SAB may not be truly thin with respect to absorption. Figure 2 compares the values calculated from eq. (2) obtained using the Reed and Ware [25] values for G with the experimental values of Cliff and Lorimer [20]; the ratios are relative to Si, i.e., B = Si. As noted by Goldstein et al. [23] the agreement is poor below 2 keV and good above 2 keV. Table 2 also compares calculated and experimental SAB values. For SMg Si' SA1 Si' STi Si' and SF.e Si, the agreement in the experimental values is generally better than 13 percent (fractional standard deviation or coefficient of variation), notwithstanding the variation in experimental configuration and conditions. With the exception of the SNa Si and SMg Si, the agreement between theory and experiment is better than 15 percent. The SMg Si value determined from eight different mineral fiber standards using the data of Beaman and File [1] was 1.7 ± 0.2 (± 14 percent). This varia- tion is primarily due to inaccuracies in the bulk chemical analysis of the mineral fibers. If IC = 1 and the S values are all relative to Si, n SA,Si1A'i~A Si,Si1i ' 255 (3)

Table 2. Calculated and experimental values of the relative sensitivity factor, SA-Si for Ka radiation. Investigator and - - - - - - - - Experimental SA-Si Values--------- Conditions 5Ma-Si SMg-Si SA1-Si STi-Si SFe-Si SCu-Si Cliff & Lorimer[13] EMMA-4 100 kY 5.77 2.07 1.42 1.08 1.27 1.58 0=0° T=45° amphibole particles Beaman & Fiie[2] EM300 80 kV 1.7 ± 0.2 1.4 ± 0.2 1.25 0-39° 7=26° asbestos fibers=0.1 um Sprys & Short[41] EM300 100 kV 7.22 1.08 1.30 silicide particles Morgan et al.[30] EM300 80 kV 3.92 1.55 1.16 1.13 1.38 f<42° 3 ym iso-atomic drops Suzuki et al.[42] JEOL 100C 400 kY 1.7 1.3 2.5 0-0° mineral fibers ---------Calcu7atedSA-SiValues--------- Goldstein et al.[22] 100 kY. 1.66 1.25 1.12 1.16 1.33 1.59 This report Eq.[11] 100 kY 1.52 1.13 1.09 1.07 1.22 1.46 Russ[4] 700 kY 2.01 1.39 1.12 0.95 1.12 1.34 0= tilt angle T - x-ray take-off angle 256

Figure 2. Relative sensitivity factors, SA Si, for Ka radiation as a function of the atomic number of element A. The curves are calculated fr.om eq. (2) and the points are experimental values from Cliff and Lorimer [20]; from Beaman [24,i. Other relative sensitivity factors can be calculated from the Si values because SA8/SC8 = SAC. If the 5 values are not relative to Si n CA ° IA/(IA + 1. 5i AIi) . (4) iB ~ e We measured the composition of a 3000A thick Cu-Sn-Cr film on a Cu TEM grid using Philips EM300 CTEM at 80 keV and a Cameca electron probe operated at 25 keV. The results are shown in Table 3 and compared with bulk chemical results. The ATEM results are seriously degraded by the secondary fluorescence and electron scattering as evidenced by the high Cu value resulting from the use of a Cu TEM grid. Off-film spectra were subtracted from the film measurements. The Cr/Sn ratio which is independent of the scattering problems is in good agreement with the chemical data (relative error = 11 percent). The Cu grid was used to demonstrate the difficulties associated with quantitation in the ATEM. As indicated previously, the results will be improved by using low atomic number grids and grids that do not contain any of the elements present in the sample. The results obtained in the electron probe, where scattering problems are minimized by the instrumental configura- tion and the use of low acceleration potential, are excellent (relative error <10 percent). From these limited data and other reported results on thin films [20,26], we conclude that the thin film model of eq. (1) is valid and capable of providing relative errors of less than 10 percent when using experimentally determined 5., values. This represents reasonably good performance when compared with the 5 percent relative error obtained using EDS systems and bulk samples [27]. However, it rust be stressed that this will only be attained in CTEMs after taking the precautions described previously. The accuracy will be best when measuring concentration ratios. The presence of oxide films or organic contamina- tion on the surface and the tendency for surface segregation and particle inhomogeneity to occur complicates and degrades quantitative results. 257 2063105052

C7 a Table 3. Experimental composition of a 3000 A thick Cu-Sn-Cr film. Method Element Neutron activation ATEM at 80 keV with SAB values Electron probe at 25 keV with SAB values and absorption corrected Electron probe at 25 keV with 5AB values but no absorption correction Composition in weight percent ~ Cu Sn Cr Cr/Sn 14.6 77.6 7.8 0.101 27 67 6 0.090 15.6 76.7 7.6 0.099 16.4 76.3 7.3 0.096 Correction of Quantitative Data It has generally been assumed that if the sample was transparent to electrons, i.e., structure was visible in the TEM image, then the sample was sufficiently thin so that the only consideration necessary in quantitative analysis was the variation in x-ray generation by the primary electron beam. The loss of ionization through backscattering will generally be negligible for sub-micro diameter mineral fibers, if the acceleration potential is above 80 keV. From figure 1, it is seen that for an 1000A film of Au the voltage could be as low as 50 keV and the backscatter fraction still below 10 percent, whereas over 50 percent would be backscattered by a bulk material. Philibert and Tixier [11] have found that continuous fluorescence is negligible and that characteristic fluorescence will be negligible if p/p ' B line t«l. p/p is the mass Ialloy absorption coefficient for the exciting radiation, B, by the material. It is not presently clear how significant the characteristic fluorescence correction is for thin films because the limited accuracy of the analysis in most CTEMs obscures the effect of characteristic fluorescence. In order to make any corrections to the data, it is necessary to know the thickness which certainly complicates the analysis and detracts from the simplicity of standardless correction. However, for particles and fibers the thickness can often be accurately estimated from the TEM image. Absorption effects in the analysis of mineral fibers were reported by Beaman and File [1] and figure 3 shows the dependence of Ix/ISi on fiber size for various minerals. The ratio of intensity ratios at one fiber radius (rl) to those at another fiber radius (r2) can be determined from Beers law. 258