Electron Microprobe/SIMS Determinations of Al in Olivine: Applications to Solar Wind, Pallasites and Trace Element Measurements

Abstract

Electron probe microanalyzer measurements of trace elements with high accuracy are challenging. Accurate Al measurements in olivine are required to calibrate SIMS implant reference materials for measurement of Al in the solar wind. We adopt a combined EPMA/SIMS approach that is useful for producing SIMS reference materials as well as for EPMA at the ~100 μg g⁻¹ level. Even for mounts not polished with alumina photoelectron spectroscopy shows high levels of Al surface contamination. In order to minimize electron beam current density, a rastered 50 × 100 μm electron beam was adequate and minimized sensitivity to small Al-rich contaminants. Reproducible analyses of eleven SIMS-cleaned spots on San Carlos olivine agreed at 69.3 ± 1.0 μg g⁻¹• The known Al mass fraction was used to calibrate an Al implant into San Carlos. Accurate measurements of Al were made for olivines in the pallasites: lmilac, Eagle Station and Springwater. Our focus was on Al in olivine, but our technique could be refined to give accurate electron probe measurements for other contamination-sensitive trace elements. For solar wind, it is projected that the Al/Mg abundance ratio can be determined to 6%, a factor of 2 more precise than the solar spectroscopic ratio.

Introduction

Accurate measurements using the electron probe microanalyzer (EPMA) require considerable care for mass fractions of ~100 μg g⁻¹ and below, primarily because of low signal to noise relative to the continuum background (e.g., Reed 2000, Donovan et al. 2011, 2016, Jercinovic et al. 2012). Conversely, precise ion-probe (secondary ion mass spectrometry, SIMS) analyses for most elements are straight-forward when their mass fraction is in the 1–100 μg g-¹ range; however, measurement accuracy requires matrix-matched reference materials. Finding suitable reference materials may be challenging, but can be made by implanting ions into suitable materials and then independently calibrating the implant fluence (Burnett et al. 2015). Figure 1 illustrates this calibration technique using a SIMS depth profile into an Al-implanted San Carlos olivine crystal: if the original olivine Al content is known, then the implant fluence can be calculated, or vice versa (Burnett et al. 2015). In this paper, we report our procedures to measure the Al content of a San Carlos olivine crystal using EPMA to produce a reference material for SIMS analysis. Except for the use of ion implants, our procedures are not unique. Bussweiler et al. (2019) determined the Al content (along with additional trace elements) of San Carlos olivine by solution ICP-MS. Batanova et al. (2015) performed a multimethod analysis of a new olivine reference material. Many others have measured trace element contents of minerals (e.g., Sobolev et al. 2007, Donovan et al. 2011, Kronz et al. 2012) but this differs from our objective of using the trace element content of Al in San Carlos olivine as a reference material for SIMS, rather than primarily evaluating trace element contents to address geological questions. We combine the EPMA measurement with the ability of SIMS to both clean surfaces and to analyse depth profiles of μg g-¹-level implants; this yields a calibrated Al implant in olivine that will enable us to quantify solar wind Al from backside depth profiles measured in Genesis (Burnett 2013, Burnett et al. 2 019) Si collector materials by Heber et al. (2014).

Figure 1.

SIMS depth profile of counts per second (cps) ²⁷Al for implanted San Carlos olivine crystal SCJ. If the olivine aluminum content is known, then the implant fluence can be calculated.

Samples

Polished epoxy mounts of 5–10 mm San Carlos olivine single crystals were made avoiding alumina and metal sample holders. SIMS analyses of samples polished with alumina were found to be hopelessly contaminated despite major efforts at cleaning. Final polishing was performed with ¼-μm diamond paste followed by cleaning with ultrasonic treatment in de-ionized water followed by methanol. San Carlos olivine crystal SCJ was implanted with a fluence of 5 × 10¹⁴ atoms cm⁻² of Al⁺ at 80 keV at Kroko Inc., Tustin, CA, USA. The epoxy mounts were coated with a thin (~100 Å) carbon coat, and carbon paint lines were painted from the sample to the holder to ensure electrical grounding during implantation. The Al fluence used produced no obvious damage to the epoxy. The implant ion beam was rastered over 10 cm producing uniform implants. A previous spot check of a high-dose Fe implant into Si by Kroko Inc. using Rutherford backscattering did not detect any inhomogeneity across the 3-inch diameter of the sample within our ability to measure (< 1%). Thus, control pieces of Si and an epoxy-mounted section of olivine from the lmilac pallasite mounted beside the SCJ sample during implantation received the same fluence. The control Si serves as a matrix-matched reference material for the analysis of Genesis Si samples.

Electron probe microanalyses

Analyses were made with the Caltech JEOL 8200 electron microprobe. As a reference material, we used the enstatite 25–25 from the Caltech reference material library, the composition of which is nominally listed as 0.080% m/m Na₂O, 33.320% m/m MgO, 4.361% m/m Al₂O₃, 54.310% m/m Si O₂, 0.840% m/m Cao, 0.170% m/m TiO₂, 0.300% m/m Cr ₂ O₃ and 6.350% m/m FeO. Wavelength scans over the Al Ka peak from the enstatite were used to determine where to measure background. The Al content of the enstatite reference material was measured relative to a synthetic MgAl₂O₄ spinel reference material (USNM) for which the Al ₂O₃ content can be assumed to be stoichiometric based on prior analytical data. Conditions used were a beam current of 40 nA and backgrounds of± 2 spectrometer units for Al on the enstatite, 60 s count time on the peak and 30 son backgrounds, and a defocused beam of 10 μm diameter. In order to be able to use the same background positions on the San Carlos olivine and the reference material, we used the enstatite rather than spinel as the working reference material. In this paper, we use accuracy as simply referring to absolute, as opposed to relative, mass fractions. Reported concentration fluences are based on working reference materials for which mass fractions and propagated uncertainties can be traced back to the USNM spinel, which we adopt as a primary reference material. Besides the obvious problem of low signal to background, several other problems required solution: (a) Al surface contamination, (b) C deposition, (c) C coat burning, (d) instrumental drifts, (e) inclusions and (f) inhomogeneity (e.g., Batanova et al. 2019).

Surface contamination

Even avoiding alumina, surface contamination of Al is a serious issue at the 100 μg g⁻¹ level. The EPMA analysis depth for Al in olivine is around 2 μm at 15 keV (Batanova et al. 2015). A 1 μm Al₂0 ₃ grain has 20% of the atoms contained in a 10 × 10 × 2 μm analyzed volume of olivine with 100 μg g-¹ Al. Moreover, an Al-bearing surface contaminant has much higher detection efficiency than Al at depth in the olivine. A relatively large rastered electron beam significantly dilutes the influence of small widely dispersed Al-rich particles. We adopted a 50 × 100 μm spot; loss of detection efficiency by the crystal spectrometers is small with this size and was mitigated by using the same rastered beam spot on the enstatite Al reference material.

Photoelectron spectroscopy measurements of Al surface contamination

Photoelectron spectroscopy (XPS) has monolayer sensitivity for surface analysis (e.g., Feldman and Mayer 1986). Monoenergetic Al 2p photoelectrons produced in a thick sample by Al K X- rays have an attenuation length of only about 2 nm (Cumpson and Seah 1997), so only near surface atoms are analyzed. Samples are only exposed to Al X-rays, so no damage or contamination results from the XPS analyses. Analyses were made on the Kratos AXIS Nova surface analysis spectrometer at the Caltech Joint Center for Artificial Photosynthesis on the un- implanted olivine single crystal SC3 to test for Al contamination. Analyses were made using a monochromatic Al Ka source (1486.7 eV) over twelve ~700 × 700 μm scan areas. Scans were collected with an analyzer pass energy of 80 eV, over the kinetic energy range ~1080–1490 eV (binding energy range 0–400 eV) at 0.5 eV step intervals, dwelling for 800 ms at each energy step for a total acquisition time of ~11 min. C 1s, Al 2p, Si 2s, Si 2p, Mg 2s, Fe 3p, Mg 2p, O 2s, and Ca 2s high-resolution XPS data were collected on all samples. Data were analyzed using Casa XPS software version 2.3.16. Measurement conditions were the same for the single crystal sapphire reference material.

XPS has only atomic per cent level sensitivity, however, as shown in Appendix A, significant Al 2p photoelectron peaks were observed on all 12 mm-sized spots with counting rates of around 500 cps (a factor of two spread). A reference single crystal sapphire sample gave 2 × 10⁴ counts. On the assumption that the Al contamination is a monolayer, the mean Al contamination level is at least 2 × 10¹³ atoms cm⁻² (Appendix A), implying that at least 2% of the Al X-rays in an electron probe analysis of SC3 will come from contamination. SC3 is a very clean sample, but since the 2 × 10¹³ atoms cm-² is order of magnitude and a lower limit, some mitigation of surface contamination is required. Reconnaissance SIMS analyses of olivine crystal SC3 indicated that the Al mass fraction was homogeneous; thus, this crystal was chosen for SIMS cleaning and EPMA analyses, as discussed below.

Electron probe C deposition and burning

To check whether Al contamination was introduced by C coating, we SIMS-cleaned a 400 × 400 μm area on high purity Si metal and coated with C in the Caltech GPS Division Analytical Facility in the standard way (nominal 20 nm coating). A 100 nA analysis of the Bremsstrahlung background (twelve measurements on each side of the peak; Donovan et al. 2011) around the position of the Al Ka peak showed a decrease in counting rate at the Al peak wavelength. This result was reproducible; however, this same decrease was not seen when the test was repeated on the Si metal sample that had not been carbon-coated. Use of a higher probe current (200 nA) yielded the same results.

Repetitive counts made at the Al Ka peak on the same spot (Figure _2; 10 μm spot data) indicated that this anomaly was a result of burning away of the C coat. To achieve low Al counting statistic uncertainties, an electron beam current of 100 nA and counting times of 10 min were used (Figure .2). Because of software constraints, the counting rate was first measured at the Al Ka peak, then measurements were made at the lowest spectrometer background setting, progressively stepping to higher spectrometer positions. By the time the background measurements crossed the Al Ka peak position (the thirteenth analysis point); the count rate was on the rising part of Figure .2 (10 μm spot data) giving the negative anomaly. The probe current was found to be stable throughout these tests (variations of <2%).

Figure 2.

Comparison of consecutive measurements on a carbon-coated silicon wafer between 10 μm spot and 50 × 100 μm raster, along with the count rate on uncoated silicon. Measurements made at Al Ka wavelength for 10 min with 100 nA electron beam current. Carbon deposition data with counting rate approaching that measured on uncoated Si wafer.

We interpret the initial decrease in Figure 2 (10 μm spot data) as due to beam-deposited hydrocarbons from the electron beam, but as the spot heated up, the C coat was burned away followed by carbon burning causes large variation in high current density (10 μm) (literally; by residual gas oxidation) and the counting rate eventually approached that of silicon without a carbon coat (dashed line on Figure 2). A complication is that the 16% decrease in counting rate is far larger than expected based on literature absorption data for Al Ka in C (Armstrong 1993, Reed 2005, his equation 8.2), but there seems no other interpretation of the data. The theoretical required C thickness to give 16% absorption is 0.9 μm, forty-five times the nominal thickness of our C coat. The discrepancy is independent of the complex time variations. Simply, the counting rate for Si metal at the Al Ka peak position is decreased by 16% when the C coat is applied.

A simple mitigation was the use of a lower current density (50× 100 μm raster, Figure 2); as noted above, this also reduces sensitivity to a single high Al contaminant particle. A steady decrease in count rate over thirty-eight 10-min analyses of C-coated Si was observed. The contamination rate depends on the current density; a larger beam area gives much lower C deposition rate. Analyses in the C removal region (beyond analysis 6 in the 10 μm spot data on Figure 2) are avoided, but given the low signal/noise for Al in olivine at 100 μg g⁻¹ (about 1/10), the small deposition decrease with the large beam spot is potentially significant and cannot be avoided, but an accurate correction can be made.

Using the rastered beam, a wavelength scan on the C-coated Si sample with uniform steps across the Al Ka peak showed no detectable Al contamination from the C coating.

The C contamination effects on olivine are far less than found on silicon (perhaps due to irradiation induced damage; e.g., Kranz et al. 2012). A test for variations in counting rate analogous to that on Figure 2 (50 × 100 μm raster) was done at the beginning of every olivine measurement session, so that a drift correction can be applied if necessary. The effects vary with instrument conditions, but only 25% of the olivine analyses show any sign of deposition, and corrections are negligible. Figure 3. illustrates the reproducibility of peak counts for twelve consecutive analyses. There were no significant uncertainties using a beam current of 100 nA and count times of 10 min, even with extra time for multiple background points. Similarly, Batanova et al. (2018), using electron energy of 25 keV and a beam current of 900 nA, did not see this effect; these conditions would produce greater C coat damage, so C coat damage is not an issue with our olivine analyses.

Figure 3.

Aluminum peak counts on San Carlos olivine SC3 using a beam current of 100 nA, count time of 10 min and a 50 × 100 μm raster. No significant uncertainties are seen, even with extra time for multiple background points.

Background correction

For the San Carlos olivine, the signal to background ratio was approximately 1/10. Therefore, it is essential to obtain accurate backgrounds, as a 1% uncertainty in background amounts to a 10% uncertainty in Al mass fraction. When measuring Al at per cent levels with the TAP crystal, a linear background correction at spectrometer positions of+ and −6 mm relative to the Al Ka peak at 90.7 mm would normally be used. The background spectrum is not linear, but at per cent levels the correction is very small and is neglected. The reason for the± 6 mm is that, at per cent levels, the background selection needs to be far enough out to eliminate interferences from: (a) the tails of the Al Ka peak, (b) a small ‘satellite’ (Al²⁺) peak at 3.9 mm below the Ka peak (2.9% of Al+ peak height) and (c) any interferences from other common elements. However, for electron probe trace element determinations, it is important to measure backgrounds closer to the peak in order to minimize uncertainties from the curvature in the background (Uercinovic et al. 2012, Batanova et al. 2015, Donovan et al. 2016).

To check for curvature, we performed analyses with two low background, two high background (equally spaced at± 1 and ± 2 mm) and one point on the Al peak. These five measurements were made in sequence from low to peak to high positions on the spectrometer to minimize the effects of C deposition and other instrumental drifts (discussed below). Our background choices were consistent with those made by Batanova et al. (2015). Figure 4 shows background and peak analyses of six spots on SC3; the uncertainties are measured counting statistics standard deviations. Figure 4 shows that the background appears linear within the restricted range of spectrometer position adjacent to the aluminum peak. Separate calculations of corrected peak counting rate were done on the data using only the inner two backgrounds, only the outer two backgrounds, and a mean of all four background points; differences were not significant.

Figure 4.

Six analyses of the San Carlos olivine SC3, run with optimized operating conditions. There was no resolvable difference when the outside background counts were used in the calculation compared with the inside background counts, or all four background counts. Symbols and lines are used to distinguish the similar analyses.

SIMS cleaning of SC3

To eliminate Al contamination issues in EPMA, the Caltech CAMECA 7f-Geo ion probe was used to surface clean eleven 100 × 150 μm spots on SC3 with the same orientation, as shown in Figure 5. (SIMS conditions: O^- 22 keV impact energy, 20 nA primary ion current, 3000 mass resolving power.) Aluminum and ²⁴Mg counting rates were followed during the SIMS cleaning to ensure all surface contamination was removed. The 100 × 150 μm area cleaned was large for SIMS analysis of an insulating sample, but excessive charging was not encountered: once steady-state sputtering was achieved, the Al and ²⁴Mg counting rates reached a steady decline due to charging, but the Al/ ²⁴Mg counting ratio was constant within a few per cent for all eleven spots, guaranteeing that the Al surface contamination had been removed. Only about 0.1 μm or less was sputtered away in the cleaning, so that surface roughness from the ion beam should be negligible. Adjustment of the mount orientation by eye for electron probe analyses was sufficient to align the 50 × 100 μm electron probe analysis spot within the 100 × 150 μm ion-probe-cleaned area. The sputtered areas required a new C coat for electron probe analysis. The enstatite reference material was cleaned and recoated along with the SIMS-cleaned areas.

Figure 5.

Backscatter electron image of San Carlos olivine grain SC3. The bright rectangles (− 100 × 150 μm) are SIMS-cleaned areas. The bright dots are fiducial marks. Both are bright as a result of charging due to the removal of the carbon coat. The sample was subsequently carbon-coated and electron microprobe analyses were performed in the SIMS-cleaned rectangles.

Instrumental drifts

Some instrumental drift was observed even when counting rates were normalized to the measured electron beam current. For analyses of the eleven SIMS-cleaned spots on SC3 in Figure 5, Appendix B gives the measured drifts of the background and peak counting rates, as well as that for the enstatite reference material measured at regular intervals within the SC3 analyses. An approximately smooth correlated drift of all analyses was observed. For the peak position, the drift was 1.1% over the course of the eleven analyses. The drift is best described by the enstatite data that have ‰ statistical precision (Figure 6). A steady increase in the Al Ka counting rate was observed with small, but significant fluctuations up to 8‰ about a linear trend. The causes of the drifts are unknown and were larger in other sessions. Based on the trend and the fluctuations in Figure 6, uncertainties in μg g⁻¹ Al from drift were insignificant over the course of a single set of peak and background measurements.

Figure 6.

Results of five replicate analyses of Al in the enstatite reference material made at regular intervals throughout the SC3 analyses shown in Figure 3. These have less than a permil counting statistics standard deviations. The uncertainty ranges are approximately the size of the data points. Relative to a smooth trend through the data, individual points differ from 0‰ to 8‰.

Aluminum mass fraction of SC3 olivine

In summary, our final optimized conditions were: (a) TAP crystal; (b) 50 × 100 μm rastered beam spot; (c) 100 nA beam current at 15 kV; (d) counting times of 10 min for each peak and background measurement; and (e) four background points per analysis, as shown in Figure 4.

As noted above, there is nothing unique in these conditions. Higher beam currents and higher electron energy could have reduced the time required for an analysis or other approaches could be used (e.g., Batanova et al. 2015, 2019); however, our selections were focused on minimizing C coat beam damage. Moreover, with modern overnight automatic data acquisition, our analysis times were practical, and Figure 6 shows that instrumental drift was not important.

Under these conditions, the enstatite reference material gave sufficiently low counting rates (3700 cps) that dead time corrections were negligible. For SC3 (69 μg g⁻¹ Al), 5% counting statistics standard deviation in μg g⁻¹ Al for a single analysis were obtained, which enabled us to meet our goal of better than ± 5% accuracy of μg g⁻¹ Al in SC3 olivine.

Figure 7 summarizes the results of sixteen analyses of SC3 from two different measurement sessions separated by one month. Points 1–11, from a single session, are all of the eleven SIMS- cleaned spots from Figure 5 with the enstatite reference material run every two or three points. In the other session, the enstatite reference material was analyzed under the same conditions, followed by three of the SIMS-cleaned San Carlos points, the enstatite reference material, two San Carlos points and the enstatite reference material. The uncertainty range bars are± 1s counting statistics (3.5 μg g⁻¹). The observed standard deviation of the data is somewhat less, 3.0 μg g-¹. All sixteen analyses agree within uncertainties, defining a uniform mean Al mass fraction for SC3 of 69.3 μg g-¹• Use of the error of the mean appears justified, giving a precision of 1.25% (0.9 μg g⁻¹). The measurement of eleven SIMS-cleaned points provides a good check on uncertainties due to contamination, Al inhomogeneity, inclusions, and anomalous fluctuations in counting rates. Our sample handling subsequent to SIMS cleaning was not perfect; some particles were observed on the cleaned SIMS spots, but these apparently were not Al-rich.

Figure 7.

Analyses of San Carlos olivine SC3 under optimized conditions of 50 × 100 μm raster, 100 nA beam current, and count time of 5 min on the peak and background. Five points were analyzed in session 1, and eleven points in session 2.

The enstatite reference material is not completely homogeneous with variations ranging from 4.3% to 4.7% Al₂0₃. However, our SC3 Al mass fractions are based on an easily located enstatite area for which the local mass fraction was calibrated relative to the spinel primary reference material. The mass fraction in this area was uniform to within 0.06 ± 0.03%. We adopt 4.70 ± 0.03% m/m Al₂0 ₃ or 0.6% uncertainty as the Al₂0 ₃ content of our local area enstatite reference material. As a check on systematic uncertainties from the matrix corrections to obtain μg g-¹ Al from the measured background-corrected Al peak counting rates, ten different correction schemes were compared but all agreed to within a standard deviation of 0.3%. Propagating the 0.3% and 0.6% with the 1.25% precision gives a total estimated uncertainty of 1.4%. The final mass fraction of Al in SC3 olivine equals 69.3 ± 1.0 μg g⁻¹ (1.4%; 1s total uncertainty).

SIMS analyses

The known SC3 olivine Al mass fraction was used to calibrate the fluence of the SCJ Al implant using the approach of Burnett et al. (2015). SC3 serves as a primary SIMS reference material used to determine the relative sensitivity factor (RSF) for Al in San Carlos olivine. The RSF is needed to calculate the SCJ implant fluence (atoms cm- ²) based on the SCJ Al implant depth profiles.

SIMS depth profiles were made for SC3 and SCJ with an o- primary beam and Al+ secondary ions using the Caltech Cameca 7f-Geo instrument. Analyses are described in detail in Appendix C. Results are summarized here.

Equations for relative sensitivity factor (RSF) and fluence, 4J

These are as follows:

$$n(x)=RSF\left(\text{Al}{/}^{24}\text{Mg}\right)(x)$$ (1)

$$\psi =\int n_i(x)dx=RSF\int \left(Al{/}^{24}Mg\right)i(x)dx$$ (2)

where n is the number of Al atoms cm-³ at depth x, ²⁴Mg is the matrix reference isotope, (Al/ ²⁴Mg); (x) is the implant (i) counting rate ratio at depth x, and 4J is the implant fluence in atoms cm^-2. Equation (1) applies either to the implant as a function of depth or to the constant (Al/ ²⁴Mg) of the SC3 or SCJ olivine. Equation (1) is the definition of RSF. Equation (2) is the basic physics relation between Al fluence and mass fraction; it is an exact equation. The implant (Al/²⁴Mg); (x) is equal to the measured Al/ ²⁴Mg at depth x minus the contribution for the SCJ olivine measured at a depth where all the implant Al has been sputtered away. The SCJ olivine Al mass fraction is assumed to be constant for a given spot, but not among different spots. Thedepth scale dx is related to the measured time step, dt and the sputtering rate S, where S is the measured sputter pit depth divided by the total sputtering time in olivine:

$$\text{dx}=\text{S dt}$$ (3)

where sputtering rates were 0.2–0.22 nm ^s-1.

In Figure 1, from the surface down to 20 nm, the profile represents a combination of surface contamination and Al transients. At greater depths, the profile is dominated by the Al implant. The implant profile has a long tail, but at depths greater than about 400 nm, the implant tail is negligible and the Al counting rate reflects that of the Al in the SCJ olivine. The olivine Al counting rate at large depths shows a systematic decrease due to charging, but when normalized to ²⁴Mg, the deep (olivine) Al/ ²⁴Mg profile is constant (Appendix C: Figure C4) showing that the Mg normalization is adequately correcting for charging.

Calculation of SCJ implant fluence

The measured electron probe olivine Al mass fraction (n) for SC3 and Equation (1) was used to calculate the RSF based on the SIMS Al/ ²⁴Mg for SC3. This RSF was used along with the implant Al/ ²⁴Mg depth profile from SCJ with Equation (2) to calculate the fluence from the SCJ depth profiles.

The implant integral in Equation (2) is the sum of two parts: a ‘main’ integral based on most of the measured Al/ ²⁴Mg profile plus a ‘surface correction’ for the surface regions of the profile affected by contamination and transients.

Quantitative SIMS analyses must be made under sputtering steady-state conditions. For SIMS elemental analyses of homogeneous mineral grains, a period of ‘pre-sputtering’ with no data accumulation is used to achieve steady-state conditions. This is not possible for an implant as some of the implant profile lies in the transient region. For olivine, large oscillating transients in the Mg depth profile are observed (Appendix C: Figure C2) until about 60 nm. An accurate correction for small residual Mg transients in the 60–150 nm range can be made (Appendix C: Figure CS), so the lower limit of the main integral is in the 50–60 nm range for the adopted profiles. At the depth of the main integral lower limit, where Mg is sufficiently close to steady state, Al is also in steady state as shown by comparison of the un-implanted SC3 Al profiles (Appendix C: Figure C3).

A surface correction for the initial parts of the implant profile lost in transients and contamination was made by fitting the profile in the 50–70 nm range to a theoretical profile (SRIM, Ziegler et al. 2010; Appendix C: Figure C6). The integral of the fitted SRIM profile from the surface to the main integral lower limit provides the surface correction to Equation (2). For the adopted profiles, surface corrections ranged from 23% to 29% (Appendix C: Table C2).

SCJ implant fluence

Two profiles of SC3 were bracketed between pairs of SCJ implant profiles. The RSF derived from the two SC3 profiles (Equation 1) agree to better than 0.2%. Folding in the uncertainty in the electron probe mass fraction, the enstatite reference material and the adopted density needed to convert μg g⁻¹ into atoms cm- ³ (n) gives an overall RSF accuracy of± 1.8%. One of the four SCJ implant profiles is anomalous, but the measured fluence is within the range of the other three profiles (Appendix C). The four adopted profiles define a mean implant fluence (Equation 2) of 4.65 ± 0.28 × 10¹⁴ cm⁻² that is within uncertainty of the nominal 5 × 10¹⁴ cm⁻² (Appendix C: Table C2) implanted by Kroko, Inc. Based on the SC3 RSF, the deep Al/ ²⁴Mg ratios for the SCJ profiles (i.e., the SCJ olivine Al) yield 112–114 μg g⁻¹ Al for SCJ (Appendix C: Table C2 and Figure C7).

Fluence uncertainties

The measured precision (standard deviation) of the fluences of the three adopted SCJ profiles is acceptable, 5.5%; this is somewhat higher than our estimated precision for a single measurement of 3.9% (Appendix C) indicating some unknown or underestimated source of uncertainty, possibly in the surface corrections. Compounding the measured 5.5% standard deviation with the 1.8% uncertainty in the RSF gives a total estimated fluence accuracy of 6.3% (1s).

Pallasite olivine Al mass fractions

The technique presented above is applied to two scientific applications: demonstrating accurate olivine Al analyses by a combined SIMS/EPMA approach and calibrating an implant reference material enabling analysis of the solar wind Al abundance. As an illustration, analyses were made of olivine in three pallasites.

In an earlier SIMS measurement session from that discussed above, four profiles were made from a polished section of lmilac, which was co-implanted with SCJ and thus has the above Al fluence as SCJ. Single profiles were also made of un-implanted olivine from the Eagle Station and Springwater pallasites. Profilometer pit depths were not measured on these samples; however, sputtering rates for the lmilac profiles can be calculated from the peak times and the mean peak depth (72.8 ± 2.5 nm) measured during the SCJ analyses discussed above. lmilac data processing was as described for SCJ in Appendix C. Results were similar except that lmilac surface corrections were smaller (15–17%) than for SCJ. Using Equation (2) and the SCJ fluence, an RSF for the pallasite analyses can be calculated from the lmilac implantation profiles. Good agreement was obtained for the four lmilac implant profiles with the derived RSF showing a standard deviation of 1.5%. Using Equation (1), this RSF can be combined with the deep (below transients) olivine Al/ ²⁴Mg to obtain μg g⁻¹ Al mass fractions for the pallasite olivines as shown in Table 1. The Al/²⁴Mg profile of Eagle Station showed at least four small (10–20 nm) Al-rich inclusions over a depth interval of 200 nm; the mass fraction in Table 1 includes the inclusions. No inclusions were resolved in any of the other pallasite depth profiles. The quoted absolute uncertainty (7.0%) comes from the implant fluence uncertainty (6.0%), the uncertainty in the adopted implant peak depth (3.3%), and the RSF precision of 1.5%. The relative uncertainty among the pallasite analyses was much smaller (< 2%).

Table 1.

Aluminum mass fractions in pallasite olivine specimens

Analysis	Al (μg g− 1)	1s
lmilac 1	58	4
lmilac 2	106	7
lmilac 3	76	5.3
lmilac4	73	5
Eagle station	40	3
Springwater	45	3

Discussion

Application to solar wind Al fluence

The work described here demonstrates feasibility for accurately measuring the solar wind fluence of Al from Genesis collector materials (Burnett 2013, Burnett et al. 2019). Precise solar wind implant profiles are available from backside depth profiling in Si collectors (Heber et al. 2014). All that is missing is a calibration of the Al implant in Si reference materials. Si samples co-implanted with SCJ could be used to intercalibrate the Si reference materials used by Heber et al. (2014). However, improved accuracy will be obtained by the use of a new, deeper, implant to eliminate the large (20–30%) surface corrections for the SCJ analyses due to deep transients in Mg and, to a lesser extent, Al. An implant at 180 keV, twice that used for SCJ, is feasible and will make surface corrections negligible. An uncertainty of 3.5% (1s) in the accuracy of the standard implant fluence and a total uncertainty of 5% in the solar wind Al fluence and 6% in the solar wind Al/Mg appears feasible. For comparison, the solar photospheric Al/Mg, among the most precise, has an uncertainty of 12%.

SIMS/EPMA measurements of Al in olivine

Trace element mass fractions in rocks and minerals are highly variable, and trace element geochemistry, to a first approximation, is an order of magnitude science. In general, high accuracy in mass fraction is rarely required. The higher accuracy required to calibrate the solar wind Al fluence is an exception.

In terms of electron probe microanalysis, potential uncertainties due to surface contaminationm ay be important for many elements and should be considered in EPMA trace element measurements. For Al, our observed uncertainties from these sources represent about 20–30 μg g⁻¹, which would be significant for Al mass fractions less than 200 μg g-¹, and perhaps more importantly, could cause a homogeneous sample to be interpreted as inhomogeneous. Our sample preparation was done with a view to avoiding Al contamination, yet significant contamination was observed. Without additional study, it cannot be ruled out that the observed contamination is unique to our sample and handling procedures; we think this unlikely.

Aluminum mass fractions in many natural olivines are relatively large. For example, Al ranges from 200 to 500 μg g⁻¹ in olivines from MORB and large igneous provinces (Coogan et al. 2014), and the issues considered here are probably unimportant in electron probe analyses of such samples. Al in mantle olivines can be lower (down to 5 μg g⁻¹; de Hoog et al. 2010), below the range of electron microprobe analysis.

The focus here is on Al in olivine, but in general, for electron probe analysis of other, normally major, elements (e.g., Na, Ca, Cr, Fe) at trace levels, similar contamination effects are highly likely. Our SIMS cleaning techniques could be adapted to give improved determinations by EPMA for other elements at the - 100 μg g-¹ level. Other issues in EPMA trace element measurement are discussed in detail in Batanova et al. (2015, 2018). High accuracy for elements at the 100 μg g⁻¹ level could be important for evaluating mechanisms of coupled substitutions.

Aluminum reference materials for SIMS analyses based on electron probe analyses of Al in olivine down to 50 μg g⁻¹ can be accurately made with the SIMS cleaning technique used here for San Carlos SC3. Below 10–50 μg g⁻¹, the much greater sensitivity of SIMS for Al indicates that this is the preferred technique when available. For samples not polished with alumina, surface contamination is easily removed by sputtering. For un-implanted samples, where a rastered beam is not required, current SIMS spatial resolution is around 5 μm, comparable to that of an electron probe microanalyzer. NanoSIMS measurements at submicron spatial resolution are probably possible. Depth resolution is easily 10 nm, much better than an electron probe, enhancing sensitivity to inclusions (e.g., Paque et al. 2017). The high SIMS precision allows easy quantitative relative measurements of zoning and heterogeneity in general, both within and among grains of the same mineral.

Matrix-matched SIMS reference materials are ideal, but in most cases where high accuracy is not required, these are unnecessary. This issue is discussed in more detail in Burnett et al. (2015); examples are known where ‘matrix effects’ on SIMS sensitivity factors are both large and small. For example, variations of up to 50% are documented for the measurement of Mn/Cr sensitivity factors with olivine Fa content (Doyle et al. 2016). In contrast, matrix effects in both elemental and isotopic compositions of B in a wide variety of phases are small (Chaussidon et al. 1997). Our experience with SIMS trace element measurements in melilite (Paque et al. 2017) shows that, compared with melilite reference materials, use of a NIST SRM 600 series glass would produce mass fraction uncertainties < 20%. It is likely that for the great majority of trace element studies of silicate minerals, SIMS analysis using NIST glasses as reference materials have adequate accuracy. When higher accuracy is required, use of calibrated ion implants as SIMS reference materials, as in this study, are feasible (see also Burnett et al. 2015).

Pallasite analyses

These are a good example of the preceding discussion, providing a ‘proof of concept’ for accurate analyses of Al in pallasitic olivine. If desired, the 7% accuracy for Table 1 could be significantly improved with additional work. The variability observed in our lmilac Al analyses of a single olivine crystal (Table 1) is not surprising despite the slow metallographic cooling rates (2.5–18 °c My-¹, Yang et al. 2010) of paIlasites. lmilac olivine is zoned in Fe, Mg, Cr, and Al (Miyamoto 1997). The Fa zoning could be modelled by a cooling rate of 20–200 °c My⁻¹, barely compatible with the metallographic rates, although the metallographic and Fa cooling rates apply to different temperature ranges. The Miyamoto core/rim Al variation was a factor of 2 over 700 μm. lmilac Al mass fractions ranged from 80 to 185 μg g-¹ with electron probe counting statistics uncertainties of 39 μg g-¹, consistent with Table 1. Mittlefehldt and Herrin (2010 and private communication) report a laser ablation ICP-MS measurement result of 80 μg g⁻¹ for Al in the lmilac olivine, consistent with Table 1. The measurement results of Mittlefehldt and Herrin (2010) for the Eagle Station (29 μg g⁻¹) and Springwater (25 μg g⁻¹) specimens are lower than our values in Table 1, but this much variability could easily be real. The observed inclusions in Eagle Station are a possible source of variation. The inclusions have small thicknesses (10–20 nm), but there are no constraints on their lateral dimensions; they could be thin exsolution lamellae, easily detected with SIMS depth resolution.

Supplementary Material

supplement A

Appendix A. XPS modelling.

supplement B

Appendix B. Instrumental drifts.

supplement C

Appendix C. SIMS analyses of San Carlos olivine.