Ultra small-mass AMS 14C sample preparation and analyses at KCCAMS/UCI Facility

We have developed techniques for accurately and precisely measuring samples containing less than a few hundred micrograms of carbon, using a compact AMS system (NEC 0.5 MV 1.5SDH-1). Detailed discussions of the sample preparation, measurement setup, data analysis and background corrections for a variety of standard samples ranging from 0.002 to 1 mgC are reported. Multiple aliquots of small amounts of CO2 were reduced to graphite with H2 over pre-baked iron powder catalyst. A reduction reaction temperature of 450 � C was adopted for graphite samples below 0.05 mgC, rather than the usual 550 � C used on samples of 0.1–1 mgC. In our regular reactors (� 3.1 cm 3 ), this reduction in temperature improved the graphite yield from � 60 to 90–100% for samples ranging from 0.006– 0.02 mgC. The combination of lower reaction temperature with a reduced reactor volume (� 1.6 cm 3 ) gave yields as high as 100% on graphite samples <0.006 mgC. High performance measurements on ultra-small samples are possible also due to a modified NEC


Introduction
An increasing demand to measure small samples at the Keck Carbon Cycle AMS (KCCAMS) facility motivated us to evaluate our ability to reduce sample sizes in terms of graphite sample preparation and spectrometer capabilities. We discuss the strategies adopted to maximize graphite yields, to set up the AMS system, and the approach used to better constrain the effect of extraneous carbon on the AMS 14 C results.

Experiments, results and discussions
Our first interest was to determinate the performance of our AMS system (NEC 0.5 MV 1.5SDH-1) under conditions of reduced carbon sample size and consequently ion beam current intensity. We initially produced smaller aliquots of graphite samples by reducing the carbon sample size from 1 mg to 0.002 mgC and verifying the intensity of the beam current produced and the accuracy and precision that these targets could achieve.

Beam current investigation
In March of 2004, we measured the first wheel of small samples graphitized using our regular sample preparation protocol [1]. It was composed entirely of NIST OX-I, OX-II and Argonne Lab Premium coal POC#3 samples ranging from 0.18-0.6 mgC. The primary normalizing standards used (OX-Is) were of regular size ($1 mgC) and produced an average 12 C 1+ beam current intensity of 40 lA, while the smallest samples delivered only 18 lA. Since the graphite samples were produced using 4-5 mg of iron catalyst independent of sample size, this decrease in beam current intensity was expected [2][3][4][5]. Eight individual measurements were obtained for each target with approximately 35,000 counts each. No dependence of 14 C/ 12 C ratios on ion beam current was observed from samples in this size range. This encouraging result led to further experiments, testing standards samples from $0.015 to 1 mgC. However, we observed that the measured 13 C/ 12 C and 14 C/ 12 C ratios were affected directly by the decrease in ion beam current intensity generated by samples <0.1 mgC.
Previous investigations with large samples demonstrated that the intensity of the ion beam current could be increased by at least 15% by decreasing the amount of catalyst used during graphitization from 4-5 mg to 2 mg of Fe (Fig. 1).
The assumption that we could generate higher ion beam current intensity for smaller samples by decreasing the Fe/ C ratio, and thereby reduce the machine-induced isotopic fractionation [3,4] was subsequently tested on samples ranging from 0.004 to 0.1 mgC. For these experiments we used conventional NEC target holders, to avoid any artifacts from target surface depth. As expected, ion beam currents from targets produced with less catalyst were initially higher compared with those from cathodes where the usual amount of catalyst was used, but they did not last long and collapsed dramatically as the sample was being sputtered by the cesium beam (Fig. 2). The overall effect on 14 C/ 12 C and 13 C/ 12 C ratios was a large decrease in accuracy when compared with results produced by graphite reduced on 4-5 mg of Fe. Therefore, all test samples reported in the following sections were prepared using 4-5 mg of Fe.
Ion source upgrading has been a major part of our inhouse development program, since the purchase of the AMS system in 2002 [6]. As part of this program, in 2003 we replaced the NEC sample holders with a new design where graphite is loaded into a 1 mm hole from the front, using a hammer and a steel pin (Fig. 2 in [7]). All data presented in the paper were measured using these holders, except for the comparison with the different amounts of Fe catalyst mentioned above. In December of 2004, the replacement of the ionizer assembly [7] provided a more efficiently focused Cs beam, where source output could be maintained until essentially all of the sample material was consumed. Measurements on ultra-small samples showed that we could measure up to 4-5% of the radiocarbon atoms in 0.002 to 0.01 mgC samples for which 12 C À beam currents of around 1 lA per lg of carbon were achieved (Fig. 3).

Combustion of ultra-small samples
There are two ways to produce small standards and blanks for comparison with small unknown samples: (a) weigh individual small amounts of reference material to match sample sizes or, (b) split CO 2 gas from a larger combusted sample ($1 mgC, for example). The latter method is easier because it only requires manipulation of the CO 2 gas sample in the graphitization vacuum line. However, it is only valid if it can be shown that it gives the same results as when small individually combusted aliquots are used. A previous study reported that no differences can be observed between the two methods [4]; however, we decided to carry out our own experiments to validate this observation using our own vacuum lines and combustion procedures.
For this purpose, OX-I samples were weighed out in the range of 0.029 to 0.051 mg to produce graphite standards ranging from 0.0035 to 0.01 mgC. Aluminum foil and the tools used to weigh out the small standards were first cleaned of carbon. The Al foil was cleaned by rinsing with methanol, then de-ionized water, 10% hydrochloric (HCl) acid and then again with de-ionized water. The tools used to measure out the standards were rinsed with 10% HCl and then de-ionized water. Both the cleaned Al foil and tools were then baked in a 550°C oven for two hours. Small standards were individually weighed out on a Sartorius M2P microbalance to ±0.001 mg. A piece of cleaned Al foil was placed on the scale tray using tweezers. A small amount of OX-I (or OX-II, IAEA-C7 or coal) was sprinkled on a cleaned watch glass. Using jeweler micro forceps and a 4X single lens magnifier a small aliquot of standard was picked up and transferred to the Al foil for weighing. Once weighed, the aliquot was picked up again and introduced into a pre-baked 6 mm Vycor tube for combustion. Silver foil or wire and cupric oxide were then added to the combustion tube according to established methods [1]. Production of a background coal sample was more problematic since the high density and high carbon content of coal made individual combustions very difficult. A 0.035 mg sample of unprocessed coal still produced many split graphite samples of 0.005 to 0.010 mgC into the graphite vacuum line and that was the smallest individual combusted coal measured out. No significant differences were observed from the AMS 14 C measurements of standards and blanks weighed out individually or CO 2 split from larger combustions, in agreement with the results reported previously [4]. Therefore, all test samples reported in the following sections were prepared using splits from CO 2 gas of larger combusted sample.

Optimization of graphite production
Our regular volume reactors have a volume of $3.1 cm 3 and are based on modified 1/4 00 Ultra-Torr tees, Swagelok plug valves, and 4.5 mm · 6 mm · 50 mm borosilicate culture tubes (Fig. 4a). Graphite targets are produced from CO 2 of large combusted standards, reduced by hydrogen at 550°C to graphite over pre-baked Fe powder catalyst (Alfa-Aesar, À325 mesh). The reduction process is monitored by pressure transducers (Omega PX139, 0-30 psi) and normally takes less than 120 min to reach completion [8]. Water produced during reaction is removed chemically by absorption in magnesium perchlorate. Details of the sample preparation process for regular size samples (0.1-1 mgC) can be found in Santos et al. [1].
Preliminary examination of the reaction yield data, by monitoring the pressure decrease during graphitization, showed that the reduction of small samples (<0.025 mgC) at 550°C was incomplete. This observation was confirmed by independent analyses of small graphitized samples of OX-I and IAEA-C6 by measuring their carbon content and isotopic composition using an elemental analyzer (EA) coupled on-line to a stable isotope ratio mass spectrometer (IRMS). Results showed a tendency toward isotopic fractionation and decreased yield, especially for samples less than 0.01 mgC. Multiple tests (described on Section 2.4) were performed to achieve the optimal reaction temperature. Lowering the temperature to 450°C allowed us to successfully graphitize samples as small as 0.006 mgC.
To decrease this sample size limit further, we built a set of small-volume reactors ($1.6 cm 3 - Fig. 4(b)). These are direct replacements for the regular reactors on the vacuum line and can graphitize 0.002 mgC samples. EA-IRMS d 13 C analyses indicated that ultra-small samples (0.002-0.01 mgC) produced in these small reactors were less isotopically fractionated than those produced in the regular reactors. To better monitor the graphitization yields and to decrease the uncertainty on the graphite target masses,  the Omega PX139 pressure transducers were replaced with more sensitive ones (Silicon Microstructures SM5812, 0-5 psi), and an additional pressure readout box calibrated for the SM5812's was installed on the line for easy switching between the two.

EA-IRMS measurements of small samples
To separate machine-induced fractionation effects from possible graphite fractionation or low yield on ultra-small graphite production, EA-IRMS measurements were performed on 0.002-0.04 mg graphite samples produced from the standards OX-I and IAEA-C6. Errors on the isotopic measurements were expected to be higher (±2&) than the usual 0.15& because of the low signals provided by the samples of low carbon content. Decreased helium flow rates and the use of many size-matched aliquots of carbon reference materials, were used to aid in the accurate integration of carbon peaks from these small samples and to provide calibration corrections. Pure pre-baked Fe was selected to be measured for background corrections, since it is used as the catalyst during graphite production. The EA-IRMS measurements showed that the d 13 C of the pre-baked Fe (À18&) was very close to the isotopic composition of OX-I. Eighty four graphite targets were prepared for these measurements to check the following variables that could potentially control the outcome of the CO 2 to graphite reduction process: (a) reaction temperatures: 450, 500 and 550°C ( The EA-IRMS measurements (Fig. 5) indicate the presence of isotopic fractionation during graphitization process (also observed by Van der Borg and collaborators [9]), possibly correlated with incomplete reduction reactions, as indicated by the nanometric measurements of yield. Note that graphite fractionation is small ($4&) even for samples with <0.005 mgC, that are graphitized in small-volume reactors at 450°C (Fig. 6(a) and (b)).
We do not understand why OX-I graphite samples produced in our regular reactors ($3.1 cm 3 ) appear to fractionate more than IAEA-C6s. Most of the IAEA-C6 and OX-I EA-IRMS results were obtained in separate runs. Although we attempted to calibrate the EA-IRMS with numerous aliquots of small and ultra-small standards as described earlier, conditions were not ideal and there may have been run to run variations. Also, peak areas were not linear with nanometrically measured graphite sample sizes, probably due to sample losses during transfer to the pre-baked (180°C) tin capsules used for EA combustion. Final results were corrected and calibrated according to the IRMS peak areas, not the nanometrically determined sizes of the graphite samples.
Despite large errors in the isotopic composition of the targets evaluated here, graphite fractionation for ultrasmall samples (EA-IRMS d 13 C) can be clearly separated from possible machine-induced fractionation effects (AMS d 13 C), because the machine effects are normally larger ($20& -Section 2.7) compared with the 1-4& shifts from graphitization effects.  ( Fig. 7(a)-(f)) were taken at the Materials Characterization Facility, UCI, using a Schottky thermal field emission FEI/ Philips XL-30 SEM with back scattered electron detector. The SEM picture ( Fig. 7(b)) shows the filamentous graphite growth on pre-baked iron Alfa-Aesar À325 mesh ( Fig. 7(a)). It is known that after iron carbide (Fe 3 C) layers form on the surface, they break up into sub-micron particles. Carbon is deposited on the rear face of these particles, which remain at the tips of the filaments as they grow (Fig. 7(b)) [10]. For small and ultra-small samples that same process takes place, however, the amount of carbon may be not enough to produce filaments, and only small patches of iron carbide (Fe 3 C) are formed. The effect can be observed on the SEM pictures of graphite samples with 0.057 mgC (Fig. 7(c) and (d)) and 0.011mgC (Fig. 7(e) and (f)) on 4-5 mg of Fe catalyst.

AMS 14 C measurement setup and data analysis
Two alternative measurement setups (Fig. 8) can be adopted depending on the size range of the unknown sam-ples being analyzed. Each possibility requires a different background correction procedure. For example, the selection of a matching-size measurement, where all normalizing standards and secondary standards are matched to the size of unknown samples (matching size method), seems very attractive since it eliminates any extra correction as is shown in diagram (Fig. 8). Realistically, it is very uncommon that all unknown samples to be measured have exactly the same sizes. Either numerous standards and blanks must be provided, or another approach must be used (the nonmatching size method of diagram - Fig. 8).
We found that the measurement of ultra-small samples can be also compromised by limited dynamic range of the current measurement system if 1 mgC OX-I standards are used to normalize unknown samples 60.01 mgC. To overcome this effect, we set up and tune the AMS system using $1 mgC graphite samples. When precision and accuracy on these targets are satisfied we switch to more sensitive current integrator ranges and perform the measurement using a smaller size of normalizing OX-I standards. To measure ultra-small samples, normalizing standards should be preferably larger than 0.01 mgC to ensure high, long-lasting beam currents. However, appropriate background correction must be applied. These issues will be addressed in Section 2.8 and Appendix A.
For data analysis, we use the stripchart module in NEC's ''abc'' analysis code for detailed scanning of data from individual measurement runs, and the Lawrence Livermore ''Fudger'' AMS analysis software for the actual analysis [11]. These tools help to determine the causes of anomalous measurements, such as accelerator or ion source sparks or other problems, and to remove them. After anomalous readings are removed, the ''clean'' results are normalized to a set of six or more aliquots of the OX-I standard and then corrected for isotopic fractionation using AMS d 13 C measurement results. The normalized results are then pasted into an Excel spreadsheet where the appropriate background corrections, based on measurements of multiple small graphite samples of 14 C-free material (coal) and an extra set of small modern standards (OX-Is), are performed (Section 2.8) to produce the final data. To verify the robustness and reliability of the corrections, an independent set of multiple secondary standards (such as IAEA C6, C7, C8 and OX-Is and II-s, TIRIs, FIRIs, etc) must be measured on the same wheel (Section 2.9).
So far, we do not have evidence that the matching size method is better than the non-matching size one or vice-versa. Our main driver for selecting a specific measurement setup has been the range of sample sizes to be measured in a given run. Note that many positions within the sample wheel are sacrificed to load standards that will help tuning and normalization, address background levels and verify the corrections applied (Fig. 8).

AMS d 13 C correction
Since our AMS system measures all carbon isotopes, we have the capability to address and correct the machineinduced fractionation when it happens. For our regular size samples, the on-line d 13 C correction clearly reduces the scatter on the final AMS 14 C results. For small and ultrasmall samples, the AMS d 13 C results differed as much as 20& from those of the original material with an overall trend (average over many runs -Section 2.9) to lighter values. However, such a trend or the larger scatter itself did not occur in every single run. So far, we do not have enough data to say definitively that the on-line d 13 C correction is always necessary for small samples, but the use of that correction certainly helped to reduce the scatter and improve the accuracy of most of AMS 14 C results shown here. For consistency, we always correct the AMS 14 C results using the AMS d 13 C values as described in Section 2.6, for any sample size.

Non-matching size method: Normalizing standards -1 mgC or smaller sizes
Three sets of standards to run as two groups: Group #1 First set -Multiple standards of 1mgC (OX-Is, OX-IIs and/or ANUs), only for setup and tuning of the AMS system.

Group #2
Second set -Multiple OX-Is and blanks of matching sizes for normalization and background correction of the small samples.
Only one background correction is necessary: for Modern carbon. The background correction for dead carbon is not required, because the normalizing standards are depleted by the same amount as the equally-sized unknowns.
Two background corrections are necessary: for Modern carbon and 14 C-dead carbon introduced during sample preparation and handling.

Matching size method: Normalizing standards -Match size with unknowns
Three sets of standards to run as one group: Group #1 First set: Multiple standards of 1mgC (OX-Is, OX-IIs and/or ANUs), for setup, tuning of AMS system and normalization of all sizes of unknowns samples.
Second set: Multiple OX-Is and blanks of small sizes to quantify modern and dead carbon blanks for correction of the small unknown samples.

Third set:
Multiple regular and small secondary standards (OX-IIs, TIRIs, FIRIs or IAEAs) to assess accuracy and precision and to check the background corrections applied.

Background corrections
The systematic trends in the normalized results from known standards and backgrounds with decreasing sample size are associated with two constant background components: (a) modern carbon -mostly introduced during sample preparation and handling, and (b) dead carbonprobably due mostly to the presence of 14 C-free carbon in the graphitizing catalyst. The latter component can be ignored when measuring unknown samples paired to small standards of identical size (matching size method), because the values for unknown sample and standard are diluted identically. Otherwise, one must make corrections for both ''modern'' and ''dead'' background components.
The modern carbon blank (m) can be quantified from the measured isotope ratios of small coal samples or any other 14 C-free material processed in the same fashion as the other samples. Isotope ratios for small coal samples, expressed as fractions of the modern radiocarbon standard (F) are shown in Fig. 9. The corrections are validated by applying them to aliquots of multiple secondary standards measured on the same wheel, with standards that are at least a half-life old providing the most sensitive tests.
The dead carbon blank (d) is quantified from DF, which are the deviations of the measured fraction modern carbon for small OX-I samples from those of large normalizing OX-I standards (Fig. 10). These are also validated when applied to multiple modern or near-modern secondary standards (such as IAEA C6, OX-II, additional OX-Is not used for normalization) measured on the same wheel.
The examination of the blank (Figs. 9 and 10) revealed that m typically varies from 0.2-1 lg, while d varies from 0.1-0.5 lg. These figures show data collected over a long period, and scatter within an individual wheel is usually significantly less. Some of these variations may be associated with graphitizer memory effects [12], and variations in the amount and cleanliness of the catalyst in our graphitization reactors, and some may be due to variations in the blank material itself.
The extraneous carbon incorporated in any sample of mass M can be corrected by the following equations: Modern carbon correction : where m is the constant amount of modern C introduced during sample processing, and M is the sample size; Dead carbon correction : where d is the constant amount of dead C introduced during graphitization, M is the sample size. The errors associated with the parameters m and d can be easily estimated from plots like these in Figs. 9 and 10. Typically they are in the range of 50% of the parameters themselves, i.e. r m = m/2 and r d = d/2. To calculate the appropriate error for the ''Modern'' and ''Dead'' corrections, Eqs. 1 and 2, respectively, we use the following equations: Finally, the relationship between the corrected fraction modern carbon for the incorporation of extraneous carbon and, the measured fraction modern carbon can be expressed as: where F and   This technique for background correction is very simple, since the mathematical formulae are simpler compared with those of other approaches [2,5,13], and the two components (m and d) can be determined and verified separately. However, for reliable results they must be measured for every wheel of small and ultra-small samples, and validated on independent sets of secondary standards of appropriate sizes and radiocarbon activity values (Fig. 11).

AMS performance of small and ultra-small samples
In order to evaluate the accuracy of these protocols, multiple small aliquots of several standards were measured on several independent wheels each containing 10-20 specially prepared small OX-Is, OX-IIs and coals spanning the size range of the ''unknown'' samples being analyzed.
Here the ''unknowns'' were multiple independent standards spanning a range of fraction modern C values: extra small OX-Is plus IAEA C6, C7 and C8. To minimize graphite fractionation, a reduction reaction temperature of 450°C was used for all samples <0.1 mgC, and reductions for samples <0.08 mgC were performed in the small-volume reactors (Section 2.3). Standards were generated from splits of CO 2 gas made with $1 mgC of material. All results plotted on Fig. 11 were obtained by the nonmatching method (Fig. 8). Fig. 11 contains dates from runs using large (1 mgC) or small (0.008-0.015 mgC) normalizing OX-Is and these are not distinguished in the figure. Data analyses and background correction procedures were applied as described on Sections 2.6, 2.8 and Appendix A.

Summary
Consistent with other studies [3][4][5] we observed that 14 C/ C ratios for small near-modern samples decrease systematically with sample size. There has been debate whether this effect is due to real isotopic fractionation, primarily a func-  Fig. 11. Fraction modern C results for the IAEA C6, C7, C8 and OX-I samples between 0.002 and $1mgC from multiple AMS 14 C measurements. These results were normalized to independent sets of six normalizing OX-I standards. For samples in the tens of microgram range, this independent set was composed of regular graphite targets (1mgC). For ultra-small samples (<0.010 mgC), the normalizing standards ranged from 0.008 to 0.015 mgC. All results have been corrected for fractionation using AMS d 13 C measurements. Blank corrections have been applied using the formulae presented (Section 2.8 and Appendix A), with uncertainties derived from the scatter in the blank data. Solid lines represent consensus values [17].
tion of machine conditions [3,4] or to dead carbon incorporated within the sample during preparation [5,14]. In our case, we observed the incorporation of contamination from ''dead'' as well as ''modern'' sources that when properly addressed and corrected (Section 2.8) and, validated by measuring secondary standards on a range of sizes and radiocarbon activities (Section 2.9) can provide reliable AMS 14 C results for samples as small as 0.002 mgC.
Our results also showed that graphite targets containing <0.025 mgC are indeed prone to machine-induced isotopic fractionation. Some of the AMS d 13 C results for small samples (several independent wheels containing only standards -Section 2.9), showed systematic trends to lighter values sometimes as much as 20& from those of the original material, and/or increased the scatter. However, such trends were not consistent for every single run, suggesting that conditions had changed from run to run. Overall, the use of on-line d 13 C corrections has improved the quality of our measurements on small and ultra-small samples, but the issue of machine-induced isotopic fractionation is complex, and only a long term evaluation can definitively clarify this issue.
As discussed in Section 2.2, we obtained comparable results from ultra-small aliquots of standards and blanks that were combusted individually and those split from relatively large CO 2 aliquots, indicating that our combustion blank is small. This use of CO 2 splits significantly reduces the difficulty of characterizing graphitization blanks and machine fractionation, as in the present experiments. However for samples which must be chemically pre-treated before combustion, the use of true procedural blanks with sizes similar to the unknowns is essential. This requirement significantly increases the uncertainties associated with the results obtained from such samples [15].
This study has shown that small quantities of CO 2 can be converted reliably into high-quality graphite targets. For samples ranging from 0.05 to 1 mgC, graphite reduction is carried out over iron powder catalyst (Alfa-Aesar À325 mesh) at 550°C into our regular reactors (volume of $3.1 cm 3 ) . For samples <0.05 mgC the optimal temperature is reduced to 450°C, and for samples <0.02 mgC we also use the small reactors (volume of $1.6 cm 3 ) to avoid any decrease in reaction yield and to minimize fractionation. AMS 14 C analysis of samples containing as little as 0.01 mgC has become routine in our laboratory, accuracy and precision sometimes approaching ±10& for 0.01 mgC samples. Depending on sample sizes and number of unknown samples to be measured either the size matching or non-size matching methods can be used. Ultra-small samples from 0.002 to 0.01 mgC can be measured with accuracy and precision of a few percent.

ðA:11Þ
For typical values m $0.5 lg, d $0.2 lg, these approximations are correct to within 30% down to sample masses M = 2 lg, within the scatter of the results for small standards and blanks in this size range.
When the standards used for normalization are relatively small (<0.025mgC -Section 2.6), they too are significantly affected by the dead carbon contamination. In this case, we have to re-write from (A. where the F is the measured isotopic ratio for the unknown sample expressed as a fraction of the (possibly small) normalizing standard. The Eq. (A.13) contains two opposing dead carbon corrections, one for the unknown sample and other for the normalizing standards. This allows the use of any sample size for the normalizing standard set: when the mass of the normalizing OX-I standards is small the DCC(M s ) term compensates the effect of d on the standards themselves. When 1mgC OX-I standards are used the term can be disregarded. When exact size matching is used DCC(M s ) and DCC(M) terms cancel.