Nitrogen Trifluoride Global Emissions Estimated from Updated

Nitrogen triﬂuoride global emissions estimated from updated atmospheric measurements

Tim Arnolda,1, Christina M. Hartha, Jens Mühlea, Alistair J. Manningb, Peter K. Salameha, Jooil Kima, Diane J. Ivyc, L. Paul Steeled, Vasilii V. Petrenkoe, Jeffrey P. Severinghausa, Daniel Baggenstosa, and Ray F. Weissa

aScripps Institution of Oceanography, University of California at San Diego, La Jolla, CA 92093; bMet Ofﬁce, Exeter EX1 3PB, United Kingdom; cDepartment of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA 02139; dCentre for Australian Weather and Climate Research, CSIRO Marine and Atmospheric Research, Aspendale VIC 3195, Australia; and eDepartment of Earth and Environmental Sciences, University of Rochester, Rochester, NY 14627

Edited by Susan Solomon, Massachusetts Institute of Technology, Cambridge, MA, and approved December 20, 2012 (received for review August 2, 2012)

Nitrogen trifluoride (NF3) has potential to make a growing contri- trifluoride can be broken down into reactive fluorine (F) radicals bution to the Earth’s radiative budget; however, our understand- and ions, which are used to remove the remaining silicon-con- ing of its atmospheric burden and emission rates has been limited. taining contaminants in process chambers (12, 13). The physical Based on a revision of our previous calibration and using an ex- and chemical properties of NF3 make it a safe gas to transport; it panded set of atmospheric measurements together with an atmo- is stable and nonflammable, and faster cleaning rates and in- spheric model and inverse method, we estimate that the global creased chamber life make it preferable to the perfluorocarbons ± · −1 ∼ emissions of NF3 in 2011 were 1.18 0.21 Gg y ,or 20 Tg CO2- (PFCs) previously used, primarily hexafluoroethane (C2F6, PFC- − eq·y 1 (carbon dioxide equivalent emissions based on a 100-y 116). Nitrogen trifluoride was also chosen because of its promise global warming potential of 16,600 for NF3). The 2011 global mean as an environmentally friendly alternative, with conversion effi- tropospheric dry air mole fraction was 0.86 ± 0.04 parts per trillion, ciencies (to create reactive F) of ∼98% of total use, compared · −2 resulting from an average emissions growth rate of 0.09 Gg y with ∼30% for C2F6 (13, 14). Thus, its use has likely played an over the prior decade. In terms of CO2 equivalents, current NF3 important role in the SC industry’s strategy to meet its voluntary emissions represent between 17% and 36% of the emissions of PFC emission reduction commitments—a reduction in total other long-lived fluorinated compounds from electronics manufac- GHG emissions of 10% between 1995 and 2010 (14, 15). fi ture. We also estimate that the emissions bene t of using NF3 over Given that its production is increasing rapidly to meet demand fl fi — hexa uoroethane (C2F6) in electronics manufacture is signi cant in end use (manufacture of SC+ devices), that it is a substitute · −1 emissions of between 53 and 220 Tg CO2-eq y were avoided during for PFCs for which emissions are already regulated, and that the ∼ 2011. Despite these savings, total NF3 emissions, currently 10% first atmospheric measurements showed a rising atmospheric of production, are still significantly larger than expected assuming abundance, NF3 is now beginning to be included in global and global implementation of ideal industrial practices. As such, there regional climate legislation (16, 17). It is therefore important to is a continuing need for improvements in NF emissions reduction 3 study the atmospheric trend in NF3 alongside other GHGs in strategies to keep pace with its increasing use and to slow its rising order to provide information that can help evaluate the success contribution to anthropogenic climate forcing. of efforts to quantify and regulate its emissions (18). In this work we supplement and extend the only published atmospheric composition | climate change | radiative forcing atmospheric NF3 record using a revised primary calibration and an automated measurement technique (5, 6). By combining our longside the control of carbon dioxide (CO2), there are sig- measurements with independent industrial production and emis- Anificant climate benefits to be gained from limiting emissions sion data, we provide optimally derived global emission estimates of non-CO2 long-lived greenhouse gases (GHGs) (1). Nitrogen from 1978 to 2011 using an atmospheric model and an inverse fi trifluoride (NF3) has a global warming potential on a 100-y time method. We discuss the signi cance of these emissions and es- scale (GWP100)of∼16,600 (see discussion below), and its use has timate how successful the use of NF3 has been with respect to its grown rapidly in the manufacture of modern electronic devices impact on Earth’s radiative budget compared with earlier SC+ (2). The potential for significant NF3 emissions was only recently industry technology. recognized (3), and was not considered in the first commitment Results and Discussion period of the Kyoto Protocol (4). Despite NF3’s recent rising importance, difficulties in making atmospheric measurements Atmospheric Measurements. We have measured archived air sam- have prevented a thorough analysis of its global abundance and ples from the Northern Hemisphere (NH) and Southern Hemi- EARTH, ATMOSPHERIC,

emission rates (5, 6). sphere (SH) to complement and compare with the measurements AND PLANETARY SCIENCES − − Nitrogen trifluoride’s radiative efficiency of 0.211 W·m 2·ppb 1 made previously in our laboratory (6). Additionally, we have ex- is comparable to many anthropogenic GHGs (7); however, its tended the in situ record of NF3 in ambient air at Scripps Institution large potential impact on climate arises from its atmospheric of Oceanography (SIO), La Jolla, CA (32.87°N, 117.25°W), lifetime, which is long on societal timescales. Prather and Hsu (3) from April 2011 to March 2012, and have analyzed the entire record to estimate average monthly mole fractions representative calculated a lifetime of 550 y (resulting in a GWP100 of 16,800); however, three separate kinetic studies have recently suggested a 1 larger sink for NF3 by reaction with O( D) in the stratosphere (8-10). Dillon et al. (9) suggested a revised lifetime of 490 y, Author contributions: T.A. and R.F.W. designed research; T.A., C.M.H., J.M., and A.J.M. performed research; D.J.I., L.P.S., V.V.P., J.P.S., and D.B. contributed new reagents/analytic yielding the GWP100 valueof16,600weusehere. tools; T.A., J.M., A.J.M., P.K.S., and J.K. analyzed data; and T.A., J.M., and R.F.W. wrote Use of NF3 began in the 1960s and 1970s in specialty appli- the paper. cations, e.g., as a rocket fuel oxidizer and as a fluorine donor for The authors declare no conflict of interest. chemical lasers (11). More recently, beginning in the late 1990s, This article is a PNAS Direct Submission. NF3 has been used by the electronics industry in the manufacture 1To whom correspondence should be addressed. E-mail: [email protected]. fl of semiconductor (SC), photovoltaic cell, and at-panel display This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. devices (from here on all three are termed SC+). Nitrogen 1073/pnas.1212346110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1212346110 PNAS | February 5, 2013 | vol. 110 | no. 6 | 2029–2034 Downloaded by guest on September 25, 2021 of the well-mixed lower troposphere at that latitude (Fig. 1). La Jolla receives polluted air masses from the surrounding region, so to estimate these average well-mixed mole fractions (that are needed for global emissions calculations) we identified the measurements that were not influenced by pollution by studying the history of the arriving air masses using the NAME (Numerical Atmospheric dispersion Modelling Environment) Lagrangian particle model (19). The reported in situ baseline abundance (dry air mole fraction) over the 12-mo period increased from 0.91 ± 0.05 parts per trillion (ppt) in April 2011 to 0.97 ± 0.03 ppt in March 2012 (error range is given as 1- σ of the filtered monthly air data). During December, none of the air masses arriving at La Jolla were classified as baseline, consistent with the recording of some of the highest NF3 mole fractions measured in air to date. Fig. 2A shows the atmospheric NF3 archive measurements made to date and the average baseline monthly mole fractions calculated from the 2011 in situ measurement data from La Jolla. The full set of archived air measurements is also documented in Table S1. Measurements of archived air tanks filled before 1975 and air entrapped in ancient ice collected from Taylor Glacier, Antarctica, found undetectable levels of NF3, and we conclude that preindustrial NF3 levels were no greater than 0.008 ppt.

Global Emissions and Atmospheric Burden. We used an inverse method to estimate emissions, using a 12-box atmospheric chemistry transport model to couple the sensitivity of semihemispheric mole fractions to global emissions (20). The emissions calculated by the inversion were then used as input into the model, and the monthly mole fraction output and growth rate for the two semihemispheres are plotted alongside the atmospheric measurements in Fig. 2, with annual average mole fractions and growth rates summarized in Table 1. The inversion-derived emissions were able to reconstruct the atmospheric history within uncertainty for the majority of measurements (Fig. S1). The onset of significant emissions is evident at the beginning of the 1990s, followed by an approximately linear increase through that decade (Fig. 3A). This finding is consistent with the beginning of NF3 use in the SC industry in the early 1990s. A significant increase in the emissions rate begins at the start of the 2000s with Fig. 2. (A)NF3 atmospheric dry air mole fractions (in ppt) measured in annual emissions increasing roughly linearly at a rate of ∼0.09 Northern Hemisphere (NH) and Southern Hemisphere (SH) archive air sam- · −2 ± · −1 ples, and mean monthly abundances calculated from in situ measurements Gg y between 1999 and 2011, reaching 1.18 0.21 Gg y in at La Jolla, CA, 2011−2012. Data from Weiss et al. (6) are shown rescaled per 2011. The emissions reported in our previous study (6), and the SIO-12 scale. Error bars are calculated as the total measurement un- corrected for the revised primary calibration (5), were ∼0.70 and certainty, which includes measurement error and the ability of the mea- − ∼0.77 Gg·y 1 in 2006 and 2008, respectively. In this work we cal- surement to reflect well-mixed air. Trends were calculated using a 12-box − culate emissions of 0.83 ± 0.19 Gg·y 1 in 2006 (∼18% higher) and model with emissions from the inversion as input. (B) Annual growth rates − 1.01 ± 0.19 Gg·y 1 in 2008 (∼30% higher). The significantly larger were calculated for each month by using the first derivative of the cubic fi ’ emissions we now calculate are due to an increased number of spline t of the model s monthly mole fraction output.

measurements, a longer data record, and the improved modeling and emissions calculations. The growth in the average global atmospheric mole fraction from near the detection limit (0.02 ppt) in 1980 to 0.86 ppt in 2011 is the result of a continuous increase in the growth rate to − almost 0.1 ppt·y 1 over the past two decades (Fig. 2B). The largest increase in the growth rate occurred between 2000 and 2006, when there was a significant relative rise in NF3 production and − − use (2, 23). Using a radiative efficiency of 0.211 W·m 2·ppb 1 from Robson et al. (7), an average instantaneous radiative forcing in −4 −2 2011 is calculated at 1.8 × 10 W·m . Thus, NF3 is currently contributing ∼0.010% of the radiative forcing due to rising CO2 concentrations since 1750, and 0.017% of the forcing due Fig. 1. Time series of NF3 in situ measurements of dry air mole fraction (in ppt) during 2011 and 2012 at Scripps Institution of Oceanography (La Jolla, to non-CO2 GHGs (1, 26). CA). Measurements representative of well-mixed midlatitudinal Northern fi Hemispheric air (black circles) were separated from the full data set (gray Signi cance of Current NF3 Emission Levels. Although the in- circles) by using the NAME particle dispersion model, and these data were stantaneous radiative forcing due to NF3 is currently small, its then used in the calculations of global abundances and emissions. atmospheric lifetime essentially makes emissions cumulative

2030 | www.pnas.org/cgi/doi/10.1073/pnas.1212346110 Arnold et al. Downloaded by guest on September 25, 2021 Table 1. Average global tropospheric NF3 dry air mole fractions (x), mole fraction growth rates (dx/dt), and global emissions ﬂux (F) with associated 1-σ uncertainties Year x, ppt dx/dt, ppt·y−1 F,Gg·y−1

1980 0.020 ± 0.021 0.000 ± 0.006 0.00 ± 0.14 1985 0.021 ± 0.023 0.001 ± 0.006 0.01 ± 0.13 1990 0.037 ± 0.026 0.006 ± 0.006 0.07 ± 0.13 1995 0.080 ± 0.029 0.012 ± 0.006 0.15 ± 0.14 1998 0.124 ± 0.030 0.018 ± 0.006 0.21 ± 0.15 2000 0.164 ± 0.031 0.022 ± 0.006 0.27 ± 0.15 2001 0.189 ± 0.031 0.028 ± 0.007 0.32 ± 0.16 2002 0.221 ± 0.032 0.037 ± 0.007 0.44 ± 0.17 2003 0.263 ± 0.033 0.045 ± 0.008 0.54 ± 0.18 2004 0.312 ± 0.034 0.054 ± 0.008 0.64 ± 0.19 2005 0.370 ± 0.034 0.062 ± 0.008 0.74 ± 0.19 2006 0.436 ± 0.035 0.069 ± 0.008 0.83 ± 0.19 2007 0.508 ± 0.036 0.076 ± 0.008 0.92 ± 0.19 2008 0.588 ± 0.037 0.083 ± 0.008 1.01 ± 0.19 2009 0.673 ± 0.038 0.087 ± 0.008 1.07 ± 0.19 2010 0.762 ± 0.039 0.091 ± 0.009 1.11 ± 0.20 2011 0.855 ± 0.039 0.095 ± 0.009 1.18 ± 0.21

For comparison with other relevant species, see table 2.1 in ref. 21, chap. 2 and table S1-1 in ref. 22, chap. 1.

with an effect lasting beyond usual societal timescales. We calculate a revised GWP100 of NF3 of 16,600 using the lifetime calculated by Dillon et al. (9), and use this to calculate CO2- equivalent (CO2-eq) emissions (Fig. 3A). Emissions of NF3 -1 in 2011 were 19.6 ± 3.5 Tg CO2-eq·y , equating to 0.06% of the CO2 emissions due to fossil fuel combustion and cement production (∼34,000 Tg·y-1) in terms of the contribution to radiative forcing over the following 100 y (27). The SC+ industry uses a suite of fluorinated gases for a range of highly specialized tasks, although the effective mix varies widely between countries and companies (14). Thus, accurately “ ” quantifying emissions using bottom-up methods (from in- Fig. 3. (A) Top-down NF3 emissions estimated optimally using the atmo- ventories, i.e., independent of atmospheric measurements) is spheric measurements and independent emission growth rates as con- difficult, and “top-down” methods (dependent on atmospheric straints. The a priori emissions used to calculate the emissions growth rates measurements) presently only allow us to quantify emissions for the inversion were estimated using data from independent sources (2, integrated globally over all industries. Using a combination of 14, 23, 24). The EDGAR v4.2 emission estimates are also shown (25). The gray shading represents the ± 1- σ uncertainty of the top-down emissions esti- top-down and bottom-up data, we have attempted to compare mate. CO2-eq emissions were calculated using a 100-y global warming po- the recent CO2-eq emissions of NF3 with those of other more tential of 16,600 (right axis). (B) The ratio of (top-down) emissions to widely studied fluorinated GHGs (Table 2). production illustrates the improved efficiency of processes during pro- ± σ The recent emissions of NF3 (in CO2-eq), which are almost duction and/or end use over time. The gray shading represents the 1- entirely from production and end use in SC+ manufacture, are uncertainty of this ratio considering uncertainty in the top-down emissions larger than those of the other fluorinated compounds from SC+ estimate and global production. manufacture, except for CF4 and possibly C2F6. Estimating emissions of CF and C F from SC+ manufacture is very difficult 4 2 6 emitted fluorinated compounds from the SC+ industry, up from

because these gases are also emitted during primary aluminum EARTH, ATMOSPHERIC,

between 13% and 28% in 2005 (Table 2). As well as considering AND PLANETARY SCIENCES (Al) manufacture (28). Assuming that the bottom-up estimates ﬂ + the pace at which NF3 replaces other uorinated gases, the rise for CF4 and C2F6 from SC and Al industries are under- + estimated [Mühle et al. (29) showed that the bottom-up estimates in future NF3 emissions depends on the SC market (likely to ﬁ from either SC+ or Al manufacture, or both, were too low], for grow signi cantly) and on progress in reducing the amount re- the lower SC+ emissions bound we used the emissions fraction leased. The global NF3 production rates are not well docu- SC+/total from the Emissions Database for Global Atmospheric mented, and estimates have been revised over successive years ∼ Research (EDGAR) v4.2 database multiplied by the most recent (3, 7, 23). The latest production estimate from industry is 12 Gg global estimates from atmospheric studies (25, 29). For the up- in 2011 (2); however, an independent report suggests sales are per SC+ emissions bound we subtracted the bottom-up estimates closer to ∼10.5 Gg (24). Using a 2011 production estimate of 12 from Al manufacture from the top-down global total estimates Gg, and scaling emissions in previous years using the relative (28, 29). Given that global emissions for C2F6 have stabilized and year-to-year changes documented in more detailed sources (23, the fact that NF3 is continuing to replace C2F6 (2, 29), it is likely 24), we calculate an emissions/production ratio of over 0.20 for ± B that CO2-eq emissions of NF3 already, or will in the near future, the early 2000s, decreasing to 0.10 0.03 in 2011 (Fig. 3 ). exceed those of C2F6 from SC+ manufacture. These emissions/production ratios (calculated from top-down Emissions of NF3 in 2010 accounted for between 17% and emissions estimates) can be compared with industry estimates 36% of emissions (in CO2-eq) of the most widely used and from NF3 production and different types of end use facilities

Arnold et al. PNAS | February 5, 2013 | vol. 110 | no. 6 | 2031 Downloaded by guest on September 25, 2021 Table 2. Global emissions of the main greenhouse gases used in semiconductor, photovoltaic cell and ﬂat-panel display manufacture (together termed SC+) From leakage during production and From all sources, SC+ manufacture, −1 −1 Tg CO2-eq·y Tg CO2-eq·y *

Gas 2005 2010† 2005 2010†

NF3 (this work) 12.3 ± 3.1 18.4 ± 3.3 12.3 ± 3.1 18.4 ± 3.3

CF4 (29) 77 77 6−46 6−52

C2F6 (29) 30 27 16−25 14−24 SF6 (30–32) 140 160 6.2 7.7

CHF3 (33) 180 130 0.9 0.6

c-C4F8 (34, 35) 5−10 6−11 3−64−7

100-y GWPs used to calculate CO2-eq emissions: NF3 16,600, CF4 7,390, C2F6 12,200, SF6 22,800, CHF3 14,800, and c-C4F8 10,300. Uncertainties are given at the 1-σ level for NF3, otherwise ±20% (1-σ) should be assumed, and very poorly constrained emissions are given with a range. *Calculated from a combination of the top-down global emission estimates and the ratios of emissions from SC+ to the global totals calculated from

EDGAR v4.2 (25). The upper bounds for the CF4 and C2F6 used emission estimates from the production of primary aluminum (28) subtracted from the top-down emissions (29). Fig. 4. Global emission savings by using NF3 vs. C2F6 for semiconductor, † photovoltaic cell, and ﬂat-panel display (together termed SC+) manufacture From 2008 (for CF4 and C2F6) and 2009 (for CHF3) emission estimates. (where a positive emissions saving indicates reduced emissions with use of

NF3 vs. C2F6). The shading covers the range of possible values for each of the (14, 23). SC+ facilities are typically characterized by the size of abatement scenarios and was calculated using the estimated uncertainties in CF emission factors (resulting from C F use), NF production figures, and the wafer they manufacture: The most modern facilities produce 4 2 6 3 our global top-down NF3 emissions. Scenarios 1 and 2 (in green and red, 300-mm-diameter wafers and use NF3 typically with conversion respectively) are described in the text. For the purpose of comparison, scenario fi ∼ ef ciencies of 98% (with some abating the remaining NF3); 3 assumes that if C2F6 had been used, its emissions would have been com- however, older facilities (producing smaller wafers) have conver- pletely abated and that actual emissions of NF3 were as calculated in this study. sion efficiencies of only ∼70% and often do not use abatement measures (14). We therefore expect that a large contribution to global emissions occurs from older SC+ manufacturing plants. to infer the possible global levels of abatement that might have The reduction in the global top-down–derived emissions/production occurred. We therefore considered two cases at the extreme ends ratio is evidence of reduced fugitive release during production of possible abatement levels (Fig. 4). In scenario 1 we considered and/or significant improvements in the utilization efficiency on zero abatement, i.e., that NF3 has not been abated and if C2F6 a global level. However, there is indication in the emissions/ had been used it would not have been abated. Under this sce- production ratios that these improvements are slowing, which, if nario we calculated a maximum emissions saving of 220 Tg · −1 fi true, means that conversion to newer facilities is slowing or that CO2-eq y in 2011. However, this nding is likely an over- the new technology is not as efficient as originally proposed. estimate because we know that abatement in some facilities does Accordingly, if NF production rates follow demand for pro- occur for NF3 and would have occurred for C2F6, and with in- 3 fi duction of SC+ devices, emissions are likely to rise significantly creasing abatement the emissions bene t from using NF3 over in the near future. C2F6 decreases. In scenario 2 we assumed a 98% abatement efficiency for C2F6 and NF3 in new facilities, and from this we −1 Alternative Historical Emissions Scenario Analysis from Electronics estimated a minimum emissions saving of 53 Tg CO2-eq·y in fi Manufacture. NF3 emissions are replacing at least some emis- 2011. However, this nding is likely to be an underestimate sions that would have occurred from continued use of C2F6, because we expect that not all modern facilities would have been which is typically converted to reactive F with efficiencies of abating optimally, and with lower abatement levels the emissions only ∼30% (13, 14). Understanding how the introduction of benefit from using NF3 over C2F6 increases. A third scenario is NF3 changed total emissions of PFCs from SC+ manufacture plotted in Fig. 4 to show the negative emissions saving had C2F6 required a hypothetical analysis of avoided C2F6 emissions. By been completely destroyed during end use; this serves to illus- using our top-down emissions and industry production estimates trate how large the estimated emissions savings in scenarios 1 to constrain the amount of NF3 used during end use, we assessed and 2 are relative to actual current emissions of NF3. Although it the amount of C2F6 that would have been used, and hence the is not possible to further constrain the avoided emissions, our emissions that would have occurred, had NF3 not been available analysis shows that the shift from C2F6 to NF3 has been signifi- SI Materials and Methods (see Fig. S2 and for full method details). cantly beneficial in terms of reducing the total CO2-eq impact of We estimate that these avoided emissions are significant and in SC+ industry PFC emissions over the past decade. 2011 totaled between 3× and 11× the global CO2-eq emissions Conclusions from use of NF3 (Fig. 4). The largest contribution to that uncertainty range is from the Our expanded data set and revised primary calibration show that abatement-level estimates—the proportion that is destroyed af- the global atmospheric abundance of NF3, a potent anthropo- ter use but before emission to the atmosphere. Given that old genic GHG, continues to rise, reaching a mean concentration in ± and new facilities convert NF3 to reactive fluorine with different the global background troposphere of 0.86 0.04 ppt during efficiencies, and that information is unavailable on the pro- mid-2011, and that its annual emission rate has risen throughout −1 portion of NF3 supplied to the different facilities, we were unable the past decade to ∼1.2 Gg·y in 2011. Thorough bottom-up

2032 | www.pnas.org/cgi/doi/10.1073/pnas.1212346110 Arnold et al. Downloaded by guest on September 25, 2021 estimates of emission rates are lacking because requirements for Pollution Analysis of in Situ Measurements. The NAME model was run in reporting are only now being initiated in legislation. Nonetheless, backward mode using UK Met Office global resolution meteorology (∼25 from our analysis of the available independent information km). Tracer particles were released at the sampling site over every 3-h pe- sources, it is clear that bottom-up emissions are currently being riod, and the 19-d back trajectory histories of these particles were estimated. fi In a similar method described by Manning et al. (19) for calculating methane signi cantly underestimated on a global scale. and nitrous oxide baseline values at Mace Head, Ireland, threshold criteria Given the available information on production rates, we calcu- were used to classify the air masses arriving at La Jolla for each 3-h period: lated that the ratio of emissions to production has decreased sig- air arriving via the continent, stagnant local air, or air from equatorial lat- nificantly since the introduction of NF3 in SC+ manufacture, itudes were classified as nonbaseline. probably because of efficiency gains in manufacturing and end use, and a shift toward semiconductor production in more modern fa- Global Emissions Calculation: Inverse Method and Atmospheric Chemistry Transport Model. Given that prior emission rate estimates of NF are likely cilities that use NF3 with very high efficiency (∼98%). Our integrated 3 global analysis therefore probably masks significant variability in inaccurate, and because our atmospheric measurements are relatively sparse, emissions performance across the globe and among industrial sec- we used a growth-based Bayesian inverse method to optimally derive global emissions (20). Instead of using absolute emission rates, this method uses the tors, which makes estimating the emissions saving over use of C2F6 more reliably understood year-to-year growth in emissions. From the total NF3 challenging and uncertain. Nonetheless, our analysis suggests that bottom-up emissions, discussed above, we calculated the annual emissions the savings in recent years from the shift to NF3 are substantial. rate growth and included this as a constraint in the inversion. Continued efforts are needed to increase the utilization effi- We calculated bottom-up emissions estimates from data that are publicly

ciency during application and to minimize emissions in production documented on NF3 production levels (2, 23, 24), and emission factors that so that the full climate benefitofNF3 use can be realized. This shift have been estimated or directly measured (14, 23, 41). Fthenakis et al. (23) will become progressively more important as the demand for discuss the reduction in emission factors (emitted/produced) from two of the largest manufacturers of NF , Air Products and Kanto Denka. Using this in- NF3 rises significantly over the coming years. Together with vol- 3 formation we applied emission factors of 10% from 1967 to 1990, dropping untary and legislative action to curb emissions of NF3, bottom-up inventories will need improving, alongside atmospheric monitoring linearly with respect to time to 7% in 1997, to 5% in 2004, 2% in 2009, and fi 1% in 2012. Calculating a global bottom-up emissions estimate for end use is and modeling work to provide independent emissions veri cation, very difficult given the different types of technology and different levels of especially on regional and national scales. abatement in use (14, 41). We calculated emissions from end use with an emission factor (emitted/supplied) of 0.02 for all years, which is in reasonable Materials and Methods agreement with the available details in the present literature (14, 23). The Instrumentation and Measurement Method. We have adapted a Medusa full time-series is shown in Table S2. The EDGAR v4.2 database published 0.1° preconcentration GC/MS system to make automated measurements of longitude × 0.1° latitude gridded annual emissions from 1978 to 2008 (for NF3 at SIO (5, 36). The measurement cycle lasts 65 min, and each sample which the origin of information is not documented) (25); however, it is bracketed by analyses of a compressed reference gas, resulting in a cali- appears that those estimates are unrealistically low relative to our bottom- brated ambient air measurement every 130 min (5). For archive samples, at up estimates (Fig. 3A; Table S2). least three separate measurements were made, and the measurement error For the inversion we computed the sensitivity of atmospheric mole frac- was calculated as the 1- σ SD of these replicate analyses. tions to global emissions using the AGAGE 12-box model (42–44). Further The measured archived air samples include NH samples collected at La Jolla details on the model structure and inversion method are described in SI

from 1973 to 1991 and at the Trinidad Head, CA, Advanced Global Atmo- Materials and Methods.WeusedaNF3 lifetime estimated by Dillon et al. spheric Gases Experiment (AGAGE) station (41.05°N, 124.15°W) from 1998 to (9) of 490 y, which considers reaction with O(1D) and photolysis by UV in 2010, and SH subsamples from the Cape Grim Air Archive collected at the the stratosphere, and a negligible reaction with OH radical (3, 8–10, 45, Baseline Air Pollution Station at Cape Grim, Tasmania (40.68°S, 144.69°E) 46). The model was run multiple times with global emissions perturbed (37). The ice core samples were collected in the 2011–2012 ﬁeld season at individually by 1 Gg in each year. The sensitivity matrix for the inversion Taylor Glacier, Antarctica, and the air was extracted on-site using previously was then constructed from these outputs so that each element contained described methods (38, 39). The air samples had ages of ∼11,360 B.P (Preboreal), the mole fraction perturbation in a particular model box on a particular ∼11,500 B.P (transition from Preboreal to Younger Dryas), and ∼11,600 B.P. month (where data exist) due to emissions in a particular year. Although

(Younger Dryas; B.P. relative to 1950 A.D.). We found undetectable levels of NF3 our box model could allow us to solve for semihemispheric emissions, the in 2-L samples of these extractions. To further constrain the preindustrial global low frequency in time and space of atmospheric measurements only abundance below the routine detection limit, we preconcentrated and mea- allowed for global emissions to be accurately resolved. The spatial dis- sured 6 L of air from one Preboreal age sample. tribution of emissions was taken from the annual EDGAR v4.2 gridded estimates. These estimates begin in 1978 with 94% of emissions originating − Calibration. We report NF3 concentrations on an updated calibration scale, from 30° 90° latitudes in the NH, and end in 2008 with these latitudes SIO-12, the preparation of which is described by Arnold et al. (5). This scale is contributing 47% of emissions. Remaining emissions originate from the based on four primary gravimetric standards prepared in a whole-air matrix lower NH latitudes (0°−30°). to minimize the risk of bias due to interference with other atmospheric The total measurement uncertainty used in the inversion was calculated as constituents; its estimated overall uncertainty is 2%. For our 2008 meas- the square root of the sum of the following squared errors: errors resulting from sampling frequency, measurement-model mismatch, measurement

urements we used a preliminary volumetric calibration scale that was pre- EARTH, ATMOSPHERIC,

pared using a digital quartz pressure gauge and a ﬁxed volume to add ∼3 precision, and scale propagation (see SI Materials and Methods for further AND PLANETARY SCIENCES

ppt of NF3 from a volumetric NF3/N2O mixture to a large aliquot of whole air details). The inversion was carried out 1,000 times using 1,000 alternative (5). At the time we recorded a small −0.597 torr zero offset in the Paros- sensitivity matrices and measurement vectors constructed from the ran- cientiﬁc 2300 torr full-scale pressure gauge but failed to apply that offset to domly perturbed parameters in the model and calibration scale, respectively. the unusually small 2.183 torr pressure measurement associated with the The uncertainty associated with the average emissions from the 1,000 inver-

required very small aliquot of NF3. As a result, we underreported the amount sions was combined with the inherent uncertainty from the inversion (taken ﬁ σ of NF3 added to the 2008 standard by 21.5%, and thus underestimated the from the error covariance matrix) to produce the nal 1- uncertainties. amount of NF3 in the atmosphere by this same percentage. The 2008 pre- To estimate the annual average global tropospheric mixing ratios and liminary standard has now been measured directly against the SIO-12 gravi- related uncertainties (Table 1), the model was run 1,000 times using the metric scale (4), the results of which show the 2008 numbers to be 19.9% too annual emissions with normal distributions based on the associated uncer- low. The revised gravimetric standard thus agrees with the zero-offset cor- tainties derived from the inversions. Each run produced mole fraction out- rected 2008 volumetric standard within the expected ∼3% combined un- puts for the lower troposphere in each of the semihemispheres. The

certainty of the two methods (5, 6). Accordingly, the NF3 measurements growth rates and associated uncertainties were calculated similarly using reported by Weiss et al. (6) that are also used in the present work have been the ﬁrst derivative of each of the 1,000 model outputs ﬁttedwithacubic corrected to the SIO-12 gravimetric scale. It must also be stressed that, except spline curve.

for the initial SIO-2008 NF3 scale, all SIO calibration scales used for atmospheric trace gas measurements in the AGAGE program are entirely gravimetric and Emissions-Saving Analysis. The following assumptions were used in this

are not vulnerable to such pressure measurement uncertainties (40). emissions saving analysis. First, emissions of NF3 during end use only were

Arnold et al. PNAS | February 5, 2013 | vol. 110 | no. 6 | 2033 Downloaded by guest on September 25, 2021 calculated by accounting for emissions during production: The upper limit ACKNOWLEDGMENTS. We thank Matt Rigby (University of Bristol) and Anita for emissions during production was taken as half the total top-down estimate, Ganesan (Massachusetts Institute of Technology) for their discussions of and the lower limit was taken from bottom-up estimates of emission factors modeling and inverse methods; Stephanie Mumma for filling standard and production (2, 23, 24). Second, utilization efficiencies were 98% and 70% tanks; our AGAGE colleagues at the University of Bristol, Empa, Massachu- setts Institute of Technology, CSIRO, and Seoul National University for their for NF3 in modern and older end-use facilities, respectively, corresponding to utilization efficiencies of 44% and 30% for C F , respectively (14). Third, when general enthusiasm and support; and Peter Maroulis and Robert Ridgeway 2 6 of Air Products Corporation for many useful discussions related to this work. abatement was considered it was only in modern facilities and assumed We are also very grateful for the time and effort spent on our submission by 98% for both NF and C F . Fourth, 10% by mass of the C F supplied was 3 2 6 2 6 the three anonymous reviewers. This research was carried out as part of the fi assumedtobeconvertedintoCF4 (14), and an abatement ef ciency of international AGAGE research program, supported in the United States by – fi 40 90% of this CF4 was assumed in modern facilities (14, 47). And fth, the Upper Atmosphere Research Program of the National Aeronautics and the production of CF4 from silicon carbide cleaning was small and con- Space Administration. Support was also provided by UK Department of Energy sistent between methods (14), and was therefore not considered. Further and Climate Change Contract GA0201, CSIRO, and the Australian Government details of the method used to calculate the emissions saving are given in Bureau of Meteorology. Ancient air collection from Taylor Glacier, Antarctica, Fig. S1 and SI Materials and Methods. was supported by National Science Foundation Award 0839031 (to J.P.S.).

1. Montzka SA, Dlugokencky EJ, Butler JH (2011) Non-CO2 greenhouse gases and climate 25. European Commission Joint Research Centre/Netherlands Environmental Assessment change. Nature 476(7358):43–50. Agency (2011) Emissions Database for Global Atmospheric Research (EDGAR) v4.2. 2. McCoy M (2011) A clean ﬁght. Chem Eng News 89(48):21–23. Available at http://edgar.jrc.ec.europa.eu/.

3. Prather MJ, Hsu J (2008) NF3, the greenhouse gas missing from Kyoto. Geophys Res 26. Hofmann DJ, et al. (2006) The role of carbon dioxide in climate forcing from 1979 to Lett 35(12):L12810. 2004: Introduction of the Annual Greenhouse Gas Index. Tellus B Chem Phys Meterol 4. United Nations Framework Convention on Climate Change (UNFCCC) (1997) Kyoto 58(5):614–619.

Protocol to the United Nations Framework Convention on Climate Change (UNFCCC, 27. Olivier JGJ, Janssens-Maenhout G, Peters JAHW (2012) Trends in Global CO2 Emis- Bonn). sions: 2012 Report (PBL Netherlands Environmental Assessment Agency, The Hague). 5. Arnold T, et al. (2012) Automated measurement of nitrogen trifluoride in ambient air. 28. International Aluminum Institute (2011) Results of the 2010 Anode Effect Survey: Anal Chem 84(11):4798–4804. Report on the Aluminum Industry’s Global Perfluorocarbon Gas Emissions Reduction 6. Weiss RF, Mühle J, Salameh PK, Harth CM (2008) Nitrogen trifluoride in the global Programme (Intl Aluminum Inst, London), pp 1–56. atmosphere. Geophys Res Lett 35(20):L20821. 29. Mühle J, et al. (2010) Perfluorocarbons in the global atmosphere: Tetrafluoro- 7. Robson JI, Gohar LK, Hurley MD, Shine KP, Wallington TJ (2006) Revised IR spectrum, methane, hexafluoroethane, and octafluoropropane. Atmos Chem Phys 10(11): fi fl radiative ef ciency and global warming potential of nitrogen tri uoride. Geophys 5145–5164. Res Lett 33(10):L10817. 30. Levin I, et al. (2010) The global SF6 source inferred from long-term high precision 8. Zhao ZJ, Laine PL, Nicovich JM, Wine PH (2010) Reactive and nonreactive quenching of atmospheric measurements and its comparison with emission inventories. Atmos 1 O( D) by the potent greenhouse gases SO2F2,NF3, and SF5CF3. Proc Natl Acad Sci USA Chem Phys 10(6):2655–2662. 107(15):6610–6615. 31. Hall BD, et al. (2011) Improving measurements of SF6 for the study of atmospheric 9. Dillon TJ, et al. (2011) Removal of the potent greenhouse gas NF3 by reactions with transport and emissions. Atmos Meas Tech 4(11):2441–2451. 1 the atmospheric oxidants O( D), OH and O3. Phys Chem Chem Phys 13(41):18600– 32. Rigby M, et al. (2010) History of atmospheric SF6 from 1973 to 2008. Atmos Chem Phys 18608. 10(21):10305–10320. fi 10. Baasandorj M, Hall BD, Burkholder JB (2012) Rate coef cients for the reaction of O 33. Miller BR, et al. (2010) HFC-23 (CHF ) emission trend response to HCFC-22 (CHClF ) 1 3 2 ( D) with the atmospherically long-lived greenhouse gases NF3,SF5CF3, CHF3,C2F6, production and recent HFC-23 emission abatement measures. Atmos Chem Phys c-C F ,n-CF , and n-C F . Atmos Chem Phys 12(23):11753–11764. 4 8 5 12 6 14 10(16):7875–7890. 11. Bronfin BR, Hazlett RN (1966) Synthesis of nitrogen fluorides in a plasma jet. Ind Eng 34. Oram DE, et al. (2012) Long-term tropospheric trend of octafluorocyclobutane (c-C4F8 Chem Fundam 5(4):472–478. or PFC-318). Atmos Chem Phys 12(1):261–269. 12. Henderson PB, Woytek AJ (1994) Nitrogen–nitrogen trifluoride. Kirk–Othmer Ency- 35. Saito T, Yokouchi Y, Stohl A, Taguchi S, Mukai H (2010) Large emissions of per- clopedia of Chemical Technology, eds Kroschwitz JI, Howe-Grant M (Wiley, New fluorocarbons in East Asia deduced from continuous atmospheric measurements. York), 4th Ed. Environ Sci Technol 44(11):4089–4095. 13. Johnson AD, Entley WR, Maroulis PJ (2000) Reducing PFC gas emissions from CVD 36. Miller BR, et al. (2008) Medusa: A sample preconcentration and GC/MS detector sys- chamber cleaning. Solid State Technology 43(12):103–114. tem for in situ measurements of atmospheric trace halocarbons, hydrocarbons, and 14. Reduction of Perfluorocompound (PFC) Emissions (2005) 2005 State of the Technol- sulfur compounds. Anal Chem 80(5):1536–1545. ogy Report (Intl SEMATECH Manufacturing Initiative, Albany, NY). 37. Krummel PB, Langenfelds RL, Fraser PJ, Steele LP, Porter LW (2007) Archiving of Cape 15. Howard-Grenville JA (2002) Institutional evolution: The case of the semiconductor Grim Air (Bur Meteorol, Melbourne). industry voluntary PFC emissions reduction agreements. Organizations, Policy and the 38. Petrenko VV, et al. (2008) A novel method for obtaining very large ancient air sam- Natural Environment: Institutional and Strategic Perspectives, eds Hoffman AJ, ples from ablating glacial ice for analyses of methane radiocarbon. J Glaciol 54(185): Ventresca MJ (Stanford Univ Press, Stanford, CA). 233–244. 16. California Global Warming Solutions Act (2006) Assembly Bill 32 (Government of 39. Petrenko VV, et al. (2009) 14CH measurements in Greenland ice: Investigating last California). 4 – 17. United Nations Framework Convention on Climate Change (UNFCCC) (2012) Report of glacial termination CH4 sources. Science 324(5926):506 508. the Conference of the Parties Serving as the Meeting of the Parties to the Kyoto 40. Prinn RG, et al. (2000) A history of chemically and radiatively important gases in air – Protocol, Seventh Session. Addendum: Part 2: Action Taken by the Conference of the deduced from ALE/GAGE/AGAGE. J Geophys Res Atmos 105(D14):17751 17792. Parties Serving as the Meeting of the Parties to the Kyoto Protocol, Seventh Session 41. Bartos S, et al. (2006) Electronics industry emissions. Intergovernmental Panel on (United Nations, Geneva). Climate Change (IPCC) Guidelines for National Greenhouse Gas Inventories (IPCC, 18. Weiss RF, Prinn RG (2011) Quantifying greenhouse-gas emissions from atmospheric Geneva), Vol 3, Chap 6. measurements: A critical reality check for climate legislation. Philos Trans R Soc A 42. Cunnold DM, et al. (1994) Global trends and annual releases of CCl3F and CCl2F2 es- Math Phys Eng Sci 369(1943):1925–1942. timated from ALE/GAGE and other measurements from July 1978 to June 1991. – 19. Manning AJ, O’Doherty S, Jones AR, Simmonds PG, Derwent RG (2011) Estimating UK J Geophys Res Atmos 99(D1):1107 1126. methane and nitrous oxide emissions from 1990 to 2007 using an inversion modeling 43. Cunnold DM, et al. (1983) The Atmospheric Lifetime Experiment. 4. Results for CF2Cl2 approach. J Geophys Res Atmos 116:D02305. based on 3 years data. J Geophys Res C Oceans Atmos 88(NC13):8401–8414. 20. Rigby M, Ganesan AL, Prinn RG (2011) Deriving emissions time series from sparse 44. Cunnold DM, et al. (1997) GAGE/AGAGE measurements indicating reductions in – atmospheric mole fractions. J Geophys Res Atmos 116:D08306. global emissions of CCl3F and CCl2F2 in 1992 1994. J Geophys Res Atmos 102(D1): 21. Forster P, et al. (2007) Contribution of Working Group I to the Fourth Assessment 1259–1269. Report of the Intergovernmental Panel on Climate Change, 2007, eds Solomon S, 45. Molina LT, Wooldridge PJ, Molina MJ (1995) Atmospheric reactions and ultraviolet et al. (IPCC, Geneva), Chap 2. and infrared absorptivities of nitrogen trifluoride. Geophys Res Lett 22(14): 22. Montzka SA, et al. (2011) Scientific Assessment of Ozone Depletion: 2010, Global 1873–1876. Ozone Research and Monitoring Project—Report No. 52, ed Ennis CA (World Mete- 46. Dillon TJ, Horowitz A, Crowley JN (2010) Cross-sections and quantum yields for the orolog Org, Geneva), Chap 1. atmospheric photolysis of the potent greenhouse gas nitrogen trifluoride. Atmos 23. Fthenakis V, et al. (2010) Life-cycle nitrogen trifluoride emissions from photovoltaics. Environ 44(9):1186–1191. Environ Sci Technol 44(22):8750–8757. 47. Yang CF-O, Kam SH, Liu CH, Tzou J, Wang JL (2009) Assessment of removal efficiency fl 24. Shon-Roy L (2012) NF3 Average Selling Prices 2003–2011 (Techcet Group, Rancho of per uorocompounds (PFCs) in a semiconductor fabrication plant by gas Santa Fe, CA). chromatography. Chemosphere 76(9):1273–1277.

2034 | www.pnas.org/cgi/doi/10.1073/pnas.1212346110 Arnold et al. Downloaded by guest on September 25, 2021