TECHNICAL SUPPORT FOR MODIFICATIONS TO THE FLUORINATED GREENHOUSE GAS EMISSION ESTIMATION METHOD OPTION FOR SEMICONDUCTOR FACILITIES UNDER SUBPART I
       Office of Air and Radiation, U.S. Environmental Protection Agency
                                       
                                  August 2012
                                       







1.   Introduction
This document presents technical support for the proposed changes to the methods for estimating fluorinated greenhouse gas (F-GHG) emissions from the plasma etching, chamber cleaning, and wafer cleaning process types at semiconductor manufacturing facilities under subpart I, Electronics Manufacturing, of the Greenhouse Gas Reporting Program (GHGRP). It also describes proposed changes to the provisions for verifying gas apportioning models for all electronics facilities and calculating abatement system uptime when default emission factors (EFs) are used to estimate GHG emissions for all electronics facilities.
This document is based on  information from Technical Support Document for Process Emissions from Electronics Manufacture (e.g., Micro-electro-mechanical systems, liquid crystal displays, photovoltaics, and semiconductors): Proposed Rule for Mandatory Reporting of Greenhouse Gases, Revised  -  November 2010 (hereafter, the November 2010 TSD, available in docket EPA-HQ-OAR-2009-0927) and non-confidential materials submitted to the EPA by the Semiconductor Industry Association (SIA).
In addition to the methods described in this document to estimate F-GHG emissions, the EPA is proposing to allow semiconductor manufacturers (and other electronics manufacturers) the option to estimate F-GHG emissions using a periodic stack testing option. For more l information about the stack testing option, please refer to Technical Support for the Stack Test Option for Estimating Fluorinated Greenhouse Gas Emissions from Electronics Manufacturing Facilities under Subpart I. 
2. Estimating F-GHG Emissions from Semiconductor Fabs Manufacturing Wafers 300 mm or Smaller
The EPA is proposing to modify the methods by which facilities are required to estimate F-GHG emissions resulting from plasma etching, chamber cleaning, and wafer cleaning processes used at semiconductor manufacturing facilities, in particular those that manufacture wafers with diameters measuring 300 mm or smaller (i.e., 200 mm and 150 mm). The EPA is proposing to remove the provisions that require the use of the "Tier 2d method" for the largest semiconductor manufacturing facilities, which requires: 
      i.       The use of default EFs to estimate F-GHG emissions resulting from chamber cleaning and wafer cleaning processes, and 
      ii. The use of recipe-specific EFs to estimate F-GHG emissions resulting from etching processes.
The Tier 2d method focuses on recipe-specific EFs for plasma etching processes because of the gaps in the available EF data for the etching process type used in the industry at the time subpart I was finalized in December 2010. 
Since then, SIA has undertaken an effort to gather and collect a much larger data set of etch EFs from semiconductor manufacturing facilities and tool manufacturers (see Section 2.1) (SIA, 2012a and SIA, 2012b). In addition, SIA raised concerns with the burden, technical feasibility, and the protection of confidential business information (CBI) associated with the Tier 2d method (and other methods based on recipe-specific EFs).  Please refer to the reference SIA, 2012a for a detailed discussion about SIA's burden, feasibility, and CBI concerns. Because of these concerns, and after evaluation and consideration of alternative methods and additional data submitted to the EPA (summarized below), the EPA is proposing to remove the Tier 2d method from subpart I.
With this change, the EPA is proposing to remove the distinction between large and other semiconductor manufacturing facilities, such that all semiconductor manufacturing facilities subject to the GHGRP and opting to not use a periodic stack testing method would be required to use F-GHG estimation methods based entirely on default EFs (i.e., requirements would be independent of facility size). In the December 2010 subpart I, the EPA had required only the largest semiconductor manufacturing facilities to use the Tier 2d method to ensure that burden was commensurate with a facility's potential emissions (November 2010 TSD). In proposing to use default EFs to estimate emissions from all process types, the burden on larger facilities would be reduced significantly, and the need to distinguish between "large" and "other" semiconductor manufacturing facilities would no longer be required. For more information specifically on the reduced economic burden from the proposed removal of the recipe-specific option, please see Economic Impact Analysis for the Mandatory Reporting of Greenhouse Gas Emissions F Gases: Subpart I.
Due to the aforementioned concerns voiced by SIA about recipe-specific EFs, the EPA is also proposing to remove the provisions that would allow all electronics manufacturing facilities (regardless of the type of electronic product manufactured or size of the facility) the option to use recipe-specific EFs to estimate emissions from all F-GHG emitting processes (etching, chamber cleaning, and wafer cleaning) and N2O-using processes (chemical vapor deposition processes and other N2O-using processes). This is also referred to as a "Tier 3 method" (See the document entitled Technical Support for Other Technical Issues Addressed in Revisions to Subpart I for more information on estimating N2O emissions).
   2.1. Developing Revised Default Etch Emission Factors
In previous subpart I rulemaking efforts the EPA has repeatedly evaluated approaches using default EFs for process types and subtypes. In general, methods based on default EFs are relatively easy to use (particularly if the number of process types and/or subtypes is limited), but they may result in estimates with high uncertainties if the EFs are not representative at both the industry and fab level.  Default EFs can fail to represent industry-wide EFs if they are based on too few process measurements or on process measurements that are not representative of the industry as a whole. Default EFs can also fail to represent fab-level EFs if actual EFs vary significantly across fabs, and if the default EFs do not capture that variability.  
Uncertainties resulting from few or unrepresentative measurements can be addressed by increasing the number of measurements and/or their representativeness of the industry; uncertainties resulting from the failure of the default EFs to capture variability can be addressed by developing factors for additional and/or different categories. Note, however, that the addition of categories (e.g., increasing the number of process types or sub-types) can introduce other uncertainties, e.g., by decreasing the number of measurements per category or requiring estimates of gas consumption for more categories.   
In the fall of 2011, as part of the process of reconsidering the Tier 2d method, SIA provided the EPA with a data set of etch EFs that was much larger than the data set available to the EPA in 2010.  The number of EF data points available to the EPA in 2010 and 2011 is shown in Table 1. 
            Table 1.  Comparison of Available EF (Utilization) Data
                                       
             Number of EF Data Points Available to the EPA in 2010
              Number of Data Points Available to the EPA in 2011
                                     Etch
                                      96
                                      976
                                 Chamber Clean
                                      163
                                      163
                                  Wafer Clean
                                       5
                                       5
                                     Total
                                      264
                                     1144
Note: The counts represent EFs related to utilization of an input gas. Counts of by-product EFs are not represented.
SIA stated that the etch data provided to the EPA in 2011 are considered representative of the industry as a whole because of the large number of data points, the presence of a measured EF for almost every gas, and the inclusion of data from the four primary suppliers of etch equipment in the data set. However, the EPA notes that it is not possible to assess fully the representativeness of the data in terms of the parameters that potentially affect EFs. Process parameters such as radio frequency (RF) power, pressure, feature types etched (e.g., gate etch, trench etch, etc...), and stabilization time were generally not available (as in the case for some data that were considered to be confidential) or were not provided to the EPA in sufficient quantities to assess the impact of such parameters on EFs (SIA, 2011a). Also, the sources of the data (i.e., the device and tool manufacturing entities), which affect its representativeness, were not known. Nevertheless, the expanded etch data set of EFs was the largest ever assembled for this industry. In light of the new etch data, the EPA concluded that revised default etch EFs could be calculated and proposed in place of the etch recipe-specific EFs required as part of the Tier 2d method. 
      2.1.1. Options for Emission Factor Development Methods and Conventions  
As a first step in developing a method to replace the Tier 2d method, the EPA evaluated the following methods and conventions for developing the default EFs proposed in the revised subpart I:
* Utilization EFs (kilograms of gas emitted/kg of gas used): The EPA calculated the simple arithmetic average of all available etch data grouped according to two different EF models (see Section 2.1.2). 
* By-product EFs: The EPA considered two conventions in assigning by-product emissions to etch process input F-GHGs for multi-gas processes: 
   o       Dominant gas convention (kg of by-product emitted/kg of dominant gas used): This convention, which was adopted in the 2006 Intergovernmental Panel on Climate Change (IPCC) Guidelines, assigns all by-product emissions to the dominant gas used in a process, or the gas with the largest mass flow for multi-gas processes. This convention is what was used in the December 2010 subpart I.
   o    All input F-GHGs convention (kg of by-product emitted/kg of all input gases used): This convention assigns by-product emissions to input F-GHGs used in a process by dividing the measured mass emitted of a specific by-product by the total mass of all input F-GHGs and assigning this by-product factor to each input F-GHG. This convention was used by SIA when compiling the newly available etch EF data (SIA, 2012b).
It is generally not possible to ascertain which input gases specifically give rise to a particular by-product for multi-gas processes, and both methods are only conventions that have some drawbacks.  In particular, while the first convention could lead to an over-assignment of by-product emissions to a dominant input gas, the second could lead to assignment of by-product emissions to input gases that do not chemically generate a particular by-product (e.g., CF4 is generally not a by-product from the use of the input etch gas NF3, unless carbon-containing films are being etched). However, because both conventions conserve mass (i.e. theoretically the same amount of by-products will be calculated by both methods for a particular recipe when using default EFs), and given that the etch data available are in the format to support the "all input F-GHGs convention," the EPA is proposing to use the second (all input gas) convention for etch EFs. Note that data were not available to the EPA to revise the chamber cleaning default EFs to follow the all input F-GHGs convention, but this is not expected to be a substantive issue at this time because the majority of the chamber clean processes for which data is available rely on only one input gas. Regardless of the convention used, the equations used to calculate by-product emissions are the same. The only difference in the conventions is how by-product emissions are attributed to input gases.
After selection of the all input gas convention, two averaging methods were considered for developing the by-product EFs:
   1. Simple arithmetic mean of all available by-product EF data: This method averages all available by-product EF data (by by-product) for each gas, wafer size, process type/sub-type combination. 
   2. Simple arithmetic mean of all available by-product EF data, including use of zeros when measurements did not detect by-product emissions: This method averages all available by-product EF data (by by-product) including measurements that did not detect by-product emissions for each gas, wafer size, process type/sub-type combination (represented by a zero value). For more information on this approach, please refer to Draft EFs for Refined Semiconductor Manufacturing Process Categories (EPA-HQ-OAR-2009-0927-0073).
The EPA compared the resulting by-product EFs from using both of these averaging methods. It was found that the number of zeros that would have to be included in the by-product EF averages (i.e., the number of utilization EFs with no reported measured by-product EF) ranged from 1 to over one hundred, depending on the grouping used (see Section 2.1.2). Including the number of zeros towards the higher side of this range will have a substantial impact on the calculated average by-product EFs. The EPA's comparison showed that including versus not including the zeros for cases where no detected by-product was reported resulted in a 38 to 45 percent difference on average for by-product EFs, depending on the grouping used (see Section 2.1.2 for grouping definitions).
The EPA questioned SIA about the method for including the zero values for non-detected by-products, and in SIA's Report to EPA on Etch Factor Proposal for Fab GHG Emissions Reporting, SIA states the following "...For example, take the case of ten separate tests conducted on an input gas X. If nine of these tests produce no detectable by-product of gas Y and one does, then the current method establishes the by-product EF based on the one test of the by-product Y from gas X and does not take into consideration the measurements where by-products are not detected. SIA makes no specific recommendation on this matter." Because the EPA was not certain whether zeros indicated that particular by-products were not looked for or whether they were looked for but not detected, the EPA is conservatively proposing by-product EFs that do not include zeros. However the EPA's approach in the December 2010 subpart I (including the zeros) may be considered further; not accounting for the zeros in the by-product EF development may result in overstating by-product emissions. Therefore, the EPA is seeking comment on the method for averaging data to develop by-product EFs in the preamble to the proposed rule.
      2.2.1. Grouping of Etch Emission Factor Data to Develop Default Emission Factors
The second step in developing the revised default etch EFs was to evaluate how individual EF data should be grouped to develop average, default EFs. This was evaluated by the EPA and SIA in looking at EF data associated with the utilization of input gases used for etch and available information for parameters known to impact EFs. The parameters for which the EPA had sufficient information on which to make conclusive decisions about their effect on etch EFs were:
   * Input gas
   * Wafer size
   * Film type etched (oxide, nitride, silicon, or metal)
Note that for 141 etch utilization EFs, or approximately 14 percent of all EF data, the EPA had information on the type of feature etched. The EPA analyzed this parameter to the extent possible and the results indicated the feature type etched was a parameter that had a potentially significant influence on EFs, but, as identified above, feature type information was quite limited. It should also be noted that SIA did a full assessment of the parameters that had the most impact on EFs, which the EPA was not able to review due to confidentiality concerns. SIA's analysis and report concluded that the three parameters listed above were the most relevant for grouping etch EFs (SIA, 2011a and SIA, 2012a). 
SIA also analyzed the best way to group the etch EF data according to the three parameters. SIA's analysis resulted in the conclusion that the best two models for grouping EFs were: 
*       Model 1: Develop default EFs by gas and wafer size for one broad etching process type, and 
*       Model 2: Develop default EFs by gas and wafer size for four etching process sub-types based on the types of films being etched (silicon etch, nitride etch, oxide etch, and metal etch).
To confirm SIA's results the EPA used two statistical approaches in analyzing the most meaningful way to group etch EF data to develop default EFs taking into account the above listed parameters known to impact EFs (input gas, wafer size, and film type).:
First the EPA ran a "best fit analysis" (i.e., an analysis of R-squared values, assuming normal distributions). This analysis indicated that two well-fitting models for developing etch EFs would indeed be Models 1 and 2 as described above. The EPA's analysis showed that:
   1. Input gas and wafer size (Model 1) provided an R-squared value of 58.6 percent, or 
   2. Input gas and wafer size and film type (Model 2) provided an R-squared value of 64 percent.
The EPA then performed an analysis of variance (ANOVA) and looked at the p-values for the different models in order to confirm that the difference in R-squared values for the two models was statistically significant, that is, that Model 2 was really providing a better fitting model. The results showed a slight reduction in uncertainty for Model 2 compared to Model 1, also consistent with SIA's analysis (SIA, 2012a). SIA analyzed other parameters such as pressure, but this information is considered confidential and was therefore not used in the EPA's analyses. See SIA, 2012a for more information. 
As another tool to assess the two models considered to group EFs, the EPA compared the uncertainties in the distributions of  the GWP-weighted EFs for each etch input gas (including utilization and by-products EFs), according to the groupings described above for Model 1 and Model 2, and using the conventions and averaging methods described in Section 2.1.1. For each GWP-weighted EF, the uncertainty at the 95 percent coverage level was calculated by multiplying the standard deviation of the factor by 1.96 and dividing by the mean. For Model 1, the uncertainty in EFs simply results from the deviation of the individual EFs around the mean (default) factor, so that, using the fitted distribution, 95 percent of individual EFs are estimated to deviate from the mean by no more than the uncertainty percentage. For Model 2 (and for the purpose of comparing Model 2 to Model 1), the uncertainty was calculated for each film type etched, then averaged over the four film types. For the EFs for which it was not possible to calculate an uncertainty value because only one data point in the database existed, a standard deviation equal to half the value of the data point (and thus an uncertainty equal to 98 percent) was assumed. The results of this analysis are presented in Figure 1 (200 mm) and Figure 2 (300 mm) below. 

      Figure 2. Uncertainty of GWP-Weighted EFs for 300 mm Etch Processes
      Figure 1. Uncertainty of GWP-Weighted EFs for 200 mm Etch Processes
As can be seen by comparing the etch EF uncertainties for Models 1 and 2, Model 2, which takes into account the film type etched (metal, nitride, oxide, and silicon), does not always provide better certainty in the EFs than the less refined Model 1 which uses one broad etching process type. For example, the average uncertainty of Model 2 for CF4 (200 mm processes, first data point on the left of Figure 1) was worse than Model 1 (69 percent vs. 54 percent). However, the opposite trend is observed for 300 mm, where the uncertainty of the CF4 EF for Model 2 was better than for Model 1 (46 percent versus 55 percent). On average (over all gases and both wafer sizes) the EF uncertainty for Model 1 is slightly higher than for Model 2 (97 percent versus 93 percent). However, opposite trends are actually observed as a function of wafer size: for 200 mm, Model 1 provides slightly lower uncertainty than Model 2 (92 percent versus 95 percent respectively); For 300 mm, Model 1 provides slightly higher uncertainty than Model 2 (101 percent versus 92 percent respectively). 
In many cases, the higher uncertainties in the Model 2 EFs are a result of spreading the data among more categories than in Model 1, lowering the number of data points in each category.  Several gas/wafer/film-type categories ended up with one data point because for most gas/wafer size combinations, one film type tends to dominate others in terms of the amount of available EF data.  As discussed above, an uncertainty of approximately 98 percent was assumed in cases where there was only one EF for a category.   (The uncertainty values for Model 2 for the cases with more than 1 data point were both lower and higher than 98 percent, with an overall average of 85 percent.)  Under the averaging convention used to determine the distribution of the Model 2 EFs, all film types were given equal weight. This led to some gas/wafer/film combinations that had a small number of highly variable values having a large impact on GWP-weighted EF uncertainties.  
Overall, in reviewing the analyses described above, it can be concluded that any reductions in bias and uncertainty related to Model 2 EFs are not always consistent and appear minimal in light of the its added complexity and the often poorly estimated uncertainties. Additionally, while these analyses indicated a slight preference for grouping etch EF data by input gas, wafer size, and film type to develop default EFs, the EPA was mindful that using this approach would increase the burden associated with apportioning gas consumption to different film types, as compared to default etch EFs grouped only by input gas and wafer size. Therefore the EPA found it necessary to further assess the EFs models in terms of how they would impact the uncertainty in overall fab emission estimates before making any decisions about which model to use in proposing revised default etch EFs. 
   2.1. Evaluation and Comparison of Emission Estimates Resulting From EF Models 1 and 2 
To further assess the two etch EF models and determine which model would be most suitable to propose to use for an emissions calculation and reporting method, the EPA analyzed the overall uncertainty of emission estimates developed using both models. For completeness in this uncertainty analysis, the EPA also considered emissions from chamber cleaning processes. As seen in Table 1 there were no new data available to the EPA to support revised default EFs for the chamber cleaning process sub-types established in the December 2010 subpart I (in-situ thermal, in-situ plasma, and remote plasma). Therefore, in this uncertainty analysis the EPA used the same chamber clean EF data used to establish the current chamber clean EFs. The EPA did not consider emissions resulting from gas consumed in wafer cleaning processes (which are required to be estimated under subpart I) because wafer cleaning F-GHG consumption is estimated to represent approximately one percent of total fab F-GHG consumption, and is expected to similarly represent a small percentage of fab total emissions (SIA, 2012a). 
To evaluate and compare the two EF models (for estimating emissions resulting from etching and chamber cleaning processes) in regards to their impact on the uncertainty in total emission estimates the EPA employed a Monte Carlo simulation method. In line with the IPCC recommendations, a Monte Carlo-based method was chosen over a propagation of error method because the uncertainty of default EFs and other input parameters can be high (>20 percent at 95 percent confidence level). The key data used in this analysis were the available etch and chamber clean EF data and gas consumption data provided by SIA to the EPA. The use of abatement systems (and the uncertainty associated with the use of abatement) was not taken into account in the EPA's analysis (i.e., all emissions estimated were unabated). 
      2.1.1. Emissions Factors Inputs
As discussed in Section 2.1, a key factor in determining the precision of a default-EF based method is the uncertainty of the EFs. While large standard deviations are generally observed for EFs in the semiconductor industry due to significant process-to-process variations (see Figures 1 and 2), it is nevertheless possible to arrive at a reasonable level of precision for fab-level emission estimates, so long as the default EFs are representative of the mix of processes run at a particular fab. Indeed, while recipe-specific EFs will significantly vary from one process or one type of tool to another, the impact of that variability will be reduced when considering that within each gas/wafer-size category, semiconductor facilities  may have multiple processes with distinctly different EFs whose differences will at least partially "average out", approximating the mean (default) EFs."  Thus, to account for this phenomenon, the EPA introduced the variable "n," which provides a measure of the representativeness and the statistical significance of the mean values of the EFs, when calculating emissions at the fab level. The EPA considered various assumptions about the uncertainty of EFs to model fab-level uncertainty: The uncertainty associated with the average default EFs was calculated as 1.96 times the standard deviation of a certain EF data group sample (i.e., wafer size and input gas, for Model 1, and wafer size, input gas, and film type for Model 2) divided by the mean and by the square root of n, where it is assumed that n represents the number of processes within an EF grouping that have distinctly different EFs. The EPA studied four cases for n: 
   1. n=NDB represents the case where n equals the number of individual data points in the data set for each wafer size and each gas (Model 1) and each wafer size, gas, and film type etched (Model 2) combination.  
   2. n=SIA3 represents the case where n represents an average count of the etch tool types (assuming each tool type runs a process or set of processes that results in a distinctly different EF) and etch process combinations by fab and assumes that the number of CVD tool chamber cleaning types equals three. The etch tool type n values were provided by SIA to the EPA.  
   3. n=SIA1 represents the case where n represents an average count of the etch tool types and etch process combination by fab and assumes only one type of CVD tool chamber cleaning (i.e., NF3 remote clean only). The etch tool type n values for this case are the same as for the case where n=SIA3.   
   4. n=1 represents the "worst case scenario"; This case was selected to model the boundary condition where only one process exists within a particular EF grouping (e.g., 200 mm, CF4, oxide etch) at a particular fab, and where the uncertainty of the overall estimate would not take into account the fact that semiconductor facilities run a variety of processes that, on average, may approximate the mean (default) EFs.

      2.4.1. Gas Consumption Inputs
In addition to the uncertainty and representativeness of the EFs, another main issue driving the precision of emission estimates at the fab level is the uncertainty related to apportioning of gas consumption. Most facilities do not track consumption on a tool or process type basis, but only measure consumption by gas at the fab (gas cylinder) level, where it is not possible to distinguish consumption of a particular gas related to one specific EF (e.g. the amount of C2F6 used for CVD chamber cleaning cannot be distinguished from the amount of C2F6 consumed for etch).  Thus, it is necessary to rely on an apportioning model to estimate the consumption of each gas for each process type or sub-type (see Section 4 for more information on apportioning model development for reporting purposes). To estimate the impact of the uncertainty related to gas consumption apportioning, the EPA relied on gas consumption data provided by SIA for the following profiles (see SIA, 2012a):
   * Fab-level gas consumption for two 200 mm fabs (Fabs B and C) and two 300 mm fabs (Fabs A and E) broken out according to Model 1 for 2009 and 2010.
   * "Industry-level" gas consumption for the 200 mm wafer size broken out according to Models 1 and 2 for 2009 and 2010. (The term industry level refers to an aggregate group of seven 200 mm fabs).
   * "Industry-level" gas consumption for the 300 mm wafer size broken out according to Models 1 and 2 for 2009 and 2010. (The term industry level refers to an aggregate group of five 300 mm fabs).
The EPA's analysis took into account the uncertainties associated with apportioning of gas consumption to process types/sub-types (see below), and total gas consumption estimates. The EPA evaluated 4 cases of assumed gas apportioning uncertainties (95 percent confidence intervals), represented by the variable f (1.96 multiplied by the standard deviation of the gas proportion), including:
   1. f = 5 percent (based on the apportioning model verification standard from the December 2010 subpart), 
   2. f = 10 percent, 
   3. f = 20 percent, and 
   4. f = SIA represents a range of apportioning uncertainties estimated by SIA, varying between 56 and approximately 240 percent (SIA, 2011c and SIA, 2012a) To estimate apportioning uncertainties, SIA looked at the variability in the difference between modeled and actual consumption for a gas at the fab-level and then used a statistical argument to translate this variability into potential apportioning errors for each process (see SIA, 2011c for the total fab-level differences). 
There were limitations to the data on apportioning error submitted by SIA: actual gas consumption estimates were based on purchase records and did not appear to account for accumulation or drawdown of on-site gas inventories. Failing to account for changes in these inventories could lead to year-to-year fluctuations in the difference between actual and modeled consumption at the fab, and such fluctuations would have nothing to do with errors in assigning consumption to different process types (i.e., apportioning error).  The range in apportioning errors provided by SIA in 2012 also differed from information previously submitted, showing at one facility that a difference of less than 10 percent was achievable when measuring and modeling gas consumption at the tool level (SIA 2011b).  For these reasons the EPA assumed various apportioning uncertainties as noted above. The EPA also assumed an uncertainty of 5 percent on total gas consumption estimates for all runs of the analysis (based on the calibration accuracy requirements in 40 CFR 98.3(i) of subpart A, which the EPA is proposing to adopt for subpart I).
      2.4.1.  Analysis Results Summary
A comparison of the results for n=SIA3 and varying assumptions for f for Models 1 and 2 (taking into account chamber cleaning emissions as well as etch emissions) are presented in Table 2. The gas use patterns modeled were industry totals provided by SIA across seven 200 mm fabs and five 300 mm fabs for 2009 and 2010; however, the uncertainties calculated here are not intended to represent industry-level uncertainties, which would generally be expected to be lower than fab-level uncertainties.   Nominal emissions (i.e., emissions calculated without accounting for uncertainties) and simulated emissions (i.e., the mean of the Monte Carlo emissions profile when accounting for uncertainties) are indicated in metric tons of CO2 equivalent (mtCO2e). Also shown on Table 2 are the lower and upper uncertainty bounds of the 95 percent confidence interval, expressed as a percentage of the nominal value. 
Several trends can be identified by comparing the results in Table 2. First, it can be observed that Model 2 systematically provides lower emission estimates than Model 1, with a difference of between 2.2 percent and 5.2 percent in nominal emissions between the two models (3.6 percent on average). This effect appears to be due to the difference in the default EFs according to the two groupings, with the EFs for Model 2 generally being slightly lower than for Model 1. While differences between emissions estimated for 2009 and 2010 can be observed (2.5 to 3.8 percent increase in nominal emissions between 2009 and 2010 for 200 mm, depending on the model, and 24.9 to 26 percent for 300 mm), this is mainly attributed to differences in utilization efficiencies of the facilities rather than changes in gas consumption patterns.
Second, it is interesting to note that the average uncertainty of the emissions estimate for the 200 mm results is 15.4 percent while it is 32.1 percent for the 300 mm facilities. This effect is attributed to the larger uncertainty observed for 300 mm EFs (98.3 percent on average), compared to 200 mm factors (86.4 percent on average). This trend is further amplified by the large uncertainty associated with the NF3 remote clean EFs for both wafer sizes (223 percent for 200 mm and 233 percent for 300 mm), taking into account that the use of this cleaning technology is more prevalent for 300 mm facilities than for 200 mm facilities. 
Third, when comparing the uncertainties of the emissions estimates for Models 1 and 2, the difference averaged over all cases presented in Table 2 is only 1.4 percent, with Model 2 generally providing slightly higher precision than Model 1. However, this trend is not consistent and in some cases the precision of Model 1 is greater than for Model 2 (in particular for 200 mm profiles). For example, the 200 mm case with f=5 (first two rows of Table 2) shows an average uncertainty (i.e. the average of lower and upper bound uncertainty at 95 percent confidence) of 13.3 percent for Model 1 and 13.9 percent for Model 2. 
Fourth, it can be observed that the upper and lower bounds of the uncertainty interval (95 percent confidence) are not symmetrical. This is most pronounced for the 2010 300 mm profile with f=SIA (last row of Table 2) where the difference between the lower bound emissions represents a -40.9 percent deviation from the nominal value while the upper uncertainty bound represents a +80.8 percent deviation from the nominal value. This phenomenon is less pronounced for 200 mm facilities and for lower values of f. It can be attributed to the more prevalent use of the NF3 remote clean technology for 300 mm facilities and to the unique characteristics of the EF profile for this technology: because the utilization efficiency of NF3 with the remote clean technology is so close to 100 percent (98.2 percent) and because utilization efficiencies cannot exceed 100 percent (it is not possible to use more than 100 percent of the precursor gas), it is necessary to truncate the distribution profile for the NF3 remote clean EF, which becomes asymmetric (approximating half a Gaussian). This characteristic, combined with the fact that NF3 used for remote clean represents the largest gas consumption in 300 mm facilities, explains the asymmetry of the emissions profiles and the fact that nominal emissions estimates can be lower than simulated emissions estimates (in particular for 300 mm). While the difference between nominal and simulated emissions is generally less than 0.5 percent, this difference for f=SIA, 300 mm, 2010, Model 2 (last row of Table 2) is 7.85 percent.
Finally, it is interesting to note that the impact of f, which reflects the uncertainty associated with gas consumption apportioning, is relatively limited between f=5 and f=20, but that this impact is well pronounced for the values of f=SIA. For example, while the average uncertainty of the overall estimates varies between 11.4 percent and 14 percent for f=5 to 20 for 200 mm (between 19.6 percent and 25.7 percent for 300 mm), the average uncertainty is 23.8 percent for 200 mm facilities with f=SIA (60.5 percent for 300 mm). The impacts of the variables f and n are further evaluated below. 
Table 2. Comparison of Analysis Results for Models 1 and 2 for n=SIA3 and Various Values of f (mtCO2e)
                                  Wafer Size
                                       f
                                     Year
                                     Model
                      Nominal Emissions Estimate (TCO2e)
                     Simulated Emission Estimate  (TCO2e)
                  % Lower Uncertainty Bound         (95% CI)
                % Upper Uncertainty Bound             (95% CI)








                                    200 mm
                                       5
                                     2009
                                       1
                                    1,288,195 
                                    1,289,421 
                                    -13.1%
                                     13.5%



                                       2
                                    1,236,741 
                                    1,235,722 
                                    -13.8%
                                     13.9%


                                     2010
                                       1
                                    1,337,585 
                                    1,337,838 
                                    -11.4%
                                     11.7%



                                       2
                                    1,267,472 
                                    1,266,101 
                                    -11.6%
                                     11.7%

                                      10
                                     2009
                                       1
                                    1,288,195 
                                    1,288,883 
                                    -13.3%
                                     13.6%



                                       2
                                    1,236,741 
                                    1,237,317 
                                    -13.4%
                                     13.9%


                                     2010
                                       1
                                    1,337,585 
                                    1,336,303 
                                    -11.2%
                                     11.7%



                                       2
                                    1,267,472 
                                    1,264,090 
                                    -12.0%
                                     11.6%

                                      20
                                     2009
                                       1
                                    1,288,195 
                                    1,288,022 
                                    -13.2%
                                     13.6%



                                       2
                                    1,236,741 
                                    1,233,985 
                                    -14.0%
                                     14.0%


                                     2010
                                       1
                                    1,337,585 
                                    1,338,610 
                                    -11.3%
                                     11.8%



                                       2
                                    1,267,472 
                                    1,264,586 
                                    -12.0%
                                     11.7%

                                      SIA
                                     2009
                                       1
                                    1,288,195 
                                    1,252,389 
                                    -33.0%
                                     20.5%



                                       2
                                    1,236,741 
                                    1,201,227 
                                    -38.6%
                                     24.4%


                                     2010
                                       1
                                    1,337,585 
                                    1,318,213 
                                    -22.5%
                                     15.3%



                                       2
                                    1,267,472 
                                    1,267,928 
                                    -18.0%
                                     18.2%
                                    300 mm
                                       5
                                     2009
                                       1
                                       437,764 
                                      439,114 
                                    -24.3%
                                     25.8%



                                       2
                                       424,431 
                                      425,184 
                                    -18.9%
                                     20.6%


                                     2010
                                       1
                                       546,737 
                                      547,474 
                                    -22.6%
                                     24.8%



                                       2
                                       534,813 
                                      534,920 
                                    -19.8%
                                     21.7%

                                      10
                                     2009
                                       1
                                       437,764 
                                      438,904 
                                    -24.1%
                                     25.1%



                                       2
                                       424,431 
                                      425,101 
                                    -18.9%
                                     20.2%


                                     2010
                                       1
                                       546,737 
                                      548,566 
                                    -23.1%
                                     25.2%



                                       2
                                       534,813 
                                      536,969 
                                    -19.9%
                                     22.1%

                                      20
                                     2009
                                       1
                                       437,764 
                                      439,969 
                                    -24.5%
                                     26.9%



                                       2
                                       424,431 
                                      425,846 
                                    -19.6%
                                     21.6%


                                     2010
                                       1
                                       546,737 
                                      548,911 
                                    -23.3%
                                     25.5%



                                       2
                                       534,813 
                                      536,907 
                                    -20.2%
                                     22.9%

                                      SIA
                                     2009
                                       1
                                       437,764 
                                      444,496 
                                    -47.7%
                                     72.6%



                                       2
                                       424,431 
                                      455,230 
                                    -38.9%
                                     75.9%


                                     2010
                                       1
                                       546,737 
                                      558,492 
                                    -49.1%
                                     78.1%



                                       2
                                       534,813 
                                      576,813 
                                    -40.9%
                                     80.8%
                                       
While Table 2 presents emission estimates at the "industry level" for an aggregate of seven 200 mm fabs and five 300 mm fabs , the EPA also analyzed emissions from the four individual facilities for which SIA provided gas consumption profiles (two 200 mm, Fabs B and C and two 300 mm, Fabs A and E). The results of this analysis are provided in Figure 3, where the nominal as well as the simulated mean emissions are presented, along with the lower and upper bound of the 95 percent confidence interval (as predicted by the Monte Carlo simulation -- P2.5 and P97.5) for each fab and each year (2009 and 2010). For this analysis, values of f=20 and n=SIA3 were retained. The results indicated that the average uncertainty (the average of the lower and upper bound uncertainty) varied between 10.7 percent and 40.2 percent, with the uncertainty of the 300 mm fabs being higher than for the 200 mm fabs (31.3 percent versus 14.3 percent over 2009 and 2010), confirming the trend observed in Table 2. Also, the emissions for year 2010 were generally higher than for year 2009, with average uncertainty values varying from 0.4 to 5 percent year over year, these trends being attributed to changes in facility production levels rather than to changes in gas consumption patterns.
                                       
Figure 3. Nominal, Simulated, Lower, and Upper Bound Emissions (P2.5 and P97.5) at the 95 Percent Confidence Interval for Fabs B and C (200 mm) and A and E (300 mm) in 2009 and 2010
As noted above while observing the results in Table 2, the impact of variable f, which represents the uncertainty related to gas consumption apportioning, seems limited at relatively small values of f (f<20  percent). To confirm this trend, Monte Carlo simulations where run on the `industry-level' 300 mm consumption profile (Model 1, year 2009) with increasing values of f=5, 10, 20, 50, 100, 150, 200 percent, and compared to the range of apportioning uncertainties provided by SIA for various gases (f=SIA). The results of the simulations are shown in Figure 4, which depicts the emissions estimates for the nominal value of the estimate, the simulated emissions, and the upper and lower bounds of the emissions estimates at the 95 percent confidence level (P2.5 and P97.5). The value of n chosen for this analysis was n=SIA3.  As can be seen, the results confirm the trend observed in Table 2, where the average uncertainty is relatively stable for f=5 to f=20 (24.5 percent to 25.6 percent), but where the uncertainty starts to go up at higher values of f (30.4 percent for f=50 to 69.9 percent for f=200). Interestingly, the upper bound of the uncertainty interval increases faster with higher values of f than the lower bound decreases, probably due to an exacerbation of the asymmetry of the profile of the NF3 remote clean gas consumption, as discussed above. Note also that the difference between the simulated and the nominal value of the estimate increases slightly with higher f values, due to the same phenomenon. When comparing the uncertainty range for f=SIA - which does not provide a fixed value for f but a range of values for different gases - Figure 4 suggests that the range of apportioning uncertainties provided by SIA corresponds to a fixed value of f of approximately 145. 
These results support the EPA's proposal (see Section 4) to set the maximum difference between the modeled and actual gas consumption to 20percent as a requirement for verifying the model used to apportion gas consumption, as this quality assurance requirement seemed to achieve a suitable balance between technical feasibility and burden, without compromising the uncertainty of the overall emissions estimates.

Figure 4. Impact of Apportioning Uncertainty (f) on Emissions Estimates for a Typical 200 mm Fab
In addition to studying the impact of the accuracy of apportioning gas consumption, the EPA investigated the influence of the variable n (as discussed in Section 2.2.1). Figure 5 presents the results of varying values of n (n=1, SIA1, SIA3, NDB), showing the evolution of the nominal value of the emissions estimate, of the simulated emissions, and of the upper and lower bounds of the emissions estimates at the 95 percent confidence level (P2.5 and P97.5) for the `industry-level' 300 mm consumption profile provided by SIA (Model 1, year 2009). The value of f chosen for this analysis was f=20. As can be seen by comparing Figures 4 and 5, the impact of n is much greater than the impact of apportioning uncertainty (f). When varying n from n=1 to n=NDB, the average uncertainty decreases from 39.4 percent to 10.5 percent.

Figure 5. Impact of Representativeness of EFs (n) on Emissions Estimates for the 300 mm `Industry Level' Gas Consumption Profile (2009, f=20)
Trends for the 3-factor model and 200 mm facilities were found to be similar to those for the two-factor model: increasing f values tend to increase the overall uncertainty (but the difference is also minimal for values of f of 20 or lower), and increasing the value of n decreases the overall uncertainty interval. In looking at the overall results of this analysis, it appears that, in light of its added complexity, Model 2 does not provide substantial improvements in the uncertainty of emissions estimates compared to Model 1. Thus, as summarized in the next section, the EPA opted to propose Model 1 for all semiconductor facilities to estimate and report emissions under subpart I.
   4.1. Development of Proposed Revised Default Etch Emission Factors
To develop the proposed revised default etch EFs, the EPA used the methods and conventions described and selected in 2.1.  This included:
   * Using a simple arithmetic averaging method to develop default utilization and by-product EFs by process type/sub-type and gas; and 
   * Using the dominant gas convention for assigning by-product emissions for chamber clean by-product EFs and the all inputs gas convention for assigning by-product emissions for etch EFs. 
As described and shown in Section 2.2, the uncertainty improvements that Model 2 could provide are minimal in light of the added complexity of it compared to Model 1, and are not always consistent. Therefore the EPA is proposing revised default etch EFs for semiconductor manufacturing facilities that correspond with the first model (Model 1) described above: default EFs by gas and wafer size for one broad etching process type and three CVD chamber cleaning process sub-types based on the CVD chamber clean process being used (in-situ thermal chamber clean, in-situ plasma chamber clean, and remote plasma chamber clean). 
Despite the larger sample of EF data available, there are still data gaps for certain wafer sizes and process types. First, little to no data exist to support establishing separate default EFs for facilities manufacturing 150 mm wafers, yet it is estimated that more than 25 facilities using this wafer size are still in operation (WFF, 2011). Consistent with the December 2010 version of subpart I, the EPA is proposing that the EFs for 200 mm technologies would apply for 150 mm as well. The EPA based the 2010 EFs for 150 mm wafers on a conclusion that EFs for 150 mm processing technology are expected to be closer to 200 mm processing technologies than to 300 mm processing technologies (November 2010 TSD). Second, the EPA did not receive any new chamber clean emissions data and, therefore, the EPA is proposing to maintain the chamber clean EFs finalized in the December 2010 rule. 
As part of the proposed revisions to the F-GHG EFs the EPA is also proposing to combine the wafer clean process type with the etch process type because both processes are very similar in principle. Both wafer cleaning and etch processes involve using plasma generated fluorine atoms or other reactive fluorine-containing fragments to remove material from wafer surfaces. While etch processes are used to selectively remove portions of material from exposed thin films (e.g., dielectric, metals), wafer cleaning relies on similar reactions to remove similar residual materials, including at the wafer edge. Additionally, wafer cleaning operations are estimated to represent about one percent of total fab F-GHG consumption, and thus are expected to represent a similarly small percentage of total fab emissions (SIA, 2012a). Lastly, the burden associated with apportioning gas consumption to the various process types is expected to be reduced by combining the wafer clean and etch process types because some gases used for wafer cleaning (e.g., CH2F2 and SF6) are also used in etching processes. 
3. Estimating F-GHG Emissions from Semiconductor Fabs Manufacturing Wafers Larger than 300 mm 
The next generation of semiconductor fabs will produce wafers 450 mm in diameter, but commercial manufacturing is unlikely to occur before 2017 (SIA, 2012c). Therefore, the EPA currently has no EF data upon which to establish default EFs for the 450 mm wafer size. To address this in the December 2010 version of subpart I, the EPA finalized provisions that required facilities to use a recipe-specific approach to estimate F-GHG emissions from all semiconductor processes used at a fab.
As described in Section 2, because of the concerns raised by SIA in regards to using recipe-specific EFs, the EPA is proposing to remove the provisions that require a recipe-specific approach for facilities manufacturing wafers larger than 300 mm. The EPA is proposing that these fabs use the same default EFs that the EPA is proposing for semiconductor fabs manufacturing wafers that are 300 mm or smaller and are proposed to no longer have the option to use recipe-specific EFs. 
The EPA, based on information provided by SIA, understands that process chemistries used for the 450 mm wafer technology will not change substantially from those used for 300 mm wafers (SIA, 2012c). While the EFs for 450 mm wafers are expected to be different from those for 300 mm wafers, no data are currently available to quantify this difference. However, this difference is expected to be, for example, less than the difference between those for 200 mm and 300 mm wafers. Therefore, at this time, the EPA is proposing to require any fabs manufacturing 450 mm wafers to use the default EFs developed for fabs manufacturing 300 mm wafers because of the similarity in process chemistries between 300 mm and 450 mm wafers. 
In the future, the EPA may develop separate EFs for fabs manufacturing 450 mm wafers. In order to provide for consistent review of technology changes in the semiconductor manufacturing industry and ensure that default gas utilization rates and by-product formation rates and DRE values accurately reflect the industry's practices in future years, especially with respect to 450 mm wafers, the EPA is also proposing to require certain semiconductor manufacturing facilities to provide a report to the EPA every 3 years, beginning in 2017, that addresses technology changes at the facility that could affect GHG emissions. The report would address how technology in the industry has changed over the previous 3 years and the extent to which any of the identified changes are likely to have affected the emissions characteristics of semiconductor manufacturing processes in such a way that the default EFs and/or DRE values in subpart I may need to be updated or augmented. For more information on this proposed provision, please see the preamble for the proposed rule amendments.
4. F-GHG and N2O Apportioning Model Development and Verification
When using an emission estimation method based on EFs in units of mass of gas emitted per mass of gas used for various process types and sub-types, fab-wide gas consumption must be apportioned to the appropriate process types and p sub-types through the use of fab-specific apportioning factors. 
The EPA is proposing to modify various aspects of the provisions related to developing apportioning factors. The provisions, as well the proposed changes and rationale for modifying the provisions are provided in Table 3 below. 
Table 3. Summary of Proposed Revisions for F-GHG and N2O Apportioning Model Development and Verification
                           December 2010 Provisions
                          Proposed Revised Provisions
                 Rationale for Proposed Modification/Addition
Apportioning Model Applicability
The same apportioning model is used for the whole facility. 
Apportioning model and verification would be on a fab-basis instead of a facility basis.
A facility with more than one fab could use different methods to estimate F-GHG emissions in each fab, use different sized wafers, or manufacture different products (e.g., MEMS and semiconductors).
Actual Gas Consumption Monitoring Period
Analyze at least a 30-day period of operation during which the capacity utilization equals or exceeds 60 percent of the facility design capacity. In the event a facility operates below 60 percent of its design capacity for the whole year, the facility must use the period of the highest 30-day average utilization for model verification.
The EPA is proposing to replace the "capacity utilization" requirement with "representative operating levels," defined to mean operating the fab, in terms of substrate starts for the period of testing or monitoring, at no less than 50 percent of installed production capacity or no less than 70 percent of the average production rate for the reporting year, where production rate for the reporting year is represented in average monthly substrate starts. 

The EPA is also proposing to modify the actual gas consumption monitoring period to clarify that it may extend up to the full year of operation.
In the December 2010 subpart I, the EPA required facilities to monitor actual F-GHG consumption for a period of at least 30 days in order to verify an apportioning model. The EPA required a minimum 30-day period to minimize burden to reporters. However, in light on new information provided by SIA, the EPA is clarifying that facilities may monitor actual F-GHG consumption for a period of more than 30 days and up to a full reporting year. 

The EPA is providing this clarification and incorporating the "representative operating level definition" because a facility may not have the information needed to confirm whether a 30-day period meets the minimum capacity utilization criteria until the end of the reporting year. Therefore, some facilities may not be able to select a 30-day period in which to monitor actual F-GHG consumption. (SIA, 2012f) 
Consumptions to be Compared for Model Verification
The modeled and actual F-GHG consumption for the F-GHGs used in the largest quantity, on a mass basis, for one F-GHG used for the etch process type and one F-GHG used for the chamber cleaning process type. The comparison made for the F-GHGs used in the largest quantity for the etch process type is the comparison that must meet the verification standard (see next row).
The modeled F-GHG consumption for the F-GHG used in the largest quantity at the fab, on a mass basis, for one F-GHG must be compared for verification purposes.  

Fabs may elect to compare the modeled and actual F-GHG consumed for two F-GHGs for verification purposes and demonstrate that the verification standard is met on an aggregate use basis for both F-GHGs if one of the F-GHGs selected for comparison corresponds to the largest quantity, on a mass basis, of F-GHG used at the fab during the reporting year.
The EPA is proposing that the comparison must be based on total F-GHG consumption (as opposed to F-GHG consumption for the etch process type) because most fabs are not configured to measure F-GHGs consumption between the etch and clean process types (SIA, 2012f), or do not have the infrastructure in place to collect this information centrally. The proposed modification would reduce burden associated with apportioning model verification.

The EPA is also proposing to give fabs the flexibility to verify the model for two F-GHGs on an aggregate use basis because, in some cases, it may be difficult to predict the F-GHG that will be used in the largest quantity for the reporting year if the expected quantities of consumption are similar.
Verification Standard
To maintain flexibility for facilities in developing apportioning models the EPA allows the development of a model based on any quantifiable metric of activity that best suits the needs or capability of the particular facility (see November 2010 TSD for more discussion). However to ensure quality of data, EPA requires facilities to meet an apportioning model verification standard of 5 percent, as measured by the difference between the actual and modeled F-GHG consumption for the F-GHG used in the largest quantity for the etch process type. 
Fabs must demonstrate that the comparison performed for the largest quantity of F-GHG(s), on a mass basis, used in the fab does not result in a difference between the actual and modeled F-GHG consumption that exceeds 20 percent relative to actual F-GHG consumption.
The EPA is proposing to relax the verification requirement because information submitted by SIA for a sample test of model verification at a fab demonstrates the difficulty in meeting the 5 percent standard. (SIA, 2011b)


Revising the verification standard from 5 percent to 20 percent is not expected to substantially impact the accuracy of emission estimates because the main source of error in emission estimates is from the EFs used, not the apportioning error (see discussion in Section 2.2.2 of this document on uncertainty evaluation). 

In looking at Figure 4 it can be seen that as long as the apportioning uncertainty is 20 percent or less, the overall uncertainty in emission estimates varies little with the apportioning uncertainty.  However, in moving to higher apportioning uncertainties (e.g., 50 percent) a trend in increasing overall uncertainty in emission estimates can be seen.
Alternative to Developing an Apportioning Model
Not addressed.

(Facilities were not given the option of measuring F-GHG consumption directly instead of using apportioning factors developed through modeling.)
Fabs have the option to develop apportioning factors through the use of direct measurement using F-GHG flow meters and weigh scales to measure process sub-type, process type input F-GHG consumption.
Some fabs may be configured such that F-GHG consumption can be measured at the process type/sub-type or at  tool level  using direct F-GHG measurement (e.g., flow meters and weigh scales) (SIA, 2012f). The EPA is proposing to allow direct measurement of F-GHG consumption as an alternative to developing an apportioning model. This provides fabs flexibility while potentially improving accuracy in emission estimates by using direct measurements in place of estimates. 
   
5. Accounting for Abatement System Uptime When Using a Default EF Method
In December 2010 the EPA finalized a method for estimating uptime of abatement systems based on the ratio of the total time an abatement system is in an operational mode and the time when F-GHGs or N2O are flowing through the connected process tool(s). While this method is the most accurate to measure abatement uptime and its impact on emissions, SIA raised the issue that it is not consistent with how industry currently tracks abatement system uptime. Complying with the December 2010 provisions would increase burden on facilities because of required installation of hardware or software to track and correlate the abatement systems' operational time with the time in which F-GHGs or N2O are flowing in through tools. (For a more detailed discussion see SIA, 2012d).
SIA suggested a simplified approach for calculating uptime and provided additional data and information on the uptime of actual abatement systems. The simplified approach calculates uptime for all abatement systems for a given gas and process type or sub-type combination, instead of by individual abatement systems. SIA also suggested that the amount of time a gas is flowing through the tool connected to an abatement system should be assumed to be equal to the maximum amount of time a tool could be operating each year (i.e., 525,600 minutes per year) (see SIA, 2012d).
The EPA evaluated the abatement uptime information submitted by SIA on actual fab abatement systems. This information included a comparison of SIA's suggested approach for calculating uptime and an estimation of how uptime would be calculated using the December 2010 subpart I method. SIA also quantified the impact of estimating uptime for all abatement systems associated with a gas/process type or sub-type combination as opposed to estimating uptime by individual abatement systems. These data and results are detailed in SIA, 2012d and SIA, 2012e. In summary, the information provided by SIA on its suggested method along with further analysis by the EPA showed that:
   1. The method suggested by SIA results in more conservative estimates of uptime for most common scenarios (when abatement system uptime is greater than 40 percent) because it assumes F-GHGs are flowing through the tool constantly. (SIA, 2012d)
   2. Uptimes estimated using SIA's suggested method are generally high.  On average they were estimated to be between 75 percent and 90 percent based on data submitted from SIA for a sample of 3 companies. (See data in SIA, 2012d)
   3. SIA's suggested method can rely on current industry practices, thereby reducing the need for fab system modifications.
   4. Calculating and applying uptime on an aggregate basis for all abatement systems associated with a gas/process type or sub-type combination, as opposed to uptime for each separate abatement system, would remove the need to apportion gas consumption and estimate GHG emissions at an individual tool/abatement system level, reducing associated burden. (SIA, 2012e)
Based on the above listed reasons, the EPA concluded that burden would be minimized by monitoring and calculating uptime by gas/process type or sub-type, instead of by individual system. The EPA is proposing to revise the abatement system uptime calculation used with the default EF method, based on the suggestion by SIA for estimating abatement system uptime that reflects current semiconductor industry practices. This approach is based on calculating the ratio of the total time that all abatement systems abating a given process type or sub-type/gas combination are not operational to the total time when the abated tools are in operation. The total time that the abated tools are in operation would be pro-rated if some tools are only operational for a portion of the year (e.g., if they are removed from or installed in the fab during the year). The proposed revised method to calculate uptime is represented in the equation below, and is applicable for all facilities using a default EF-based emission estimation method (regardless of the type of product manufactured).
UTij=1-pTdijppUTijp
where:
UTij	= The average uptime factor of all abatement systems connected to process tools in the fab using input gas i in process sub-type or process type j (expressed as a decimal fraction).
Tdijp	= The total time, in minutes, that abatement system p, connected to process tool(s) in the fab using input gas i in process sub-type or process type j, is not in operational mode (as defined in 40 CFR 98.98 of subpart I), when at least one  of the tools connected to abatement system p is in operation.
UTijp	= Total time, in minutes per year, in which abatement system p has at least one associated tool in operation. For determining the amount of tool operating time, a facility could assume that tools that were installed for the whole of the year were operated for 525,600 minutes per year. For tools that were installed or uninstalled during the year, a facility would prorate the operating time to account for the days in which the tool was not installed, and would treat any partial day that a tool was installed as a full day (1,440 minutes) of tool operation. For an abatement system that has more than one connected tool, the tool operating time would be 525,600 minutes per year if at least one tool was installed at all times throughout the year. If a facility had tools that were idle with no gas flow through the tool, the facility could calculate total tool time using the actual time that gas is flowing through the tool.
i	= Input gas.
j	= Process sub-type or process type.
p	= Abatement system.
Because the EPA is now proposing that uptime be tracked and associated with a particular gas, it is necessary to also propose an equation for estimating the uptime for emissions of by-products that are not also used as input gases. While the calculation method for the uptime associated with by-product emissions is the same as the method presented in the above equation, different inputs are needed that correlate to uptime for by-products as opposed to utilization emissions. For example, for estimating uptime for a particular by-product the variable UTjk (The average uptime factor of all abatement systems connected to process tools in the fab which emit by-product gas k, in process sub-type or process type j) would replace UTij in the above equation, and so forth. 
6. References
IPCC (2006) 2006 IPCC Guidelines for National Greenhouse Gas Inventories. The National Greenhouse Gas Inventories Programme, The Intergovernmental Panel on Climate Change, H.S. Eggleston, L. Buendia, K. Miwa, T Ngara, and K. Tanabe (eds.). Hayama, Kanagawa, Japan.  Available at: http://www.ipcc-nggip.iges.or.jp/public/2006gl/pdf/3_Volume3/V3_6_Ch6_Electronics_Industry.pdf. 
WFF(2011). Semiconductor Equipment and Materials Industry (2011) World Fab Forecast, May 2011 Edition. (For purchase)
SIA (2011a). SIA Alternatives to Recipe-Specific Testing, August 11, 2011. Presentation from SIA/SIMI to EPA.  Available in EPA docket EPA-HQ-OAR-2011-0028.. 
SIA (2011b). Semiconductor Engineering Models for Apportionment of Fluorinated Greenhouse Gases to Process Categories (Verification Tests to Demonstrate Difficulty of achieving 5 percent limit). June 16, 2011. Available in EPA docket EPA-HQ-OAR-2011-0028.
SIA (2011c). SIA/ISMI Presentation of Etch Alternative, November 17, 2011 and Additions & Corrections: ISMI/SIA Alternative to Recipe- Specific Testing of 17 Nov 2011. Available in EPA docket EPA-HQ-OAR-2011-0028.
SIA (2012a). Report to EPA on Etch Factor Proposal for Fab GHG Emissions Reporting. February 28, 2012. Available in EPA docket EPA-HQ-OAR-2011-0028.
SIA (2012b). International SEMATECH Manufacturing Initiative Environmental Safety and Health Technology Center Etch Process Equipment Emissions Characterization Data. February 6, 2012. Available in EPA docket EPA-HQ-OAR-2011-0028.
SIA (2012c). SIA's Response to EPA Etch Alternative Proposal, April 9, 2012. Available in EPA docket EPA-HQ-OAR-2011-0028.
SIA (2012d). SIA Briefing Paper on Abatement Issues: Destruction Removal Efficiency (DRE). January 10, 2012. Available in EPA docket EPA-HQ-OAR-2011-0028.
SIA (2012e). Questions for SIA on Abatement System DRE, Uptime, and Related Requirements. March 5, 2012. Available in EPA docket EPA-HQ-OAR-2011-0028.
SIA (2012f). SIA Revised Proposal to Amend the Apportionment Model Validation Criteria in 40 CFR 98.94(c).March 15, 2012. Available in EPA docket EPA-HQ-OAR-2011-0028.
