Overall Air Emissions Control Strategy Tim Higgs Intel Corp. 145 S. 79 th St. MS: CH10-22 Chandler, AZ (480)

Transcription

1 Overall Air Emissions Control Strategy Tim Higgs Intel Corp. 145 S. 79 th St. MS: CH10-22 Chandler, AZ (480) Introduction Air program requirements for semiconductor manufacturing facilities have increased significantly in recent years. In the past, facilities only had to comply with traditional state and local permit programs whose requirements were relatively few compared to today. New federal programs such as the Title V operating permits rule, compliance assurance monitoring (CAM), and the various rules implementing the hazardous air pollutants (HAPs) program have created new requirements and burdens both for those facilities directly covered by the rules and those interested in avoiding them. An additional future challenge for some facilities will be the semiconductor MACT rule for control of hazardous air pollutants, which is now under development. Since many of the more burdensome requirements of these rules apply only to those sources defined as major by the Clean Air Act, many semiconductor facilities have chosen to remain minor sources by maintaining emissions below the appropriate thresholds. This strategy can help minimize burdens and potential restrictions on flexibility that these programs impose on major sources. Even minor sources, however, still must maintain rigorous air programs which can firmly establish their minor source status. Therefore, many of the basic elements needed in a sound air program are the same for major and minor sources. These include thorough understanding and documentation of emissions, selecting and maintaining the proper control equipment, and developing permits which balance enforceability with flexibility. As fab manufacturing sites increase in size and new technologies are introduced at an ever increasing rate, the importance of a well managed air program will continue to increase in importance.

2 Emissions Characterization The first basic element of a complete air program is a thorough understanding of plant wide emissions. Many of the requirements that apply to a facility are directly tied to the chemicals and quantities emitted, and most air permits require reporting of all regulated emissions on at least an annual basis. In addition, if a facility intends to establish itself as a minor source of emissions, the facility must be able to adequately demonstrate that emissions are indeed below the major source thresholds. For these reasons, a complete and well documented emissions profile of the facility is a critical first step that must be taken to establish and comply with the overall air program. Semiconductor fabs tend to be relatively complex emissions sources. A fab may have dozens or even hundreds of individual manufacturing tools which can be emission sources. Individually most of these are quite small, but in total the emissions can be significant. The primary pollutants of concern originating from the fab are typically volatile organic compounds (VOCs) and hazardous air pollutants (HAPs). There are also emission sources outside the fab which must be considered. These include fuel burning equipment such as boilers, generators, and VOC abatement devices as well as miscellaneous other sources like chemical storage tanks and wastewater treatment operations. Fuel burning equipment is a source of the typical combustion byproduct pollutants (NOx, CO, SOx, PM), and is also a small but not insignificant source of additional VOCs. The other miscellaneous operations can be sources of VOC and HAP emissions. All sources must be considered in developing the overall emissions profile. There are many different techniques available for determining the emissions, all of which have their strengths as well as limitations. Common techniques include the use of emission factors, direct stack testing of individual tools, standard end of pipe stack tests, and chemical mass balance. Understanding the strengths and weaknesses of each approach will help the plant engineer decide which is best for a particular application. Emissions from fuel burning sources such as boilers, emergency generators, and VOC abatement devices are arguably the most straightforward of the emissions calculations for

3 a fab. Emission factors for each of the combustion byproducts are readily available. These typically come directly from the equipment manufacturer, or from EPA s AP-42 Compilation of Air Pollutant Emission Factors document. i These factors are then usually multiplied by the fuel consumption of the unit(s) to determine total emissions. In some cases, emission factors may be expressed as a function of hours of operation, as opposed to fuel use. This is common with emergency generators. In those cases, the emission factor is multiplied by hours of operation to determine total emissions. These methods for determining combustion emissions are well established and have been used successfully for some time. Nonetheless, there are certain limitations the engineer must be aware of when making use of the methods. First, emission factors tend to be average values developed from many different units over long periods of time. The actual emissions from a given unit or set of units at a given time can be significantly different from the emission factor. This is especially true of AP-42 emission factors, which are a compilation of many different types of units operating under many types of conditions. Data supplied directly from a manufacturer of a unit is typically more specific to the unit in question and is generally considered to more accurately reflect emissions from that particular unit. Even in that case, however, the reported emission factors are either averages over a broad range of conditions, or are specific to one particular operating condition. For example, the NOx or CO emission rate specified by a boiler manufacturer is usually the emission rate when the unit is at full fire. In a low fire or idle mode, the emissions can be very different. Likewise, emergency generators will have very different fuel consumption (and therefore emissions) rates at full load vs. no load conditions. These variations must be comprehended when performing emissions calculations, and in determining the appropriate permitted plant wide emission limit for the facility. For facilities with relatively low emissions of these pollutants, it is advisable to simply set permitted emission levels well above expected actual emissions. For facilities which are struggling to maintain emissions below a key threshold (such as a major source level), this may not be an option. In this case, it may be wise to perform actual stack testing of the units to ensure that emissions are accurately characterized.

4 For fab operations, determining emissions of VOCs and HAPs is the primary challenge. A number of methods have been used for these calculations. Chemical mass balance is possibly the most widely used method for determining VOC emissions. This method involves determining the amount of VOC containing material that enters the fab, and the amount that exits as hazardous waste. The difference between the two is assumed to be emitted to exhaust. If the VOC exhaust is sent to a control device, the emissions are then adjusted to take into account the removal efficiency of the device, based on prior testing or other information about control device performance. The mass balance method has many pluses. For starters, it is relatively simple - chemical use and waste generation data are routinely collected by most fabs, so additional data collection is not required. The method is also very inclusive all emissions, whether stack or fugitive, will be counted by an overall chemical mass balance. Mass balance also has limitations which must be understood. For starters, some VOCs can experience significant dragout to wastewater in a fab. This is especially true of materials like PRS and ethylene glycol which are used in wet stations. Since the wastewater losses are difficult to quantify, a mass balance will usually assume that this lost material was emitted to air, which can result in a significant overestimation of emissions. Studies performed at Intel have indicated that far more of these materials go to wastewater than to air emissions. Other limitations with mass balance have to do with inaccuracies in tracking chemicals in and out. Chemical use records often show what was purchased in a given time period, as opposed to what was actually used. Waste records usually show when waste was shipped, as opposed to when it was generated. This can result in differences in inventory, on both the chemical and the waste end, from the start to the end of a mass balance period. At large factories with large chemical throughput volumes, these errors can become significant especially if mass balance is done over periods much shorter than one year. Therefore, large factories which are trying to maintain emissions below relatively low thresholds may need to consider methods that are more precise than mass balance for determining emissions.

5 Mass balance is generally not a viable method for determining emissions of most HAPs. There are several reasons for this. First, the largest quantity of HAP use in most fabs is in dilute liquid solutions of hydrofluoric and hydrochloric acids. Waste from these operations is sent to the facility wastewater systems along with many other fab liquids. Therefore, quantifying the amount of HF or HCl sent to the wastewater system is difficult if not impossible. Even if this could be quantified, the emissions from these dilute solutions are so small compared to the total use, that they would be well within the margin of error of most mass balance calculations. Therefore, a standard evaporation calculation can be useful for estimating emissions from wet benches. One such calculation is found in EPA document 560/ Estimating Releases and Waste Treatment Efficiencies for the Toxic Chemical Release and Inventory Form. ii Even for a large fab, emissions from these sources are expected to be less than 1000 lbs./year prior to the scrubbers. The primary source of HAPs emissions is usually from plasma tools such as chemical vapor deposition chambers and etchers. In these cases, the HAPs are usually the result of byproduct formation from the use of chlorinated and fluorinated compounds, such as when HF is formed from the partial destruction of SF6. Since there are multiple possible byproducts that can be formed from the incoming compound, and since there is also often significant solids deposition from these processes, mass balance is not likely to be a useful emission calculation tool. For these operations, emission factors can be developed which express the pollutant generation as a function of the incoming chemical use. Some companies have developed their own emission factors for those tools, based on the specific processes operated by that company. This may not be a practical option for others. In many cases, emissions information is now being developed by the fab tool manufacturers. This can be used as a source of emission factor data. A third potential source of HAPs is from liquid organic materials, such as photoresists or strippers which contain HAP components. In these cases, mass balance is typically an option only if the HAP component makes up a significant portion of the overall organic waste stream. Organic HAPs which are mixed with other liquid organic wastes from the

6 fab may be such a small portion of the total waste stream that accurate quantification may be difficult, and the mass balance may be inaccurate. In such a situation, emission factors can be used to determine the individual compound emissions. Potential sources of those emission factors include actual testing of the tool, obtaining data from manufacturers, or modifying an individual tool waste collection system in such a way that a mass balance can be performed around it with the results being applied to similar tools. Traditional end of pipe stack testing is generally considered to be the least favored option for determining overall annual emissions, due to a number of limitations. Standard EPA methods do not exist for many of the HAP compounds, and where they do exist the HAP emissions may be too dilute to be detected at the end of the stack. The HAPs that are likely to exist in the highest concentration at most fabs are HF and HCl. For these compounds, the standard EPA methods are not chemical specific and simply measure the total fluorides or chlorides present in the stream. Interferences from other chlorine and fluorine containing compounds can cause overestimation of HAP emissions. Accurately quantifying mass VOC emissions with stack tests requires the use of two methods simultaneously a total hydrocarbon method such as EPA method 25, and a speciated method such as method 18. Data from a total hydrocarbon method cannot be converted to a mass emission rate unless detailed information on specific chemical constituents is known. On the other hand, speciated methods only collect data for the particular species analyzed, leaving doubt about whether other species were present and not captured by the method. Finally, both HAP and VOC emissions from fabs tend to be variable due to the complex nature of the operations. Therefore, it is difficult to say whether a stack test taken over a finite period of time truly represents an average level of emissions which can be extrapolated out over a full year. These techniques can help provide a thorough overall understanding of facility emissions. This data can be used to help understand the specific regulatory requirements that apply to a facility and can also help the factory target potential areas for emissions reductions. A detailed emissions breakdown can also prove invaluable in understanding and optimizing performance of control devices, and in selecting new devices.

7 Control Equipment Control equipment plays a crucial role both in understanding facility wide emissions and in maintaining compliance. In order to understand and minimize total emissions, device performance must be both optimized and accurately comprehended. Semiconductor fabs essentially rely on two types of air emission control devices wet scrubbers and VOC abatement devices. Wet scrubbers are used for control of inorganics, such as acids and chlorine. Some of these inorganic materials are listed HAPs. A recent set of control device surveys was performed by the Semiconductor Industry Association (SIA) as part of the project to develop a Maximum Achievable Control Technology (MACT) standard for the industry. The surveys did not identify any type of control other than wet scrubbing for inorganic emissions. The surveys included more than 170 manufacturing sites in the U.S. The same surveys, however, found a number of different approaches used for controlling VOCs. The VOC abatement approaches basically all relied on oxidation, adsorption or some combination of both. The SIA survey only collected detailed control device data from a few of the larger facilities, but a subsequent benchmarking effort performed by Intel gathered additional data on these systems. That study identified 125 VOC abatement devices in use in the industry in the U.S. It is not known how many total devices are in use in the industry, but given the fact that the total number of facilities identified in the SIA survey was around 170, this is believed to be a substantial portion of the total units. Of the 125 systems, 53% utilized a rotary concentrator wheel followed by thermal oxidation. Another 34% used direct thermal oxidation, without an initial concentration step. 8% of the units used fixed bed concentrators followed by oxidation, while 3% used fluidized bed adsorption systems followed by condensation of the organics. The final 2% used point of use catalytic oxidation systems. Other than the catalytic oxidation systems, all of the other devices were centralized, end of pipe systems rather than point of use.

8 Regardless of the exact type of system being used, understanding the control efficiency being achieved can be a challenge. Most wet scrubber systems are rated by the manufacturer as being 95%+ efficient at removing a key target compound (e.g. HF or HCl). However, this rating usually is applied to inlet concentrations of 5 10 ppm, which is far higher than the average semiconductor fab inlet. VOC exhaust streams generally have much higher concentrations than scrubbed exhaust streams, but still often operate below the realm where optimum efficiency is achieved. Understanding actual efficiency (and therefore actual emissions) requires information about performance at low inlet concentrations, as well as an understanding of the limitations of stack test methods. The MACT surveys mentioned above attempted to develop information on the performance of control devices under actual fab operating conditions. Wet scrubber efficiency from the reporting facilities ranged from 21 97%. The facility reporting 21% efficiency had a very low inlet concentration, which likely contributed to the low reading. All other facilities reported overall efficiencies of at least 72%. Reported VOC abatement efficiencies ranged from 26 99%. Again, the facility reporting 26% was experiencing a very low inlet concentration. All other facilities reported efficiencies of at least 60%. It is important to note that the efficiencies reported were overall averages based on all the data points collected by that facility. Individual readings at each facility varied considerably, depending on the inlet concentration at the time of the test, among other things. A detailed discussion of these results is provided in a paper included in the SSA `99 proceedings. iii To be consistent with the MACT definition, SIA then tried to define the average performance achieved by the top five performing sources. Since efficiency is clearly dependent on inlet concentration, the committee was not comfortable with defining performance simply by an efficiency number and assuming that number could be met under all conditions. Instead, the committee came up with a performance metric that expressed performance as either a specified efficiency or a maximum outlet concentration. Standard statistical techniques were then applied to determine the likelihood of a given device meeting that performance, based on the MACT survey data.

9 The conclusion of this effort was that a properly performing wet scrubber should have a 90% probability of achieving either a 93% removal efficiency, or an outlet concentration of no more than 0.75 ppm. Clearly, this conclusion says that a device experiencing a very low inlet concentration (e.g. 1 ppm or less) will not be expected to routinely achieve a high removal efficiency. Similarly, for VOC abatement devices the study concluded that a properly performing unit should achieve a 93% removal efficiency or an outlet concentration of 6.8 ppm. There are two known factors which limit the ability to achieve high removal efficiency at low inlets. One is simple device performance. At some low level, treatment devices will begin to approach a limit below which they can no longer effectively remove pollutants. The precise level will vary by the type of unit and the conditions it is experiencing, but the results of the MACT survey provide a good analysis of where the average device will approach this level. The other factor involves stack test method limitations. Even if device performance is actually very good at low concentration, as the detection limits are approached, the ability to accurately measure efficiency decreases. A concentration reading of 2 times the detection limit has an error margin of + 50%. An efficiency calculation involves two stack readings (inlet and outlet) which means this error will be compounded. For this reason, it is recommended that efficiency calculations only be used if the inlet concentration for the particular reading is at least 5 times the minimum detection limit for the test. Additional understanding of the total overall performance can be improved if information is available concerning the amount of time the devices experience the various inlet conditions. For example, it may be known that scrubber efficiency is 90% when the concentration is 1 ppm or higher, 80% when the concentration is between 0.5 and 1 ppm, and unmeasureable when the concentration is less than 0.5 ppm. To truly understand overall performance, it s necessary to understand how the average efficiency can be determined from this information. The most accurate way to do so is to calculate a weighted average which takes into account the percent of the pollutants being generated in each of these ranges. This of course requires some information about the amount of

10 time that the device is operating in each range. It is then important to develop the weighted average not based on the number of hours the unit operates in each range, but on the total pounds of pollutants treated in each range. This distinction is important because a device operating at 0.1 ppm for 1 hour is actually treating only 1/10 th the amount of pollutants it is treating during an hour when operating at 1 ppm. Therefore, the total impact on overall efficiency at the low concentration is much less than at the high concentration. This can be an important factor to consider if inlet concentration to a device is highly variable, as it often is for wet scrubbers. For systems where the inlet concentration is more consistent, this becomes less important. Often, that is the case for VOC exhaust streams. Permitting Properly permitting a facility is very dependent on understanding actual emissions and control device performance. When that is fully comprehended, it is important to develop a permit that establishes accurate limitations that the facility can comply with and that provide the type of flexibility needed for the facility to make the continuous changes and improvements that are so crucial to this industry. Plant site emissions limits (PSELs) can be a useful tool to achieve this flexibility. PSELs essentially set an emissions cap over the entire facility, instead of setting stack or unit specific emission limits as permits often did in the past. Ideally, the PSEL approach should be combined with an express pre-approval that allows certain changes as long as the PSEL is not exceeded. This has the benefit of allowing a fab to make changes without delay, while still maintaining a high level of environmental protection since overall emissions still remain below permitted levels. Although this approach was viewed as unconventional in the past, more and more agencies who are aware of the flexibility needs of the semiconductor industry have begun to accept these types of permits.

11 Several factors are important to consider when establishing PSELs. Obviously, a thorough understanding of emissions and control device performance is crucial as described above. With this information, it is possible to set PSELs that are sufficiently above existing limits to ensure continued compliance, yet still sufficiently low to avoid major source thresholds if that is desired. It is also important to understand exactly what emission sources are covered by the PSEL. Paradoxically, even though PSELs are intended to apply plant wide, there can be units at a source that are not covered by the limit. Most local regulations acknowledge that certain activities are considered insignificant. Office activities, custodial chemicals, and emission units below certain size thresholds are common examples. This benefits the permittee because it means that resources do not have to be spent trying to quantify these difficult to quantify (and undoubtedly very small) emission sources to demonstrate compliance with the PSEL. In essence, these emissions don t count toward the PSEL. However, according to EPA guidance, they do count toward major source thresholds. Therefore, if a site is interested in avoiding major source thresholds it is necessary to demonstrate that the PSEL plus any insignificant emissions are below that threshold. This is not usually a problem if the permitted levels are well below the thresholds, since emissions from the types of activities mentioned above can typically be safely assumed to be very minor. In addition, EPA is developing a list of trivial activities (including some but not all of these) which truly don t count toward major source thresholds. However, there are cases where this can be an issue. For example, some locations exempt emergency generators from permitting if they are used fewer than a certain number of hours per year. EPA guidance states that potential to emit calculations from generators should assume 500 hours per year of operation, in the absence of any other permit limits. A site which had a NOx PSEL close to the major source threshold, and also had significant unpermitted generator capacity not covered by the PSEL could be seen by the agency as having a potential to emit above major source thresholds. These types of subtleties must be closely evaluated to ensure that the desired thresholds are avoided. Flexibility is also important in the compliance demonstration terms of the permit. Usually a permit will specify the type of compliance demonstration method that is to be

12 used, such as emission factors to calculate NOx emissions, or chemical mass balance for VOCs. If these are specified to rigidly, then the permittee may not be able to make changes to their methods if better data or measurement techniques are developed. When using emission factor methods, it is advisable to avoid putting the specific emission factor in the permit for this reason. For any compliance demonstration method, it is useful to have permit language which states that the method can be changed if data is developed that suggests a change is warranted. The permitting agency will often require language specifying that the change can only be made after agency review and approval, but is usually open to this sort of review being done without a formal permit revision procedure. This accomplishes the objective of being able to incorporate new information that more accurately reflects emissions, while providing the agency with assurance that the new information is accurate. Specifying the compliance provisions for control devices is also a very important permit element. Often a control device is required to meet a specified removal efficiency. As discussed above, understanding actual control device efficiency can be complicated so care must be taken to avoid accepting permit requirements which cannot be consistently met. Sometimes a permit condition will simply state a control device efficiency requirement, and state nothing further about averaging times, inlet concentrations, or other operating conditions. In that instance, the condition can be interpreted to mean that the control device must continuously be operating at or above that specified efficiency. A 1 minute period where the efficiency was below that level could be interpreted as a violation. As discussed above, control devices in the semiconductor industry are often operating at concentrations below the optimum treatment realm and at those times are achieving less than optimum efficiency. These situations do not represent a true environmental issue, since concentrations (and therefore emissions) are very low, but they could represent a technical permit violation if the language is not carefully written. Properly worded control device provisions can address this issue. One way is by specifying that the control device achieve either a specified removal efficiency or a maximum outlet concentration (e.g. 90% or 1 ppm outlet). Another method is stating that

13 the removal efficiency requirement applies only when the inlet concentration is at a minimum level (e.g. 90% at inlet concentrations of 5 ppm or greater). In any case, control device efficiency requirements should always specify a minimum averaging time of at least several hours. This will help avoid violations as a result of unusual events like very short duration emission spikes or simple anomalies in the data. Summary Air quality programs are among the most complex environmental requirements a facility has to deal with. The requirements, and therefore the potential for violations, are numerous. In addition, the factories themselves are very complex making comprehension of emissions and all related requirements difficult. A thorough and detailed understanding of emissions sources, control devices, and compliance requirements is essential to maintaining a high quality environmental program.

14 References i EPA AP-42 Compilation of Air Pollutant Emission Factors Volume I: Stationary Point and Area Sources, Fourth Edition ii Estimating Releases and Waste Treatment Efficiencies for the Toxic Chemical Release and Inventory Form, EPA 560/ ; U.S. Environmental Protection Agency, Office of Pesticides and Toxic Substances: Washington, D.C., 1987; pg. 6-4 iii SSA `99 proceedings Semiconductor Maximum Achievable Technologies (MACT Challenges) T. Higgs