 |
 |
 |
 |
|
Comments? Suggestions? Send us your feedback.
MicroMagazine.com |
CLEANROOM FACILITIES/TECHNOLOGIESJustifying a continuous contamination monitoring systemRoger Welker, Lighthouse Worldwide Solutions Traditionally, many approaches have been taken to measure contamination—or the factors believed to influence contamination levels—in cleanrooms. It is universally recognized that the cleanroom must be positively pressurized with respect to the general factory environment. In most cleanrooms, a pressure gauge or inclined tube manometer would be permanently installed on an outside wall. Once a day or once a shift the pressure would be read and the reading recorded. In this way, the cleanroom would be audited on a sampling basis. Other examples can be cited. Rotating-vane or hot-wire anemometer measurements would be taken near the face of HEPA filters to confirm linear discharge velocities, verifying that the room air recirculation system is functioning correctly. This would involve taking a large number of measurements and so would be done infrequently. Often such a survey would only be taken as part of an annual room certification. Room air velocities at the workstation level would be collected from a smaller number of sampling points and often manually surveyed once a week. Some environmental conditions are recognized to be more critical than room pressure or air velocity and would be checked at least once a shift (e.g., relative humidity) or as often as once a batch (e.g., starting pH of a bath). For critical contamination parameters, a continuous monitoring system would often be built into the process equipment or its dedicated environmental enclosure (e.g., to monitor temperature in a stepper). The continuous monitoring system could be easily justified because a clear link between the process parameter and yield could be made. Airborne particle contamination has long been considered important to measure and control. As a consequence, many manufacturing processes would require that airborne particle measurements be taken every day or every shift. However, the traditional methods used for sampling airborne particle contamination often produce erroneously low particle count results. In addition, the infrequency of particle count measurement makes it difficult or impossible to correlate results with yield. These limitations are often used to justify minimizing the frequency of manual surveys and have been cited as evidence that automated continuous contamination monitoring is not justifiable. Prior discussions of continuous contamination monitoring systems have tended to focus on the data management software1 or make the tacit assumption that a system will be bought without a discussion of how to justify one to skeptical management.2 Occasionally, clever methods have been developed for reducing the cost per sample point of a continuous monitor, but these still avoided discussing a method to demonstrate their necessity.3 It is often difficult to justify the large capital investment necessary to install a continuous contamination monitoring system. Often this is the result of faulty historical data where sampling practices precluded obtaining a true characterization of the workplace. There is a fear that hasty installation of a system will result in placement of sensors in locations where they are not needed. Finally, there are questions about the types of sensors that should be used, the required resolution, and other technical concerns that make decision making difficult. In order to overcome these difficulties, there needs to be a method which will permit one to objectively determine where and what kind of continuous contamination monitoring system is required. This article describes such a method and illustrates its use in several case study applications. Traditional Airborne Particle Measurements In the traditional approach to monitoring airborne particle contamination, an operator moves a conventional, self-contained optical particle counter to the workstation and places an isokinetic sample probe at some convenient location on the station, usually held in place by a stand. The conventional, self-contained optical particle counter usually contains a vacuum pump, power supply, and display, and often a printer. Inclusion of these features often results in a large, heavy particle counter. As a consequence, the conventional counters are mounted on a lab cart to facilitate moving about the cleanroom. This setup is conspicuous to production personnel. In addition, the isokinetic probe and its stand are frequently bulky and difficult to locate close to the product or process. Because of this, the probe is often arbitrarily placed on the workstation in a location driven more by convenience than any other consideration. Production personnel at the stations almost always cease all activity and move away when the particle count sample is to be taken. This results in elimination of actions that may be generating contamination during normal production, thus lowering the particle count in the sample. More often than not, sampling via the conventional approach can only obtain contamination associated with the cleanroom or clean bench. This is often described as cleanroom-idle sampling, that is, sampling in which the contribution of equipment and personnel is not counted in the total. Contamination associated with materials handling, load/unload operations, personnel-generated contamination, and the like are seldom included in such sampling. Improper censoring of data often takes place as well. When the particle count operator observes the rate of particle count, as long as the counts arrive at a relatively steady rate, the count is allowed to proceed. However, the particle count operator will usually terminate the count if a sudden burst of particles occurs, especially if the operator can associate the burst with some undesired activity, such as someone walking by. Such improper data censoring may be repeated as often as is necessary until an acceptable result is obtained. Quite often the only count considered acceptable is one that is below the class limit for the area being sampled. These two factors—sampling during workstation idle periods and improper data censoring—result in a historical particle count database that makes the cleanroom and its workstations appear to be well in control with respect to particles. Adverse actions are generally taken on the basis of such data. First, the manual particle sampling frequency is reduced so the labor cost associated with particle count sampling can be lowered. It is difficult to justify sampling more often when the data indicate the areas are in control from a particle count perspective. The facility's monitoring points are often divided into two or four subgroups, cutting in half or quartering the sampling frequency. In extreme cases, such as in very large assembly operations, this may be carried to the extreme that each sample location is visited only once per month. Second, given the apparent compliance with airborne particle limits, all appears to be in control, and switching from a manual sampling protocol to a capital-intensive continuous monitoring system simply cannot be justified. In order to correct the historical database and develop a more accurate description of the work area, a new sampling strategy must be developed. In the early stages of implementation, this approach should be designed to minimize capital expenditure. The strategy must also deal with the two chief factors affecting the accuracy of the particle count: sampling the wrong place at the wrong time and improper data censorship. Critical and Busy Sampling Critical and busy sampling can be described and defined in the following ways.
The critical location often positions the inlet to the particle counter in a place where laminar airflow does not exist. This approach works well since the bulky isokinetic probe can be eliminated, allowing greater freedom in placement of the inlet to the particle counter. The inlet tubing to the particle counter should then be attached to the workstation with brackets, tie wraps, or other means. This ensures repeatability of the sample location and protects the tubing from getting loose and interfering with the process. The hardware needed to implement critical and busy sampling costs a few dollars per workstation and takes minutes to install. Figure 1 shows an example of critical and busy particle sampling hardware installed on a magnetic recording head test stand.
The particle counter outlet end of the tube should then be terminated at some point on the workstation that permits the operator to attach the conventional particle counter to the sample tube, without disturbing the process. This allows sampling to continue without stopping the process, an approach referred to as busy period sampling. Modified Data Collection Protocol Once the low-cost critical and busy sampling hardware is installed, a new particle count protocol must be adopted in which data censoring is not allowed. The particle count operator observes and records the activity at the workstation during each sample. If no product is in production and the station is unoccupied, the sample is labeled as taken during Stage 1 operation, or a cleanroom-idle sample. If product is being processed, but no production personnel are present, the sample is labeled as taken during Stage 2 operation. If product, people, and tooling are present, the sample is labeled as a Stage 3 operation, fully operational and fully populated. If the process takes place within a tool or enclosure which effectively prevents contamination generated by the operator from getting on the product, however, the sample is labeled a Stage 2 sample. If the inlet to the particle counter is scraped, the tubing is bumped, or the inlet otherwise disturbed so that the count is invalidated, the sample is so notated. These occurrences indicate that the installation of the critical and busy sampling hardware should be corrected. By eliminating the data-censoring option, the rejection of otherwise valid data is also eliminated. In addition, by labeling the stage of operation for each sample, possible contamination sources can be diagnosed. For example, if Stage 1 particle counts are a significant fraction of Stage 3 counts and the Stage 3 counts are out of specification, the facility itself would be a fruitful place to begin searching for the source. Ongoing Use of Critical and Busy Sampling When a sample location is identified as out of specification with respect to airborne particles, a second phase of investigation should begin. The stand-alone counter may be used like a Geiger counter, sniffing out specific particle generation points. If these points can be located, fixed, and kept under control by manually sampling at some tolerably low frequency, then a continuous monitor is not justifiable. However, the critical and busy sampling hardware and protocol should continue to be used. If workstations are identified that require continuous monitoring, then the continuous monitoring system is used with the critical and busy sampling system. Whenever an alarm goes off, the manual particle counter is brought back to the location and used again in the Geiger counter mode. Case Studies of Traditional Versus Critical and Busy Sampling The following three case studies offer comparisons of traditional sampling with critical and busy sampling. Case Study 1. Table I shows the results of sampling two sets of identical workstations on adjacent disk-drive head-gimbal assembly production lines in a Class 10,000 ballroom-style cleanroom. The data listed are the average and standard deviation of particle concentration, in particles per cubic foot (>=0.5 µm diam). With the traditional manual sampling protocol, the stations had been found to comply with the airborne particle count requirements of a Class 10,000 cleanroom. The particle count increases slightly using the critical and busy sampling protocol but not enough to change the conclusion that all stations comply with Class 10,000. There are only slight differences between stations on line A versus line B. Results similar to those listed in Table I and plotted in Figure 2 are often found in mixed-flow cleanrooms. The general room contamination dominates the contamination generated at the individual station. Such data indicate a continuous monitoring system might not be necessary for these stations.
Table I: Traditional versus critical and busy sampling in a mixed-flow, Class 10,000 cleanroom (particles/cu ft >0.5 µm).
Case Study 2. Table II lists comparative data for two identical sets of Class 100 workstations in the same cleanroom as the first case study. These Class 100 stations are located under vertical laminar-flow (VLF) units, effectively isolating each station. When sampled with the traditional approach, none of the nine stations from either line A or line B exceeds Class 100. Conversely, the particle count averages and standard deviations increase for all 18 stations when sampled using the critical and busy protocol. In 7 of the 18 cases, critical and busy sampling shows stations far dirtier than Class 100. A comparison of workstation 6 on line A versus its identical counterpart on line B reveals that the line B station is almost 10 times dirtier. Figure 3 plots some of the data listed in Table II in order to illustrate differences between the two sampling protocols.
Table II: Traditional versus critical and busy sampling in Class 100 vertical laminar-flow workstations in Class 10,000 ballrooms (particles/cu ft >0.5 µm). This example illustrates two common results of the use of critical and busy sampling in laminar-flow work areas. First, the emissions from the individual stations are evident because the mixing effect of the non-laminar-flow cleanroom is eliminated. Second, the differences between pairs of otherwise identical stations can be detected. Case Study 3. This example discusses operations under Class 100 VLF units in a Class 1000 cleanroom at a disk-drive facility. Average values are shown in Table III, omitting standard deviations, because of the limited sample size of the survey. All 20 workstations sampled using the traditional approach easily meet Class 100. The boast in this facility was that most of the stations would meet or exceed Class 10. The critical and busy samples indicate most do not even satisfy Class 100 requirements. The worst discrepancy is found in location 18, where Class 1000 was exceeded. Note that using the critical and busy sampling approach, there are sufficient particles available to allow use of a 0.5-µm-resolution, 0.1-cu ft/min optical particle counter at nearly every station. If the data collected using the historical method were utilized, either a 0.3- or 0.1-µm-resolution counter would probably have been chosen, greatly increasing the cost of continuous monitoring.
Table III: Average obtained by traditional versus critical and busy sampling in Class 100 VLF units installed in a Class 1000 cleanroom (particles/cu ft >0.5 µm). Figure 4 plots some of the data listed in Table III. The figure illustrates an important feature of the critical and busy sampling approach in workstation 7. In order to sample using the traditional protocol, the operator had to open the doors to the work cell. The safety interlock would stop the machinery inside, eliminating its contamination contribution. The critical and busy sampling hardware was mounted so the operator could connect to the sample tube without having to open the enclosure. Thus, the machinery would continue to operate, allowing the detection of its contribution.
Figure 5 shows a plot of traditional versus critical and busy sampling in hybrid workstations, where an operator and automated tooling work together. A comparison of Figures 4 and 5 illustrates a fundamental belief in a way seldom demonstrated as clearly: people are a major contamination contributor. Trend, Cyclic, and Burst Patterns of Particle Generation. In addition to the average particle concentration prevailing at a workstation, trend, cyclic, and burst patterns of particle generation are also of major concern.4 When sampled over a long duration, the average particle concentration may appear to be within control limits, but looking at the data in more detail often reveals unwanted particle concentration behaviors in the workstation. Upward trends in particle count are undesirable because they may, at some future moment, exceed the control limits. Examples of upward trending can be seen where stations gradually become dirty between deep cleaning intervals. Since the rate at which stations become contaminated is not perfectly constant, it is not easy to predict when the next deep cleaning should be scheduled. A continuous monitoring system may be useful in such a situation. Cyclic patterns of particle generation are a special case, where the bursts have a repeatable pattern. These patterns can usually be associated with specific activities on the station. If associations can be established, fixes are often easy to develop and implement. Experience has shown that a cyclic pattern can be adequately controlled using manual monitoring in conjunction with critical and busy sampling hardware. Random contamination bursts are observed in nearly every cleanroom and are often associated with sudden, catastrophic events. A good example is shedding from an electric motor, such as the example illustrated in Figure 6. This stepper motor was continuously monitored for more than a week. The >=0.5-µm counts downwind of the motor started in the 15—30 particles/cu ft range, but cleaned up quickly to 1—3 particles/cu ft. Two large bursts can be seen. Each sample reflects the average over 10 minutes, collected at 0.1 cu ft/min. Averaged over 7+ days, the electric motor produced only 16 particles/cu ft. The second burst exceeded Class 100 for 25 hours. With a once-a-week manual sampling plan, the chance of detecting this burst is only 1 in 7. The first burst, with a duration over Class 100 for 5 hours, had only a 1 in 37 chance of being detected with weekly sampling.
The following two case studies examine the results of extended-duration manual monitoring under different conditions. Case Study 4. Ten different Class 100 workstations in 10 different disk-drive, disk-media, and semiconductor factories were monitored using critical and busy sampling hardware.5 These data were also compared to traditional monitoring results. Sampling was of sufficient duration that the percent compliance could be calculated. Percent compliance is the percent of time that a station is below its particle count limit when it is monitored. High percent compliance is considered to be good. Stations with very low percent compliance are highly likely to be detected in a traditional, once-a-week particle sampling protocol, as illustrated in Figure 7.
Case Study 5. In this study, a manual optical particle counter was used to sample a workstation in a disk-drive facility using the critical and busy sampling hardware for several hours. The data were collected once a minute in a Class 100 VLF clean bench located within a Class 10,000 ballroom. The particle count operator observed and recorded the activities at the station but did not interfere with the actions of the production operators in any way.
Table IV: Extended duration manual particle monitoring data (particles/cu ft >=0.5 µm). Table IV is a partial summary of the particle concentration data, averaged to the nearest 5 particles/cu ft (0.5 µm) and the particle count operator's notes of the activities in the station. The results are also plotted in Figure 8. The operator's notes provide a very clear understanding of what is happening. Wipedown is a relatively messy process since it stirs up large amounts of contamination. Setup or waiting for work in progress (WIP) generates little contamination. Assembly and especially soldering generate large amounts of airborne contamination. The arrangement of the items on the station is not fixed. For convenience, the second operator moved the solder fixture and fume extraction system, with disastrous results. During soldering, the first operator averaged 370 particles/cu ft, the second averaged more than double that amount, or 746 particles/cu ft.
This case study illustrates an example where a continuous monitoring system may be justifiable. Workstation layout must be such that the reach and comfort of the operator can be accommodated. A continuous monitoring system might be a useful tool to keep particle counts under control after such rearrangements. Case Study 6. Magnetoresistive (MR) heads are extremely electrostatic discharge (ESD)—sensitive devices. Modern static-safe facilities thus require many ESD protection tools to allow for their safe manufacture. Air ionizers are one of the most important tools provided for these static-safe work areas. Charged-plate monitors have traditionally been used to gauge ionizer performance. During weekly audits, the charged plate measures discharge times and float voltages. In this procedure, the ESD technician places the sensor of the charged plate as close to the intended product location as possible. The air ionizer is occasionally found to have drifted out of balance and need service. This service may consist of cleaning the emitter points on the air ionizer, but simply cleaning the emitter points is sometimes inadequate and the ionizer must be manually balanced. One of the less-understood features of ionizers' performance is that they interact with the environment. That is, grounded objects on the workstation below the ionizer tend to drain charge to ground. The polarity and amount of charge drained off is a function of the distance to and position below the ionizer's emitter points. Relocating objects on the station thus can change the balance of the ionizer. This can happen frequently in a development site, where tooling and stations are often changed because of changes in products or process flow. In order to more fully characterize these changes, an electrostatic charge monitor equipped with a 20-pF plate was installed at a station for 4 days. Each 10-minute sample was scanned for the maximum positive and negative voltage swing. Figure 9 depicts the variation in float potentials measured for one station in the development cleanroom. The layout of the stations was checked every shift with any changes noted.
Figure 9 shows that larger variations were observed in the station during the first shift (observations 0—50, 160—210, etc.) than during other shifts. The development cleanroom was being used as a first-shift production area and was practically empty otherwise. Note, however, that an engineering experiment was performed on the second shift of the third day. In this experiment, a tall measurement stand was placed on the station, almost directly under the ionizer. The station was rearranged to accommodate the stand. However, when the stand was taken away at the end of the second shift, the station was not returned to its original layout. Hence, on the third day the balance in the station swung to a strong positive imbalance. Case Study 7. In this example dealing with continuous airflow monitoring, a very large cleanroom in a disk-media factory was equipped with 54 modular laminar-flow units. Once a week a cleanroom technician would do a velocity survey, measuring the linear air velocity discharged from the filters in each module. Approximately every other week, at least one of the modules would be found to have very low or no air velocity. This creates unwanted horizontal airflow. Clearly, discovering this just once a week is undesirable, so the challenge is how one would design a cost-effective continuous monitoring system to watch for the condition. The answer lies in the design of the cleanroom, as illustrated in the plan shown schematically in Figure 10.
The design of the cleanroom lent itself to definition of four airflow zones, labeled A, B, C, and D in Figure 10. From 10 to 16 laminar-flow modules supplied air to these zones. It was immediately recognized that airflow from any module would change the horizontal airflow through the restricted areas defined by the return plenums. In order to provide a module flow monitoring system, five hot-wire anemometers were installed in the restricted locations, numbered 1 through 5 in the figure. Such anemometers are frequently used to measure vertical air discharge velocity from the HEPA filters. In this application though, they were mounted to monitor horizontal rather than vertical flow. After installation of the horizontal-flow monitors, no imbalance condition went unnoticed for longer than a single shift. Of course, the monitors would not tell the cleanroom technician which module had failed, but they would show which intersection between zones was out of control. The technician would then go to the out-of-control intersection and determine the direction of the horizontal flow. This would then determine in which zone a module had failed, allowing the technician to quickly survey and locate the failed unit. Conclusion An effective method has been developed to permit the assessment of the need for continuous contamination monitoring. The method can be used for sampling and measurement of airborne contamination, monitoring voltage balance in a static-safe work area, and monitoring the function of laminar-flow modules. This method optimizes the placement of sample points in order to allow a correct characterization of the workplace to be made. The data obtained permits the selection of particle counters or other sensors using a low-cost strategy. References 1. Pariseau D, "The Future of Cleanroom Monitoring Systems," CleanRooms, 9(1):39, 1995. 2. Livingston J, Bower R, Pochy R, and Branst L, "Using an Automated Cleanroom Monitoring System to Maximize Contamination Control," MICRO, 15(9):113—123, 1997. 3. Fardi B, "An Evaluation of a Cost-Effective and Efficient Airborne Particle Monitoring System," in Proceedings of the 38th Annual Technical Meeting of the Institute of Environmental Sciences, Mount Prospect, IL, IES, pp 38—44, 1992. 4. Bzik TJ, "Statistical Management and Analysis of Particle Count Data in Ultraclean Environments," in Proceedings of the Microcontamination Conference, Santa Monica, CA, Canon Communications, pp 93—118, 1985. 5. Query CF, "Continuous Monitoring in Cleanrooms: A Guide for the First-Time User," presented at Asia-Pacific Magnetic Recording Conference, Singapore, Data Storage Institute, July 1998. Roger Welker is founder and principal scientist of R.W. Welker Associates. He was most recently senior director of application technology for Lighthouse Worldwide Solutions (Milpitas, CA). Before joining LWS, Welker spent 15 years in high-technology development and manufacturing at Micropolis, Seagate, and IBM. Before that, he spent 11 years in applied R&D, focusing mainly on applications of fine particle technology. He holds a BS in physical chemistry from the University of Maryland (College Park). Welker has authored or coauthored more than 60 papers and is a member of the Institute of Environmental Sciences and Technology, the American Association for Aerosol Research, the Electrostatic Overstress/Electrostatic Discharge Association, and the Data Storage Institute. (Welker can be reached at 818/368-0557.)
MicroHome |
Search | Current Issue | MicroArchives Questions/comments about MICRO Magazine? E-mail us at cheynman@gmail.com. © 2007 Tom Cheyney |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Figure 1: An example of critical and busy particle sampling hardware installed on a magnetic recording head test stand.
Figure 2: Traditional versus critical and busy sampling in a mixed-flow, Class 10,000 cleanroom.
Figure 3: Traditional versus critical and busy sampling in Class 100 vertical laminar-flow workstations in a Class 10,000 ballroom.
Figure 4: Comparative sampling data for Class 100 vertical laminar-flow workstations in a Class 1000 cleanroom.
Figure 5: Comparative sampling data for hybrid automation in Class 100 vertical laminar-flow workstations in a Class 1000 cleanroom.
Figure 6: Particle bursts from an electric motor.
Figure 7: Traditional versus critical and busy sampling and percent compliance in Class 100 workstations.
Figure 8: Extended-duration manual sampling data from Table IV.
Figure 9: Charge level in a static-safe cleanroom workstation.
Figure 10: Plan view of a horizontal flow—monitored, vertical laminar-flow cleanroom.