MANAGEMENT STRATEGIES

Using an automated cleanroom monitoring system to maximize contamination control

Jack Livingston, Ray Bower, Rocco Pochy, and Lee Branst

Cleanroom monitoring is an essential component of the contamination control programs that are critical to achieving high yields in the semiconductor and disk-drive industries. Today, the manual monitoring methods typical of the recent past are being replaced by sophisticated automated systems that can provide real-time, continuous data from multiple sensors installed adjacent to process tools. In addition to providing alarms of catastrophic events in visual, audible, or electronic format or as pager notification to engineers and operators, such a cleanroom monitoring system (CMS) can be used to identify trends and provide early warnings of contamination problems, correlate contaminant excursions with other cleanroom parameters, and reduce failure-analysis time.

Mounting stand (center) for particle counter located near lot rack and stocker.

This article describes one semiconductor manufacturer's experiences with an automated CMS. Motorola's MOS 12 fabrication facility in Chandler, AZ, elected to install the system to provide automated real-time monitoring of cleanroom particle levels, to help ensure environmental compliance, and to provide direction on where ongoing improvements could be made. While the applications of the CMS at MOS 12 are similar to those that could be implemented at other high-tech manufacturing facilities, it should be remembered that each fab is unique in equipment, procedures, and production flow.

Automated CMS Specs

When MOS 12 was being planned and built in 1994, contamination control personnel decided to move beyond the manual techniques commonly used in cleanroom particle monitoring. In the manual method, cleanroom or quality assurance personnel carry portable counters throughout the fab to measure particle levels at different sites. The counters are moved to a new location after a specified period or left at critical sites for unattended, continuous data collection. While effective, this technique is manpower intensive, making it difficult to collect sufficient data for trend analysis and troubleshooting. Having a CMS allows personnel to be used for selecting and analyzing relevant data and for fixing problems, rather than for simply collecting data.

The system that Motorola obtained from Lighthouse Associates (Milpitas, CA) contains two basic modules, the engineering control station (ECS) and the sensor interface unit (SIU). The ECS microcomputer runs Windows-based system software, which performs operator-designated tasks and can display real-time data in a variety of formats, including graphs, charts, tables, and sensor maps. The SIU controls the data collection interface to the sensors. It can accommodate sensors with different interfaces and manages data collection by adjusting reporting rates in response to facility conditions. One ECS can manage multiple SIUs and can manipulate data from any remote location. At MOS 12, two SIUs control approximately 100 sensors (200 would be maximum), including two multiport units monitoring 12 locations each. The CMS can trigger data collection based on relay inputs from process equipment such as door or robot movements. This relay control permits data collection only when a critical condition arises.

An area control unit (ACU) provides remote connectivity, data redundancy, and point-of-use graphics terminals. The ACU and ECS can access each other via modem or RS-232, RS-422, or RS-485 interfaces. When a connection is terminated, the units continue to operate independently; they then automatically resynchronize their data when the connection is reestablished. Thus, the ACU continues to collect data even if data transmission to the ECS is interrupted.

About 70 of the CMS sensors can detect particles >=0.3 µm, while the others can detect particles >=0.1 µm. In data displays, sensor locations are identified by fab bay or actual unit location, as needed. Typically, there are six sensors per bay, almost all of which are placed in areas where wafers are exposed and therefore vulnerable to ambient air conditions. Roughly 70% of these sensors are placed adjacent to a tool load port; the others are put near a table surface where the wafers are removed from their carriers before being loaded into a tool. The monitored bays contain such wafer-processing equipment as diffusion furnaces, etch stations, photolithography systems, ion implanters, deposition systems, and wet processing tools. Wafer inspection stations are also monitored.

There is a separate automated building monitoring system, which collects data on temperature, atmospheric pressure, humidity, and so forth. These building data are fed into the CMS for access by contamination control engineers. If, for example, a temperature or pressure shift or spike indicates that something in the environment has changed, then an investigation can be launched to identify the cause of the deviation.

Mounting stand (center from bottom) for particle counter located at load port height in atch bay.

As MOS 12 continues to add new equipment and processes, the CMS is also upgraded. Originally only individual sensors were used with the system. Then, about two years ago, the plant purchased a sample multiport sensor for integration with the CMS. The new sensor optimized signal timing between devices and permitted the sampling of multiple locations using a single sensor unit. With this unit, a mechanical selector allows a choice from among the dozen input ports spread throughout a bay. The disadvantage of this setup is that samples are taken sequentially instead of simultaneously. On the other hand, many more locations can be sampled for the cost of one sensor and a few accessories, thereby enabling wider coverage for the same or less expense. The success of the multiport sensor trial led Motorola to purchase a second 12-port unit. (The CMS is not limited to 12 sensors in a multiport block, but this number was germane for the MOS 12 facility.) These two sensor units are now in place below the bay floors with tubing running through the floor to the sensing locations.

Early Warning and Trend Analysis

The CMS at MOS 12 can be used in a number of ways. Most important, it can provide a real-time alarm in the event of a severe contamination event so that immediate action can be taken. An alarm is usually not triggered unless all the sensors in a particular bay fall below the Class 10 compliance criteria; that is, if the particle level in a bay averages above 10 particles/cu ft over a specified time interval. Fortunately, such occurrences are exceedingly rare, which means that the CMS is used primarily to collect data. This information is compared to baselines that have been set for each monitoring location, and preventive measures are taken when there are deviations or trends away from those baselines.

Figure 1: Example of a monthly bar graph showing the percentage of time each photolithography bay was in compliance with Class 10. (Data have been normalized.)

Figure 2: Example of a weekly bar graph depicting compliance data by bay and worst sensor. (Data have been normalized.)

Data Displays. The collected data are output in a series of reports and graphs, which are produced on monthly, weekly, and daily schedules. The monthly and weekly charts are used to monitor progress toward eliminating problems and associated high particle counts and to return to compliance with standards. While particular problems or trends are reported to management immediately, the monthly bar graphs compiled for management (see Figure 1) consolidate in a readily observable form the tremendous amount of data that can be generated by the CMS. Weekly reports, examples of which are given in Figure 2 and Table I, provide a means for tracking performance in each bay over a convenient time interval and are displayed for everyone in the plant to see. Another type of chart plots data for individual sensors in a bay (see Figure 3). This chart can cover any time interval desired and is generated as needed. When several sensors are charted together, each is identified by a different type of line (dashed or dotted, for example); lines of different colors can also be used to make identification even easier.

Table I: Example of a weekly summary report showing particulate excursions. The table indicates performances for each bay and highlights the worst sites. (Data have been normalized.)

Figure 3: Example of a weekly chart of particle performance measured by a sensor in a photolithography bay. (Data have been normalized.)Mounting stand (center from bottom) for particle counter located at load port height in etch bay.

Perhaps the most useful report is the list compiled daily of the 10 sensors with the worst performance in the previous 24 hours in terms of the highest particle counts and percentage of readings that are higher than 10 particles/cu ft (see Table II). These reports are examined by contamination control personnel, who make decisions about which counts are unusually high or indicate a troublesome trend. One way they examine the data is by plotting the particle counts for a sensor versus time. Are the high counts steady or are there occasional bursts of high counts? Is there a trend toward higher counts or is there a wide oscillation in the data? The answers to these questions determine which areas need to be investigated further. There is not necessarily a cutoff level below which readings are ignored; rather, the compliance team uses its experience to form judgments regarding which readings justify some type of follow-up action.

Table II: Example of a daily listing of the 10 sensors reporting the worst performance over 24 hours (midnight to midnight), by particulate count and noncompliance percentage. (Data have been normalized.)

The CMS can also track particle counts over shorter time intervals. For example, instead of showing daily data, a table can have data points for every 10 minutes over a portion or all of the prior 24-hour period. Table III lists the particle counts from four sensors at 10-minute intervals over 1 hour.

Table III: Example of a report of actual readings for a 1-hour period in a photolithography bay.

Examples of CMS Applications. On one particular day at MOS 12 the worst-ranking sensor indicated a high percentage of compliance failures, and personnel decided the situation warranted further attention. When this determination is made, standard procedure dictates that a portable counter be taken into the bay and separate, manual measurements be taken of the areas around the sensor site. This investigation determined that particles were being generated because of an integration problem between a piece of equipment and the equipment installation site. After appropriate remedial steps were taken, the contamination was eliminated. Sometimes, if a graph has a certain signature, the data can be correlated with some unusual event, such as an equipment artifact. (There are, of course, a number of other possibilities.)

Another incident reflects a situation in which a daily report displayed high particle counts from a sensor in a diffusion bay. A review of the associated graph determined that these readings were consistent with a real event. A portable counter was then brought in to make measurements in various locations around the sensor site, beginning at the ceiling where the flow of clean air enters the area. Was the air entering the area clean or were there already particles in it? In this case, the investigation revealed that particle counts were high directly beneath certain ceiling locations, and further examination determined that some of the spun-glass HEPA filters had been damaged during recent maintenance work when equipment had touched the filter faces. After the filters were repaired and the maintenance people were trained in the proper way of operating in the vicinity, the particle counts decreased and there was no further damage to the air supply system. This incident shows the value of using a CMS to make continuous measurements within an area. If a portable counter had been used to take random readings, it most likely would have taken a long time to discover this problem since the suspect areas may not have been checked.

Another application example involves a photolithography bay. In this case there were two incidents of high counts within a general area. After it was determined that the counts were real, particle levels were verified with a portable counter, again starting at the ceiling. It was discovered that contamination was emanating from that area in bursts, from both the filter edges and the filter media itself. Further investigation revealed two causes of this problem: First, a few of the HEPA filters were no longer seated properly. Even though these filters are seated in a gel, they can come loose over time, and a few of these had. Second, the airflow through these particular filters was too high. An excess of air pressure can cause the filters to expand and leak. Once the filters were reseated and the airflow was adjusted, the particle problems went away—for a while.

Not long thereafter, the CMS indicated that a second contamination event was occurring in this location, albeit with lower particle counts. The same exploratory procedure was undertaken and the airflow was found to be satisfactory. Then the investigators determined that the high particle counts were concentrated in the worktable area, which had become a very highly used location. It turned out that the design of a particular portion of the work surface was not optimal. Although this situation was known previously, it had not caused any problems before. Now, however, the increasing amount of work being done in the area was causing particles to accumulate in an air vortex instead of being blown downstream as they would have been with a properly designed airflow. The solution was to clean up the work area and redesign the airflow pattern in the vicinity.

Another example of a graph designed from CMS-generated data is shown in Figure 4. This display presents data from all the sensors in an etch bay collected over the July 4 weekend. Production activity essentially ceased on the afternoon of July 3 and did not resume until early July 7. This lull is reflected in the decrease in particle counts, which indicated that the area was in compliance with Class 1 for the period. Although these data do not identify a particular contamination source, they support the theory that a substantial portion of ambient cleanroom particles are generated by personnel and production-related activity.

Figure 4: Example of a weekly chart of particle performance measured by the six sensors in an etch bay. (Data have been normalized.)

Beyond Scheduled Reporting

In addition to providing displays of sensor-derived data, the CMS can receive and display data from the building monitoring system (see Figure 5). This capability sometimes allows contamination control engineers to identify potential problems that facilities engineers are not aware of. For example, on one occasion there was an increase in particle counts throughout a bay at a time when there was no construction work or other unusual activity in the vicinity. A subsequent investigation of airflow in the area revealed that the air was stagnant, and facilities personnel then discovered that a fan-operation sensor had failed to detect that the fan had stopped. Restarting it eliminated the immediate airflow problem, but the incident indicated that facilities-sensing capabilities needed improvement.

Figure 5: Example of a chart of data from a building monitoring system, showing temperature in an equipment chase between two cleanroom areas. These data are fed into the CMS for analysis by cleanroom personnel.

While the CMS reports enabled MOS 12 personnel to solve the above-mentioned problems quickly, there can be an embarrassment of riches as far as data collection capability is concerned. The issue becomes how to manage the tremendous amount of information that is collected. In a few cases, the sensors took samples every minute for extended periods. How can these data be used as a starting point for taking appropriate actions and as a measure of ongoing performance? The CMS permits data to be exported to spreadsheet programs such as Lotus or Excel. The sensor names, date and time ranges, and method of data-point averaging are all configurable, and these selections can then be stored in a separate file for quick editing to produce another spreadsheet with related data. In this way, the sensor data can be compressed, used for comparisons, or prepared for more-complex statistical functions.

Conclusion

The experiences highlighted above represent only some of the capabilities of the CMS. In addition, the system has the ability to track sensor performance, permits parameter modification during system operation, and includes security features to limit access in certain areas of the system. Current and previous system configurations can be compared, and data redundancy and system storage provide backup in the event of component failure.

The installation of the CMS at MOS 12 has provided much greater area coverage and more frequent data sampling than would have been possible with a manual system. In addition, it has allowed personnel greater opportunity to discover and investigate particle events and to plan for, design, and integrate ongoing upgrades to cleanroom facilities and equipment.

Jack Livingston is involved in contamination control at the Motorola MOS 12 fab in Chandler, AZ. An employee of the company since 1973, he recently completed a tour at Sematech in the Contamination-Free Manufacturing and Factory Integration programs. Livingston received his MS in physics from Arizona State University. He is a member of the Institute of Environmental Sciences and Technology and has written several articles pertaining to defect and contamination reduction. (Livingston can be reached at 602/814-4904.)

Ray Bower is an equipment specialist at MOS 12. He has worked in manufacturing groups at Motorola since 1989 and now specializes in cleanroom engineering functions, including system administration for the MOS 12 cleanroom monitoring system.

Rocco Pochy is vice president of engineering at Lighthouse Associates, Milpitas, CA. Previously, he held positions at NASA/Ames Research Center, Lawrence Livermore National Lab, Sun Microsystems, and Lighthouse World Solutions. Pochy has a BS in physics and computer science from San Jose State University, where he is studying for an MS.

Lee Branst is a high-technology communications specialist at Caracal Communications, which he founded in 1986. He holds a BS in science management from Case Institute of Technology and an MBA from Weatherhead School of Management, both of which are parts of Case Western Reserve University.

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit