Product In Action

Automating Slurry Delivery Reduces Labor, Improves Data Collection, Increases Yields

The PanelView 550 controller gives users numeric input options along with its graphical electronic display.

Downtime can be deadly in the semiconductor business, largely because of the high costs of manufacturing and the enormous market pressures to deliver product in a timely manner. Critical system failure can put wafer processing so far behind that lost time becomes difficult to recoup. Serious breakdowns in production are often caused by seemingly simple things, such as pump or valve failure and filter loading.

The process is further complicated by the volatility of the chemicals used in the fab. For example, the slurries employed during chemical-mechanical polishing (CMP) are often blamed as a major source of wafer defects, scrapped product, and production downtime. The quality of slurry from the same batch, lot, and manufacturer can vary dramatically. Some lots may have a higher percentage of larger, potentially damaging particles than others, which requires that each batch of slurry be tested individually. Outsourcing this testing is inefficient because it can leave a production line idle. Moreover, slurry does not maintain its particle size over time as the molecules cluster together and begin to agglomerate. Without proper monitoring, even a good batch of slurry can turn bad.

Engineers at Kinetics Systems's Phoenix-based Chempure Systems unit—a supplier of turnkey state-of-the-art pumps, titrators, blending, and distributing systems through its MEGA Systems product line—knew they had to improve things drastically. When they examined the problems plaguing slurry delivery and CMP processing, they realized that most of the guesswork and manual measurement had to be taken out of the existing process.

"We realized that slurry consistency varied from the top of the barrel to the bottom, from batch to batch," explains Jeff Wilmer, Chempure's R&D director. "Everywhere we looked in the process, we kept coming back to particle size and slurry inconsistency." His company's goal was to find a better way to stabilize the chemistry and to improve the consistency and quality of the polishing slurry.

An initial area of concern was that while nonideal particles represent a very small percentage of the total particles in slurry, they are responsible for most of the microscratches on the wafers. "Traditional metrology packages don't even detect the presence of the damaging nonideal particles," says Wilmer. He knew that eliminating the microscratches would require a new method of identifying nonideal particles.

Kinetics engineers focused their research on the slurry distribution system, from the source to the tool. Their solution involved real-time monitoring and sampling of the slurry to allow manufacturers to immediately determine the particulate composition before it was introduced into the line. Prequalifying slurry wasn't a new idea, but automating the delivery process was. The group developed a technique in which the slurry travels first to an automated supply line analysis unit before it is delivered to the polishing system. Slurry can be analyzed—and disqualified—at this station safely without contaminating the line or affecting production. Automation eliminates the inefficiencies and partial inaccuracies of manually checking the slurry, providing the user with more-reliable slurry output.

When the engineers dissected the slurry supply line, they evaluated the concentrations of nonideal particles throughout the system's entire circulation flow. They tested slurry pumps, filters, and the complete circulatory loop to identify problem areas, and developed a system that eliminated postbarrel agglomeration. Design flaws in the system that promoted agglomeration were eliminated, such as tortuous paths in the slurry flow and slurry drying after contact with air. After building a slurry supply line prototype in their R&D lab, the engineers refined the process by testing actual lines.

Early in development, Kinetics chose Rockwell Automation (Milwaukee, WI) to provide the automation platform for the slurry-qualifying and blend/distribution tools, largely because of Rockwell's broad product line. The adaptability of Rockwell controls and software also gave Kinetics the opportunity to integrate into existing systems. "Many of our existing customers had already been using Rockwell Automation products in their lines," says Wilmer. "The adaptability of their software made integrating with their lines easier."

Kinetics and Rockwell jointly developed an automated slurry monitoring, analysis, and distribution program, using an Allen-Bradley SLC 500 system, Rockwell's RSView32 GUI software, and an Allen-Bradley PanelView 550 for local operator interface. The resulting software collection and data acquisition (SCADA) programmable controller system features sensors and alarms applied to the slurry tanks which give the fab improved system monitoring and control. The sensors measure particle sizes at different levels in the slurry barrel. If the percentage of scratch-causing particles exceeds the user-established limit, this information is sent to the SLC system and an alarm graphically alerts those individuals monitoring the system. Then, instead of using the contaminated slurry or shutting down the line, the redundant system automatically switches to a different barrel of slurry. The alarms can be routed to a pager, e-mail system, or voice annunciation through Rockwell's RSMessenger, in addition to being displayed and controlled by the PanelView 550 and RSView32.

"We collaborated on devising an automated solution to the semiconductor polishing process," notes Wilmer. "Kinetics researched how automation would improve the blend and dispense systems, and Rockwell provided the platform to achieve the desired results."

The Kinetics system is redundant, so if a pump, filter, or valve fails, another will automatically take over, dramatically reducing downtime. The RSView32 and RSMessenger control an alarm system that notifies users if an event occurs. Users can create set points that teach the SCADA control system to recognize how the system behaves when it is failing and how it should react. If a primary pump fails, for example, the SLC immediately recognizes the failure and switches to the backup pump while triggering an alarm, alerting the user that a pump must be replaced. Meanwhile, the manufacturing process has not missed a beat because of the automated redundancy. A detailed alarm message appears on both the PanelView monitor and on a remote monitor (placed in the subfab, in an office, or both), which graphically displays the problem and lets the engineer know what part needs to be replaced.

 

RSView32 gives the user a graphical, real-time representation of the system and alarm functions. Users can access the system remotely without inspecting the entire system manually.

The automated system can also predict system failure. The user sets operating ranges in the program, based on preset indicators of pressure, flow, and other parameters, which are used to diagnose the beginning of system failure. If the SLC detects fluctuation in the system's readings, it warns the operating engineer of a potential problem. The engineer can address and rectify the situation before wafers are damaged or the system shuts down.

This automated process can also operate with other instruments to maintain consistent concentrations of the chemicals used throughout the polishing process. Over time, these volatile chemicals can degrade from simple exposure to environmental conditions. Instruments used to gauge the chemical concentration within the slurry can signal subsystems to deliver precise amounts of the required chemical dosage to replenish the main delivery system. The SLC- or PLC-driven subsystem determines the amount of a specific chemical necessary to bring the current concentration back within the specified set points and controls the dosing system.

RSView32 displays the particulate levels throughout the system and automatically collects the data for historical reports. Users can then compile productivity reports based on the graphical data and find correlations. For example, if a high percentage of nonideal particles cyclically develops near a certain pump and the frequency of yield-impacting microscratches peaks during these intervals, users can adjust the SLC alarm set points of this pump to signal this situation when it develops.

"In the past, engineers didn't have the ability to look at the background data to see why production was poor for a given day," says Wilmer. "With all of the historical data now automatically available to our customers, they can monitor the distribution systems to look at circumstances in every blend that comes through the distribution stations."

"Our customers, including several major semiconductor manufacturers, have seen improvements in yield," he continues. "We have provided them with a powerful tool that increases yield, reduces labor, and improves data collection."

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