Today, Statistical Process Control (SPC) is the gold standard of quality control because it helps manufacturers to maximize production of on-spec product with minimal waste and rework.

SPC uses graphical control charts to determine when a process should be adjusted. The following guidelines can help silicon chip and solar cell manufacturers make the best use of SPC for cleanroom quality control:

•      List the criteria that will maintain a stable environment and assure product quality to focus monitoring on critical process control conditions.
•      Structure the testing process to focus on these criteria and avoid extraneous testing.
•      When troubleshooting, think outside the box or farther down the line.
•      Use computational fluid dynamics (CFD) modeling to help identify the source of a problem.

Here are insights on how:

List the criteria that will maintain a stable environment. It’s important to narrow the focus to monitoring only those conditions that are critical for maintaining process control. Specific SPC monitoring criteria are determined by the type of facility. The major environmental factors that affect the quality of a silicon chip or solar cell are airflow temperature, humidity, air pressure, particles, light wavelength and electromagnetic interference (EMI). Therefore, these should be the key monitoring criteria in any facility.

Conductivity is sometimes overlooked as a monitoring requirement. During design and construction, the facility should have been properly grounded to control conductivity, directing energy to ground at the proper rate without creating sparks. Nevertheless, conductivity should be monitored to assure that the grounding system is properly functioning.

Monitoring criteria should be based on the project’s requirement for manual versus automated monitoring based on monitoring frequency. This can be determined by comparing the costs of manual monitoring with the capital and life cycle operational costs of an automated system. If the quality specifications of the product require hourly points of monitoring, it makes good economic sense to use an automated system. However, when data monitoring is less frequent, the payback period on the high capital cost of automated systems may be too long. Similarly, the data from an initial period of monitoring may demonstrate that the frequency can be reduced, based on data that demonstrate consistent achievement of performance/quality standards. Monitoring should be fine-tuned based on the data as the SPC implementation period progresses.

Structure the testing process to focus on each of the criteria identified above in an orderly sequence along the critical path. Not only is it important to understand what to test, but also how to test it. If a fault is discovered toward the end of the process, the product has to go through a long and expensive incubation process. This is made worse today because a smaller wafer has more capacity than it did in the past and therefore each batch represents a more expensive loss.

In most cases, SSOE, a top five design firm serving the semiconductor industry, sets up a fishbone (cause and effect) diagram or uses another statistical tool to help owners determine the types of issues that could be creating a problem. Troubleshooters can check off the potential causes one by one until they discover a demonstrable cause. When the test is well-structured, troubleshooting is a simple matter of using the tools and methods that process control experts routinely use to ferret out a problem.

Brainstorming can contribute to understanding all of the potential sources of process control problems. Some can be logically eliminated without testing until the field is narrowed, after which a series of tests can be implemented. For example, if there have been temperature variations, monitoring can be set up to determine the precise variation. If the problem is particle control, some of those particles can be captured and sent out for analysis. The chemical composition of the particles suggests the most probable sources.

All too often, it is easy to get distracted from the real possibilities along the critical path. For example, as wafers and dyes get smaller and
line widths get narrower, there are some sectors in the industry that must test for and control particle contamination even at the molecular level.

When troubleshooting, think outside the box or further down the line. Particles can be blown long distances from the source because of eddy currents and other phenomena that occur around devices. Vibration is another common source of product contamination, and its source is often unexpected and sometimes distant from the process line. One must examine all of the potential sources of these problems with an open mind, as in the following examples.

In one case, an inspection of the product showed particle contamination of some of the wafers. The suspected source was a contaminated tool; however, during troubleshooting, SSOE engineers discovered that an ionization emitter had failed, which created a charge on the wafers and emitted particles that were attaching to them. When the emitter rods were replaced, the problem was solved.

In another case, a particle source was reported to be a light station, but the actual source was found at a point 15 feet further down the bay. Engineers discovered that the installation of the laminar flow hoods in the space was not parallel, which affected airflow and caused the particles to travel down the side of the wall and fall out at the light station.

Use CFD modeling to help identify the source of a problem. A good Computational fluid dynamics (CFD) model creates a visual representation of what is occurring in the facility but which cannot be seen. It also lets engineers compare the results of various facility solutions during the design phase, thus avoiding additional construction costs to correct the problem. Unfortunately, many CFD models are generated based on incomplete data. As a result, the modeler may make assumptions that are not based on field-gathered information. For example, the design documents may not show a wall of conduit that acts as a barrier in a certain area of the facility, so the model shows free airflow through that area. To ensure that the data are complete, a modeler must ask the right questions—what objects may be obstructing flow or altering the pattern?

Take field measurements to complete the data required to develop a valid model. They also know that the entire facility affects the 10-square-foot process area they have been engaged to model; therefore, they factor in the appropriate variables in a wider area. They also interface with the many facility experts who have access to all of the information, including the facility’s air-balancing team, as well as engineers and other contractors with experience in the facility. In fact, these experts’ involvement in the process of CFD modeling and the completion of its information is a vital step in the integrity of that model.

It is now universally recognized that SPC can be applied to any process where output of a product meeting specifications—conforming product—can be measured. At its full potential, the process is an enormous aid to cost effectiveness because it can make as much conforming product as possible with a minimum amount of rework or scrap.

 

Dewayne Galyon is a technical specialist and associate at SSOE Group. With nearly 30 years of experience, Dewayne leads, performs, and participates in reviews of multi-discipline projects, critical investigation boards, operational readiness reviews, facility turnover and startup, environmental compliance assessments, and time and materials estimates.