Computational fluid dynamics predict and optimize airflow
By Hank Hogan
San Jose, CA — How do you control particulate and molecular contaminants in a cleanroom? To paraphrase a song, the answer is blowing in the wind. The movement of contaminants follows the flow of air. Those same air currents also influence temperature distribution, smoke movement, operator safety, energy costs and air quality.
The challenge is to predict and optimize airflows. Doing so not only saves time and money in a new facility, but it also is beneficial when rearranging equipment in an older cleanroom. Ideally this optimization should be done as early in the design stage of a project as possible. In the past, such predictions were based on rules of thumb or scale models.
Today with the advent of more powerful computing platforms and software tools, there`s another emerging approach: computational fluid dynamics. That was evident at a recent airflow modeling seminar hosted by Flomerics Ltd. (San Jose, CA).
“We consider airflow modeling essential,” says Douglas Cansdale, a cleanroom design specialist with Industrial Design Corp. (IDC/Portland, OR), during his presentation. IDC designs and constructs cleanrooms around the world.
Some of the applications covered in the seminar included semiconductor cleanrooms, tool minienvironments, antiseptic fill rooms and medical facilities. In the 40-plus person audience were representatives of the semiconductor, pharmaceutical and disk drive industries. There were also interested engineering consultants and researchers from a national laboratory.
One presentation that showed the power of computational fluid dynamics concerned the placement of smoke detectors. A large semiconductor manufacturer initially put smoke detectors directly above equipment in a chase. During live testing it was discovered that the smoke detectors didn`t work. That is, a fire could be burning in the chase and the smoke detectors would not sound an alarm. Moving the detectors away from the equipment and close to a return air inlet fixed the problem. It was clear from animated computational fluid dynamics simulations why this happened. The return vent was sucking in so much air that just above the equipment in the chase was a dead zone. As a result little smoke actually made it up to the region. In contrast, smoke concentrated at the return air vent.
Understanding why something happens is useful, but more vital is the ability to predict.
How accurate are computational fluid dynamics simulations?
In computational fluid dynamics a room is divided into smaller volumes, and the Navier-Stokes partial differential equations that govern fluid flow are solved iteratively for each region until the cumulative error drops below a threshold. There are both generic computational fluid dynamics software packages and ones tuned to a particular type of application, such as building ventilation. The movement of air, particles, contaminants or heat can be traced.
The typical cleanroom consists of multiple air inlets and outlets, along with scattered machinery and other barriers. All of these impact airflow. Heat sources also influence air movement. According to the seminar`s panelists, the results from properly optimized computational fluid dynamics software can be quite good if the boundary conditions are known. This means the simulations have to correctly place equipment and furniture, along with accurately including heat sources, air inlets and outlets. Without attention to such boundary conditions, the simulation may be nowhere close to reality.
For small simulations running on fairly powerful workstations, the time to reach convergence for a ventilation-tuned computational fluid dynamics package can run from a half hour to four hours. Larger or more detailed simulations require overnight runs. In all cases, the accuracy of the simulation depends on how well the virtual model correlates to the real one. However, when there`s a difference between measurements and simulations, it`s not always the simulation that`s wrong.
“Good engineers do not blindly accept measurements or simulations,” warned David Tatchell, chief executive at Flomerics (Surrey, England), during his overview of the field`s history. Tatchell has been involved in computational fluid dynamics since the late 1960s.
One of the key advantages that computational fluid dynamics has over more traditional methods is that visualization tools can be applied to the results. For instance, standard Food and Drug Administration (FDA) practice requires a smoke test as part of the proof of proper ventilation in some pharmaceutical applications. A video clip from Chiron Corp. (Emeryville, CA) showed such a smoke test. The same test was simulated with computational fluid dynamics software and then put through a visualization package. The visualization technique released massless particles into the simulated air stream. Different colors indicated different particle velocity vectors. Not only did the simulations agree quite well with the actual results, but tracking the air movement was easier because different colors were used.
It may be awhile before smoke tests are dispensed with, if ever, but both the seminar panelists and audience felt that the FDA may someday opt for simulation enhanced by visualization.
As for the future, a pressing issue raised by David Hope, microcontamination manager for Intel`s Advanced Fab Design (Hillsboro, OR), concerns the upcoming generation of 300-mm fabs. Today`s typical semiconductor cleanroom design consists of multiple floors, a gowning room and distributed fans. While Hope estimates that only 30 percent of the cost of a new facility will be in the building and cleanroom, saving money by changing factory design is still important. Intel`s practice is to replicate successful factories/processes. Hence any cost savings, or penalty, will likewise be replicated.
Reducing the number of floors, moving the gowning area out of the factory, and other changes may save money. This can`t be done blindly. It can be effective to build scale models and mockups, but it can also be very expensive. That`s one of the reasons why companies such as IDC are making use of computational fluid dynamics.
However, no one is yet ready to depend solely on virtual airflows. The approach followed by Huntair Inc. (Tigard, OR) seems typical. The company designs, manufactures and installs cleanroom air handling technology. Bruce MacGibbon, director of research and development for Huntair, doesn`t try to predict accurate absolute numbers for various parameters using computational fluid dynamics. Rather computational fluid dynamics is used to assess various designs and select the most promising.
As MacGibbon remarked, “we use computational fluid dynamics as a time saver on the initial design.”
Hank Hogan is a freelance writer based in Austin, TX. He has written for New Scientist, High Technology Careers, Electronic Components, and Multichannel News International. He was previously a semiconductor process engineer.