The dream, reality, and future of intelligent process control
11/01/2000
Devon Kinkead, Extraction Systems Inc., Franklin, Massachusetts
Brad Van Eck, International Sematech, Austin, Texas
Advanced equipment and process control (AEC and APC) are keys to future success in wafer processing, but the industry is adopting the basics of AEC, APC, and so-called e-manufacturing at a glacial pace relative to that of our technology and financial demands. To whit, a recent gathering of process control professionals — who harvest, interpret, and build control algorithms using a fab's equipment and process data streams —expressed its dissatisfaction with the present level of data interpretation and machine-level algorithms.
The process control community is right; process equipment and metrology suppliers are not delivering the capabilities our customers need to construct advanced process control systems.
For example, does CD inspection equipment communicate with the track and scanner to trace the source of CD variation? Does the scanner communicate with the track to optimize wafer flow, reduce delay times, and increase productivity? Can the CD inspection equipment prompt the lithography cluster to look for the root cause of an increase in CD variation? Unfortunately, these capabilities do not exist in any sophisticated form in today's equipment.
 |
Production economics of 130nm-300mm technology have driven the metrology section of the International Technology Roadmap for Semiconductors (ITRS) to call for order-of-magnitude reductions in the time allowed to spot both trends and source problems. Not surprisingly, in situ real-time monitoring is seen as the solution to this pressing issue. However, it just produces data; the data must be harvested, interpreted, and yield an output that can help solve a problem. Data streams alone are not enough.
It is open to argument that any process must have a defined scope or it is ultimately unmanageable. Where does the responsibility of the metrology equipment end and that of the fab begin? On present course, will we be able to find a process problem in a lithography cell before another 140 300mm wafers go by? We don't think so and here is why.
Standard metrology logic inspects, filters, classifies, and analyzes; it does not typically take the next step to suggest a root-cause-focused action. For example, if a monitor detects an excursion, does it send an e-mail or page to appropriate engineers who, upon arrival, find that the monitor has already run root-cause-focused diagnostics? Has the monitor e-mailed the diagnostic data and suggested remedies to the equipment maintenance group?
This level of intelligence and communication is possible with today's technology and can be a highly valued benefit if it is built in from the inception of the equipment and process control architecture.
One obstacle is that equipment vendors must bring their products to market faster than ever. The priorities of any new equipment development process are: make it work; make it work most of the time; make it convenient to use; and make it cheap. Higher-level software intelligence probably falls under the third priority, "convenience," but, typically, competitive pressures to reduce cost push it lower in priority.
To develop the needed advanced process and equipment control algorithms, it is likely that a fab's process control people can only do about one-third of the job because the deep tool-level process knowledge required to contribute another third of the work resides with the equipment vendor who is still working on 98% uptime or a 6% factory cost mandate. The metrology or sensor supplier who understands measurement and its interfaces adds the remaining one-third. An effective advanced process control strategy that can be readily implemented into high-volume manufacturing needs all three partners to work together to automate the entire process. These partnerships have been historically difficult to arrange.
How do we bridge this chasm and what is the cost if we don't? Rapid diagnostics, notification, and the communication of suggested solutions must appear as requisite functions on the first draft of a new metrology tool's form, fit, and functional specifications. The specification development may go as follows:
- When defining what a product will and won't do, consider fault detection processes upstream and downstream of the monitored process.
- Discussions with target customers should include all relevant users of the tool's output data, including process owners, equipment, yield management, health and safety, IT, and facilities engineers.
- Write a paragraph on a day in the life of a customer to prove that the customer's problem is well understood. The primary goal is to remove obstacles that prevent attacking the right problem quickly. Removing these obstacles requires accurate and reliable monitoring technology and intelligent application-specific deployment. Attacking fast demands high levels of automation, communication, and intelligence from the monitor's software.
- A clear understanding of the customer's problem-solving process is critical. This step will define benefits that critical features of the tool will deliver.
- Honesty and reality will intrude into an otherwise visionary process. It is important to define the perimeters of the product's functionality, thereby identifying the needs that the product fails to satisfy.
- Rapid diagnostics, notification, and the communication of remedies and solutions should rise to the surface as highly valued customer benefits.
Until this type of dialogue happens early in the design process, it will be difficult to determine system performance patterns and, ultimately, to trace the source of problems in a matter of hours, as industry roadmaps require.
Looking forward, fab process engineers, who have customer clout, will have to lead the cooperative charge to integrate metrology and equipment suppliers to automate processes during process development and pilot production. Clarifications of the priority of advanced equipment and process control from the customer are realistically the only way this synthesis will happen. Customer-centric "e-manufacturing" programs have already begun to reduce the cost of test equipment repair by replacing the number of on-site service calls with remote technical support.
International Sematech is supporting these efforts by sponsoring an annual symposium focused on AEC and APC. As for the cost of lethargy: In the time it took you to read this article, that scanner just patterned another dozen 300mm wafers.
 |
Devon Kinkead is a founder and president and CEO of Extraction Systems, 10 Forge Pkwy., Franklin, MA 02038; ph 508/553-3900, fax 508/553-3901, e-mail [email protected].
 |
Brad Van Eck is the project manager of RTP tool development, sensor integration for improved OEE, and the AEC/APC XII symposium chair at International Sematech.