e-Manufacturing: Still a long way to go
11/01/2003
There is a lot of talk these days about building lights-out semiconductor factories with total factory-wide control of scheduling and processing. Fab operators would be there only to monitor operations and perform maintenance. All process recipes would be automatically adjusted based on feedback and feedforward signals from built-in sensors and metrology. At the end of the line, device parameters would uniformly meet tough specs from lot to lot, wafer to wafer, and within wafers. Yields would be in the high 90s, boosted by rapid fault detection, classification, and analysis, allowing quick response and correction to changes in chamber characteristics, nonuniform plasmas, pressure variations, or any other root causes. This vision of chip factories run by computers, networks, and sophisticated software is described by the term e-Manufacturing.
How far is the industry from reaching this utopian status? Advanced 300mm fabs are already highly automated, and some of the pieces for e-Manufacturing are in place. But, as a number of presentations at the recent Sematech AEC/APC Symposium (Colorado Springs, CO, Sept. 15–17) indicated, there is still a very long way to go.
While some key standards exist, Semi has a host of new standards moving toward approval, especially for connectivity and software. Process tools with standard software interfaces geared to the middleware-based architecture of the new 300mm fabs allow functional communication between the fab's management execution system (MES) and each tool's advanced process control (APC)/sensors, according to Gary Behm, who described the intricate workings of IBM's new 300mm fab in East Fishkill. Open architecture will enable the fab to use best-of-breed process tools on the line, while providing plug-and-play flexibility, particularly in foundry operations, according to Mark Liu, VP operations, TSMC. Wider use of standards is critical to cutting costs for high mix chipmaking in foundries, he added. While APC is being applied widely throughout the new 300mm fabs, much more integrated metrology must be added to monitor processing and apply corrections.
As automated processing proliferates in fabs, it is vital that users get good data from their tools. In the past, a human operator could check outputs and take proper action, but in an automated system, the data filtering is done by computers, which are not as adaptable, pointed out James Moyne of Brooks Automation. Tests on data quality from 300mm process tools revealed some serious problems. Aside from poor documentation, outputs did not conform to even present SECS/GEM standard documents. For example, a 300mm etcher gave all sensor outputs in ASCII code with no context, requiring a translation step to determine what was being measured. The tool showed values of 2 when shut off, indicating an unknown offset. A CMP tool did not expose sensor values through the SECS/GEM port at all for end point detection. A standard request for the state of all internal units while running wafers got back all "zero" responses. Each tool tested had different areas of noncompliance, according to Moyne.
The industry has not even agreed on definitions for time stamping and other basic parameters, so a data quality task force co-chaired by Moyne and John Pace of SI Automation, plus Brad van Eck of International Sematech, is pushing to address such problems quickly.
Harvey Wohlwend explained how International Sematech's goal in the e-Manufacturing program is to accelerate productivity improvements using data-driven, real-time, automated decision-making across all fab processes (see SST, Oct., p. 42, for a more complete discussion). Industry-wide adoption of a set of common guidelines, user requirements, and standard interfaces is needed for all components of fab data systems in order to make the vision of e-Manufacturing a reality.
Moyne pointed out that other industries, such as petrochemicals and automakers, are already far ahead of the semiconductor industry in automating their factories. TSMC's Liu said that his company simply would not buy tools that did not comply with agreed-on standards and open-architecture protocols. Speakers from IBM and Intel echoed this message. With that kind of clout behind the push for e-Manufacturing, hiding process data from users and using proprietary software and interfaces, now a common approach in tool design, may be like fighting the tide rather than riding the wave. It's time to get with the program.
 |
Robert Haavind
Editor in Chief