The rocky road to 300mm automated factories
01/01/2003
We've all seen the commercial with the greasy mechanic warning: "You can pay me now, or you can pay me later......"
This is a lesson the semiconductor industry is learning the hard way. For years there has been talk about developing built-in metrology to monitor processes, either in situ or in line at the tool location. Some metrology companies hooked instruments to tools to show that it could be done. Process toolmakers offered to add metrology to their offerings, and found some interest. When they revealed the extra cost, though, the answer was almost always, "Forget it."
It was much cheaper to have process engineers develop recipes and then have fab operators implement and monitor them via test wafers than to actually build all the sensing, instrumentation, metrology, data collection, networking, and control and analysis software that would be needed to automate wafer processing. One test wafer out of 25 could be analyzed in the lab, and if any trouble showed up, the offending process could be shut down for troubleshooting and corrective action.
Now a few highly automated 300mm wafer-processing plants are on stream. The value of semi-processed wafers is climbing rapidly, from the BMW to the Porsche range and beyond. Fabs can no longer afford to toss out a 25-wafer lot because of misprocessing, whether from operator error or a glitch in the gas line or process chamber. If a key parameter is off target anywhere, fab operators need to know about it quickly and corrective action needs to be taken before another wafer is processed. With hundreds of tools in a large wafer plant, that is an awful lot of sensing, data collection, and analysis. Chipmakers are finding they are way behind in coming up with all the technology needed to automate their operations fully, ramp to yield quickly, and steadily improve yields.
Getting it right is critical. Bad runs due to human error or any other cause are unacceptable. Test wafers are impractical, except for process development. Chipmakers cannot afford to spend hours dissecting a wafer, followed by more hours trying to find where and why something went wrong. These factories need instant data that can be easily analyzed, and the data must be delivered over networks compatible with tool software and with management execution systems. To accomplish this, each major fab builder is developing its own unique fab-wide architecture, making industry-wide solutions elusive. Semi has set up task forces to tackle various parts of the problem, but the needs are racing far ahead of agreed-upon standards and solutions.
At a recent Sematech automatic equipment/process control (AEC/APC) conference, many of the world's leading chipmakers — Intel, IBM, TSMC, Texas Instruments, AMD, Infineon, Toshiba, Motorola, LSI Logic, and more — described work they were doing, and encouraged the industry to come up with new and improved technology, hardware and software, for their highly sophisticated new factories.
IBM described a detailed list of specifications that would allow a vendor to develop a product off-site that would plug-and-play when hooked to its flexible network architecture.
This is not the first industry to struggle toward automated factories. Some decades ago, GE built a nearly totally automated robotics plant in Virginia to show off its prowess. Everything was automated, whether it made sense or not. The result was a costly disaster. General Motors built a highly automated auto assembly plant in Hamtramack, Michigan. It was another colossal bust. GM later partnered with Toyota in the much more successful New United Motor Manufacturing plant in Fremont, California, with a better balance between automation and workers.
The semiconductor industry has enough smart people to design well structured, highly efficient, smoothly running 300mm wafer-processing plants. But if each is unique, there will be little incentive for developers to come up with the thousands of new, and often innovative, cost-effective solutions needed to knit them all together into seamless, reliable manufacturing systems. Already, small companies with good ideas must struggle to survive until the fabs are willing to buy their devices or equipment in volume. Without some uniformity and standards, everything will become a custom job, slowing progress and diminishing market incentives.
Never has the need for industry-wide collaboration on a global scale been more acute. If the semiconductor industry does get its act together, and homes in on optimal, well-coordinated solutions from process sensors, metrology and data collection, up to fab-wide control and e-manufacturing, the results may provide spin-offs throughout the industrial world. The world will be watching us.
Robert Haavind
Editor in Chief
Change is coming to SST
The technology needed for the industry to maintain its momentum is getting ever tougher as we reach the 130nm node and push toward 90nm and 65nm. You will find new departments in Solid State Technology, geared to helping the industry meet its goals. These include: Front-End-Of-the-Line (FEOL) and Interconnect (BEOL) each issue, and a quarterly column on Test-Assembly-Packaging (TAP), all critical areas for advancing technology. Since so many of our loyal readers have moved up into management, SST will include an occasional article on Fab Management. Also, at the back of each issue you'll find Perspectives, a new column of commentary by knowledgeable observers on current critical issues. Watch for them, and let us know what you think.