We need better factories
06/01/2000
 |
Making complex integrated circuits is the most demanding manufacturing task the world has ever known (outside the field of biology!). Not only do chipmakers have to deal with immense complexity, but they have to do it using a constantly changing mix of materials, tools, and processes. Because of steadily shrinking pattern sizes, each step forward tends to be much harder than the ones before. We have been forced to go beyond optical microscopes to scanning electron microscopes and even to atomic force microscopes in order to "see" what we are doing. Patterns are being exposed reliably down to almost half the wavelength of the light used to print them, a feat thought impossible only a short time ago. Yet the relentless pace — doubling the number of devices in the same area every 18 months — goes on, in spite of the ever-more formidable barriers.
As this process continues, just doing the same things better works for many aspects. Occasionally, though, the industry runs into a step function: something just cannot be done the same way any longer. At the top of this list for a long time has been optical lithography. As a result, attention, resources, and money have flowed into next-generation lithography. But light has proven so resilient that so far it has done the job better and longer than anyone could have hoped. The focus has not been so strong on some other potential step functions that may prove just as important. One of these is operation of the chipmaking factory, or fabrication plant.
In recent times, recognizing that it might take more than shrinking circuit features to continue the pursuit of the performance/cost benefits of Moore's Law, chipmakers have been increasingly concerned about, first, yield, and then, tool productivity. Yields are now often well up in the 90% range, and far less time is needed to ramp to high yield. Currently the spotlight is turned on every aspect of a process tool's activity, well beyond just "down time." Even standard aspects, such as routine maintenance and set-up times, are being considered as nonproductive, and intense study is shortening or eliminating all periods when a tool is not actually carrying out a process step successfully.
These efforts have centered on isolated tools and processes - but what about the factory as a whole? Within the next few years, a major transition to 300mm wafers is planned. While the industry played with the idea of putting just 13 wafers into a cassette to avoid making them too heavy for humans, it appears instead that the 25-wafer pod, weighing so much that it must be moved by robots, will be standard. Already many fabs move cassettes between bays and stockers with automatic guided vehicles or along overhead rails. In the future, even movement into and out of individual cells and tools will be robotized.
Experience has shown that a totally automated factory can prove to be quite inflexible. Much work remains to be done if the 300mm wafer fabs now being planned are to be cost-effective and efficient. Strict adherence to all kinds of standards will be needed to get all the infrastructure and tools to work smoothly together. Computer control of the plant will become necessary, and software will have to be much more intelligent and compatible across all operations. This is a tall order for an industry where vendors often try to distinguish themselves by unique designs, and operate their equipment with proprietary software. If three companies work on a standards committee, three standards might emerge. That culture will have to change if future semiconductor fabs are to be better than today's, and that may be essential in order to stay along the Moore's Law track.
Automation of fabs offers real opportunities to make huge strides in improving many aspects of factory layout, design, and operation. The industry needs to work cooperatively to get the most out of the transition.
Then, somewhere out ahead, we will probably have to learn to build chips by studying biology!
Robert Haavind
Editor in Chief