Chipmakers still seek an edge in fabrication technology

10/01/2002

Is all chip manufacturing going to be done in foundries a few years from now? Is there no longer any competitive advantage in chipmaking skills?

These are very important questions for our industry, particularly for integrated device manufacturers (IDMs) planning to build new 300mm fabs or upgrade existing plants. If it didn't make any difference where their wafers were actually processed, they could become design houses, like Cisco in networking, which contracts out all its manufacturing. Some observers, such as Bob Johnson of Gartner Dataquest, say that chipmaking has become essentially a commodity. Even the most advanced designs can now be manufactured in foundries, he suggests, so the margins on chipmaking will shrink and the real value will be in the designs or intellectual property (IP).

If foundries can match the advanced technology of even the best IDMs, why wouldn't it make more sense to have them make all chips? The foundry business model should do better at keeping costly equipment running because of diverse customer lists, while some IDM plants could become nearly idle during down cycles.

There seem to be at least three good reasons why IDMs want to keep their own wafer-processing plants. One reason is that they continue to see chipmaking skills as a potential differentiator. Many executives at IDMs, such as Intel, IBM, Texas Instruments, and STMicroelectronics, continue forward with plans for 300mm fabs, although they may assign some chipmaking to foundries. A commodity chip giant like Samsung in South Korea is unlikely to have much, if any, dependence on foundries.

Certainly IBM has gained advantage from its early commercial development of silicon-germanium, copper, silicon-on-insulator, and other technologies. This allows IBM to offer higher-performance, faster, and lower-power chips based on special fab capabilities. While IBM has operated a high-level, boutique type of foundry in Burlington, VT, it is now expanding its foundry services to design firms that want to take advantage of cutting-edge processing capabilities, especially at a new 300mm fab in Fishkill, NY.

Motorola is pushing its capabilities in putting compound devices right into a CMOS process flow. STMicroelectronics has developed a number of generations of BiCMOS capability that give it a big advantage in mixed signal devices. Intel plans to use strained silicon and other advances based on its processing skills to shrink devices and push its microprocessors ahead of the competition.

While foundries might have made good progress toward making devices in the 130nm generation, they had to develop processes appealing to a wide customer base, and some of the more challenging chips are proving difficult to make at high yields. Designers have to tailor their circuits to this fairly general-purpose process capability. That may be OK for many designs, but not for the key circuitry that gives an IDM its edge in the marketplace.

Another area of concern is that at times of peak demand, an IDM might not be able to get all the chips it needs. Having its own fab gives an IDM control not only of processes but also of what wafers get processed when. That control can be lost at a foundry when things get very busy.

A third hang-up involves IP, where so far there has been more talk than action. Putting several circuit designs together on the same chip can be a nightmare. Different EDA software may have been employed for various types of circuits, and test programs may have been devised for different testers. How do you verify and test such a chip? These incompatibilities have greatly limited the IP marketplace, and efforts by the industry to achieve broad compatibility have not been very successful. An IDM can set and enforce standards within its organization to enable convenient reuse. A foundry may have a library of circuits using similar software and geared to its processing capabilities, but that greatly limits the choices of the circuit designers. In many cases, that may be adequate. But it is unlikely to meet the demanding requirements of an IDM trying to gain competitive advantage.

The answer here, as is often the case, is a mixed bag. While the number of circuits treated as commodities and made by foundries may grow, the foundry share is likely to reach a maximum, perhaps at 50% or a little higher. Key circuits give a major IDM an edge, although as the cost of a new fab rises toward $3 billion, more chip companies will be sharing fab costs and risks through alliances and partnerships. Manufacturing of commodity DRAMs and core microprocessors is unlikely to be turned over to foundries. Some other key chips will continue to be made in-house, so dedicated wafer-processing plants with advanced capabilities will exist well into the future.

Robert Haavind,Editor in Chief