Investors in the semiconductor industry never learn, they just keep building wafer fabs. Indeed, they seem to have come up with their own variant of Parkinson's Law. Instead of work expanding to fill the time available for its completion, it's capital over-expanding to fill all available demand.
DRAM over-expansion already is a textbook case. The same thing happened on a smaller scale in graphics chips, SRAMs, and even parts of the ASIC business. Now it's the entire foundry industry that threatens to become one big glut, even though most foundries already are running well below capacity.
Nevertheless, everyone and his brother is rushing now to jump into the foundry business. Today's Common Wisdom holds that a paradigm shift is occurring in what's called "the semiconductor business model." That is, the foundries will crowd the integrated device manufacturers (IDMs) as the dominant type of chip producer.
Ever quick to jump on the latest chip vogue, capital is rushing to build new foundries or convert old fabs into foundries. Much of that frenetic activity is happening in Taiwan. Pacific Wire & Cable Co., which has never run a chip operation, has purchased sub-0.25-micron logic technology from Nan Ya Technology Corp. to use in building a major 8-inch wafer foundry.
Nan Ya itself is rushing to complete a new foundry that can turn out 30,000 wafers a month. And while the island's United Microelectronics Corp. (UMC) has shelved for now its 10-year building plan, it still will open the new foundry fabs that it already has under construction.
In Oregon, LSI Logic Inc. is opening a new 8-inch fab that will employ sub-quarter-micron processing to turn out "mega-quantities" of custom chips for customers. While this fab arguably may not be considered a foundry, LSI Logic still will compete head-on with other foundries in making leading-edge chips that either are jointly designed or designed by the customer.
This tidal wave of new foundry capacity will be augmented substantially by underutilized DRAM plants that are being converted to the foundry business in an attempt to improve their returns. This change in business plan is now underway at several Taiwanese and South Korean DRAM fabs. Acer Inc. is converting the former TI-Acer fab to a logic foundry, PowerChip Semiconductor and Mosel-Vitelic are switching their memory production line over to logic for foundry customers, and LG Semicon plans to shift some of its DRAM lines to foundry work.
UMC, which took over a DRAM fab last fall from Nippon Steel Corp., is changing it into a logic foundry. A state-of-the-art DRAM fab being closed down in North Tyneside, England, by Siemens AG's Semiconductor Group, may be sold to someone who wants to turn it into a foundry.
Threatening to turn all this into an even bigger glut is the accelerating effort by some of the big foundries to shrink their processes to 0.18-micron feature size as quickly as possible. That will end up having the same kind of impact on chip output as die shrinks did on DRAMs.
While there's no question now that demand for foundry production is exploding, capacity -- both new and converted -- could easily and quickly outpace it. Just as in the recent DRAM disaster, such overcapacity could easily result in a wave of price-slashing and red ink.
What all this means is that it is going to take more agility and prowess than ever to run a successful foundry -- even in the next market boom.