Silicon is one of the most abundant elements on earth, but that's scant comfort for the makers of silicon semiconductors. While some investors worry about a possible glut in the $56 billion market for dynamic random-access memories (DRAMs), experts at Dataquest Inc. and Rose Associates point to an issue that could have longer-term implications: a looming shortfall in silicon wafers that will crimp the ability of chipmakers to fully satisfy demand.
How severe the shortage will be and how long it will last are in dispute, but many believe that wafer supplies will get tight by late summer, then intensify in 1997 and perhaps 1998. Dataquest reckons the shortage could stretch into the next century, at least for the latest 200-millimeter size (chart).
There's no way to avert a short-term deficit, because wafer suppliers cannot expand production as fast as chipmakers can. From now until 2000, the worldwide chip industry will build 140 fabs--or fabrication plants--says Strategic Marketing Associates Inc., a market-research firm. While a new fab can go up in just 12 months, building a wafer-making plant takes half again as long. Some chipmakers might take a hint from Micron Technology Inc., which recently postponed the opening of its next fab, and delay cutting the ribbons on their own new plants. But that won't ease the supply problem, says Clark J. Fuhs, head of materials, forecasting, and silicon research at Dataquest. "Our forecast of tight silicon supplies already takes into account delays in some DRAM fabs planned for 1997 and 1998."
STRETCHED THIN. A further complication is that wafer makers have their own supply problems. The plants that supply them with raw material--polysilicon, an ultrapure form of sand--take the longest of all to build. "You need at least 24 months to bring a new polysilicon plant onstream," says Duane O. Townley, a vice-president at polysilicon-producer Hemlock Semiconductor Corp.
Polysilicon suppliers have fallen unusually far astern because of the dramatic growth in the chip market. The chip business has historically gone through cycles, with demand slackening every three years or so. That gives silicon production capacity time to catch up. But the chip industry hasn't had an off year since 1990. Instead, sales have exceeded the normal growth rate of 15%--topped off by a 44% explosion in 1995, to $155 billion. Yet the Semiconductor Industry Assn. was predicting as late as October, 1994, that chip output in 1995 would climb only 15%. As a result, says Daniel J. Rose, president of Rose Associates, a Los Altos (Calif.) consulting firm, "the whole food chain is stretched right to a thin hair."
The semiconductor industry can seem inscrutable even under the best circumstances. Chipmakers build the world's most expensive factories to turn out industry's cheapest products: transistors, the key components in all microchips. The cost of building wafer fabs doubles every three years--the tab for a world-class fab now stands at $1.25 billion and should hit $3 billion by 2000--yet the price of transistors drops 20% to 30% a year. The ever-cheaper chips continually create new markets that more than offset the smaller revenues from each chip.
ODD MATH. Lately, the strange arithmetic of chipmaking has caused more head-scratching than usual. Lingering concerns about a chip surplus intensified on Feb. 26, when Micron Technology, one of the top-10 DRAM makers, announced a delay of several months in the opening of its newest chipmaking factory in Utah. The next week, Fujitsu Ltd. delayed construction of its next fab, in northeastern England. On Mar. 6, Texas Instruments Inc. warned that industry sales in 1996 might not match forecasts. Why? All three pointed to the prices of DRAMs, which have fallen as much as 40% since December. DRAMs represent one-third of the total chip market.
The plunge in DRAM prices has had repercussions in the stock market. In January, Tokyo's Nomura Research Institute downgraded the stocks of many Japanese DRAM producers. That triggered a sell-off by Japanese speculators. The same thing happened on Wall Street--where the most extreme case was that of Micron. Its stock peaked at more than $94 last September, but has since plummeted to around $30.
Despite the indications of a silicon shortage, wafer producers will likely be cautious in adding capacity. They don't want to repeat the fiasco of a decade ago. In 1984, as chip sales zoomed 46%, wafer and polysilicon companies plowed millions into expansions, then watched in horror as chip sales plummeted in 1985. Ever since, there has been a glut that has forced suppliers to scrape by on margins of 2% to 6%, says Rose. As the overcapacity dwindled--from 112% in 1993 to 50% in 1994 and to 17% last year--profits finally picked up. By the end of 1995, "our margins had improved to 25%," says Chief Financial Officer James M. Stolze of MEMC Electronic Materials Inc., a leading wafer maker.
DRAM DEMAND. The current surplus of DRAM chips probably won't alleviate the wafer pinch, because the glut is temporary. Late last year, DRAM producers got caught by an unexpectedly quick technology shift from conventional, "fast page mode" (FPM) chips to a speedier type of DRAM circuitry dubbed "extended data out," or EDO. Demand for the latter was sparked by computers based on Intel Corp.'s fastest Pentium chips and rapidly took hold. The swing from FPM to EDO "happened almost overnight," says James J. McGregor, a senior analyst at chipwatcher In-Stat Inc. As a result, DRAM suppliers are scrambling to crank out more EDO chips--and steeply discounting FPM designs.
Once FPM inventories are gone, prices could stabilize or even increase before returning to their usual decline. But prices won't hold the line, as they did for three years before the blip created by the FPM glut. Prices defied the historical trend of steady declines only because supplies were tight--a consequence of stingy capital spending in the early 1990s. Now, DRAM makers such as Samsung, Hyundai, and NEC have brought many new fabs on line, with more to come.
That's good news for buyers and sellers of multimedia PCs. To handle more daunting sound and video effects, these computers typically come stuffed with up to four times as much memory as the PCs of just a couple years ago, when 4 megabytes of DRAM was considered adequate. Sales of multimedia PCs haven't lived up to expectations, in part because prices remain so high. The cost of 16 megabytes of DRAMs can represent a third of a PC's total price.
LOCKED IN. Lower DRAM prices should stimulate demand for memory chips and power-packed PCs in the months ahead. Indeed, sales of DRAMs are projected to outpace the rest of the chip industry in 1996. Last year, world semiconductor sales jumped an impressive 44%--but the DRAM business was even more spectacular, soaring 70% above 1994. For this year, the Semiconductor Industry Assn. puts DRAM growth at 40% and overall industry growth at 26%.
To cover their futures, DRAM producers and other big chipmakers are locking wafer suppliers into long-term contracts that guarantee deliveries for up to five years. Wafer makers, in turn, are striking similar deals with their polysilicon suppliers--or simply acquiring them. Because captive suppliers can't say no to their owners, the polysilicon crunch should ease by 1997, says Dataquest's Fuhs.
The wafer shortage could also ease by late 1997, if chipmakers decide to change their ways. Today, one of every eight silicon disks on the production line is a test wafer, used to verify that the "real" wafers are processed correctly. More automated production-control systems could slash this waste. Like a cobbler's shop, chip fabs often depend on reliable but old-fashioned equipment. While this is changing, Dataquest does not expect the adoption of better control systems to spread fast enough to prevent a tightening of wafer supplies in the decade's waning years. And if undercapacity increases by 2000 to the 25% shortfall predicted by Dataquest, some fabs may be locked up out of necessity, not choice.