By Otis Port Nothing sums up semiconductors better than Moore's Law -- the dictum that the computing power of chips will double every 18 months or so, at no increase in cost. Gordon E. Moore first laid down his law in 1965, eight years after he co-founded Fairchild Semiconductor and three years before he helped create Intel.
For four decades, advances in technology have steadily fulfilled his vision, and the chip industry has racked up an enviable growth record: an average of 16.5% a year, after smoothing out the sector's notorious boom-and-bust cycles.
IRONIC TIMING. Then came the Semiconductor Industry Assn.'s (SIA) annual "forecast" dinner on Nov. 6. With a twinge of paradox, the SIA presented its Lifetime Achievement Award to Moore -- and then indicated to small chipmakers that the glory days of Moore's Law have ended, at least for them. Nobody said this in so many words, but it's implicit in the SIA's latest outlook. Growth will slow, and in the future only a handful of big companies will be able to afford the hugely expensive equipment needed to make the high-performance chips.
At first glance, the association's forecast -- a consensus estimate by the 70 companies that produce 90% of the world's chips -- looks like more of the same: The industry is now emerging from its worst slump, and sales growth could hit 20% next year and again the year after. But the real news was the bombshell dropped by W.J. "Jerry" Sanders III, chairman of Advanced Micro Devices (AMD), who delivered the SIA forecast.
From now on, Sanders said, you can kiss that 16.5% annual growth figure goodbye. Instead, he predicted, "average annual growth rates in the range of 8% to 10% will be the norm." Increases of that magnitude are certainly nothing to sneeze at -- for many industries, 9% is only a dream. But it means the chip crowd is slashing its expectations nearly in half -- maybe more, considering that SIA forecasts have usually been a bit optimistic.
END OF AN ERA. The heady growth of yore can't continue, Sanders explained, because the industry has become too big. No new gizmos on the horizon could sustain the 16.5% surge in demand that was triggered by the PC, video games, and other consumer-electronics products -- not when the baseline point has climbed to $142 billion.
Sanders didn't soft-pedal the ramifications of this contention. "It's a sea change for our industry," he admitted. To survive, many chipmakers will be forced to hatch new strategies, such as specializing in niche markets. And for companies that can't do better than No. 3 in their chosen markets, Sanders had stark advice: "Pack it in."
Companies that heed his suggestion won't necessarily disappear. They can join the fabless crowd -- the 200-odd companies that develop chip designs, then farm out production to a contract manufacturer with a so-called wafer fab, or chip factory.
SHRUNKEN PROSPECTS. Actually, going fabless can boost profits. Many converts to the approach soon notch better margins than before, says Jodi Shelton, executive director of the Fabless Semiconductor Assn. For example, since Xicor closed its fab in 1998, gross margins have climbed steadily, from 14% to last quarter's 54%. But the $70 million company also pared its 1,200-person staff by 87%, and total revenues have dipped 34%. That will become a familiar pattern, predicts Thomas J. Engibous, CEO of Texas Instruments. Many chipmakers, he says, "will fade into a different business model [and end up] smaller companies."
The industry's shrunken prospects will also have a huge impact on equipment suppliers, says Carl R. Johnson, president of market researcher Infrastructure. "Consolidation is inevitable," he warns, because only the top seven or so chipmakers can afford to keep on buying the latest gear for turning out chips with ever-thinner circuit lines. The rest, says Johnson, may never go below today's cutting edge of 130 nanometers, or 1/800th the width of a human hair. "The smaller guys are going to have to learn to live with products that use existing process technology," he says.
Moore's Law has two facets: Using the thinner lines to stuff more power onto the same size chip, or using thinner lines to shrink the size of an existing chip. Since size determines 95%-plus of a chip's cost, shrinking the size by 50% cuts costs by that amount. The latter aspect is the key to chips being used in lower-cost gadgets. The former is the key to supercomputers on a chip, gigabit memory chips, and system-on-chip circuits.
FLATTENING OUT. What everyone will probably have to live with is the chip industry's cycle of ups and downs every few years -- although the magnitude of the swings could diminish. In the coming upturn, the SIA sees global sales cresting at a record of $206 billion in 2004, up from $142 billion this year. Then, after leveling off for a breather in 2005, the industry will be set for the next boom -- or perhaps another downturn.
Savvy investors will still have silicon waves to ride, but not the huge cresting ones of the last four decades. Port covers technology for BusinessWeek from New York