By Olga Kharif Sometimes, it really is better to be safe than sorry. Take computer chips. Last year, Nvidia (NVDA) and ATI Technologies (ATYT) -- Nos. 2 and 3, respectively, among graphic-chip suppliers worldwide -- took boldly different tacks with their new designs. ATI played it safe and released its new Radeon 9700 Pro chip -- an apple that didn't fall far from the tree. Its intricate components were located within 0.15 micron (one-thousandth the width of a human hair) from each other, no change from previous ATI chip specs. But Nvidia decided to get fancy. It designed its NV30 chip around 0.13-micron technology, which turned out to be far more difficult to make.
Result: Nvidia is paying the price for its risk. In its second quarter, ended July 27, the outfit blamed 0.13 micron production on lower-than-expected gross margins of 28%. Turns out contract manufacturer Taiwan Semiconductor (TSM) needs about 11 weeks to turn out a batch of NV30 chips, vs. six weeks for more traditional designs. That cost Nvidia plenty in late deliveries, with no commensurate improvement in sales or market penetration, says Michael McConnell, an analyst with Pacific Crest Securities. An Nvidia spokesperson says problems associated with the 0.13-micron process have been worked out. Yet, partly due to its production woes, Nvidia's stock has fallen 30% since June, to around $17.60 as of Sept. 3.
Meanwhile, ATI's gross margins stand at 32.9% -- five points above Nvidia's. "We made the right process choices," says Matt Skynner, ATI's director of marketing for what are known in the industry as "desktop discrete graphics." The stock, which has almost doubled since June, was trading at around $15 as of Sept. 3.
BEHIND THE CURVE. The same scenario has played itself out across the industry over the past year, with many players finding that it pays to stick with the tried and true. That's a 180-degree turn in thinking for high-end chipmakers. For years, outfits that didn't migrate -- and fast -- to the newest and greatest in chip-production technology were regarded as laggards. The accepted wisdom was that new manufacturing processes always offered higher yields and lower costs.
That remains the prevailing view on Wall Street. "It's very clear that investors, at this point, prefer companies that are investing in new technologies," says Merrill Lynch analyst Brett Hodess. "Otherwise, they think they won't be able to produce the right products." But profitability, growth, and high returns no longer seem to go hand in hand with high-risk investments in new manufacturing processes, if the the past year is any indication.
Inside the industry, fast technological migration is no longer regarded as closely tied to a success, says Ray Bingham, president and CEO of chip-design software powerhouse Cadence (CDN). For a lot of chipmakers -- especially those whose chips go into products with a lifespan of less than two years -- waiting until new manufacturing technologies prove viable is starting to look like a smart strategic and financial move. New processes no longer guarantee good yields and often require costly chip redesigns.
BIG BLUE'S BLUNDER. For example, making 0.13-micron chips requires the use of more expensive copper instead of the usual aluminum, explains Charles Byers, director of worldwide brand management for Taiwan Semiconductor (TSM), which will receive 20% of its revenues from 0.13 technology by yearend. New materials can lead to unforeseen glitches, and even the tiniest of chips -- 0.09 micron in size -- can overheat. Says Byers: "As you go deeper and deeper into the frontier, the trees get thicker."
Product yields for the latest generation have been worse than ever, at around 40%, vs. 85% to 90% for mature processes, estimates Len Jelinek, an analyst with chip consultancy iSuppli. And raising that number will take lots of time and money.
The risks are huge: IBM's (IBM) microelectronics business lost a larger-than-expected $111 million in the second quarter. John Joyce, IBM's chief financial officer, told investors in a July conference call that the loss was due in part to low production yields at a new chip plant. The plant stamps chips out of silicon-plate wafers that are 300 mm in diameter, vs. the standard 200 mm. Theoretically, the use of larger wafers should have increased chip output and reduced cost per unit. Now, IBM doesn't expect the division, which was profitable in 2002, to break even for 2003, Joyce said during the call.
WHO NEEDS THEM? Many smaller semiconductor companies, already weakened by the worst downturn in the industry's history, can ill afford those problems. So they're erring on the side of caution. Pixelworks, a Tualatin (Ore.)-based supplier of chips for televisions, now expects to migrate to next-generation technology in 2005 -- five years after its move to 0.18-micron production, says CEO Allen Alley. In the past, it shifted to new technologies every three years, he says.
With demand still light, that lag probably won't matter much. Pixelworks' customers aren't demanding smaller or better functioning chips, Alley says. The semiconductor industry is recovering slowly, and unit shipments should show only single-digit percentage increases for most components over the rest of the year, according to chip consultancy IC Insights. So new production technologies specifically designed to pump out more chips don't really provide an advantage.
Tinier chips also have lost some of their appeal in other sectors of the industry. Cell phones, for instance, probably can't get any smaller -- if they did, they would be hard to use, say ergonomics experts. The only reason to keep shrinking them is to stuff more functions into the same piece of silicon while reducing heat output. But chips built for DVD players, TVs, and other devices already perform at speeds and capacities that seems to suit the market, and might not need more features any time soon, says Jelinek.
KEEPING THE FAITH. Many chipmakers might be wise to spend their limited resources on improving their existing production processes instead. They also might want to focus on tighter chip integration, says Cadence's Bingham. Cadence plans to introduce a software platform allowing manufacturers to optimize chips to work together as part of a whole system in the fourth quarter, he says. Last May, it company acquired K2 Technologies, which makes software that helps chipmakers deal with problems like low yields in new manufacturing processes. Says Bingham: "Companies just don't have the wiggle room financially to have designs not work."
Not all producers will slow down in the race to develop quicker, faster, better chips. Intel (INTL), the king of computer microprocessors, needs to improve its chips' functionality to meet the ever-increasing demand for running smarter software and video on PCs. And Texas Instruments (TXN), the market leader in cell-phone chips, "is steadfast in its drive to deliver new-process technology every two years," says Peter Rickert, a process-development platform manager. TI, which has six fabs producing 0.13-micron chips, is developing 0.09-micron and 0.065-micron processes, he says.
For the rest of the industry, however, embracing change is quickly becoming a "question [of] when," says Taiwan Semiconductor's Byers. "Until 2001, companies would get there quickly regardless. Now, many have returned to business fundamentals." Perhaps that's not a bad thing. Kharif covers the chip industry for BusinessWeek Online from Portland, Ore.