by James Montgomery, News Editor
A recent report from Gartner suggested that the amount of capacity that has been announced to be online between 2007-2008 will fall well short of what the industry will need to have ready for production — by as much as a whopping 700,000 wafers/month capacity.
Bob Johnson, research VP for Gartner Dataquest, has raised his expectation for how many brand-new 300mm fabs will be announced for initial production in 2007-2008, up to 16 from 13 in previous projections. But that’s not enough — he also suggests that projected semiconductor demand will require 21 new 300mm fabs in 2007, and 18 in 2008.
“Approximately nine additional new fab projects for 2007 and 14 for 2008 are needed to provide sufficient capacity to meet projected demand,” bringing to 39 the total number of new fab projects ramping over the next two years, Johnson wrote. Assuming 30,000 wafers/month capacity, that’s roughly 690,000 wafers/month of brand-new capacity that will need to be ramped up in less than 18 months.
That sounds like a lot — but is it really time to worry about not having enough capacity? Or, put another way, is this amount of as-yet unplanned capacity unusually more than what the industry is used to for any given year?
“This is the normal flow of fab announcements — no more, no less than standard,” Johnson told WaferNews. He pointed out that in April 2004, Gartner similarly noted that there would need to be 8-10 new fab announcements made over the next few months to meet projected demand — and by the firm’s next update three months later, seven new fabs were planned, along with other major expansions.
Gartner’s latest projections for new production capacity [see chart] suggest capacity will increase about 9%-11% a year, roughly in step with projected semiconductor industry growth, barring major shifts in revenue per sq. inch of silicon, he says.
Generally, leadtimes for fabs are 12-18 months out, and anything announced before that is risky, Johnson says, partly because of regulatory and financial considerations. If someone announces a new fab project to come online in 2009, but then six months later pushes those plans back to 2010, “they’ll be brutalized” by investors, he says. Only a few chipmakers have the resources and long-term planning abilities to be able to confidently project a complex multibillion project like a 300mm fab, Johnson adds. Intel, for example, is well along the path to 45nm manufacturing, so it has started building in Israel and elsewhere.
Also, bringing capacity online is not a one-time event — most facilities are ramped up over months or even years. “There’s a difference between new capacity, and coming online,” he said, adding that “it takes a number of years for capacity to come online, [and] only a few people can afford to both plan and build fabs in that amount of time.” In fact, the amount of new production capacity coming online doesn’t equate to the number of new fabs — e.g., some of the projected overall capacity in 2006 has been slowly ramping since 2004. (Soitec’s new 350 million euro, 300mm Fab 3 in Singapore, for example, is scheduled to start production in mid-2008, but won’t reach full capacity of 1 million wafers/year until 2010.)
Moreover, Johnson points out that an extra 10 or so fabs each year really isn’t a huge number anymore. “We can’t think of this industry in terms of 1985 capacity numbers,” he says, noting that by the end of this year there will be some 67 300mm fabs in operation worldwide.
Recent concerns over slowing investments from chipmakers, namely foundries and memory makers, may suggest a pressing need for less capacity, not more. Johnson notes that memory firms’ capacity investments will slow down significantly at some point (they currently represent 45% of total semiconductor industry capex, vs. just 22% of revenues), but says an anticipated strong year for memory in 2008 will spur investments for capacity to be ready in 2007. A rise in forecast capacity will be matched by higher semiconductor production, maintaining a balance through 2007, and a rise in 2008 matching anticipated semiconductor market growth, according to his report.
Ultimately, Johnson says the current gap between currently announced capacity over the next two years and what is forecast to be needed is “the normal flow for fab announcements,” and he is confident there will be six to seven new fab projects announced in the next six months. (In fact, just days ago, reports surfaced that Promos is set to spend $5 billion on a pair of 300mm fabs in Taiwan.) “If we believed the sales guys, then business would fall off a cliff 9-12 months out,” Johnson says, explaining that chipmakers may be planning 1-2+ years out for a new line requiring implanters and etchers, but those plans are well in advance of the purchasing process, where leadtimes are typically six to nine months. “So, take it on faith that the industry is going to keep going.” — J.M.