by James Montgomery, News Editor, Solid State Technology
Entering the early part of 2007, everyone knew that semiconductor capex was heavily slanted toward memory spending, and hoped that as memory firms started to pull back, other spending (e.g. IDMs and foundries) would pick up in 2H07. But a foundry spending rebound instead has been repeatedly pushed back. Now, after posting 4Q sales forecasts ranging from good to “lackluster,” all four top pure-play foundries — TSMC, UMC, SMIC, and Chartered — say they will actually spend less in 2008, anywhere from -10% to -25% or more.
The reasons given were clear — improve productivity in the lines already running, focus on migrating more mature processes (e.g. 0.13-micron) down to 65nm, and get ASPs back up and improve profitability — send a great message to the investor community. But the real reason could be a much bigger shift in the leading-edge foundry business model.
On the sidelines till 45nm?
Based on the foundry’s explanations during 3Q conference calls, it’s clear that their 65nm business isn’t gaining traction as had been hoped, outside of a few core high-volume customers. Foundries may be hanging onto equipment they’ve bought in the past few months, and readying it to be used for 2008 production runs instead, notes Gartner analyst Kay-Yang Tan, in a research note. He adds that the firms are anticipating slower momentum in 2009, so a 2H08 rebound in capex is unlikely.
At 45nm, though, the foundries are expecting broader adoption. Compared with 90nm and 65nm, “we are engaged in with, I would say, significantly more customers at the 45 nanometer technology,” said TSMC CEO Rick Tsai, during his company’s 3Q07 conference call with analysts. With chip designs for leading-edge processes becoming increasingly complex and costly (e.g. DFM and complex lithography steps/tools), perhaps the majority of foundry customers are sitting on the sidelines, waiting for 65nm tech to mature to more affordability, and 45nm process manufacturing to mature, where there will be a better balance between performance benefits and costs.
Fundamental shift in foundries’ biz model
The more likely explanation, according to analysts, is that we’re witnessing is a fundamental shift in the business model for foundries offering leading-edge manufacturing. Except memory companies and leading-edge logic firms like Intel, IC makers “are not moving aggressively to leading edge technology as they did in the past,” noted Bob Johnson, research VP for Gartner Dataquest. “The rush to the leading edge is not as pronounced as it was a few years back.”
Johnson noted that foundries, especially ones with leading-edge capacity, once rushed to develop leading-edge capacity believing customer demand would flock to them once it was in place. Now, pure economics has taken over, and these foundries are realizing that for developing leading-edge processes, they only need to work with third-party EDA firms for design packages/rules to lay out; put enough tools in place for early prototyping for their best customers with high-volume complex chips (e.g. nVidia); and invest in more volume capacity as designs/commitments come in.
“The business model will shift from, ‘If we have it they will come,” to ‘We’ll have it when they arrive,'” Johnson quipped.
Foundry execs’ comments also seem to support this explanation. TSMC’s plans to reduce capex as a % of sales to perhaps <20% in 2008, from historically ~25% of sales, "is definitely not a one-off phenomenon," noted Tsai, in his conference call's Q&A session. "We have seen most part of this year where supply in the leading edge have outstripped demand in a very significant way," added Chartered president/CEO Chia Song Hwee, in his company's 3Q conference call, adding that his company's goal now is to "make sure that the demand and supply are in moderation."
Bill McClean, president of IC Insights, also agrees that this is the most likely scenario, where the slow 65nm adoption “will see a big pickup” next year “and build up as demand warrants.” At this point he sees a splitting market of fabless customers: those for whom 0.13-micron or 90nm is just fine, with no need to look at 65nm “for a while” — and another group of customers that really needs the benefits of leading-edge technology and is already clamoring for 45nm. As a result, “the middle node [currently 65nm] gets ignored for a little while until other group catches up,” he notes. He expects this trend to continue to “move down the line” — 90nm customers eventually will move on to 65nm, while the other group shifts to 32nm and beyond (at which point in-between 45nm business stalls), etc.
This mentality of not overspending, trying to match expansion to demand, is essentially a “build-to-order” strategy, McClean noted, a model that’s worked well in the automotive industry — “I order a black Cadillac, it’s built that day on the line, there’s no inventory” — but not in the IC industry, which relies upon a number of immutable processes, e.g. process lag, wafer starts, getting processes in place, etc.
Another possible endgame to this business model should be discussed, that the customers who are waiting on the sidelines for better cost/functionality benefits to catch up with their needs, may never get into the game at all. “The cost of implementing a design at that level [65nm and below] is becoming prohibitive,” Johnson said, noting that DFM and lithography issues make it far more complex and costly than ever before to get a set of masks that’ll print a chip design, and then one actually has to get the design to work. For companies needing chips for consumer products that have relatively short lifecycles, these costs are just not worth it. “Eventually what will happen is, some of these firms […] can get integration benefits with multiple chips in a package vs. system-on-chip,” he explained, and that means “fewer designs going to the leading edge.”
The dangers of upside risk
There is risk in reducing capex requirements for the next year, though, if IC demand ends up being better than anticipated. Chipmakers could quickly find themselves coming up short in capacity in 1H08, forcing them into reactionary capacity additions in 2H08 that could stretch into 2009.
Among a number of factors McClean cites that could cause 2008 to be rosier:
– A big year in PCs. Finally, the corporate adoption of Vista picks up steam. With DRAM prices consistently <50% what they were a year ago, PC makers have no problem dropping 2GB of memory into their systems.
– Talking telecom. The market for cell phones could get even hotter, with new applications for video, mobile TV, and things like the just-tipped “Google Phone” platform, to be licensed for handsets produced by Motorola and Samsung. “If the high-end cell phone and communications business really picks up, then we’ll see a different story” in 2008, McClean said.
– Notebook surge. Notebook unit sales are poised to follow 2007’s 40% unit sales growth with a similar performance in 2008, according to one analyst, which would be good news for some big fabless companies (e.g. Broadcom, nVidia), and foundries still get a pretty big portion of sales from the computing sector.
– Consumers on the go. Don’t forget about still-popular consumer electronics like iPods and other music players, and PDAs.
– The big macroeconomic picture. Worldwide GDP is expected to be generally flat in 2008 (3.7%), in line with the 20-year average, with the big four (US, Europe, China, Japan) seen maybe slightly slower individually. The picture could get rosier if any of them do better, or if other growth regions (e.g., India, Latin America) outperform expectations.
The art of the head-fake
Though the foundries seem to be in lockstep to reduce 2008 capex, it’s not unthinkable that one of them might be trying to catch the others off-guard by playing coy, then in a cunning strategic move reverse course and in early 2008 commit more investments in a bid to gain market share (betting on the aforementioned rosier demand). “If I was TSMC that’s what I would do,” McClean said. “I’d enjoy the other three toning it down, then step on the gas further. It’s been proven over the past 30 years that companies that put in capacity gain share.”
If anyone is looking to break ranks and bump up spending in 2008 it would likely be TSMC, taking the opportunity to “become even more ‘Intel’ foundry than they are — do what they want to do, spend and spend, and gain share,” McClean said. And because TSMC is bigger than its rivals by multiple factors, there wouldn’t be much that UMC, SMIC, and Chartered could do to respond, McClean explained. UMC’s utilization rates are >90% but the business is still pretty volatile. SMIC is “struggling financially” and is looking to deemphasize the low-margin DRAM business it had taken on. And Chartered the smallest of the group and trailing the others (45nm is not expected until at least mid-2009), would face an uphill battle convincing investors that there’s pent-up customer demand just waiting for Chartered to put in more leading-edge capacity, he noted.
Nonetheless, McClean points out that in recent years TSMC has generally kept to its word with budget forecasts, and all the foundries are keen to reverse the trends of falling revenues/wafer, or at least slow down the declines. He points out that there have been other cases in the IC industry where groups have tried to collectively practice such discipline, but there’s always been a maverick to mess things up — most notoriously in the memory sector. “The DRAM guys sit out every couple of years, but suddenly Samsung throws more investments into capacity, and then everyone else has to ante up and compete with them,” McClean said. If the big 4 foundries can stay on the same page and ease capex, “if nobody breaks ranks it can hold,” he pointed out. “It can be more profitable for all of them.” — J.M.