Foundries Flex with Customers Via Reserve Capacity
By Ann Steffora Mutschler
Marking a shift in the foundry model, TSMC CEO Morris Chang said during the recent TSMC technology symposium that TSMC is the most likely foundry to have reserve capacity to respond to unexpected surges in the market.
At first glance, it seemed like an innocuous comment, leading to speculation as to the true meaning.
“It was only just recently that TSMC has changed their strategy of walking a very close line on 100% capacity utilization,” observed Joanne Itow, managing director of manufacturing research at Semico Research Corp. “To use the term ‘reserve capacity’ — peak capacity and then extra on top of it — is a different way of looking at things right now.”
Today’s markets are much harder to plan for than in the past. “Consumers are fickle,” she said, pointing to the tablet as one example. One result of the March 11 Japan quake could be pent-up demand, and “with the foundry market serving so many more customers, that impact is going to be felt with more exaggeration,” she explained.
Itow also pointed out that TSMC has been very effective in terms of planning their capacity with a huge team of people doing market analysis, as well as staying in very close contact with customers. “They listen to their customers, both from the perspective of what their customers are saying in terms of what they think they need, and what TSMC thinks they need. Everyone expects to be wildly successful in the market, but everybody can’t be. And because TSMC has such a broad view of the market I think they can make some pretty good assessments as to how much capacity they really need to add.”
Itow said there’s a different planning strategy with regard to advanced technologies, versus some of the more mature capacity. “Moving forward from here, we’re going to see some very healthy capacity utilization rates for the advanced technology nodes. And all the foundries are continually looking at ways to keep their mature capacity filled, so we’re hearing some emphasis on MEMS, although TSMC didn’t seem to put as much emphasis on that as I hear from Global Foundries. Global Foundries seems to be trying to be really aggressive in that market.”
Could this also be a way of smoothing over an overcapacity situation? “Of course,” said Dean Freeman, research vice president at Gartner. “How else do you justify the capex spending? When you are looking at costs roughly in the area of roughly $100 million per thousand wafer starts per month for leading edge, if you are putting in 10,000 wafer starts a month of reserve capacity just so that when you have market swings, or you have crises like this in Japan, you can respond… you’re basically taking $1 billion and saying, ‘Alright, we’re going to have this reserve and we’re O.K. with it.’”
But, given that TSMC has historically run at very tight capacity, the reserve does give a little bit more of a buffer to absorb new customers while they are adding capacity, he said.
Freeman believes the whole foundry industry needs to add this capacity so they can show their clients that they’ve got the capability. But Freeman added that he wonders as clients keep rolling forward, do the foundries keep this reserve as a kind of buffer as the industry continues to go down technology nodes? “Will we always have this 10 to 30 thousand wafers per month (roughly about a six month excess capacity) buffer that allows them to absorb these clients changing their minds, allows them to when things get really tight to say, ‘Let’s open up this line when we add extra capacity in order to meet the demand that we have out there?’”
Overall, he views this strategy change as a good thing for the capital equipment industry. “If we have six-month overcapacity and there isn’t any equipment being ordered that’s $3 billion. I think once you build this in and if you keep going forward, you don’t have as big a capex swing as maybe you would have had historically.”
Upon further investigation, Chang’s statement was in fact representative of the entire foundry industry changing strategy to accommodate leading-edge customers. Putting to rest the speculation, an anonymous source close to the foundries confirmed the reserve capacity scenario is indeed “a fundamental change in the business strategy and it’s not just TSMC but other foundries as well. It has to do with the shift of the product mix that they are manufacturing, more towards a microprocessor, more towards a higher end chipset.”
This strategy also reflects the fabless companies like Qualcomm and Broadcom dictating more of the terms of not only how the wafer builds are going to go, but details on packaging as well. As such, in order to be competitive in this new landscape, there’s a change in business strategy on the part of the foundries and most famously TSMC. “They are deliberating running at a lower utilization rate. Where they once would be having a strategy of trying to run at close to 100% utilization of the line, now they are aiming deliberating at more like 85 to 90% of the line. And it really varies for the leading-edge capacity where most of the shots would be called, so to speak, that they would run at a much lower utilization rate to accommodate those high maintenance customers,” the source continued.
The foundries want to be front and center with those customers to be able to accommodate their needs and there is a competitive advantage to being able to have a bit of extra capacity as compared to what it’s been historically.
“It’s an interesting landscape where we are seeing the foundry model really having a new wave to it: it’s a resurgence of the foundry model and it has to do with a much richer chipset coming their way. If you back up a little bit more to the end markets and think about why that is, it has to do with more of a mobile computing architecture where you have your smartphone and your tablet and its talking into a datacenter – all of those things require more ARM processors. The ARM processor phenomenon has really changed the foundry opportunity,” the source concluded.
Tags: capacity, foundry, GlobalFoundries, TSMC