by Pete Singer, Editor-in-Chief, Solid State Technology
Dan Hutcheson, president of VLSI Research, gave a wide-ranging talk on the economics of the 450mm wafer transition at The ConFab yesterday. He discussed lessons learned from the mishandling of the 300mm transition, why chipmakers see a need for 450mm, why equipment suppliers are reluctant to support it and why it’s nearly impossible to predict the ROI of 450mm. He then proposed what may be a viable model for 450mm development, similar to the foundry model implemented by TSMC.
Mistakes in 300mm development were costly
Hutcheson lead off his talk noting that the 450mm transition has become a “very emotional” topic for our industry. “We know that historically all the wafer size transfers are risky for the leaders. If we go back to the 100mm and 150mm transition, IBM and Intel lead that in the mid-80s, and they really paid for it. They wound up having a very difficult time being competitive. IBM had the same problem when they led the charge at 200mm in the early ’90s,” Hutcheson said. “For 200mm and 300mm, it was the equipment suppliers that largely had the burden and they wound up paying for that very dearly and a lot of them are still struggling to pay off the investment they made in 300mm.”
One of the mistakes made in the 300mm transition was that the development had to be done twice. “We missed the planned launch date of 1997 and we had to scale back on spending. We didn’t make 0.25μm so all the gear had to start over again and be resized. We actually designed two 300mm generations,” Hutcheson said.
“More importantly, Hutcheson said, is what happened to equipment demand. “Back between 1995 and 2000, what we saw invariably was between 800-1200 units shipping depending on where we were in the cycle. Post-300mm, that was cut in half to about 500 units. Everybody wonders why steppers cost twice as much — it’s because they only ship half as much today as they used to.” Thus, the only way to increase revenues is to raise prices, he said. “The real issue with 450mm is that we potentially run into the same thing.”
Another problem with the 300mm transition is that it didn’t really return higher revenue growth for the equipment suppliers. “The chip industry has been growing nicely since 2000,” said Hutcheson. “In fact it’s been in its longest upturn in its history. The equipment industry has been essentially flat. It’s still struggling to get past this 2000 peak of $50B. This is why the equipment industry really sees 450mm as being a very painful transition, because they don’t want to go through that again. The bottom line is they have to report to the shareholders and their shareholders are not really prepared to fund 450mm development this time around. There’s just no financial case for the semiconductor equipment suppliers to do this.”
Hutcheson said that economics don’t really matter in the 450mm argument because it’s simply impossible to predict return-on-investment (ROI). “Semiconductor companies are going to need it sometime,” he said. “The supply side feasibility says 2020+, the demand side says 2014. I don’t think there’s any possible way to forecast the return-on-investment.” Historically, he said, it comes down to “a matter of faith. Why does this magical 30% number come up? These guys say we’ve done all this modeling and we show we can hit 30%. It’s because that’s what number we started with. It’s the same number we had when we started the 300mm program, it’s the same number for 200mm and that’s historically what we’ve gotten,” he said.
An example given was that of a factory with 500 tools. “What you really worry about is aggregation error. You have 500 tools and you have 10 first-order effects in measuring the cost of every one of those tools in your factory. That means you have 5000 error sources. If I have a 10% error on every variable, every tool, that gives me a 500% error potential in trying to forecast this transition by cost/benefit analysis,” Hutcheson said. “When we start to factor in the errors from materials, the number of suppliers in the industry, the number of process steps, you need accuracy levels in the parts-per-billion to avoid aggregation error problems, so you really can’t predict this.”
One actual example illustrating how assumptions and reality diverge is that of that wafer steppers. “We had 100 wafers/hr on 200mm litho tools and the argument was that they were going to go to down to 50-60wph with 300mm. Where are they today? They’re 150 wph — so that’s a 300% error on one set of tools. This is why you can’t predict this,” said Hutcheson.
Cycle time is another example. “We went from 8 to 12 week cycle times on 200mm. Today we’re in the 4-8 week range. Every variable is unproven. So when you try to predict this stuff you have so many sources of error, you just can’t beat it and that’s why it’s basically a matter of faith,” said Hutcheson. “There’s just no way to predict development cost. And there’s no reliable way to forecast the ROI.”
“450mm really isn’t going to happen until we develop a new development model because we know that the one company funding it doesn’t work,” Hutcheson said. “We also know that the consortia, equipment industry funding it doesn’t work. We also know the equipment industry is cutting back R&D expenditures as a percent of sales. They’ve been doing it for several years now. So the equipment industry is not going to fund this cycle.”
On the other hand, if chipmakers are willing to truly invest in equipment development, equipment suppliers will have little choice but to comply. “The real thing here is that the chipmakers are going to have to walk the talk,” Hutcheson said. “They’re going to have to start spending money. They’re going to have to start placing orders for development tools.” And this is already starting to happen, he noted, pointing to the recent announcement from Intel, TSMC, and Samsung that they’ve reached a letter of agreement to be committed to 450mm. “Now it’s really easy to cut a press release, and it’s certainly a lot cheaper than cutting a $5B check,” he joked. Nonetheless, “the difference is that this is the first time that the Asians, especially Taiwan and Korea, have ever signed on to work in a coalition to develop the new wafer size. It’s a fundamentally strategic thing. They still have to come up with the money and do something with it but we’re certainly a lot closer. If you look at 300mm, the US guys said we want to do it, the Japanese guys said we want to go it alone and they created Selete, which wound up doubling the expenses for a lot of people,” Hutcheson said.
What development model will work? “I think we need something that is a low-risk high-ROI activity, and we’re not seeing this happen yet,” Hutcheson said. “What we need to do is take a page out of history like what the foundry guys did. They got together several companies and they all put 20% in, and that funded the first foundry fabs. It’s a perfect testbed to work out the automation issues with 450mm. Partners can share the output at a discount just like they did with the foundries.”
“If we do it right, you can do it in the $5B-$10B range in terms of total development costs,” he projected. — P.S.