Time not ripe for 450-mm wafers
By Ann Steffora Mutschler, Senior Editor -- 10/24/2006
For a technology that may be at least a decade away from widespread adoption, the 450-mm wafer size has attracted a lot of attention.
Both proponents and opponents have vocally expressed their views, and now it looks like the consensus is that production for 450 mm is about eight to 10 years away. First, many equipment manufacturers say, investments in 300 mm need to be recouped before equipment providers and semiconductor makers will begin to seriously consider 450-mm wafers.
“The best thing right now for the industry is to maximize our investment in 300 mm and extend it further,” says Iddo Hadar, chief technology officer for the foundation technology group at Applied Materials. “For the great majority of semiconductor manufacturers and equipment makers, 450 mm doesn’t make any sense.”
“Intel and Samsung think they need 450 mm right now to stay on the economic curve, so Intel is driving this, trying to drive the industry forward,” says Dean Freeman, research director for semiconductor manufacturing equipment at Gartner Dataquest. “On the other hand are the device manufacturers and equipment manufacturers finally getting margins on 300 mm saying, ‘You must be out of your mind.’”
Another consideration is how much the equipment will cost when the transition happens, Freeman says. “Our estimate is 11 fabs will be able to afford 450-mm equipment, including Intel, the IBM consortia [IBM, Chartered Semiconductor and Samsung], Toshiba, TSMC and possibly STMicroelectronics.”
Recognizing that the timing of moving to 450-mm wafers is the $64 million question, Ted Vucurevich, senior vice president and chief technology officer at Cadence Design Systems’ advanced research and development group, says, “I don’t see it happening before 32 nm and really not likely until the 22-nm process node. The adoption of new process nodes for margin improvement has bifurcated the industry into memory, processor and high-volume SoC providers as ‘fast adopters’ and ‘the rest’.
“The fast adopters are able to follow the cost reduction available at the next process node, while the rest can’t,” he explains. “Every time we increase wafer size, we in effect double the net number of chips. Given the retooling costs associated with moving to 450 mm, the threshold of adopting a larger wafer doesn’t really make sense before 22 nm for fast adopters,” Vucurevich says.
Barriers abound
Challenging adoption of 450-mm wafers is the economic battle that the equipment manufacturers would have to face, since the adoption gap between fast adopters and the rest will mean longer time frames for equipment manufactures to recover development costs of 450-mm tools.
Freeman points out that only Intel and maybe Samsung have the volumes that 450 mm would benefit at this time. “Applied says no, but if Applied starts, the rest of the industry will too,” he notes.
Intel contends that productivity needs improvement, to which Sematech subsidiary ISMI has proposed “300-mm Prime” to improve efficiency by increasing utilization of each piece of equipment in a fab. (For more, go to “ISMI Exploring 450mm Manufacturing Transition” and “ISMI Updates on 300P.”) Members of the International Sematech Manufacturing Initiative (ISMI) have approved key concepts of the plan to move the industry toward 450mm wafers from the 300mm wafers that are currently the leading edge.
On the technology front, Jim Kupec, chief operating officer at e-Silicon and former president of UMC USA, says that a move to 450 mm will be “like moving a train sideways. It’s not just incremental pieces plugging in here and there; you have to convert the whole factory chain at the same time.”
“When that’s going on,” he explains, “there are groups of things that become barriers and need to be solved all at the same time. For instance, when the industry went from 200 to 300 mm, automation had to be solved because the wafer boxes were so heavy that operators couldn’t pick them up anymore without hurting their backs. So they had to come up with automation schemes to move the wafers through the fab that was never done before.”
Similarly, these types of issues will come up again in the eventual shift to 450-mm wafers.
Although the barriers seem daunting, there will be a time when it is clearly right to move to a larger wafer size, particularly given Gartner’s estimate that it will cost the industry $16 billion in equipment and process development to move to 450 mm. Primarily equipment manufacturers will absorb the bill.
“I think what you’ll see is a much saner move to 450 mm, and it will be well thought out. In the ‘70s and ‘80s the industry tended to take more risks,” Freeman says. “Now it is very risk-averse.”
Applied’s Hadar believes the discussion must include an examination of the problem that needs to be solved.
“One of the reasons people have considered transitions to larger wafers in the past was the die size increase. As it turns out, that is absolutely irrelevant right now as a rationale and that actually became clear halfway through the 300mm transition. This occurred because technology moved fast enough to avoid the die size increase,” he says.
Another reason given to move to 450mm was that the industry was moving so fast, it was expected that there would be a far greater number of fabs – but this is not the case. It turns out the growth of the industry is more modulated now and the scale of fabs is so significant that the ‘number of fabs’ argument has fallen apart, he continues.
“Where equipment suppliers and semiconductor companies need to put the most energy is making sure we move technology advancements as fast as possible and make it available to the largest number of fabs. If that happens, if they are able to get it ramped in production as quickly as possible, then the benefits will be there for the chip makers and ultimately for the end users and all that cost structure will improve by itself,” Hadar explains.
It comes down to the money
Ultimately, Cadence’s Vucurevich believes economic pressures will drive 450mm adoption.
“Fast adopters have applications that require extremely large volumes, such as cell phones. Imagine that you couldn't add more value by adding more features to the phone, how do you reduce the cost basis of the silicon to lower the cost of phones?” he questions.
“For the rest, the margin benefits of 450mm production are much harder to achieve. One specific challenge is that the number of average chips sized at 7-millimeters squared in a good die in a 450mm wafer lot is much greater than 300mm. As a result, ‘split’ lot production has to be improved to allow aggregation of demand,” Vucurevich adds.
Finally, moving to 450mm wafers has an impact from the design perspective. “When you go down a process node, you get two times as much function in the same space. However to fill that increase in space, you actually need to do more than two times as much design work. In effect, if you both go down a process node and increase the surface space of silicon that increases not only the number of people but also the amount of time it takes to design by more than a factor of two,” he added.
Additional resources:
450mm Wafers: Not the Best Solution
SEMATECH (SEmiconductor MAnufacturing TECHnology), a semiconductor manufacturing research consortium
© 2009, Reed Business Information, a division of Reed Elsevier Inc. All Rights Reserved.