With stories of Fermi delays, not to mention the dismal initial supply of Radeon HD 5800-series chips, it’s fair to say that the 40nm revolution hasn’t taken the world by storm. However, TSMC, which provides AMD and Nvidia’s 40nm chips, has revealed that it has plans to double the supply of 40nm chips by the end of the year.
TSMC’s senior vice president of R&D, Shang-Yi Chiang, recently discussed the issues with the 40nm switchover at the TSMC Japan Executive Forum. The first hurdle was that TSMC apparently severely underestimated the demand for the new chips from the outset.
“40nm technology happened to be a very high demand in the early stage,” said Chiang, adding that “we saw the customer demand ramp up so quick, more than what we had seen before for 40-nanometer, and we are working very hard to make up the volume.”
This isn’t going to be easy, though, as there’s only one TSMC fabrication plant (Fab 12) taping out 40nm chips at the moment, and Chiang says that this can only produce 80,000 12-inch wafers per quarter. However, Chiang promises that this quantity will double by the end of the year, with extra chips coming from Fab 14.
Of course, the other big issue for the 40nm switchover was TSMC’s yield issues last year. Chiang says that these problems are now behind the chip maker. “We resolved this yield problem in the second half of last year,” said Chiang. AMD also backed this up when we asked the company about its supply of 40nm chips for the Radeon HD 5800-series.
“It’s consistently improving,” AMD’s graphics product manager Dave Baumann told THINQ when asked about the company’s supply of 40nm chips. “If you look from pre-Christmas to post-Christmas, it’s kind of been night and day availability scenarios for the 5800-series, and we expect that to continue.”
Chiang also explained the reasons for the manufacturing problems, pointing out that the move from 45nm to 40nm was a tougher job than previous switches in manufacturing processes. “This is the first time we began to use 193nm shrink immersion,” says Chiang, “that means the photo resist during exposure will be merged in water,” which he says can result in high numbers of potential defects.
He also says that TSMC is now using its third-generation low-k material for 40nm chips, explaining that the second-generation low k material became “quite fragile,” resulting in “a lot of potential issues in the package side. “
However, while TSMC claims to have sorted out all its yield issues, it looks as though the current trickle of 40nm chips isn’t going to expand until the end of the year. This at least explains the amount of time we’ve had to wait for Fermi, as well as Nvidia CEO Jen-Hsun Huang’s recent admission that Fermi production wouldn’t hit “full stride” until at least May this year.
This production schedule was also recently confirmed by Nvidia CFO David White at the Goldman Sachs Technology Conference. Although Nvidia has announced that it plans to unveil the first high-end Fermi cards at PAX 2010 on 26 March, these are expensive high-end cards that look as though they’re going to be available in limited quantities at first. Mainstream, lower-cost cards based on the Fermi architecture won’t arrive until “the mid-part of the year,” says White.
We’ve asked Nvidia for comment on its supply of 40nm chips, and we’ll update you when we get more information.