A few days ago, IBM and AMD pretty much announced that they had 65nm figured out.
According to the article, they’re going to need to try out what they came up with on 90nm AMD processors from Fab 30 first before attempting to do this at 65nm.
What is rather more interesting is what is NOT in the article: any mention of IBM bringing this to manufacture for any of its chips.
This Is A Delay
Although this is being trumpeted as an advance, if you look at the events surrounding this announcement, it looks more like a rather delayed one.
65nm should have been down pat by now by IBM. They certainly could use some now. The initial XBox2 chips should have been 65nm. They aren’t, they’re 90nm chips. Expecting 90nm chips to run at 3.2GHz is probably why yields haven’t been good enough on them to prevent XBox 360 shortages, or so says Microsoft.
AMD’s new Fab 36 “was supposed to be “was designed to be a 65-nm from the start.” Who said that? An AMD exec, only eight months ago, and he said AMD was “certainly on schedule” at the time.
Instead of XBox 360s coming with 65nm chips, and AMD pumping out 65nm chips within six months, we’re going to see AMD playing guinea pig with these new processes, which is a rather big indicator these processes aren’t completely cooked yet.
This looks to be the main reason why Fab 36 won’t start off at 65nm, not any AMD-only failing.
Unfortunately, it also means that if AMD has problems implementing the new technology, 65nm chips may be further delayed, for AMD and IBM. So this doesn’t just affect Hammers and PowerPCs, it affects XBox 360 and PS3, too.
Meanwhile, in contrast to the big delays getting 90nm Prescotts out, Intel seems to be having a fairly easy time of it with their 65nm process.
This shouldn’t come as too much of a surprise. Both AMD and IBM have struggled with SOI throughout. AMD had delay after delay with Hammers, and couldn’t make more than a million Hammers a quarter until early 2005. IBM failed several times to meet Apple’s (modest) CPU demands for G5s, a factor that helped lead to the Apple switch.
Yes, AMD’s current lead over Intel is in large part due to the use of SOI, but it has been a decidedly mixed blessing for both companies, and the problems are continuing.
One Silver Lining
Maybe these new techniques will prove to be pretty cooked. They are designed to increase speed, so maybe the guinea pigs will run faster. Just, just just maybe, those overachieving Opterons and/or (more likely) those upcoming “CG” stepping A64s will be the guinea pigs. After all, they are supposed to be made in Fab 30, and if they have any hope of keeping to a 2H 2006, they would need to start the trials sooner rather than later.
It’s a good question for anyone in a position to ask AMD to ask, early and often.