Apparently, the folks at AMD Japan are a bit more talkative than those elsewhere, and the folks at PCWatch. (If you prefer a bit more English in your articles, take a look at what XBitLabs has to say here and here, and you might want to look here, too.
Here’s what you need to know:
We ought to see the first 65nm products around the turn of the year, and the first ones will be x2 products.
AMD will start futzing around with 45nm chips at the beginning of 2008, which will probably mean a product launch towards the end of the year.
The main process (as opposed to architectural) improvement will be the adding of germanium to the SOI sauce. This will improve transistor performance about 15% over current models.
(The AMD claim of 42% is just silly because that figure comes from a prior generation of CPUs. You don’t want to know how much better next year’s processor than last year’s; you want to know how much they are than today’s CPU. To get that figure, just divide 142%/124%, and you get 115%, or 15% better.)
We don’t really know how well AMD is doing with 65nm right now. In another piece of silliness, they made a claim that Fab 36 yielded mature yields right away, as if this were some great achievement. Well, of course they did, what they were measuring was making 90nm CPUs, an old process technology. No R&D required, all you have to do is do the same thing you’ve been doing next door for a year or more. In contrast, the other squiggly line on the graph represent the time it took to actually develop the process from beginning to end.
(Speaking of that graph, notice that there’s no percentages or time indicators on that graph. This makes it completely useless, you don’t know what the squiggles really represent. It does, however, seem to verify that AMD had a hell of a time with the initial SOI, with very subpar yields for a long, long time.)
Up to this point, Fab 36 seems to have been ramping up more slowly than expected. A number of financial analysts pointed this out, and it’s pretty hard to argue with them. If you don’t buy equipment, you can’t depreciate it.
AMD is claiming to be “on schedule,” but since they never publicly set any specific milestones for Fab 36 until the end of the year; it’s easy to be “on schedule” when there is no (public) schedule.
This isn’t necesarily too bad, but it does mean equipment installation at Fab 36 will come later rather than sooner, and will likely be pretty hectic towards the end of the year, with big, sudden jump ups towards the end.
Keep that in mind when you see a graph which seems to indicate that Fab 36 will have, say, around 30% 65nm production around September. It’s probably 30% of not a whole lot. After all, why would AMD make millions and millions of 65nm processors, then wait three months to launch the product to sell them?
Nonetheless, by early 2007, Fab 36 is supposed to have enough wafer starts, and a high enough percentage of those wafers ought to be for 65nm chips, for there to be serious 65nm production —– at Fab 36.
What happens to Fab 30 production? AMD doesn’t say a word about that, and that’s a problem. During 2007, Fab 36 will be replacing Fab 30 capacity much more than it will augment total CPU production (unless you think 90nm chips are going to continue selling like hotcakes throughout 2007). As we’ve said before, Fab 36 will make fewer (though bigger) wafers than Fab 30, and the shift to a high proportion of dual-cores will eat up most of the usual production increase from a process shrink.
Fab 36 will only be able to make a lot more CPUs than AMD is today once it goes to 45nm.
There’s been some mumbles that AMD wants to invest more money in Fab 30, but it is incredibly late in the game for them not to have made up their minds on what they’re going to do with it. I suspect AMD wants some more German financial help, which hasn’t been forthcoming yet.
Conclusions? I don’t think all is going perfectly, but probably well enough to meet the timeframes. I would say, though, that based on the information we have (and there could well be some missing pieces that could change matters), there isn’t going to be any huge boosts in production any time soon.