AMD To Build Fab With UMC
There’s actually two parts to this deal.
The first is that UMC will make some AMD chips if AMD needs some extra chips and can’t make enough at Dresden (CPU production in Austin is being phased out). That’s handy, but probably not major.
The bigger part of the deal is building a new fab together. AMD can’t quite afford two big fabs, so it’s planning on one-and-a-half.
The important point for people to realize is that:
1) Fabs cost a LOT of money (AMD has already spent a bit over two billion dollars on Dresden) and
2) You don’t say, “Let’s build a fab!” and have it anytime soon.
Three And A Half Years
This was announced the other day. Even if all goes according to plan, it won’t start making processors for another three-and-a-half years (and it will take more time after that to get it to full production).
That’s nothing unusual, but just to show you how crazy the CPU-making business is, this new fab will start operations based on the .065 micron process.
AMD is still at .180 micron. It’s now shifting to .130 micron. Eighteen-twenty four months from now, it will then go to .010 micron, and another eighteen or so months after that, we see this fab.
So this fab is only going to show up three generations down the road.
If Something Will Save AMD A Lot Of Money In 2005, Won’t The Same Thing Save Intel A Lot Of Money in 2002 and 2003 and 2004?
This fab will also be a 300-mm wafer fab. What does that mean?
CPU-making is a little like making pancakes. You pour the mix into the skiddle, and you make pancakes. The bigger the skiddle, the more pancakes you can make at one time.
Right now, both AMD and Intel use 200mm, or eight-inch skillets. When you hear about 300-mm wafer fabs, that boils down to going from eight- to twelve-inch skillets.
Why bigger skillets? It saves money. How much? Let Hector de J. Ruiz, president and chief operating officer of AMD, tell you:
“A 300-mm megafab can achieve cost savings of substantially more than 30 percent compared to today’s 200-mm manufacturing facilities.”
That sounds great. Only one problem. Intel knows that too, and is doing a hell of a lot about it now. Matter of fact, Intel is just about ready with two, and plans on have four of these 12-inch skillets, I mean fabs, operating by the end of 2003.
AMD’s executives have been hollering a lot lately about how their processors are much smaller than Intel’s, and thus a lot cheaper to make.
A Northwood takes about 130 sq mm of space; an AMD Thoroughbred will take up only 80mm, or about 40% less space.
A ballpark figure on how much it costs Intel to make a Northwood is about $60. Looking purely at die size, that would make the AMD Thoroughbred cost about $36.
Now take that 30%+ cost reduction and apply it to the cost of a Northwood. Now a Northwood would cost more like $40 to make. Not much difference, is there?
AMD execs don’t mention that part. 🙂
Sure, there are other factors involved, but Intel going to 12-inch skillets will go far to reduce most of the cost advantage AMD is claiming that they’ll have (though this won’t really start kicking in until we get to .100 micron). That means largely neutralizing the price war weapon for AMD in a couple years.
So getting an alternative supplier and an advanced facility is all good news, but that easily-overlooked little tidbit will have a major impact on AMD’s prospects and tactics in a couple years.