Elmer Phud: Re: "Yes you do"
Hmmm. A bit of a cryptic reply. I will try to explain myself more fully.
Things I know:
According to AMD, Fab 30 starts about 5000 wafers per week.
Thoroughbred die is 80 mm^2.
Clawhammer die will be about 106 mm^2.
Barton die size will be about 115 mm^2.
AMD sells about 8 million CPUs per quarter, or about 615,000 per week.
Things I can reasonably speculate about:
Using a reasonable model for edge exclusion, there should be about 330 Thoroughbred die candidates per wafer. Hammer should be about 248 die candidates, and Barton should be about 224.
Things I do not know:
How much capacity at Fab 30 is given to Hammer development and production.
How much capacity at Fab 30 is given to Barton development and production.
How many processors per week AMD is producing now.
Assumptions:
If it is assumed that 100% of Fab 30 production is for Thoroughbred, and AMD is still selling 615,000 units per week, and Durons are completely phased out, then yield calculation is simple. 615,000/5000 - 123 units per wafer, or a yield of 37%
If it is assumed that 75% of the Fab capacity is given to Tbred production, the yields are about 50%.
If it is assumed that 50% of the Fab capacity is given to Tbred, then the yields are about 75%.
I have no information to help me decide which of the above assumptions are closest to being correct. Of course, one could say that no matter which processes are in production, they are producing about 615,000 unite per week and you can get yield from that. The problem with that is that we do not know the breakdown of the different die sizes for the different processes. One could also argue that the problem could be bounded by worst-case. In that assumption, the worst-case average yield is 37%. However, this tells us nothing about the yields for mature processes like Tbred "A" versus new processes like Tbred "B", Barton, or Hammer.
