John,
<<How can you say 'no'? The example of Motorola producing and selling 0.5um+ microcontrollers at under $1 each gives a contrary answer.>>
Motorola can sell 0.5um processors at less than $1 because the die size of a 68HC05 is very small even at that geometry. By the way, HC logic gates sell at under $0.10 and they are on 3.0um processes. Obviously, if the die size is very small, there are diminishing returns in shrinking the die further.
In the case of AMD and Intel, AMD has to go to 0.25um in order to survive. As someone pointed out, the Fab 25 building should be as depreciated as any of Intel's 0.35um plants, so the steppers, etchers, and other equipment are the main extra depreciation cost. Until AMD ramps the 0.25um process up fully, this depreciation will be a considerable per piece cost (thus the $700 million break-even figure). Once AMD ramps up 0.25um later this year, this will be less of a factor.
As for Intel continuing to compete with 0.35um products, which ones do you suggest? Would the Pentium/MMX yield well at 266 and 300 MHz? If so, then this might have been a smarter move in the near term, but I think Celeron was a better strategic move. Intel is more concerned with converting people to the Slot 1/Pentium II mindset at the moment than it is on squeezing out big profits on the low end. Intel realizes that AMD would eventually increase 0.25um yields and that AMD would have to be defeated in the marketplace rather than the fab.
I don't know whether Celeron will be successful, but it will be priced low enough to keep AMD in the red for a while longer. If Celeron is successful (even if it costs Intel money), AMD will be forced to adopt Slot 1 (and endure Intel's lawsuit).
Richard |
