For qualification perhaps? Sgolds has some interesting points on the subject.
I think the conversion to .13 is actually further along than you indicate (you postulate half .13 in Q2).
Dresden was never planned to be 100% .18; rather it was planned as a 50M device plant at .13. So, some percentage of space was never allocated to .18 and is never converted to .13 - it is built at .13 from the 'gitgo'. To speak of a .13 conversion for equipment that doesn't exist is more than a linguistic nit - it changes the whole way one thinks of the fab planning and implementation!
This way of approaching the problem leads to dividing Dresden into many sections:
v% .18
w% .13 bulk converted from .18
x% .13 SOI converted from .18
y% .13 bulk
z$ .13 SOI
OK, we now have 5 processes and we haven't even talked about products. I will limit the product discussion to cores since a given core built on a given line can give us three products (mobile, desktop, server) which happens at a later stage. The products (I may not have the 7th generation on .13 named right):
T-Bird on .18
Palamino on .18
T-Bred on .13 bulk
Barton on .13 SOI
Clawhammer on .13 SOI
Sledgehammer on .13 SOI
Now, you can see that the combinations of product on processes becomes quite complex. Management decisions can control some of the complexity by deciding which product goes on converted lines and which product goes on brand new lines.
To anticipate the conversion rate, one must first determine whether we are speaking of production for existing product (such as converting .18 T-bird to .18 Palamino) or future product (perhaps converting .18 T-bird to .13 SOI Clawhammer). By now, you should be anticipating where this is all going: If you take a producing line down to convert for future product as part of the R&D to manufacturing process, that production comes out of the quarterly mix and reduces output. Presumably, that is made up somewhere else by adding new production from areas that never produced anything in the whole history of the world (previously unused fab space)!
Thus, you could go from 100% .18 to 50/50 .18/.13 and actually decrease production if you wanted. More likely, AMD has been bringing up .13 production and retiring .18 production at a rate intended to balance the needs of introducing new product (which reduces current production) and maintaining existing production levels. More total production lines are in operation, but not all of them are utilized for current shipping product.
You can do a similar analysis simply on the transition from T-bird to Palamino: Some new lines are started just for Palamino, some lines continue to product T-Bird (probably ended by now), and some lines are being readied for future product.
One last wrench to throw into the equation - production at Austin is being wound down at the same time.
Anyhow, I wouldn't attempt to put hard numbers on any of the variables (v, w, x, y, z) above, but I think this is exactly how AMD has been building out Dresden this year in a way which has kept the processor supply relatively constant over three quarters while increasing the fab utilization at Dresden. These are the factors that some here have not considered when expressing surprise that Dresden continues to build out without a radical short term impact on AMD's total units shipped.
Postscript: One can see the flexibility this gives management in their planning. Had 2001 been a better year overall, AMD could have increased production to supply more of the market. Instead, they undertook more concurrent development of future technologies so that we will have a faster move to .13 SOI on the heels of .13 bulk. I think it was very smartly done!
ragingbull.lycos.com |
