kash, re:<AMD needs to adress the capacity issue soon>
In Q3, AMD sold 3.8M CPU's with an average die size of about 76.4mm^2
I believe they can double wafer starts before they run out of space in Fab 25. (In the middle of Q3, AMD was getting about 160 good chips per wafer, which means there were ~24000 wafers in the quarter or 1825 per week. However, since it takes 8-12 weeks to produce, package and sell a chip, this reflects the wafer starts at the beginning of the quarter, rather than the average. It is widely assumed that Fab 25 is capable of 4,000 to 5,000 wafer starts per week, so, taking the midpoint, with a constant die size, production could increase by 146%. This is also verified by brokerage reports which lament the fact that Fab 25 is running at less than half capacity.
If AMD builds 90% K6-2's and 10% K6-3's in the fourth quarter, the average die size will increase 9.6% to 83.7 mm^2. Production in units is supposed to increase by (4.5-3.8)/3.8 = 18%. In Q1'99, assuming half K6-3's (117 mm^2) and half K6-2's (80 mm^2), the average die size would go up to 99mm^2 an 18% increase, and production is supposed to go no higher than 5M (another 11%). So the required increases are as follows:
............DieSize.....Production.......Total...Total rel.to Q2
Q3 to Q4....9.6%........18%...............29%.....29%
Q4 to Q1....18%.........11%...............31%.....69%
Q1 to Q2....18%.........0%.................18%.....99%
Available increase, assuming 4500 wafers/week capacity: 146%
So, it appears that to produce the product mix expected in Q2'99 (a few K6-2's, mostly Sharptooth and a few K7's) AMD needs to ramp from 1825 wafers per week to 3650 wafers per week. This is below the maximum capacity which has often been cited for Fab 25 of 4000-5000 wafers per week.
There isn't much cushion to allow for yield slippage as the K7 is introduced. But remember, the K6-2 was also a new product at the beginning of the 3rd quarter. On the pessimistic side, the K7 "line" may require more equipment than the K6-2 or K6-3, making 5,000 wafers/week unattainable, and 4,000 wafers per week a more accurate maximum.
The K7 on 0.25 micron technology will probably never be more than 15% of the units produced at Fab 25 (much higher revenue %!) unless Fab 30 is slow coming up to steam, so its not necessary to convert more than 1/3 of Fab 25 over towards producing this product.
PS - I'm not an AMD employee or privy to inside knowledge except the *160* number, and that was verified +- 10% by a public domain source as well.
Petz |
