K8 Manufacturability
As i am totally d'accord with you for the definition of the term, and agree on that it still has to be proven here is my take on it:
We have seen working silicon out in the wild since over a year. We do not know how far it scaled until now. Admittedly, there are indications that it is not going as quick as AMD hoped it would do. And we dont know anything about yields - apart from history: When IBM began their SoI-process, yields were poor. So, better we assume poor yields and lower than expected clockspeeds to prevent us from major disappointments. If it turns out lil bit better, fine.
What I talked about in terms of an evolutionary approach was meant not only conceptual, but processwise as well: Migration of a Fab from bulk to SoI gives AMD the degrees of freedom they are dependent on (as it is the only one they have): As far as I understand (corrections most welcome) the SoI-process uses (almost?) completely the same tools as for bulk (though operated with different parameters), so what they promise for migrating to 64bit-computing with the product they have as well: The flexibility of transition whenever and to what extent ever is suitable for them. Iaw: As long as they have poor yields and low clockspeeds, they have the option to adress the low-volume markets (Server and Workstations) only for a while and continue to serve the volumes with bulk-products. As soon as they achieve higher binsplits, Top-End Desktop can be adressed. As soon as yields get better, Mobiles, from a certain yield Performance-Segment can be adressed. (The latter, leave alone Value, will maybe only be adressable on 90nm anyway). However there is the restriction of doing this successfully only within the window of opportunity. So the crucial determinant of manufacturability (not only) of Hammer is timing. Based on the assumption that at performance-specs (that is from working silicon, not from the drawing boards) meet at least the dimensions published so far, this window is still open for a while.
The essence of all that is:
1. No doubt AMD is right in saying they are betting the farm on Hammer and on SoI at the same time - however this is trivial - whatever they would have done else would have meant nothing else. From the considerations above the path they chose with SoI and Hammer and staying on 200mm seems to be the most cautious way of going forward (if not the only one, as they cannot keep up the pace of shrinking).
2. For the final evidence that K8 is manufacturable the real challenge lies way ahead: Seeding the market with some hundredthousand 130nm-Hammers a quarter (from whatever yields) I do not consider as a manufacturing-problem; the real challenge is getting out decent yields and binsplits out of the 90nm-process in 2005 - where Intel believes the acknowledged adantages for SoI on 130nm will diminish as the bottlenecks move to somewhere else to be adressed by other means - from what nobody knows it can be done on SoI-wafers. So we still have some two years left speculating on this one. For me, it is really all about confidence in the people working on it - as i feel being far off being capable of judging about what is doable by what means and when. Thisrespectively, the closer collaboration with IBM announced lately is maybe much more important than this years launch-issues.
K.
News 
Market Data 
Discover 
AI
