"When design and manufacturing are "tightly coupled",
the design rules will tend to be optimized for the particular application. Examples would
be the number of metal layers, thickness and pitch of metal layers, device
performance (Idsat versus Ioff), and many other "trade-offs". "
These variations can be captured in ".1" variations, as in "775.1" vs. "775.2," etc.
Within a company, they understand these variants, and presumably use them themselves.
All that remains is for them to accurately convey to the designers, whether at HP or Intel Israel or Intel Bali what these design rules are. This is precisely like having an accurate machine model to supply to compiler and application writers.
An argument that in-house coders can use information not available to outside coders is just an argument that the design or design rules are not accurately being represented.
Fix that problem, for internal use if for no other reason.
"If you seperate
these, then manufacturing will "dictate" the rules and will not be as
aggressive. "
This is really no different from Intel Bali having to adapt to a process given to them. Sure, the process and design guys meet to optimize a process, but only rarely does a specific processing trick dictate a design, or vice versa. (I remember a couple of such attempts, and they didn't do so well. And, even then, the manifestation of the tricks (what the process parameters came out at) could have been given to outside designers with no loss of leadership.)
I am unpersuaded that processor design and processor manufacturing must be done by the same company. (BTW, I think H-P just plain balked at making the decision to build a billion dollar wafer fab...they punted. Sun has always used foundries--with pretty good results until recently. IBM covers both bets. The other makers are simply too small to matter. And since H-P bought Alpha, H-P's punt gave us IA-64 as the one true processor...unless it's not the best approach, in which case a major opening for another company or country will have been opened.)
"I think we are seeing some of this with the AMD/UMC "partnership"."
Perhaps. More plausibly, though, is that this is just another case of normal stumbles AMD made frequently even without any partners.
"Another example is HPQ ... They started their "outsourcing" with INTC
fab'ing their first part, then went to IBM for their second part, and HPQ is
now coming back to INTC. The reason HPQ has come back is the need for
a CPU type performance process."
Are you saying IBM's Power4 and the new East Fishkill plant are not examples of a CPU-type performance process? That would be news to a lot of people.
--Tim May
Market Data 
Markets 
AI
