Dear Kpf:
APM helps too. Yes it does. It helps to accelerate development, but it cannot substitute a development fab.
Yes it can. APM allows AMD to make virtual production lines, maximizing equipment capacity. They can also make a virtual rocket lot line. One rocket line and one production line are all that is needed for process R&D. The rocket line allows the die designers quick turn around for functionality testing and debugging. The production line allows the determination of the manufacturing ability of the design. IIRC, Rocket lot yields are lower than production lot yields.
Because of APM, the equipment that is used in the above lines are also used in virtual production lines producing shippable output. Where Intel and your way, a fab of a certain clean room size would have 4 production lines, AMD with APM can make 6 virtual production lines and the two virtual development lines.
An Intel development fab used to have just one production line and one rocket lot line. Lately they have two or three production lines in that fab. The only reason why Intel can't do development and production in the same fab is their copy exactly production line philosophy. The new process line won't be exactly like their old one. An even on the same process, a CMW line is different than a P4 or a CD line.
That is the big reason AMD developed APM. With virtual lines, an Opteron line and a Turion line can share much equipment. A 65nm Opteron and a 90nm Turion line can also share equipment. So a development fab with its two lines can be put in place using spare equipment capacity of the production fab. Even equipment qualification during a ramp can be accomodated using spare production equipment capacity.
Umm. Are you aware if you say Fab 36 was partly a production fab you do accuse AMD for lying and subsidy fraud? In fact, the opposite is true: They had not enough heads, tools and waferstarts available to shrink its 90nm-architecture until now: Parts of the design of products they introduced last week are still 90nm. Let alone making its new architecture work.
Once one virtual line was used for 90nm 300mm wafer production, Fab 36 became a production fab. They still had to qualify new equipment. They still has a rocket lot line that was used from time to time. Rocket lot lines are only in use after a new spin has the masks made. After they produce a lot or two of the new spin, they are idled while the testing, debugging and fixing goes on. That frees up the underlying equipment for use in other virtual lines. The virtual lines also switch their paths over time so lot 47 and lot 48 of the same virtual production line may go through different paths through the actual equipment.
So in answer to your question, no lying was done. And since Fab 30 is always going to be a 200mm fab at 90nm, 90nm lines still can be developed there. And through APM, 90nm lots and 65nm lots can pass through the same equipment. All that happens is that a 90nm mask is loaded instead of a 65nm mask and APM loads the settings for the 90nm step rather than the 65nm settings for that step. In fact it can load different settings for each piece of equipment at the same step in the production line process. Litho machine 94J may have different 90nm settings for that step in the virtual production process than Litho machine 95Q gets.
So the new architecture works because APM virtualizes the production process. hat is how a 20000WPM 200mm facility was able to process 30000 200mm wafers in a month.
Oh yes, and heads, with it. I've seen they trimmed headcount by a double-digit figure, as well.
Max, the Axe doesn't do a measly 10% reduction by getting rid of a small subdivision of a company and some attrition that naturally occurs. He axes half of middle management, a few large divisions and sacks a bunch of fiefdoms. 10% is just a warm up of his axe. So far all you had is a Nick, the Needler.
Pete |
