Paul, Questions related to AMD's foundry agreements with UMC. When a foundry has multiple clients that it is trying to satisfy on a given process, are there significant trade-offs in process parameters that must be made to allow for production of the different chips on a single line? If UMC, for example, were trying to balance process parameters between NVIDIA graphics chips and AMDs processors, what types of trade-offs would be made and what impact would those trade-offs have on each party?
-- Don't mean to butt in, but I'd like to throw my 2 cents if ok with Paul and you..... It's not uncommon to run multiple products on the same process production line. intel manufactures both P3 and P4, and soon also IA-64 Madison all on the same 0.13u Process line. It is a matter of which Masks you select for your Litho Steppers.
-- Even if you have to say.... deposit different film thicknesses, or bake in diffusion furnace for a longer or shorter times for the different layers, Implant at different energy levels, these are normally recipe adjustments to the tool, not necessarily introductions of different chemicals and or process steps for the different products. These "Product Adjustments" are usually worked out in development.
I suppose the foundry could dedicate a given line in a fab to a particular high volume customer, but wouldn't that be rather inefficient unless the customer could guarantee maximum line utilization?
-- This would seem to be very costly, unless, as you say using very high utilization..
A related question: If the process parameters at AMD's Dresden fab and UMC are significantly different for a given node, what complications might arise that would affect yield and performance? Would the chip designers have to modify the original design parameters for production on the UMC process?
-- I think you have hit the main issue nail on the head. This is where the foundry deal will make or break. Intel follows a "Copy Exactly" method, where each High Volume Production Fab uses EXACTLY the same equipment, procedures, everything, as the development Fab. Transferring the process is therefore made much easier. If it works here..... It should work there.....
Not so in this deal. Unless IF...... IF.....by some coincidence, and/or plan, UMC might have the exact same Process equipment that AMD developed the process on, if not.... there needs to be not only a transfer of the process, but there may need to be a "Process Characterization" done on some of UMC's incompatable equipment. Adjusting the recipe chemical flows and/or energy levels, to achieve the same thicknesses and/or critical dimensions.
The ReCharacterization, might go well, it might not. That I think will be most critical. Can AMD and UMC get the process transferred from a different toolset, and get similar performance from UMC's toolsets, at a similar (or lower) cost.
My WAG.... Maybe....
Semi |
