Advanced Micro Devices Inc (AMD)

[Dan3]

Re: It does make sense to go to 300mm. He who starts first
will pay it off first and make more money first

It also makes sense to go to a low power SOI process. And he who starts first will pay it off first and make more money first.

In the case of Intel's 300mm wafer FABs, they were expecting a huge bubble of Y2K replacements and that 200mm2 chips on a 90nm process would dissipate less heat than they did on a 130nm process. Instead, a 100mm2 chip is dissipating more on 90nm than a 130mm2 chip did on 130nm.

So their whole move to 300mm wafers is kind of pointless - they would have been able to produce as many parts as they can ever sell on the existing 200mm lines. While they probably save $10 or so in marginal costs on each chip produced, they've spent $6 Billion, or so, converting to 300mm wafers. Allocate those FAB costs over the chips produced for the period in which the lines aren't obsolete (not totally, but as high margin, leading edge process) and the incremental cost is about $20 per chip (figuring $6 billion in extra costs spread over 300 million CPUs). Those 300mm wafers have raised Intel's costs by $10 a chip rather than lowered them by $10 per chip, as a simple marginal cost analysis would seem to indicate.

Meanwhile, AMD's far smaller investment in SOI lines is letting them move into server and notebook markets that used to be inaccessible to them - and Intel's bet on the wrong technology has resulted in Intel starting to lose whole segments. Massively parallel machines are in server rooms where heat and power dissipation is often the primary constraint. And the same thing goes for portable workstation class notebooks - a very small segment now, but one that should grow dramatically along with the move to notebook systems.