Re: I think this paragraph is a bit optimistic
I wrote In Q4 of next year... Intel get its head well above water for 1 to 2 quarters. But by then
Q4 2001 plus 1 to 2 quarters is Q1 or Q2 of 2002. I think we're in agreement on that one, I should have been clearer in the way I wrote it.
Re: From what I can see, Intel's 130nm process will have smaller Leffs and similar benefits. AMD's 130nm process will have doodads like SOI. They'll probably be more or less on par, though I'm tempted to crown Intel here if only by an Angstrom.
Intel has about a 15% smaller feature size on a 6 layer process that lacks SOI and requires deep etching and complete filling of the trenches. AMD has a moderately larger feature size, allowing the use of shallower etching but with SOI and lower capacitance these larger transistors should be as fast or faster than Intel's, and current consumption lower. The big question will be whether SOI is harder, or etching and filling those deep narrow trenches. AMD could end up with an easier yielding process that is also faster, or Intel could have the advantage, or it could be a "tie."
If AMD chooses to use all 9 layers, the chips would be harder to FAB, but smaller. At these frequencies, speed of light issues are relevant and a smaller die size may keep AMD from hitting speed path walls that will stop Intel's larger 6 layer die. If all works out perfectly (it never does) AMD would halfway to the performance of the .10 generation while Intel is stuck back at the .13 level.
Note that this article about the why of SOI is from 1998, so some of the expressions (like "getting to" 1 GHZ are a bit dated), but it still gives a nice summary of why SOI is so appealing to AMD.
IBM will combine copper interconnects with SOI transistors in a range of MPUs next year, and expects performance gains of 20 to 30 percent from the shift to SOI. By applying copper, SOI and low-k interlevel-metal dielectrics to the gigahertz processor design unveiled last February at the International Solid State Circuits Conference, IBM expects to be able to push commercial processor speeds to the gigahertz range in two to three years—faster than competitors such as Intel Corp.
And because SOI shines particularly brightly at low-voltage operation, the technology will make it possible to overcome the heat and power-dissipation problems in high-performance ICs, and offer the mobile market a means of delivering reasonable performance at single-volt supply voltages.
....Keith Diefendorff, who was the chief PowerPC architect at both Motorola and, later, Apple before joining Micro Design Resources, said IBM's announcement is "a really big deal" that puts pressure on Intel.
"You can argue that IBM has gotten a couple of steps ahead," he said. "Copper is a really big deal, and IBM got there well before Intel. They have repeated that with SOI, and the combination of both copper and SOI results in performance gains equal to almost a whole process generation." At 0.18-micron design rules, IBM can apply copper interconnects and SOI-type transistors and get "a pretty big advantage," said Diefendorff.
eetimes.com
Regards,
Dan |
