Intel Corp (INTC)

[greg s]

AMD's Dresden fab is reason for delays

From the Inquirer, so take it with a grain of salt:

AMD's Dresden fab is reason for delays

Trouble at the chip mill

By Kay Nine: Thursday 19 September 2002, 16:27

Lots of people have scratched their heads over the last few weeks, worrying about AMD.
Why was the XP 2600+ introed as a complete surprise? Why is it still not available from the grey market? Has Hammer got to be that late? The sad answer is: Yes. AMD's problem these days seems to be Fab30, not the design of CPUs.

The Dresden facility was originally built with one clean room for a Cu process. The first samples, then K6-2, were made in November 1998 but never hit the streets - the yields were just too low. It took till October 1999 for the first Athlons to be produced in Dresden.

In the beginning of 2001, AMD had already started to expand the fab. Another clean room was needed for SOI. In Q4 2000 AMD started serious research in the Material Analysis Lab in Dresden.

The plant has become a kind of all-purpose-playground for AMD, it plans to even move parts of the packaging to Dresden. With all this research and expansion going on Fab30 still had to maintain a constant flow of wafers, especially when AMD gained significant market share with its Athlon XP processors. And it managed to do so, which is quite an achievement.

AMD was on par with Intels offerings on a performance basis for a long time – but this has changed.

These days people have to make the choice between an Athlon XP 2600+ or a Pentium 4 2.8 GHz - if they can find the AMD part at all.

Intel cut its prices drastically twice this year and will do so again, as soon as the hype-dreaded 3.06 part arrives. Even enthusiasts these days may choose Intel again as the price has dropped, full-featured mobos are available - and AMD lost part of its marketshare according to Mercury – see here.

The next big thing for AMD has to be Barton - a Tbred with double the cache. But large caches are hard to make in high quantities. If you have too many dead cells (there's always a few spare ones) on the die, the whole CPU is worthless. This means very tough yields.

AMD introed the Tbred at 1.46 MHz as a mobile part first in April 2002. Why mobile? My guess is, because there is little demand there. The first desktop version of Tbred was available in June at 1.8 GHz, model numbered 2200+. AMD skipped a lot of speed grades and announced the 2600+ at 2.13 GHz late August. The latest Tbred came with some minor changes to the core, which means that with the original design in four months there was only a 23 per cent increase in clock speed that AMD was able to manufacture from the same process.

Remember: The desktop version with the original TBred core started at 1.8 GHz and was never accelerated. Within three months AMD was at least able to crank out 2.13 GHz with a redesign, an 18 per cent increase.

Now for Hammer. First protoypes of the Clawhammer were widely available in May at 800 MHz and were running very smoothly, showed no patches or unusual voltage supplies on the board etc. Many people predicted the final product to ship at the end of the year. When Intel "enabled" the market for the Pentium 4, first test systems were widely available in May/June 2000.

The Willamettes then were running at 900 or 1000MHz. The chip was then introed late November 2000 at 1.4 and 1.5GHz - at least a 50 per cent increase in clock speed in six months! This is due to excellent manufacturing and good yields.

Even if AMD would intro Clawhammer at only 1.6GHz and 3400+, as was rumoured since the original announcement, it would have to double the clock speed from the prototypes in six months. It seems highly unlikely this was possible given that Fab 30, currently AMD's only plant to make the SOI Hammers, was always slow in ramping up clock speeds and never had good yields with a brand new design.

I think this is why Barton and Hammer were delayed recently. AMD has to ramp up clock speed very fast with the current TBred to keep up with Intel. But can it? µ

http://www.theinquirer.net/?article=5473

greg

- #board-369
- #board-1329