IBM yields better

[Sxotty]

At first I thought ha maybe Nv left so they have better yields

Speaking during a briefing by telephone, John Kelly, who runs IBM's chip business, said that in the past few months the number of flaws found in the chips it makes have decreased.

"Lately our defect densities have been improving quite rapidly," Kelly said. The facility started off last year with good improvements in its production of useable chips, but that then declined due to some design issues, he said.

"We seem to have turned the corner on those (design issues). We believe we understand them and we are now making rapid progress on them. But we still are not quite to our target -- to our objective -- but we are getting very close," Kelly said.

http://biz.yahoo.com/rc/040512/tech_ibm_1.html
Then after reading it seems that maybe instead they figured out what the problem was...

 

[thatdude90210]

At first I thought ha maybe Nv left so they have better yields

Then after reading it seems that maybe instead they figured out what the problem was...

Your first statement still could be right, fitting over 200mil transistors on .13 might have been the design issue he was talking about.

 

[Evildeus]

Well don't know if it's a new figure, but:

Last quarter, IBM achieved a yield rate of over 70% for its own ASICs made with the process, the sources indicated.

From Digitimes

 

[AlphaWolf]

Well don't know if it's a new figure, but:

Last quarter, IBM achieved a yield rate of over 70% for its own ASICs made with the process, the sources indicated.

From Digitimes

I wonder what the breakdown of that would be. What's the highest transistor count chip that IBM produces for itself?

 

[Dave Baumann]

The issues have never been with IBM's own processing, always with the customer models - this is what they have been working on to improve.

 

[Evildeus]

I'm not sure the wording of the article is good, cos i don't think an external partners could know the internal figures of IBM on IBM's ASICs.

I thought it was about that particular external IC figures.

 

[Dave Baumann]

The wording is perfectly clear - 70% yields on IBM's own chips. Reports have been saying this for a long time, and IBM have addressed it themselves.

 

[Evildeus]

Ok, then nothing new in here?

 

[Dave Baumann]

Not clear. It was only a few weeks ago that IBM themselves stated that the customer yields have improved - is this report indicating that futher yield improvements have been made since then, or a customer is considering moving back because of those previously announced yeild improvements?

 

[Evildeus]

Yes i agree it's not clear.

But it seems that the source is currently an IC using both UMC and IBM. So he shouldn't have difficulties to compare both solutions.

It seems that UMC (and TSMC) have still better yieds, but less room for capacity according to the same source.

that's why i was wondering if the yield rate is coming from the source, won't it be on their yield rate?

 

[Pete]

The issues have never been with IBM's own processing, always with the customer models - this is what they have been working on to improve.

Seems strange, considering Apple is reportedly unhappy with the number of CPUs they're getting from IBM. Does IBM just have a small capacity, or is the G5 just immensely popular?

 

[KimB]

Your first statement still could be right, fitting over 200mil transistors on .13 might have been the design issue he was talking about.

No, because they were talking about reducing defect densities. That has nothing to do with which chip is being fabricated.

 

[kemosabe]

Yields might have improved somewhat, but there are now rumblings about difficulties ramping 6800 chips. TSMC's capacity issues have been reported ad nauseum, but I find it strange that wafer allocation would be constrained for NVDA at IBM.

http://www.xbitlabs.com/news/video/display/20040530115849.html

 

[Evildeus]

Yeah, digitimes article says exactly the contrary :/