IBM to produce Nvidia chips

[Glonk]

http://news.com.com/2100-1006-994072.html?tag=fd_top

Sorry if this has been posted already.

IBM to produce Nvidia chips


By John G. Spooner
Staff Writer, CNET News.com
March 25, 2003, 9:00 PM PT


Nvidia has signed a pact with IBM Microelectronics to help it avoid putting all of its graphics chips in one basket.
The chipmaker on Wednesday will announce an agreement in which IBM will serve as an additional manufacturer for the GeForce family of graphics chips. Nvidia's current manufacturer is Taiwan Semiconductor Manufacturing Co. (TSMC).

Nvidia wanted more manufacturing capacity and was looking to avoid the pitfalls of working with just one producer, said Nvidia spokesman Derek Perez.



As previously reported, Nvidia had to delay the introduction of the top-of-the-line GeForce FX processor in part due to TSMC, its only manufacturer at the time, making a transition from a 180-nanometer manufacturing process to a new 130-nanometer process. The change to a new process allows chipmakers to produce smaller, faster and less power-hungry chips.

When looking for an additional foundry, Nvidia also sought out technology that would help it boost performance, Perez said. This gave IBM a leg up on the competition.

IBM is rivaled only by Intel in its chipmaking technology and research and development prowess. The Microelectronics division, working with IBM Research, has introduced a number of new chip-manufacturing technology breakthroughs over the years. IBM also recently opened a new chip plant in East Fishkill, N.Y.

"I think (IBM) wowed us with a lot of things," Perez said. "Having all of its technology expertise in our back pocket was a no-brainer for us."

The chipmaker plans to continue its work with TSMC, even for its highest-performance chips, said Perez, who noted that the manufacturer has produced more than 200 million Nvidia graphics chips over the last five years.

Analysts believe it's in Nvidia's best interest to be able to shift production between two different foundries.

"I suspect what Nvidia is doing is diversifying its foundry base a little bit so it doesn't get trapped in a situation where it only has one foundry to deliver a product," said Dean McCarron, analyst with Mercury Research. "That changes its manufacturing situation pretty significantly, as when dealing with more than one foundry, you have the ability to start partitioning your manufacturing. It gives you more flexibility."

IBM Microelectronics will be qualified to build any future GeForce graphics processors, Perez said.

The IBM chip division will begin manufacturing unspecified Nvidia GeForce chips next summer using its 130-nanometer process, Perez said. The graphics chipmaker will not use IBM's more exotic manufacturing technologies--such as the silicon-on-insulator technique, which helps reduce power consumption and boost performance--at first, but may use them in the future.

The new contract is a big win for IBM. Nvidia is one of IBM's first manufacturing customers after a reorganization of the IBM Technology Group. The restructuring, which took place last summer, sought to expand the Microelectronics chipmaking division into a manufacturing and design services arm.

Since then, IBM has opened the Fishkill plant as a chip foundry, put in place a program to license its PowerPC processor and begun to license various chipmaking technologies to outside companies, including Sony. IBM has also teamed with Singapore's Chartered Semiconductor to share technology and manufacturing capacity.

One of IBM's first chip technology licensees since the reorganization is Advanced Micro Devices. AMD licensed IBM's chipmaking technologies, including silicon-on-insulator, for future processor manufacturing.

IBM also builds chips for cell phone makers Qualcomm and Xilinx.

The terms of the Nvidia-IBM agreement were not disclosed, but it is expected to be worth hundreds of millions of dollars over several years, a source familiar with the deal said.

 

[MuFu]

Heh, I had been waiting for something like this to happen. Even if SOI is out of the question, IBM have a mature 0.13u low-k process. No doubt it will be used for NV40. Perhaps they will tapeout to an IBM fab in light of the TSMC/NV30 problems.

I wonder if ATi will reach out Intel. They already have that broad cross-licensing agreement.

MuFu.

P.S. Spot the typo - 180nm?

 

[pcchen]

Intel didn't make chips for other companies with very few exceptions. IIRC most are server products with high margin, such as Alpha and PA-RISC. So I think it is unlikely Intel will fab for ATI.

 

[RGB]

IBM's fabs are great - very high quality. The main downside is that the high quality comes at a high price, so this will make it even harder for Nvidia to lower prices and still get decent margins.

 

[Glonk]

IIRC, Intel for a short period of time had a "fab-for-hire" business, but closed it down a couple months ago.

 

[Me]

I heard that AND had some spare capacity awhile back though. They signed a deal with motorola to share development cost and fab space IIRC.

 

[MuFu]

IBM's fabs are great - very high quality. The main downside is that the high quality comes at a high price, so this will make it even harder for Nvidia to lower prices and still get decent margins.

See that's the thing - nVidia would probably be more inclined to place their high-end ASIC databases in the hands of IBM and pay a premium for the increased likelyhood of decent yields. In the long run that would probably be cheaper than having TSMC spit out crappy dice. Meanwhile TSMC handle the less complex ASICs. Of course they wouldn't detail it in an official announcement.

MuFu.

 

[speng]

This deal with Nvidia still does not stop IBM from making ATI chips.

Just as TSMC made both Nvidia and ATI chips.

IBM would surely welcome making ATI's chips, more money for the foundry and I still think they haven't completely filled it's capacities.

Smart move by Nvidia, especially if they can license some of the emerging technologies coming out from IBM.

Speng.

 

[Evildeus]

Yes smart move and interesting for the future.

 

[Dave Baumann]

This deal with Nvidia still does not stop IBM from making ATI chips.

IIRC IBM Kinda already do: Flipper.

 

[CeiserSöze]

Flipper production is handled by NEC. IBM is producing the Gekko though

 

[epicstruggle]

any ideas why it took nv so long to go with ibm. I would have thought they would have changed ships along time ago?

later,

 

[Dave Baumann]

Sounds as though NV35 may be one of the first at IBM. Here's an alternative report:

http://eet.com/semi/news/OEG20030325S0049

By David Lammers

EE Times
March 26, 2003 (12:41 a.m. EST)


AUSTIN, Texas — Nvidia Corp. has struck a multi-year foundry agreement with IBM Corp.'s microelectronics division that brings the leading graphics chip maker into IBM's growing list of foundry customers.
Qualcomm, Nvidia, Xilinx, and two other unnamed top-ten fabless companies are now in IBM's foundry customer list, an IBM spokesman said. The Nvidia win will help IBM fill its new East Fishkill, N.Y. 300-mm wafer fab, which is now ramping.

An Nvidia spokesman said "we are in no way dumping TSMC. We are going to continue to engage in that relationship to the fullest extent possible. The reason we signed a deal with IBM is that they are a recognized technology leader, and as a fabless company Nvidia can leverage the brand new facility IBM has at East Fishkill."

IBM is winning customers to its high-end foundry initiative, said Joanne Itow, foundry analyst at Semico Research (Phoenix). However, Itow said the IBM-Nvidia relationship does not mean that TSMC is losing its grip.

"Nvidia saw capacity get tight in 2000, and really desires to have a second source of manufacturing capacity. I don't think this comes from any big problems Nvidea had with TSMC. Most of TSMC's big customers are very satisfied, and even though TSMC is relatively expensive compared with some other foundries, its customers still see TSMC as the best place to go, both for service and fairness to its customers," Itow said.

Nvidia (Santa Clara, Calif.) markets three versions of its current GeForce FX graphics design. Nvidia plans to make one or more of those chips at IBM Microelectronics (East Fishkill, N.Y.). Other members of the GeForce FX family will still be made at Taiwan Semiconductor Manufacturing Corp. (Hsinchu, Taiwan) The agreement does not call for IBM and TSMC to both manufacture the same devices, an Nvidia spokesman said.

IBM will start out this summer making Nvidia graphics products on its 0.13-micron bulk CMOS process with the fluorinated silicate glass (FSG) dielectric, rather than the more challenging low-k SiLK dielectric.

IBM expects to begin making its first 90-nm products for internal use late this summer. The three-year agreement will bring Nvidia into the 90-nm technology realm, where different flavors of low-k will be used at IBM and TSMC.

Nvidia had some difficulties last year bringing up its high-end graphics chips on TSMC's 0.13-micron FSG process, but those early problems are largely resolved, Itow said. IBM has struggled as well to get high yields on its 0.13-micron low-k process for FPGA-maker Xilinx Corp.

Nvidia did at one point run some test products at United Microelectronics Corp. (UMC, Hsinchu, Taiwan), but the Nvidia spokesman said nothing was brought to manufacturing there. At TSMC, Nvidia has made about 200 million processors over the past five years, and the partnership has helped Nvidia grow its revenues to $1.9 billion in 2002.

He declined to say which devices will be made at IBM, or when Nvidea plans to transition to IBM's 90-nm process.

In a statement, Jen-Hsun Huang, president and CEO at Nvidia, said "IBM brings a new model of advanced technology coupled with close customer integration" to the partnership. Michel Mayer, general manager of the IBM microelectronics division, said "we have a deep customer collaboration with Nvidia. It's the kind of support that is normally reserved for IBM's server group and very select large OEM partners."

 

[LeStoffer]

Thanks for the links. I thought that this was key info from Daves link:

IBM will start out this summer making Nvidia graphics products on its 0.13-micron bulk CMOS process with the fluorinated silicate glass (FSG) dielectric, rather than the more challenging low-k SiLK dielectric.

So probably no ultra-fast NV35 is coming right up.

 

[nutball]

The Register reckons IBM will be producing 90nm parts later this year (though I assume this won't be in viable quantities?). That could be quite significant for NV40 :?

 

[Dave Baumann]

The news reports state that IBM will be producing 90nm chips this year internally - i.e. for IBM. When the 90nm process will become available to their customer foundry is another question.

 

[Dave Baumann]

http://www.beyond3d.com/forum/viewtopic.php?t=5052

(Oh how I long for for phpbb 2.2 with its 'merge thread' function!)

 

[Sabastian]

Doesn't Intel consider IBM to be ... competition?

If so this may spur them to help ATi. Further ATi has some sort of relationship with Intel but just what that is based on is not so clear. Something about Intel stepping on ATi toes. Also Intel has a dislike for Nvidia and this love in between IBM and nvidia may cause Intel to move closer to ATi. Hard to say though I am not sure that ATi moving even closer to Intel would be in ATIs best intersest. This move by Nvidia doesn't bode well for TSMC Nvidia being their largest customer. Also a shift in fab companies, unless this has been in the works already, will cause delay in NV35 production. It will be interesting to see just how all this turns out at any rate.

 

[MuFu]

I doubt TSMC are in danger of losing much business from nV. They will always be able to undercut IBM when it comes to churning out crappy mainstream ASICs by the thousand.

MuFu.

 

[Sabastian]

I doubt TMSC are in danger of losing much business from nV. They will always be able to undercut IBM when it comes to churning out crappy mainstream ASICs by the thousand.

MuFu.

Yeah but they will lose them margins on the high end wafers. This also speaks volumes about TSMCs ability to be competitive on the high end in the future or at least what nvidia thinks. OTOH nvidia will have a better fab and thats something they desperately needed it appears.