IBM chip group still recovering from 90nm transition
How long do you think IBM will be stuck?
IBM chip group still recovering from 90nm transition
By Tom Krazit, IDG News Service
July 20, 2004 7:35 am ET
IBM Corp.'s chip business was profitable in the second quarter, but its major customer is still plagued by delays in obtaining a sufficient quantity of the company's new 90-nanometer processors, signaling that IBM isn't all the way out of the woods just yet.
For the second straight quarter, Apple Computer Inc. used its earnings conference call to testify to IBM's inability to deliver enough PowerPC 970FX processors to meet demand. Due to chip supply problems, Apple was forced to delay the introduction of a new iMac system until September, months after Apple had hoped to have the system available and too late for the important back-to-school shopping season.
Earlier supply problems with the PowerPC 970FX chip forced Apple to delay shipments of new XServe servers and the new Power Mac G5 systems, the company said in April.
IBM combined the Microelectronics Group with its server business to create the Server and Technology Group in January.
Then, on its own first-quarter conference call in April, IBM acknowledged that it was suffering from yield problems at its new 90-nanometer manufacturing facility in East Fishkill, New York. Yield is a measure of how many working processors can be cut from a silicon wafer.
The company broke out the former chip group's results separately in its first-quarter financial report, which showed the former chip group lost money.
Overall yields improved substantially in the second quarter at the East Fishkill plant, IBM said last Thursday. The chip group was profitable on a pro-forma basis, using accounting rules not certified by the U.S. Federal Accounting Standards Board, more commonly known as FASB.
However, second-quarter yields did not appear to be enough to satisfy Apple's demand for both Power Mac systems and new iMac systems. Apple had hoped to have new iMac systems based on the PowerPC 970FX chip available around the time it announced the delay in early July. Apple blamed IBM, citing manufacturing problems for delays in the new iMac introduction as well as the availability of Power Mac systems that were announced in June.
An IBM spokesman declined to comment on Apple's remarks during that earnings conference call. In May, IBM held a conference call to proclaim that its yields were improving and that it believed the company had "turned the corner" on the yield problems.
The transition to new 90-nanometer technology has hit a few snags across the industry, as most process technology transitions have done over the last decade. The 90-nanometer designation refers to the average size of the structures on a given chip. A nanometer is one-billionth of a meter.
The launches of Intel Corp.'s first 90-nanometer desktop and notebook chips were delayed a few months, but the company appears to have ironed out its transition issues. Intel is actually yielding more 90-nanometer Prescott desktop processors than it had expected, forcing the company to cut back on production, it said last week.
Advanced Micro Devices Inc.'s 90-nanometer Opteron processors are
expected in the third quarter. Company executives claim they have not seen any problems with the move to 90-nanometer technology, attributing the smoothness of the transition to its use of silicon on insulator (SOI) technology in its manufacturing processes.
But IBM has struggled with the jump to 90 nanometers. It also uses SOI technology, a process in which chip makers add a thin layer of silicon oxide to their silicon wafers and then build the transistors over that layer. This is supposed to reduce leakage current from the transistors.
Power leakage has been a primary concern in the 90-nanometer transition. Transistors are getting so small that electrical current can leak out of the transistors as heat, which creates problems for system designers. In previous process technology transitions, chip makers have been able to reduce power consumption, but many 90-nanometer chips consume just as much power as their predecessors. Intel's Prescott chip actually consumes more power than its older counterpart at similar clock speeds.
Apple was forced to use a liquid cooling system to deal with the heat generated by the 2.5GHz PowerPC 970FX introduced this year. The 2.5GHz G5 processor, Apple's name for the 970FX chip, consumes about as much power as its slower predecessor but it is much smaller, according to Apple product managers. This means the heat arising from the processor is more concentrated, and a more sophisticated system was needed to control that heat, Apple said when launching the new Power Mac systems.
IBM told Apple that it expects the yield problems to disappear by the first quarter of 2005, Apple said on its earnings call. The spokesman declined to confirm that statement, and also declined to comment about specific manufacturing issues that contributed to the poor yields.
IBM chip group still recovering from 90nm transition
By Tom Krazit, IDG News Service
July 20, 2004 7:35 am ET
IBM Corp.'s chip business was profitable in the second quarter, but its major customer is still plagued by delays in obtaining a sufficient quantity of the company's new 90-nanometer processors, signaling that IBM isn't all the way out of the woods just yet.
For the second straight quarter, Apple Computer Inc. used its earnings conference call to testify to IBM's inability to deliver enough PowerPC 970FX processors to meet demand. Due to chip supply problems, Apple was forced to delay the introduction of a new iMac system until September, months after Apple had hoped to have the system available and too late for the important back-to-school shopping season.
Earlier supply problems with the PowerPC 970FX chip forced Apple to delay shipments of new XServe servers and the new Power Mac G5 systems, the company said in April.
IBM combined the Microelectronics Group with its server business to create the Server and Technology Group in January.
Then, on its own first-quarter conference call in April, IBM acknowledged that it was suffering from yield problems at its new 90-nanometer manufacturing facility in East Fishkill, New York. Yield is a measure of how many working processors can be cut from a silicon wafer.
The company broke out the former chip group's results separately in its first-quarter financial report, which showed the former chip group lost money.
Overall yields improved substantially in the second quarter at the East Fishkill plant, IBM said last Thursday. The chip group was profitable on a pro-forma basis, using accounting rules not certified by the U.S. Federal Accounting Standards Board, more commonly known as FASB.
However, second-quarter yields did not appear to be enough to satisfy Apple's demand for both Power Mac systems and new iMac systems. Apple had hoped to have new iMac systems based on the PowerPC 970FX chip available around the time it announced the delay in early July. Apple blamed IBM, citing manufacturing problems for delays in the new iMac introduction as well as the availability of Power Mac systems that were announced in June.
An IBM spokesman declined to comment on Apple's remarks during that earnings conference call. In May, IBM held a conference call to proclaim that its yields were improving and that it believed the company had "turned the corner" on the yield problems.
The transition to new 90-nanometer technology has hit a few snags across the industry, as most process technology transitions have done over the last decade. The 90-nanometer designation refers to the average size of the structures on a given chip. A nanometer is one-billionth of a meter.
The launches of Intel Corp.'s first 90-nanometer desktop and notebook chips were delayed a few months, but the company appears to have ironed out its transition issues. Intel is actually yielding more 90-nanometer Prescott desktop processors than it had expected, forcing the company to cut back on production, it said last week.
Advanced Micro Devices Inc.'s 90-nanometer Opteron processors are
expected in the third quarter. Company executives claim they have not seen any problems with the move to 90-nanometer technology, attributing the smoothness of the transition to its use of silicon on insulator (SOI) technology in its manufacturing processes.
But IBM has struggled with the jump to 90 nanometers. It also uses SOI technology, a process in which chip makers add a thin layer of silicon oxide to their silicon wafers and then build the transistors over that layer. This is supposed to reduce leakage current from the transistors.
Power leakage has been a primary concern in the 90-nanometer transition. Transistors are getting so small that electrical current can leak out of the transistors as heat, which creates problems for system designers. In previous process technology transitions, chip makers have been able to reduce power consumption, but many 90-nanometer chips consume just as much power as their predecessors. Intel's Prescott chip actually consumes more power than its older counterpart at similar clock speeds.
Apple was forced to use a liquid cooling system to deal with the heat generated by the 2.5GHz PowerPC 970FX introduced this year. The 2.5GHz G5 processor, Apple's name for the 970FX chip, consumes about as much power as its slower predecessor but it is much smaller, according to Apple product managers. This means the heat arising from the processor is more concentrated, and a more sophisticated system was needed to control that heat, Apple said when launching the new Power Mac systems.
IBM told Apple that it expects the yield problems to disappear by the first quarter of 2005, Apple said on its earnings call. The spokesman declined to confirm that statement, and also declined to comment about specific manufacturing issues that contributed to the poor yields.
Comments
I'll expect my Dual Core 970MP shortly after MWSF. I've been wanting a Quad for sometime now.
This was probably good. A little embarrassment never hurts to keep egos in check. IBM Engineers learned a bit about adversity in 2004.
The real question is how the 65nm transition is affected by the 6 month stall in the 90nm transition.
It is because of their perseverance in microprocessor technology that they landed Sony who expects results, and on top of that, Microsoft, and Nvidia who are also expecting things from them.
I know IBM is a huge company, but did they take on too much here? Do they have that much personnel to dedicate enough resources to maintain this kind of production in the same manner that they earned these customers?
All of them accept Apple doubted the PPC, and IMB architecture, but once it started to come around now they all want something, and the company that has really stood by the PPC, and IBM for the longest, (and through the worst of times of their own "processor wise" I might add) looks like one that is getting the short end of it all. (Apple)
It's Big business, and Sony will sell a ton of processors in the PS3 - you know they will, and the XBOX is still pretty popular in the US. So how does Big business look at this situation?
Cell processors I imagine need a staggering amount of R&D, and So does Microsoft's processor, and it's pretty obvious so does the PPC 90nm, Dual Core, and so will a transition to 65nm eventually.
So it's pretty much off my original thread topic, but I have to go back to questioning if IBM has enough personnel with the experience, know how, and creativity to dedicate full time resources to all these things individually, and collectively.
As much as I love the PPC, Velocity Engine, and the 970 G5 PowerMacs I worry about IBM.
Originally posted by onlooker
As much as I love the PPC, Velocity Engine, and the 970 G5 PowerMacs I worry about IBM.
<shrug> This is exactly the kind of crucible that causes great things to happen. They've done a lot of astounding groundwork, and now its time to transform it into a lot of product. Putting a fire under their toes is the catalyst that is needed to make it happen. I think the potential benefits to Apple from IBM working with Sony & MS are enormous, and in the background Freescale is still plugging away. Apple's processor future looks brighter than any time since about '96 when PowerPC hit its high relative to x86.
The issue of supply meeting demand, therefore, will be how strong the demand turns out to be. My guess is that the G5 iMac will be as strong in the innovation area as the G4 (if not stronger) and it will take a while (past Christmas) to meet that demand. If so it will be like the iPod mini and Airport express - if you don't order fast you will be waiting a while.
Another interesting point, related to price. While a lower entry level price is highly desired, can Apple even begin to meet the demand if an AIO entry level G5 is, say, $1,000 - $1,200? Will the yields accommodate the demand for a slower G5? Has IBM been stock piling slower (1.6?) G5s that can go into an entry level G5 iMac? (Does IBM consider the slower chips to be "slag" that they can sell to Apple at a very cheap price?) Will Apple want to favor the top of the range G5 iMacs in terms of availability because of the increased Gross Margin?
Lots of questions and only about 6 weeks before we get some answers.
Originally posted by Programmer
The real question is how the 65nm transition is affected by the 6 month stall in the 90nm transition.
Bring that topic up again when the G6 is around the corner. Most likely the G5 will remain a 90nm chip. Might also stay a single core. The G6(a power6 deriative) will reach the 65nm since it is based off the Power6 projected to be 65nm. The G6 has the better chance also to be a dual core as well.
Originally posted by quagmire
Bring that topic up again when the G6 is around the corner. Most likely the G5 will remain a 90nm chip. Might also stay a single core. The G6(a power6 deriative) will reach the 65nm since it is based off the Power6 projected to be 65nm. The G6 has the better chance also to be a dual core as well.
In case you hadn't noticed, the 9xx series is being more aggressive on process than the POWER series. POWER4 started at 180nm and migrated to 130nm, and the POWER5 is starting at 130nm and will migrate to 90nm. The designs are also beefed up for reliability and robustness. The 970 started at 130nm is has moved to 90nm, and the POWER4-derived 970MP (which looks like an awfully believable rumour) will be 90nm. I'd be surprised if the POWER5-lite doesn't start at 90nm and go to 65nm fairly quickly. I think we'll see 65nm sooner than most people seem to think, and it won't be POWER6 initially.
I have more confidence in IBM producing results, than I'll ever have in motorola!!!
Originally posted by quagmire
Bring that topic up again when the G6 is around the corner. Most likely the G5 will remain a 90nm chip. Might also stay a single core. The G6(a power6 deriative) will reach the 65nm since it is based off the Power6 projected to be 65nm. The G6 has the better chance also to be a dual core as well.
Given the problems at 90nm, I would very surprised if we see 65nm on the original plan. Intel was planning on moving to 65nm for some of the chips they cancelled, now they're focusing on dual core.
Remember we were all excited about SSOI and how much better it would be when we moved to 90nm? Hmmm, looks like it wasn't as good as hoped. We'll need some innovation much better than that for 65nm IMO. The good news is that IBM spends a LOT on R&D in this area, so we have some hope, but not in the next year I suspect.
Originally posted by feverishaaron
I suspect that the FX chip won't last too long. The 980 is already in development, and it is based on the POWER5, which is even farther along.
Care to back that up? (links to macosrumors don't count)
Originally posted by kiwi-in-dc
Given the problems at 90nm, I would very surprised if we see 65nm on the original plan. Intel was planning on moving to 65nm for some of the chips they cancelled, now they're focusing on dual core.
Remember we were all excited about SSOI and how much better it would be when we moved to 90nm? Hmmm, looks like it wasn't as good as hoped. We'll need some innovation much better than that for 65nm IMO. The good news is that IBM spends a LOT on R&D in this area, so we have some hope, but not in the next year I suspect.
I agree that the move to whatever process is next after the 90 nm process may be much longer in coming than anticipated and may require some new designs that were not envisioned in the prior roadmaps.
With a bit of luck IBM will get the yields on the current process headed up sometime soon. Yields were always the "bugaboo" that hampered AMD in competing with Intel. Unless the yields get up by early next year Sony, in particular, will be unhappy because the PlayStation 3 will need a bunch of processors to meet expected demand.
I think they are figuring out why it yielded so poorly in the transition, and what they learn will make the next transition that much easier because they will know where the problems are, and hopefully anticipated them.
I think they will figure it out soon enough, and use what they learned on dual core 976's, or 980's. Hopefully by MWSF, but I would like to see a PowerMac incremental update before then.
The important lesson learned by Apple & IBM is to wait until a new process is going full speed before releasing it for sale. If we see a new chip at next year's WWDC you can be sure that there will be a good supply before it is announced.
Check this article out. This is exactly what I was talking about only worse. THis guy thinks IBM is screwing Apple intentionally over these processors. I have to wonder if he's right. He makes good points.
Originally posted by onlooker
http://www.linuxinsider.com/story/34994.html
Check this article out. This is exactly what I was talking about only worse. THis guy thinks IBM is screwing Apple intentionally over these processors. I have to wonder if he's right. He makes good points.
Nice conspiracy theory, but I think it is just that. For Steve to be as gun-ho as he was about 3 Ghz (especially after the G4 into fiasco) I wouldn't be suprised if he had a contract with IBM for a delivery of that speed by this time, and not just an estimate of the clock speed. Also, IBM's chip devision is seperate from their hardware devision and Apple's contracts are with the chip devision. Apple probably has quota's listed in their contract with the chip devision which they are not able to meet in the fastest clock speed, or are just able to meet, and thus those chips are not available for the chip devision to "sell" to the hardware devision for use in their own computers. Add in that IBM's servers that the 970 is used in is a compleatly differnt type of server (a blade) which requires even greater heat constraints than the PowerMac does which may make the high end chips unusable in a fully loaded blade server.
My observations with internal devisions of the same company is that they don't "play" well with each other. Their bosses are looking to increase profitability of their own devision (and thier own bonus), sometimes at the expese of other devisions within the same company.
As to the speculation on the Cell microprocessors I didn't know that there was enough information out on this archetecture to come up with the conclusions that the author has. For example, I have not seen definate information that the Cell processors were based on the PowerPC core.
OS X is suposed to be an extreamly protable OS. Sure the might loose the AltiVec optimizations, but a lot of the advantages that that brought (as I understand it) are now bieng offloaded to the GPU. Also, if a program does not take advantage of AltiVec it see's no advantage from it bieng on the processor. I don't think that it would be a disaster if Apple were to abandon AltiVec as a dead technology today if they could achieve real world performance from a new chip that exceeded the gains that it brought to the OS and other productivity software. And to the backward compatability, due to the cocoa model (again as I understand it) all the developers would have to do is have the existing code recompiled for the new chip to make it run. Potentially removing any AltiVec optimized code, though I think that it is already set up to run even if the AltiVec engine is not available (like on an older G3 computer), and adding any new optimizations to the code for increased performance.
And last, but defianatly not least, is that the Cell Archetecture is targeted towrd game consoles. This is closer to the embeded market than the server/workstation market that IBM is targeting the Power, and their direvatives archetecture to. I would imagine that they will have a lot of performance gains in their design for their intended market but will be too specialized to work extreamly well as a general purpose CPU. They will also be set for a longer development time and product life span. Consoles have a shelf life of about 4 years before they are replaced, while computers are in a continuous upgrage cycle based on the availability of faster CPU's which tend to be released every 6-12 months.