Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/japan-to-set-up-new-semiconductor-outfit-with-ibms-help.17054/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Japan to set up new semiconductor outfit with IBM's help

Looks like an R&D joint venture. They haven't committed to licensing the technology from IBM yet.


Rapidus will work to get a licence from IBM to manufacture sub 2-nanometre chip technology in Japan once they develop the capability, according to people familiar with the talks.
 
Any idea what is IBM's strategy here with R&D on advanced nodes like 2nm - What are they hoping to achieve? They don't do any manufacturing... purely relying on Samsung for their chips, right? Then why waste money on R&D like this?
 
Any idea what is IBM's strategy here with R&D on advanced nodes like 2nm - What are they hoping to achieve? They don't do any manufacturing... purely relying on Samsung for their chips, right? Then why waste money on R&D like this?

I think IBM wants to build a institution like Interuniversity Microelectronics Centre (IMEC) in Belgium to pursue advanced manufacturing technology in semiconductor industry.

One of the problems I can see is IMEC operates as a non-profit organization to serve its partners and sponsors. This model doesn't work under IBM's situation because at the end of day IBM is looking for as much profit as possible.

IBM probably believes the government is/should be its semiconductor research organization's financial hacker. I'm guessing again.
 
Last edited:
I think IBM wants to build a institution like Interuniversity Microelectronics Centre (IMEC) in Belgium to pursue advanced manufacturing technology in semiconductor industry.

One of the problems I can see is IMEC operates as a non-profit organization to serve its partners and sponsors. This model doesn't work under IBM's situation because at the end of day IBM is looking for as much profit as possible.

IBM probably believes the government is/should be its semiconductor research organization's financial hacker. I'm guessing again.
Both Intel and AMD depend on TSMC, in part or in whole. I think Japan (and maybe IBM as well) are attempting to extricate themselves from CCP control, in a darker future in which the CCP has seized control, in whole or part, of Taiwan.
 
Both Intel and AMD depend on TSMC, in part or in whole. I think Japan (and maybe IBM as well) are attempting to extricate themselves from CCP control, in a darker future in which the CCP has seized control, in whole or part, of Taiwan.
IBM is independent of the PRC and ROC. After GF doped out they went to their other partner (Samsung) for manufacturing
 
I think IBM wants to build a institution like Interuniversity Microelectronics Centre (IMEC) in Belgium to pursue advanced manufacturing technology in semiconductor industry.

One of the problems I can see is IMEC operates as a non-profit organization to serve its partners and sponsors. This model doesn't work under IBM's situation because at the end of day IBM is looking for as much profit as possible.

IBM probably believes the government is/should be its semiconductor research organization's financial hacker. I'm guessing again.

Do fabs pay IBM royalty for using it's technique?
 
IBM has always developed the base process technology for its own server business and continues to do so - thus this 2nm breakthrough, which is legit. As early in the 2000s its proprietary processor volumes were insufficient to offset the skyrocketing costs of fabrication - just think about the huge cost bump in the migration from 200mm to 300mm fabrication -, IBM first designed the processors and operated as a foundry for Nintendo, Sony, MS Xbox plus Apple (as it was running on PowerPC). Aggregating these volumes added to its own proprietary needs, IBM could amortize the capex needed to support the processing and fabrication cost for its own server business that was extremely profitable (mainframes, UNIX servers) even if the IBMs server processor volumes were no match with X86. On top IBM's processor R&D delivered the process technology that went into AMD, Samsung and STMicro in a development alliance. Once Apple defected to Intel for its Macs and the console businesses (Nintendo, XBOX and Sony) eventually migrated off too, IBM could not simply offset those volumes and lost cost competitiveness. Thus the spin-off of its CAPEX hungry fabrication business, pooled fabs together with AMD fabs, sold off to Mubadala who created Globalfoundries.
Even then, Globalfoundries and Samsung would rely on the continued output of IBM Process Technology R&D to get below 10nm. Once GF de-committed the sub 10nm roadmaps, IBM, who needed sub 10nm for the competitiveness of its servers, collaborated with Samsung and enabled Samsung to compete with TSMC. Remember Smasung was king in memory, not so much in logic. Fact is, today's IBM Mainframe and Power Processors are fabricated at Samsung.
IBM never abandoned its R&D efforts - nor would CFIUS probably ever have agreed to the Mubadala transaction, had it included the R&D portion. Net - IBM is more than legit in process technology. Its processor technology and design teams are second to none. IBM just monetizes its investment in mainframes, Unix servers and related SW stack - for those of you who read an I&E and a balance sheet, these are two of the most profitable HW businesses to this day.
If I were Hitachi, Fujitsu and the Japanese government and wanted to catch up on fabrication and logic processor technology, IBM would be a prime address to call on.
 
IBM used to be a powerhouse for innovation, including semiconductors (device structures, processing technologies, technology applications, etc.).
15-20 yrs ago they were earning $4-5B a year revenue on licensing various technologies to various companies/industries (including the then semi superpower called "Intel" ;-). Not sure if they've been continuing on that path for the last 10-15 yrs since they seem disappeared from the semi competitiveness scene (even though I still think IBM has deep pockets with perhaps some hidden gems, from licensing perspective). I'd not be surprised IBM has been doing research on the sub-5 or 3 nm space but I am totally not sure how this partnership with Japan will work out, since Japan has been out of advanced logic for decades and how are these technologies to be commercialized for revenue...
 
IBM used to be a powerhouse for innovation, including semiconductors (device structures, processing technologies, technology applications, etc.).
15-20 yrs ago they were earning $4-5B a year revenue on licensing various technologies to various companies/industries (including the then semi superpower called "Intel" ;-). Not sure if they've been continuing on that path for the last 10-15 yrs since they seem disappeared from the semi competitiveness scene (even though I still think IBM has deep pockets with perhaps some hidden gems, from licensing perspective). I'd not be surprised IBM has been doing research on the sub-5 or 3 nm space but I am totally not sure how this partnership with Japan will work out, since Japan has been out of advanced logic for decades and how are these technologies to be commercialized for revenue...
Thanks to @ECLUS. Sounds like obviously IBM has never pulled their foot from the R&D gas pedal. Good for them!
 
IBM has always developed the base process technology for its own server business and continues to do so - thus this 2nm breakthrough, which is legit. As early in the 2000s its proprietary processor volumes were insufficient to offset the skyrocketing costs of fabrication - just think about the huge cost bump in the migration from 200mm to 300mm fabrication -, IBM first designed the processors and operated as a foundry for Nintendo, Sony, MS Xbox plus Apple (as it was running on PowerPC). Aggregating these volumes added to its own proprietary needs, IBM could amortize the capex needed to support the processing and fabrication cost for its own server business that was extremely profitable (mainframes, UNIX servers) even if the IBMs server processor volumes were no match with X86. On top IBM's processor R&D delivered the process technology that went into AMD, Samsung and STMicro in a development alliance. Once Apple defected to Intel for its Macs and the console businesses (Nintendo, XBOX and Sony) eventually migrated off too, IBM could not simply offset those volumes and lost cost competitiveness. Thus the spin-off of its CAPEX hungry fabrication business, pooled fabs together with AMD fabs, sold off to Mubadala who created Globalfoundries.
Even then, Globalfoundries and Samsung would rely on the continued output of IBM Process Technology R&D to get below 10nm. Once GF de-committed the sub 10nm roadmaps, IBM, who needed sub 10nm for the competitiveness of its servers, collaborated with Samsung and enabled Samsung to compete with TSMC. Remember Smasung was king in memory, not so much in logic. Fact is, today's IBM Mainframe and Power Processors are fabricated at Samsung.
IBM never abandoned its R&D efforts - nor would CFIUS probably ever have agreed to the Mubadala transaction, had it included the R&D portion. Net - IBM is more than legit in process technology. Its processor technology and design teams are second to none. IBM just monetizes its investment in mainframes, Unix servers and related SW stack - for those of you who read an I&E and a balance sheet, these are two of the most profitable HW businesses to this day.
If I were Hitachi, Fujitsu and the Japanese government and wanted to catch up on fabrication and logic processor technology, IBM would be a prime address to call on.
All true except for the statement that "IBM is more than legit in process technology". IBM's processes are not anywhere close to production ready or even feasible (similar to IMEC nodes). Samsung has to do significant legwork to adapt and rearchitect IBM's designs just to make them manufacturable (and at least for the past few gens even this hasn't been happening).

I also never really understood the purpose of this R&D. I can't imagine they drive much value from it. They are only making royalties off of Samsung, and the node that IBM "gets" isn't exactly what they designed either (preventing IBM from gaining any IDM advantages from Samsung nodes). Presumably POWER CPUs would also be better if they had the superior performance of an "equivalent" TSMC node, as well as not having to wait as long for the Samsung node to reach the maturity necessary for making their HUGE POWER systems. I saw an interview where the IBM folks were talking about how this R&D let's them "see into the future" (of semiconductors) and "design their chips around this". I am not in the design side of the business, but I don't really understand this. Are IMEC whitepapers and your own testchips/foundry partner roadmaps not enough to plan out your future architectures?
 
Last edited:
You can read on their website what they do - for example on advanced semi R&D https://research.ibm.com/topics/semiconductors#our-work
or ... just look at the processor design of the last generation of mainframes and the last generation or Power Chips. Here is the Power 10 spec https://newsroom.ibm.com/2020-08-17-IBM-Reveals-Next-Generation-IBM-POWER10-Processor - remember it was released in 2020 - thus the next 'rev' should be imminent and more advanced. Or look at mainframe processors here - https://research.ibm.com/blog/telum-processor. These are different beast for people who are used to AMD or Intel Designs, but the technology levers applied in their design are top notch - just optimized for IBM's use cases. And as said - these technologies refresh at a different rhythm than a PC market - mainframes and Unix servers typically refresh every 2 to 2 and a half years. Thus stay tuned.. if history is to go by, they will deliver 5nm and below fairly soon.
 
IBM is a semiconductor company in the same way Bernie Madoff was a money man
All true except for the statement that "IBM is more than legit in process technology". IBM's processes are not anywhere close to production ready or even feasible (similar to IMEC nodes). Samsung has to do significant legwork to adapt and rearchitect IBM's designs just to make them manufacturable (and at least for the past few gens even this hasn't been happening).

I also never really got this R&D. I can't imagine they drive much value from it. They are only making royalties off of Samsung, and the node that IBM "gets" isn't exactly what they designed either (preventing IBM from gaining any IDM advantages from Samsung nodes). Presumably POWER CPUs would also be better if they had the superior performance of an "equivalent" TSMC node, as well as not having to wait as long for the Samsung node to reach the maturity necessary for making their HUGE POWER systems. I saw an interview where the IBM folks were talking about how this R&D let's them "see into the future" (of semiconductors) and "design their chips around this". I am not in the design side of the business, but I don't really understand this. Are IMEC whitepapers and your own testchips/foundry partner roadmaps not enough to plan out your future architectures?
All true except for the statement that "IBM is more than legit in process technology". IBM's processes are not anywhere close to production ready or even feasible (similar to IMEC nodes). Samsung has to do significant legwork to adapt and rearchitect IBM's designs just to make them manufacturable (and at least for the past few gens even this hasn't been happening).

I also never really got this R&D. I can't imagine they drive much value from it. They are only making royalties off of Samsung, and the node that IBM "gets" isn't exactly what they designed either (preventing IBM from gaining any IDM advantages from Samsung nodes). Presumably POWER CPUs would also be better if they had the superior performance of an "equivalent" TSMC node, as well as not having to wait as long for the Samsung node to reach the maturity necessary for making their HUGE POWER systems. I saw an interview where the IBM folks were talking about how this R&D let's them "see into the future" (of semiconductors) and "design their chips around this". I am not in the design side of the business, but I don't really understand this. Are IMEC whitepapers and your own testchips/foundry partner roadmaps not enough to plan out your future architectures?
IBM is a different beast: assume you have a multibillion dollar software business that rides on top of your proprietary processor architecture. And assume this Software runs - to this day - the most demanding processes in the finance and other sectors. And assume you have a multibillion dollar maintenance business related to that HW and assume you have to stay competitive as you are basically minting money here. Assume further that after missing the PC processor business and after somewhat missing the cloud business you have established an early lead in Quantum.
Would you then propose to the board to make yourself dependent on a processor designer and an external fab to save cost?
Here is what you would have done in 2015 - have Intel design your processors for mainframes and then give fabrication out externally to save Capex. IBM did not go to Intel for designs but did go fab-less with the creation of Globalfoundries.
Just look at what happened: Intel is struggling to regain the lead from AMD, Globalfoundries was supposed to deliver sub 10nm fabrication to IBM and AMD and bailed. AMD went to TSMC and got a leg up on Intel that f'..d up their processor and process roadmap, IBM - having kept its own processor design and process R&D, was able to control its destiny on processor design and to get SAMSUNG to produce its sub 10nm chips. Net: IBM is an odd beast in technology. IBM monetizes its R&D in its downstream HW, SW and Services plus it monetizes the same R&D in royalties to third parties. Had IBM relied on Intel five years ago it would probably have saved a ton of money but probably jeopardized the competitiveness of its downstream business.... makes no sense. They are pretty smart in what they do. A jewel to this day..... but not easy to understand fully if seen just from a semiconductor industry logic.
 
IBM is a different beast: assume you have a multibillion dollar software business that rides on top of your proprietary processor architecture. And assume this Software runs - to this day - the most demanding processes in the finance and other sectors. And assume you have a multibillion dollar maintenance business related to that HW and assume you have to stay competitive as you are basically minting money here. Assume further that after missing the PC processor business and after somewhat missing the cloud business you have established an early lead in Quantum.
Would you then propose to the board to make yourself dependent on a processor designer and an external fab to save cost?
Here is what you would have done in 2015 - have Intel design your processors for mainframes and then give fabrication out externally to save Capex. IBM did not go to Intel for designs but did go fab-less with the creation of Globalfoundries.
Just look at what happened: Intel is struggling to regain the lead from AMD, Globalfoundries was supposed to deliver sub 10nm fabrication to IBM and AMD and bailed. AMD went to TSMC and got a leg up on Intel that f'..d up their processor and process roadmap, IBM - having kept its own processor design and process R&D, was able to control its destiny on processor design and to get SAMSUNG to produce its sub 10nm chips. Net: IBM is an odd beast in technology. IBM monetizes its R&D in its downstream HW, SW and Services plus it monetizes the same R&D in royalties to third parties. Had IBM relied on Intel five years ago it would probably have saved a ton of money but probably jeopardized the competitiveness of its downstream business.... makes no sense. They are pretty smart in what they do. A jewel to this day..... but not easy to understand fully if seen just from a semiconductor industry logic.
By this logic their gambit has failed. Samsung can't use IBM's IP (it is not designed to be easily mass producible), furthermore Samsung's nodes are always behind schedule. By extension IBM is not independent since the technology they get and when it comes out is not (directly) in IBM's hands. If IBM spent $0 on R&D they could still use TSMC (or maybe Samsung depending on how dependent they are on IBM's IP, however since they haven't dropped IBM all together they are probably pretty dependent on it for at least components/materials research).

Also how is Samsung any different than GF. If anything it is worse since GF at least had 14HP which literally had IBM's14nm FEOL (even if they had to replace IBMs nonfunctional BEOL with GF's 7nm). As far as I know Samsung's nodes don't ever just take IBM's designs (I can't remember if it was for 7nm or 5nm where IBM just cranked EUV dose to get yields and Samsung had to go through gymnastics reengineering things just to get functional yields)
 
Nothing is immutable. TSMC is up today-agreed and yes, nothing is ever as simply as cut and paste or 'shoehorning in' to bring something to high yields and scale production. But - Leadership changes happen at major disruptions... stay tuned. Has always happened like this in my 35 years in the business.
The point remains - having the skills and capabilities proved invaluable to IBM and no, they have not failed - they were able to sustain their source of monetization without major hick-ups, because they still had sufficient autonomy. Are they fully independent? - Never said that and nobody is. Not even TSMC is. Samsungs efforts are still in early stage and yes, they have their work cut out for them. But would not write them off. Intel' IFS is a gambit - but watch that space with all the investments coming online. They are actually executing on their process roadmap and have deep pockets and 'Chip Acts'......
 
Nothing is immutable. TSMC is up today-agreed and yes, nothing is ever as simply as cut and paste or 'shoehorning in' to bring something to high yields and scale production. But - Leadership changes happen at major disruptions... stay tuned. Has always happened like this in my 35 years in the business.
The point remains - having the skills and capabilities proved invaluable to IBM and no, they have not failed - they were able to sustain their source of monetization without major hick-ups, because they still had sufficient autonomy. Are they fully independent? - Never said that and nobody is. Not even TSMC is. Samsungs efforts are still in early stage and yes, they have their work cut out for them. But would not write them off. Intel' IFS is a gambit - but watch that space with all the investments coming online. They are actually executing on their process roadmap and have deep pockets and 'Chip Acts'......
I wish I could work at Samsung's R&D department and see how IBM's R&D compares to TSMCs/Intel's components research departments, and how much many changes need to be made to make it HVM capable. The vibe I get is that IBM's IP is worse in this regard, but given I don't work at Samsung I couldn't say anything authoritative on the matter. As for the potential of TSMC declining, or intel declining if they used intel back in 2015, couldn't IBM just switch foundry to whomever is the new leader. After all they switched from IBM >14HP > GF 7nm > SF 7nm. For this reason I see literally no risk to going to TSMC. In what way is this IBM Samsung relationship"invaluable" to IBM. There are people much smarter than me working there, so the decision has to make sense for some reason, but every argument I've ever heard from IBM has been unconvincing.
 
Back
Top