Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/samsung-foundry-strategy-meeting-key-finding-the-reason-google-missed-out.23063/page-2
        )

    [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] => 2021770
            [XFI] => 1050270
        )

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

Samsung Foundry, Strategy Meeting Key 'Finding the Reason Google Missed Out'

benb said:
This is UNTHINKABLE at Samsung or Intel.
Click to expand...
The more high stakes the node race become, the lower in priority the yield race become.

Today's leading edge process in tomorrows mainstream, and the day after tomorrow's mature node. If you want it to keep make money for 20 years, you need to be able to do what you have described, but also less pressure on engineers being whipped to race new nodes and features.

TSMC goes fast literally because they historically didn't crazily chase being first, for the sake of being first. The more resources are taken for node chase, the less resources left for yield chase.

I see the big clog on the market for the last planar node, and I see TSMC very much seeing it too, but still not being willing to upgrade 40nm, and improve service with sane shuttle offering (while GloFo keeps eating more of 22-45 market)
 
benb said:
Many fabs have this same problem, which boils down to process maturity: It is rare these days for a process to be mature enough at introduction, but only TSMC permits significant changes after introduction. This is their super power. At TSMC, in the middle of the 7nm node, if you want to introduce EUV lithography, you can, and there is a system in place (informed by and approved by customers) to support this. This is the most extreme case imaginable, and every other process in the fab is equally empowered to change, after the node is introduced, when it brings benefits. This is UNTHINKABLE at Samsung or Intel. The lead site dictates, other sites follow. The result: Slow ramp of yield, relative decline of competitiveness.
Click to expand...
From what I hear this is changing at Intel. I've heard a lot of churn regarding Copy Exactly from people I know. It sounds like there is a push to view Copy Exactly as a required methodology for process transfer, and unlike in the past remove the handcuffs once the process is up and running in HVM at another site. This will require significant changes in Intel's systems to keep process change control across their fabs, but will make them much more competitive if they can pull it off.
 
Paul2 said:
TSMC goes fast literally because they historically didn't crazily chase being first, for the sake of being first. The more resources are taken for node chase, the less resources left for yield chase.
Click to expand...

Intel Foundry made the same mistake Samsung and Globalfoundries made, they went after TSMC. I told the Intel people first off to go after Samsung Foundry, a much easier target. HNA-EUV is a glaring example. Instead of trying to beat TSMC, follow TSMC. TSMC followed Intel for many years before passing them. There is no express lane in the foundry business.

Here is the problem now, the ecosystem cannot be toyed with. IP companies do not make money if the customers do not come. A lot of money was spent porting IP to Samsung 3nm and Intel 18A with no customer follow through. IP companies will not make that mistake again. To top that off, Rapidus is investing ridiculous amounts of money in building an ecosystem (IP) for limited capacity manufacturing. Just what is an IP company to do? Take the money!
 
Paul2 said:
The more high stakes the node race become, the lower in priority the yield race become.

Today's leading edge process in tomorrows mainstream, and the day after tomorrow's mature node. If you want it to keep make money for 20 years, you need to be able to do what you have described, but also less pressure on engineers being whipped to race new nodes and features.

TSMC goes fast literally because they historically didn't crazily chase being first, for the sake of being first. The more resources are taken for node chase, the less resources left for yield chase.

I see the big clog on the market for the last planar node, and I see TSMC very much seeing it too, but still not being willing to upgrade 40nm, and improve service with sane shuttle offering (while GloFo keeps eating more of 22-45 market)
Click to expand...

"The more resources are taken for node chase, the less resources left for yield chase."

Will this observation also apply to Intel's 5N4Y?
 
hist78 said:
"The more resources are taken for node chase, the less resources left for yield chase."

Will this observation also apply to Intel's 5N4Y?
Click to expand...

Yes, because the people who do serious yield optimisation, and new processes are more of less the same.
 
Daniel Nenni said:
Intel Foundry made the same mistake Samsung and Globalfoundries made, they went after TSMC. I told the Intel people first off to go after Samsung Foundry, a much easier target. HNA-EUV is a glaring example. Instead of trying to beat TSMC, follow TSMC. TSMC followed Intel for many years before passing them. There is no express lane in the foundry business.
Click to expand...
If this article is to be believed, it seems like under Lip-Bu Tan Intel might be shifting towards a focus on Samsung's share of the foundry market.
https://www.trendforce.com/news/202...ry-connect-in-seoul-sparks-sk-hynix-hbm-buzz/
 
You must log in or register to reply here.
Back
Top