Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/samsung-3nm-1st-generation-reaches-stable-yield-in-mass-production.17417/
        )

    [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
)

Samsung 3nm 1st generation reaches 'stable' yield, in mass production

Fred Chen

Moderator

Regarding next-generation GAA processes, the 3nm first-generation process is currently being mass-produced with a stable yield and based on the first-gen mass-production experience, the development of the 3nm second-generation process is progressing rapidly.
 
Of course it is; they have "perfect yields". You can't get better than 0 defects/cm^2; hence the stability. All jokes aside I do hope we get a nice information dump on at least 3GAE (and hopefully 3GAP) some time soon. I'd love to look at the choices they made and what sort of decisions they made to address the fundamental issues that HNSs need solved.

Here's hoping that 3GAP family is a good entry point for Samsung and that they can make up lost ground.
 
Apparently, they made an SRAM in late 2021, but did not show any small SRAM cell size when published at beginning of 2022. https://ieeexplore.ieee.org/document/9601158 So GAA is not a shoo-in shrink.

I think they always compared their 3GAE with their 5nm process and it showed somewhat lacking density improvements, but had more than a generation upgrade power efficiency. So this is expected a long ago. I'm actually quite curious about how 3GAP will work. Can it achieve density improvements over 3GAE? Can they attract customers who didn't like FinFETs before?
 
Can they attract customers who didn't like FinFETs before?
I believe the largest beneficiaries of these advanced logic technologies are the FPGA companies, but I would appreciate it if somebody will chime in on this.

The FPGA fabric is made up of CLBs (configurable logic block) and switching matrices between CLBs. The SRAM that configures the CLBs can can be programmed in slowly (serially). These horribly inefficient programmable things benefit greatly by more densely packed logic, even if the RAM isn't scaling.

The flaw in my argument is that Apple is paying for much of the R&D. Does Apple have a Verilog app that allows the users to program their phones?
 
I haven’t heard of anyone of note using Samsung 3nm. Anyone heard anything else?

Samsung missed the early adopters, as usual. Even QCOM and Nvidia switched back to TSMC. As I said before TSMC N3 will be the biggest node in TSMC FinFET history since it is the last FinFET node and there is no competition. GAA is a promising alternative but the ecosystem is not ready for the majority of companies. Remember, FinFET is open source but GAA is not so it will require custom models etc... The TSMC Symposium is in two weeks, we will know more then, but until Samsung has identifiable customers for GAA it is all just press releases, smoke and mirrors.
 
Samsung missed the early adopters, as usual. Even QCOM and Nvidia switched back to TSMC. As I said before TSMC N3 will be the biggest node in TSMC FinFET history since it is the last FinFET node and there is no competition. GAA is a promising alternative but the ecosystem is not ready for the majority of companies. Remember, FinFET is open source but GAA is not so it will require custom models etc... The TSMC Symposium is in two weeks, we will know more then, but until Samsung has identifiable customers for GAA it is all just press releases, smoke and mirrors.
Do you know if Samsung is sending folks to help customers transition to GAA designs?
 
Samsung missed the early adopters, as usual. Even QCOM and Nvidia switched back to TSMC. As I said before TSMC N3 will be the biggest node in TSMC FinFET history since it is the last FinFET node and there is no competition. GAA is a promising alternative but the ecosystem is not ready for the majority of companies. Remember, FinFET is open source but GAA is not so it will require custom models etc... The TSMC Symposium is in two weeks, we will know more then, but until Samsung has identifiable customers for GAA it is all just press releases, smoke and mirrors.
How much do you think we will get on N2 at the symposium Daniel? I am beyond curious as to if we will get in depth details this year of N2 and beyond.
 
There’s some kind of huge untold story here with 3GAA. What makes a breakthrough new technology receive so little commercial interest?
 
How much do you think we will get on N2 at the symposium Daniel? I am beyond curious as to if we will get in depth details this year of N2 and beyond.

I think this year's symposium will again be all about N3. N3 will be a VERY long node, absolutely.

8L66pLqdjNibGw54ERoo5e-1200-80.png
 
I do not know. I will ask around about Samsung GAA design starts. The only ones I know about now are Intel 20A.
You’ve been saying that a bit recently. Are you getting indications 20A not going to be HP only node like intel 4 is, or that in spite of this some customers would rather have that than 18A? Or are you using 20A instead of 18A as the catch all name for that whole family of nodes?
 
You’ve been saying that a bit recently. Are you getting indications 20A not going to be HP only node like intel 4 is, or that in spite of this some customers would rather have that than 18A? Or are you using 20A instead of 18A as the catch all name for that whole family of nodes?

Sorry, I meant 18A, the foundry version.
 
I think this year's symposium will again be all about N3. N3 will be a VERY long node, absolutely.
The plan for tsmc N2 HVM is 4 phases in Hsinchu and few phases in Taichung. According to the latest rumors from supplier chains, tsmc have postponed Taichung site and 3 phases in Hsinchu for N2 HVM to finished in 2026 or later and will only have one phase in Hsinchu ready in 2025.
If these are true which mean N2 will only significantly contribute to revenue till 2026. tsmc's earning call next week should address these rumors.
 
Back
Top