Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/samsung-says-it-will-beat-tsmc%E2%80%99s-4nm-production-capacity.18751/
        )

    [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 SAYS IT WILL BEAT TSMC’S 4NM PRODUCTION CAPACITY

Daniel Nenni

Daniel Nenni

Admin
Staff member
3nm chips

Samsung has been working on improving its 4nm chip-making process for a while now, and it seems that the company is making significant progress. According to Sammobile and 9to5google, Samsung’s 4nm process is now on par with TSMC’s, and the company is confident that it will beat TSMC’s 4nm production capacity soon. Samsung’s steady progress in improving its 4nm chip yield may earn the company new orders from key clients such as Qualcomm and NVIDIA.

Smartphone chipset market review
Img Src: Nextpit

According to TrendForce, the company’s contract chip unit is gaining market share. It now appears that the huge funding of its foundry business is now paying off. In addition, Samsung is expected to start mass production of chips using 4nm process technology in the United States by the end of 2024. It will be ahead of TSMC’s Fab 21 plant in Arizona.

In a special speech at the Seoul National University, Samsung co-CEO Kyung Kye-hyun said the company is optimistic about its position in the semiconductor market. He also claims that the company is ready to compete more aggressively than before.

SAMSUNG TO MASS PRODUCE THIRD-GEN 4NM SEMICONDUCTOR CHIPS SOON​

Samsung is all set to begin mass production of the third-gen 4nm chips during the first half of 2023. The company has invested $3.8 billion to increase its 4nm production factories to improve yields. Samsung’s Pyeongtaek location spans nearly 2.89 square kilometres. It is as wide as the total space of chip manufacturing sites in Hwaseong and Giheung combined. This new facility will help Samsung to increase its production capacity and meet the growing demand for its 4nm chips.

GIZCHINA NEWS OF THE WEEK​

Samsung’s upcoming fab near Tyler, Texas, will be the company’s first leading-edge production facility in the U.S. in years. Samsung has high hopes that it will enable it to meet the needs of its many U.S. customers and challenge Intel and TSMC’s foundry services.
According to the report, if all goes according to plan, the new factory of Samsung’s 4nm process technology (SF4E, SF4, SF4P, SF4X and SF4A) will start mass production by the end of 2024. While this is unlikely to have an immediate impact on the foundry market, Samsung can arguably beat TSMC to the punch with its 4nm process in the US. However, Samsung Foundry will still lag behind Intel Foundry Services, which plans to start producing chips at its 20A (2nm class) node in 2024 and its 18A (1.8nm class) node in 2024-2025 produce chips.

SAMSUNG FOUNDRY’S 4NM-CLASS PROCESS TECHNOLOGY YIELD IS NOW HIGHER THAN 75%​

Samsung Foundry’s 4 nm-class process technology yield is now higher than 75%. This is a significant improvement from the past, and it shows that Samsung is making steady progress in improving its 4nm chip yield. The “yield” in chip production refers to the number of chips produced on one wafer compared to the maximum number technically possible. A higher yield means that more chips can be produced from a single wafer, which can help to reduce the cost of production and increase profitability.

According to TrendForce data, Samsung’s foundry’s market share in the foundry chip manufacturing market soared from 9.9% in the first quarter to 11.7% in the second quarter of 2023, with revenue reaching $3.234 billion, up from $2.757 billion in the first quarter. While TSMC maintained its dominance, its market share fell to 56.4% with revenue of $15.656 billion.
Conclusion

Samsung’s steady progress in improving its 4nm chip yield is a positive sign for the company. The company is confident that it will beat TSMC’s 4nm production capacity soon, which will help it to win new orders from key clients such as Qualcomm and NVIDIA. Samsung’s investment in its 4nm production factories and the new Pyeongtaek facility will help the company to increase its production capacity and meet the growing demand for its 4nm chips. With a yield of over 75%, Samsung is making significant progress in improving its 4nm chip yield, which will help to reduce the cost of production and increase profitability.

www.gizchina.com

Samsung Says It Will Beat TSMC's 4nm Production Capacity

Samsung's progress in the 4nm chip-making process puts it on par with TSMC. Learn how Samsung aims to surpass TSMC's production capacity
www.gizchina.com www.gizchina.com
 
Daniel Nenni said:
Samsung’s steady progress in improving its 4nm chip yield is a positive sign for the company. The company is confident that it will beat TSMC’s 4nm production capacity soon, which will help it to win new orders from key clients such as Qualcomm and NVIDIA. Samsung’s investment in its 4nm production factories and the new Pyeongtaek facility will help the company to increase its production capacity and meet the growing demand for its 4nm chips. With a yield of over 75%, Samsung is making significant progress in improving its 4nm chip yield, which will help to reduce the cost of production and increase profitability.
Click to expand...
Samsung's 4nm is about tsmc's N6~N7 so I'm not sure how Samsung's 4nm can take over tsmc's N4 orders?
On the other hand, tsmc's N5/N4 family should be already >130k wafers/month (not include Arizona). Samsung will build >130k wafers/month for its 4nm?
 
Maximus said:
Samsung's 4nm is about tsmc's N6~N7 so I'm not sure how Samsung's 4nm can take over tsmc's N4 orders?
On the other hand, tsmc's N5/N4 family should be already >130k wafers/month (not include Arizona). Samsung will build >130k wafers/month for its 4nm?
Click to expand...
I think a bit more nuance is valuable. PPW it is definitely not in the same league given the single digit improvements from 5LPE to 4LPE and how we saw how poorly the re-branded 5LPE QCOM parts did compared to the same SOC on N5P. I don't know if SF4 fixes the shortcomings, but maybe it does. But credit where credit is due, 4LPE is fairly dense. Larger gate pitch than N5 by like 10%ish and a smaller minimum metal pitch. Having a single fin std cell option (even if it isn't denser) should also be a nice goodie for designers. Given how it stacks up I would have to imagine that certain customers might see SF4 as "close enough" or "basically equivalent" for their purposes. With that said I don't think those customers are of the sexy sort like QCOM/MTK/NVIDIA.

TLDR; like intel 4, I think SF4 is kind of hard to classify (funnily enough for the opposite reasons). I suppose that isn't necessarily a bad thing though, as not every node needs to be a carbon copy of TSMC's node. Besides the fact that technological divergence is expected when you have these isolated groups developing nodes, I think it is also essential to drive the industry forward.
 
Fred Chen said:
The more recent 4nm yield reports seem to show lower values: https://www.gizchina.com/2024/02/05/samsung-exynos-2400-chip-manufacturing-process-yield-rate/

"Samsung Exynos 2400 currently has a yield rate of about 60%."

(TSMC N4P is said to be about 70%)
Click to expand...
Yield is king and basis for everything! Intel is well familiar with that at Q10.

10% given the die size makes for a pretty heavy discount one company needs to make to be competitive.

Second it is well know yield is highly correlated to reliability.
 
Fred Chen said:
The more recent 4nm yield reports seem to show lower values: https://www.gizchina.com/2024/02/05/samsung-exynos-2400-chip-manufacturing-process-yield-rate/

"Samsung Exynos 2400 currently has a yield rate of about 60%."

(TSMC N4P is said to be about 70%)
Click to expand...
The main reason for bringing it back here is it's the same author reporting a lower number, seems he forgot his earlier 75%. Probably thinks he's hearing reports for the first time again.

75% isn't even a good number, though the reported TSMC value is around there.
 
Fred Chen said:
The main reason for bringing it back here is it's the same author reporting a lower number, seems he forgot his earlier 75%. Probably thinks he's hearing reports for the first time again.

75% isn't even a good number, though the reported TSMC value is around there.
Click to expand...
The percent yield is tied to die size and defect density as well is inherent process window for systematics.

if you look at year downs of various companies process and dimensions one can make their own conclusions far deeper than just % yield
 
Doesn't seem like leading edge AI players are very price sensitive when what everyone wants is the latest, fastest, most power efficient process. What can Samsung even offer at this point other then bargin bin prices. No fabless design house can afford to fall behind their competitors at this critical time and its hard to see how one would justify going with an inferior process. I feel like I haven't heard of any of the big dogs going with samsung since their 8nm node.
 
Maximus said:
Efficiency of Samsung's 4nm is about on par with tsmc's N6.
So technically the yield/cost comparison should be done with N6. But it would be more miserable...
Click to expand...
Samsung's most recent report outs indicate that SF4 is finally competitive with N4 on a power-performance basis (and it has always been competitive on a density basis). They also seem to have good improvements to variation so assuming any DD issues that might have existed are now sorted out, I think SF4 is finally the real deal. Of course finally being competitive with N5 family four years later isn't exactly what you like to see, and it is so late that it will likely miss out on a lot of HPC and all of the premium mobile SOCs. Value smartphones/industrial/IOT chips might be made on SF4, and presumably Samsung will start shifting their product mix to have more SF4 exynos or QCOM chips to replace older N5 MTK/QCOM SOCs. Which I think is big deal.

Currently Samsung Foundry has significantly less scale/revenue than even intel foundry, and if you want to talk about 7nm class nodes and below the situation gets even worse. As a result I think SF4 family of nodes and the logic only Texas fabs will be one of the linchpins Samsung will need to build the healthy foundation their logic business needs in order to continue to innovate.
 
nghanayem said:
Samsung's most recent report outs indicate that SF4 is finally competitive with N4 on a power-performance basis (and it has always been competitive on a density basis). They also seem to have good improvements to variation so assuming any DD issues that might have existed are now sorted out, I think SF4 is finally the real deal.
Click to expand...
IMG_1286.jpeg

There is still a significant p/w gap between SF4 (Exynos2400) and N4 (8 Gen3/Dimensity9300)based on large core comparison.

Ref:
 
Maximus said:
View attachment 1953
There is still a significant p/w gap between SF4 (Exynos2400) and N4 (8 Gen3/Dimensity9300)based on large core comparison.

Ref:
Click to expand...
It doesn't help that Samsung's SOC design teams are infamously bad. I don't know much about how the different android SOCs stack up with each other, but do the google SOCs on Samsung fair any better since they are only partially designed by LSI? Regardless it is interesting that the results are that poor even with that in mind. This is something worth monitoring to see if SF4P or SF4X chips continue this trend or if they show a big improvement like you would expect from the most recent papers. I am not as in-tune with all of SF's node variants (and the constantly changing names makes it even harder to keep track of). But it would seem that a product that is SF4 based is using an older process from 2022 back when we know Samsung's 4"nm" was not super hot. Conflictingly I don't recall there ever being any products launched as 4LPP. I only remember ever seeing 4LPX (aka 5LP_ but rebranded to 4"nm" so QCOM looked less bad) and 4LPE (real 4"nm"). So maybe what we see on the based SF4/4LPP E2400 is what "fixed" 4"nm" looks like. If that was the case, that would be very sad. TLDR it is unclear to me if the E2400 is on a version of SF4 that is "good" or "crummy" and I want to see what SF4P or SF4X look like before condemning the SF4 family to being eternally bad.

1716832284223.png
 
nghanayem said:
TLDR it is unclear to me if the E2400 is on a version of SF4 that is "good" or "crummy" and I want to see what SF4P or SF4X look like before condemning the SF4 family to being eternally bad.

View attachment 1954
Click to expand...
IMG_6506.jpeg

Samsung official website says Exynos 2400 is made by 3rd gen 4nm/SF4P.
semiconductor.samsung.com

Exynos 2400 | Mobile Processor | Samsung Semiconductor EMEA

The Samsung Exynos 2400 is revolutionizing mobile gaming and on-device AI, offering advanced graphics, powerful performance and enhanced power efficiency.
semiconductor.samsung.com semiconductor.samsung.com
 
Why is samsung LSI seemingly run so incompetently? Does anyone have any vision inside/worked with them can speak on it. I know the PDK's have historically been rough to say the least
 
You must log in or register to reply here.
Back
Top