Friday, September 24th 2021

Samsung Working on Attracting more Foundry Customers by Improving Customer Structure and Process Node Breakthroughs

Samsung is by far Samsung's largest foundry customers and this is no secret, but now it seems like the company wants to gain more customers to help pay for the costs of operating a cutting edge foundry. A little over a decade ago, Samsung was part of the Common Platform technology alliance together with GlobalFoundries and IBM, which allowed companies to almost pick either foundry based on a common design kit and common process technologies. It made Samsung an attractive foundry option, but the alliance didn't last.

As we know, Nvidia gave Samsung a try with Ampere and there were a lot of reports of yield issues and what not early on. This seems to have persuaded Nvidia to move back to TSMC for Lovelace and Hopper, which is a big loss for Samsung. However, it seems this was also something of a wakeup call for Samsung, as the company is apparently looking at making some internal changes to its customer structure so it can handle third party customers in a better way.
Samsung is ready to start producing chips on its 4 nm node this year, with its 3 nm GAAFET node expected to live in 2023, something that was revealed earlier this year. This is at least six months to a year behind TSMC's 3 nm node which is expected to go into production in the second half of 2022. According to DigiTimes, TSMC already has 60 EUV tools for its 3 nm node in its possession, whereas Samsung in comparison is only expected to have 20 at the most and isn't expected to have more than 30 by the end of this year.

Maybe the biggest issue for Samsung is that its foundry business isn't making money, as the second quarter profits were only US$268 million, with TSMC raking in over US$13 billion at the same time. Even SMIC and UMC are more profitable than Samsung and it's likely that some of this is down to the fact that third parties are wary about using Samsung's foundries due to a wide range of reasons, least not that Samsung is often a direct competitor. Samsung has considered spinning off its foundry business in the past and this might still come to pass, but it's apparently not something the company is considering at the moment.

Samsung has a lot of work ahead of itself to not only stay competitive with TSMC, but also to convince potential customers to choose them over its competitors, even more so now maybe, as Intel is also vying for a slice of the foundry business cake. With TSMC having announced that it'll be investing US$100 billion between 2021 and 2023, in capacity expansion, or roughly twice that of Samsung's sales from its foundry business, it's going to be tough for Samsung to stay competitive without more business. Rumours are that Tesla might be moving from TSMC to Samsung and we also know that Google is working with Samsung on its Tensor SoC. However these are still fairly small business opportunities and Samsung needs some high-volume customers to be able to keep up with TSMC long term.
Source: DigiTimes
Add your own comment

23 Comments on Samsung Working on Attracting more Foundry Customers by Improving Customer Structure and Process Node Breakthroughs

#1
Ferrum Master
You forgot to mention their Exynos + RDNA2 upcoming ambitious venture. And it will be on their 4nm process.

There were some benchmark leaks already, if I ain't wrong.
Posted on Reply
#2
Bruno Vieira
The generous marketing. Samsumg 5nm = TSMC 7nm (if you disagreem please read the Anandtech Galaxy S21(s) reviews compatring), so the upcoming 4nm will probably compete with the older 5nm. The best thing for them is that TSMC cant service the demand. The #2 cant complain right now.
Posted on Reply
#3
TheLostSwede
Ferrum MasterYou forgot to mention their Exynos + RDNA2 upcoming ambitious venture. And it will be on their 4nm process.

There were some benchmark leaks already, if I ain't wrong.
Isn't that a Samsung product though? It's not something they're making for a third party.
Bruno VieiraThe generous marketing. Samsumg 5nm = TSMC 7nm (if you disagreem please read the Anandtech Galaxy S21(s) reviews compatring), so the upcoming 4nm will probably compete with the older 5nm. The best thing for them is that TSMC cant service the demand. The #2 cant complain right now.
The picture is courtesy of DigiTimes and not something I produced.
Samsung's node names aren't really the concern here, bur rather how they are going to be able to attract customers that brings in extra revenue. Samsung manufacturing for Samsung isn't going to bring in sufficient funds for the foundry to keep developing new nodes, which long term means that Samsung's foundry business is either going to take a bigger and bigger chunk of Samsung's profits to stay competitive, or they're going to have to slow down and might even end up in a similar situation as GloFo where they simply stop developing new nodes due to the costs involved.
Posted on Reply
#4
Ferrum Master
TheLostSwedeIsn't that a Samsung product though? It's not something they're making for a third party.
IMHO besides that Exynos, the next gen Qualcomm 898? Is being made also.
Posted on Reply
#5
TheLostSwede
Ferrum MasterIMHO besides that Exynos, the next gen Qualcomm 898? Is being made also.
Right, that might be the case and it would be the kind of business that Samsung needs.
Posted on Reply
#6
Bruno Vieira
TheLostSwedeIsn't that a Samsung product though? It's not something they're making for a third party.


The picture is courtesy of DigiTimes and not something I produced.
Samsung's node names aren't really the concern here, bur rather how they are going to be able to attract customers that brings in extra revenue. Samsung manufacturing for Samsung isn't going to bring in sufficient funds for the foundry to keep developing new nodes, which long term means that Samsung's foundry business is either going to take a bigger and bigger chunk of Samsung's profits to stay competitive, or they're going to have to slow down and might even end up in a similar situation as GloFo where they simply stop developing new nodes due to the costs involved.
I wasn't comment on you, your report is great. I'm sorry if my comment let you think otherwise. The naming shananigans is something even Intel adopted recently.
Posted on Reply
#7
Richards
This is fud from taiwan.. digitimes is partly owned by tsmc's founder so you wont expect good stories from them
Posted on Reply
#8
RandAlThor
Cmon Nvidia, give them another chance and buy some more capacity.
Posted on Reply
#9
TheLostSwede
RichardsThis is fud from taiwan.. digitimes is partly owned by tsmc's founder so you wont expect good stories from them
Which part is fud? Not everything is based on the DigiTimes story, as I'm not stupid enough to believe everything they write without checking up on what I can.
As for revenue etc. those are public figures that Samsung has to publish, so it's not like DigiTimes published some secret insider info in the article.
Considering that there's only one company that makes the machines that are used at these types of nodes, i.e. ASML and the fact that they can only produce so many machines, it's not strange that they've prioritised their best customer. It's also no secret that each node shrink is getting more and more expensive to develop.
So please, enlighten me, what exactly is fud here?
I'm not saying you have to agree with this point of view, but you have to provide a compelling argument to go with your comment.
Posted on Reply
#10
prtskg
I personally think Nvidia is lucky they went with Samsung. Had they went with TSMC, they couldn't have increased their wafers supply easily.
Posted on Reply
#11
TheLostSwede
prtskgI personally think Nvidia is lucky they went with Samsung. Had they went with TSMC, they couldn't have increased their wafers supply easily.
Well, it's kind of a situation where they're seemingly doomed either which way when it comes to supply, as one partner is presumably having manufacturing problems and the other is running at full steam and still can't meet demand from its other customers. So having some supply is obviously better than none, but I'm guessing Samsung isn't making money on Nvidia as they took one more than they could chew.
Posted on Reply
#12
DeathtoGnomes
TheLostSwedeWell, it's kind of a situation where they're seemingly doomed either which way when it comes to supply, as one partner is presumably having manufacturing problems and the other is running at full steam and still can't meet demand from its other customers. So having some supply is obviously better than none, but I'm guessing Samsung isn't making money on Nvidia as they took one more than they could chew.
I'd wager Nvidia wrote the book on how to screw Samsung with that.
Posted on Reply
#13
Richards
TheLostSwedeWhich part is fud? Not everything is based on the DigiTimes story, as I'm not stupid enough to believe everything they write without checking up on what I can.
As for revenue etc. those are public figures that Samsung has to publish, so it's not like DigiTimes published some secret insider info in the article.
Considering that there's only one company that makes the machines that are used at these types of nodes, i.e. ASML and the fact that they can only produce so many machines, it's not strange that they've prioritised their best customer. It's also no secret that each node shrink is getting more and more expensive to develop.
So please, enlighten me, what exactly is fud here?
I'm not saying you have to agree with this point of view, but you have to provide a compelling argument to go with your comment.
Not on this story but digitines always posts fud on Samsung saying clients are gonna leave them for tsmc... they still bitter nvidia picked Samsung for ampere
Posted on Reply
#14
Vayra86
DeathtoGnomesI'd wager Nvidia wrote the book on how to screw Samsung with that.
LOL yeah well, that's the highwaymen robbing the thieves guild then. Its not like Samsung is holier than thou :D
Posted on Reply
#15
TheLostSwede
RichardsNot on this story but digitines always posts fud on Samsung saying clients are gonna leave them for tsmc... they still bitter nvidia picked Samsung for ampere
I don't think I said any such thing, besides Nvidia, which it seems to be fairly public knowledge that they're going back to TSMC.
I'm not sure what you read, but what I wrote about was Samsung's struggle with attracting big bucks customers for their foundry business, which again isn't a secret.
How you went from that to fud is a bit strange...
Posted on Reply
#16
Tardian
Samsung Electronics is in talks with Tesla to make Tesla's next-generation self-driving chips based on Samsung's 7-nanometre chip production process, a South Korean newspaper reported on Thursday. Since the beginning of this year, Tesla and Samsung have discussed chip design multiple times and exchanged chip prototypes for Tesla's upcoming Hardware 4 self-driving computer, the Korea Economic Daily reported, citing sources with direct knowledge of the matter. If Samsung wins the order, it is expected to make the chip on its 7-nanometre production process, the report said. Samsung already produces chips in Tesla's current Hardware 3 computers.
Tesla Chief Executive Elon Musk said in the company's AI Day event in August that Tesla would introduce new hardware for its self-driving computer for its Cybertruck electric pick-up truck in "about a year or so."
In the chip contract manufacturing industry, Samsung is second to TSMC (2330.TW) which had 52.9% of market share compared to Samsung's 17.3% as of end-June, according to analysis provider TrendForce.
www.reuters.com/business/autos-transportation/samsung-talks-with-tesla-make-next-gen-self-driving-chips-korea-economic-daily-2021-09-23/
Interesting times.
Posted on Reply
#17
TechLurker
I could see AMD maybe buying up some capacity, if only to help ramp up Ryzen and future GPU availability as well as provide a back-up should something again happen to TSMC (whether it's flooding, a brownout, or a war with the mainland). Such as APUs and embedded components on Samsung since they're still monolithic, and maybe some lower end SKUs that they don't have enough bins for (entry level Ryzens or "budget" GPUs). Doubly so considering their active partnership on the mobile side of things.
Posted on Reply
#18
zlobby
Ferrum MasterYou forgot to mention their Exynos + RDNA2 upcoming ambitious venture. And it will be on their 4nm process.

There were some benchmark leaks already, if I ain't wrong.
You got sauce for the 4nm part?
Posted on Reply
#19
Tardian
zlobbyYou got sauce for the 4nm part?
TSMC N4: due in 2022, easy to implement
TSMC’s N5 is the so-called long node that is set to be used and evolve for years to come. The next step in N5’s program is called N4, which is expected to enter risk production in Q4 2021 and volume production sometimes in 2022.
TSMC’s 4nm process technology will extend usage of EUV lithography tools, which will allow the firm to reduce masks count and costs further. The company has not disclosed any information about expected improvements, but N4 expected to be good enough to motivate customers to use the technology.
One of the most important aspects of N4 is that it features N5-compatible design rules, SPICE (simulation program with integrated circuit emphasis) models and IP. To that end, it will be very easy for SoC developers with 5nm designs to adopt TSMC’s 4nm technology and even re-use some of the building blocks they already have.
www.techradar.com/au/news/the-future-of-leading-edge-chips-according-to-tsmc-5nm-4nm-3nm-and-beyond

I have no inside knowledge.
Posted on Reply
#20
zlobby
Tardian
I have no inside knowledge.
So, Samsung are going to use TSMC for their Exynos chips?
Posted on Reply
#23
las
RandAlThorCmon Nvidia, give them another chance and buy some more capacity.
Nvidia's plan is probably to use several fabs and several different nodes to max out GPU delivery and flood the market (gain marketshare)

1) Keep releasing RTX 3000 series (and maybe soon, 3000 SUPER series) from Samsung 8nm
2) Maybe re-introduce Turing at 12nm TSMC (RTX 2060 with 12GB rumour + other SKUs. Because everything sells right now. GTX 1660 Super sells for 400+ dollars where I live)
3) Use TSMC 5-6nm for RTX 4000 series in Q3 2022

Nvidia already gained 3% marketshare in 3 months, they now sit at 83% dGPU marketshare.

Intel dGPUs in Q1 2022 targetting AMDs primary segment; Low to Mid-end. If they go with aggressive pricing, AMD will have a hard time on the dGPU market going forward. Intel will want to gain marketshare fast, and they have to sell GPUs at low margins to do this.

AMD used to have a few good value cards every generation - cards that sold well - think RX580,570,480,470 - AMDs best selling GPUs in the last 5 years .. yet 6600 and 6700 series have horrible pricing and availability because AMD downplays GPU output at TSMC and this is one of the reasons why Nvidia gained marketshare fast. AMD 6000 series mostly exists on paper, insanely low output.

All this is the reason why Nvidia went Samsung in the first place. Good decision looking back. If Ampere had been on TSMC 7nm GPU availablity would be hell instead of just poor.
Posted on Reply
Add your own comment