Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/the-third-star-samsung-electronics-wants-to-dominate-cutting-edge-chipmaking.14815/
        )

    [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
)

The third star Samsung Electronics wants to dominate cutting-edge chipmaking

soAsian

Active member

"The firm has around 15% of the foundry market, compared with more than 50% for TSMC, which plans to spend $100bn over the next three years on new capacity. SE’s non-memory chip revenues make up only 7% of total sales (though that is up from nothing in 2005 and the company also makes some other specialised processors for sensors and the like). The share of profits is even lower."
 
It looks to me (a non-semiconductor professional) that Samsung will be in a tough spot. Here's my two-cents...

1. Memory business MAY shrink in the future
I wonder with the introduction of Apple Silicon (specifically the M1, M1 Pro, M1 Max) to the desktop space will herald a change for laptops and desktops in the future. Given the remarkable 400 Gb/s of memory bandwidth the M1 Max (vs the desktop DD4 memory getting about 18 - 35 Gb/s) has shown and the efficiency benefits of having the CPU and memory together on a single SoC chip, adding Samsung-manufactured DRAM into your desktop may be on its way out in the not too distant future.

2. The NAND Apocalypse (as written in Dylan Patel's great article on SemiAnalysis - The Impending Chinese NAND Apocalypse)
Chinese companies have now joined the memory sector with Yangtze Memory predicted to grow rapidly, flood the market with cheap memory and hurt the margins of ALL the current leaders (Samsung, SK Hynix, Kioxia/WD, Micron). As noted in the Economist article, Samsung's memory business brings brings in nearly half of its operating profits.

Also note the recent plans for expansion plans of Micron ($150 billion over the next decade) and SK Hynix ($106 billion for a fab complex south of Seoul). Lots of spending = lots of product = product glut?

3. Intel Foundry (IDM 2.0)
The other IDM in the room. Intel will pursue the same costumers as Samsung and TSMC. While Samsung's current process node technology APPEARS ahead of Intel at the moment, Intel predicts they will leapfrog Samsung in process node technology in 2022/2023 with Intel 4. Fabless companies may welcome a 3rd player in the game but the truth (according to ASML's CEO) is that as process nodes become more and more expensive....one of the three will eventually drop out. If you have the worst process node (in terms of density and performance ) and poor yields, then you're in going to be in trouble when there's an alternative to TSMC.

4. Apple
The company EVERYBODY wants to do business with. The Kingmaker...and the company Samsung fought for 7 years in an epic patent fight. The company that doesn't seem to forgive or forget their suppliers (see Nvidia) once the bridge is burned. Why would Apple consider using Samsung if their 2nm process node technology in 2025 appears to be roughly equivalent (or slightly worst) than TSMC's current 5nm?!

5. TSMC
No conflicts of interest with costumers. Reliable partner to their costumers. High yields on ALL their process nodes. A generation or two ahead of Samsung Foundry.
 
Agree for most part except #1, Samsung's profits in DRAM mainly come from mobile (LPDDR), server (DDR/HBM) and graphics (GDDR/HBM), not plain ol desktop memory. It's not like M1 devices aren't using LPDDR memory from Samsung/Hynix/Micron so they're not going away. Nothing to replace standard DRAM either in the near term horizon either, and unlike NAND, no threat from Chinese fabs.

#2, most likely will be a glut with all the spending. YMTC's impact will be interesting to see and there's still the matter of the consolidation, possibly between Kioxia and WD.

#3, depends on how much weight we place on their roadmaps, actual dimensions and yield. Intel sure is barking loudly about their goals and execution strategies, but can they really pull it off after wanking off for so long? And how many EUV litho systems will they be able to get their hands on, especially high-NA variants? Seems like Samsung's got their hands on more units than Intel does. We know Samsung has a dedicated EUV line in Hwaseong, does Intel? Just open ended questions, I don't have the answers to them.

#4, I don't think Apple is as shallow as people think, they just go with whoever makes a better product at a better price point at the quantities they need and when they need it. It's business at the end of the day, and Samsung has faltered since somewhere around the 10nm node. They wouldn't consider Samsung if their 2nm process is worse than TSMC's current 5nm, at least not for their leading edge chips. If Samsung made a competitive alternative, had good yields and could delivery the quantities needed on time, why not?

#5, Agreed. Small factors to consider would be how stable electricity generation and water availability will be in Taiwan going forward with their own renewable energy mandates/targets. Always a risk for earthquakes as well. But for most part, as long as they don't stumble, it's their race to win. The only other question is, it is necessarily a good thing for one company to have a monopoly? I'm not comfortable with having all eggs in one basket with ASML/TSMC, but that's just me, just another non semiconductor professional!
 

"The firm has around 15% of the foundry market, compared with more than 50% for TSMC, which plans to spend $100bn over the next three years on new capacity. SE’s non-memory chip revenues make up only 7% of total sales (though that is up from nothing in 2005 and the company also makes some other specialised processors for sensors and the like). The share of profits is even lower."
When Samsung burned Apple, they made a huge mistake that will be hard to recover from if ever. Like Morris Chang has always stated "We don't compete with our customers". Ethics is everything and when Samsung through missteps created the Apple/TSM alliance they made a horrible mistake that they will continue to pay for.
 
That's valid for semiconductor foundry work from Apple. Apple still sources the majority of their screens from Samsung. And half of the RFPCB for iPhones. And camera components and DRAM and NAND, etc. Ethics is important, I absolutely agree. But so is acknowledging one's own arrogance. In that sense, it appears Apple isn't arrogant enough to source components from Samsung as needed, which is a good thing. (y)
 
Agree for most part except #1, Samsung's profits in DRAM mainly come from mobile (LPDDR), server (DDR/HBM) and graphics (GDDR/HBM), not plain ol desktop memory. It's not like M1 devices aren't using LPDDR memory from Samsung/Hynix/Micron so they're not going away. Nothing to replace standard DRAM either in the near term horizon either, and unlike NAND, no threat from Chinese fabs.

#2, most likely will be a glut with all the spending. YMTC's impact will be interesting to see and there's still the matter of the consolidation, possibly between Kioxia and WD.

#3, depends on how much weight we place on their roadmaps, actual dimensions and yield. Intel sure is barking loudly about their goals and execution strategies, but can they really pull it off after wanking off for so long? And how many EUV litho systems will they be able to get their hands on, especially high-NA variants? Seems like Samsung's got their hands on more units than Intel does. We know Samsung has a dedicated EUV line in Hwaseong, does Intel? Just open ended questions, I don't have the answers to them.

#4, I don't think Apple is as shallow as people think, they just go with whoever makes a better product at a better price point at the quantities they need and when they need it. It's business at the end of the day, and Samsung has faltered since somewhere around the 10nm node. They wouldn't consider Samsung if their 2nm process is worse than TSMC's current 5nm, at least not for their leading edge chips. If Samsung made a competitive alternative, had good yields and could delivery the quantities needed on time, why not?

#5, Agreed. Small factors to consider would be how stable electricity generation and water availability will be in Taiwan going forward with their own renewable energy mandates/targets. Always a risk for earthquakes as well. But for most part, as long as they don't stumble, it's their race to win. The only other question is, it is necessarily a good thing for one company to have a monopoly? I'm not comfortable with having all eggs in one basket with ASML/TSMC, but that's just me, just another non semiconductor professional!
Thanks for your points, peter! I wasn't able to find any information regarding the memory built into Apple's M1 devices...so my assumption was that the memory was actually from TSMC rather than from Samsung/Hynix/Micron. My viewpoint partially relies on that to be true. Imagine Samsung no longer able to supply their LPDDR memory in the future to the laptop market if Intel/AMD begin to follow Apple's SoC approach.

Of course, this is just speculation on my part. 😁
 
Thanks for your points, peter! I wasn't able to find any information regarding the memory built into Apple's M1 devices...so my assumption was that the memory was actually from TSMC rather than from Samsung/Hynix/Micron. My viewpoint partially relies on that to be true. Imagine Samsung no longer able to supply their LPDDR memory in the future to the laptop market if Intel/AMD begin to follow Apple's SoC approach.

Of course, this is just speculation on my part. 😁
Just think of all the memory on cell phones, the LPDDR memory is stacked onto the SoC (system on chip) in PoP (package on package) resulting in a SiP (system in package). Fortunately (or unfortunately, depending on your PoV), the LPDDR memory is still made by the big three so I don't see their influence waning anytime soon. TSMC does make some less conventional memory solutions but so far, nothing beyond that. They've expressed interest in acquiring a memory business back when Toshiba was up for sale, but haven't followed through with any of it. At least for now, I would imagine they lack the IP to directly counter the big three. I don't see them directly entering the memory business, just too much upfront investment and space needed to detract from their main business and too cyclical for them to fathom. Not saying it can't happen though!
 
The joker in the sleeve is the monolithic 3D. Samsung is by far a the most ahead of the crowd company when it comes to mono 3d. They have working mono 3D DRAM in the lab, few more radical NAND rethingkings, and of course the had mono 3d logic, and SRAM working in labs at least 10 years ago.

Take a look at just how long their horizon is. Their first playings with OLED happened in nineties, and since then they had ~100 PhDs working non-stop for decades on end until they got a working, serially manufacturable OLED panel.

Now, they are getting more pressure in the OLED space, and QLED space, and now they are already ready to unpack another jocker, which they kept in the sleeve for years.

The same will be in the semiconductor space.
 
To paraphrase Andy Grove, the #2 is the (not so) secret weapon against the #1. In Grove’s case the #2 were Intel’s Taiwanese engineers, the #1 was Japan. Now let’s see if the analogy holds- Korea/Taiwan.
 
Thanks for your points, peter! I wasn't able to find any information regarding the memory built into Apple's M1 devices...so my assumption was that the memory was actually from TSMC rather than from Samsung/Hynix/Micron. My viewpoint partially relies on that to be true. Imagine Samsung no longer able to supply their LPDDR memory in the future to the laptop market if Intel/AMD begin to follow Apple's SoC approach.

Of course, this is just speculation on my part. 😁
So far there isn't any teardown published for Apple M1 Pro and M1 Pro Max yet. But there are 2021 iPad Pro teardowns and 2021 iPhone 13 teardowns available by Tech Insights. For the units they have analyzed, the LPDDR4X memories are made by SK Hynix and storage NAND memories are made by Kioxia (Toshiba spinoff). For Apple's huge volume, it may have additional memory suppliers for different models and regions.

The third URL link below is a 2020 M1 MacBook teardown. Its LPDDR memory is made by SK Hynix and storage SSD is made by Western Digital.



 
Back
Top