Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/samsung-and-tsmc-seeking-to-spend-their-way-to-worldwide-domination-of-advanced-ic-technology.13902/
        )

    [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 and TSMC Seeking to Spend Their Way to Worldwide Domination of Advanced IC Technology

Daniel Nenni

Admin
Staff member
Combined, the two companies are forecast to represent 43% of global semiconductor industry capital spending this year.

IC Insights recently released its March Update to the 2021 edition of The McClean Report. In addition to presenting its 2021 IC market forecast revisions by product type and rankings of the top 40 IDM and top 50 fabless IC suppliers in 2020, the Update contains semiconductor industry capital spending forecasts by company for 2021 and the latest global forecast for spending through 2025.

Keeping up with producing leading-edge IC technology has become increasingly expensive over the past 25 years. The investment required to implement the most advanced process technologies for logic devices has now driven out all but three companies—Samsung, TSMC, and Intel, from the leading edge portion of the market. Moreover, of these three manufacturers, only two can truly be considered to be at the leading edge (Samsung and TSMC), with both in volume production of 7nm and 5nm ICs. In contrast, Intel is not expected to be in high volume production of 7nm devices, in its own fabrication facilities, until 2022, at which time Samsung and TSMC are forecast to be producing commercial quantities of ICs using 3nm process technologies.

Historically, the IC companies that had the highest level of capital expenditures were also the companies that were able to produce the most advanced devices. Although Intel has been among the top two leading semiconductor industry capital spenders for 25 out of the past 27 years (Figure 1), the company spent only about half of what Samsung spent in 2020 and is expected to once again fall far short of what both Samsung and TSMC are expected to spend this year (about $28 billion each).

Samsung first spent more than $10.0 billion in semiconductor capex in 2010. After spending $11.3 billion in semiconductor capex in 2016, Samsung’s 2017 outlays for the semiconductor group more than doubled to $24.2 billion.

Samsung’s semiconductor capital spending has remained very strong since 2017 with outlays reaching $21.6 billion in 2018, $19.3 billion in 2019, and a massive $28.1 billion last year. The sheer magnitude of Samsung’s spending over the 2017-2020 timeperiod ($93.2 billion) is unprecedented in the history of the semiconductor industry! At $93.2 billion, this amount was more than double the $44.7 billion spent by all the indigenous China semiconductor suppliers combined over this same timeframe. Although Samsung has not provided guidance for its 2021 outlays, IC Insights estimates that the company will keep its spending essentially flat with 2020.

Top Semiconductor Spending Leaders.png


Figure 1

TSMC is the only pure-play foundry to offer leading-edge technology. It is seeing very strong demand for its 7nm and 5nm processes, which represented 47% of the company’s sales in 2H20. Most of its current investment is targeting additional capacity for its 7nm and 5nm technologies. Illustrating how quickly TSMC moved to more advanced processes, the company’s 5nm products represented 8% of its total 2020 sales ($3.5 billion), after essentially having no 5nm revenue in the first half of last year.

On January 14, 2021, TSMC dropped its “bombshell” news that it plans to ramp up its capital spending this year to $25-$28 billion, a 60% increase using IC Insights’ expectation for $27.5 billion in outlays by the company. At $27.5 billion, the company’s average quarterly capital spending rate this year would be $6.9 billion, more than double what the company spent in 4Q20.

It now appears that both Samsung and TSMC realize the golden opportunity that is currently before them. While Samsung began its spending surge in 2017, TSMC will begin what is likely to be a huge multi-year ramp of spending in 2021. Combined, IC Insights expects Samsung and TSMC’s capital expenditures will reach at least $55.5 billion this year and represent an all-time high percentage of total semiconductor industry outlays held by the top two spenders (Figure 2). With no other companies presently able to match these huge spending sums, Samsung and TSMC will likely put even more distance between themselves and their competition this year with regard to advanced IC manufacturing technology.

Top CAPEX Semiconductor Spends.png


Figure 2

Can governments like the EU, U.S., and China invest in their indigenous IC industries and catch up in the IC technology race with Samsung and TSMC? Considering how far behind they are, IC Insights believes that governments would need to spend at least $30 billion per year for a minimum of five years to have any reasonable chance of success. Is there the willingness and/or ability to follow through on such a commitment? Moreover, for China, even if the money were available, they would certainly be hindered by trade issues prohibiting some of the most critical pieces of process equipment from being sold into the country.

Without extremely quick and decisive action by other IC producers or governments, Samsung and TSMC are well on their way to world domination of leading edge IC process technology, the cornerstone of all of the advanced consumer, business, and military electronic systems of the future.

Report Details: The 2021 McClean Report
The 2021 edition of The McClean Report—A Complete Analysis and Forecast of the Integrated Circuit Industry was released in January 2021. A subscription to The McClean Report includes free monthly updates from March through November (including a 180+ page Mid-Year Update), and free access to subscriber-only pre-recorded webcasts through November. An individual user license to the 2021 edition of The McClean Report is available for $5,390 and a multi-user worldwide corporate license is available for $8,590. The Internet access password and the information accessible to download will be available through November 2021.
https://www.icinsights.com/services/mcclean-report/pricing-order-forms

PDF Version of This Bulletin
A PDF version of this Research Bulletin can be downloaded from our website at http://www.icinsights.com/news/bulletins/
 
Samsung CAPEX spend includes memory. What % is memory though? And how much of the China CAPEX spend is in use? There are a lot of idle fabs in China. It seems to me that there may be a glut coming after the pandemic hording subsides.
 
Samsung dominates memory and its a BIG reason why they can keep matching TSMC's steps. For Samsung, (in my opinion) the big questions are:
1. Can Samsung Foundry catch up/surpass TSMC in process technology? If not, can they have continued success in capturing TSMC's clients (like Nvidia & Qualcomm) to make profit?
2. Will Intel start outsourcing some of their products to Samsung?
3. Can they continue to hold back the Chinese upstarts (particularly YMTC for memory, BOE for display, and Huawei/Xiaomi/Oppo for smartphones)? If Samsung losses market share/profits, can their Foundry business continue to be competitive?

For TSMC, its big questions are:
1. Can they continue to hold off Samsung/Intel for the lead on process technology? At what point will Apple decide that fast is "fast enough" and stop funding TSMC's new processes?
2. Will Intel outsource (more) to TSMC?
3. Will they consider building a new fab in the EU or Japan? I honestly would be surprised if TSMC does NOT decide to eventually build a fab in the EU given that it's under "preliminary investigation" by the EU for "anti-competitive practices". A TSMC fab in Europe would most likely placate EU officials' antitrust concerns by alleviating their national security concerns. I imagine that the EU would STRONGLY prefer a TSMC fab over a Samsung fab given the technology lead and that Samsung may also be a competitor to many European fabless firms.
 
Samsung dominates memory and its a BIG reason why they can keep matching TSMC's steps. For Samsung, (in my opinion) the big questions are:
1. Can Samsung Foundry catch up/surpass TSMC in process technology? If not, can they have continued success in capturing TSMC's clients (like Nvidia & Qualcomm) to make profit?

Samsung has been in the foundry business for a long time. Certainly before the iPhone came to town. The culture is just different. I worked with them for many years on various projects, same with TSMC, UMC, SMIC and GF. The foundry business really is all about collaboration and the customer experience and TSMC is hard to beat. Right now Samsung Foundry's biggest market niche is that they are not TSMC. Second is low cost and those two market segments are very difficult to scale profitably.

Also know that Nvidia and QCOM use both Samsung and TSMC so they did not really capture them. Nvidia and QCOM are playing foundries against each other for best pricing.

Last point is that the foundry business is changing. System companies (Apple, Amazon, Huawai, etc... ) dominate the semiconductor industry and they will not use Samsung due to competitive reasons. Samsung is one of the largest systems companies in the world and they compete with just about everyone.

Sound reasonable?
 
For TSMC, its big questions are:
1. Can they continue to hold off Samsung/Intel for the lead on process technology? At what point will Apple decide that fast is "fast enough" and stop funding TSMC's new processes?

Good question. So far the answer is yes and I think as long as Apple has iProduct competition the answer will continue to be yes. I think TSMC adding AMD and Intel to their inner circle also strengthens TSMC's position with Apple. Apple also has a huge semiconductor ego to feed.

 
Memory and advanced logic foundry processes are *very* different, with limited commonality between them. Samsung is certainly the world's biggest spender on memory technology which is why it's #1 on the semiconductor spenders list, but I suspect TSMC has been spending more than Samsung (and Intel...) on logic process technology for some time, and certainly will be in the near future after their recent big announcement.

Which is why they're leading the technology race right now. Yes Samsung have announced 3nm GAA first but analysis shows their process isn't any better than TSMCs 3nm FinFET, if anything it's a bit worse, and this has been the same story for several years now e.g. Samsung announced EUV first but TSMC won the race to get it into high-volume high-yield mass production and with higher density and lower power.
 
2. Will Intel outsource (more) to TSMC?

I certainly hope so. It really will be a test of strength. Intel and AMD will be head to head on a level playing field at the process level so it will be all about design. Intel manufacturing will be head to head with TSMC, nothing like some internal competition to boost innovation on both sides.
 
For TSMC, its big questions are:
3. Will they consider building a new fab in the EU or Japan? I honestly would be surprised if TSMC does NOT decide to eventually build a fab in the EU given that it's under "preliminary investigation" by the EU for "anti-competitive practices". A TSMC fab in Europe would most likely placate EU officials' antitrust concerns by alleviating their national security concerns. I imagine that the EU would STRONGLY prefer a TSMC fab over a Samsung fab given the technology lead and that Samsung may also be a competitor to many European fabless firms.

Yes, I'm sure they will, but only with government funding like TSMC gets in Taiwan. But remember, these are token political gestures, seedlings that can grow if the climate is right. The TSMC fabs in China are not growing. The TSMC fab in AZ has a very good chance of growing. It all depends on how the host country collaborates politically and financially with TSMC.
 
I certainly hope so. It really will be a test of strength. Intel and AMD will be head to head on a level playing field at the process level so it will be all about design. Intel manufacturing will be head to head with TSMC, nothing like some internal competition to boost innovation on both sides.
Do you think TSMC really want to Open the leading edge for Intel?TSMC has more way more orders then they possible can produce. Personally i think yes they increase orders at TSMC for grafics but not CPUs
 
Do you think TSMC really want to Open the leading edge for Intel?TSMC has more way more orders then they possible can produce. Personally i think yes they increase orders at TSMC for grafics but not CPUs

TSMC upped CAPEX for Intel this year ($28B). Remember, it takes longer for Intel to design a chip than it does for TSMC to build a fab. Hopefully Intel will build the client side chips at TSMC and save the server business for Intel fabs.
 
The sabras and the helgas don't work with each other.

I expect more outsourcing to tsmc, more fake news from mike rogoway and more broken promises from pat gelsinger. pat gelsinger says stuff that was declared not to happen before he says it.
 
TSMC upped CAPEX for Intel this year ($28B). Remember, it takes longer for Intel to design a chip than it does for TSMC to build a fab. Hopefully Intel will build the client side chips at TSMC and save the server business for Intel fabs.
Yes the increased CAPEX but i dont see the link dircretly to Intel. Intel knew they have a new CEO in February. Before there was no decision Made. So TSMC increased for Intel before and descion was Made at Intel? Pat is a IDM person. Ist sure for shareholders they anounce more TSMC, but just for grafics and other stuff.

on the Design Point view, Intel has exactly one Core Design for all CPUs. They sure can not Dual source this design...
 
Yes the increased CAPEX but i dont see the link dircretly to Intel. Intel knew they have a new CEO in February. Before there was no decision Made. So TSMC increased for Intel before and descion was Made at Intel? Pat is a IDM person. Ist sure for shareholders they anounce more TSMC, but just for grafics and other stuff.

on the Design Point view, Intel has exactly one Core Design for all CPUs. They sure can not Dual source this design...
That shipped sailed at the time of 14nm
 
Intel has a big problem on their hands either way.

If they go more to TSMC, for what product and what process? They can of course outsource more of their "other" stuff, like GPUs, FPGA, Ponte ponte vecchio, but the real question is if they outsource their main buisiness CPUs. And here it gets tricky. Both of design standpoint and process standpoint. Let met explain.

Currently Intel design one core design (Skylake, Sonnycove, Willow etc) for their own process 14nm, 10nm, 7nm. If they go to TSMC they need to basically design different Cores because they can not really outsource everything to TSMC. What product would be sutible for TSMC? Of course no low end stuff. Laptop? No really because high volume. Desktop? Nope. Entprise? Yes, but they need not only a good process, they need also a good interconnect (EMIB, Infinity Fabric) to go Chiplets.

Then comes to question if TSMC is willing to give Intel access to leading edge. If Intel wants to continue to develop their own leading edge process (7nm, 5nm, 3nm) i personally see no way that TSMC opens their secrets to a competitor. If Intel abandom their leading edge, the CAPAX increase from TSMC is way to low.

Pat Galsinger is totally aware of the strengh of having their own fabs. I see only a minor increase at TSMC.
 
The decision to outsource to tsmc was already made.

Pat galsinger is a poor leader that will ruin his subordinates lives as well as their families.
 
The decision to outsource to tsmc was already made.

Pat galsinger is a poor leader that will ruin his subordinates lives as well as their families.
Yes the decision was made to Outsource for sure...but for what?
I think we dont see any CPU designs at TSMC
 
Apple, amd, and mediatek uses tsmc for cpus so it's done all.the time.
 
The problem Intel have with the big Xeon server CPUs (Intels' cash cow) is that they're huge monolithic die (up to ~600mm2?) for which a process with low defect density is essential to get any yield at all. This is one of the killer reasons they haven't been able to switch over to their 10nm process, the yields are OK for smaller die but not for big ones (the Intel 14nm process is now very mature so yield isn't a problem).

The only way to get process yield up is to shift massive numbers of wafers through the fabs, which TSMC always do and Intel have done over the years with 14nm -- but not with 10nm yet (and maybe never will...).

So they can't move the big Xeons to their own 10nm process yet (maybe never?), and it's unlikely that their own 7nm process (if it ever goes to mass production) will have good enough big-die yield for some time (if ever?).

If Intel move to TSMC and keep the monolithic Xeon die then they will get usable yield, but it'll still be a lot lower than AMD's chiplet approach with a ~100mm2 8-core CPU, and it's no longer in their inhouse fab where they can hide the cost (and tweak the process) so they'll end up with maybe 2x cost for the same number of cores in the same process compared to AMD.

Or they could also go to chiplets, but this is a massive architecture change because CPUs need to be designed for this approach from the start, it can't be bolted on afterwards, and they're years behind AMD with a chiplet-ready architecture.

It's difficult to see how Intel can dig themselves out of this hole...
 
The problem Intel have with the big Xeon server CPUs (Intels' cash cow) is that they're huge monolithic die (up to ~600mm2?) for which a process with low defect density is essential to get any yield at all. This is one of the killer reasons they haven't been able to switch over to their 10nm process, the yields are OK for smaller die but not for big ones (the Intel 14nm process is now very mature so yield isn't a problem).

The only way to get process yield up is to shift massive numbers of wafers through the fabs, which TSMC always do and Intel have done over the years with 14nm -- but not with 10nm yet (and maybe never will...).

So they can't move the big Xeons to their own 10nm process yet (maybe never?), and it's unlikely that their own 7nm process (if it ever goes to mass production) will have good enough big-die yield for some time (if ever?).

If Intel move to TSMC and keep the monolithic Xeon die then they will get usable yield, but it'll still be a lot lower than AMD's chiplet approach with a ~100mm2 8-core CPU, and it's no longer in their inhouse fab where they can hide the cost (and tweak the process) so they'll end up with maybe 2x cost for the same number of cores in the same process compared to AMD.

Or they could also go to chiplets, but this is a massive architecture change because CPUs need to be designed for this approach from the start, it can't be bolted on afterwards, and they're years behind AMD with a chiplet-ready architecture.

It's difficult to see how Intel can dig themselves out of this hole...
Totally agree
Sapphire Rapids already uses 4xDies without an IO die. Similuar to the first epyc AMD Rome. Interesting times for intel and we should get an update next week from Intel CEO about the process going forward
 
Back
Top