Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/three-tsmc-weaknesses-balancing-its-many-strengths.17407/
        )

    [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
)

Three TSMC Weaknesses (balancing its many strengths)

benb

Well-known member
Youtube creator Marton Barcza (TechAltar & Friday Checkout), building on interviews of Asianometry's Jon and Bloomberg's Tim Culpan, finds that:
-TSMC inverts the usual, lower, value of manufacturing in industry structure
-TSMC collection of leading and trailing edge customers cooperatively maximizes fab investment returns
-92% of < 10nm chips are made at TSMC
And yet has three weaknesses
-Politics: Japan, USA and China want what they have, and may get it
-Messing Up: Gate last and EUV resolved in TSMC favor, perhaps the next thing won't; Morris Chang shoes are hard to fill
-Cost: 28nm was the least expensive cost/gate; in one (rather narrow, I think) scenario this could transform TSMC's massive advantage in leading edge capa into a massive disadvantage.

Watch it here:
 
It also doesn’t match intel’s margins or statements from 45 thru 14nm eras. They claimed they were always getting sizable cost per transistor reductions. The results show that as they moved down to smaller nodes they made better quad cores with much smaller diesizes and saw their gross margins soar. To top it off the intel equivalent nodes were much more complicated than their competition. HKMG last/self aligned vias at 32nm, finFET/SAC at 22nm vs 28nm, and finFET/more advanced lithography at 14nm vs 20/16nm.

The claim that it is the best bang for your buck especially after taking into account the design costs and the fact that nodes like N7 haven’t finished depreciating yet is more plausible.
 
Last edited:
Youtube creator Marton Barcza (TechAltar & Friday Checkout), building on interviews of Asianometry's Jon and Bloomberg's Tim Culpan, finds that:
-TSMC inverts the usual, lower, value of manufacturing in industry structure
-TSMC collection of leading and trailing edge customers cooperatively maximizes fab investment returns
-92% of < 10nm chips are made at TSMC
And yet has three weaknesses
-Politics: Japan, USA and China want what they have, and may get it
-Messing Up: Gate last and EUV resolved in TSMC favor, perhaps the next thing won't; Morris Chang shoes are hard to fill
-Cost: 28nm was the least expensive cost/gate; in one (rather narrow, I think) scenario this could transform TSMC's massive advantage in leading edge capa into a massive disadvantage.

Watch it here:

There are several issues with this YouTube video:

1. In the video the author claimed one of the scenario that can hurt TSMC is that TSMC may spend $20 billion to build a brand new fab but can't find a customer who is willing to place an order. He misunderstood how TSMC's business works. TSMC always finds customers and collects commitments first before building a new fab.

2. When Apple, AMD, Nvidia, Mediatek, or Qualcomm are considering a next generation of leading edge process, they are not just looking at the rising per transistor cost as the author implied. PPA (power, performance and area), competition, market opportunities, revenue growth, and profit potential are all important or more critical than the per transistor cost. The author thought the ever rising per transistor will stop companies adopting new process nodes. This is not true.

3. MediaTek is a spinoff of UMC, not the other way around.

4. Morris Chang, like many TSMC early employees and key staffs, did receive stocks as part of his compensation.

5. Morris Chang retired from TSMC in 2018 and doesn't hold any position ever since. The author thought Morris hasn't passed the responsibility completely to others at TSMC. This is far from the truth.
 
Last edited:
I, for one, only one I guess, was impressed with Marton's analysis, the sources and the cogency.

The thing that struck me wasn't the details. hist78 is right about those. It's the first point, that manufacturing is usually lower value, but TSMC remains, like Wiley Coyote, unaware of this.

The expansion of globalization (edit: fablessness) for decades may have given TSMC a special position, but when this expansion ends, which seems to be ending now, perhaps things will be different.
 
Last edited:
TSMC is arguably not first to nanosheets, so does being first to something really mean anything?
Localized at that point in time it does matter - assuming you being first doesn’t come at the cost of everything else. 14LPP beat 16FFL to market and Samsung snagged a bunch of customers that they wouldn’t have otherwise had. The follow up 7LPP was plagued with throughput issues during it’s early days and customers bailed for N7. In short being first can be a big advantage, but only if the overall package is good. Being first also won’t do you much long-term good if the folowups fall behind the competition.
 
I, for one, only one I guess, was impressed with Marton's analysis, the sources and the cogency.

The thing that struck me wasn't the details. hist78 is right about those. It's the first point, that manufacturing is usually lower value, but TSMC remains, like Wiley Coyote, unaware of this.

The expansion of globalization (edit: fablessness) for decades may have given TSMC a special position, but when this expansion ends, which seems to be ending now, perhaps things will be different.
The video was by no means bad. It was reasonably well researched as well. There were just some things that didn’t add up for me. But that is no different than this very forum where we might agree with many of somebody’s thoughts and opinions, but not all of them.
 
Localized at that point in time it does matter - assuming you being first doesn’t come at the cost of everything else. 14LPP beat 16FFL to market and Samsung snagged a bunch of customers that they wouldn’t have otherwise had. The follow up 7LPP was plagued with throughput issues during it’s early days and customers bailed for N7. In short being first can be a big advantage, but only if the overall package is good. Being first also won’t do you much long-term good if the folowups fall behind the competition.
So we watch eagerly how Samsung and Intel execute their GAA rollout.
 
The origins of TSMC are rarely reported correctly. Morris Chang was not a one man army, there were others that made this possible. And it was not just the Taiwanese government, TSMC would not have happened without the Phillips investment/partnership.

One of Morris Chang's strengths was partnerships and that is what really made TSMC what it is today. Collaboration is a big part of TSMC's secret sauce and that came from Morris. What did not come from Morris is the strong competitive spirit TSMC has today and that is now a critical part of TSMC's dominant market position, that was CC Wei.

One quick example, Morris insisted on calling the first FinFET process 16nm to show respect for the Intel 14nm which was a better process. Then Samsung came out with a comparable process to TSMC and called it 14nm. TSMC then had to explain over and over that their 16nm is competitive with Samsung 14nm. It really was a sales blocker and that is why Samsung 14nm did so well. Now even Intel follows TSMC with FinFET process node naming.

Current CEO CC Wei came up through the ranks, not unlike Morris Chang. CC has a PhD from Yale and worked at TI. In fact he worked for Wally Rhines who worked for Morris Chang. He then worked for ST Micro and Chartered Semiconductor before joining TSMC in 1998. CC Wei is the right CEO for TSMC in this ultra competitive time. No knock against Morris but give credit where credit is due. CC Wei is the CEO for the future of TSMC, my opinion.

You might want to check our book on the history of the fabless industry https://semiwiki.com/books/Fabless 2019 Version PDF.pdf It is a much more accurate account on why we are fabless.
 
One quick example, Morris insisted on calling the first FinFET process 16nm to show respect for the Intel 14nm which was a better process. Then Samsung came out with a comparable process to TSMC and called it 14nm. TSMC then had to explain over and over that their 16nm is competitive with Samsung 14nm. It really was a sales blocker and that is why Samsung 14nm did so well. Now even Intel follows TSMC with FinFET process node naming.
I've always wondered about that. From my understanding 14LPP was the slightest bit denser with a bit worse PPW than 16FFL. It is also my understanding that 14LPP snagged more customers mostly because 16FFL had a slower ramp. Obviously finance folks/Wall Street just see the number and ask about why is TSMC behind. However surely chip designers know better, why would the number matter to them? Nobody is buying 4LPE under the understanding that it is more dense than N5P, so why would the smaller number have gotten Samsung extra business?

As for choosing 16 out of respect for intel, was that something Morris said to explain why their arbitrary number was smaller than Samsung's when investors were howling about it? Or was it something that Morris said when 16FFL was first unveiled?

Current CEO CC Wei came up through the ranks, not unlike Morris Chang. CC has a PhD from Yale and worked at TI. In fact he worked for Wally Rhines who worked for Morris Chang. He them worked for ST Micro and Chartered Semiconductor before joining TSMC in 1998. CC Wei is the right CEO for TSMC in this ultra competitive time. No knock against Morris but give credit where credit is due. CC Wei is the CEO for the future of TSMC, my opinion.
Agreed CC Wei is criminally underrated. Morris wouldn't have handed the reins to him if he didn't think he was ready/better for TSMC.
 
As for choosing 16 out of respect for intel, was that something Morris said to explain why their arbitrary number was smaller than Samsung's when investors were howling about it? Or was it something that Morris said when 16FFL was first unveiled?

It was hotly debated inside TSMC and Morris (engineer at heart) won. And yes Intel jammed this naming convention down TSMC's throat in presentations and such. Remember when Intel said the pureplay business was dead and they could not go where an IDM can go? I guess TSMC was in their rearview mirror.... :ROFLMAO: That was a big motivator in the quick FinFET process hopping TSMC did so be careful what you say, it may come back and bite you in the market share.

I have no idea what will happen with GAA but my guess is that TSMC will again own it like they did FinFETs. FinFETs was a slow start but a big finish for TSMC. It's all about the ecosystem. TSMC has so much push from so many customers and partners it really isn't fair, but then again it is.
 
The origins of TSMC are rarely reported correctly. Morris Chang was not a one man army, there were others that made this possible. And it was not just the Taiwanese government, TSMC would not have happened without the Phillips investment/partnership.

One of Morris Chang's strengths was partnerships and that is what really made TSMC what it is today. Collaboration is a big part of TSMC's secret sauce and that came from Morris. What did not come from Morris is the strong competitive spirit TSMC has today and that is now a critical part of TSMC's dominant market position, that was CC Wei.

One quick example, Morris insisted on calling the first FinFET process 16nm to show respect for the Intel 14nm which was a better process. Then Samsung came out with a comparable process to TSMC and called it 14nm. TSMC then had to explain over and over that their 16nm is competitive with Samsung 14nm. It really was a sales blocker and that is why Samsung 14nm did so well. Now even Intel follows TSMC with FinFET process node naming.

Current CEO CC Wei came up through the ranks, not unlike Morris Chang. CC has a PhD from Yale and worked at TI. In fact he worked for Wally Rhines who worked for Morris Chang. He them worked for ST Micro and Chartered Semiconductor before joining TSMC in 1998. CC Wei is the right CEO for TSMC in this ultra competitive time. No knock against Morris but give credit where credit is due. CC Wei is the CEO for the future of TSMC, my opinion.

You might want to check our book on the history of the fabless industry https://semiwiki.com/books/Fabless 2019 Version PDF.pdf It is a much more accurate account on why we are fabless.

The ecosystem, such as Open Innovation Platform, is critical to TSMC's success. While TSMC holds 40%+ net profit margin, many TSMC's suppliers, partners, and customers are having 20% to 30%+ net profit margin too. Somehow TSMC created a value chain that everyone participated can/may make a lot money.

On the other hand, Intel's major customers (Dell, HP, Lenovo, Asus, Acer) often see their net profit margin hoovering around single digit or low teens. This type of business model and value proposition are questionable.
 
On the other hand, Intel's major customers (Dell, HP, Lenovo, Asus, Acer) often see their net profit margin hoovering around single digit or low teens. This type of business model and value proposition are questionable.
This is universal to every merchant chip vendor. Being an electronics manufacturer has always and will always be a low margin activity. Samsung is the one exception I can think of, but that is because they own the whole supply chain (batteries, displays, processors, logic manufacturing, DRAM, NAND flash, assembly, product design, as well as having a strong brand that allows Samsung to charge a premium). Foxcon Dell and the like have to worry about competitors that have access to the same technology, at the same pricing, and a far lower barrier to entry then say designing a chip or fabbing them. There is also less room for differentiation forcing electronics manufactures to compete on price.
 
Last edited:
Some details of the 14nm/16nm timeline. For 6 months, Samsung LSI's Exynos was the only 14nm Finfet mobile SOC on the market, and this was quite a surprise at the time.

The Samsung S23 announcement this month, in contrast, did not reveal a 3nm Exynos on GAA. In fact, complete opposite: No Exynos, only Qualcomm Snapdragon 8 Gen 2 on TSMC 4nm, everywhere in the world. This is also surprising. Perhaps Samsung 3nm GAA is not truly ready yet in 2023.

From wikipedia: https://en.wikipedia.org/wiki/14_nm_process
----------------------------------
In February 2015, Samsung announced that their flagship smartphones, the Galaxy S6 and S6 Edge, would feature 14 nm Exynos systems on chip (SoCs).[27]

On March 9, 2015, Apple Inc. released the "Early 2015" MacBook and MacBook Pro, which utilized 14 nm Intel processors. Of note is the i7-5557U, which has Intel Iris Graphics 6100 and two cores running at 3.1 GHz, using only 28 watts.[28][29]

On September 25, 2015, Apple Inc. released the iPhone 6S & 6S Plus, which are equipped with "desktop-class" A9 chips[30] that are fabricated in both 14 nm by Samsung and 16 nm by TSMC (Taiwan Semiconductor Manufacturing Company).
 
This is universal to every merchant chip vendors? Being an electronics manufacturer has always and will always be a low margin activity. Samsung is the one exception I can think of, but that is because they own the whole supply chain (batteries, displays, processors, logic manufacturing, DRAM, NAND flash, assembly, product design, as well as having a strong brand that allows Samsung to charge a premium). Foxcon Dell and the like have to worry about competitors that have access to the same technology, at the same pricing, and a far lower barrier to entry then say designing a chip or fabbing them. There is also less room for differentiation forcing electronics manufactures to compete on price.

That's why Apple walked away from Intel to use Apple's M1 and M2. Apple can not only capture more profit but also significantly differentiate Macs form Intel based PCs.
 
Last edited:
I’m curious at which point IFS is scaled enough to make not only profit for Intel but decent margins. Also there is the question of capacity. Utilization will presumed to be tight in boom times because Intel will want to push through as many of its own products through one would think. Where does that leave IFS customers? They have to compete with Intels own designs for capacity allocation id imagine. I’m aware their would be contracts and Intel wouldn’t just screw over its customers but it could delay or refuse to sign additional allocation if it thinks it is in its best interest to allocate future capacity to its own products. If Intel makes more money makes its own chips how could they possibly expect to know how much to allocate to customers? You all are welcome to skewer the hell out of my ramblings as I was just thinking out loud.
 
Back
Top