They depend on ASML lithography machines, right? Intel, Samsung and other foundries could/do also buy ASML's machines, so what gives TSMC their advantage?
Semiconductor industry has the deepest and most complex value chain in the economy.
TSMC has always been pure fab company and it pays off now. Samsung and Intel are also in chip design business and do other things. They are less focused. TSMC's lead is not guaranteed. Any new technology node can fail and remove them from competition if they make wrong choices.
I'm not exaggerating when I say that designing a new technology node for mass production is like a moon program. TSMC designs new node every few years, builds a fabs that cost 20+ billion. ASML machines are the most expensive tooling there but they are just part of the whole.
Intel and Samsung are almost as good, but small differences have huge impact in the final result and timing. Small percentages in final yield can make a difference.
A bit of an aside, but your comment made me realize how much we don’t know about the process used to create the technologies that underpin our civilization.
We (laymen) have no idea about the incredible complexity behind manufacturing a single CPU chip.
> we don’t know about the process used to create the technologies that underpin our civilization.
It's been like that for a while. That's why we need everyone else to function as a society.
I've watched a couple blacksmithing videos on youtube but I still don't really know how smelting iron works, and I wouldn't be able to produce a steel if my life depended on it. That's literally 2500+ year old tech.
It was science back then, it is science now. Using technologies at the time. 2000 years from now, some people will look at our era as barbarian and unscientific.
Many years ago I sat next to the CTO of Broadcom on a plane. We struck up a conversation and when I learned he worked for a chip designer I asked the question that I assumed was the most difficult part. I asked how the engineers laid out the circuits physically on the chip. He laughed and told me the engineers write software that defines the parameters and functions and the software lays out the chip (I know I am oversimplifying). I was dumbfounded to learn that hardware design was really software development. Today I think Google has an open chip design effort where you write chip specs in Python (from memory).
They even have their own languages for coding the chip logic ( Verilog is one).
My partner did chip design and layout. The used a tool call magma (We have some mugs from a conference she went to) but they got bought by synopsis. Very expensive, very specialized software
This was probably not true many years ago, or at least was a simplification. Even with some automated routing, there are many subtleties like high frequency electronics; not trying to have your radio equipment routing through your power supply, etc.
It is a mix: digital logic is almost entirely laid out by machine, but analog circuits and the underlying logic cells (as well as certain specialised digital blocks like memories) are a lot less automated. The most expensive part isn't doing that: it's doing the validation and simulation of the design so that it'll have the best chance of working after you've spend 5-6 figures and many months on actually building it (and then there's the cost of testing it in reality and figuring out what isn't actually working and how to fix it: the first batch of chips still extremely rarely works).
Most people don't even know where their water comes from. Most people that drive do not know how a combustion engine works. Even people watching TV are only vaguely aware it's a series of static images chained together to make it look like its moving.
People in general have a barely functional understanding of computing devices, let alone their operating systems or hardware components, at an usage level.
Their fabrication process is so far removed that I don't think most people would even be able to know where to look for the information, even if they could access and understand all of it.
That also reminded me of _Connections_, which notably made the same point in _the late 1970s_! It isn't like technology's gotten any easier to understand -- in total, "top to bottom" -- since then.
What is a node in this context and what does designing a new node entail? Does this design process mean TSMC can do something with ASML machines that another company with the same machine couldn't do?
The important thing to understand is that at the level of chip manufacturing, you’re doing advanced materials science. These aren’t just abstract logic gates and structures built out of them, such as SRAMs. At the manufacturing level, the logic gates are nano-scale 3D structures made of different materials and in different shapes. For example, at this scale, the shape of the transistor gates has a big effect on performance: https://en.wikipedia.org/wiki/Fin_field-effect_transistor. That materials science is highly guarded secret sauce in companies like TSMC. NVIDIA and Apple will tell TSMC what logic gates go where (or often at an even higher level—specifying certain SRAM blocks going here or there). But TSMC has to actually build the nano-scale material structures that comprise those logic gates.
The ASML machines perform the actual photolithography. They expose the die mask to the photoresist coated silicon wafer. Using that technique, you can built up complex 3D materials incorporating different layers and shapes. But the ASML machine doesn’t know how to make a transistor. It’s kind of like a 3D printer in that way.
> NVIDIA and Apple will tell TSMC what logic gates go where (or often at an even higher level—specifying certain SRAM blocks going here or there).
That's not quite how design works; it's much much more detailed. At the end of the day, TSMC's customers send them a "GDS" file that is a complete physical representation of the die they want manufactured. It describes every top-to-bottom later of the manufacturing process. (TSMC will take the GDS layers and split them up or combine them or do other operations, depending on some process details, but the customer will also check and sign off on that).
It's not just telling TSMC where to put standard cells or SRAM. And while TSMC does offer their own standard cells and SRAM, customers can and often do design and use their own. Analog/RF design is even more detailed, since that's done at an individual transistor level.
TSMC is mainly in the business of design and selling a manufacturing process for making transistors not logic gates.
I was under the impression that it’s unusual for fabless chipmakers to do transistor level design anymore, and that they mostly rely on the standard TSMC libraries. But I may be wrong about that. The critical point I was trying to make is that the ASML machine doesn’t know how to make transistors. That’s something TSMC does: https://www.anandtech.com/show/16041/where-are-my-gaafets-ts...
No that's definitely not the case. For digital design, in general, you're not doing anything transistor-level. But those digital designers are using standard cell or memory libraries which are created at a transistor level. I believe the foundries also charge for their own standard cell libraries, and there are many other vendors that a design company can choose from like cadence, synopsys, even ARM.
For analog/RF design, of which there is still an enormous amount, it's always transistor level.
Theoretically, yes, but in practice, no. Creating a manufacturing process is a very complex and time consuming process that takes a huge about of research and development, investment, etc to achieve profitable yields. For bulk CMOS processes, which is what digital circuits are made on, this is especially the case. Foundries like TSMC can't afford to let customers design custom transistors, and customers generally don't have the expertise to do so. There's a huge economies of scale benefit from having customers all use the same process.
That being said, there is a little room for tweaking. The manufacturing process has a ton of variation in it, and the center point / average of that variation can be moved around a little. I think most companies just take what they get, but I'm betting the big players (Qualcomm, Apple, Nvidia, AMD) all do internal tracking of the process variation when they get product back, and give feedback to the foundries to make changes to optimize their own yield.
There's also been a recent push in the very new and advanced processes for "Design-Technology Co-Optimization", where the digital circuit design (i.e. standard cells and memory) and the process technology design happens together. We got here because all of the low hanging fruit has been picked and now companies are chasing single digit percentages in yield and PPA (power, performance, and area) improvements. It's a collaboration between foundry and customer that happens before the process is even released, so again - big players only, and it's still more tweaking than custom transistors.
For other types of processes, meaning non-bulk CMOS, customers can definitely design their own transistors. This is especially common in RF applications where you're often making chips with a few transistors. In some cases, it can be just changes in the shapes and/or dimensions of the transistors in the GDS, relative to what the foundry recommends. For this type of custom transistor, the foundry still controls the material science details of how the process happens. In other cases, though, customers are in control of everything from the transistor dimensions to the chemical concentrations and methods used for fabrication.
So in that case there is no materials secret sauce guarded by TSMC, with the customer either submitting the GDS file with all the info, or signing off on the modified version.
Sorry my comment wasn't very clear about this. There's still secret sauce in the materials. The GDS is purely dimensional. It's just a bunch of shapes. But how those shapes get translated into actual silicon, and how various chemicals are used and in what concentrations to manipulate the silicon to make the transistors function is the secret sauce.
It's also important to note how crazy the ASML EUV machines are, they each take a megawatt of power to produce about 100 watts of EUV. The infrastructure just to run these machines is staggering. TSMC uses about 5% of Taiwans total power I believe.
I like you 3d printing reference. You can have 2 people with the same printer trying to print the same part. What else goes into getting a good part? Proper bed leveling, bed heating, choice of material, part cooler, nozzle temperature, feed rate, speed, acceleration, layer thickness, infill pattern, and many other settings and choices. And that's just an off the shelf printer melting plastic...
Ya this analogy really clicked with me. To really torture the analogy, Apple or any other TSMC customers hands them an .obj of their chip, and TSMC acts as the slicer and converts that into the GCode that ASML’s “printers” understand. And just like how some slicers have better overhang and infill algorithms TSMC has their own secret sauce for telling the ASML machines what to “print”.
Does that expansion of the analogy work or did I just move the analogy further from the truth?
It's a good analogy but there's one more major aspect. The silicon wafer doesn't sit under a single "printer" the whole time. It has to be moved around between countless different machines all doing different specialized tasks. So there is a huge logistical challenge as well. It is like combining 3D printing with an international airline.
This might be pedantic nit-picking, but: I think bed leveling/heating, nozzle temperature, choice of material, etc. are parameters that the printer manufacturer should have optimized (and, IME, good ones do). I think the end user's chief responsibility is slicing (infill patterns, layer thickness, etc. as you noted), indeed it is in this sense the end user can be said to be in the same position as Intel/Samsung, not so much the hardware maintenance but knowing the slicing tricks for getting complicated geometry to come out just right.
For example, when you're making a cube, the sharp ends are places where bad things happen. When making sharp movements, the nozzle will tend to leave ugly trails and in other times cause warping. So here you can do a trick to save yourself: mouse-ears (extra material around the important edge, so that the bad artifacts happen instead on additionally-created non-core-geometry). At ground level, you use brims.
The interview that I posted above discusses copper interconnects that IBM research introduced.
TSMC had previously used a "spin-on" dialectic technique at a previous node, and reverted to CVD because of problems.
They were able to beat all other manufacturers to market with copper interconnects (including IBM), because they avoided spin-on, which worked well in testing, but not in production.
They had great luck in gaining this prior experience.
>I'm not exaggerating when I say that designing a new technology node for mass production is like a moon program.
It's insane. I had a friend tell me about what goes on in these asml machines. Essentially the targeting system that moves the wafer around cannot have any vibration so I kid you not the platform is floating and controlled by magnets. And this is just the etching machine.
The etching process that uses a laser to blast a moving molten tin sphere to shape it so the next laser blast that vaporizes it produces parallel EUV light is pretty insane, too. See https://www.youtube.com/watch?v=5Ge2RcvDlgw
I worked on a campus where they were building prototype manufacturing processes with these tools.
The building those tools were housed in had a foundation that was iirc 30’ of a specialized concrete mix. I was chatting with some of the construction engineers over coffee and the consensus was that 10,000 years from now, some archeologists would be pondering wtf this giant concrete platform was for.
I highly recommend browsing through the videos in Asianometrys channel if you want easily digestible summaries on this topic: https://youtube.com/@Asianometry
A recent interview with Shang-Yi Chiang, former Vice President of R&D at TSMC (also held positions at TI, HP, and SMIC) had insightful commentary on the speed of bringing up a new node.
"We all take two years to develop one generation, how come you guys can do it in one or one-and-a-half year?" And they asked if some of your customer transfer technology to you or what not? And I told him, "No," I told him that, "That's not true." I think he probably implied we steal technology from customer, the way he talk.
And I say, "I'll tell you why." I said that, "When we develop one node, basically you have some learning cycles. First, you do some simulation. And you have some idea, then you run wafers to prove that. So, you run a group of wafers according to simulation and you have some splits. The wafer runs through the fab, they come out and you measure them, you analyze them, and you try to improve and you run this again. This again, you run. So, this is learning cycle." At that time, "It takes about six learning cycle, roughly, to complete one generation." Of course, you had some short loops and not just one. I said that, "My R&D wafer in the fab run much faster than yours, because my R&D engineer works three shifts and you only work one shift. So, your R&D wafer move eight hours a day, my work/move 24-hours a day. So, my wafers go three times faster, even if you are twice smarter than me, I still beat you up." <laughter>
> my R&D engineer works three shifts and you only work one shift
> All RnD fabs move wafers 24/7
Are you talking about the same thing? It's one thing to let an experiment run overnight with three technician shifts. It's another thing to have three research shifts.
It could also be a metaphor for doing more research than the competition on how to best use the same equipment, so I wouldn't get too hung up on it.
> Why did this “3x shift advantage” deliver gains only in the past 5 years?
Intel, and the US, had an enormous head start. Gordon Moore started at Shockley, which was founded in 1955. William Shockley was awarded the Nobel Prize in 1956 for inventing the solid state transistor. Moore then went to go help found Fairchild Semiconductor in 1957, which invented the CMOS process. Intel was founded in 1968. TSMC wasn't founded until 1987.
In 1968 when Intel was founded, Taiwan was a poor country with little capital. Its GDP per capita was around $300. Taiwan's GDP per capita in 1987 was $5,300 in today's money. The US's was almost four times higher at $20,000.
However, in the end stages of RCA, they engaged in a semiconductor technology sharing agreement with a consortium of Taiwanese companies (that did not include TSMC, which did not yet exist).
At the end of this agreement, UMC had better technology at higher yields than RCA.
This technology transfer had a profound impact upon Taiwan.
"Taiwan managed to persuade RCA to agree in 1976 to transfer semiconductor technology."
I’m pretty sure the big players like Intel and Samsung put their r&d eggs into the G450C coalition to build bigger wafers. TSMC saw that as a threat, did not join and invested more on its own. In the meantime, the 450 thing collapsed for a variety of reasons and the billions of dollars invested went up in smoke and left Intel and other behind.
It's not the whole story, sure, things are complex... but I think you calling it "made up" is a bit absurd. I'd attribute most of the lack of competition with Intel to a lack of capital; which, thanks to the success of TSMC's largest customers (who directly compete with Intel) Apple and AMD is no longer a problem.
Intel did have a huge advantage on everyone else 20 years ago, but ten years ago they more or less started sitting on their asses, and the rest of the world caught up and is passing them by. I don't know why something simple like running 24/7 R&D would not explain it.
TSMC's decision to use "Black Diamond" CVD was driven by previous failures of "spin-on" dielectric that allowed them to deliver copper interconnects before anyone else.
However, the R&D cycle, which failed for "spin-on," otherwise allowed TSMC to deliver faster, as they ran in multiple shifts, allowing accelerated focus on the new node.
The others might move wafers 24/7 but they don't have three shifts of R&D scientists and engineers working to continuously analyze the results/improve yield/performance 24/7. TSMC does.
Manufacturing equipment isn't plug-and-play like an office laser printer. It's more like a paint brush and you're responsible for the results. It seems they spend more time learning how to get the most out of their tools.
Even if they stole technology they'd need a time machine to be where they're at now since they're ahead of others. I wouldn't put it past Intel to imply they stole from their upcoming node but, if so, it's not like Intel has that out yet...
3 separate shifts iirc. But don't Intel run 24/7 too? Or is that just for prod? I remember some plant (not Intel, a memory plant) had an issue and lost power for a few hours and lost a shite load of memory wafers and took a couple days to come back into prod.
Intel certainly do run machines 24/7 but perhaps those R&D engineers/PhD's don't, so maybe only the techie's running the machines have full coverage...?
As pure designers yeah but Chinese government is about to do to chips what they did to solar and batteries so I wouldn't count them out yet. Might be 5-10 years before they catch up but they are by no means out of the fight.
Not only do they currently have the largest share of current technologies they are going to be first to market with next-gen batteries too with CATL and BYD both entering mass production of sodium-ion batteries this year. These have lower energy density for now but that will likely chain as improvements land.
Japan and Korea dropped the ball is what happened, it wasn't any singular error but many unforced errors over a good 10 year period that led to them being now substantially behind.
You mean Chinese battery companies have the largest Lithium battery market share in China where foreign competitors aren't allowed to compete? Last I checked CATL makes mostly NCM, largely based on LG Chem's tech, and then LFP.
>> Japan and Korea dropped the ball is what happened, it wasn't any singular error but many unforced errors over a good 10 year period that led to them being now substantially behind.<<<
You don't seem to understand the battery market. Japan and South Korea have been in the battery business at lease since the mid 1990's. Japan's manufacturing has declined over time, but LG Chem is the largest EV battery manufacturer in the world. Of course, that's the global market share, outside China -- ie, there won't be any Chinese battery in North America.
Considering the overall investment goes into dozens of billions, it seems a little hard to believe they wouldn't source worldwide talent to make it 24/7 as well. Also they probably source top talent. I'm always skeptical of those "we just work a little harder" explanations.
Except in manufacturing, getting your yield up and tuning a process is basically 'just work'. You define a space of parameters you want to explore and then you brute force it by building samples at each parameter step and then analyzing it.
This can be pipelined and parallelized to some extent, but then you have to convince enough PhD level employees to do night shifts, because each process step is basically a miniature physics or chemistry experiment that has to be monitored and tuned constantly (at this stage).
It's one thing to make one perfect transistor, it's a totally different ballgame to make 10 billion perfect devices with better than 90% tool uptime, and an essential component of closing that gap is brute force experimentation.
I just think that if TSMC can find such PhD level employees in Taiwan, why couldn't other competitors find them in places way more abundant in those types of people? Convincing them is a matter of paying them more, there are plenty of highly educated physicians who work night shifts. And it's not like TSMC could hide this secret sauce either, so you would think that before dumping another 20 billion in a new investment, Intel would consider simply replicating what they could easily observe from TSMC.
I remember reading something that TMC has links with the local universities to train talent, and nearly guarantee them a job in the end... Ireland has an Intel Fab here, and I don't think there are many courses around here doing chip development... Mind you, not sure Intel Ireland does chip development, but could be wrong... they have something like 6k employees, god only knows what they are doing...
The short answer is during a time of massive profits of semi companies, most cashed out. One (TSMC) has kept reinvesting gains into R&D for the last 20 years. Lithography via ASML is the cutting edge advance, but there are many other small details too. Both in the technology itself but also in the complexity of integrating that technology to work together in one process.
I think that one of the greatest advantages that Korean, Chinese, and Taiwanese companies have is that most of them seem to be somewhat immune to Harvard Business School ideas, and Jack Welch was never worshipped as he was in the West.
Keeping the bean counters under leash is good for building long-term value.
Samsung, as a conglomerate with all the affiliates makes up 20% or so of Korea's economy. The entire state is motivated to keep it going,so a financial analysis of some 20 year old in wall Street don't have such a huge impact. TSMC is the same, it's a strategic company in taiwan,so again the entire country ensures it's not going anywhere. In the west, GE and similar companies are important, but not that important,so idiots like Jack Welch get more mileage than they'd get in Asia.
Business schools teach efficiency. Efficient as in 'put in a dollar, what gets you the most back in the shortest time period?' It is oriented to appeasing stockholders (investors), who want to see earnings grow. If your company is losing money, cut cost centers and maximize revenue to up the profits for the quarter so that the stock doesn't go down. This usually means cutting jobs, defunding research, and making things cheaper. Another way to accomplish this is to sell off specialized parts of the company for a quick cash infusion. Another big thing is the need to quantify -- you need to be able to put the numbers on a chart and if you can't then it is worthless. R&D cannot be quantified like that so it, along with things like IT and information security are seen as cost sinks ready to be slashed.
All of these add up to have terrible effects on companies that rely on research and having a workforce of highly trained professionals who are entrenched in your institutional knowledge. For example, you can't just fire a materials science engineer who is a specialist in silicon crystal seeding to save money one quarter and then hire another one when needed.
> It is oriented to appeasing stockholders (investors), who want to see earnings grow.
This strategy appeases short-term shareholders, at the cost of significantly penalizing long-term holders. It's not a simple case of "appeasing shareholders".
My point is that shareholder capitalism doesn't necessarily lead to the short-term focus you describe. Shareholders can in fact be the group with the longest-term focus of all. They can still be there and care about the company long after the current C-suite are all gone.
While it's certainly possible for that to be true, it isn't in reality. The average holding period of a stock is 5.5 months.
People have proposed to not give people voting power as shareholders until they've held the stock for a period, because there is this huge population of investors that care nothing for the long term viability of thr company.
I don't think I have communicated what I meant. Stock traders would be people who trade stocks looking for return on investment through trading, not from investing in a company by holding stock.
Business school type leadership rarely works in high tech. Important decisions are technological decisions.
It's easier to pick senior engineers from the industry and give them business education than vice versa. There is also special field called industrial engineering that trains people to manage and lead industrial processes.
When it comes to research, it always costs a substantial amount to accomplish anything at all.
So the cost is always prominent, and the return is difficult to fully attribute.
Forces at work that can lead to undue cuts.
Engineering companies need to be run by engineering/businesspeople selected internally from a deep bench of capable candidates. As succession occurs technical leadership can better maintain continuity of what made the company competitive to begin with.
For the bean counters it can be impossible to realize potential in many ways, not just research. Concrete costs will always be ripe for quantitative elimination without consideration of the crippling effect on upside outcome that can not be recovered in less than a few years (or generations) after overcompensating funding has been restored.
Which could happen . . . right?
No conincidence how many different kinds of hard technology are referred to in "generations" of advancement.
Bean counters are supposed to be good at math but they're usually no engineers.
When these start to come into leadership positions of an engineering company it can be a very bad sign.
It could be worse.
Some of the top financial operators are not even bean counters, or leadership material of any kind. More like social climbers who've moved up a corporate ladder without much distraction from any effort other than the social climbing itself. Even worse when the only reason they're financial is greed.
Look what happened to Boeing when the engineering culture no longer extended all the way to the top.
Still an engineering company, with some of the world's most outstanding engineers doing very advanced things very successfully. Just not as much as it once was. And not as much as it could be.
Well acccounting companies should be run by accounting/businesspeople which does seem to work, and bean counters shine until generations later at a place like Arthur Andersen where it looks like they were replaced by social climbers at the top and oh, well.
After surrendering to the United States, Japanese companies were taught Harvard Business School ideas as part of a reconstruction effort.
While that is a very tiny piece of what is referred to as Japan's Economic Miracle, it is enough to explain the difference in exposure and susceptibility to Harvard Business School ideas compared to Korea, China, and Taiwan.
Also their industry still survived despite the massive bubble popping etc. Instead of making fabs they build most of the important tooling and chemicals in semi-conductor manufacturing. Namely the photo resists and the specialised tooling for inspecting and repairing masks, including the insane EUV masks.
Before the CCP existed, Taiwan was part of China, and was occupied for many years by Japan. After the Chinese civil war and the rise of the CCP, the defeated nationalist government moved to Taiwan. It is only then that any US alliance with Taiwan began.
China had never ruled over the island. It called it a province for ~ 7 years before Japan took Taiwan and actually ruled over it. The US was allied with KMT after they took over from the Qing dynasty. And continued to be allies after CCP pushed KMT out. When China attacked Taiwan the U.S. stepped in.
The US had a strategically ambiguous alliance with Taiwan a long time before TSMC. Take a look at a map of the Pacific. Color China red, color Taiwan blue. That really bottles up China's navy. Change Taiwan to red: now China has easy access to the Pacific.
China is surprisingly confined in its waters. Strange as it seems with all that coast line, China (and thus a huge chunk of world trade) has to transit confined sea pathways before it can reach oceans.
Conversely, should China ever gain control over Taiwan, that fact and the implied retreat of Western navies (US primarily) from those waters would make China's Navy the guarantor of the global (trade) order in its most important economic region. Today, it is the US Navy that fills that role.
Geography remains hugely important in shaping history. Per some views, the very rise of Europe was primarily due to navigational advances. Technologies that allowed circumventing strategic geographical realities that had stood for thousands of years: the land corridors of Asia and the Mediterranian sea & ports.
That's not something that Kissinger has ever gotten to decide and he has historically been frequently wrong. Kissinger is about as meaningless of a reference point as you could use in that sphere.
He hasn't had potent influence over US foreign policy in over four decades. They barely pay attention to anything he has said in the last several decades, and the left certainly ignores him.
There are multiple reasons, but the single biggest reason is: Apple
A customer that is willing and able to ramp to huge volume a small size chip (important for good yield), and even willing to ramp with relatively low yield (due to high iPhone margins).
Time to volume production is generally gated by yield for foundries, and Apple provides an intermediate target with the small mobile chips.
Don't underestimate the psychological effects underlying vendors' convergence on TSMC
It's like an extreme version of "nobody got fired for going with IBM"
Decision makers, even and sometimes especially in groups are not always rational. There are many cognitive biases that might make going against TSMC too difficult to justify
Also historically they have been the only option for __nm manufacturing, although transistor size is beginning to become irrelevant <10nm with different tricks having more of an impact on performance and efficiency
TSMC shares deep history with ASML via Philips (Philips was the first investor in TSMC, and ASML was started off as a Philips spin off).
Most of today's advanced litho (EUV) was actually funded/developed in American labs (Berkeley lab etc). The US was ahead in litho for awhile...but GCA folded and ASML acquired Cymer (based in San Diego, develops light sources for EUV) and SVG.
TSMC has a couple of advantages: 1) Focus exclusively on manufacturing, not design, 2) Large volume customers/products (Apple, AMD, NVIDIA etc), but especially Apple/iPhone. The process technology /manufacturing is tightly coupled to design specs. This provides a faster manufacturing learning curve.
-Intel screwed up by missing mobile / pushing out EUV adoption. Since they also design their own chips, they compete with any potential foundry customers in several markets (ex: data center), which is why they haven't been able to grow their foundry services division that much. TSMC is friendly with everyone and does one thing: manufacture chips in the fab.
Intel is attempting to catch up now by placing several orders for the next version of EUV, High NA EUV. They've also stated that with their planned process innovations ("RibbonFET" and "PowerVIA"), they will have a better process than TSMC at the 18A node (1.8 nm).
-For Samsung, their core business is more memory/displays. Memory lags logic in process technology innovation.
On the logic side, Samsung has made the leap to Gate-all-around transistors (next evolution of transistor, succeeding FinFET) before Intel/TSMC but seems to be facing yield problems.
TSMC is the top dog now, but things are going to get interesting as we go <3, 2 nm...
TSMC's fabs are entire cities of manufacturing most of the necessary components right next to each other, along with millions of resident workers. You would have to replicate all of that, not just the assembly line.
TSMC isn't competitive. That's a misconception. Back in the days Intel was ahead and by a huge margin. TSMC go by the slow and steady model.
Intel made mistakes and lost its gain and here we are today. Intel has always been a lot more aggressive and risk taken. They might catch up again. It's just a long process.
As to Samsung etc they were never really in the game. They only got remotely competitive by heavy poaching and stealing of trade secrets from TSMC / Intel (and merging IBM fabs).
At the end of the day it's not just ASML machines (that take forever to deliver). It's a huge investment.
I'm not sure I understand how they're not competitive. Staying steady while your opponent is making mistakes is, in fact, a competitive strategy — straight out of Aesop's Fables. Intel doesn't get to be the "competitive" one just because they're playing the role of the Hare.
Your example of Hare suggests Intel was sleeping. I don't think they were.
And no, TSMC is staying steady no matter what. That's how they've been operating from the start - regardless of if the "opponent" is making mistakes. It's not a strategy in that sense where they're competing. They're doing their own thing and the competitors either died by themselves or failed.
They're not competitive as in they aren't competing with Intel. Otherwise they'd go for counter measures or what not. That's never been their strategy. Cutting edge wasn't their core business originally.
Competitive in economics and the market means they can provide a good at a lower price than competitors. Either TSMC has no competitors, or they are beating them. If they don’t have competitors, it’s a sign other newcomers can’t compete, so they’re still competitive.
That's like saying if everyone fails an exam and 1 student failed a bit less that they're great or that public transport in American is great because there's none in the Amazon Forest.
IF we go at it that way everyone is competitive - just label the target market differently and claim to be on top / no competitors.
It think I'd start by saying that TSMC hasn't historically been in the position it currently occupies and that while TSMC has a good lead, it's not like Samsung isn't close.
A big part of TSMC's current status is probably Apple. Back in 2015, TSMC was slightly behind. The iPhone 6S was dual-sourced from TSMC and Samsung with the TSMC chips being 16nm and the Samsung ones being 14nm. Intel had launched 14nm parts a year earlier.
We don't know what agreements that Apple has made with TSMC, but it seems reasonable to think that Apple made a big multi-year commitment to TSMC potentially with up-front money that TSMC could use to get to the position it now occupies. Apple's ability to commit years in advance can really change things. TSMC could confidently invest in their future without wondering "will someone pay for it?"
Samsung has been struggling a bit with its foundry and I think part of that is that companies don't want to commit to Samsung. While Qualcomm and others would like an alternative to TSMC, they're not looking to commit to Samsung. They'd rather take a wait-and-see approach, keep their options open, and bail on Samsung at the first sign of trouble. That leaves Samsung unable to make the kind of long-term planning that TSMC can. In a way, this becomes a self-fulfilling prophecy since Samsung simply won't have the same commitment.
Apple decided that they wanted state-of-the-art foundry access and they chose TSMC to become that. That's not to say that TSMC isn't talented - Apple wouldn't choose a company that wasn't. It's simply to note that they have a very symbiotic relationship. Apple has the margins and commitment to pay for new advances so TSMC can confidently spend money knowing it has a guaranteed buyer at high rates. Once it has that advance, it can sell that capacity to other companies a year later at lower rates and make more money. This becomes self-reinforcing.
Now, many companies that have such a great self-reinforcing profit maker end up stumbling or resting on their laurels. Intel did. So it's not guaranteed that being the big fish means you'll continue being the big fish. TSMC seems reasonably committed to continuing to spend the money necessary to remain in its position and keep Apple happy.
Intel got taken over by bean-counters who preferred the short-term profits one could get by not investing as much in the future of their foundry. Even if you have the ability to spend money to remain competitive, that doesn't matter if you decide not to. For a while, this was fine for Intel. They were making good money, AMD was a mess, and x86/x64 was still king. Of course, not investing in the future will eventually bite any company.
With Samsung, there have been complaints from Samsung and AMD about their yields on the latest processes, but I think it's reasonable to think some of that is because Samsung can't confidently invest in new processes like TSMC can. That's not to say it isn't hard work for TSMC, but a lot of work is easier when you know it will pay off. "If you build it they will come," is sometimes true. TSMC hasn't needed to take that gamble since their Apple partnership started - Apple is guaranteed business. When we're talking about stuff that requires very long-term investments, having basically guaranteed business is a huge advantage.
TSMC has executed very well and I don't want to take that away from them. At the same time, they benefitted from an Intel run by bean-counters who wanted to milk profits from x86/x64 dominance rather than investing in their foundry and the fact that Apple doesn't want to use Samsung as much as possible and could make huge commitments years in advance.
I guess I might reframe the question: if you were Samsung, how would you remain competitive with TSMC? Sure, you can spend money, but will the customers come? Likewise, if you're Intel, the question is: how long will it take to regain our competitiveness? TSMC didn't build its company in a day. It was many years of top-notch execution. It will take Intel several years to get back on track - and it's not like TSMC will be resting during those years either.
I think it’s a mistake to focus on ASML. Manufacturing silicon chips is incredibly complex, yes some of the complexity is handled by ASML but that’s one tool in a massive chain. Think about it partly in the same way as car manufacturers, point at a part in my car? Bosch probably made it. That doesn’t mean my car is a Bosch. TSMC is about bringing together the entire tool chain. To compare them to their biggest competitors- the reason Intel failed partly because they made a bet on technology for the next generation of chips and it didn’t pan out, but the underlying reason was their processes failed.
There are many great comments in this thread. Finally in 2023 we are getting somewhere in these topics.
I will ignore the Tech side of it. ( You dont buy ASML and expect it all to work, that is like saying you buy a computer and the results of the programmers are all the same. ) You should spend a weekend or two on Youtube Channel on Asianometrys, SemiWiki or Semi Engineering.
On the Business Side of things. With what the ex-CEO Morris Chang called the Grand Alliance, meaning TSMC being Foundry only they will have to work with every industry members to get things work.
TSMC also do not have their own product to compete against their customers ( Pure Play ). Compare to Intel and Samsung, do you prioritise your own SKUs first? where you might earn much higher margin, think Xeon and Exynos SoC? or Global Foundry's case, AMD's order. How do you also balance the capacity planning with so many different customers? This is something that TSMC does extremely well compared to all in the industry.
You then have an economy of scale question. New node are now fast approaching $20B capital expenses, and if you do it elsewhere and not in Taiwan. That could easily be $30B simply because you have to replicate some of the infrastructure already in place along with higher operating cost. Add in another $10B+ of R&D. Ignoring the political side of things, could the industry sustain two Fab players both on leading edge? With an expected ROI within 3 years time. And if your answer is somehow yes, given the trajectory of continuing increasing cost of leading edge node, will it be sustainable in 5 years time? ( In case anyone is wondering, Intel are using their Profits from x86 CPU to catch up to TSMC )
TSMC's culture is also very engineering centric. You have people with Dr. in engineering and expect them to learn about Management or Finance. Rather than some Marketing and Sales people. They are also pragmatic enough with technology and advancement not just sticking to their Roadmap aka Intel 10nm. Most of their Marketing and PR were hired in the recent 5 years because the amount of fucking bull shit from Mass Media they have to endure.
They also work long hours. And are organised with much higher efficiency. Most of these are organisational and management problems. You can think TSMC are moving more like Startup speed within compare to Intel with lots of bureaucratic layers. ( Which Gelsinger is trying hard to break )
If TSMC wasn't around, I think it could be argued that China would have tried to take over/invaded long before now. So if you are the government of Taiwan, helping TSMC out is in your best interest.
I'm not arguing with your point, but the impact moving a small amount of chip production to the US will have.
"Let's just move TSMC to the US" is a naive solution to the problem most people with little industry experience arrive at. No real effort is being made to move any significant portion of final fabrication out of Taiwan and into the US, and the latest and greatest processes are being developed exclusively in Taiwan. Progress on 5nm fabs in AZ will probably be excruciatingly slow, dependent on significant numbers of Taiwanese worker exchanges, and produce a chip 2 generations behind cutting edge by the time it's operational.
There's some backlash in Taiwan over "piercing the silicon shield", but it looks like it could end up being a way to get the US and Taiwan to cooperate even more intensely than they already do.
No... China doesn't want TSMC they want Taiwan... and thrusting the world into the technological dark ages for a decade by blowing up the leading semiconductor factory isn't a favorable political stance to take.
Because they took upon a very hard task, and kept punching it year, after years, through, low margins, highly cyclical market, extremely low returns on capital, tricky clients, and a marathon like RnD expenditures.
They are the leader because everybody else dropped out, and went to make money on something easier, like making Wordpress websites.
The US moved up the stack. We are a nation of software engineers who are losing the knowhow and knowledge on how to build the things that actually enable software.
These ChatGPT reply bots are actually kind of useful to remind one of what a limited perspective these generic AI models can offer without prompt engineering.
Compare this answer to the one given by Nokinside. It’s much more like storytelling, highlighting the possibility of failure lurking behind every process node success. The ChatGPT one is basically just a string of meaningless fluff words like “competitive edge” and “pushing the boundaries”.
TSMC has always been pure fab company and it pays off now. Samsung and Intel are also in chip design business and do other things. They are less focused. TSMC's lead is not guaranteed. Any new technology node can fail and remove them from competition if they make wrong choices.
I'm not exaggerating when I say that designing a new technology node for mass production is like a moon program. TSMC designs new node every few years, builds a fabs that cost 20+ billion. ASML machines are the most expensive tooling there but they are just part of the whole.
Intel and Samsung are almost as good, but small differences have huge impact in the final result and timing. Small percentages in final yield can make a difference.