34 ms·
Ask HN: Why is TSMC so competitive in semiconductor fabrication?
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?
- adastra22 3y agoIt’s not how big your machine is, it’s how you use it. TSMC has the know-how.
- Nokinside 3y agoSemiconductor 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.
- YourDadVPN 3y agoWhat 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?
- Nokinside 3y agohttps://en.wikichip.org/wiki/technology_node https://en.wikichip.org/wiki/technology_node
- rayiner 3y agoThe 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 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.
- phkahler 3y agoI 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...
- ryanschneider 3y agoYa 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?
- anon84873628 3y agoIt'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.
- pen2l 3y agoThis 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.
- chasil 3y agoThe 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.
- elteto 3y agoA 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.
- hnfong 3y ago> 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.
- hexane360 3y agoWhat's even crazier is it's 2500+ year old tech that we turned into science in a period of about a hundred years (1860s-1960s).
- mantas 3y agoMore like labeled it as science.
- xref 3y agoKinda like all these “doctors” who just swaggered in and labeled brain surgery as science, we’ve been trepanning for thousands of years!
- mantas 3y agoIt 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.
- deleted 3y ago
- Freire_Herval 3y ago>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.
- prewett 3y agoThe 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 https://www.youtube.com/watch?v=5Ge2RcvDlgw
- nwellinghoff 3y agoVery cool video. Worth a watch
- Spooky23 3y agoI 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.
- f0e4c2f7 3y agoThe 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.
- elzbardico 3y agoI 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.
- sigbottle 3y agoWhat's the difference between HBS ideas and these Asian companies? I have no background in economics so idk. Like what advantages do both have? From this comment it reads like HBS has no merit at all but I'd imagine that's not the entire picture
- Nokinside 3y agoBusiness 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.
- ramraj07 3y agoI highly recommend browsing through the videos in Asianometrys channel if you want easily digestible summaries on this topic: https://youtube.com/@Asianometry https://youtube.com/@Asianometry TSMC analysis playlist: https://youtube.com/playlist?list=PLKtxx9TnH76SRC7ZbOu2Nsg5mC72fy-GZ https://youtube.com/playlist?list=PLKtxx9TnH76SRC7ZbOu2Nsg5m...
- chasil 3y agoA 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> https://www.computerhistory.org/collections/catalog/102792671 https://www.computerhistory.org/collections/catalog/10279267...
- psychphysic 3y agoTaiwan has invested in it like it was it's military. And it worked, they got to the point that the US would defend them from a Chinese invention.
- throwaway2990 3y agoThe US has been an ally of Taiwan since before the CCP existed.
- yccs27 3y ago"Being an ally" and "Ready to defend against China" are very different qualities.
- qwytw 3y agoThe main reason Taiwan is still independent is that US was willing to defend it for all those years.
- rowls66 3y agoBefore 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.
- throwaway2990 3y agoChina 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.
- menshiki 3y ago> Before the CCP existed, Taiwan was part of China This is simply not true.
- re-thc 3y agoTSMC 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.
- pohl 3y agoI'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.
- re-thc 3y agoYour 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.
- diceduckmonk 3y agoCompetitive 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.
- 3y ago
- dev_tty01 3y agoRead "Chip War" by Chris Miller.
- shanghaikid 3y ago[flagged]
- icapybara 3y agoThanks ChatGPT
- pavlov 3y agoThese 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”.
- civilized 3y agoWhy does every ChatGPT answer sound like it's just about to say "buy my ebook, 10 Secrets of SEO and Social Media Marketing"
- govolckurself 3y ago[dead]
- obscurette 3y agoBecause most of these books are actually generated?
- TrianguloY 3y agoDefinitely not the tooling. Seriously, it's really horrible. Is there no competition to force them to improve it?
- TYPE_FASTER 3y agoInteresting article in Wired this month: https://www.wired.com/story/i-saw-the-face-of-god-in-a-tsmc-factory/ https://www.wired.com/story/i-saw-the-face-of-god-in-a-tsmc-...
- cmarschner 3y ago[flagged]
- lmpdev 3y agoDon'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
- flipper88 3y ago[dead]
- baybal2 3y agoBecause 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.
- Freire_Herval 3y agoThe 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.
- GeekyBear 3y agoThey bet big on EUV at the right time? It was certainly said to be the next big thing for a very long time before it was finally ready to be used in production.
- AChamarthy 3y agoTSMC 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...
- mdasen 3y agoIt 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.
- deepnotderp 3y agoThere 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.
- bick_nyers 3y agoIf 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.
- FooBarWidget 3y agoSo are you saying that the US is actively trying to get invaded by moving TSMC out of Taiwan?
- bick_nyers 3y agoNo... 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.
- futhey 3y agoI'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.
- Freire_Herval 3y agoChina won't invade the US for the same reason the US won't invade China. It's just another cold war.
- DeBraid 3y agoStrongly recommend this deep dive podcast from Acquired https://www.acquired.fm/episodes/tsmc https://www.acquired.fm/episodes/tsmc
- mooreds 3y agoAnd if you don't have 2.5 hours, Throughline did a good 47 minute episode as well: https://www.npr.org/2022/10/05/1127015286/silicon-island https://www.npr.org/2022/10/05/1127015286/silicon-island
- SilverBirch 3y agoI 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.
- cjbprime 3y agoTSMC'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.
- ksec 3y agoThere 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 )
- urthor 3y agoSemiconductor fabrication is winner takes all. Intel won. Everyone else but TSMC gave up in logic. Samsung mostly focused on memory. TSMC got Apple as a customer. Intel choked its lead by paying outrageous dividends and cutting R&D expenditure. TSMC has done well. They've also done a Steven Bradbury.