“Samsung announced at the Foundry Forum earlier this year that it plans on spending $116 billion over the next decade in an effort to develop the chip production capacity and expertise that will serve the needs of key players in the 5G, automotive, machine learning, blockchain, and high-performance computing markets.”
I suspect SMIC or another Chinese foundry will get there over the next 10-20 years. China justifyably considers it a strategic priority and a matter of national security and will put in as much money as it takes to pull people and IP out of Taiwan, South Korea, and the US to make it happen. The closest parallels are display fabrication, where BOE and CSOT have swallowed the LCD market and BOE is on track to compete with Samsung on OLED market share over the next five years. Similarly, there are a handful of fabs in China now starting to ramp DRAM and flash on competitive process nodes.
Unbundling the engineering model as the current system is highly integrated - do certain segments of the process better than the incumbents. This would be primarily a business model innovation, not an engineering one.
New architectures, as Intel have shown lately, their architecture isn't perfect. This requires a new set of design thinking from the ground up and challenging existing assumptions about Von Neumann architectures and instruction sets. Risc V gives some hope to this idea.
Non-silicon (Photonics, GaN, Diamond, Quantum, etc.) computing technology would require new skills with different materials that the incumbents don't possess. Still years away though.
Commodity EUV and further process simplification would greatly reduce the barrier to entry but requires new uninvented engineering technology and practices.
Risc V is made to be boring. It does nothing to "challeng[e] existing assumptions about Von Neumann architectures and instruction sets". It is also very unlikely to become the next high perf ISA for tons of reasons, it is more suited for embedded stuff.
Risc V doesn't touch on Von Neumann architectures. That could be addressed by tech like Memristors as an example but to put it lightly, they're not ready yet.
But the belief that to be a competitive fab house you have to churn out high performance chips is in itself one of the existing unquestioned assumptions of chip manufacture which may not pan out in the longer term. The idea of the generic CPU may very well become seen as a luxurious, wasteful idea once Moore's law properly runs out of road in a few years. Specialization will breed new ISAs, even boring ones.
> The idea of the generic CPU may very well become seen as a luxurious, wasteful idea once Moore's law properly runs out of road in a few years. Specialization will breed new ISAs, even boring ones.
This is already the case and I suspect the current general structure will continue mostly unchanged: you will still need your general purpose high perf generic CPU for the mostly the same workloads we use them for today (and that's including to run legacy software), and for now there is kind of only one broad successful approach to design them (for mass produced things, at least). Then in embedded chips you can use basically anything, and you also have way less stable ISA in chips more dedicated to massively parallel compute.
Even with JIT you can not really multiply the basic GP CPU ISA ad infinitum, because for bulk system code JIT is not that viable (even if it is for big apps). Also, this is basically attempting to deport the stable interface problem in another layer, but you can not necessarily remove all the features that made it possible to have a stable ISA, given tons of them are also needed for perfs. And they are since a very long time. So for even just semi-fast general purpose CPUs, I suspect the race is mostly over (hypothesis: higher level computer topology unchanged -- if you switch to e.g. chip stacking, things could change more)
For all the other cases, and you are right they are also massively important, things will continue to evolve in tons of directions.
> I can’t think of how there is going to be a new entrant to the market that competes with the current three’s processes.
There will probably be a continuing growth of demand for making chips, so I wouldn't be surprised to see some folks get into the market with "old" tech. So it's unlikely any one will jump in right at the bleeding edge, they may work their way up to more modern processes over time.
I could see a niche of 'home grown' fabs in various countries / regions that server local demand that is paranoid about supply chain security.
Depending on how automated you make the fabs, I wonder if it would be possible to sell "limited runs" of chips to people, kind of like book self-publishing but with VHDL/Verilog.
You know what I like to think about sometimes: all of TSMC's facilities, except for one each in Washington and Singapore, are easily in range of mainland A2/AD weapons. Even if Taiwan is unhit, the silicon will be no good if it can't get off the island.
When I read the title, I was actually hoping for a video tour of one of their fully automated 300mm Fabs.
Unfortunately it seems that video material of cutting edge fabs on the web is scarce, and what is available does not really show the fab as a whole, but only details instead.
Well, I guess everybody is afraid of giving away IP...
One thing I don't understand is why fabs got uprooted from USA. There doesn't seem to be cost savings if you try to offshore a multi-billion dollar fab because labor is not the main cost. So fab in China/Taiwan/USA should cost more or less same money to run.
As a sibling comment points out, there is still significant fab capacity in the US.
There are a LOT of things to untangle. The first thing is the difference between leading edge fabrication, and everything else. If you look at this wiki list of active fabs (https://en.wikipedia.org/wiki/List_of_semiconductor_fabricat...), you'll see lots of plants still building stuff at 45nm+, and a large chunk building in 22-45nm.
There are only 3 main leading edge manufacturers', TSMC, Intel, and Samsung (I'm discounting Global Foundry since they publicly announced they're not chasing smaller process nodes). All of them concentration their fabs in their home country. Intel has some fabs in Ireland (I don't really know why..), Israel (Intel's Israel team is top-notch and a real contributor to Intel's success, so I guess that makes sense), and one in China (cause that's the world we live in). All of Intel's leading edge plants (10 and 7nm) are in the USA.
Of the non-leading edge plants, you'll also find that the majority of companies also concentrate their fabs in their home country, with American companies being a bit more dispersed. I would evaluate that the majority of fabs reside in a company's home country. I wager a bunch of the greater American dispersion comes from mergers/acquisitions. There's been a lot consolidation in the chip industry (at every level and vertical) in the last few decades.
I think what you're seeing isn't so much companies moving their fabs off shore. What you have is that most companies are no longer willing to invest their own R&D into pushing fabrication processes, so they stop building new fab plants, and instead look for a fabrication house to buy from. And today there's really only one choice, TSMC, which happens to be Taiwanese.
I've been reading Fabless: The Transformation of the Semiconductor Industry lately, and there's a whole bit in there about TSMC.
TSMC was the first company to be a "pure-play" company that did no design work and only manufactured orders. This was back in the days when the guys at Intel and IBM had the attitude that to be a real player in the industry, you had to have your own fab. But there was an increasing amount of fabless design houses in the US that would piggyback off the excess fab capacity that Intel/IBM had after they were done with their own chips, and TSMC targeted that demand by tables companies for manufacturing capacity directly.
The takeaways I got was fabs moved away from the US because TSMC and the like figured out an innovative business model that let them focus on a single part of the supply chain that provided substantial cost savings for all these new fabless companies in the US. Meanwhile, the old established players like Intel/IBM with fab experience weren't willing to hop on this new segment and kept their fab capacity mainly for making their own designs. Although TSMC started in a high capital low return segment, as opposed to fabless that could produce and capture most of the value in a chip design, the IP itself, TSMC eventually became so good at what they do and process nodes became so expensive to develop that no other US companies had the incentive to enter this segment from scratch and lose money for years before catching up with TSMC for frankly not very high returns. The only US company keeping up with TSMC is Intel, and that's because they've been manufacturing from the start and have enough volume to make investing in a fab and spreading that fixed expenditure over all their chips worth it.
Transistors, microchips and fabs were all born and thrived in USA. One would think USA would have first mover advantage because this is very very hard to replicate tech but for some reason, fabs got completely eradicated from USA.
Have you ever wondered what's inside a new $20 billion Fab made of that makes it so expensive? Where is that equipment designed, developed and manufactured?
There exists extremely high-tech sector called "Semiconductor Capital Equipment Industry". All the instruments and equipment involved in deposition, lithography, etch and clean, inspection, metrology, and prosess control is super expensive.
Most of that technology is in the hands of Japanese (Nikon, Canon, Hitachi High-Technologies,Tokyo Electron and too many others to name), ASML Holding is big European player and largest supplier of photolithography systems. American companies involved are Applied Materials (their Semiconductor Systems), Lam Research and KLA-Tencor.
Yes absolutely, but why not protect these innovations by operating these machines completely under our own supervision?
It's a lot easier to duplicate a production chain if you have full access to the machines (which come with extensive training). And you can even copy the process on a machine-by-machine basis, which makes it much simpler.
1) You do realize a large amount of people here aren't american?
2) Many (including myself) believe that a tightly coupled international trade is the by far most efficient way we have of avoiding violent conflict between countries.
There are still plenty of fabs in the USA. The Wikipedia page on the list of fabs puts over a hundred in the US[0]. I remember at one point when Samsung made most of the A Chips for Apple, the fab was actually in Texas though they do have a bunch in South Korea as well.
Employing 48k qualified people (like TSMC did in 2018) isn't cheap anywhere, and it's likely 2-3 times more expensive in the US. Maybe $1.5B/yr in Taiwan and $5B/yr in the US?
Indeed, running a public company in Taiwan is known not to be a pleasant thing.
Taiwan government is very activist when it comes to henpecking corporations on "governance standards." Boards of a lot of important companies are de-facto micromanaged by the state.
A stark contrast to the US, where laxity on that front is negated by mindbogglingly complex securities regulations.
For US, you can still play your game, just pay enough for lawyers. For Taiwan, not so much. Choose your poison.
Maybe, but the biggest coal-fired power plant in the world is there for example (look up Taichung Power Plant) and those things aren't known for being clean.
ROI = Return on Investment. Most people in practice calculate that in excel as a percentage of profit over their capital investment.
Smart business people know there's more important things than a dimensionless number, but hey, Peter Principle and "math is hard." See Black-Scholes and the 2008 recession for how wrong that can go.
I used to do math tutoring for economics students. Boy, that was an eye-opener! I had to explain over and over what independent variables meant, and how that affected for example derivatives. As bad as p-hacking and the social sciences.
Econ Student: "Can I just plug these lists of numbers in the formula and write down the answer?"
Me: "What are the numbers measuring? Are they independent?"
Typically a process step means a completely new fab. Thus when you Build a new one everything is on the table: cost of land, cost of labor, skill level of labor, distance from HQ... So when intel build new fabs in Penang and later PRC these all went into the decision. The old fabs are still around making older-generation parts.
Ah. So what they did at the Penang site was bonding, testing and packaging? (Not sure I'm using the correct terminology.)
That makes more sense now that I think of it. Can't imagine there was a steady supply of electricity there back in the early 70s. (Got any source though?)
I did wonder if Apple would build a fab; I mean what else do you do with all that cash you've got sloshing about, and they must use a fair chunk of output of a fab.
The article ended right when it promised to get interesting.
It set up the stage with the construction of the fab, the market, the international tension -- and then nothing.
I would've really loved some insight about what the international influences mean for day to day operation, some technical challenges etc.
(Or this would be the perfect setup for a murder mystery. One of the key engineers is found dead, and her ex-boyfriend, US and Chinese spies all could have a hand in it. My fantasy running wild again... :D )
Yes, it's ASML and their technology partners in Europe (Zeiss, Trumpf, many others).
ASML is really at the core of bleeding edge semi. Their lithography equipment is basically unchallenged in the industry. Noone can make a 7nm CPU without them.
Disclaimer: I've worked in several semiconductor companies mentioned in this article as hardware guy.
Actually it's quite liberating to be the provider of technology. You control the whole market. Specially true for some of the companies which are virtual monopolies (over 95% of the market is owned by them). In EUV side, which is the future (happening now), some vendors are the only option. Fabs can pressure them, of course, but at the end of the day they are the only vendors available. I find it quite funny, on one hand TSMC is one of those "jump? Where and how high" kind of customer but OTOH there is no one else they can turn to. I find this kind of relationship quite funny and unique. I am not aware of any other industry where the vendor has this much power. All because they are the only supplier. Maybe in defense?
Developing a state of the art wafer scanner for EUV lithography would probably take decades, many billions of dollars (not tens but more likely closer to $100B), violate thousands of patents and the end result would 100% be inferior to what the entrenched players (which for EUV lithography is de facto only a single company, ASML) have on offer today, which means it would actually put you at a competitive disadvantage as a foundry, compared to competitors that just buy the state of the art machines. Add to that it already takes huge effort and investment to run a fab, semiconductor manufacturing isn’t a process where you just get some gear, then press a button and wait until the chips roll out. It takes billions and months to years to just switch to a new process node. That’s TSMC’s core business: using the tools to make chips, not make the tools themselves, and that’s already extremely hard. It’s not a coincidence there are basically only 3 big players left that can compete at the most advanced process nodes.
Development on ASML NXE EUV scanners started ~20 years ago and only since very recently are they being used for high volume production. There is literally zero chance anyone would be able to profitably build the same thing from scratch again, at least not unless they find some radically different way to make semiconductors.
The fabs are in Taiwan. The people who operate the fabs are in Taiwan.
The people who do the R&D for all the processes live in Taiwan. The people who create the Intellectual Property live in Taiwan.
Analogously, many decades ago Mitsubishi Group, a Japanese company, bought Rockefeller Center. Ha ha. Good luck with that. If things didn't work out it's not like they could disassemble 30 Rock and put it on a boat to Japan. Unsurprisingly they no longer own it.
It's not like before WWII when the Japanese bought one of the elevated subway lines in NYC and disassembled it for steel and shipped it back home to build battleships.
TSMC's assets are in fabs that are difficult to disassemble without great expense. And the other important assets go home every night. "Foreign investment" isn't able to spirit them out of the country.
TSMC is a Taiwanese company. Who gives a fuck what some bookkeeping entries in some computer ledgers say about "foreign investment".
> It's not like before WWII when the Japanese bought one of the elevated subway lines in NYC and disassembled it for steel and shipped it back home to build battleships.
> I found a Wikipedia discussion, but it is by no means dispositive:
It seems fairly so on the claim that Japan bought the el, dismantled it, and shipped the steel home; it is less dispositive on the question of whether some of the steel either reached Japan or freed up other steel for shipment to Japan when it was dismantled by it's actual civic owners, who had no connection to Japan.
Shareholders of such things are sometimes proxies for the true powers hidden behind. Money is not a democracy. There are national interests at stake here, and then I'm not talking about Taiwan's national interest.
Shareholders typically don't vote on things like that directly, but they can vote for Directors who usually have the power to fire/hire CEO's who then make decisions like new locations.
