A few highlights from a conversation with a friend who works for Zeiss and regularly deals with ASML (take this with a grain of salt, I may have misunderstood some of it and he also may not be an expert in all of these areas):
- The lenses (actually mirrors for EUV) only demagnify by a factor of 4. So your wafer template is already extremely small and costs millions to produce.
- 13.5nm is pretty much the smallest wavelength you can reasonably handle. If you want to go smaller, you have to build a particle accelerator. (He heard that from a colleague, wasn't sure how much of that was actually true.)
- The prices in this Twitter thread are pretty spot on. The EU wants to invest into chip making to become more independent in this area. But the sums they're talking about wouldn't even pay for one such machine. Politicians don't seem to be aware of the dimensions we're talking about here.
- Also in the thread, ASML has a monopoly on this tech. Others have not invested into EUV and by now, it's pretty much impossible for anybody to catch up.
- Everyone's hiring like crazy atm. He started home office during Covid and now Zeiss says he can't even go back to the office because they don't have enough office space anymore. He doesn't mind. Everyone's distributed all over the place anyway so whether he does conference calls at the office or at home makes no difference.
- He said they're not very affected by the chip shortage.
A lot of this tech is the result of more than 30 years of research, starting with national labs (DOE funded Sandia, LLNL, etc), then consortiums of Intel, AMD, etc (EUV LLC), in fact, ASML was a member of EUV LLV and specifically, because some of the techniques they're using was originally pioneered by DOE Labs and EUV LLC members, they had to get approval from DOE/EUV AFAIK.
Also, the core laser tech, tin droplets, was actually pioneered by Cymer, but they bought Cymer wholesale. So part of the issue is they vertically integrated a lot of the very difficult components of their machine by buying smaller companies.
This would be like TSMC buying ASML and preventing other fabs from getting access. It's definitely bad for the market, and EUV is so damn expensive it took 30 years of R&D and a consortium of companies to even get to this point. The competing technologies, using different Lasers, would need enormous funding to get them to work.
EUV LLC built a 100-nm EUV prototype almost 20 years ago using a different technique, but gave up on that path. There's also electron beam lithography (EBL) which is maskless AFAIK, but I don't think it ever made it to production either. I guess Intel and other players decided it was better to just own ASML stock.
So the world seems stuck with ASML, unless of course, China gets a hold of one of the machines and decides to just clone it and ignore patents.
> - Also in the thread, ASML has a monopoly on this tech. Others have not invested into EUV and by now, it's pretty much impossible for anybody to catch up.
Having basically the entire tech economy relying on one unique supplier is just insane.
I hope their production facilities are very well protected and decentralised otherwise one natural disaster in the Netherlands or an Unabomber type terrorist attack and the semiconductor shortage we are facing now would be nothing compared to what would happen then.
Except I'd imagine that most people won't feel the terrorist attack quite so quickly, giving folks likely enough time to recover. If they got bombed up, all the foundries would be fine and can continue production of chips as usual. Of course, the machinery need sophisticated maintenance and replacement parts, but most of those can be produced by non specialist equipment (like normal mills and lathes and stuff for all the mechanical structures). Zeiss plays a big part with the optics and stuff so that would mostly be covered as well.
I think the key is that, the machines that make extremely advanced machines are not that advanced in of themselves.
You don't have to go down that many layers from the machines that foundries use (arguably objects among the very apex of human technology) to bog standard fabrication machines like lathes and mills and whatever they use to cast metal. We have lots of those around the world. It will definitely take a bit of time, but I don't think it will be all that utterly devastating.
> Also in the thread, ASML has a monopoly on this tech. Others have not invested into EUV and by now, it's pretty much impossible for anybody to catch up.
This is essentially true, but people need to keep in mind this is not like when we say Apple has a monopoly, or something.
The capital equipment market is very different than consumer markets. For example in 2019 it's estimated ASML sold 26 EUV machines, in total. Analysis are super excited because this might go to 41 in 2021.
ASML makes the machines that make the chips. They can do their own runs of whatever chips they need, because they have to fully test the multi-million dollar machines before they disassemble them and ship them out.
That's not how it works. The light source and steppers are important, yes, but there's much more to a fab. Think of all the chemistry, for one thing.
What you claim is basically like saying a robotics company could just build their own trucks, because the truck companies use their robots for the truck factories.
They have to fully and completely test out the machines before they ship them out.
So, yes — they do have their own fabs in house. They built them.
Disclaimer: I worked a 6 month contract for ASML in Eindhoven, and I talked extensively with their engineers. They told me what they did as part of their testing process.
Sure they could. But someone still need to design the chips and take them from silicon to a packaged chip. They obviously do that to a limited degree, but a test run to see if everything is within spec is very different to producing actual usable finished chips. Or at least they don't have to go the full on complex state of the art chip design just for testing.
They have the masks and all the chemicals and other materials, equipment, and devices necessary to run the machines that they build. They get masks and other information as needed from their customers. They have to be able to demonstrate full end-to-end performance of the machine to the satisfaction of the customers, before the machines can then be disassembled and shipped out.
They make very few machines per year (measured in double digits), and each one takes a team of people many months to make.
Each team spends months on-site at the customer to prepare for the machine to be made, then goes back to Eindhoven to actually build it (which takes months), then it gets disassembled and shipped out to the site, then the team goes back to the site and spends more months rebuilding the machine and tweaking it for final operations. The whole process takes years to go from start to finish. That’s for one machine. And if you want to move it a foot to the left, then you’ve got to go back and tear it down and go through the rebuild process all over again. These machines are that sensitive. Just you walking into the clean room changes the atmosphere where the machines are running, and risks causing them to become misaligned and need recalibration.
And that’s for the regular non-EUV machines. Each one of these costs about 50 million bucks or more, and that’s just for the core machine itself, and not all the ancillary equipment that is needed. EUV machines cost two to three times that amount.
The technology to do EUV depends much more on things that are not chips, than they do cutting edge chips that could only be manufactured by the likes of TSMC. Their actual requirements for the chips themselves that go inside the machines are relatively low. With ASML, It’s the physical engineering processes that are truly superhuman.
So, they can easily make the chips in-house that they need. It’s more a matter of whether they want to do such a small run to get the chips they need, because they don’t need that many chips per machine, and they make just a tiny handful of machines each year.
OTOH, they can also easily get supplies from anyone they want, because of their critical position in the chain.
> - He said they're not very affected by the chip shortage
My guess as to why is that they probably have a very predictable/steady chip supply pipeline and are low enough volume and high enough margin that they can easily outbid anyone and it wouldn't have a meaningful impact on the bottom line.
That's not exactly difficult. Particle accelerators are all over these days, from medical applications(radiotherapy with liniacs) to materials inspection(portable cyclotrons).
X-ray producing synchrotrons are generally pretty large and expensive. There has been thoughts of using them for lithography and microfabrication in the past (see LIGA) that haven't really panned out. There are companies working on more advanced laboratory sources, but mostly for analytical applications (EasyXAFS, Sigray). There is a company working on a lab scale synchrotron (lyncean tech).
Ah yes, ASML. I have a number of friends and ex-colleagues who work there. Unsurprisingly, due to their importance, ASML are extremely well-known inside of the Netherlands. It's one of those companies that spun out of Philips Electronics and became very successful (NXP is another good example).
A lot of research and development is still going on in the area where Philips had their research labs. If you're curious to see what kind of high-tech is being worked on in the "Dutch silicon valley" I suggest you Google "Brainport Eindhoven" or "High Tech Campus Eindhoven".
Kinda off-topic but what's it like working at ASML? They are hiring for a lot of software roles at the moment. I've been mulling about applying there but from the outside it looks very bureaucratic (at least that's what the Glassdoor reviews seem to indicate).
As someone who currently works there it is indeed bureaucratic. But I also feel like it is an environment that focuses less on cost/money and optimizes for the highest quality. I.e., there is room to explore the best way of doing something and there is less pressure to do it ASAP or for the lowest price.
Ah fair enough, thanks for the insight. I've worked in a heavily regulated domain for the past few years and just wanted a change to something different. Perhaps I'll still give it a shot if nothing else turns up.
I applied there, they asked me what I would do if there was too much work for the week. Of course I answered I would just work nights, np. No, they said: Go to your team lead tell her/him you can do one thing and discuss what to choose. They don’t want people getting burned out, they are too important. I liked that.
In most interview situations, the interviewer is looking for someone who will "go the distance" to deliver no matter what, i.e. someone who will work nights
You can use this question to show your strength, you can say if its a one off situation you will go the distance but if its happening regularly then you will try to raise questions like why are projects always going into crunch mode, are we not estimating the tasks properly etc.
This shows that you are a team player but also willing to push back.
Did it seem to you as if they "held that against you"?, I mean, that you said you'd work (a lot?) extra? Maybe they could look at that as indicating some lack of life experience,
Or was their intention with that part of the interview primarily informational, I mean, that they wanted you to get to know their company better?
They immediately told me it was a sort of trick question and explained why (they value my sanity and health). Perhaps it was also a way to advertise themselves to me? They sure made a good impression with this.
If it happens once or twice a year, I still wouldn't be bothered by it. In fact, I'm such a procrastinator that I sometimes do it because of my own tendency to postpone things.
I love working at night, but not in addition to working the previous day. Has a bad impact on everything unless it's truly a one-off emergency or something.
You have to clock your hours there and are pushed to take time off for any overhours you have made.
On the other hand, customer escalations are expected to be handled relatively quickly.
It really is. The machines are highly sophisticated and downtime is very costly. Code bases are heavily documented and have a bureaucratic process to protect them. ASML does have less crucial software projects that have a more normal process.
Some of the jobs look pretty cool, can you share what is the salary range to be expected? The website is silent about that, which in Europe almost always means underwhelming numbers.
I am a contractor who works for them, but have lots of friends working as direct employee. (Netherlands, software engineer mostly)
They pay you 12+1 salaries (1=you get half of your salary as bonus 2 times per year, as vacation bonus).
You get 40 days of vacation (Netherlands has 25 days of vacation standard, %99 of companies)so people are very relaxed.
ASML is able relocate you from another country, moving all your house (I think it was up to €5.000 costs). Bringing family etc. They show you around city, assign a guide to explore city, help for municipality stuff etc.
They also give a temporary (2months) accommodation until you find a place, or until they move your house.
They always hire because they are growing all the time.
They have people working since 20 years or even 25 years.
To me, frankly, it's like the Google of the Netherlands, because some people try to get hired by them and retire there. It's a good company in my experience.
Thanks for elaborating on all the perks, but frankly your posts misses the most important bit of info, namely the salary ranges. I don't care about vacations or relocation package if the salary is mediocre and I have to work there until I'm old (i.e. if they don't pay enough for me to save money and retire early/earlish).
Unrelated to this thread, but I always found 'optiver' such a weird name for a Dutch company. Sounds to me like "optiefer". "Optiefen" is a Dutch way of saying "fuck off".
Optiver works in a vastly different and super tiny sector (in employee count) which is renowned for having higher salaries than any other sector (including tech) and is a massive outlier.
Yes, though there's also a couple other examples. In my experience the biggest barrier to europeans getting fairly compensated is just that people believe it's impossible and so don't seek out the industries that do.
Of course you will just get help when you get cancer and homeless people get help when they need it. It takes some money away from the higher paying jobs, and may make the salary underwhelming compared to the US. But then again, you won’t need a million dollar house in the right neighborhood to get your kids into a decent school.
The US has taxes too (and their government does plenty of throwing money at things.) And many companies there still pay for health insurance. I think this argument doesn’t really make sense: taxes or other costs of doing business don’t sufficiently explain the difference in pay.
The wealth gap is a lot bigger in the US and here you can walk into a hospital even when you’re homeless and leave debt free. Anyone can apply for social security of about 1000 eur/month. People are only homeless when they have mental issues preventing them from following the steps to get them out and they can get help at any point. This is pretty different from the US.
I don’t disagree with what you wrote but I don’t understand what it has to do with differences in software engineer salaries between the US and the Netherlands.
I think GP meant more taxes to cover for country infrastructure (roads, power, etc) , schools, universities, health insurance, pensions, ie the 'all inclusive for everyone' package which is more common in Europe than in the US.
Some of these taxes are paid for by employers.
Edit:and they cover costs for everyone, including non employees (they're generic taxes, not employer-funded health insurance). So you have 'not the whole population' funding the whole population.
Basically, it's more expensive for the company to hire you, so you get less.
I remember talking to a company which could hire me either in the US or in the EU. US paid better because cost of employment was lower, but at the expense of an overall less complete package for me and my family which I would have had to complete with my own money .
In the end, you can't really compare raw salaries - you need to look at what's included, and what kind of cost of living you will incur.
The GP suggested that wages are lower in Europe because of higher taxes. So to be clear, I am only arguing about the former thing and what the taxes are used for is basically irrelevant to me.
My complaint is that the argument about tax just isn’t good at explaining the difference. Eg, choosing basically at random from a quick search[1], one might pay $100k in employee taxes on $265k gross (37%) in the Bay Area. I don’t have a good idea of the employer payroll taxes but if I guess about 8% that’s another $20k the employer needs to pay. Then they pay health insurance (rough guess at $20k per year, but not sure about this average numbers were like this including employee contributions so I’m assuming that plans are good but maybe employees are healthier than average and have smaller families bringing down the costs).
So that comes out to $305k from the employer and $165k to the employee.
Suppose a company spends the same amount in Berlin. That’s €260k. Let’s say there’s 20% or so of employer payroll taxes[2], and we’ll bump it up to 25% for extra private health insurance (no idea if that is reasonable; the 20% includes some compulsory insurance) giving a salary of €208k. And then at 50% effective tax rate[3] that comes to €104k = $120k net, which isn’t really so different from the Sam Francisco example. But you don’t (I think) see many salaries like that in Germany. They would look more like €100k gross[4], so where is all that other money going? I don’t believe there are 160% employer payroll taxes, and businesses generally pay taxes on profits not revenue. So there must be some increase in other costs for the business but I don’t understand what could explain such a big difference
I think the actual argument must be about competition and demand for labour. But I don’t really have a strong argument. Salaries are higher not far away in Switzerland. There is some theory that salaries are high (at some companies) in the US outside of coastal cities because one could believably threaten to move to one, but Europeans can’t typically make such a threat. But many Germans could surely threaten to move to Switzerland, and many do commute across the border. So I don’t see how it works.
Edit: I realise I picked Germany instead of the Netherlands. But I think tax rates are higher in Germany?
[3] made this number up. I suspect it is actually hard even with a relatively high income to pay such a high effective tax rate. I tried pulling out an online tax calculator but there are a lot of options and I didn’t know what was reasonable. 50% feels like a reasonable overestimate to me.
[4] this is mostly coming from HN. Maybe the Germans earning €200k don’t write comments complaining about salary differences between the US and Europe.
In the US, the rule of thumb seems to be that employer cost is about 1.3 what the employee gets. So using your quora example, if the employee got 265k (165k after all taxes) , it cost the employer 345kusd.
In France, the same 345kusd will turn into 172kusd for the employee, and after income tax, assuming single, no children, will become 120kusd.
Basically, for the employee to land the extra 45k (and match the us salary) , the employer will have to pay an extra 140kusd.
Now, this is a factor, but maybe not enough. I'll speculate that it's a combination of :
- lower cost of living in some parts of Europe (housing in the bay area vs German cities say) means less demand for for higher salaries. It's worth noting that gafas wanting to lower wages for remote workers seem to imply exactly this.
- some things are already paid for by your taxes (university, unemployment insurance, health, retirement, public transportation, etc), which will tend to keep wages lower as well because you don't need to take the money from your post income tax salary. This is why where your tax money goes matters. If tax money went entirely, say, to a pelican conservancy associations or waging war instead of all of the above , I would definitely want to be paid better.
- productivity questions aside, more vacation time so employer wants to pay less
- people commute, but very few people leave their country (language, culture, etc). Germany is a rich country, it's not obvious that your life will be spectacularly better in Switzerland (it might be though!). Sure, Switzerland has very high wages, but cost of life is very high, hence the commute. I will also venture a cultural aspect, which is purely speculative and based on my observations : Americans tend to move a lot more than Europeans.
Yes, pre-tax income is also different, even if you use "pre-tax" in the broadest sense (cost per employee, but without expenses like office space that happen to scale worth employee count but are not related at all to the contract). But if you look at cost per employee you still see much higher numbers in the US, that's what you were talking about, right?
I think the explanation is that the US simply has enough global dominators in their field (e.g. Google basically eating up the non-Chinese ad market over the whole range from local business directories to the brand awareness stuff that used to make TV stations a big business) that they define the market, pulling up the bids of weaker bidders, whereas in other places outliers looks ASML are so much of an exception that the bids they need to get whoever they want hardly influences market rate.
Well to counter that, when comparing similar employees within my international company between the US and the EU, the US employees are about 1.5 times more expensive (this is the money that an FTE will cost a project). Not sure why that is, perhaps it is really an inequality issue and maybe the cleaners are cheaper in the same company in the US than in the EU? I don't know those details.
> People are only homeless when they have mental issues
Not to say there's not a relative greater share of the homeless population who has "mental issues", but it's definitely not the only factor at play. Many situations can lead to homelessness.
I "walked the city" with some former homeless people with my colleagues as part of a societally relevant activity. They explained us that most cities have places where you can get a free postal address, then you can ask for "bijstand" (lowest level of support here), and for rent subsidy and care subsidies. Then you can apply for a social rent housing, which by then you can definitely afford. But going through these steps requires some level of focus and dedication. What you see is that people with severe ADHD (and a plethora of other issues) for example, simply can't sleep in a public sleeping place (like at the Salvation Army), too much crowd and strange people. They prefer the streets. Often they don't trust the state or the healthcare system and avoid seeking help even with festering wounds. One woman we spoke with was a raging alcoholic, one of her kids murdered the other one. These are tragic, heartbreaking stories. These people are completely locked up in their trauma.
I have a group of friends who all worked there (San Diego office, formerly Cymer), half of whom have quit. They complained of massive inefficiency (very bureaucratic, lots of meetings), relatively low pay, difficulty moving around, feeling overworked but still shipping buggy software, and some difficult colleagues.
I also happen to know a project manager there who had a bit of a rocky time getting started (overworked), but I think is now happy there.
To be honest, my friends have put me off from working there, but who knows, it could just be a biased impression and maybe the company is working to turn things around.
My major background is in electron microscopy. Eindhoven is also a massively important microscopy center. This is because Philips used to build great TEMs once upon a time, which was spun out as Philips Electron Optics in the mid 90s.
That company merged with FEI of Portland, OR to become one of the largest electron microscope company. FEI a few years back was bought out ThermoFisher to become a group company. They still command a huge portion of the electron microscopy market.
Philips history in 6 chapters, written up by RF engineer with 30 years at Philips/NXP https://www.maximus-randd.com/technology-history.html TLDR Commodore level management. It ends with picking inferior but marginally cheaper third party components for TVs instead of using own internal products developed specifically for the role, in effect sinking R&D budget and destroying market leader position.
WSP (World Standard Pinning) "standardization" story with management naive dreams of selling own products in Asia without cannibalizing home European market while simultaneously moving Consumer Hardware Development to Asia .. where all the hardware was thereafter designed with Asian parts first was priceless.
Philips TV division deciding to second source, skipping Philips Tuners division thus eroding own company market share and profit. Then in turn Philips Tuners division commissioning a cheaper clone of Philips MOPLL from Siemens with a huge minimal volume quota, consequently eroding Philips Semiconductors market share by 25% and forcing Tuner division to manufacture inferior products with already obsolete part. Just beautiful, <Chef's Kiss>, Pure genius!
> The most devastating development, however, was purely internal and of severe structural impact.
They are focusing on healthcare and personal care products and are doing pretty well there. Their MRI scanners are well regarded, for instance.
Other electronics like televisions are still produced under the Philips name, but that's just TP-Vision making them in China and slapping the Philips name on it.
Lighting has also been spun off into a separate company (https://www.signify.com/global) which has a license to produce Philips branded products - similar to the TPVision - Philips TV arrangement. Philips itself is fully focused on healthcare these days (and some consumer adjacencies like Toothbrushes and Baby care stuff).
I feel like ASML is the company that almost everyone who would care has heard of at this point. Granted, I hadn't heard of them 3 years ago, even though I work directly with HW designers.
Its just funny how they're going through the hype cycle in their level of public awareness. Give it 3 years and (hopefully) the "general public" won't have to care about them again, because chip shortages are resolved
:)
> US is already blocking them from exporting to China
This might actually be a good thing for China. In a "necessity is the root of invention" kinda way. Ultimately this might be good to the world too, competition is generally a good thing.
This is true in general. In this case, however, there are no efforts I am aware of in the industry that can catch them. Their technological lead is so great, in so many subsystem and science areas, it's likely they will be defacto monopoly in this space for 20 years.
I don't understand this. They made it, surely someone else can too? Are we on the verge of a bunch of industries using hyperspecialization as a monopoly justifier? I mean this sounds just like what I used to hear pre-Snowden about DARPA and the NSA etc, 20 years ahead of the public.
I know. It's difficult to comprehend the scope of science that is required to make this system work as a production machine.
For example, the surface they are printing the chip circuits on has layers of material, say, tungsten, that have thicknesses measured in atoms. As in, 'I need a layer of tungsten 100 atoms thick'. At that scale, the surface is 'bumpy' to the point that you must re-focus each time the wafer/lens moves to the next region on the wafer. This is a mechanical adjustment at atomic scale that has to happen very fast, or you'll never get enough wafers through the machine to pay for it. The solution to just this one subsystem of moving and refocusing required a decade of R&D and dozens of scientists and engineers to get fast enough, reliable enough, repeatable enough to be useful in a production setting.
Yes, others can make a system like this, with a similar amount of R&D spend and personnel and time. They would have to move faster than ASML to ever catch them. They'd have to budget 20 years, 100s of staff and a very large capital budget. In 20 years, ASML will be that much further down the road.
When the DoD was deeply involved with SV instead of just purchasing top binned COTS? At that time they basically were operating with hardware that far ahead, but that was always relative to the general consumer, it didn't stop companies from paying insane prices for marginal improvements that only HFTs and research labs would care about. Nobody would have bothered trying to bring something so expensive to the broader market - remember the multi-core games Intel played for years before AMD started knocking bread off their table?
Also, are you implying that the lack of leaked supercomputing documents is evidence of a commercial technological parity, or did I miss a big reveal (totally possible, haven't been paying attention)? Because as I recall he didn't pull a Manning style dump of operations-level intel (regardless of legality) - he instead coordinated with Greenwald to expose illegal strategic programs (and occasionally a debatable legal one that the DNI lied about in the prior news cycle - which was pretty funny). So I didn't expect any leaks about bleeding edge crypto hardware - because most domestic communication wouldn't need space age tech to break (A5/1 has always been hobbled).
China flat out bribing, kidnapping or blackmailing ASML employees seems perfectly viable. Or sneaking operatives into ASML to steal trade secrets. China is the best in the world at this and they now have huge incentive to try even harder
I wouldn't consider higher wage for engineers and scientists the same as bribing. I mean, one of silicone valleys key strength is the ability for companies to poach from competitors in order to catch up, this cross pollination of ideas is both beneficial to the industry AND individuals.
bribing isn't paying higher wages, it would be paying a huge amount of money for somebody to stay at ASML as a spy and giving you crucial information for money
Isn't it better to just hire them, so they bring you all their years of technical and operational knowledge?
Of course, it requires the new external hires be given enough influence and autonomy. It remains to be seen if Chinese firms are able to integrate international talents effectively.
Why steal if money is no object? If there truly was enough motivation in whatever circles these decisions get made in, they could just offer all the top folks at ASML an eight figure or nine figure salary package. The fact that this hasn't happened yet suggests that it's not as critical as it's made to sound like, or at least those key decision makers don't believe so.
Ironically, the very thing that everyone assumes will allow China to dominate appears to be hamstringing them: politics.
China can say they have a priority on domestic semiconductor production, and they can throw obscene amounts of money at the problem.
But ultimately, something that takes 10+ years to develop isn't appetizing for China's current political climate. It needs to be the best, yesterday.
Consequently, they'll continue throwing money at whoever promises results next year, and starving whoever says it can't be done but offers a realistic plan to eventual dominance.
Capitalism sucks in many ways, but decentralized command economies (national government commands, local supplies) are about the worst of all possible worlds for high knowledge, long lead time, heavy research industries.
Which is why you saw the Soviets do pretty well at heavy industry, but not so well at computers.
>national government commands, local supplies are about the worst of all possible worlds for high knowledge, long lead time, heavy research industries
Yet DARPA and company are behind like half of modern IT technology.
>Which is why you saw the Soviets do pretty well at heavy industry, but not so well at computers
Soviets were fairly behind the US in electronics to begin with, and they had massive production problems everywhere, including heavy industry(turns out that setting every price and production quota from Moscow wasn't really effective).
It's just that heavy industry was easier.
Then what? Probably a fight that will result in one of the greatest blows to the global economy ever, as well as the collapse of China as we know it today.
The Taiwanese will destroy all the fabs before China can take them. We already have a chip shortage, can you imagine what it will be like to cut current supplies by another 40-50%? The resulting sanctions alone would cripple China.
Hence, status quo will remain, and Taiwan will continue to be a prop for nationalist speeches in China and the US defense industry will continue to make a mint off selling them old gear.
Destroying TSMC fabs is in neither US nor PRC or Taiwan's interest, it's the only point of consensus and mexican standoff leverage between all parties that prevents a TW war from escalating too far. IMO US and PRC would rather bargain some sort of sharing agreement over TSMC even in event of military actions in exchange for not glassing TW than have everyone lose access to the fabs.
> imagine what it will be like to cut current supplies by another 40-50%
And despite everyone's aforementioned desire to preserve TSMC, ultimately this is still a trivial amount if you consider that the East Asian semi supply chain accounts for 75% of global IC share and is completely within PRC missile range if there was ever an expansive east Asian theatre over Taiwan. Losing TSMC in war over TW will set the world back 1-2 nodes and 10 years, losing regional fabs and supply chains (in SK or JP) in a broader war over TW will knock out decades worth of capex. Global semiconductor industry is a 500B market, trillions in terms of industries affected, most of which benefits the west. PRC has massively more leverage in semi mutually assured destruction, so much so that TSMC is ultimately expendable if PRC is ever forced to act upon TW, because losing TSMC doesn't even scratch the surface of semi disruption PRC can cause US/West, who again are disproportionate beneficiaries of current IC status quo.
Obviously no one wants to destroy the fabs, but realize they would be destroyed if it was evident that Taiwan would be taken. That is part of the Taiwan playbook.
If China invaded Taiwan then US companies would have a hard time moving manufacturing outside of China so maybe demand for semiconductors would go down. The problem would just be the supply of everything downstream falling off a cliff.
I’m not sure the status quo will remain. The Chinese respond very emotional whenever Taiwan gets mentioned. Don’t bank too much on China acting rational.
America pays the price for its short sighted hubris chasing profits and thinking we could turn China into a liberal democracy if we threw enough money at them
This just comes off as a “I can see the problems with that approach from 40 years ago because my hindsight is 20/20” kind of statement.
It’s also worth considering that the cost of living in the US has been lowered because China has been the factory for our WalMart / Target / IKEA goods. If we didn’t open up China to the US / WTO markets, the world would be a very different place and that counterfactual is pretty difficult to calculate.
The cost of living hasn't been lowered, because one of the chief mandates of the Federal Reserve is making sure the cost of living never decreases. Instead, the savings went to a combination of asset holders and make-work jobs. This is one of the major causes of the growing class divide - our manufacturing economy was destroyed, but the people who were put out of work were still expected to get the same income somehow.
The cost of living in our timeline is cheaper than the timeline where China was never opened for trading with the West. I never stated it suggested that it caused a negative interest rate in an absolute sense.
To some extent, as with any imperfect feedback loop. But not nearly to the degree that is implied by the statement, as most of the savings were eaten up by our own policies.
Also cost of living for who? The now-unemployed swaths of the country certainly didn't see their cost of living go down in terms of hours worked.
Your Cost of Living argument is lame; the term has a definition which you have overloaded to mean something completely different.
Yes, there are people who will lose jobs for all sorts of reasons, including competitive advantage of different regions (eg China and Mexico have labor that is far cheaper than Americans are willing to work for) or the euphemistic “creative destruction”.
I don’t feel sorry for horse buggy whip salespeople who had to search for a different purpose in life. They already won a lottery by being born in the USA in the most affluent time in history. The economy doesn’t owe anyone anything.
The only thing I wished we as a country had done during multinational trade pacts was not to put the onus on individual employees for proving jobs lost to offshoring; there was moral hazard and the evidence to prove causation was hidden from those who needed the evidence. Other than that, these trade pacts simply accelerated the changes that were going to happen anyway. It’s better not to live in a one-industry town where the winds of change can decimate the Un-diversified local economy.
> Your Cost of Living argument is lame; the term has a definition which you have overloaded to mean something completely different.
Your usage of the term has a loaded definition. You're trying to thread a needle between the nominal cost of living (which has increased) and effective cost of living (which has increased for some), presumably by appealing to CPI. But you can't appeal to a measure that's from the exact paradigm I'm criticizing - of course things look consistent and reasonable from inside that paradigm (but its results on the real world are problematic).
> I don’t feel sorry for horse buggy whip salespeople who had to search for a different purpose in life. They already won a lottery by being born in the USA in the most affluent time in history
I'm not arguing that we should have held back offshoring, and you can spare me the race-to-the-bottom privilege indictment.
> The economy doesn’t owe anyone anything
This is a passive voice, while I'm criticizing a specific actor. "The economy" didn't decide that prices have to go up year over year, and in fact they would have naturally gone down with offshoring. Rather, the government decided that prices still needed to go up year over year, despite the gains made from offshoring.
If prices had been allowed to decrease, then the communities that lost manufacturing jobs would have at least seen some benefit from goods costing less as their own expenses would have gone down. But rather than the promised benefit of "cheaper goods" being realized throughout society, those gains were collected centrally far away from the people most impacted.
Savings from China trade are fairly trivial, calculated in a few hundreds dollars per family per year.
And if the US need a cheap factory hub there's Mexico.
No, the long term goal was the same as that in relation to the USSR under Yeltsin.
1) Privatise State assets and
2) Create a billionaire class whose fealty was to Western Capitalism not the Russian state.
The CCP saw this and acted accordingly.
Seeing as you need ASML people to run these effectively, maintain them, repair them - I'm guessing we will see a massive global semiconductor shortage if China invades that makes this year look like a cakewalk.
Can geolocation easily be falsified? Sure you can jam GPS, but can you easily falsify it?
I'm not an expert, but I was surprised to learn that GPS chips aren't as simple as I imagined. E.g. to get a "normal" (civilian I guess) licence you have to manufacture your chips so that they shut down if the object is moving too quickly (otherwise it could be used for missiles).
It can be done for probably less then a thousand dollars, even.
It's also pretty trivial to detect the simple falsification schemes, but outside of exotic military and research projects, none of the easily available GPS SoCs bother.
> was surprised to learn that GPS chips aren't as simple as I imagined.
On that you are very right.
I'm often amazed by GPS, or any of the other GNSS technologies really.
A fleck as big as a thumbnail which can communicate with multiple satellites wizzing-by to lock your location within 10s of meters. The fleck contains the antenna, the RF frontend, signal locking and tracking circuit, and a small computer to provide the navigational solution from the observables. And you get all this wonder super cheap.
The whole system, the space vehicles, the ground segment is of course very expensive, but somehow the militaries of the world decided that they will you use all of that for free. If you would write this in a sci-fi I would trash your manuscript immediately. Except somehow it is not science fiction. It is reality.
Then TSMC's intellectual property becomes public property. An effort immediately begins to clone anything relevant.
Apple, Samsung, Intel, AMD, Texas Instruments, Nvidia, ASML, IBM, and so on. All are now welcome to take TSMC's intellectual property and use it in any way they see fit.
It'll be a spectacular race, and those companies have the resources to pursue it properly.
The same goes for all intellectual property in Taiwan. It all immediately becomes public property, free for anyone to utilize.
Who says? The US Government does, and the Europeans will likely agree. It'll usher in an epic confrontation politically and economically with China. The world will - practically overnight - split into two major factions, as with the Cold War.
Any people inside TSMC that are able to escape, or that are already in the US/Europe/etc, go to work helping the US and its allies to clone TSMC's technology. That includes spending the required sums to build fabs to accomodate that outcome. It'll be relatively easy for TSMC insiders to dump all relevant company trade secrets over to the US and its allies, which would be guaranteed to occur.
Not “any” one inside TSMC know the company trade secrets. One of the reasons to why TSMC is ahead is due to the fact that many years of accumulated knowledge and experience is distilled into a few minds.
Because so much trade is done in dollars. And because companies globally want access to U.S. financial markets, money transfer systems, companies and markets, the U.S. has the ability to tell even foreign companies what to do.
> The U.S. has incredible, unexpected power over global companies and trade.
True, but the more it uses it for self-serving purposes the more the rest of the world tries to find alternatives, weakening that soft-power in the long-term.
"rigged with explosives" is likely a bit of an exaggeration.
It would not be difficult to conduct a campus-wide evacuation then have the Taiwanese Air Force destroy the facility with targeted weapons. It might also be the case that the facility's precise coordinates are well-known to Taiwanese military's artillery units and the entire place could be reduced to rubble in minutes.
Normal SOP is to keep the explosives, which do have shelf-lives, in a magazine a short distance from where they need to be emplaced. Perhaps look up how they Swiss routinely included niches in bridges and the like to place them in case of invasion.
If China invades Taiwan before enough TSMC plants are built in America to meet American, and by extension - Western - demand for electronics, then it's war. Period, end of story.
There is no fucking way in any timeline in any alternate universe that Western electronics companies can allow China to put their businesses on hold, because it isn't just a bunch of AMD and Qualcomm CPUs... it's every automotive company in the Western world. It's every appliance manufacturer in the Western world PLUS South Korea.
China will go right back to 1820. Damn near dead fucking last on the world stage.
The Communist party leadership would probably kill Xi Jinping before allowing him to make such a monumental mistake. And nuclear war isn't an option either, because all of China will end up completely uninhabitable if even a single major American city gets targeted.
And it won't just be "America vs. China" either, it'll be "Every Western nation vs. China". And China will lose. Badly.
People forget that nuclear MAD works both ways when they write these power tripping comments.
If China takes control of chip supply the US will have to play along and try to apply soft power, never war, because China is a nuclear nation with 4 people to every 1 in the US.
It hasn't even been a month since Afghanistan proved (again) that even when you have overwhelming force you might not have a favorable outcome.
> And nuclear war isn't an option either, because all of China will end up completely uninhabitable if even a single major American city gets targeted.
No, by far, China is the only nation which may ever consider taking the first strike.
There is Mao's famous saying, but also the fact China is super crazy about nuclear war preparedness.
Every major city has regular NBC drills, stockpiles of everything, and huge shelters.
All it takes is someone figuring out how to do the same thing without all the specialized parts or bulk. That's the main cost. This is the way of tech: it starts expensive, full of specialized parts that only a few suppliers produce, then someone figures out how to do it cheaper.
At $1B a pop + 50% for lifetime maintenance, the incentive is there.
The problem with semi is that they will move to a machine that costs 2 billion a pop within 2 years. The output of this 2 billion dollar machine will be greater than the cost increase.
Semi is heading to a winner take all market based on capital expenditure. Even TSMC can’t afford to in house the work ASML does and keep pace with new developments.
ASML's first EUV machine contract closure was for $15 bil, and the contract explicitly stipulated that $3bil was just for the purpose of acquiring three specialty Boeing 747s for shipping the machines to the clients.
Each chunk of the machine must be shipped in vacuum. It takes about 40 chunks to make a complete EUV lithography machine. Each is about the equivalent mass of a school bus.
Why the need for special 747s? Can't they put each chunk into a shipping container sized steel box and seal it vacuum tight, then ship that with regular 747s?
Splitting some of the parts into smaller parts will expose the insides to outside air which brings some serious cleanliness issues. The mirrors especially will get destroyed instantly by heat if there is any dirt on them when they are flashed by EUV.
I imagine they do the disassembly and packaging in a clean room. (At least for now the technicians require air :D) Then the packages are vacuum pumped, sealed, then shipped. Then at the destination fab they are assembled again in a clean room.
If I remember correctly, the machines are designed to fit the plane and vis-a-versa. So they will be smaller on top to fit the curvature of the plane body, for example
I suspect their equipment is susceptible to vibrations on a regular 747. I bet most of what makes their 747's special is dampening. Also, packing and unpacking the equipment increases the chance of things going wrong.
ASML is a great company. What should be stressed is that the lenses are coming from Zeiss SMT, which is a strategic partner of ASML since 1997. In fact, they are so important that ASML bought a significant part of the Carl Zeiss subsidiary some years ago. You can't get a pretty good video about the relationship and EUV at: https://www.zeiss.com/semiconductor-manufacturing-technology...
Schott (glass) / Zeiss (lens) combo is extremely important in a lot of industries. Most people probably heard about them through photography or spectacles.
Apparently, they zap the tin droplet with a soft pulse first to flatten it before hitting it with a big pulse to make the plasma.
A former colleague of mine worked on tuning the laser-hitting -droplet process. Happens inside a vacuum of course, but after some time the vacuum would always drop. Turns out that when you miss the droplet often enough and thus hit the steel walls of the chamber, eventually you'll punch through the walls. And not be able to hold a vacuum this way.
The vacuum chamber is likely aluminum. Tin melts at 232 C, aluminum at 660 C. Vaporizing the tin could require enough power in the laser to melt the aluminum. It seems feasible to me, anyway, and is certainly sounds like the kind of thing that happens in equipment development.
The reason ASML is one of the few successful tech companies in Holland is because this type of engineering is so far distanced from consumer products. As soon as consumers come into play it triggers our HR departments to introduce a whole slew of non-engineering positions who pretty much tend to find power and kill/stifle technical innovation.
It's the reason Phillips (and many other great Dutch companies) has become an innovative shadow of what it used to be.
The US seems to have a better understanding how great engineering happens when talented engineers are running the show.
I was bullish on Cymer back in the mid to late 90s. The stock did nothing (well maybe doubled) until ASML bought the m, perhaps then doubling the stock price. I was so frustrated because Cymer’s light source was the enabler of smaller and faster chips. Anyone have a theory why Cymer never had a great ROI, compared to other equipment makers? Nikon and Cannon were already enormous, and Applied Materials, KLA-tencore, and others were already big and faced competition.
This isn't true at all. I just double checked and you can view the thread logged out no problem. Twitter's UI is admittedly garbo so depending when you clicked on the thread you may have needed to click an expand link.
It's not consistent, I've been getting these login-walls for a while on Twitter when browsing incognito and now it started to show up on my normal windows if I'm logged out.
It's some kind of A/B test or progressive rollout, they are definitely forcing me to login to view threads and there is no way around it on the official Twitter website. I've been using nitter.net instead.
Yep, I feel bad for many of the less technical "rubes" who still believe they're somewhat anonymous on the internet if they, e.g. use Incognito (which Google went out of their way to market as such), or buy some dubious VPN service, or use disposable emails on Reddit, Twitter, etc. accounts.
You are simply not anonymous on any site you visit or search you make or place you go with your phone. The tech giants and their "data sharing partners" can correlate you a dozen ways, and most of that is by design via "features" like JS APIs and browser data that is included in every request.
Those of us paying attention know that Twitter, Reddit, Google, FB, and the rest will turn over your real ID to law enforcement, no questions asked.
And it won't be long now 'till some conservative whistle blower from Google or Twitter leaks the proof that the tech giants - via the more politically inclined activists and operatives who now work in their executive offices - have created blacklists which are used to ensure no wrong-thinkers (e.g. Trump supporters) are hired or employed at their own companies. And then it will be leaked that these same blacklists were traded and sold off to their "partners", and are being used for similar purposes in all of corporate America.
And from there, we know how this story ends, though I'm not sure the activists enabling it understand that things won't turn out well for them either.
I think it's a matter of the data lords not yet playing their hand. We know they have this data but they haven't used it yet in any high profile case. They're just still in the extend phase.
I agree with your sentiment there's political blackmail at work in big organizations, but you may be surprised to learn that it's not all/just anti-Trump. Trump is rightfully seen as a racist tyrant-to-be in liberal circles, but still has and had a lot of institutional/corporate support (which got him in power in the first place, and helped him attempt a coup early this year).
Just imagine what life is like as an anti-nationalist anti-capitalist worker. In most fields it's NOT an easy position to hold, because you've got both the good-thinking liberals and the hardcore fascists against you. And here we're just talking about ideas which are not written on your forehead, but the situation globally is really bad for women and people of color (for example) in a lot of industries/institutions.
You need to login to show the whole thread. I hit the same wall. Twitter was already a terrible medium for these kinds of posts, now it just got worse.
I didn't need to log in: I got the rest of the thread by following the link "see rest of this thread". That link though is easy to miss, what with being somewhere mid-page and being surrounded by elements that are visually much louder, and I, too, wish that no one here would submit or upvote links to Twitter.
Twitter seems to be doing some kind of A/B testing. Clicking the "See rest of this thread", pops up a login window, which on being closed takes you back to the initial page.
I had this issue, sussed it in about 10 min. If you're in a Chrome-based browser (incl Brave), DL the Ublock Origin extension, then: Extension options > "My Filters" > add one of the filters in the reddit thread linked.
(https://www.reddit.com/r/uBlockOrigin/comments/p5lgcv/twitte...)
I was thinking that it’s inefficient to transport equipment from the Netherlands to mostly Asia, but any place they’d be located would be importing part from around the world. And the cost to ship one of these things is probably a rounding error in the price.
It’d be interesting for them to partner with Antonov to build a second An-225. Maybe the economics would make sense instead of needing to ship via multiple 747s.
I think you answered your own question: 747s are a stable, mature platform. Shipping is a rather insignificant cost, so it doesn’t make sense to take risks unnecessarily.
One aspect is meeting the overlay and CD targets throughout the stack to meet yield. Having a process that can find defects and more importantly practices to minimize the defects in your process is a big piece. Typically the process given to mass production would be only half solved and then the fab is trying to refine it in ramp.
I used to manage a fab of high end ASML tools and there is a non trivial list of things to continue and solve. You have reliability issues that require scheduling long (2-5 weeks) downtime to fix.
An upstream defectively issue in a spinner tool might lead to taking one dispense or develop node out of flow to improve the defect rate but in turn tank the Scanner efficiency by 40% as it waits to output wafers and that backs up the imaging. The reality is the ASML tools given to fans don’t have everything ironed out at rhe start. So fabs see the reliability issues in real time but there’s no time for the fab to take it out of production for 3 months to address a major part replacement. These all lead managers to make bad choices that continue to dig you in a hole.
At the time I was supporting 14nm it was well known TSMC had world class software tools and practices to minimize their fabs defects and maximize higher order control of overlay and CD bias. This allows them greater flexibility to take a tool down and fix the issues rather than live with them or have to use a band aid.
EdiT: I did not work at intel so can only speculate why they are behind but it was also well known TSMC photo engineers were worked 80+ hrs/week always . They were paid less so they hired more of them and they worked a lot longer so as a result they had better fab processes and support tools.
Fabs are just a wee bit more complicated than buy the machine and plug it in. Everything is very tightly integrated under insanely precise requirements. Changing equipment is akin to redesigning and reconstructing a non trivial portion of the fab. And that's just integrating these lithography machines, which are but one piece of a huge puzzle.
Yes. But it will take a non trivial amount of time for them to change course, and they could still stumble on any number of other issues on the integration.
A bit on how this played out: what they call their first 10nm node which in its third iteration has been named Intel 7 was more aggressive than TSMC's first 7 nm node, and failed for many years to economically produce chips. Both use 193 nm UV lithography, and then TSMC made a more aggressive than Intel's node using some EUV from ASML, and both TSMC nodes worked.
Now TSMC is two major nodes ahead with 3 nm risk production credibly scheduled for this year and mass production next year, with Intel said to be buying a lot of that. What was Intel's still delayed 7 nm node, the first to use EUV, is now named Intel 4. I've not looked at it closely, but it looks like something equivalent to TSMC 3 nm nodes is scheduled as Intel 20A for angstrom and 18A, sometime in 2024 and 2025.
I have yet to see anything that convinces me Intel will regain its ability to make state of the art logic chips, which for many generations was one of their most important advantages, allowing them to beat "smarter" CPU designs with their own CPUs being manufactured 1-2 nodes ahead of everyone else. It will be interesting if some day the true story of how this happened is revealed.
Doesn’t this mean the Intel is done? Intel has never designed chips with innovation that matches the quality of AMD, Apple, and Nvidia. I just don’t see how they survive as a fabless.
Doesn’t this mean the Intel is done? Intel has never designed chips with innovation that matches the quality of AMD, Apple, and Nvidia.
That's frankly not true, for example see them along with a few others not including the above companies reviving the out-of-order technology IBM first developed for its System 360 supercomputers, for Intel first in the Pentium Pro.
Culturally I doubt they can shift to fabless although we'll be able to tell if they're really trying by seeing mass layoffs of managers, and I've very concerned they didn't recruit a fab guy to be their CEO although I wonder if no one qualified was willing to take the position. Another argument against that is their placeholder CEO who'd previously been their CFO actually had a clue, I suppose wasn't part of one of the factions involved in the fab debacles, decided to go fabless, and his strategy was junked by the new CEO.
In general I don't see a path to victory for them given my guess that they won't be able to regain their ability to move to new nodes that can make economical chips. Although I haven't investigated their third iteration of their 10 nm node, now Intel 7, need to spend more time reading Semiwiki etc. and also see if there's any sign they're going to get their first EUV node, Intel 4, working in any practical time frame, it's already delayed.
On the other hand, AMD has done it again, spent a huge amount of their capital on this time an FPGA company; at best that means they believe they don't have anything better to devote resources to, which is bad news for their AMD64 ecosystem. If you've followed their history going back decades, it was said a while ago over the long term they've never made money for their stockholders, and they've repeatedly gained serious leads only to blow them, raise your hand if you once used one of the 100MHz 486DX4s. Just like Intel's cultural problems, I suspect this aspect of AMD's culture will give Intel breathing room sooner or later, as it did when K8 got long in the tooth.
Apple isn't even in the running unless they decide to start selling their ARM SoCs to third parties. Who gives a damn about those if you're not in the Apple ecosystem, which has it's own problems. Nvidia if allowed to buy ARM will I expect destroy that ecosystem, certainly make it uncompetitive, including eliminating competition they would be facing thus likely making them complacent.
So what is your final say? You say they have a chance at executing an IBM scale pivot, but a low one? You also say that AMD historic inability to generate profits will give them some breathing room?
So, intel will recover because it will gain breathing room and it will benefit from a pivot?
It seems like a stretch.
For one it seems like shifting a company used to an accommodating internal node users to one that has to retool and shift to TSMC/Samsung is a major ask. For two it seems like major profits are around the corner for AMD. This is a company that has historically had <1 % of the markets intel participate in. Those numbers have grown by factors and there is no sign they will stop.
Would I count Intel out? Maybe not. Maybe the US gov will provide a lifeline. But there is not a doubt in my mind that Intel could stumble again and loss the manufacturing advantage permanently. That seems like it would be a mistake that would lead to another 5 years of market share loss.
I certainly have no "final say," I'm just pointing out some of the forces including internal cultural ones that I believe may have effects on the outcomes.
IBM has nothing to do with the business or technical questions, I'm referring to the https://en.wikipedia.org/wiki/Tomasulo_algorithm which was first used in the late 1960s in the floating point unit of the IBM System/360 91 and much later used for the Pentium Pro.
AMD's historical inability to generate profits pertains to its historical inability to succeed well for very long, it's an independent measure of that phenomena. So no matter how well AMD is doing right now, it's in their DNA to fail any time now; doesn't mean it's going to happen, but I'll repeat again that paying a huge sum for an FPGA company is a very bad sign, much worse then their purchase of ATI which did not end up helping their CPU efforts, if anything distracted from the latter. Another sign AMD "will stop" is the reported generally lower quality ecosystem, but I have no good reading on that.
Your comment about a "major ask" is just repeating why I previously said, "Culturally I doubt they can shift to fabless although we'll be able to tell if they're really trying by seeing mass layoffs of managers...."
How much of AMD's "major profits around the corner" will be put into the FPGA unit? It was an all stock deal so they at least won't be having to pay down debt on it, but to restate my main point about it, it's a vote of no confidence in their CPU business, doing it comes at tremendous opportunity costs.
Never heard that money was the core problem for Intel's failure to move to new nodes, so the US government doesn't have the right sort of lifeline to provide. Intel is currently stumbling and I personally guess that won't stop, but that's based on very thin data and what I see the new CEO trying to do.
Or not, if the denials about the rumor about them buying Global Foundries directly from its owner Mubadala Investment Company are correct. If managed well, which is never the way to bet in tech company acquisitions, that would give them invaluable expertise in the business of being a foundry.
I know they have had fabs, but they decided to go with TSMC recently, and that's what I was wondering about. Going with TSMC doesn't mean TSMC does their designs too, does it?
The lithography machines are just one part of a very complex machine that is a semiconductor fabrication facility. So basically they failed in one of the many other things.
IIRC They have bought some but they were hesitant to depend on what was then unproven technology. Intel thought they could achieve smaller chips using the previous generation’s technology.
That article does not say that the first 15 went to Intel. It says that Intel placed an order for 15 in 2015. But by the end of 2017, ASML's cumulative shipments of EUV machines was only 16, and I'm pretty sure TSMC had at least two of those. As of a year ago, TSMC had half of all the EUV machines that have been shipped, and more than half of the EUV production capacity (on account of newer EUV machines having higher throughput).
> why don't they produce semiconductors themselves
Apart from the different field of expertise at play (think manufacturing cooking equipment vs running a full-scale canteen), there's also environmental concerns. Looks like ASML has outsourced much of the production of key components to 3rd-party countries.
Electronics manufacturing has a considerable environmental impact (see articles about chemical soil/water pollution in china) and global north countries are happy to have others exploit their population and damage their environment in their place. This is similar to how the "recycling" industry operates.
It could be done with a merger with TSMC (or Samsung), but at this scale these companies are critical for the countries themselves. A merger like this would be extremely political.
This video is in the same vein, and does a great job putting this machine into perspective in teh chip making process, but it's sadly a decade old now.
Seems like quite a risk to only have one company provide these machines.
Semiconductor companies seem to follow the same trend, the major contributing factor seems to be the patent portfolio needed to secure cross-licensing deals.
It looks like patents are actively preventing new companies in the semiconductor business.
I can’t find it, but I’ve seen people extrapolate that to compute when there would be room for only one fab in the world. Seems we’re there for one part of a fab.
Indeed. I've heard that there was a lead time of about 12 years on the EUV machines at ASML. I had a tour of one off their facilities once (generic recruitment thingy). Back in early 2000s they were working to correct for errors where the margin of error on their error detector was greater than the error that needed correcting.
(The wave length used was about an order of magnitude greater than the lirhography result; I have no recollection of how they claimed to handle that.)
They are pushing the boundaries of what's possible, using incredibly long lead times. I don't see how any start-up can compete with that, even in the absence of patents.
Canon and Nikon make similar machines too, but they've lost market share ever since ASML released their dual scan system. And at the moment EUV doesn't appear to be a mature enough market to support competitors.
That, and upfront investment costs. Very few companies have a few spare billion to pour into setting up the infrastructure, plus deal with resource deals, regulations, politics etc.. I'm pretty sure that a lot of countries would prefer to be a bit more independent on the semiconductor side, but getting up and running is hard.
> It looks like patents are actively preventing new companies in the semiconductor business.
As far as I know, the really deep technology on the CPUs is actually not patented - it would be nearly impossible to reverse engineer or proof an infringement, but you'd need to reveal your methodology to file the patent in the first place.
There are a number of companies that can produce lithography machines, just not on the same level as ASML (at the moment anyway). Even if we ignore any patent/IP and upfront investment issues, designing and actually constructing machines which are on par with ASML's is not an easy task and will take years of research, assuming the project succeeds in the first place. What ASML is doing now is the culmination of years of research and investment and some pretty specific know-how. It's kind of like trying to replicate NASA's James Webb telescope project: many of the basic principles aren't very hard, but actually building the thing is.
I agree it's a tall order to compete with ASML, patents notwithstanding. So yes patents are not everything, even though cross-licensing is beneficial, rather than invent it yourself.
Actually the James Webb telescope, which has a scheduled launch this year, started out in 1996.
Given the 25 years it has taken so far and a huge amount of money, then it's not too far fetched to imagine someone replicating a space telescope.
It's really not. The cost of production was too high compared to other markets which reduced demand. Which is exactly what happens with all government projects.
The lithography is just a step of the process. I'm just a layman but there's a lot of chemistry involved in just the processing of materials like coating and etching the wafers, particle bombardment, etc. Even the conditions and environment in which the lithography is done makes a big difference.
I suspect the following might be a big part of the secret sauce:
a) Having a culture where EE work is more prestigeous than software work, leading to smart people going there.
b) Having spent decades on scaling up the national university system for engineering educations, starting from the 80s electronics design/manufacturing boom.
c) Having a low salary level, internationally speaking.
The combination of a large, smart and affordable work force for something like this is pretty hard to beat.
What I meant was TSMC uses these machines. Anyone else (except the Chinese I guess?) can also use these machines. What is the TSMC secret sauce that no one else can match?
It's not a single thing, it's everything. State of the art fabs require precision control of an incredible number of process steps each with an incredible number of tunable dimensions. These machines are just one piece of the puzzle. A key piece, but buying the machines does not a fab make.
Think of these machines as being what a paint brush is to an artist; it's just a tool. Anyone can buy a paintbrush, but it takes much skill and experience to paint a masterpiece.
Even if you have a masterpiece in your possession to try and copy, you will fail.
Semiconductor manufacturing seems to have a very big winner-takes-all advantage, it takes enormous amounts of capital and 3-5 years to start a new fab, and the incumbents own a complicated web of secrets and patents that are basically impossible to navigate.
Calibration and following working procedures. The lithography machines have endless knobs to tune. It has to be done well to get a decent yield out of a factory.
TSMC and Samsung are two companies that have competitive technology node in EUV (Intel struggling).
It means they have invented ways to use these machines to print extremely small transistors and that are fast, energy efficient and have good yield (less failures in the process).
- The lenses (actually mirrors for EUV) only demagnify by a factor of 4. So your wafer template is already extremely small and costs millions to produce.
- 13.5nm is pretty much the smallest wavelength you can reasonably handle. If you want to go smaller, you have to build a particle accelerator. (He heard that from a colleague, wasn't sure how much of that was actually true.)
- The prices in this Twitter thread are pretty spot on. The EU wants to invest into chip making to become more independent in this area. But the sums they're talking about wouldn't even pay for one such machine. Politicians don't seem to be aware of the dimensions we're talking about here.
- Also in the thread, ASML has a monopoly on this tech. Others have not invested into EUV and by now, it's pretty much impossible for anybody to catch up.
- Everyone's hiring like crazy atm. He started home office during Covid and now Zeiss says he can't even go back to the office because they don't have enough office space anymore. He doesn't mind. Everyone's distributed all over the place anyway so whether he does conference calls at the office or at home makes no difference.
- He said they're not very affected by the chip shortage.