Hacker News Reader

Comment by detaro | Hacker News Reader

There is no spare capacity to build highest-end process fabs, and there is huge demand for capacity in 22/28nm too.

"spare capacity" ? what do you mean ? Taiwan and South Korea are enough to supply the world demand? Well, it is still hard to get your hands on top-notch chips at a reasonable price whatever the news is saying...

You need certain machines for EUV. There's only one manufacturer (ASML) for those, and they're at capacity. And even if they wanted to scale up, there's only one manufacturer (Zeiss) for the EUV mirrors required for those machines, and all their capacity for the next years is already bought as well.

DUV seems to be economical down to about 7nm though.

From what Intel put out in marketing materials, even at 14nm they had a very unreliable process with DUV. So, it might not be economical for the type of IC produced in these factories.

So it means that chinese SMIC announced '7nm' implies they have EUV working?

No, they don't have EUV. They have 7nm which they're doing with DUV, but they're hitting the same issues Intel did with abysmal yields and there's no guarantee that SMIC will be able to solve something Intel was never able to.

... unless they jump onto the EUV train. That said heard that japan may come back as a top-notch silicium country with... nanoimprint?!

They're sanctioned, they can't just "jump"

come on... ofc they would have to do the whole R&D.

They do have EUV machines, albeit unreliable. Like you yourself mentioned the yields are the problem here.

This is really more of an engineering problem, you need more cycles for quality and reliability. But it's not something unsurmountable. Intel had no reason to try hard enough, Intel is never going to get sanctioned.

'Necessity is the mother of all innovation' might come to play here. If it comes down to living vs dying, people go to no ends to make things happen.

Source? I've not seen anything _anywhere_ about anyone other than ASML producing EUV lithography tools. DUV yes, but not EUV

I am announcing 1nm chip rihht here right now! Trust me I wouldn't lie.

The nm measure is a marketing tool. Intel nm is 'larger' because they were measuring gates differently.

I only take nm claims as comparison of products of same company.

TSMC (and Intel) use DUV for their 7nm.

TSMC's DUV 7 processes are reportedly very good.

And I thought Intel's 14 was good too (after some ramping), which is why they stayed on it and iterated on it for so long, it was the 10nm process they had so much trouble with and the problems there are not generally reported to be with lithography but materials (cobalt, among other things, which they dropped in later nodes).

So I'm not sure what you are trying to argue. DUV has _proven_ to be economical down to about 7nm.

Why can't ASML scale up? I don't understand this. I think because they are a monopoly they can work slower, they don't need the extra profits.

ASML is (now) a company everyone has heard of, but like GP hinted at with their Zeiss comment, their supply chain is crazy deep as well. Getting everything scaled up across companies and countries is far from trivial.

That's true, but this is one of the most important techs we have. From the recent news, I see that TSMC wants ASML to scale up as well. TSMC announced that it will cut its capital expenditure by 10% due to supply problems. This announcement tanked ASML's share price.

My comment is that ASML is hitting their profit targets, they have backlog until 2024. If I were leading ASML or an employee of it, I wouldn't want to scale up. I am already hitting my targets, it's cozy, I have backlog until 2024, I have no competition.

The current situation is good for ASML and bad for us.

My comment is equally applicable for all companies in the supply chain that are monopolies.

Apparently ASML's internal processes are a mess, with everything consistently blocked by technical debt. I suspect this is the biggest blocker to their ability to scale their production faster.

this makes me wish I were like the Harvey Keitel character in Pulp Fiction, so I could straighten my tie, say "I'm on it", and hop into my vintage Mercedes convertible to go fix it.

It’s a Honda and it’s not a convertible. Memory is strange.

US model, so it was an Acura NSX. Don't know if the one in the film had it, but some years of the NSX had a removable targa top.

Frustrating your consumers by not being able to supply their orders is how you stop being a monopoly. Eventually your backlog is long enough that it becomes worth your clients figuring out how to do what you do in-house, or a startup spots an opportunity. Being oversubscribed is a nice position to be in, but it doesn't mean you can rest on your laurels. (Plus, even if you're hitting your profit targets, leaving a boatload of money on the table has gotta be frustrating for at least some of ASML's employees and insiders).

Scaling up too much is a huge risk. Sure, people want a lot more EUV machines right now, but what if they don’t later? You can be left with a lot of expensive capital outlay that is going unused.

The semi industry is known for being very boom/bust so it’s best not to scale up too quick lest it kill your company.

This is a convincing argument.

The governments of the world should be throwing money at that so there's no way the company would go bankrupt. Perhaps 5 years of revenue. This seems clearly worth subsidizing.

It’s not software - they can’t just sprinkle some Kubernetes on it and scale. There are complex global supply chains involving extreme cutting-edge technology, engineering, and research.

They have a build process video. A lot of knowledge is in folks heads as well from what they said - and nothing is "mass produced" really. And it's all at the absolute limits.

Zeiss covers their mirrors and lenses. The smoothness is extreme: "If you were to enlarge such a mirror to the size of Germany, the largest unevenness – the Zugspitze, so to speak – would be a whole 0.1 millimeters high."

Then they have positioning / tilt accuracy.

"If one of these EUV mirrors were to redirect a laser beam and aim it at the Moon, it would be able to hit a ping pong ball on the Moon’s surface."

What they don't say is what the yield is on these. I've heard they have to try and make X to get y that can hit all the specs.

In the machines themselves didn't they have to build in both an electron microscope and an atomic force microscope for defect detection?

And then the environment they operate in is terrible from a wavelength absorption energy / contamination (tin?) issues etc.

I feel this could be applied to the production of EUV mirrors:

https://phys.org/news/2022-10-telescope-mirrors-sharpen-view...

Why didn't they simply build more chips in 2020 instead of shuttering all of those automobile factories?

At some point you're running into actual, real constraints, real bottlenecks in the supply chain, that will take years to resolve. You can't just scale up on short timeframes, no more than you can make a baby in a shorter time frame.

I know I might seem like a flat-earther with this, but I don't understand the chip crisis either. The explanation that many chipmakers went bankrupt after the automobile industry scaled down their orders seemed logical to me. But now that time has passed and we still have a chip crisis makes me think that there is something else.

We know that ASML doesn't ship high tech machines to China, it only ships low tech ones. China in return holds the world hostage by not producing enough chips. With enough subsidy, I think we could ramp up production to meet demand.

I also don't understand that the key companies are monopolys. It all seems a bit planned to me. Place the chip manufacturing company in Europe, chip factory in Taiwan, assemble in China. I feel like someone planned to distribute these technologies and infrastructure to avoid having one country having it all.

I know I cannot support my argument, but this is a gut feeling I have.

OTOH, nobody can reasonably explain why there are unsolveable real bottlenecks in the supply chain and why it would take years to resolve. Throw resources and people at it. This is not just one company with limited resources and a single goal. We should fund the semiconductor industry.

chips are critical for so many things, each "zone" of the world should have its own capabilities to build leading-edge chips, and probably with international collaboration on R&D (like ITER).

I don't think the global economy can "fund" properly more than the current number of actors of this industry, namely each of those zones would have to consider what would be its "local" leading-edge chip manufacturing as a "military defence"-thingy: making money out of it would be optional.

Yes, I agree, but the underlying IP is top secret. No one in the world can replicate it. I believe the leading forces want it to remain this way. (China shouldn't get its hand on high tech UEV machines because they might reverse engineer them, enforce export control laws etc.)

Because they're already working on the next gen (High NA EUV) which are a lot faster than the current EUV machines so developing extra capacity for the current gen will in the medium term reduce supply.

> Well, it is still hard to get your hands on top-notch chips at a reasonable price whatever the news is saying...

it's really not and actually the slowdown is so bad that AMD has had to reduce production, as well as other companies reducing memory/flash wafer starts etc.

it's not just intel or fake news, availability really hasn't been a problem for a year or more at this point, and if anything we're starting to shift into the "glut" phase of the bullwhip cycle.

It depends on what you're looking at. I develop electronics hardware, and it is still incredibly difficult to get a lot of ICs.

Texas instruments still has not caught up, and their chips are used just about everything, including on graphics cards.

yeah that's fair, the embedded market (both power, microcontrollers, and others) is still fucked up, but, the leading-edge market is in oversupply at this point and companies are starting to pull back production hard.

Reply on news.ycombinator.com