> It does make it baffling why Intel is unable to produce 5nn chips with the same ASML machines TSC is using. It must be management failing Boeing style.
ASML machines do not produce chips, they're just ones doing the etching. There is a ton around what you do with it, and how all the rest of the chip is produced.
They don't come with pre-programmed transistor gate templates. Each fab company has its own libraries of extremely specialized designs, doing different optimizations and tradeoffs. As mentioned plenty of times before, the consensus seems to be that Intel tried to pile up too many new improvements, which backfired. TSMC took a more incremental and safe approach. You have to appreciate that a defect that impacts high-volume yield today could have its roots in a design that was put on the roadmap years before.
> As for the rest of Europe. Im sure they can politically do this. But where will the money end up?
If H2020 is anything to go by, the majority of that money will be salaries of people working on big research projects. At worst the project outcome is useless. I don't see how any of the way these projects are structured benefit the pocket lining of managers.
A point of clarification.
ASML does lithography, which creates the patterns for the chips. The big players in etch (removing layers of material following the pattern) are Lam Research and Applied Materials, both American companies.
And yet it's enough to spark a geopolitical war to block China from acquiring these machines[1], which basically means China has massively ramped up it's EUV investment if you look at semiengineering investment news.
At the same time you can trust that TSMC secrets are no longer secret[2]. A lot of the Taiwanese companies are actually quite bad at security despite Taiwan have some top notch security engineers. I guess it's just old industries general problems.
Sorry but it looks as if you had a view and then googled articles to back this up. The second article is about IP theft and gives no reason to doubt TSMC's integrity. Nor does it give any reason to think Taiwanese companies are worse at opsec than e.g. US or Dutch players.
Sorry it's you that didn't read it. China has had a massive cyberop and stole pretty much every secret TSMC had. It doesn't matter whether Taiwan is trustworthy or not if their cybersecurity sucks. Good job for not reading, not understanding and then using a random fallacy to respectfully diss me.
> the consensus seems to be that Intel tried to pile up too many new improvements, which backfired. TSMC took a more incremental and safe approach
Can you point to some industry literature that explains this?
If this is true, it is a lot more surprising that Intel is talking about going fabless. You build a lot of expertise by rolling the dice on new techniques that don't pan out on the first iteration. People wouldn't be so pessimistic about Boeing if they had just been overly aggressive in a new airframe design, rather than failing to safely mount new engines on an old design.
The consensus on HN (search for Intel 7nm, cobalt, ...). It's all speculation at this point. I doubt that Intel would publish details on their processes and its defects so soon.
Intel buying some TSMC capacity does not mean it is going fabless. Nor does it mean Intel is abandoning being vertically integrated.
I like your Boeing remark. It's better to fail at doing something aggressive than to fail to innovate.
> It's better to fail at doing something aggressive than to fail to innovate.
I don't think that's a particularly useful way to divide up failures. I think it's better to fail in a way that doesn't kill people than a way that does. Whether you kill people by being too conservative or too aggressive with your design is pretty irrelevant compared to that.
It's paywalled but Charlie later published something else referring to that article as talking about COAG problems so I don't feel bad sharing the info at this point.
> People wouldn't be so pessimistic about Boeing if they had just been overly aggressive in a new airframe design
Well... yes. That's called judgement, and execs at firms like Boeing get paid positively stupid sums of money to provide the good kind. Which they didn't.
"What's the big deal? If I had left the vat of acid unattended in a locked room, everything would have been fine."
Are they talking about going fabless? I thought they’ve only discussed making some parts at TSMC, not all of them.
They’re not on the first iteration at this point. They’ve been on 14 nm since 2014. 10 nm and to a lesser extent 7 nm is failing after many iterations and many years.
>People wouldn't be so pessimistic about Boeing if they had just been overly aggressive in a new airframe design, rather than failing to safely mount new engines on an old design.
Unfortunately Boeing doesn't have a huge list of customers that are willing to spend what it costs to switch to a new airframe. They sell to airlines what airlines want to buy. Airlines wanted a 737 with higher bypass engines and that's exactly what they got.
Boeing deserves what they get for pushing a faulty design out the door but they didn't do it as some nefarious cost cutting measure. They did it because American and Southwest said they wanted nothing to do with a "797" and threatened to switch to Airbus.
> they didn't do it as some nefarious cost cutting measure. They did it because American and Southwest said they wanted nothing to do with a "797" and threatened to switch to Airbus.
ASML machines do not produce chips, they're just ones doing the etching. There is a ton around what you do with it, and how all the rest of the chip is produced.
They don't come with pre-programmed transistor gate templates. Each fab company has its own libraries of extremely specialized designs, doing different optimizations and tradeoffs. As mentioned plenty of times before, the consensus seems to be that Intel tried to pile up too many new improvements, which backfired. TSMC took a more incremental and safe approach. You have to appreciate that a defect that impacts high-volume yield today could have its roots in a design that was put on the roadmap years before.
> As for the rest of Europe. Im sure they can politically do this. But where will the money end up?
If H2020 is anything to go by, the majority of that money will be salaries of people working on big research projects. At worst the project outcome is useless. I don't see how any of the way these projects are structured benefit the pocket lining of managers.