Being able to produce state of the art semiconductors is arguably the most important manufacturing ability for a country to have. The US does not want to be dependent on Taiwan or South Korea to build what is nowadays a cornerstone economic driver (compute) and cornerstone defense tool (compute).

So to put it simply; the US wants to be sure it can still make H100's even if the rest of the world goes to shit.

> Being able to produce state of the art semiconductors is arguably the most important manufacturing ability for a country to have.

I don't think that is quite right. It is a very important ability. I don't think it is the "most important" ability. If you have semiconductor fabs but not dry docks to build capital ships you will be in for a world of hurt. If you have semiconductor fabs but not agriculture to feed your people you will be in a world of hurt. If you have semiconductor fabs but not the ability to cast solid-fuelled rocket engines for your missiles you will be in a world of hurt.

It is one of the many important abilities. The reason we are talking about it is not because it is the "most important", but because it is at danger of being lost. We don't talk about the other equally very important abilities (like dry docks for giant ships, agriculture to feed the nation, or solid fuel casting, or a myriad of other things) because nobody worries about those going away.

I am not sure capital ships are as critical now. With drones/hypersonics it seems they are too vulnerable for use in any peer conflict. Pretty sure if China/US were to wage war in 5 years then all the capital ships from both sides in the conflict zone will be scrap within 24 hours. Send 100 hypersonics per ship and one will hit.

The parents point is more that for a functioning state, you need all of those things. If one of those things is at risk externally, bringing that one internal first makes sense.

Carriers are still very important to US force projection and hypersonic missiles are really overblown. We also seem to be readily able to take out existing hypersonic ballistics.

Also a 5 year war with the US and China that starts with a multiple thousands of missles? That’s just going to be a nuclear exchange and last not very long at all if one sides detects any launch like that.

We'll see how the SinkEx on the ex-USS Tawara goes at RimPac 2024.

I'd be surprised if they don't use some ASBMs.

But larger ships are built to take an incredible amount of punishment. It typically takes a heavyweight torpedo to crack them (hence why ASW is a primary skill set for navies). The physics of getting a 1/4 ton+ non-nuclear warhead (torpedo class) highly maneuvering are rough.

And there's a reason the Navy developed and deployed SM-6, and is now adding SEWIP Block III...

https://m.youtube.com/results?search_query=rimpac+sinkex

Just recently a major ship was sunk, though: https://en.wikipedia.org/wiki/Sinking_of_the_Moskva

Did they make mistakes?

Still having a conscript-based navy?

Loading their capital ships decks with exposed, extremely large anti-ship missiles?

And they likely still would have been able to tow it back to port, if the weather hasn't been bad.

Are you asking if the underfunded Russian military made mistakes in manning a ship, building a ship, designing a ship? Surely, it was meant to be rhetorical

Technically, the Russian military didn’t make any mistakes in designing or building that ship since it was inherited from the Soviet Union. Although make the mistake was having it still in operation…

> Send 100 hypersonics per ship and one will hit.

What about all of those submerged ships that will lay waste to their opponent with the weapons they carry?

Sure, subs also have a place. I’m just saying that conventional warfare with high tech weapons favors resource decentralization. Right now things are easier to blow up than defend.

> If you have semiconductor fabs but not agriculture to feed your people you will be in a world of hurt.

The US certainly has that capability - should food supply ever become critical, slaughter cattle, pigs and sheep and redirect the grains used to feed them to the population. The amount of grains would feed 800 million people [1].

[1] https://news.cornell.edu/stories/1997/08/us-could-feed-800-m...

Arguably the most important, not the most important. You can make a strong argument for why it is, is all that I am saying.

Tha vast majority of military chip requirements can be met with old processes. You don't need a 5nm chip with a 13 billion transistors in a F-35 or for the guidance module in a cruise missile.

Very modern chips are more suited for intelligence work no weapons systems per se. The physics of flight, artillery and balistic missiles is pretty much well understood, we don't need machine learning for that. Some modern systems use computer vision as a terminal guidance system, but again, you don't need state of art semis for that.

Even from a designing a chip for ordinance perspective, there's a lot of meat on the bone to make chips just that much more accurate and that much less susceptible to interference. And making those more and more reliable, cheaper, and smaller is also a big deal.

Stuff like this wasn't realistic from systems made in the 90s or even 00's.

https://youtu.be/vY9rJBL1S2Y?si=Dq1QJp90Up6UKLaQ

I'm not a military expert by any means, but I imagine 5nm chips/ML are immensely useful in designing F-35 and cruise missiles.

For CDF (Computational Fluid Dynamics) yes, but ML not so much. This is a relatively new technology and this sector is slow to change.

Add in the fact of the black box nature of ML, and it becomes a pain to anything that requires a certification.

Zero of the chips in the F-35 are 5nm or below. These aircraft use surprisingly old process nodes.

Apples and oranges. The F35 is probably using 15-20 year old technology based on program start. Some cruse missiles use 1970s era technology at the newest based on program start and publicly available documents (since there were some updates).

The defense department is far larger than missiles and planes.

Also nuclear weapons simulation is what keeps the DoE continuously buying new world leading super computers.

You can find context for the recent push here: https://en.wikipedia.org/wiki/CHIPS_and_Science_Act

Thank you, this is exactly what I needed!

Sometimes the fab location might be older but the fab itself might have gone through much retooling throughout the years. I doubt the TI fabs in Sherman and Dallas in 1965/1966 are running all the same equipment as back then.

There are a lot of interesting fabs in the US making some pretty bleeding edge products, but often not digital microprocessor chips. A lot of the more bleeding edge are analog/RF kind of stuff, especially GaN and GaAs stuff.

A modern fab is basically a super, super clean warehouse. The equipment and layout can easily be changed up for a new generation of products.

Intel is costantly modernizing one of its fabs in PDX. From what I have heard from a neighbor working at it, Intel brings most new equipment here, uses it for a certain time, then packages it up and ships to Arizona, so they don't have to pay sales tax on it, since its now used...

Intel gets a sweet break on property taxes too, basically paying a flat fee per year. Otherwise, they could never afford the taxes on $20-40 billion of equipment in a single fab. The state makes up for it by taxing 20,000 well paid employees at a 10% income tax rate.

Something I think I should have better understanding of:

My impression is that automotive and military applications mostly use older, more mature, cheaper modes. Are the fabs for these older nodes mostly re-purposes formerly cutting edge fabs, or do they go around building brand new (higher volume?) fabs for these nodes?

I guess I’m wondering if the capacity to build automotive chips in 10 years will be limited by the ability to build cutting-edge chips nowadays. Or maybe if TI (or whoever) can, like, borrow a couple near-retirement TSMC guys in a couple years and spin up some brand new old-tech fabs.

It depends on the application. Some use an unconventional process because of application requirements e.g. radiation hardening. Special purpose fabs are often subsidized to some extent to keep them running. Also, the US military will upgrade the silicon of existing systems if the situation warrants it e.g. it is cheaper to use more modern silicon than to maintain a fab for the old silicon.

Most weapon systems use old silicon that needs to be robust in all military operational environments. They don't benefit from having more modern silicon. Terminal guidance systems on hypersonic missiles that do kinetic intercept of other hypersonic missiles often use something like a MIPS R3000/4000 class CPU and a DSP of similar vintage.

ISR systems (intelligence, surveillance, reconnaissance) benefit from state-of-the-art silicon because that data is extremely large and analysis is time sensitive. Since pervasive ISR at scale is a cornerstone of modern military operations, having the best silicon and software for this type of computation is strategic. In these cases, it is often the latest commodity silicon.

Since we're talking about chip fabs as strategic military assets, I'm curious how vulnerable they are to sabotage.

E.g., how difficult would it be for a nation state to damage one in a way that requires a major replacement part from ASML, with long lead time?

And with or without it clearly being an intentional act of sabotage?

Intel has older fabs on that list - including one that was built in 2003 - that are capable of 7nm production processes. Some of the older domestic plants have been significantly updated over the years.

It's mostly about having enough capacity so if China goes nuts and takes over Taiwan the world isn't held hostage.