The current most advance fabs are place in Taiwan ( along the coastline which is nearer to China ) if those fabs are destroyed, there will no longer be newer CPU generation for a decades due to the recipe to make these chips are lost in war. Companies which relies on the advance of semicon will halt but will eventually recover as other players ( Intel, SMIC, Samsung or the parts of what’s left of TSMC ) catches up with progress made in the fabs in Taiwan.
There are really 2 tiers of semiconductor manufacturing, and for the advanced nodes Taiwans TSMC is in the lead with Samsung close behind, and Intel is lagging. The less advanced (aka older, larger) nodes manufacturing capability are more spread out.
Those fabs are a fairly significant part of the whole situation.
China wants them because they have those sweet ASML EUV machines that they are sanctioned from buying.
The US needs those fabs to stay controlled by Taiwan because they don’t want the world’s most advanced contract fab to be overseen by a hostile country.
Taiwan needs those fabs because it makes them important for the US to defend. And because those fabs are so important to everyone to who wants high end chips, just having them is a good defense.
I remember reading somewhere that Taiwan has rigged all the TSMC fabs to explode and flood in case China invades it. The fabs however are less important than the innovation and technical knowledge to run it and is stored elsewhere and also with the human resource. These can be quickly transported to ensure they can setup elsewhere and prevent the tech from falling into China's hand.
The machines may be rigged to explode, flood, or otherwise hose themselves, but active sabotage is largely unnecessary. If China takes Taiwan by force, two things will happen: 1) the US will heavily sanction anyone who trades with them, so even if they manage to keep the TSMC fab online it won't do them any good in a commercial sense; and 2) without ongoing service and support from ASML and many other Western vendors, the equipment will work for about as long as the leased aircraft that Russia stole from Airbus's and Boeing's customers.
They might get six or twelve months of production out of TSMC's facility, and then things will start to break that they will not be able to fix easily. Invading Taiwan will cause economic damage measured in hundreds of billions of dollars at a bare minimum, and much of that damage will be suffered by China.
Bottom line, it would be insanely stupid of China to attack Taiwan. Of course, many people (including myself) said the same thing about Putin and Ukraine, so...
>and much of that damage will be suffered by China
Signiciantly more will be suffered by US/west whose advanced industries extracts disproportionate patent fees and have extreme reliance on TW semi for short to medium term. We're talking trillions in knock on effects. Proximity of semi production / supply chain concentration in TW and broader east Asia is PRC's trump card in war. During peace US has trump card in sanctions / tech control, but during war PRC has leverage of simply setting the advanced economies back 30 years which will actually close relative gap. PRC might even pull ahead if they manage to mass rollout of 28nm indigenous fabs and complete indigenous semi supply chain before anyone else. That's without mentioning during war, PRC has incentive to attack SKR/JP fabs to draw US into first island chain where they're weakest to honor security commitments. If things ever get heated with US, it would be insanely smart for China to threaten TW/JP/SKR as well. And if PRC attacks TW, I surmise there's going to be three way side deal between PRC, US and TW leadership to maintain TSMC production continuity in exchange for evacuations for TW leadership to US who would otherwise get the noose for treason. Which itself makes the notion that TSMC is rigged to blow counterproductive - surival of TSMC is TW elites golden ticket off the island - US isn't going to look kindly to a TW admin that explodes US economy and tech lead.
People can live without smartphones, but can't live without food. In case of war, westerners will need to postpone buying some electronics or clothes, but the Chinese will resort to severe food rationing.
That is old data, today China produces more than enough to feed its population. They produce 650 million tons of grain a year which is more than enough for their population, and is a net exporter of meat.
To make it easier to understand, total world grain production is about 2250 million tons for a population of 7.8 billion. At worst China would need to ration more expensive meals, but people wouldn't starve.
Someone else addressed per capita self sufficiency below. But operationally, this strategy is likely only if US has ability and appetite to enforce blockade from other producers. And likely only viable short term before PRC finishes conventional deterrence. As we can see with UKR, most of the world ignored sanctions on food and energy, including west. Keep in mind PRC is buying a lot of ag products from US because of Phase One trade deal. Medium long term, they'd rather decouple and build up other producers.
Another Q one should ask is how vunerable US is to comparable counter action. Because US agriculture is highly industrialized and automated whereas PRC ag is in the process of automation and still reliant on labour. Sufficient cyber attack on US ag infra during harvest season and US is forced to ration. What happens to automated John Deere tractors when they have no GPS to harvest. Same with familiar threats of embargoing PRC oil because US is self sufficient on the metric that there's enough fossil fuel resources in the ground. US is only as self sufficient as how much it's 130 refineries can produce. If US interdicts PRC energy, don't be surprised if PRC takes down comparable levels of production state side.
Abandoning trade with China will hurt us, certainly. But unless the Xi regime is overtly suicidal, that's all that will happen, and we'll eventually get over it.
As for cyberattacks from China, those can only do so much, and in any case they can be mitigated by cutting the fiber. OTOH, shooting down GPS sats will lead to war. They know better than that. I hope.
Legally, so is blockading PRC, including food and energy imports which is not an unfamiliar discussion in US policy circles. As for cyber, US DoD / Biden also explicitly communicated to both PRC and RU that cyberwarfare on critical infra = actions comparable to war. I don't know if cutting fibre is a joke, there's millions of PRC diasphora, unless US disconnects from internet globally, that's assuming PRC agents aren't state side already. So the real question is, does the US know better than to intervene when PRC attempts armed reunification with TW, which seems increasingly likely while bills are being drawn in congress to boost TW into major non-Nato ally? Because if US does, there's no telling how things will escalate, especially when modern connectivity and increasing PRC missile capabilities means CONUS is no longer safe. Incidentally, I think both sides knows this, on paper but frequently, better judgement gets overriden by domestic politics.
It's unlikely China or Russia, or anyone, has the capability to shoot down GPS sats. A number of countries, the US included, have demonstrated capability to shoot down satellites, low orbit satellites. GPS satellites are in geosynchronous orbit, which is a whole different species of animal. Shooting down an LEO satellite can be done with an SM-3 missile which has a mass of 1.5 tons. That's significantly lower than the mass of a rocket needed to put the satellite in orbit, and that's possible because SM-3 is not designed to reach orbital speeds, only orbital heights. Its primary mission is anti-ballistic missile defense and ballistic missiles can reach thousands of kilometers altitude, but not tens of thousands.
In theory one can design a missile that can reach GSO heights without reaching GSO speeds, but that requires a dedicated project, which would be extremely expensive, and very difficult to keep under wraps. All missile launches are pre-declared to minimize the risk of a nuclear war. You can find a list of all orbital and suborbital launches on wikipedia (e.g. [1]). If anyone tried to test a missile that can reach GSO heights, they couldn't keep it a secret.
One can then try to shoot down GSO satellites with full size rockets, like Falcon Heavy. But there are 31 GPS satellites out there, and 30 Galileo. I doubt either Russia or China (or even America) have 60 full size rockets available to do a blitzkrieg strike against the GPS satellites.
I remember reading somewhere that some country is working on laser weapons to knock out satellites. That makes more sense than using a missile as all you want is to disable a satellite, not necessarily destroy it and add to the space junk ...
The US heavily sanctioning China would basically be the end of America's economy, it isn't a realistic option. There also won't be enough public support for it as long as conflict is contained to the APAC region.
Furthermore, it probably also means the disintegration of US global sanctioning power. Europe would rather trade with China than the USA, assuming it's mutually exclusive.
Pick any Walmart, Target, or other big box store and see how much of their stock is not Chinese made. Cutting trade would basically shut all of them down and there would be a revolt.
> 2) without ongoing service and support from ASML and many other Western vendors, the equipment will work for about as long as the leased aircraft that Russia stole from Airbus's and Boeing's customers. They might get six or twelve months of production out of TSMC's facility, and then things will start to break that they will not be able to fix easily.
Right. Fab tools require constant maintenance. I'm sure they keep a stock of repair kits and spare parts, but probably no more than a few months supply. Their supply chain ops (procurement and engineering) sets min/max stock levels of these parts, but I'm willing to bet that there's no more than a few months supply on hand. And then equipment gets dirty, wafers get scrapped, yield plummets, etc.
Serious question: why is the US trying to (or is already preventing) China from acquiring EUV chip tech? What’s the reason for such an aggressive intervention in a free market?
Which is why forced birth is making a comeback - the only way to ensure local production now is to ensure you have a permanent underclass you can squeeze for your profits.
As globalism unwinds, there will be an open desire to roll back child labor and environmental laws as well as birth choice.
We'll get factories back in the USA - and wish we hadn't.
Or we'll see a massive expansion in prison labor, even worse than we have now.
> Which is why forced birth is making a comeback - the only way to ensure local production now is to ensure you have a permanent underclass you can squeeze for your profits.
That makes no sense. If the right was trying to build an underclass of workers, then the easiest way to do it would be to not build a wall at the southern border.
The simple explanation here is the correct one: they oppose abortion on social grounds.
Just earlier this year they were complaining that Chinese companies were ripping off their IP [0] and IIRC even within the last 2-3 years they were themselves hesitant to send bleeding edge tech over for similar reasons. I wonder what has caused their tune to change.
That said, if I'm reading the tea leaves properly...
I think the last couple years were a wake-up call to the US that not only are they overly dependent on other countries for critical components, but that China is far closer to becoming 'self sufficient' than us (1).
Ukraine was a bit of a wake-up call to us, that we are too afraid of nuclear escalation (2) but also to China that embargoes can happen more quickly than you'd expect and hurt almost as fast.
What this means however, is we are now both in a race for self sufficiency on this front.
... Really random question, I wonder why Mexico doesn't have more fabs?
----
1 - I think the US has a weird definition of 'self sufficiency' mind you. For instance in some cases it seems the attitude is 'Oh we did that again and can ramp up.' As a general comparison, I'm betting the possibility of getting a '100% China content' vehicle is far higher than getting a '100% USA content' vehicle.
2 - Without getting into it, there was a short window early on in the Ukraine conflict where I would have doubted Putin's willingness to go nuclear. IMO It's one of those cases where the would police card may have been worth playing, were it not for our missteps in using it over the last 20 years.
> These can be quickly transported elsewhere to ensure they can setup elsewhere and prevent the tech from falling into China's hand.
Wouldn't it still take quite a long time to get a new fab to the point of being productive? These machines aren't exactly sitting in storage by the hundreds, and I imagine you'd also need to dial in all the processes pretty well to get good yields? (though I have no idea what that entails...)
>Do I understand correctly the much of the worlds semiconductor fabrication is in Taiwan
Not straightly true. The worlds "leading edge" Semiconductor Fabs is located and dominated by TSMC. Yes. But Semiconductor as an industry includes many things not leading edge like small chips used in appliances as well big volume commodities like NAND and DRAM. Those are mostly in South Korea, Japan and then US. ( Along with Singapore and Germany in case anyone wants to nitpick. )
If TSMC get destroyed, it will likely set back the whole Semiconductor industry by 10 years. Simply rebuilding the Fabs capacity, leading edge or not would take 5 - 10 years with perfect project planning and execution. Not to mention no player in the industry has shown they are capable to providing similar quality, yield, services to TSMC with similar prices at an acceptable margin.
I don't understand what part of the sentence you've quoted makes you think that TSMC hasn't put a lot of thought into their price points already.
Regardless of how true quoted sentence is, it's also obviously true that if they changed prices to $1billon for every single chip sold then they would sell close to zero chips (ignoring that such a crazy move would probably see an attempt to replace leadership, or a takeover attempt, or whatever). And equally, if they lowered their prices to give each sale only 1% of their current profit margins, they would sell more chips.
Every time they set a price, they'll be thinking about where the line is to get that balance right. And they already know they're the industry leader, and of course they know it in a lot more detail than that simple statement, so they're taking that into account (along with things like demand, how close the competitor products are, etc.)
Of course that's not to say that they've necessarily made the ideal choice every time, and maybe if they raised prices they would make more money overall. But how does the fact that them literally disappearing would set back the industry make this more or less likely to be the case?
Why are you convinced your armchair strategy is superior to decades worth of analysis by one of the most successful companies in the world?
TSMC has no desire to become Apple because they already capture so much of Apple's revenue. I guarantee you they have put enormous thought into the exact optimal price point, as building fabs is extremely capital intensive.
I stand by my comment - I could write the exact same response to the paragraph you added as to your original one (with minor tweaks from "changing prices" to "entering new business areas").
They can make more by selling higher volumes at a lower margin.
> Or why don't they build a consumer brand? They could fade out Apple if they start building their own consumer products.
There is a lot more going to Apple's products than TSMC's fabs? To say nothing of their business with MediaTek, AMD, Qualcomm, Broadcomm, Nvidia, Sony and others [1].
This is the same failure Intel fell into, compounded tenfold. (If we build best, we should eat the designers' cake too.)
> They can make more by selling higher volumes at a lower margin.
Higher margins are made closer to the consumer.
> (If we build best, we should eat the designers' cake too.)
Apple is eating software developer's cake ... (well, 30% of it).
The greatest businesses on the planet all rely on brand recognition. TSMC has none and relies only on its technological expertise which at some point might be surpassed by a competitor. There is a reason why Apple wants to own the entire supply chain, and they are smarter about it because they start at the other end (close to the consumer) where margins are larger.
Which is consistent with and irrelevant to making “more by selling higher volumes at a lower margin.” Higher margins are made by Michelin-starred restaurants than TSMC. That doesn’t mean they should expand into fine dining.
Chasing margins at the expense of profits is bad business. Cannibalising Apple would mean losing volume, from Apple and others, which in turn undermines TSCM’s scale-based advantage.
This isn't about Michelin restaurants versus McDonald's. This is about McDonald's versus farmer joe who makes the best potatoes which are essential to McDonald's business.
If tsmc were run by Apple's board, we would all be paying 30% of our income for the privilege of using modern computers. A much better strategy, just look at Apple's balance (and Apple products aren't even as exclusive as tsmc's).
> were run by Apple's board, we would all be paying 30% of our income for the privilege of using modern computers
This was IBM. For fundamental, structural reasons, they were dethroned. TSMC doesn’t have the option of eating Apple’s cake without giving up the scale that makes it competitive. Again, it’s similar fresh-MBA thinking that tanked Intel. (TSMC wouldn’t exist hadn’t Intel given it the neutral market space.)
"> were run by Apple's board, we would all be paying 30% of our income for the privilege of using modern computers
This was IBM. For fundamental, structural reasons, they were dethroned."
I think they did it to themselves, first by eschewing virtual memory in the System/360 and first models of the System/370, thus losing the higher education market and what people eventually moved to like UNIX, thus their big machines become a specialized niche of the whole computer market. Then in moving from a preference of leasing to sales which gave them a one time goosing of revenue which they didn't prepare to lose when the cycle of that ended.
They did do well in PCs for a while, but the unit that did that in Florida eventually got blue washed (there's lots more to the story of course).
"Again, it’s similar fresh-MBA thinking that tanked Intel."
Also very unsure of this, I don't see that resulting in the 10 nm now Intel 7 debacle. More like their decades long high level very bad engineering management eventually caught up with their crown jewel of fabrication. The not so fresh MBA I assume CFO who was made an interim CEO actually had a plan which might have succeeded.
I don't think the current CEO's plan has much if any chance at success, especially since I don't see him making the brutal moves necessary to change the company culture to work as a foundry. In particular, I note while he worked for Intel for a long time, he's a design instead of fabrication guy, the latter is what they needed and I wonder if they couldn't hire anyone good from that field.
> they did it to themselves, first by eschewing virtual memory in the System/360 and first models of the System/370, thus losing the higher education market and what people eventually moved to like UNIX, thus their big machines become a specialized niche of the whole computer market
Yup. They went from neutral generalists to niche specialists.
TSMC is, today, a neutral generalist with the scale of Apple, AMD, Qualcomm, Samsung and the others combined. You’re (EDIT: Nope!) advocating trading that for Apple’s profits. Short-term versus long-term greedy.
> McDonald's versus farmer joe who makes the best potatoes
Replying to a parent comment of yours (EDIT: Also wrong!). McDonald’s has scale. TSMC has scale. Farmer Joe competing with McD’s makes sense. Tyson competing with McD’s does not; it will lose, in the long run, McD’s as well as other restaurant chains as customers with that play, undermining its home-turf edge while it competes where it has none.
"You’re advocating trading that for Apple’s profits."
Look at who you're replying to, I'm not advocating that at all!
The Semiwiki guy credibly claims TSMC has reached this worldwide pinnacle of chipmaking in part by being trustworthy and creating an ecosystem based on trust. That of course includes being fair with your customers, which has allowed the latter to fund the creation of new fab lines and nodes knowing TSMC won't screw them over once that money is spent.
> This was IBM. For fundamental, structural reasons, they were dethroned.
They were dethroned because they opened up the market to clones, and allowed Microsoft to gain control over the software side.
See, this is about control. TSMC has less control over the market than their clients do. And if they don't act on it, it may become their demise, just like with IBM.
It's an internet meme that some goof at US Army War College picked up on and wrote a paper suggesting TW should blow up their own fabs to deter PRC invasion. Which is dummy take since for PRC, TW is a political problem and US can always strangle fab operations via supply chain sanctions. TW media saw this as serious traction and pushed some articles basically saying "leave TSMC alone". TW, at least TW leadership won't blow up TSMC for the simple reason that it'll destroy US advanced economy, and if they ever want a safe ride off the island to retire in US, it would be in everyones interest the fabs stay untouched.
Does anyone not think those exercises were designed to send a political message, specifically in response to US politicians visiting Taiwan?
Yes they were military exercises, but quite blatantly it wasn't China actually needing to do any testing or training of their military, except using it to send a political message (both domestically and internationally).
(Which isn't my giving an opinion on the likelihood or not of China ever using military force against Taiwan, just a response to your ridiculing the idea that recent actions weren't political).
The economic damage a war would cause is immense. TSMC however would contribute surprisingly little to the economic damage.
TSMC has a big volume and an edge in node process. However, the vast majority of chips in use are in embedded usecases. Most non-general-computing hardware containing chips doesn't need 4nm. It would chug just fine with two decade old chips. Even military hardware uses very outdated chips.
Catching up to a modern semi fab would cost time and money, but making more outdated chips isn't that much of a problem. There's a certification issue, but in such a scenario surely certification would be relaxed.
As for general purpose computing, servers always have the option of scaling horizontally. The biggest issue would be with phones and tablets. Unfortunate, but tiny compared to the other costs of war.
It would mean no new gaming consoles, Apple devices, AMD CPUs, high performance FPGAs, GPUs by either NVIDIA or AMD, WIFI chips from Broadcam, no Qualcomm ARM CPUs, maybe also no cheap Mediathek CPUs for smartphones.
IIRC, NVidia and Qualcomm fab at Samsung. Others would have to compete for limited remaining quantity or use slightly inferior chips.
All this difficulty with high end smartphones and gaming consoles is completely insignificant compared to the 'normal' economic damage a war would cause.
NVIDIA does some manufacturing at Samsung however certain products are only made on TSMC. Such as the coming RTX 4000 series which is made entirely on TSMC 5nm.
I missed some announcement though that Qualcomm is moving more towards using Samsung.
> All this difficulty with high end smartphones and gaming consoles is completely insignificant compared to the 'normal' economic damage a war would cause.
This 'difficulty' would set Moores Law back more than half decade outside of Intel and Samsung smartphone chips. The next smallest large scale node being 12nm at Global Foundry and keep it frozen there for at least 2-3 years in total leaving as decade behind Moore's Law. You underestimate how much damage that causes. Even if China completely wipes out Taiwans 1% share of GDP that impact is miniscule compared falling 10 years behing on Moore's Law.
> Even if China completely wipes out Taiwans 1% share of GDP that impact is miniscule compared falling 10 years behing on Moore's Law.
I could not tell if it was a tongue in cheek statement, but if not, can you explain why falling 10 years behind on Moore's law would be such a big deal.
Moore's Law is dead. We were mostly stuck on 14nm for years and... it had no obvious effect on economic growth. A better way to put it: What kind of application must have a very small node and would not be able to run economically without a tiny node?
Embedded doesn't care all that much. Servers can scale horizontally. What's left is various consumer products, and even here you can easily make a smartphone on 12nm, it would just be slower. Hmm.. Maybe VR and locally-run ML must have a tiny node size. Hardly big markets. For all else, there'd be a temporary efficiency loss, but we'll make do and catch up within a decade. You could argue delaying the future would have a running effect, but it seems to me mostly leisure activities would be affected and investment/production would be fine?
The Moore's law effect is insignificant compared to the damage of a conflict between the world's #1 manufacturer and #1 consumers.
Across all industries? The supply chain is looking, and TSMC isn't the only fab in Taiwan. It would also impact car production, factory machinery production, solar panel production, PC production, medical device production, etc...
And all of this happening across almost all industries simultaneously? That's not a recession event - that's a depression event.
I'm guessing they have some less modern stuff. Maybe.
But. BUT.
UMC is kinda an unsung player here to some extent. They are another Taiwanese Fab, and do a lot of the older process stuff.
I'm guessing that whatever plans Taiwan's government has for TSMC Facilities in a Kobayashi Maru scenario with China, UMC probably gets some version of it as well.
Lets also assume that if things get to that state with Taiwan/China, at minimal we'd probably have a larger embargo against China (another source of such chips) and there's a possibility of trade disruption with other Asian players like South Korea.
This makes no sense. The economic damage of removing over half the chip market in a short period of time would be immense. Even if we don't need high end chips, which we absolutely do need, Taiwan and TSMC still make plenty of all sorts of chips that go into every day devices. If TSMC falls, along with Taiwan, we're looking at the cost of everything electronic soaring. It'll make the run on graphics cards due to crypto miners a quaint memory.
The older tech is way more tractable, and there will also be chip recycling to help. Economics will ensure supply catches up, except the highest end chips.
I find the focus on chips bewildering, since there are so many more critical issues like 'how do we prevent a stock market run and the crash of every company which outsourced manufacturing to China?' or (if you live in China) 'where do I get a food ration card?', if this gets real bad. Semis are merely an annoyance.
The other issues are important as well, but semis run everything we use. For instance, nearly half the world's display drivers are made in Taiwan. These can be made elsewhere but will take time to find production capacity.
Do we have the technological capability to recycle old semis to work with current systems to keep things running? Do enough people have this skillset? I'm sure we figureit out, but I doubt it there are enough people to really get it moving. With a severe lack of chips we'll see systems falter or all out crash as components wear out and can't be easily replaced. I think you're underestimating how much runs on what Taiwan produces and I don't think "economics" can bring more production online in a meaningful time frame.
Semis run everything, but: they wear out slowly, there's a long line of consumer semis for reuse, and few systems require the latest semi tech; There's extra capacity in the West for the older semi tech.
It's understandable HN community would focus on semis, but compared to 'financial crash worse than 2008' and 'Would all people be able to get food next month?' semis aren't that important. Semi production would be a slow long-term crisis we would be able to adjust to if we could handle effectively the initial very big short-term crises resulting from a massive war.
As Keynes put it: "In the long run we're all dead". If we couldn't handle the initial short-term economic crises we'll have bigger problems than long-term semi production adjustment. If we can, we could handle adjusting semi production and use. It's not that it isn't an issue, it's just way less important and less difficult than the other issues.
it's not. South Korea has fabs as well and Japan too in Asia. China is also making progress in developing the own nodes (with TSMC's help for older nodes), and Europe and the US have significant capacity at higher nm nodes. TSMC is getting a lot of press because they manufacture the leading edge nodes at the moment, that are used in high profile products.
At least in theory both 20A and 18A are significantly superior to what TSMC has planned for its initial N2 node. Backside power delivery is the largest single factor but Intel will also likely have a density advantage on top of that. TSMC's N2 is a very conservative node.
Um, can we just talk about feature density… as I understand it backside power is a huge breakthrough and Intel will probably be a year before TSMC get to similar feature size with technology that is much more uncertain.
We don't know how good Intel 18A is yet. It could be equivalent or only slightly better than TSMC's N3E or further enhancements of TSMC 3N. Keep in mind that the
base TSMC N3 node is scheduled for volume production in Q4 2022. So the TSMC is still likely quite ahead.
This is all assuming that everything goes well for Intel. If 18A is simply better than TSMC N3E, then TSMC won't have anything to compete until 2025 - 2026 when they release N2.
For TSMC, they have real paying customers that need to plan years ahead so all their date estimates are very conservative. For Intel, they're desperately trying to catch up so they're much more aggressive in their estimates.
According to Anandtech, TSMC's N2 could come in 2025 if things go well, which is roughly the same as 18A. So basically, if things go PERFECTLY for Intel, they'd be in the lead for a few quarters - not 2 years like you said.
It can happen as these things are mostly about money. They stopped aggressive investing in certain areas to save money and had a lot of stellar quarters in succession.
Even TSMC expects Intel to get back on track in a few years because, again, they started investing again. Please note Intel is about to get huge subsidies from the US (~50b) and EU (~20b) real soon. It's free money for their precious fabs and R&D. I am very much against the EU giving them anything but a kick in the nuts but oh well, gotta keep the US and Israel happy
We already buy everything from China. Pretty much everything you own in your home is made in China already, including the computer you're using to browse HN and the router you use to connect to internet. If China and the US severe relationships, the world will go into a depression anyways.
Intel CEO Pat most certainly finessed the government into giving Intel a huge payout by exaggerating this threat.
There is a crucial difference between IC fabrication and almost everything else: we don’t even have the capability to fabricate at the most advanced nodes. On the other hand, we are capable of manufacturing everything else; we just need a bit of time to set things up and ramp up capacity.
If we don’t gain that capability now, it will take us too long to recover when push actually comes to shove.
The USA developed the technology, holds the patents, and licensed them to ASML. They basically are our machines. That’s why Trump was able to block the China deal.
> we don’t even have the capability to fabricate at the most advanced nodes.
I'm not sure that's entirely true. If we define "most advanced nodes" as 5nm-class fabrication, certainly Intel is doing that well now, just not in volume production. Intel 4 based Meteor Lake powered on and runs Windows as of earlier this year.
No capability to build and not ready for volume manufacturing aren't really the same class of problem.
So Intel is 2 years behind volume production of 5nm by TSMC and Samsung. To me, that sounds like a significant delay in capability.
Anyways, I think the CHIPS Act was the right move at the right time. SMIC is quickly catching up thanks to near unlimited funding from the Chinese government. We need to play the same game if we ever want to catch up to TSMC.
> We already buy everything from China. Pretty much everything you own in your home is made in China already, including the computer you're using to browse HN and the router you use to connect to internet. If China and the US severe relationships, the world will go into a depression anyways.
Not as bad as if Taiwan's exports to the USA are stopped.
We can build motherboards / pick-and-place machines in the USA. We literally cannot make the chips that are coming out of Taiwan right now. Samsung is the #2 IIRC, so maybe South Korea can pick up the slack in a hypothetical Taiwan-conflict.
> Intel CEO Pat most certainly finessed the government into giving Intel a huge payout by exaggerating this threat.
US Government is also got TSMC (Taiwan's semiconductor fabs) to build an Arizona fab-lab. Look, even Taiwan is scared of possible conflict and is hedging on the physical safety of the USA.
Hopefully conflict doesn't happen, but the powers that be are making smart moves here. Hopefully it will be enough.
"US Government is also got TSMC (Taiwan's semiconductor fabs) to build an Arizona fab-lab. Look, even Taiwan is scared of possible conflict and is hedging on the physical safety of the USA."
I don't think that's the case, as I understand it this comes from the US wanting to onshore some of TSMC's capabilities and Taiwan screwing up electricity production on the island.
It does Taiwan the government and ruling class no good whatsoever to place assets beyond the PRC's military reach, rather the incentives are entirely opposite of that. The more valuable TSMC's fab lines there are, the more incentive the US has to help with the ROC's defense, or war if it comes to that. Which in the case of torpedoes from attack subs can even be done with some plausible deniability.
The government and people of Taiwan understand that TSMC is a major reason mainland China hasn't bulldozed their tiny island yet. It's their ace in the hole. They have a monopoly on arguably the most important technology in the world right now, and the US and EU will protect them as long as they hold that monopoly. There's no way they'd actually physically offshore that golden goose. It's a brilliant move by Taiwan, and I highly doubt they'd voluntarily give it up.
> There's no way they'd actually physically offshore that golden goose. It's a brilliant move by Taiwan, and I highly doubt they'd voluntarily give it up.
5nm Fab is being built right now in Arizona though by TSMC. They're beginning to offshore fabs.
I'd assume TSMC chose the practical optimum there: given current events, having all leading fans physically in Taiwan was no longer a tenable option.
So they re-balanced for a volume play.
If some (read: national security volume) fabs in EU / US are enough to assuage governments, but most (read: global economically important volume) are still in Taiwan, then that's the next best thing.
It will be if China invades Taiwan. They will not be able to keep the TSMC facility in business for very long, assuming it survives the attack to begin with.
Where life will get interesting is in South Korea, at the Samsung fab. They will have the only 3-nm process in the world if TSMC goes away.
Money doesn’t smell, i.e. it does not have an intrinsic “purpose” to it. More exactly, the money that Intel would have been invested out of their own pockets in order to build those fabs has most probably now been diverted to those dividends, because there’s new money for those fabs being provided by the taxpayers.
It is absolutely possible to earmark government funding and arguments to the contrary are cynical at best and deceptive at worst.
You can offset internal Rd investment with gov money and use that elsewhere, sure, but when the magnitudes are large and immediate, that doesn't play. You get one check, and it better be spent on the intended cause in X fiscal years or you pay it back.
The tech community has suffered under the Cost Plus and other workaround bullshit for long enough that the cynicism is inbred. But there's plenty of gov funded RD money that does precisely what it was invested to do.
There’s no cynicism about it (not that cynicism would be intrinsically something bad, but that’s another discussion), but that’s the thing, the Government saying and actually checking that X sum is spent by a certain company on objective Y doesn’t make any sense, because once the X sum of money enters the company’s coffers its “substance” changes, it isn’t “X” anymore, it is “drowned” in the anonymity of said company’s available (or achievable) funds (funding). Again, I have no doubts that the Government will actually check that the money marked as “Z”, money which had been given to said company, is actually spent on objective Y, but it’s all illusory, because that money has stopped “being” Z once it entered the company’s accounts.
To put it another way, it’s similar to the futility of those well-meaning people who give some homeless person $5 on the express condition that they spend those $5 on food, not on alcohol or drugs. Which said homeless person is all too happy to do, he/she will gladly spend those gifted $5 on food, which leaves the other hipotethical $5 which he/she had in their pocket all available for buying drugs or alcohol, money which otherwise would have been spent probably half and half (2.50 on food, 2.50 on drugs).
At least in the counter-example I gave the result of largesse is a homeless person better fed, which is good (and also more intoxicated, which is not so good), but in this example with Intel you have taxpayers filling the pockets of one of the biggest tech companies in the world.
is this spoken from experience or assumption? Your examples sound like you are assuming that money sent to companies is sent like a check, and no further action is taken.
I cannot claim authority over all instances of gov funding, but I have managed funds from the government for Rd use and they absolutely do audit, and they absolutely do require an accounting down to the last penny, and they absolutely do review plans vs actions. And they absolutely do get refunds for money not spent. And, critically, they absolutely do shutdown follow on funding if they do not like what they see.
And after two months more high na will be sent to tsmc , one machine is just a testing machine , for production many are needed this is just a pr stunt
It isn’t. TSMC has multiple customers lined up with the volume required to ramp these processes to high volume, Intel only has their own designs and whatever foundry customers willing to be an Intel guinea pig.
TSMC also has extensive experience with these machines compared to Intel.
Pat is just making shit up and hoping their stock price doesn’t crater even more.
Some interesting things in there ($50 million for artificial photosynthesis, what do they know). Plus lots more than that for nuclear. Don't know how much, they purposely broke it into lots of smaller pieces in lots of different paragraphs, like they always do.
All in all a good bill, even though it's textbook business as usual.
Nanometer process indicators long ago became more marketing than reality. I was hoping (irrationally, it seems) that when we got down past integer nm readings we’d just let the terminology float, and go to version numbers or something.
Alas, the false geometries march on down to the next unit.
Let's be real: half of the drivers in computer tech involve numbers that have dubious isolated value: 144hz monitors; gigahertz processors; number of cores; and now process scales.
Once the first few cycles hook into a number it becomes a marketing gimic to sell next year's version.
119 comments
[ 3.0 ms ] story [ 189 ms ] threadIf those fabs become unavailable to the west do we know what that would do to the world economy?
China wants them because they have those sweet ASML EUV machines that they are sanctioned from buying.
The US needs those fabs to stay controlled by Taiwan because they don’t want the world’s most advanced contract fab to be overseen by a hostile country.
Taiwan needs those fabs because it makes them important for the US to defend. And because those fabs are so important to everyone to who wants high end chips, just having them is a good defense.
They might get six or twelve months of production out of TSMC's facility, and then things will start to break that they will not be able to fix easily. Invading Taiwan will cause economic damage measured in hundreds of billions of dollars at a bare minimum, and much of that damage will be suffered by China.
Bottom line, it would be insanely stupid of China to attack Taiwan. Of course, many people (including myself) said the same thing about Putin and Ukraine, so...
The US economy will be in ruins, and nothing can stop a large scale military conflict.
Signiciantly more will be suffered by US/west whose advanced industries extracts disproportionate patent fees and have extreme reliance on TW semi for short to medium term. We're talking trillions in knock on effects. Proximity of semi production / supply chain concentration in TW and broader east Asia is PRC's trump card in war. During peace US has trump card in sanctions / tech control, but during war PRC has leverage of simply setting the advanced economies back 30 years which will actually close relative gap. PRC might even pull ahead if they manage to mass rollout of 28nm indigenous fabs and complete indigenous semi supply chain before anyone else. That's without mentioning during war, PRC has incentive to attack SKR/JP fabs to draw US into first island chain where they're weakest to honor security commitments. If things ever get heated with US, it would be insanely smart for China to threaten TW/JP/SKR as well. And if PRC attacks TW, I surmise there's going to be three way side deal between PRC, US and TW leadership to maintain TSMC production continuity in exchange for evacuations for TW leadership to US who would otherwise get the noose for treason. Which itself makes the notion that TSMC is rigged to blow counterproductive - surival of TSMC is TW elites golden ticket off the island - US isn't going to look kindly to a TW admin that explodes US economy and tech lead.
People can live without smartphones, but can't live without food. In case of war, westerners will need to postpone buying some electronics or clothes, but the Chinese will resort to severe food rationing.
To make it easier to understand, total world grain production is about 2250 million tons for a population of 7.8 billion. At worst China would need to ration more expensive meals, but people wouldn't starve.
Another Q one should ask is how vunerable US is to comparable counter action. Because US agriculture is highly industrialized and automated whereas PRC ag is in the process of automation and still reliant on labour. Sufficient cyber attack on US ag infra during harvest season and US is forced to ration. What happens to automated John Deere tractors when they have no GPS to harvest. Same with familiar threats of embargoing PRC oil because US is self sufficient on the metric that there's enough fossil fuel resources in the ground. US is only as self sufficient as how much it's 130 refineries can produce. If US interdicts PRC energy, don't be surprised if PRC takes down comparable levels of production state side.
As for cyberattacks from China, those can only do so much, and in any case they can be mitigated by cutting the fiber. OTOH, shooting down GPS sats will lead to war. They know better than that. I hope.
This is in event of an war.
Legally, so is blockading PRC, including food and energy imports which is not an unfamiliar discussion in US policy circles. As for cyber, US DoD / Biden also explicitly communicated to both PRC and RU that cyberwarfare on critical infra = actions comparable to war. I don't know if cutting fibre is a joke, there's millions of PRC diasphora, unless US disconnects from internet globally, that's assuming PRC agents aren't state side already. So the real question is, does the US know better than to intervene when PRC attempts armed reunification with TW, which seems increasingly likely while bills are being drawn in congress to boost TW into major non-Nato ally? Because if US does, there's no telling how things will escalate, especially when modern connectivity and increasing PRC missile capabilities means CONUS is no longer safe. Incidentally, I think both sides knows this, on paper but frequently, better judgement gets overriden by domestic politics.
It's unlikely China or Russia, or anyone, has the capability to shoot down GPS sats. A number of countries, the US included, have demonstrated capability to shoot down satellites, low orbit satellites. GPS satellites are in geosynchronous orbit, which is a whole different species of animal. Shooting down an LEO satellite can be done with an SM-3 missile which has a mass of 1.5 tons. That's significantly lower than the mass of a rocket needed to put the satellite in orbit, and that's possible because SM-3 is not designed to reach orbital speeds, only orbital heights. Its primary mission is anti-ballistic missile defense and ballistic missiles can reach thousands of kilometers altitude, but not tens of thousands.
In theory one can design a missile that can reach GSO heights without reaching GSO speeds, but that requires a dedicated project, which would be extremely expensive, and very difficult to keep under wraps. All missile launches are pre-declared to minimize the risk of a nuclear war. You can find a list of all orbital and suborbital launches on wikipedia (e.g. [1]). If anyone tried to test a missile that can reach GSO heights, they couldn't keep it a secret.
One can then try to shoot down GSO satellites with full size rockets, like Falcon Heavy. But there are 31 GPS satellites out there, and 30 Galileo. I doubt either Russia or China (or even America) have 60 full size rockets available to do a blitzkrieg strike against the GPS satellites.
[1] https://en.wikipedia.org/wiki/List_of_spaceflight_launches_i...
Furthermore, it probably also means the disintegration of US global sanctioning power. Europe would rather trade with China than the USA, assuming it's mutually exclusive.
https://www.usnews.com/news/world/articles/2022-07-30/swiss-...
Right. Fab tools require constant maintenance. I'm sure they keep a stock of repair kits and spare parts, but probably no more than a few months supply. Their supply chain ops (procurement and engineering) sets min/max stock levels of these parts, but I'm willing to bet that there's no more than a few months supply on hand. And then equipment gets dirty, wafers get scrapped, yield plummets, etc.
The oligarchs in the west decided to outsource key industries to developing countries to increase their profit margins.
Now they're scrambling to fix those mistakes.
As globalism unwinds, there will be an open desire to roll back child labor and environmental laws as well as birth choice.
We'll get factories back in the USA - and wish we hadn't.
Or we'll see a massive expansion in prison labor, even worse than we have now.
Correlation not causation.
The simple explanation here is the correct one: they oppose abortion on social grounds.
There's no such thing as a free market in our world. It's just a myth. Everything is regulated or under the purview of treaties or laws.
Just earlier this year they were complaining that Chinese companies were ripping off their IP [0] and IIRC even within the last 2-3 years they were themselves hesitant to send bleeding edge tech over for similar reasons. I wonder what has caused their tune to change.
That said, if I'm reading the tea leaves properly...
I think the last couple years were a wake-up call to the US that not only are they overly dependent on other countries for critical components, but that China is far closer to becoming 'self sufficient' than us (1).
Ukraine was a bit of a wake-up call to us, that we are too afraid of nuclear escalation (2) but also to China that embargoes can happen more quickly than you'd expect and hurt almost as fast.
What this means however, is we are now both in a race for self sufficiency on this front.
... Really random question, I wonder why Mexico doesn't have more fabs?
----
1 - I think the US has a weird definition of 'self sufficiency' mind you. For instance in some cases it seems the attitude is 'Oh we did that again and can ramp up.' As a general comparison, I'm betting the possibility of getting a '100% China content' vehicle is far higher than getting a '100% USA content' vehicle.
2 - Without getting into it, there was a short window early on in the Ukraine conflict where I would have doubted Putin's willingness to go nuclear. IMO It's one of those cases where the would police card may have been worth playing, were it not for our missteps in using it over the last 20 years.
[0] - https://www.theregister.com/2022/02/11/asml_chinese_rival_ip...
Wouldn't it still take quite a long time to get a new fab to the point of being productive? These machines aren't exactly sitting in storage by the hundreds, and I imagine you'd also need to dial in all the processes pretty well to get good yields? (though I have no idea what that entails...)
Possibly years.
I honestly don't know if you can 're-clean' a clean room if it's flooded with the right things. And the right things might just be seawater.
(IDK. Not a fab engineer by any stretch.)
Not straightly true. The worlds "leading edge" Semiconductor Fabs is located and dominated by TSMC. Yes. But Semiconductor as an industry includes many things not leading edge like small chips used in appliances as well big volume commodities like NAND and DRAM. Those are mostly in South Korea, Japan and then US. ( Along with Singapore and Germany in case anyone wants to nitpick. )
If TSMC get destroyed, it will likely set back the whole Semiconductor industry by 10 years. Simply rebuilding the Fabs capacity, leading edge or not would take 5 - 10 years with perfect project planning and execution. Not to mention no player in the industry has shown they are capable to providing similar quality, yield, services to TSMC with similar prices at an acceptable margin.
If that is true, why doesn't TSMC crank up its prices?
Or why don't they build a consumer brand? They could fade out Apple if they start building their own consumer products.
Regardless of how true quoted sentence is, it's also obviously true that if they changed prices to $1billon for every single chip sold then they would sell close to zero chips (ignoring that such a crazy move would probably see an attempt to replace leadership, or a takeover attempt, or whatever). And equally, if they lowered their prices to give each sale only 1% of their current profit margins, they would sell more chips.
Every time they set a price, they'll be thinking about where the line is to get that balance right. And they already know they're the industry leader, and of course they know it in a lot more detail than that simple statement, so they're taking that into account (along with things like demand, how close the competitor products are, etc.)
Of course that's not to say that they've necessarily made the ideal choice every time, and maybe if they raised prices they would make more money overall. But how does the fact that them literally disappearing would set back the industry make this more or less likely to be the case?
TSMC has no desire to become Apple because they already capture so much of Apple's revenue. I guarantee you they have put enormous thought into the exact optimal price point, as building fabs is extremely capital intensive.
They can make more by selling higher volumes at a lower margin.
> Or why don't they build a consumer brand? They could fade out Apple if they start building their own consumer products.
There is a lot more going to Apple's products than TSMC's fabs? To say nothing of their business with MediaTek, AMD, Qualcomm, Broadcomm, Nvidia, Sony and others [1].
This is the same failure Intel fell into, compounded tenfold. (If we build best, we should eat the designers' cake too.)
[1] https://www.gizmochina.com/2021/12/15/tsmc-top-10-customers-...
Higher margins are made closer to the consumer.
> (If we build best, we should eat the designers' cake too.)
Apple is eating software developer's cake ... (well, 30% of it).
The greatest businesses on the planet all rely on brand recognition. TSMC has none and relies only on its technological expertise which at some point might be surpassed by a competitor. There is a reason why Apple wants to own the entire supply chain, and they are smarter about it because they start at the other end (close to the consumer) where margins are larger.
Which is consistent with and irrelevant to making “more by selling higher volumes at a lower margin.” Higher margins are made by Michelin-starred restaurants than TSMC. That doesn’t mean they should expand into fine dining.
Chasing margins at the expense of profits is bad business. Cannibalising Apple would mean losing volume, from Apple and others, which in turn undermines TSCM’s scale-based advantage.
If tsmc were run by Apple's board, we would all be paying 30% of our income for the privilege of using modern computers. A much better strategy, just look at Apple's balance (and Apple products aren't even as exclusive as tsmc's).
This was IBM. For fundamental, structural reasons, they were dethroned. TSMC doesn’t have the option of eating Apple’s cake without giving up the scale that makes it competitive. Again, it’s similar fresh-MBA thinking that tanked Intel. (TSMC wouldn’t exist hadn’t Intel given it the neutral market space.)
This was IBM. For fundamental, structural reasons, they were dethroned."
I think they did it to themselves, first by eschewing virtual memory in the System/360 and first models of the System/370, thus losing the higher education market and what people eventually moved to like UNIX, thus their big machines become a specialized niche of the whole computer market. Then in moving from a preference of leasing to sales which gave them a one time goosing of revenue which they didn't prepare to lose when the cycle of that ended.
They did do well in PCs for a while, but the unit that did that in Florida eventually got blue washed (there's lots more to the story of course).
"Again, it’s similar fresh-MBA thinking that tanked Intel."
Also very unsure of this, I don't see that resulting in the 10 nm now Intel 7 debacle. More like their decades long high level very bad engineering management eventually caught up with their crown jewel of fabrication. The not so fresh MBA I assume CFO who was made an interim CEO actually had a plan which might have succeeded.
I don't think the current CEO's plan has much if any chance at success, especially since I don't see him making the brutal moves necessary to change the company culture to work as a foundry. In particular, I note while he worked for Intel for a long time, he's a design instead of fabrication guy, the latter is what they needed and I wonder if they couldn't hire anyone good from that field.
Yup. They went from neutral generalists to niche specialists.
TSMC is, today, a neutral generalist with the scale of Apple, AMD, Qualcomm, Samsung and the others combined. You’re (EDIT: Nope!) advocating trading that for Apple’s profits. Short-term versus long-term greedy.
> McDonald's versus farmer joe who makes the best potatoes
Replying to a parent comment of yours (EDIT: Also wrong!). McDonald’s has scale. TSMC has scale. Farmer Joe competing with McD’s makes sense. Tyson competing with McD’s does not; it will lose, in the long run, McD’s as well as other restaurant chains as customers with that play, undermining its home-turf edge while it competes where it has none.
Look at who you're replying to, I'm not advocating that at all!
The Semiwiki guy credibly claims TSMC has reached this worldwide pinnacle of chipmaking in part by being trustworthy and creating an ecosystem based on trust. That of course includes being fair with your customers, which has allowed the latter to fund the creation of new fab lines and nodes knowing TSMC won't screw them over once that money is spent.
Also not the guy who made potato analogies.
In the long run Apple's strategy may work much better because if Intel and others catch up, TSMC will be a replaceable commodity.
They were dethroned because they opened up the market to clones, and allowed Microsoft to gain control over the software side.
See, this is about control. TSMC has less control over the market than their clients do. And if they don't act on it, it may become their demise, just like with IBM.
That’s surely the reason for their political exercises in the region these days.
Yes they were military exercises, but quite blatantly it wasn't China actually needing to do any testing or training of their military, except using it to send a political message (both domestically and internationally).
(Which isn't my giving an opinion on the likelihood or not of China ever using military force against Taiwan, just a response to your ridiculing the idea that recent actions weren't political).
TSMC has a big volume and an edge in node process. However, the vast majority of chips in use are in embedded usecases. Most non-general-computing hardware containing chips doesn't need 4nm. It would chug just fine with two decade old chips. Even military hardware uses very outdated chips.
Catching up to a modern semi fab would cost time and money, but making more outdated chips isn't that much of a problem. There's a certification issue, but in such a scenario surely certification would be relaxed.
As for general purpose computing, servers always have the option of scaling horizontally. The biggest issue would be with phones and tablets. Unfortunate, but tiny compared to the other costs of war.
All this difficulty with high end smartphones and gaming consoles is completely insignificant compared to the 'normal' economic damage a war would cause.
I missed some announcement though that Qualcomm is moving more towards using Samsung.
> All this difficulty with high end smartphones and gaming consoles is completely insignificant compared to the 'normal' economic damage a war would cause.
This 'difficulty' would set Moores Law back more than half decade outside of Intel and Samsung smartphone chips. The next smallest large scale node being 12nm at Global Foundry and keep it frozen there for at least 2-3 years in total leaving as decade behind Moore's Law. You underestimate how much damage that causes. Even if China completely wipes out Taiwans 1% share of GDP that impact is miniscule compared falling 10 years behing on Moore's Law.
I could not tell if it was a tongue in cheek statement, but if not, can you explain why falling 10 years behind on Moore's law would be such a big deal.
Embedded doesn't care all that much. Servers can scale horizontally. What's left is various consumer products, and even here you can easily make a smartphone on 12nm, it would just be slower. Hmm.. Maybe VR and locally-run ML must have a tiny node size. Hardly big markets. For all else, there'd be a temporary efficiency loss, but we'll make do and catch up within a decade. You could argue delaying the future would have a running effect, but it seems to me mostly leisure activities would be affected and investment/production would be fine?
The Moore's law effect is insignificant compared to the damage of a conflict between the world's #1 manufacturer and #1 consumers.
And all of this happening across almost all industries simultaneously? That's not a recession event - that's a depression event.
why would (non-autonomous) car need state of art chips?
I'm guessing they have some less modern stuff. Maybe.
But. BUT.
UMC is kinda an unsung player here to some extent. They are another Taiwanese Fab, and do a lot of the older process stuff.
I'm guessing that whatever plans Taiwan's government has for TSMC Facilities in a Kobayashi Maru scenario with China, UMC probably gets some version of it as well.
Lets also assume that if things get to that state with Taiwan/China, at minimal we'd probably have a larger embargo against China (another source of such chips) and there's a possibility of trade disruption with other Asian players like South Korea.
I find the focus on chips bewildering, since there are so many more critical issues like 'how do we prevent a stock market run and the crash of every company which outsourced manufacturing to China?' or (if you live in China) 'where do I get a food ration card?', if this gets real bad. Semis are merely an annoyance.
Do we have the technological capability to recycle old semis to work with current systems to keep things running? Do enough people have this skillset? I'm sure we figureit out, but I doubt it there are enough people to really get it moving. With a severe lack of chips we'll see systems falter or all out crash as components wear out and can't be easily replaced. I think you're underestimating how much runs on what Taiwan produces and I don't think "economics" can bring more production online in a meaningful time frame.
It's understandable HN community would focus on semis, but compared to 'financial crash worse than 2008' and 'Would all people be able to get food next month?' semis aren't that important. Semi production would be a slow long-term crisis we would be able to adjust to if we could handle effectively the initial very big short-term crises resulting from a massive war.
As Keynes put it: "In the long run we're all dead". If we couldn't handle the initial short-term economic crises we'll have bigger problems than long-term semi production adjustment. If we can, we could handle adjusting semi production and use. It's not that it isn't an issue, it's just way less important and less difficult than the other issues.
I could be wrong but I thought that some of the ML stuff for military is causing more demand on the high end?
If it is successful Taiwan is left with a hole to fill as its economy is completely dominated by its semiconductor industry.
EDIT: okay so if Intel get 18A done in 2024 that is equivalent to N2 which has recently slipped to 2025/2026.
I say equivalent but as we all know the real feature density is what matters and I’d love if someone had some expected comparisons on this.
Intel's roadmap is Intel 4 (TSMC N5P) --> 20A (TSMC N3) --> 18A (TSMC N3E)
Intel, who hasn’t shipped anything using EUVL, plans to be two years ahead of TSMC, within two years.
How is that realistic?
We don't know how good Intel 18A is yet. It could be equivalent or only slightly better than TSMC's N3E or further enhancements of TSMC 3N. Keep in mind that the base TSMC N3 node is scheduled for volume production in Q4 2022. So the TSMC is still likely quite ahead.
This is all assuming that everything goes well for Intel. If 18A is simply better than TSMC N3E, then TSMC won't have anything to compete until 2025 - 2026 when they release N2.
For TSMC, they have real paying customers that need to plan years ahead so all their date estimates are very conservative. For Intel, they're desperately trying to catch up so they're much more aggressive in their estimates.
According to Anandtech, TSMC's N2 could come in 2025 if things go well, which is roughly the same as 18A. So basically, if things go PERFECTLY for Intel, they'd be in the lead for a few quarters - not 2 years like you said.
Even TSMC expects Intel to get back on track in a few years because, again, they started investing again. Please note Intel is about to get huge subsidies from the US (~50b) and EU (~20b) real soon. It's free money for their precious fabs and R&D. I am very much against the EU giving them anything but a kick in the nuts but oh well, gotta keep the US and Israel happy
I'm 100% against tax payer funded chip fabs. It's just a handout to huge corporations.
Already happening: https://semianalysis.substack.com/p/intel-cuts-fab-buildout-...
Also, look for huge executive bonuses in the near future.
The gov't is specifically paying for new fabs that are already being built.
Not to mention this is a national security issue given the possible Chinese invasion of Taiwan.
Intel CEO Pat most certainly finessed the government into giving Intel a huge payout by exaggerating this threat.
If we don’t gain that capability now, it will take us too long to recover when push actually comes to shove.
Regardless, I’d go further and argue that we should be building up the capability to manufacture our own EUV lithography machines.
As far as I am aware, it’s the exact opposite: over the years, ASML licensed some of their patents to US companies.
[1] https://asiatimes.com/2022/07/how-china-could-leapfrog-us-ch...
If you think about it, ASML wouldn’t be this critical to semiconductor fabrication if it didn’t control the patents for its core tech.
I'm not sure that's entirely true. If we define "most advanced nodes" as 5nm-class fabrication, certainly Intel is doing that well now, just not in volume production. Intel 4 based Meteor Lake powered on and runs Windows as of earlier this year.
No capability to build and not ready for volume manufacturing aren't really the same class of problem.
Anyways, I think the CHIPS Act was the right move at the right time. SMIC is quickly catching up thanks to near unlimited funding from the Chinese government. We need to play the same game if we ever want to catch up to TSMC.
Not as bad as if Taiwan's exports to the USA are stopped.
We can build motherboards / pick-and-place machines in the USA. We literally cannot make the chips that are coming out of Taiwan right now. Samsung is the #2 IIRC, so maybe South Korea can pick up the slack in a hypothetical Taiwan-conflict.
> Intel CEO Pat most certainly finessed the government into giving Intel a huge payout by exaggerating this threat.
US Government is also got TSMC (Taiwan's semiconductor fabs) to build an Arizona fab-lab. Look, even Taiwan is scared of possible conflict and is hedging on the physical safety of the USA.
Hopefully conflict doesn't happen, but the powers that be are making smart moves here. Hopefully it will be enough.
I don't think that's the case, as I understand it this comes from the US wanting to onshore some of TSMC's capabilities and Taiwan screwing up electricity production on the island.
It does Taiwan the government and ruling class no good whatsoever to place assets beyond the PRC's military reach, rather the incentives are entirely opposite of that. The more valuable TSMC's fab lines there are, the more incentive the US has to help with the ROC's defense, or war if it comes to that. Which in the case of torpedoes from attack subs can even be done with some plausible deniability.
The TSMC factory in Arizona is full speed ahead. Its literally in Arizona right now, undergoing construction / hiring / spinning up.
5nm Fab is being built right now in Arizona though by TSMC. They're beginning to offshore fabs.
Look up any information on "TSMC Fab21".
So they re-balanced for a volume play.
If some (read: national security volume) fabs in EU / US are enough to assuage governments, but most (read: global economically important volume) are still in Taiwan, then that's the next best thing.
Where life will get interesting is in South Korea, at the Samsung fab. They will have the only 3-nm process in the world if TSMC goes away.
TSMC isn't giving up their cutting edge technology, just what will be considered an old node size by 2024.
You can offset internal Rd investment with gov money and use that elsewhere, sure, but when the magnitudes are large and immediate, that doesn't play. You get one check, and it better be spent on the intended cause in X fiscal years or you pay it back.
The tech community has suffered under the Cost Plus and other workaround bullshit for long enough that the cynicism is inbred. But there's plenty of gov funded RD money that does precisely what it was invested to do.
To put it another way, it’s similar to the futility of those well-meaning people who give some homeless person $5 on the express condition that they spend those $5 on food, not on alcohol or drugs. Which said homeless person is all too happy to do, he/she will gladly spend those gifted $5 on food, which leaves the other hipotethical $5 which he/she had in their pocket all available for buying drugs or alcohol, money which otherwise would have been spent probably half and half (2.50 on food, 2.50 on drugs).
At least in the counter-example I gave the result of largesse is a homeless person better fed, which is good (and also more intoxicated, which is not so good), but in this example with Intel you have taxpayers filling the pockets of one of the biggest tech companies in the world.
I cannot claim authority over all instances of gov funding, but I have managed funds from the government for Rd use and they absolutely do audit, and they absolutely do require an accounting down to the last penny, and they absolutely do review plans vs actions. And they absolutely do get refunds for money not spent. And, critically, they absolutely do shutdown follow on funding if they do not like what they see.
why? isn't it in the interest of the EU to not be dependent on China for chips, if they take over Taiwan to corner that market?
Intel will be the first customer of ASML's post/next-gen EUVL systems:
* https://www.asml.com/en/news/press-releases/2022/intel-and-a...
* https://seekingalpha.com/article/4496925-asml-fuelling-intel...
TSMC also has extensive experience with these machines compared to Intel.
Pat is just making shit up and hoping their stock price doesn’t crater even more.
I would be curious to read the article again in 2025.
https://www.commerce.senate.gov/services/files/1201E1CA-73CB...
Some interesting things in there ($50 million for artificial photosynthesis, what do they know). Plus lots more than that for nuclear. Don't know how much, they purposely broke it into lots of smaller pieces in lots of different paragraphs, like they always do.
All in all a good bill, even though it's textbook business as usual.
Alas, the false geometries march on down to the next unit.
Once the first few cycles hook into a number it becomes a marketing gimic to sell next year's version.