They are currently basically one of at most a handful of people who can do that. They need to face the music and get their elbows out; now is a good time.
> It took humanity thousands of years to come up with calculus.
Depends on how you see it. During most of humanity's existence, we were hunters and gatherers, with simple and small societies - there wasn't much need for advanced math and there was not really a possibility for a society to feed a thinker. Only when complex societies like ancient Greece and Rome were possible thanks to early industrialized farming, there was a possibility for people to do nothing "productive" towards the society itself surviving but instead invest resources into advancing society.
Ancient Greece was thousands of years ago. Some complex societies are even a bit older. For example the Assyrian empire started around 2600BCE, and I guess there are older ones still.
And thanks to all that effort the Chinese now likely have all the blueprints. The only reason why they still don't have EUV is because this stuff is difficult to bring up even if you have all the documentation - screw something up even a little, and it's not going to work. But it certainly isn't going to take them 20 years to replicate it.
The nature of patents is that it is mostly public information anyway and free to use after the patents expire. That works both ways of course, the Chinese have some pretty fancy technology in other fields that manufacturers in the EU or US might benefit from.
Of course with chips, patents and know how that e.g. Intel or ASML have are two things.
Ultimately this is good as it drives companies to continue innovating. Intel's current problems are related to them forgetting to do that.
Precisely. A billion ways to get things wrong, and if you get it wrong it doesn't work. And the info to get it to work is unlikely to be in the patents. You have to have folks who worked with this stuff before, which is difficult, but doable.
And as far as "innovating" in this field, we're almost at the end of the run - there are, as far as I'm aware, no processes planned beyond 3nm, and even 3nm is astonomically complex and expensive.
>> patents and know how that e.g. Intel or ASML have are two things.
Shouldn't that make the patent invalid? I thought the requirement is that the patent should give enough info so that "person skilled in the art" can reproduce the invention.
At the time I worked for an ASML contractor, the joke was that the chinese already have all the ASML code, just they couldn't compile it. This sounded so true, because the ASML code is the worst pile of c built over almost 40 years that the world has ever seen and because the chinese were rather blatant in their attempts to steal whatever they can from ASML.
Not personal, but chinese agents would wait in front of the ASML buildings and target ethnic chinese workers offering them money and honor if they come back to the motherland and work there. Stories about stolen laptops and so on.
This reminds me. I have an anecdote from my 90-something year old uncle who coordinated and managed construction efforts for the power grid in a post-WWII socialist-bloc country between 1940-1990. At first they relied on the blueprints left behind by Germany, and then they sourced their power generators and components from the USSR. But within several decades, it became clear that the USSR stuff is just bad. Like straight up unreliable and faulty bad. So the question became about whether to build their own, which the engineers were very confident they could do. Although it seemed feasible in principle, after many years of failures it just turned out that manufacturing these things was way too difficult to do reliably domestically. They had underestimated the entire intellectual investment necessary. Buying from the West, from manufacturers with decades of experience became the only option. And this was a long time before the iron curtain even started to collapse. And this was a process managed by socialist/communist loyalists. They were buying tech from the West while still projecting an image of strength and control.
Regimes trying to copy something strongly reminds me of that anecdote. They usually manage to waste a lot of resources only just barely catching up, copying, and then they still end up buying the stuff that works from the companies that have the decades and the experience doing it.
Making this chip tech sounds orders of magnitude more complex.
Reading the article I only think about what a major fumble it was geo-strategically that the people with good ideas went to Taiwan.
China is in a _much_ different place than the USSR was: for one thing hundreds of thousands of Chinese have advanced degrees from Western universities. A lot of them have industrial experience as well (as well as probably copies of trade secrets they could divulge if the CCP really insists - we even catch the dumbest of them in the process of doing so from time to time). As if this wasn't enough, the West has outsourced darn near everything manufacturing related to China. If they want 5nm, they'll have 5nm, within 5 years or less. And if they can't have it, they'll just take Taiwan.
"The U.S. Department of Energy (DOE) has announced an agreement with ASM Lithography (Einhoven, Netherlands) that would allow the company to participate in the extreme ultraviolet (EUV) lithography development effort now underway in Livermore, Calif."
EUV is a joint research by US Dept of Energy and several US companies. ASML got invited to join the research based on conditions lay out by US Dept of Energy.
Samsung's big cash cow is it's memory business, it's responsible for only 20% of revenue but 50% of profits. They're doubling down on the logic chip business with massive investment in their 3nm gate-all-around architecture.
One issue is there's a significant conflict of interest between their device and CPU businesses. Apple left them precisely because Samsung is a competitor in handsets, and other handset manufacturers are also wary of buying from them for the same reason.
As a side note it is true TSMC deserves a lot of credit for the advantages of the Apple M-series CPUs. Without the 5nm TSMC process they wouldn't be anywhere near as impressive. However Apple has been working very closely with TSMC on it, and their long term financial commitments been a big factor in TSMC being able to move forward so decisively on developing the process and investing in future nodes and capacity. Apple ditching Samsung as a supplier has had a big influence on the shape of the CPU industry. The (currently) top comment by WithinReason highlights part of this dynamic.
> One issue is there's a significant conflict of interest between their device and CPU businesses. Apple left them precisely because Samsung is a competitor in handsets, and other handset manufacturers are also wary of buying from them for the same reason.
I don't think what bothered Apple is relevant to other phone makers. Case in point, Qualcomm doesn't mind having Samsung fab their SoCs, and even their modems, a domain where Samsung has repeatedly tried to catch up!
Anyway in the current Android smartphone landscape, everyone is using ARM reference CPU designs.
It's depressing, it seems like the high end Android handset market has disappeared. What we have now are midrange and low end, and what passes for high end are really mid range phones dressed up with a few premium features or gimmicks.
> What we have now are midrange and low end, and what passes for high end are really mid range phones dressed up with a few premium features or gimmicks.
Ultimately no one can shake their culture, it follows you like your shadow.
Google's is cheap and low-end. They have become the Walmart of tech companies.
There was also a point in Samsung’s past when they had been selling Nand chips to the market, then decided to hold all the capacity for themselves (or maybe Apple and themselves), then somewhat recently opened it back up again - but nand (and CPU) chips aren’t easily interchangeable - so a reliable partner is more important. Dram chips on the other hand are much more interchangeable.
A lot of what I read about TSMC's current lead is that it's based on trust, including creating a complete ecosystem based on that. As you note Samsung is not so trustworthy a company for outside competitors, and I'd add reports it's not always honest about what it's accomplishing for its foundry customers, or would be ones.
> As a side note it is true TSMC deserves a lot of credit for the advantages of the Apple M-series CPUs. Without the 5nm TSMC process they wouldn't be anywhere near as impressive.
We're going to look back on this story as one of the craziest product launches of all time. I genuinely have no idea how Apple secured 5nm so early (and blocked other customers from getting a piece until 2022), I have to imagine that it was a last-minute decision they pushed out to TSCM, judging from their response. So many different parts of the M1 are pushing boundaries we've never hit before (for better and worse), so it will be deeply interesting to see how Apple scales a CPU that barely exists in it's current form. Some of my friends in the manufacturing biz were absolutely dumbfounded by the fact that Apple fit 64 gigs of memory on the M1 Max (the thing is surrounded by memory chips), which makes me extremely curious about their next pro hardware.
Fun thought experiment, too: Would the M1 have still been such a knockout hit on TSCM 7nm? It certainly wouldn't be as good as it's current incarnation, but I'd bet dollars-to-donuts that a big.LITTLE ARM chip would have worked pretty well on last-gen silicon.
>I genuinely have no idea how Apple secured 5nm so early (and blocked other customers from getting a piece until 2022), I have to imagine that it was a last-minute decision they pushed out to TSCM, judging from their response.
I would not be at all surprised if they laid the groundwork for this partnership back in 2015 or even earlier. The only reason Apple dual-sourced the A9 was because TSMC didn't have the capacity to deliver the full requirement, but Apple had already been trying to get off Samsung for years.
Apple has a long track record of early, heavy capital investment in core technology manufacturing capacity and long term contracts with suppliers. This serves the dual strategic purposes of securing exclusive access to new tech, and driving down the long term costs of that tech. They did this with retina displays, in that case Apple literally owned the manufacturing equipment operated by Sharp.
In order to build out fab capacity TSMC needed long term high volume guaranteed orders. Apple delivered that. There was nothing last minute about any of this, I think it's always been the plan.
Ruthlessness. Apple did the same thing when the iPhone first came out with the higher end capacitive touch screens. When it was released, all the competitors raced to catch up but found that the supply was gobbled up by apple for years. If you compared any android touchscreens with apple ones for the first 3ish years the apple ones were night and day better.
Apple is also willing to invest in technology in return for exclusivity for awhile.
>One issue is there's a significant conflict of interest between their device and CPU businesses. Apple left them precisely because Samsung is a competitor in handsets, and other handset manufacturers are also wary of buying from them for the same reason.
Commodore manufactured the 6502 with its MOS subsidiary, but I don't believe that Apple or Atari hesitated from using the 6502 in their computers.
It seems like the industry, players and technology involved have evolved and matured significantly since 6502 CPUs were relevant. Is Commodore or Atari even still around?
If Samsung came out with a contemporary CPU with 2x the compute per watt of the M1, Apple would probably be interested in it.
Notably, it seems like Intel had the capability and head start to have been able to develop such a CPU by now, but instead they held back progress to maximize their profits. It was apparently a bad long term strategy, as they've now lost their lead. The market moved on around them.
But other companies, like Rockwell Semiconductor, manufactured perfectly good 6502 clones. If MOS tightened the screws, Atari or Apple could've simply headed to any other number of suppliers. The same can't necessarily be said for a competitive processor today.
I wasn't aware of the connection between volume and technological progress in semiconductor manufacturing:
“When you have volume orders, then you can do yield experiments, you can improve your yield, and yield is everything because that’s how you cover your costs,”
"Foundries that can juggle multiple clients and technologies can swiftly advance on their competitors. For example, TSMC’s scale allowed it to master extreme ultraviolet lithography faster than anyone else, which reduced the number of steps it took to make advanced chips and boosted the throughput of its fabs."
The operators that calibrate them and pick parameters, do. And some machines have internal long-term calibration and feedback (no need to call it Machine Learning), so in a way they learn on their own too.
Obviously you're correct about that, but those productivity gains from those sources are unlikely to be accurately predicted by Wright's Law, which is what this thread is about.
If you scale up production by just building more of the same machines that's right, but over time with new process nodes and increasing fab sizes there are plenty of opportunities to apply new skills to improve automation and scale effects.
I thought it was a more general principle and applies almost any task. For example, the software written today is done at a larger volume and more cheaply than at any other point in history and can do more things than ever before (heck printing “hello world” or adding 2 numbers was a feat on the original machines).
My understanding is that it’s a general phenomenon. The more we do something, the faster we get at it (either as a species or individuals), the more tools we accumulate to help us smoothie out any bumps, tackle larger projects more efficiently, etc.
Semiconductor manufacturing R&D and pathfining includes many expensive dead ends. Massive volume is required to justify the gigantic upfront cost to develop a new process every 2-4 years (overlapping with previous and next process node).
Economist would say that volume gives you:
1) Increasing marginal returns after massive fixed capital investments.
2) increasing returns to scale from variable investments when volumes increase.
Every CPU project I worked on at Intel from 1997-2010 was cancelled. That's 13 years of one senior engineering salary that never turned a profit, let alone the thousand other people on the related teams. One could argue the failed learnings helped guardrail successful products, but my internal perspective says, Nah. Intel throws/threw a staggering amount of money at failed projects. Wondering if Gelsinger will shake out the numerous boondoggles that litter that company.
> Itanium shipped, and I have a soft spot in my heart for it, VLIW was brilliant. The hate is administered by knownothings, it was a CPU for engineers.
Around the 90s, the big OOOE scare and push to move away from it seemed to be based on projections showing wire delay and super-linear scaling structures like schedulers would bite OOOE very hard and make it nonviable.
This might have been technically true -- Pentium Pro reorder buffer was 20 entries while new cores might have only 10 times that, about an order of magnitude. Processor widths only by about 2x. IPC has increased about an order of magnitude as well (ignoring SIMD). While transistor density increased 3 or 4 in that time.
So I don't see a huge problem with that general prediction. What I wonder about is why on earth they thought in-order VLIW could solve this in a better way. Itanium scaling ran out of steam well before the out of order designs. And huge widths never happened or probably didn't help much because they're always just going to get stuck on taken branches, data dependencies, etc.
I liked Itanium when it first came out, it was a floating point monster and even on integer it shaped up pretty well against the likes of POWER4 in the TPC-C wars, exchanging the lead for several generations. But it became apparent that it depended on heroically fast (and therefore sadly tiny) L1 cache which halted its frequency scaling. It depended on a heroically large L3 cache which was became a problem in the multi core era. All because it dealt badly with dynamic hazards like cache misses, and executing beyond them to generate memory level parallelism.
Sure it had advanced loads which were very cool on paper, but never really were such a capable substitute.
I don't hate EPIC/VLIW. Some people hate Itanium because they claim it killed of a lot of diversity in the CPU market. I don't think that it did, and wouldn't hate it even if I thought it did. I can see why it failed though, and that's because it was technically inferior to out of order designs for most of its target markets.
It would be fascinating to teach a 3xx level course on Itanium aspirations, design, and failures, because I think they literally speak to everything that was happening in microarch (and much of what has happened since) from 1990 - 2020.
Do cancelled projects at that level of expertise also contribute to institutional knowledge? I presume you generate information from near misses, and the resulting intuitions get applied and shared when you work on the next project with the next team.
Even if every Intel CPU project was success, it would not matter when Intel fabs are not competitive.
Intel decided not to do EUV when others did and squeeze yet another node from older technology and it backfired phenomenally.
Even when everything is done right, the risks are high. The competitive high-end node business went from 4 to 2 in one generation. From Intel, GF, TSMC, Samsung to TSMC, Samsung. Intel is trying to get back and they have to burn billions.
Guess Intel is moving with the times. Won’t be the first time they overhauled their business, I heard they used to be DRAM makers before jumping into CPUs.
The latest roadmap from TSMC now points to 2023 for 3nm. This actually put Intel in the same time frame as TSMC. So incidentally both Intel and TSMC will be on 4nm in 2022, and 3nm in 2023.
Note: Intel 4nm is a Full, new Node while TSMC 4nm is an optimisation of their 5nm.
Of course the roadmap and time table from Intel is often not a reliable indicator and also doesn't tell you whether it is product shipment or if they are only ramping. My guess is that Intel is running as fast as they can to reach those target. That is also not including capacity difference with node from both companies.
I think Intel under Pat Gelsinger is making the same transition as Andy Grove did to during his era.
They changed their branding. Their 10nm node was renamed to Intel 7, their 7nm is now Intel 4. Since amount of nanometers doesn't really mean anything anymore that's good PR strategy that nicely represents state of their nodes because density on Intel 7 and TSMC 7nm is pretty similar anyway.
I believe Intel didn't strictly rename Intel 10nm to Intel 7, it's a refinement of Intel 10nm that is Intel 7 (they were previously calling it 10nm+ if I recall correctly).
Intel rebranded their x nm to be TSMC's y nm. This may seem a bit subversive but it's not ridiculously unreasonable as the nanometre figures are complete bullshit, and the density measures are also sufficiently synthetic as to be mostly useless when comparing a similar node anyway (i.e. there's more to it than just shoving transistors on a chip)
There are very interesting ways that we could compare architectures, but it would never get past the marketing department and it might reveal capabilities that they would rather have a secret, so we are stuck with a single meaningless number.
At the top right of the graph you can see that it's currently around 9% in the minus in the pre-market.
A lot of sites don't update the graph for pre-market movements and only show price movements from when the market is open. On yahoo finance you can see pre-market movements if you open the full-screen graph (simply click the graph or click "full-screen" on mobile). https://finance.yahoo.com/quote/INTC?p=INTC&.tsrc=fin-srch
I'd bet PC sales are down because of the lack of GPU's. Mining hurts the rest of the industry because they aren't selling as many high end MOBO's, CPU's, cases, etc etc. Sounds like they need to twist arms to preference PC's over mining.
I know the focus of the article is Intel and TSMC but I kept wondering about Samsung. If it's following the same foundry model that TSMC is, why does it only have a small fraction of the market (compared to TSMC)? Shouldn't it be doing as well as TSMC?
Edit: it is maybe a little more complex than that, because much of that is memory, and memory and logic processes may not be able to share the same learnings and developments.
"... That represented 15.0% of the world’s total capacity and about two-thirds of it was used for the fabrication of DRAM and NAND flash memory devices. ..."
As others pointed out, they both do a lot of volume. But TSMC are ahead at the leading edge.
Complete speculation, but I would guess this is mostly due to Apple. Apple wants the best process and they’re able to pay for it. Further, Apple is mistrustful of Samsung so they were happy to drop them and switch to TSMC.
I think this highlights the difficulty Intel will face. Samsung’s foundry is independent from Samsung’s phone business, and I don’t think anyone has alleged otherwise. But even if Intel does manage to keep its two sides at arms length, will that be enough to attract business from competitors like Apple and AMD? I’d guess no.
- Do other firms trust Intel with their IP? Apple has a new feature for their next generation - Intel quickly becomes aware of this and this information leaks into Intel's product plans.
- Is Intel a reliable foundry partner? TSMC have a 30+ year track record, Intel have none as a third party foundry. Will they change their strategy again?
- Will other firms get a fair prioritisation of resources in the fabs? Or will Intel's own products be favoured?
I really really hope not. Considering what Intel is willing to do in software to sabotage AMD performance, [0][1] I shudder to think what they'll do in hardware.
Regulation of anti-competitive behavior seems to have been extremely relaxed lately. Amazon's marketplace and Amazon Retail seem at odds with each other, yet they're allowed to coexest, and everyone is fully aware of the anti-competitive behavior. Intel's chip business needs to be seperated from their foundry business, and the 2 can't be allowed to comingle with each other.
The incentive should be that a competitive free market is better for prosperity in the long term. If anti-competitive practices are allowed, the losers are the consumers, whose interests the government are supposed to represent.
From another point of view, looking at the government as an institution rather than through its role as an advocate of the people, you could also make the case that the government should oppose any massive consolidation of wealth/power as a threat to its own power.
FAANG companies each individually have higher annual revenue than most countries' GDPs. In the absence of government intervention, there wouldn't be much to stop them from turning into British East India Companies and establishing banana republics, raising private armies, etc.
In an extreme scenario, some kind of corporate Genghis Khan could hypothetically consolidate the S&P 500 into a single megacorp if the DOJ allowed it to happen, and that company's revenue would be about half the US GDP (making it the third largest "economy" in the world). It would effectively be a hostile state right in our own backyard. Would we even have the ability to enforce taxation if they said no? What would stop them from wielding that incredible wealth to effectively buy a majority stake in government?
Although GP was joking, if AMD was a customer of Intel there's no way Intel would sabotage that relationship and make other customers lose trust in them.
Sure, it can happen, and something like it probably has.
The point is that doing it and being caught out would be something approaching corporate suicide, and sane leadership would not risk it.
This places it in the bucket of 'unlikely, high impact' risks for potential clients. They'd want to design process and contractual language to try to make such perfidy apparent and not bet the farm, but I doubt it would be a deal breaker.
I would think IP concerns would dominate in such an arrangement.
> would be something approaching corporate suicide
No it would not. Just like in any other case where a corporation had to pay a huge fine - they do it, or it even gets lowered significantly, and then move on. The executives who were responsible retired long ago and the motto of shareholders and management always is "look to the future and don't bother me with old problems".
You make quite an assertion based on what - the opposite has already happened many times. Risk that has to be paid for much later often is worth it, especially when it's going to happen to a new team. Not that the risk for a CEO is worth worrying about (for them) even if they were to get caught, the term "golden parachute" exists for reasons.
> Gelsinger will have to change Intel’s culture and, to some extent, its technology, both of which are deep-seated. “He has to turn a huge ship around,” said Robert Maire, president of Semiconductor Advisors.
Yes and even harder they will have to convince potential customers that their culture has changed too.
Edit: This also adds the complexity of constantly managing trade-offs between (lower margin) external customers and the (higher-margin) in-house CPU business - it would be so easy to get that balance wrong.
I think you've been downvoted because of English grammatical errors, but if I understand your point correctly, I whole-heartedly agree:
All intel needs to do is make competitive chips at a good price. In past years, they resorted to more anti-competitive efforts, but at the end of the day, if you make good stuff and sell it at the right price, people will buy the crap out of it.
Lets look at who "people" are here - TSMC's biggest customers over the last three years:
AMD, Nvidia - direct Intel competitors
Hi-Silicon - no due to being based in PRC
Mediatek - based in Taiwan
Which leaves Apple, Broadcom and Qualcomm.
I suspect they will get some business from the last two but Apple clearly have a deep long term relationship with TSMC and have invested a lot into that relationship. Can't see it being easy to get them to move.
Maybe Intel is counting on political pressure to get the big US firms to use them?
There's a lot of interest from the big clouds on custom sillicon, I'm sure part of any govt cloud contract could include stipulations for use of sillicon manufactured or atleast verified on the western hemisphere.
That's a good point but isn't that just switching from a high margin x86 product to a lower margin Arm based CPU - seems like a defensive measure at best.
It looks like everyone is opening a fab in the next couple years ( for many reasons, one of them is political ) so I would guess in the next couple years ordering chips will be "cheap" because most fabs are desperate to recoup at least some of the gigantic investment and margins will be razor thin.
So my guess would be in the next decade, the real money will be on the design and IP not the manufacturing. ( excluding the bleeding edge half-a-nm stuff )
Everyone is opening a fab, but it only "counts" if you in fact succeed at making the chips. Otherwise, you've only spent $5 billion on a failure.
Intel has fallen behind because it failed to deliver its 10nm (now called 7nm) process in 2015. Intel tried-and-tried, but failed, to the point where supercomputers (ie: Aurora) were delayed because of the manufacturing mishap.
Just because everyone is __planning__ to open a fab doesn't mean that they'll be successful at it. If Intel fails to reach 5nm (TSMC-name), then TSMC will win.
In fact, TSMC has already started making 5nm chips called the Apple M1 Pro.
---------
Intel has finally delivered 10nm, but they were 5 years late. If it takes another 5 years to deliver their version of 5nm, then TSMC will be at 3nm by the time Intel reaches 5nm
-----------
Its all about manufacturing actually. All of these chip designs can be ported over to different manufacturers, so long as the different manufacturers are __CAPABLE__ of implementing the designs.
Everyone is buying the same machines from ASML. Machines aren't the issue, its process. Its nitty gritty manufacturing details that no one will talk about (because those details are the "moat", the secret sauce that makes TSMC do something that Intel can't).
It looks like Intel has figured out 7nm, just much later than expected.
Intel takes huge leaps, it’s how they kept 5e it lead for so long, but TSMC makes incremental progress and it’s paid off now that things are getting extremely difficult.
Actually, it's more than just the executive slots.
Intel is institutionally paranoid, and Grove is part of that fault. The old guard, even in engineering, is retiring/dying and have not groomed successors because that could have threatened their political position.
Can Gelsinger really fix this? I suspect probably not.
USA paid its best engineers to do advertisements and finance speculations. It means USA will be best in the world at advertisements but naturally will lag behind in other fields.
Or in other words, Intel fell into the same management trap as other legacy tech providers. They focused on paying industry wide 'median' salaries and bled talent.
Intel and TSMC are all using lithography machines by ASML, so what is the difference? Are they using the same lithography machines? Is there a difference in production processes used at TSMC vs Intel? Does anyone know any details?
https://wccftech.com/intel-vs-tsmc-chip-war-has-started-and-... Here they say that TSMC has a lead in the number of EUV lithography machines. " in 2021 and 2022, TSMC will have acquired 40 - 50 EUV machines while Intel will have procured 13 - 20 machines. Estimates based on Dutch company ASML, who is the sole supplier of the machines, suggest that TSMC had anywhere in between 30 - 63 EUV scanners by the end of last year." But Intel is playing catch-up "even as TSMC currently leads in machine inventory, Intel's future inventory could render this advantage moot. Mr. Gelsinger's belief that Intel aims for "leadership by 2024" seems to follow this timeline as well."
There was an interesting thread about this in the past if someone wants to dig it up, because offhand, a lot of us were a bit perplexed by the notion of Taiwan having water difficulties, and Intel building in Arizona, which is not known for an abundance of water.
Arizona solves the issue of a hypothetical PRC attack on the island of Taiwan.
PRC will have enough military strength to attack Taiwan and challenge US Carriers. We don't know if they'd win but... enough strength that they're a real threat... at least near China (with assistance from land-based hypersonic missiles and whatnot). PRC probably doesn't have the military strength to attack Arizona.
I think geopolitics is about US wanting to have control over chip production (being able to start or stop it, being able to ban certain countries and companies from getting access to the latest technology). I doubt conflict between China and Taiwan means anything here. China is not going to attack Taiwan which has US support.
> China is not going to attack Taiwan which has US support.
China will have a Navy that (by tonnage) rivals the US's Navy in a few short years. US still will have more capital ships but there's big, unanswered questions about what the proper tactics will be in the next great naval battles.
Would we rather have the US's "Supercarrier" strategy? Big ships with lots of money put into each one? Or would China's "little ships" strategy be superior?
Really: one question. Will China's hypersonic missiles truly be a capital-ship killer? If their weapon can truly kill a US Supercarrier, (or more specifically: if China __believes__ it can kill the US Supercarriers) then China will probably attack Taiwan.
China is building a navy for _some_ reason. Taiwan seems like the obvious target.
------
China's increasingly large Navy is a big enough deal that the USA has given away its nuclear submarine secrets to Australia to counter-act the threat. That doesn't really change the picture too much (US always had nuclear-submarines available), but its clearly a degree of saber-rattling going on right now between US and China.
I think the idea is that, though I'm sure the AZ site is unsatisfying as far as water goes, they can always pipe water in and it was picked for other reasons.
everyone already recycles water, but you can't make ultrapure water ultrapure again after using it to rinse off chemicals without significant work. "recycling" here means they use the waste water for evaporative cooling. that 2MW of power has to go somewhere and you don't dissipate the waste heat with a big heatsink, you evaporate water with it to get rid of the heat, as with most industrial facilities.
we had an actual intel plant manager who actually ran one of intel's fabs on one of the previous threads about this and HNers of course deigned to tell him his business, as usual. Love it.
The current shortage is making countries like the US realise that semiconductors are a critical commodity for the function of their advanced economies.
The US wants fabs back on their soil and will probably be willing to burn money to make it so.
Will the US be satisfied with just having foreign operated fabs on US soil? Or will the US want home growth (and firmly under US control) companies like Intel to have similar capabilities.
Frankly, IMHO things in the long term aren’t looking that great for TSMC. Some of their business is going to get eaten by “local” fabs - even if they can’t do leading edge nodes they will eating into TSMC’s business on the lower end.
There is also the danger of overcapacity destroying any profitability in the fab business altogether - guess who will get bailed out by the US government if that happens; hint: not TSMC.
> There is also the danger of overcapacity destroying any profitability in the fab business altogether - guess who will get bailed out by the US government if that happens; hint: not TSMC.
The boom/bust cycle is nothing new in semiconductors. It's a fairly regular thing. Also, keep in mind that consolidating and repositioning during downturns is how TSMC became TSMC.
I don’t mean temporary overcapacity. Overcapacity could be near permanent / long lasting as governments step in to prevent their local fabs from failing as they want to maintain local production capability no matter what.
In general, if semiconductors become a national security issue, I think there will be a fundamental change in the nature of the business. It will not be business as usual for TSMC.
> In general, if semiconductors become a national security issue, I think there will be a fundamental change in the nature of the business.
Maybe, who knows what the future will bring.
But I've heard this before [1] and once recessions hit, and budgets get cut, that kind of thing quickly becomes less tenable.
And then someone realizes that it takes a constant stream of incoming parts to keep the equipment inside any factory running for any meaningful amount of time, and that supply is not protected, and neither are their suppliers...
But I agree, things will not stay as they are. The only thing usual about this business is change.
[1] The US semiconductor manufacturing sector was under serious assault by the Japanese in the 80s. That was my first "trade war" in this industry...
That was before it was seen as a potential national security issue.
Semiconductors could end up like the US defense industry, which the US feeds with a continuous amount of money (recession or not) and is not outsourced for obvious reasons - but its products are sold to others; other countries outsource to the US. The US buys military tech on occasion but rarely ever directly buy the hardware in significant numbers - they will manufacture it in-house. It's great if it turns a profit but that's not the point.
Every country has their own defense industry if they can afford it.
That was just talk. With the current shortage there is plenty of evidence. Things are a lot different now too given how widely use semiconductors are.
That said, it remains to be seen how seriously the US government is about this. The $50 billion USD bill hasn’t passed through Congress yet. But if the US government does decide to make huge moves, things might change in a big way.
TSMC is no doubt pretty worried hence their setting up of fabs in the US now in an attempt to hopefully ward off US government action.
>Frankly, IMHO things in the long term aren’t looking that great for TSMC. Some of their business is going to get eaten by “local” fabs - even if they can’t do leading edge nodes they will eating into TSMC’s business on the lower end.
TSMC's Q3 shown bulk of their revenue come from higher end node. 7nm along is at 34%. the only other options for high end node is Samsung and Intel.
I don't see any long term problem for TSMC if they keep going at its current pace.
Not many foundry can do high end. GlobalFoundries vow out at 7nm a long time ago.
It is only a matter of time before PRC annexes the island of Taiwan and TSMC production is hugely disrupted. The USA did not win the last two proxy wars against China (North Vietnam and North Korea). This time will be no different.
There is no objective measure of "having won" the war. There never is because war results in extremely mixed outcomes. All that matters is the victors have the power to subjugate the people of the land, people's beliefs and consequences be damned.
There's an argument to be made that annexing Taiwan would actually be detrimental to the PRC: doing so would be hugely disruptive to the global supply chains that form a major source of income to China.
China is willing to decimate its own economy to achieve long term objectives. This includes annexing Hong Kong, shutting down celebrity culture, fintechs, and the tech industry.
I don't disagree that the PRC has their own best interests in mind. However, I don't see how they can be oblivious to the fact that going to war with and occupying Taiwan would not only tie up a significant part of their military, but would also immediately give their current trading partners both reason and rationale to cut off trade. Not to mention the brain drain from Taiwan that would follow shortly after.
If anything, I'm far more worried about the PRC exercising the soft power it's slowly been building in eg. part of Africa.
The idea is to get Taiwan during the tail-end of the USA's upgrade cycle.
China is building new ships today: roughly 3 aircraft carriers, and a huge number of Destroyers that will (on a tonnage basis), outnumber the entirety of the US Navy. They aren't capital ships, but a mass of lower-quality ships would probably still work if those hypersonic missiles can really kill our capital ships. (A big question in of itself: do China's hypersonic "ship killing" missiles work? If so, can the USA's defense systems shoot them down? If not, then why wouldn't China attack Taiwan?)
Meanwhile, the US Navy is just about to upgrade over the next 15 years. This means that a number of our ships will be docked, and the ships that are remaining will be older designs.
If China times it correctly, it can attack Taiwan and force the USA to defend with its obsolete ships, and before the new ships come online. In effect: the USA has already "spent" our money / resources on the upgrade process, but we haven't gotten the benefits of it yet.
----------
This is a once-in-a-generation opportunity to attack. Once the USA's fleet is upgraded, the opportunity closes (as USA will have newer / better technology ships, while China's old Carriers/Destroyers will be at the disadvantage. And China won't reclaim the advantage until it spends a huge amount of time and money on their own upgrade cycle).
Furthermore: this "delay in upgrade cycles" was caused by Bill Clinton's budget in 1996. It takes many years for a budgetary issue to come up, and the USA has spent huge amounts of money on its military from 2001 onwards (because of Afghanistan / Iraq). As such, no such "upgrade gap" will exist for 20+ years at the minimum after this opportunity passes.
Will China be willing to wait another 50 years for another opportunity like this? Or will China actually go for it? And remember, China has already dedicated incredible amounts of money towards building a new fleet. China's fleet is a "sunk cost", their only decision now is whether or not to use it.
The Nazis and Imperial Japanese lost the war the minute they made strategic blunders like... pissing off USA + Soviets __simultaneously__.
It turns out that the human ability to make strategic blunders is very possible. So if your best argument is "it'd be a strategic mistake to do X". Well... history is filled with horrible strategic blunders.
It'd probably be against American interests and a strategic blunder to divert forces from Afghanistan into Iraq on an notion of WMDs for example. Everyone makes strategic blunders all the time. I don't fault people for making them, I just try and study them and understand why it happens.
----------
There's something fascinating about war machines, like carriers, hypersonic cruise missiles, and destroyers. You have to see them in action to understand the intoxicating effects of these machines.
Once you understand how these machines can hypnotize a person... how they can make you feel like you can accomplish anything... then, and ONLY then, will you understand why Wars get started.
China's got Hypersonic cruise missiles. They know it. We know it. China knows the USA doesn't have them (!!!). That knowledge is intoxicating. China sees that as a military advantage. Will they be intoxicated by that knowledge to the point where they attack Taiwan?
Well, no one knows for certain. But they've looked at those missiles and have __already decided__ to build more ships than the entirety of the US Navy. So what does that tell you?
> The Nazis and Imperial Japanese lost the war the minute they made strategic blunders like... pissing off USA + Soviets __simultaneously__.
This is an important point and why I believe Wolf warrior diplomacy runs contrary to China's national interests. It has caused countries who previously viewed China as merely a competitor to view it as a direct threat and to take appropriate action. You know you've gone too far when your actions have resulted in Korea and Japan aligning with each other about the need for closer ties and to align with the U.S. on regional security.
The US officially doesn't rule out escalating from conventional to nuclear war against other nuclear-armed countries. Even if the possibility is remote, I imagine that nobody is suicidal enough to want to test the possibility.
And what if China performs the calculus: decides that Tactical nuclear weaponry is fine and that the USA isn't dumb enough to escalate into Strategic nuclear weaponry?
There's a big difference between nuclear powers lobbing 100kilo-ton bombs at each other's navies in the Pacific, and nuclear powers using 10-Megaton bombs at each other's cities.
Given that China has decided to make a large mass of cheap Destroyers, what do you think China is thinking about this next war? Tactical nuclear weaponry would favor the one who has a more dispersed naval force, and be a bigger problem for USA's multi-billion $$ supercarriers, would it not?
-------
Lets say China nukes a military Supercarrier. Will the USA respond by nuking Beijing? Of course not. Knowing that USA is tactically in the disadvantage from using nukes (due to the composition of our fleet vs the Chinese fleet), we know that USA won't escalate to using even tactical nukes.
On the one hand, that makes sense; on the other, it would still be the first use of nuclear weapons since WW2. The last close calls, like Cuba, didn't even involve any bombing, just the threat of one power's blockade against another's ships, and that one is widely seen as having come close to mutual destruction. I don't think that actually lobbing nukes at each other would be less dangerous.
Also, accidents happen, and it's not clear how secure strategic weapons are against everything from a zealous airman to a submarine cut off from its command and having to make a judgment call. Daniel Ellsberg's book (The Doomsday Machine) suggests that this kind of accident during a crisis has come close to triggering a full nuclear war several times.
To summarise, I'm sure the tactics look good for China, but escalating to nukes still seems like a suicidal move, considering the risk of accident or miscalculation.
The DF-21, available maybe 10 years ago, would have the range to hit ships all the way out to Japan or so. That's really all you need to know: China is trying to develop (or already has developed) "ship-killing" hypersonic missiles.
But of course, no one knows if they work vs US ships (can US Ships deflect the missile? Do we have flares or something that the missile will mistake for a ship? Can the US shoot down the missile before it reaches a ship? No one knows).
You can see in the Eastern/Southern theaters of China, they have roughly 500,000 Army troops vs Taiwan's 80,000 troops. (China has a military of over 1-million, but most are deployed elsewhere). China also has:
"The PRC has the largest navy in the world, with an overall battle force of
approximately 350 ships and submarines including over 130 major surface combatants. In
comparison, the U.S. Navy’s battle force is approximately 293 ships as of early 2020. China
is the top ship-producing nation in the world by tonnage and is increasing its shipbuilding
capacity and capability for all naval classes."
So PRC already outnumbers the USA in terms of ships. In a few more years, PRC will outnumber the USA in terms of tons. China is definitely going for "build more" and not necessarily "build better". There hasn't been a modern naval battle since like WW2 between major powers, so no one really knows if this is a good or bad strategy. But its certainly a strategy.
USA of course has our 10x Nimitz supercarriers + 1x Ford supercarrier (going through our carrier-upgrade cycle currently, +F35 project I'm sure you're aware of). USA has a number of "Amphibious assault ship", which are sorta-kinda like carriers (except for the Marines instead of Navy), but do contain VTOL aircraft like F35-B variant.
I don't think anyone really knows for sure how the battle will play out. The US ships (ie: the "Supercarriers") are incredible war machines, and even the smaller Amphibious assault ships would probably go toe-to-toe with even the Chinese carriers. But how does that play out with China's missiles and airforce?
I don't think the timing is favorable for China. Based on the papers I've read, China is about 15 years away from having a sufficient amphibious capability needed to invade Taiwan. That very likely means a sustained blockade, which will be extremely difficult politically, militarily, and economically. Surely such an invasion would be swiftly followed by harsh sanctions from many of China's trading partners. Considering the systemic risk China is presently baking into their economy by not allowing defaults to occur, what will be the appetite for adventurism of this sort with things unsettled at home? Is Xi willing to risk completely destabilizing China's economy to capture Taiwan? That sounds extremely reckless and unlikely, but stranger things have happened.
China has a policy of making everything in-house and already requires exit visas for its people. Brain drain can't happen if you are not allowed to leave.
> ...after a couple of years of unrest and social instability.
If China wins Taiwan vs the USA, there wouldn't be any unrest or instability. The Chinese people would be proud of their accomplishment, and the rise of China as a military power.
Like... what "instability" occurred from Russia's annexation of Crimea? None what so ever.
>Knowing that the PRC tends to take the long view,
They do, and that's the problem. The shot clock is ticking down quickly: fertility rate in PRC just hit 1.3 births per woman. Total population is projected to halve by the end of the century. The current cohort of military-age males is shrinking, and there won't be another one to replace it. The PLA will never be stronger than it is today.
If Xi wants to take Taiwan, he has to do it now, because he's not going to get another chance. PRC manufacturing output is cratering due to power shortages and the global shipping and semiconductor convulsion, so the economy is going to get clobbered anyway. They're already doing recon flights through the ADIZ and running freedom of navigation naval patrols. If there's going to be a Taiwan annexation, it'll happen in the next 12 months.
The US Navy is hapless and the Taiwanese army is supine. Evergrande is collapsing and there's already a wide array of tariffs and sanctions against PRC leadership. They accidentally caused a global pandemic and shrugged off any political ramifications. They've already taken half a dozen hits this year and survived. Why not take another?
It depends what you mean by long term objectives. The supreme objective of the CCP is to stay in power in China. All and every other objective is subsidiary to this.
The CCP decided many decades ago that economic growth was absolutely necessary for them to stay in power because it is required in order to maintain social stability. An economy with hundreds of millions of unemployed peasants in the countryside, and feeble industries was not sustainable. This is why they purged a bunch of ideological marxists and embraced communism with Chinese characteristics, AKA Capitalism.
For a long time the CCP maintained that 7%+ economic growth was necessary in order to absorb the unused labour pool and maintain social stability.
Needless to say a war with Taiwan and the US, plus blockades by the rest of the world would be a somewhat risky situation from a social stability point of view. It would be a complete throw of the dice, and for all Chinese people love gambling with money, that's not the CCP way with politics.
Would the world respond extremely negatively to a Taiwan scenario? Or much more negatively than a Hong Kong scenario? (Which literally just happened, and not much seems to have changed as far as I can tell)
We know the USA would respond militarily. But how many other countries would be pulled into the conflict? Germany still trades with Russia despite the Crimea annexation, right? Clearly USA / UK / Australia are making plans but is there anyone else who'd join in the fight?
Like, realistically, what would happen under a Taiwan scenario? Do you think McDonalds and Apple will just leave China? Will Tencent games be banned in the USA? Will USA really make League of Legends and Genshin Impact illegal during the war?
Or, would the world continue as usual? Would US Video Gamers continue playing their games, McDonalds continue China operations / Apple continue making their iPhones while the whole military stuff happens? Frankly, I see this as the more realistic scenario.
We can already see how video gamers reacted to Hong Kong: with a bunch of companies capitulating to Chinese demands.
Hing Kong is a completely different situation, it's always been mainland Chinese territory as Britain only leased it. Full mainland control was always the plan, it was just a matter of time. Also the political administration, elected under a system implemented under British rule, supported the recent changes.
I'm in no way justifying what the CCP did to Hong Kong, but frankly it was inevitable. That is not the case with Taiwan, which has a population and democratically legitimate government completely committed to independence.
An invasion of Taiwan would be much more similar historically speaking to the German invasion of Poland, or the Korean war. It's utterly and completely unacceptable, and would be legitimate grounds for total war. Should China succeed in conquering the island it would not be possible to counter-invade, but an indefinite blockade of transport into and out of China and total interdiction of it's air and sea traffic both civilian and military would be entirely proportionate and the minimum I would expect. See the blockade of North Korea for an example. We'd buy our phones and video games elsewhere.
> but an indefinite blockade of transport into and out of China and total interdiction of it's air and sea traffic both civilian and military would be entirely proportionate and the minimum I would expect
And China outnumbers the US warships ship-for-ship currently, and China is outbuilding more warships ton-for-ton than even the USA.
China also has a larger (though inferior) air force, as well as a huge number of (untested) missiles. But the USA won't have many bases to stage our Air Force from (Taiwan is so close to China, and so far away from Japan / South Korea). So really, its China's Airforce + China's Missile-force + China's Navy vs the US Navy + US Marines + maaaayyyyybe a very stretched thin US Air Force.
If the USA wants to enforce a naval blockade, they'd have to win a war against China. That's a lot of advantages to China frankly.
I'm certain the US could beat China in a "fair fight" for Antarctica. But the issue with the Taiwan scenario is that the Chinese military is growing ever stronger, and its conceivable that China could beat the USA in that region. If China does happen to win militarily, then a US-sponsored blockade would be impossible.
The blockade would mostly operate in the Pacific Ocean and Indian Ocean. Interdicting all Chinese sea traffic, and all over-ocean air traffic globally is entirely doable. The Chinese military has zero practical operational experience, while the US has an extensive network of global bases and experience up to the eyeballs.
There is simply no way China is going to contend with the US globally. Zero trade with the US, Europe, Japan, South Korea and Australia is a given. A sea blockade of trade with Africa, South and Central America, and Australia is fine. That leaves land and air trade only with Asia. How are they going to get oil? China's not going to establish military domination over the Middle East any time soon.
As a Brit I don't see us staying neutral, or Australia. Hence the nuclear subs Australia is buying. In a conflict like this, they would use them.
If China annexes Taiwan it will be for reasons that have nothing to do with TSMC.
There may be a short term benefit but if they get cut off from ASML (Europe), Applied Materials (US), etc. it's going to be tough to keep their machines in service and stay leading edge. And those companies have very deep international supply chains too and may also get sanctioned.
I couldn't care less. Supply chains are already massively disrupted, I can't get the electronics I care about at affordable prices and lead times, and everyday costs are out of control in my country (USA).
i believed people is overplaying China invasion. TSMC is a business. a saying goes "business man have no border".
TSMC have a lot of fabs in Taiwan. that doesn't mean TSMC as a business can't build fab elsewhere.
TSMC is already building a fab in AZ USA and reach an agreement with Sony to build a fab in Japan. TSMC also have fabs in China and India govrt is looking to Taiwan for a fab in India.
> In fact, TSMC has already started making 5nm chips called the Apple M1 Pro.
This is not a recent development. TSMC has been mass producing 5nm chips for slightly over a year now. The first 5nm chip on the market was the Apple A14 (iPhone 12 line) released almost exactly one year ago, and soon thereafter (November 10th, 2020) the first M1 Macs.
Samsung has had their ""5nm"" chips out since early 2021 with the Exynos 2100 and Snapdragon 888. It's in the ballpark of TSMC's process but not as dense.
Intel hasn't even announced their first mass market 7nm (or ""Intel 4"", sigh) products yet. ""Intel 4"" is inferior to Samsung's and TSMC's 5nm processes on paper, and it'll likely not even be next year that we're going to see products with them.
I don't think Intel's foundries will be competing on the leading edge of the market. They'll have be trying to have value focused production with their 14nm and 10nm processes, I would guess. There should be plenty of demand there still, but it'll be a good while - if ever - before Intel is leading again.
Intel's numbers are theoretical. They're still relying heavily on multi patterning for their current chips. They have yet to demonstrate competency with EUV. Intel 4 is also projected for 2023. Tsmc are producing 5nm chips right now.
> Intel said Thursday its "Alchemist" graphics chips will be made by TSMC using the latter's newly named "N6" chipmaking technology, an upgraded version of its "N7" technology.
I mean, I guess they're already "winning". Intel's own chip-making group has begun to use TSMC. Hopefully Intel can figure something out about what its like to be a customer, so maybe they can be a better supplier?
But yes: customers, market share, and knowledge. By having more customers, TSMC can perform more experiments on more leading edge nodes, like 3nm or smaller.
Up until the kaboom, microprocessors have become a commodity. Granted one with huge startup costs, but a commodity nonetheless. (There's a reason why Intel used to be top forbes company, and why today it is still stuck in the $100-200B market cap range compared to $2T amazons & apples.)
Could you elaborate? AMD spun their fabrication arm off into GlobalFoundries a while back, and they seem to be doing okay. I thought that would be a good precedent for Intel divesting its fabs too, but I'd love to hear any counterpoints to that.
AMD spun off GlobalFoundries to save itself, and suffered greatly for many years afterward. Let's not pretend it was some kind of masterstroke done by a company that was firing smoothly on all cylinders.
Do a Google search for the stock price. AMD announced the spinoff in October 2008.
This makes sense, because of the move to application-specific ICs.
Products used to be generic ICs on custom boards. That's shrinking to a single application-specific IC. You can usually get all the transistors you need for a product onto one application-specific part. But that part will not be a commodity.
Much of this is semi-custom. You plug together a few pre-defined CPU cores, a GPU, a video decoder, some software defined radios, etc. and you have a phone.
Having more fabs in more locations is a good thing, for robustness.
192 comments
[ 3.1 ms ] story [ 224 ms ] threadChina's dithering might be centred partly around considerations of how they can overtake a country while keeping its silicon fabs in tact.
Depends on how you see it. During most of humanity's existence, we were hunters and gatherers, with simple and small societies - there wasn't much need for advanced math and there was not really a possibility for a society to feed a thinker. Only when complex societies like ancient Greece and Rome were possible thanks to early industrialized farming, there was a possibility for people to do nothing "productive" towards the society itself surviving but instead invest resources into advancing society.
Of course with chips, patents and know how that e.g. Intel or ASML have are two things.
Ultimately this is good as it drives companies to continue innovating. Intel's current problems are related to them forgetting to do that.
And as far as "innovating" in this field, we're almost at the end of the run - there are, as far as I'm aware, no processes planned beyond 3nm, and even 3nm is astonomically complex and expensive.
Shouldn't that make the patent invalid? I thought the requirement is that the patent should give enough info so that "person skilled in the art" can reproduce the invention.
Any stories you have (and are allowed to) tell?
Regimes trying to copy something strongly reminds me of that anecdote. They usually manage to waste a lot of resources only just barely catching up, copying, and then they still end up buying the stuff that works from the companies that have the decades and the experience doing it.
Making this chip tech sounds orders of magnitude more complex.
Reading the article I only think about what a major fumble it was geo-strategically that the people with good ideas went to Taiwan.
"The U.S. Department of Energy (DOE) has announced an agreement with ASM Lithography (Einhoven, Netherlands) that would allow the company to participate in the extreme ultraviolet (EUV) lithography development effort now underway in Livermore, Calif."
https://www.eetimes.com/u-s-gives-ok-to-asml-on-euv-effort/
EUV is a joint research by US Dept of Energy and several US companies. ASML got invited to join the research based on conditions lay out by US Dept of Energy.
One issue is there's a significant conflict of interest between their device and CPU businesses. Apple left them precisely because Samsung is a competitor in handsets, and other handset manufacturers are also wary of buying from them for the same reason.
As a side note it is true TSMC deserves a lot of credit for the advantages of the Apple M-series CPUs. Without the 5nm TSMC process they wouldn't be anywhere near as impressive. However Apple has been working very closely with TSMC on it, and their long term financial commitments been a big factor in TSMC being able to move forward so decisively on developing the process and investing in future nodes and capacity. Apple ditching Samsung as a supplier has had a big influence on the shape of the CPU industry. The (currently) top comment by WithinReason highlights part of this dynamic.
I don't think what bothered Apple is relevant to other phone makers. Case in point, Qualcomm doesn't mind having Samsung fab their SoCs, and even their modems, a domain where Samsung has repeatedly tried to catch up!
Anyway in the current Android smartphone landscape, everyone is using ARM reference CPU designs.
Ultimately no one can shake their culture, it follows you like your shadow.
Google's is cheap and low-end. They have become the Walmart of tech companies.
We're going to look back on this story as one of the craziest product launches of all time. I genuinely have no idea how Apple secured 5nm so early (and blocked other customers from getting a piece until 2022), I have to imagine that it was a last-minute decision they pushed out to TSCM, judging from their response. So many different parts of the M1 are pushing boundaries we've never hit before (for better and worse), so it will be deeply interesting to see how Apple scales a CPU that barely exists in it's current form. Some of my friends in the manufacturing biz were absolutely dumbfounded by the fact that Apple fit 64 gigs of memory on the M1 Max (the thing is surrounded by memory chips), which makes me extremely curious about their next pro hardware.
Fun thought experiment, too: Would the M1 have still been such a knockout hit on TSCM 7nm? It certainly wouldn't be as good as it's current incarnation, but I'd bet dollars-to-donuts that a big.LITTLE ARM chip would have worked pretty well on last-gen silicon.
I would not be at all surprised if they laid the groundwork for this partnership back in 2015 or even earlier. The only reason Apple dual-sourced the A9 was because TSMC didn't have the capacity to deliver the full requirement, but Apple had already been trying to get off Samsung for years.
Apple has a long track record of early, heavy capital investment in core technology manufacturing capacity and long term contracts with suppliers. This serves the dual strategic purposes of securing exclusive access to new tech, and driving down the long term costs of that tech. They did this with retina displays, in that case Apple literally owned the manufacturing equipment operated by Sharp.
In order to build out fab capacity TSMC needed long term high volume guaranteed orders. Apple delivered that. There was nothing last minute about any of this, I think it's always been the plan.
Apple is also willing to invest in technology in return for exclusivity for awhile.
Commodore manufactured the 6502 with its MOS subsidiary, but I don't believe that Apple or Atari hesitated from using the 6502 in their computers.
If Samsung came out with a contemporary CPU with 2x the compute per watt of the M1, Apple would probably be interested in it.
Notably, it seems like Intel had the capability and head start to have been able to develop such a CPU by now, but instead they held back progress to maximize their profits. It was apparently a bad long term strategy, as they've now lost their lead. The market moved on around them.
“When you have volume orders, then you can do yield experiments, you can improve your yield, and yield is everything because that’s how you cover your costs,”
"Foundries that can juggle multiple clients and technologies can swiftly advance on their competitors. For example, TSMC’s scale allowed it to master extreme ultraviolet lithography faster than anyone else, which reduced the number of steps it took to make advanced chips and boosted the throughput of its fabs."
Contrary to Moore's law, it's not slowing down.
My understanding is that it’s a general phenomenon. The more we do something, the faster we get at it (either as a species or individuals), the more tools we accumulate to help us smoothie out any bumps, tackle larger projects more efficiently, etc.
Economist would say that volume gives you:
1) Increasing marginal returns after massive fixed capital investments.
2) increasing returns to scale from variable investments when volumes increase.
Where were they cancelled? I know about Larrabee and Itanium of course. But the others? Would they have succeeded in the market if pushed?
Or was Intel just going in all directions a bit randomly without strong purpose?
Larrabee is Xeon Phi. Aren't they shipping today?
These never saw the light of day, only one taped out. I don't know why, I was too busy engineering to care:
- a Pentium3 + on-die, integrated graphics
- a 7.2 GHz pentium 4 (yes, 7.2 GHz. in 2002)
- the successor to the i740 graphics which was cannibalized for the i815 chipset
- another Pentium 4 variant which I can't recall what it was, but we spun up an entire arch team to work on it
- variants of Itanium (McKinley, Madhall)
> Or was Intel just going in all directions a bit randomly without strong purpose?
DING DING DING.
Around the 90s, the big OOOE scare and push to move away from it seemed to be based on projections showing wire delay and super-linear scaling structures like schedulers would bite OOOE very hard and make it nonviable.
This might have been technically true -- Pentium Pro reorder buffer was 20 entries while new cores might have only 10 times that, about an order of magnitude. Processor widths only by about 2x. IPC has increased about an order of magnitude as well (ignoring SIMD). While transistor density increased 3 or 4 in that time.
So I don't see a huge problem with that general prediction. What I wonder about is why on earth they thought in-order VLIW could solve this in a better way. Itanium scaling ran out of steam well before the out of order designs. And huge widths never happened or probably didn't help much because they're always just going to get stuck on taken branches, data dependencies, etc.
I liked Itanium when it first came out, it was a floating point monster and even on integer it shaped up pretty well against the likes of POWER4 in the TPC-C wars, exchanging the lead for several generations. But it became apparent that it depended on heroically fast (and therefore sadly tiny) L1 cache which halted its frequency scaling. It depended on a heroically large L3 cache which was became a problem in the multi core era. All because it dealt badly with dynamic hazards like cache misses, and executing beyond them to generate memory level parallelism.
Sure it had advanced loads which were very cool on paper, but never really were such a capable substitute.
I don't hate EPIC/VLIW. Some people hate Itanium because they claim it killed of a lot of diversity in the CPU market. I don't think that it did, and wouldn't hate it even if I thought it did. I can see why it failed though, and that's because it was technically inferior to out of order designs for most of its target markets.
Intel decided not to do EUV when others did and squeeze yet another node from older technology and it backfired phenomenally.
Even when everything is done right, the risks are high. The competitive high-end node business went from 4 to 2 in one generation. From Intel, GF, TSMC, Samsung to TSMC, Samsung. Intel is trying to get back and they have to burn billions.
skills have to pivot I guess
Note: Intel 4nm is a Full, new Node while TSMC 4nm is an optimisation of their 5nm.
Of course the roadmap and time table from Intel is often not a reliable indicator and also doesn't tell you whether it is product shipment or if they are only ramping. My guess is that Intel is running as fast as they can to reach those target. That is also not including capacity difference with node from both companies.
I think Intel under Pat Gelsinger is making the same transition as Andy Grove did to during his era.
did Intel skip 7 and 5nm?
https://www.anandtech.com/show/17027/intel-reports-q3-2021-e...
A lot of sites don't update the graph for pre-market movements and only show price movements from when the market is open. On yahoo finance you can see pre-market movements if you open the full-screen graph (simply click the graph or click "full-screen" on mobile). https://finance.yahoo.com/quote/INTC?p=INTC&.tsrc=fin-srch
https://anysilicon.com/five-semiconductor-companies-hold-53-...
Edit: it is maybe a little more complex than that, because much of that is memory, and memory and logic processes may not be able to share the same learnings and developments.
Complete speculation, but I would guess this is mostly due to Apple. Apple wants the best process and they’re able to pay for it. Further, Apple is mistrustful of Samsung so they were happy to drop them and switch to TSMC.
I think this highlights the difficulty Intel will face. Samsung’s foundry is independent from Samsung’s phone business, and I don’t think anyone has alleged otherwise. But even if Intel does manage to keep its two sides at arms length, will that be enough to attract business from competitors like Apple and AMD? I’d guess no.
- Do other firms trust Intel with their IP? Apple has a new feature for their next generation - Intel quickly becomes aware of this and this information leaks into Intel's product plans.
- Is Intel a reliable foundry partner? TSMC have a 30+ year track record, Intel have none as a third party foundry. Will they change their strategy again?
- Will other firms get a fair prioritisation of resources in the fabs? Or will Intel's own products be favoured?
[0] https://ol.reddit.com/r/hardware/comments/ik9lw1/updated_int...
[1] https://www.agner.org/optimize/blog/read.php?i=49
^ For those interested in the storied history of Intel hurting AMD
Obviously it doesn't work that way in practice.
FAANG companies each individually have higher annual revenue than most countries' GDPs. In the absence of government intervention, there wouldn't be much to stop them from turning into British East India Companies and establishing banana republics, raising private armies, etc.
In an extreme scenario, some kind of corporate Genghis Khan could hypothetically consolidate the S&P 500 into a single megacorp if the DOJ allowed it to happen, and that company's revenue would be about half the US GDP (making it the third largest "economy" in the world). It would effectively be a hostile state right in our own backyard. Would we even have the ability to enforce taxation if they said no? What would stop them from wielding that incredible wealth to effectively buy a majority stake in government?
A lot of the anti-trust breakups result in several good businesses and maybe some competition/consumer benefits too.
The point is that doing it and being caught out would be something approaching corporate suicide, and sane leadership would not risk it.
This places it in the bucket of 'unlikely, high impact' risks for potential clients. They'd want to design process and contractual language to try to make such perfidy apparent and not bet the farm, but I doubt it would be a deal breaker.
I would think IP concerns would dominate in such an arrangement.
No it would not. Just like in any other case where a corporation had to pay a huge fine - they do it, or it even gets lowered significantly, and then move on. The executives who were responsible retired long ago and the motto of shareholders and management always is "look to the future and don't bother me with old problems".
You make quite an assertion based on what - the opposite has already happened many times. Risk that has to be paid for much later often is worth it, especially when it's going to happen to a new team. Not that the risk for a CEO is worth worrying about (for them) even if they were to get caught, the term "golden parachute" exists for reasons.
Yes and even harder they will have to convince potential customers that their culture has changed too.
Edit: This also adds the complexity of constantly managing trade-offs between (lower margin) external customers and the (higher-margin) in-house CPU business - it would be so easy to get that balance wrong.
All intel needs to do is make competitive chips at a good price. In past years, they resorted to more anti-competitive efforts, but at the end of the day, if you make good stuff and sell it at the right price, people will buy the crap out of it.
AMD, Nvidia - direct Intel competitors
Hi-Silicon - no due to being based in PRC
Mediatek - based in Taiwan
Which leaves Apple, Broadcom and Qualcomm.
I suspect they will get some business from the last two but Apple clearly have a deep long term relationship with TSMC and have invested a lot into that relationship. Can't see it being easy to get them to move.
Maybe Intel is counting on political pressure to get the big US firms to use them?
So my guess would be in the next decade, the real money will be on the design and IP not the manufacturing. ( excluding the bleeding edge half-a-nm stuff )
Intel has fallen behind because it failed to deliver its 10nm (now called 7nm) process in 2015. Intel tried-and-tried, but failed, to the point where supercomputers (ie: Aurora) were delayed because of the manufacturing mishap.
Just because everyone is __planning__ to open a fab doesn't mean that they'll be successful at it. If Intel fails to reach 5nm (TSMC-name), then TSMC will win.
In fact, TSMC has already started making 5nm chips called the Apple M1 Pro.
---------
Intel has finally delivered 10nm, but they were 5 years late. If it takes another 5 years to deliver their version of 5nm, then TSMC will be at 3nm by the time Intel reaches 5nm
-----------
Its all about manufacturing actually. All of these chip designs can be ported over to different manufacturers, so long as the different manufacturers are __CAPABLE__ of implementing the designs.
Everyone is buying the same machines from ASML. Machines aren't the issue, its process. Its nitty gritty manufacturing details that no one will talk about (because those details are the "moat", the secret sauce that makes TSMC do something that Intel can't).
It looks like Intel has figured out 7nm, just much later than expected.
TSMC has competent leaders like Morris Chang and Mark Liu.
Intel had lost it's way after Andy Grove gave up the helm. Pat Gelsinger grew up under Gordon and Grove and the old guard is back.
Intel is institutionally paranoid, and Grove is part of that fault. The old guard, even in engineering, is retiring/dying and have not groomed successors because that could have threatened their political position.
Can Gelsinger really fix this? I suspect probably not.
Absent any geopolitical difficulties... but that's a whole other discussion.
It takes a lot of water to make the ultra pure water needed to wash the wafers. Without access to lots of water they will too will have issues.
PRC will have enough military strength to attack Taiwan and challenge US Carriers. We don't know if they'd win but... enough strength that they're a real threat... at least near China (with assistance from land-based hypersonic missiles and whatnot). PRC probably doesn't have the military strength to attack Arizona.
China will have a Navy that (by tonnage) rivals the US's Navy in a few short years. US still will have more capital ships but there's big, unanswered questions about what the proper tactics will be in the next great naval battles.
Would we rather have the US's "Supercarrier" strategy? Big ships with lots of money put into each one? Or would China's "little ships" strategy be superior?
Really: one question. Will China's hypersonic missiles truly be a capital-ship killer? If their weapon can truly kill a US Supercarrier, (or more specifically: if China __believes__ it can kill the US Supercarriers) then China will probably attack Taiwan.
China is building a navy for _some_ reason. Taiwan seems like the obvious target.
------
China's increasingly large Navy is a big enough deal that the USA has given away its nuclear submarine secrets to Australia to counter-act the threat. That doesn't really change the picture too much (US always had nuclear-submarines available), but its clearly a degree of saber-rattling going on right now between US and China.
we had an actual intel plant manager who actually ran one of intel's fabs on one of the previous threads about this and HNers of course deigned to tell him his business, as usual. Love it.
https://news.ycombinator.com/item?id=28683036
https://youtu.be/Dq04GpzRZ0g
Geopolitics have been figured out. TSMC's next 5nm plant will be in Arizona. https://appleinsider.com/articles/21/06/01/tsmc-starts-const...
The current shortage is making countries like the US realise that semiconductors are a critical commodity for the function of their advanced economies.
The US wants fabs back on their soil and will probably be willing to burn money to make it so.
Will the US be satisfied with just having foreign operated fabs on US soil? Or will the US want home growth (and firmly under US control) companies like Intel to have similar capabilities.
Frankly, IMHO things in the long term aren’t looking that great for TSMC. Some of their business is going to get eaten by “local” fabs - even if they can’t do leading edge nodes they will eating into TSMC’s business on the lower end.
There is also the danger of overcapacity destroying any profitability in the fab business altogether - guess who will get bailed out by the US government if that happens; hint: not TSMC.
The boom/bust cycle is nothing new in semiconductors. It's a fairly regular thing. Also, keep in mind that consolidating and repositioning during downturns is how TSMC became TSMC.
https://www.youtube.com/watch?v=Z7QkIECEkVc
In general, if semiconductors become a national security issue, I think there will be a fundamental change in the nature of the business. It will not be business as usual for TSMC.
Maybe, who knows what the future will bring.
But I've heard this before [1] and once recessions hit, and budgets get cut, that kind of thing quickly becomes less tenable.
And then someone realizes that it takes a constant stream of incoming parts to keep the equipment inside any factory running for any meaningful amount of time, and that supply is not protected, and neither are their suppliers...
But I agree, things will not stay as they are. The only thing usual about this business is change.
[1] The US semiconductor manufacturing sector was under serious assault by the Japanese in the 80s. That was my first "trade war" in this industry...
https://www.washingtonpost.com/archive/business/1985/09/22/s...
Semiconductors could end up like the US defense industry, which the US feeds with a continuous amount of money (recession or not) and is not outsourced for obvious reasons - but its products are sold to others; other countries outsource to the US. The US buys military tech on occasion but rarely ever directly buy the hardware in significant numbers - they will manufacture it in-house. It's great if it turns a profit but that's not the point.
Every country has their own defense industry if they can afford it.
That said, it remains to be seen how seriously the US government is about this. The $50 billion USD bill hasn’t passed through Congress yet. But if the US government does decide to make huge moves, things might change in a big way.
TSMC is no doubt pretty worried hence their setting up of fabs in the US now in an attempt to hopefully ward off US government action.
TSMC's Q3 shown bulk of their revenue come from higher end node. 7nm along is at 34%. the only other options for high end node is Samsung and Intel.
I don't see any long term problem for TSMC if they keep going at its current pace.
Not many foundry can do high end. GlobalFoundries vow out at 7nm a long time ago.
https://investor.tsmc.com/english/encrypt/files/encrypt_file...
https://www.predictit.org/
If anything, I'm far more worried about the PRC exercising the soft power it's slowly been building in eg. part of Africa.
China is building new ships today: roughly 3 aircraft carriers, and a huge number of Destroyers that will (on a tonnage basis), outnumber the entirety of the US Navy. They aren't capital ships, but a mass of lower-quality ships would probably still work if those hypersonic missiles can really kill our capital ships. (A big question in of itself: do China's hypersonic "ship killing" missiles work? If so, can the USA's defense systems shoot them down? If not, then why wouldn't China attack Taiwan?)
Meanwhile, the US Navy is just about to upgrade over the next 15 years. This means that a number of our ships will be docked, and the ships that are remaining will be older designs.
If China times it correctly, it can attack Taiwan and force the USA to defend with its obsolete ships, and before the new ships come online. In effect: the USA has already "spent" our money / resources on the upgrade process, but we haven't gotten the benefits of it yet.
----------
This is a once-in-a-generation opportunity to attack. Once the USA's fleet is upgraded, the opportunity closes (as USA will have newer / better technology ships, while China's old Carriers/Destroyers will be at the disadvantage. And China won't reclaim the advantage until it spends a huge amount of time and money on their own upgrade cycle).
Furthermore: this "delay in upgrade cycles" was caused by Bill Clinton's budget in 1996. It takes many years for a budgetary issue to come up, and the USA has spent huge amounts of money on its military from 2001 onwards (because of Afghanistan / Iraq). As such, no such "upgrade gap" will exist for 20+ years at the minimum after this opportunity passes.
Will China be willing to wait another 50 years for another opportunity like this? Or will China actually go for it? And remember, China has already dedicated incredible amounts of money towards building a new fleet. China's fleet is a "sunk cost", their only decision now is whether or not to use it.
It turns out that the human ability to make strategic blunders is very possible. So if your best argument is "it'd be a strategic mistake to do X". Well... history is filled with horrible strategic blunders.
It'd probably be against American interests and a strategic blunder to divert forces from Afghanistan into Iraq on an notion of WMDs for example. Everyone makes strategic blunders all the time. I don't fault people for making them, I just try and study them and understand why it happens.
----------
There's something fascinating about war machines, like carriers, hypersonic cruise missiles, and destroyers. You have to see them in action to understand the intoxicating effects of these machines.
Once you understand how these machines can hypnotize a person... how they can make you feel like you can accomplish anything... then, and ONLY then, will you understand why Wars get started.
China's got Hypersonic cruise missiles. They know it. We know it. China knows the USA doesn't have them (!!!). That knowledge is intoxicating. China sees that as a military advantage. Will they be intoxicated by that knowledge to the point where they attack Taiwan?
Well, no one knows for certain. But they've looked at those missiles and have __already decided__ to build more ships than the entirety of the US Navy. So what does that tell you?
This is an important point and why I believe Wolf warrior diplomacy runs contrary to China's national interests. It has caused countries who previously viewed China as merely a competitor to view it as a direct threat and to take appropriate action. You know you've gone too far when your actions have resulted in Korea and Japan aligning with each other about the need for closer ties and to align with the U.S. on regional security.
The US officially doesn't rule out escalating from conventional to nuclear war against other nuclear-armed countries. Even if the possibility is remote, I imagine that nobody is suicidal enough to want to test the possibility.
There's a big difference between nuclear powers lobbing 100kilo-ton bombs at each other's navies in the Pacific, and nuclear powers using 10-Megaton bombs at each other's cities.
Given that China has decided to make a large mass of cheap Destroyers, what do you think China is thinking about this next war? Tactical nuclear weaponry would favor the one who has a more dispersed naval force, and be a bigger problem for USA's multi-billion $$ supercarriers, would it not?
-------
Lets say China nukes a military Supercarrier. Will the USA respond by nuking Beijing? Of course not. Knowing that USA is tactically in the disadvantage from using nukes (due to the composition of our fleet vs the Chinese fleet), we know that USA won't escalate to using even tactical nukes.
Also, accidents happen, and it's not clear how secure strategic weapons are against everything from a zealous airman to a submarine cut off from its command and having to make a judgment call. Daniel Ellsberg's book (The Doomsday Machine) suggests that this kind of accident during a crisis has come close to triggering a full nuclear war several times.
To summarise, I'm sure the tactics look good for China, but escalating to nukes still seems like a suicidal move, considering the risk of accident or miscalculation.
But I can conceive a hypothetical Chinese Admiral who disagrees with the both of us. And that's all that it will take.
https://en.wikipedia.org/wiki/DF-21#DF-21D_(CSS-5_Mod-4)_Ant...
The DF-21, available maybe 10 years ago, would have the range to hit ships all the way out to Japan or so. That's really all you need to know: China is trying to develop (or already has developed) "ship-killing" hypersonic missiles.
But of course, no one knows if they work vs US ships (can US Ships deflect the missile? Do we have flares or something that the missile will mistake for a ship? Can the US shoot down the missile before it reaches a ship? No one knows).
-------
For China's military composition: https://media.defense.gov/2020/Sep/01/2002488689/-1/-1/1/202...
You can see in the Eastern/Southern theaters of China, they have roughly 500,000 Army troops vs Taiwan's 80,000 troops. (China has a military of over 1-million, but most are deployed elsewhere). China also has:
"The PRC has the largest navy in the world, with an overall battle force of approximately 350 ships and submarines including over 130 major surface combatants. In comparison, the U.S. Navy’s battle force is approximately 293 ships as of early 2020. China is the top ship-producing nation in the world by tonnage and is increasing its shipbuilding capacity and capability for all naval classes."
So PRC already outnumbers the USA in terms of ships. In a few more years, PRC will outnumber the USA in terms of tons. China is definitely going for "build more" and not necessarily "build better". There hasn't been a modern naval battle since like WW2 between major powers, so no one really knows if this is a good or bad strategy. But its certainly a strategy.
USA of course has our 10x Nimitz supercarriers + 1x Ford supercarrier (going through our carrier-upgrade cycle currently, +F35 project I'm sure you're aware of). USA has a number of "Amphibious assault ship", which are sorta-kinda like carriers (except for the Marines instead of Navy), but do contain VTOL aircraft like F35-B variant.
I don't think anyone really knows for sure how the battle will play out. The US ships (ie: the "Supercarriers") are incredible war machines, and even the smaller Amphibious assault ships would probably go toe-to-toe with even the Chinese carriers. But how does that play out with China's missiles and airforce?
How many rabid PRC-haters do you think Taiwan harbours? Well, it's more than one...
von Braun worked for the Nazis to make missiles during WW2. Then was picked up by USA / NASA to make missiles / ICBMs / Space Program.
It turns out that a surprising number of engineers are rather flexible when it comes to ethics. They just want to work on cool things.
...after a couple of years of unrest and social instability. Which as simonh pointed out (https://news.ycombinator.com/item?id=28959591) is exactly what the CCP does not want.
Knowing that the PRC tends to take the long view, I don't think they'd take the chance regarding Taiwan.
If China wins Taiwan vs the USA, there wouldn't be any unrest or instability. The Chinese people would be proud of their accomplishment, and the rise of China as a military power.
Like... what "instability" occurred from Russia's annexation of Crimea? None what so ever.
They do, and that's the problem. The shot clock is ticking down quickly: fertility rate in PRC just hit 1.3 births per woman. Total population is projected to halve by the end of the century. The current cohort of military-age males is shrinking, and there won't be another one to replace it. The PLA will never be stronger than it is today.
If Xi wants to take Taiwan, he has to do it now, because he's not going to get another chance. PRC manufacturing output is cratering due to power shortages and the global shipping and semiconductor convulsion, so the economy is going to get clobbered anyway. They're already doing recon flights through the ADIZ and running freedom of navigation naval patrols. If there's going to be a Taiwan annexation, it'll happen in the next 12 months.
The US Navy is hapless and the Taiwanese army is supine. Evergrande is collapsing and there's already a wide array of tariffs and sanctions against PRC leadership. They accidentally caused a global pandemic and shrugged off any political ramifications. They've already taken half a dozen hits this year and survived. Why not take another?
The Taiwanese military has had decades to prepare for a Chinese invasion; it would be surprising if they didn't have a similar setup in place.
[0] https://en.m.wikipedia.org/wiki/Stay-behind
[1] https://en.m.wikipedia.org/wiki/Operation_Gladio
The CCP decided many decades ago that economic growth was absolutely necessary for them to stay in power because it is required in order to maintain social stability. An economy with hundreds of millions of unemployed peasants in the countryside, and feeble industries was not sustainable. This is why they purged a bunch of ideological marxists and embraced communism with Chinese characteristics, AKA Capitalism.
For a long time the CCP maintained that 7%+ economic growth was necessary in order to absorb the unused labour pool and maintain social stability.
Needless to say a war with Taiwan and the US, plus blockades by the rest of the world would be a somewhat risky situation from a social stability point of view. It would be a complete throw of the dice, and for all Chinese people love gambling with money, that's not the CCP way with politics.
We know the USA would respond militarily. But how many other countries would be pulled into the conflict? Germany still trades with Russia despite the Crimea annexation, right? Clearly USA / UK / Australia are making plans but is there anyone else who'd join in the fight?
Like, realistically, what would happen under a Taiwan scenario? Do you think McDonalds and Apple will just leave China? Will Tencent games be banned in the USA? Will USA really make League of Legends and Genshin Impact illegal during the war?
Or, would the world continue as usual? Would US Video Gamers continue playing their games, McDonalds continue China operations / Apple continue making their iPhones while the whole military stuff happens? Frankly, I see this as the more realistic scenario.
We can already see how video gamers reacted to Hong Kong: with a bunch of companies capitulating to Chinese demands.
I'm in no way justifying what the CCP did to Hong Kong, but frankly it was inevitable. That is not the case with Taiwan, which has a population and democratically legitimate government completely committed to independence.
An invasion of Taiwan would be much more similar historically speaking to the German invasion of Poland, or the Korean war. It's utterly and completely unacceptable, and would be legitimate grounds for total war. Should China succeed in conquering the island it would not be possible to counter-invade, but an indefinite blockade of transport into and out of China and total interdiction of it's air and sea traffic both civilian and military would be entirely proportionate and the minimum I would expect. See the blockade of North Korea for an example. We'd buy our phones and video games elsewhere.
And China outnumbers the US warships ship-for-ship currently, and China is outbuilding more warships ton-for-ton than even the USA.
China also has a larger (though inferior) air force, as well as a huge number of (untested) missiles. But the USA won't have many bases to stage our Air Force from (Taiwan is so close to China, and so far away from Japan / South Korea). So really, its China's Airforce + China's Missile-force + China's Navy vs the US Navy + US Marines + maaaayyyyybe a very stretched thin US Air Force.
If the USA wants to enforce a naval blockade, they'd have to win a war against China. That's a lot of advantages to China frankly.
I'm certain the US could beat China in a "fair fight" for Antarctica. But the issue with the Taiwan scenario is that the Chinese military is growing ever stronger, and its conceivable that China could beat the USA in that region. If China does happen to win militarily, then a US-sponsored blockade would be impossible.
There is simply no way China is going to contend with the US globally. Zero trade with the US, Europe, Japan, South Korea and Australia is a given. A sea blockade of trade with Africa, South and Central America, and Australia is fine. That leaves land and air trade only with Asia. How are they going to get oil? China's not going to establish military domination over the Middle East any time soon.
As a Brit I don't see us staying neutral, or Australia. Hence the nuclear subs Australia is buying. In a conflict like this, they would use them.
There may be a short term benefit but if they get cut off from ASML (Europe), Applied Materials (US), etc. it's going to be tough to keep their machines in service and stay leading edge. And those companies have very deep international supply chains too and may also get sanctioned.
You sound like you have a dog in this fight.
TSMC have a lot of fabs in Taiwan. that doesn't mean TSMC as a business can't build fab elsewhere.
TSMC is already building a fab in AZ USA and reach an agreement with Sony to build a fab in Japan. TSMC also have fabs in China and India govrt is looking to Taiwan for a fab in India.
This is not a recent development. TSMC has been mass producing 5nm chips for slightly over a year now. The first 5nm chip on the market was the Apple A14 (iPhone 12 line) released almost exactly one year ago, and soon thereafter (November 10th, 2020) the first M1 Macs.
Samsung has had their ""5nm"" chips out since early 2021 with the Exynos 2100 and Snapdragon 888. It's in the ballpark of TSMC's process but not as dense.
Intel hasn't even announced their first mass market 7nm (or ""Intel 4"", sigh) products yet. ""Intel 4"" is inferior to Samsung's and TSMC's 5nm processes on paper, and it'll likely not even be next year that we're going to see products with them.
I don't think Intel's foundries will be competing on the leading edge of the market. They'll have be trying to have value focused production with their 14nm and 10nm processes, I would guess. There should be plenty of demand there still, but it'll be a good while - if ever - before Intel is leading again.
According to transistor density information this is false (https://en.wikipedia.org/wiki/5_nm_process)
TSMC 5nm density: 173 MTr/mm^2
Samsung 5nm density: 127 MTr/mm^2
Intel 4 (estimated): ~200 MTr/mm^2
Got a source? I'm willing to change my mind, but the data doesn't agree with your assertion from my point of view.
Samsung's process has always lagged TSMC and INTC. It's part of why the 3080 is so hot/power hungry.
What do they win? More maket share, more revenue, more profits, better margins, competitors goes out of business?
I know very little about this industry, but is the market big enough for all players to 'win'?
https://www.reuters.com/business/intel-details-mixed-source-...
> Intel said Thursday its "Alchemist" graphics chips will be made by TSMC using the latter's newly named "N6" chipmaking technology, an upgraded version of its "N7" technology.
I mean, I guess they're already "winning". Intel's own chip-making group has begun to use TSMC. Hopefully Intel can figure something out about what its like to be a customer, so maybe they can be a better supplier?
But yes: customers, market share, and knowledge. By having more customers, TSMC can perform more experiments on more leading edge nodes, like 3nm or smaller.
Who wants one company to completely integrate its entire supply chain, producing everything from start to finish?
Imagine the cost if all companies did this. Imagine how it would suck for startup companies to get a foothold in this world.
1. Intel fabless: open up x86 like ARM. work on Xscale (ARM) and RISC-V (SiFive).
2. Intel foundry: a pure play foundry with no tie to its fabless side to win customers trust.
the foundry ecosystem is going to hurt Intel. the current trend is everyone design its own chip and outsource to foundry for production.
edit: look like the deal to buy SiFive fall apart but the idea is Intel fabless can focus on chip design regardless the platform.
Do a Google search for the stock price. AMD announced the spinoff in October 2008.
What would make Intel more successful at operating a foundry than anyone else? Or is it just a way to make the best of assets they already own?
Intel is very successful with their 14nm, with a engineer CEO like Pat. I think Intel can catch up on node process.
Much of this is semi-custom. You plug together a few pre-defined CPU cores, a GPU, a video decoder, some software defined radios, etc. and you have a phone.
Having more fabs in more locations is a good thing, for robustness.