Hacker News new | past | comments | ask | show | jobs | submit login
TSMC to Build Advanced Semiconductor Factory in Arizona (wsj.com)
300 points by JumpCrisscross 23 days ago | hide | past | web | favorite | 113 comments

We've moved most comments moved to https://news.ycombinator.com/item?id=23187698, the current thread on this, excepting the comments that were about this probably being a rumor (now clear that's not so), the comments confusing Taiwan for China, the political/nationalistic flamebait, and sundry unsubstantives.

TSMC official announcement. 5nm fab, 20000 wafers/month, 1600 jobs.


Can US $12 Billion cost be right? Going to be difficult to make their money back on what will by then be an old node, with 20,000 wafers per month.

I think this is a false rumour. The digital office of Taiwanese president is vehemently against letting fab tech out of the island. The woman running it is super sharp, can tell from knowing people who worked under her personally.

It is one of their biggest insurance against mainland aggression. Any conflict with Taiwan is suicidal for the mainland economically. Plus, add the fact that 3/4 of China's own light industry is within 1000km radius of Taiwan.

TSMC has already bough land and started construction for its 5nm fab, and laid first stone for 3nm fabs in Taiwan.

Stopping 3nm fab construction now is completely out of the question unless they are ready to throw out few billion dollars cash out of the window, at the time when they took biggest debt obligations in their history.

Only 2nm node is left on the roadmap, but there are big, big doubts of it materialising any time soon. We are talking about at least 8-10 years.

With all my knowledge of the industry, such deliberations never result in decisions like "Hey, lets set aside 20-30 billion USD to lay still for 10 years, and hope for everything happen according to plan."

Semiconductor fab companies are extremely, extremely risk averse, and snap judgements are completely uncharacteristic of the industry.

They don't do anything without having top tier economists, and finance people have a go at the task for years on end. Financial planning of fab construction takes years.

What is much more likely is them taking over a smaller fab in US, or building a limited scale project to make specialty chips for rad-hard, or US military hardware companies on US soil. This is what I myself been hearing for much longer than Trump been in office.

I think you are wrong, the geo-political tectonic plates has not only shifted but collided violently in Jan 2020.

US is brining supply-chain by hook or crook.

Taiwan needs Americas protection and no other country can be its security guarantor not even Japan. So US has leverage.

I would be surprised if TSMC does not open multiple facilities in US.

There's no way they are that stupid. TSMC alone has the power to shut down a huge number of electronics companies in both China and the US. Remember the HDD crisis? Anything happening to TSMC would dwarf that. Nope, Taiwan is not giving up their biggest playing card. That would amount to suicide.

I don't think anyone would be stupid enough to trust promises from a US administration right now, but I can potentially imagine a (far?) future scenario where Taiwan agrees to build fabs in the US in exchange for official "Taiwan is a separate country and definitely not part of China" recognition from the US. A lot of stars would have to align though, and even then it would be a pretty big risk for Taiwan and possibly still not worth it.

BTW, daily reminder that Taiwan is a separate country and definitely not part of China. :)

One might say that the Taiwanese government is the one that should be in power on the mainland as well :)

A law is in the pipeline (first reading in Taiwan) to remove that claim. I call it one-china law. Taiwan Finally recognise china as the Sovereignty of the mainland. Total cut off.

The china strangely you may think object. It is like British is passing a law finally to give up USA and USA object. But china do. And they know once cut off the USA can recognise Taiwan, unlike in the past it is a bit of conflict. What is Taiwan - oh the whole china. Now what is Taiwan? An island ruled by an independent state from at least 1949 (or legally from 1911 one should say). It even has a us-Taiwan act link even after 1979s.

It is good timing to try this move. But before that ...And hence all civilian infra is moving out from china by their allies and USA may have to think about getting key assets duplicate from Taiwan.

No one Outside china want war. But you have to prepare for it and avoid a total disruption of you own country and life. Chip is strategic. And the skill set not just the tools.

Higher post same idea as my previous post. Taiwan need a deal as USA cannot be trusted in international conflict. It is ok for Korea and Japan perhaps, but would USA gave up Taiwan when the first USA soldier died. We are talking about a country that has nuclear bomb and the means to bomb USA mainland.

Some deal needed.

I'm not even sure China wants war, but they need a way to relinquish claims on Taiwan without losing face, or relinquishing their claims to the south china sea - nor frankly do I think that invasion of Taiwan is likely, China has not had a military action in 50 years effectively, I dont see one as likely now.

I won't say that

If TSMC brings state of the art fabrication to the US, doesn't that mean they're subject to seizure by the Defense Production Act we've all learned so much about, TSMC's corporate fate becomes subject to CFIUS approval, and because of that, Taiwan loses a significant amount of political leverage to protect itself?

LOL, that's literally the case with any foreign investment in physical assets.

Tesla factory in Shanghai - if there is a war, all bets are off. CCP will raid the factory within the next 20 mins and employ military rule in the factory in about an hour's notice.

Physical assets are in some ways the ultimate thing to own. Even services that run on data centers which are physical things. Semiconductors are physical objects that your Dropbox runs on.

Which is exactly why this rumor makes no sense.

This rumor makes perfect sense. Taiwan is looking at US to shield it from Chinese influence. Moving manufacturing to US and building on top of strong American-Taiwanese collaboration (which already exists) only makes sense. The US has strong laws for IP protection, has proper constitutional values and doesn't steal technologies on a whim without due process. Taiwan has benefited greatly from the US and the US has benefited greatly from Taiwan.

I just wish American president was someone like the Taiwanese president. She is a strong, powerful and smart lady.

If the American government believed that TSMC facilities in America would meet America's demand for silicon, then wouldn't America lose that incentive to ensure Taiwan wasn't steamrolled by China? Supposing such a war did take place, TSMC might be unable to continue operating in Taiwan but still able to operate in some capacity in America. Would TSMC scrap the fabs in America just to spite the American government for the betrayal of Taiwan? That doesn't seem likely to me.

I don't doubt that the Taiwanese president is smart, which is why I doubt this story.

> wouldn't America lose that incentive to ensure Taiwan wasn't steamrolled by China

No, the opposite. Real trade enables balanced long-term interdependency, growth and innovation, unlike "free" trade focused on extraction and asymmetrical flows. The most valuable assets brought to the table by Taiwan and the USA are culture and these are complementary in the long term.

Because the US is so famously focused on the long-term and would totally protect a country from getting wiped out because of "shared culture."

Taiwan needs all the leverage it can get, and giving the US the gift of no longer needing it would be the dumbest thing TSMC could do in the long and short term.

Not shared culture, complementary culture.

A real partnership can create more leverage, not less, instead of holding one card until it is overtaken by events.

US laws generally only hold for US citizens and US companies.

The US had no issues using the NSA to spy on German IP and even bug the cellphone of Angela Merkel. And well, given that no consequences ever materialized, it won't seem like they'd need to stop at any time.

So given that, thinking that the US has "strong laws" is a tad naive. You know, since literally the opposite already happened.

There must be something lacking in this viewpoint, that US law is a liability for non-Americans, when you consider how entities around the world issue financial securities that are subject to US law when they could just not do that. I'm thinking, for example, of reading about Argentina issuing bonds governed by US law, in US currency.

I implore you to read the US Supreme Court opinions in its full glory: https://www.supremecourt.gov/opinions/slipopinion/19

Moving manufacturing to the US takes away leverage. It makes no sense.


Please follow the site guidelines in the future.


Last I checked, the DPA doesn't allow seizure, just forced manufacturing for a price. I'm not sure about the specifics, but it sounds like the theory is that there's at least a little profit in the price used under the DPA.

No law could permit uncompensated seizure as it would violate the 4th Amendment. Forced labor would violate the 13th Amendment. A better to way to look at it is that the DPA sets the rules for commercial activity--if you want to undertake commercial activity, you have to do so subject to the regulating law, which in the case of the DPA includes the possibility of building and selling whatever the government requests. You could always choose to just walk away, of course.

The takings clause is quite interesting and I imagine that between the expansive national security powers granted to the President and Congress's assent, the Court and the Executive might find a way to justify such a taking perhaps by seizing the foreign ownership of the corporation.

Of course the takings clause could also be satisfied by compensating TSMC, which perhaps due to sanctions or congressional action could be tied up indefinitely.

> No law could permit uncompensated seizure as it would violate the 4th Amendment.


> Forced labor would violate the 13th Amendment.

Not if the laborers were convicts, but, yes, generally.

US is making a lot of noise about it, but the current administration lacks the competence to actually do it. The US generally doesn't do this kind of state-directed enterprise dirigisme, outside of Boeing.

Remember Foxconn Wisconsin? https://www.theverge.com/2020/4/12/21217060/foxconn-wisconsi...

Falling behind in state of the art of fabs is seen by the DoD as the same kind of strategic issue that leads to them propping up Boeing.

> the same kind of strategic issue that leads to them propping up Boeing

I don't quite understand what you mean by this, what is the context of US support for Boeing?

Here's the backstory on Boeing and the military.

US air doctrine is that they will always buy air dominance, so Boeing will be funded for military programs. However, the US govt. historically has had the influence to tell military aircraft mfgs. what to prioritize, and what lines to drop.

An example is Northrop. They were told to drop the popular F-5 and F-20, and focus on the B-1 program. They were also told to merge with a competitor, but declined.

It's weird when you first study it, but the US Air Force wants zero risk missions into enemy skies.

So that's why they're ok with a handful of $200 million fighters and $2 billion bombers, if they're stealthy enough not to be seen.

(Stalin on the other hand is often attributed with saying, "Quantity has a quality of its own." I would argue a balance is needed between a perfect fighter, in the case of the F-22 with under 200 airframes mfg., and having 10,000 slightly less perfect ones, like the F-16.)

W: F-22 https://en.wikipedia.org/wiki/Lockheed_Martin_F-22_Raptor

If the public cannot stomach allied casualties, losing too many pilots might end a war even if the material cost of the planes was low. Quantity may have a quality of it's own, but maintaining public morale seems just as important.

Oddly, at the moment there are people on American streets demanding casualties in order to get the malls open again.

(Coronavirus death toll now larger than Vietnam)

That is a reason why you move so that when the time come (which is not chosen you we are not talking about us invaded china here but china just blockage of Taiwan say), you have option. If Taiwan has a strategic assets at least you do not risk a war ...

Guess that is why Taiwan also would not want the asset out.

There have to be a political deal. If USA and Taiwan is competent.

You'd think, if the people in charge had any sense, but it's really about lobbying, and Boing is much better at that than Intel.

What's your assessment of ERI (Electronics Resurgence Initiative), especially their sponsorship of R&D for open-source EDA toolchains and open IP blocks for chiplets, e.g. OCP ODSA? There are many universities and tech companies participating and a budget north of $1B.

TSMC US-manufactured FPGAs could be a good compute platform for open-source hardware designs that can be rapidly iterated.

ERI: https://eri-summit.darpa.mil/what-is-eri

OpenCompute ODSA: https://146a55aca6f00848c565-a7635525d40ac1c70300198708936b4...

Toolchains and IP are not the issue . We're talking about actual manufacturing.

There is a secure foundry system for DoD use: https://semiengineering.com/a-crisis-in-dods-trusted-foundry...

GlobalFoundries only goes to 14nm. Many applications benefit from 7nm and below for size/weight/power.

> Toolchains and IP are not the issue . We're talking about actual manufacturing.

Of course. Let's say you're TSMC with limited capacity on your bleeding edge node, operating a price auction that drives up revenue against your fixed investment. Would you create another fab to compete with yourself, adding extra capacity, reducing price competition, and revenue?

Now let's say you can create a new fab that does not compete directly with the existing product line, e.g. because most of the capacity of the new fab will be occupied with new customers who could not previously make use of the Taiwan fab. Where would these customers come from? They can't only be the US customers who use the existing "secure foundry", as that would mean supply/competition problems for GlobalFoundries.

Any large supply increase in actual US fab manufacturing is going to require a large demand increase in actual US fabless semiconductor companies. Open toolchains and IP are a necessary, but not sufficient, step on the path to expanding the commercial US semiconductor industry, which would also benefit secure use cases.

This is only tangentially related to the current administration. Onshoring manufacturing has been a supposed policy goal of the President since 2016, and that hasn’t amounted to squat.

The tectonic shift is that it has suddenly become a priority for the permanent, unelected (i.e. real) US government for a different set of reasons.

Sounds like Foxconn is using the Wisconsin subsidies as part of an offshore money sink. Didn't Wisconsin evict landowners to build these empty buildings?

Beyond turning policy knobs it's not really in the administration's hands though. There's a lot middle and higher level managers saying "gee I really looked like an idiot when I couldn't deliver on my objectives because I was relying on supply in China" who are looking to avoid a repeat. Some stuff is gonna shift back regardless of what the administration does.

Definitely true in general, but it would surprise me if the Trump administration wasn’t involved in making this particular project happen, just on the strategic importance alone.

FYI the OP of this thread is "Mr Paul Ni" based in Guangdong who has repeatedly posted on Weibo and other Mainland media from multiple accounts statements such as "Han man rising to the top like the Sun, taking the impure foreign man's wealth and women", "fentanyl opioid war retaliation against the Imperialist West", "Chinese race rise to rightful power", "Hong Kong cockroaches should be crushed", "unleashing the virus on clueless West" and other stuff like that.

Just to provide some context into possible motivation behind some of the statements they are making.

> US is brining supply-chain by hook or crook.

What if the current administration don't last till 2021?

Or more likely, what if the word of the current administration don't last till 2021?

With 40 million in unemployment queue and nation feeling vulnerable and paranoid.. Americans are not fucking around no matter what ideological stripe they come in.

Some talk softer than others but the direction is baked in.

> Americans are not fucking around

As a government with a consistent policy? I'll wait until 2021.

> I think you are wrong, the geo-political tectonic plates has not only shifted but collided violently in Jan 2020.

I don't think so. In 10 years, there will be no Trump in the office, and I doubt than even US based companies are ready to stake so much money on US government money continuing to come in 10 years term.

> Taiwan needs Americas protection and no other country

Their biggest defence against China is their mutual economic annihilation, along with most of world's light industry.

Remember, it was USA who made them loose their UN seat, and sabotaged their nuclear weapons program to keep them dependent, and let China into the WTO. Since at least Jiang Jingguo, Taiwan has been rather ambivalent with USA, and are wary of being set up

Trump is not the first to raise the point on this exact matter, in fact that was going back and forth at least since Bush Jr as I know.

Lastly, the biggest, and most obvious argument about your sentiment is that if American political establishment (with Trump, or without him) were truly serious about committing to confrontation with China, they should've started with barring US companies making business with Chinese multinationals, or at least obvious state companies.

Without seeing that first, you can be assured that US will not run into collision with China, or the politically connected big money in US would be first to run.

> barring US companies making business with Chinese multinationals, or at least obvious state companies.

CFIUS actions have been reported in the tech news for the last 12 months, in multiple sectors.

The movement to bring supply chains back to the US, or at the very least out of China, is far bigger than Trump.

Not just US, the entire world is waking up to CCP's iron grip.

Europe, Korea, Japan, India and Australia are planning to recede from China's dependence on manufacturing.

This is an amazing time to be alive, I encourage people to look into manufacturing disciplines. As Elon put it (agree or disagree with him, but he is right about this) - "Someone's gotta build stuff. If you're not doing it, someone else is." Physical things don't just pop out of thin air. Every single thing except your body and natural substances are manmade. Someone decided how its gonna be built and how it will come into existence.

Completely agree.

I believe that we need to reframe the debate around manufacturing from one about jobs to one about national security.

This mentally frees us to fully invest in manufacturing automation, which will neutralize any labor cost advantages China and other countries may have.

Except that Elon is moving deeper and deeper into China...

for the first time in a long time both party agree on this. i believed this is the U.S. failure for believing in China. U.S. bring China into WTO, open up trade and hollow out Midwest. what American get in return is Emperor Xi. CCP is not going to change and Xi already have more powers consolidated into him than any other previous CCP leader even Mao.

American's dream for a open and democratic China is long dead.

> American's dream for a open and democratic China is long dead.

There already is an open and democratic China across the Taiwan straight. And the USA has a looong history of rather ambivalent relationship with it, starting from the Chinese civil war (to the surprise of many, US has sided with Mao for quite a number of times)

For long, US policy was to purposefully keep the Republic of China weak, vulnerable, and dependent.

Were the US even 1% genuine in their commitment to seeing democratic China, they would not be consistently setting up Taiwan to be undermined politically.

Because during the cold war Taiwan wasn't nearly big enough to counterbalance the USSR. the PRC was, and when the US got bloodied in Vietnam and needed to pull out they wanted to make sure the USSR couldn't step into the hole they left behind and start pushing across the rest of Southeast Asia. Cue the PRC. Right after the US left, the PRC had their own war with Vietnam because they thought Vietnam was working with the USSR to conquer the rest of Southeast Asia and surround them, and the PRC and the US ended up working to isolate Vietnam over its invasion of Cambodia and put pressure on the USSR. With China holding down the fort in East Asia, the US could pull back its military spending from 10% GDP to 5% GDP in the 70s while the USSR got bankrupted by a new military threat.

Doesn't look like US politicians today are willing to throw Eastern Europe under the bus to flip Russia against the PRC like they threw Taiwan under the bus though.

Even without Russia as a threat, China is incomparable to the USSR. Its productive potential dwarfs that of the US manifold.

I would be interested in seeing the supporting arguments and data for your latter claim. At a glance US gross manufacturing output lags China by about 7%, but per capita US manufacturing output is ~300% that of China. Between the Monroe Doctrine and NATO states, however, the "US manifold"'s productive potential is where it appears accurate assertions of relative dwarfism can be made. Without intimate familiarity of geopolitics, I would estimate all save 3-5 of the top 20 nations by manufacturing output fall clearly in the US sphere of influence.

This is because when comparing manufacturing outputs, you should look at it in PPP terms, not in absolute dollar terms, due to devaluing and other factors. When you want to calculate how many tanks, missiles, bombs and airplanes, PPP output is what matters. Or if you want to take a page out of the Soviet playbook, Gross Material Product.

The US manufacturing sector is 11.6% of GDP, so 2.33 trillion dollars PPP. In China, it accounts for 40% of the GDP, so more or less 10 trillion dollars of manufacturing output PPP (!). This is a manifold advantage.

The reason why this is to be done is because if you wanted to build say, a tank in the US vs China or Russia, you would need many more dollars for the exact same tank. So measuring manufacturing output must be done PPP.

During the cold war both the US and the USSR were self sufficient in terms of wheat and oil. China is one naval battle in the Indian ocean away from having a taste of Imperial Japan's oil shortages in WW2. If the USSR practiced state capitalism then we might still be locked in the twilight struggle with them today.

No, and no. There is no reason to believe that the USSR would have been able to develop beyond their level of development, that many capitalist nations aren't able to surpass, given the geopolitical and political situation. It seems to me that the fall of the USSR was due to political issues, and an unsustainable political system, more than anything.

China also doesn't require any naval transport from oil shortage. Indeed, it very conveniently has a major oil exporter to it's north. So unless you would wage war on the Russian and Chinese mainland, you have no hope of completely starving China of oil. You might reduce the amount of oil they have, but they have enough reserves to last until a tripling of Sino-Russian pipelines.

We're discussing a recent change in tone, behavior and overall policy. What you're describing is the product of previous administrations working hard to get on (the People's Republic of) China's good side, cozying up to China and trying to integrate it into the Democratic community.

This is now precisely what is changing.

The article says it would be a 5nm fab and the core reason behind it would be to sell to America which is getting more and more worried about dependence on Asian supply chains. As a US government backed initiative, it seems likely that the US would help financially in some way.

5nm fab will require an investment of $15B+. Moreover, having it in the middle of nowhere in Arizona, without a semiconductor supply chain there will also add massive cost of building a support cluster by themselves.

Intel itself has to haul all consumables for their US fabs in Arizona from Asia.

So, add few more billions for the uncertainty on the supply chain side.

Plus, having a 2-3 year project deadline looks to me as feasible, but still very tight. TSMC's 5nm fab in Taiwan has effectively took 4 years given that they still in the final stages of setting up its internals. In US, they will have to do all of that without the massive support of Taiwanese industry, and no local expertise.

Also, by completing a 5nm project 3-4 years from now, who they will be selling the capacity to, when biggest cash cows will move to 3nm, and there will be nobody to take their place for existing 5nm capacity? The cost of tapeout on 14nm is already so extreme that leaves only around 30 companies globally in the game, it will be much worse for 5nm.

Now, the biggest showstopper — TSMC is a public company with a lot of pension fund, and sovereign fund type investors. How they are going to make them swallow that? Them saying that out of the blue will be a bombshell.

Lastly, Taiwanese government has long subjected its semiconductor industry to zealous supervision, and policy guidance. Such move will have to pass by the regulators, which Taiwan has many.

So, without US government putting at least $20B cash, and big guarantees on the table, the project seems to be in the same gimmick rank as Foxconn's Wisconsin plant.

Commercially, it makes not much sense to me unless something really surreal is happening on the political, or financial side of the matter.

Intel has 4 fabs in Arizona, and one in New Mexico =. They have more fabs in Oregon (6) but it's not like TSMC will be building from scratch out in Arizona.


Microchip has fabs in AZ. Before LSI sold off their Oregon plant, ONN’s most advanced fab was in Phoenix.

NXP has Freescale’s old fabs as well. I think generally speaking, AZ is well represented in terms of fab expertise.

Intel has the majority of its fabs in Arizona. What are you talking about?

We'll have to wait to see what happens, but $20B cash is absolutely nothing to the US government if we are talking about securing vital supply chains. And US-Chinese relations aren't going so hot so it feels like ripple effects on the chip industry are inevitable.

Well, having a fab in the middle of nowhere alleviates concerns related to vibration, noise, and water supply, no?

And pollution. Semiconductor manufacturing is dirty, dirty business.

Not sure about having plenty of water in Arizona, but they've got plenty of silicon. :)

5nm will be a more budget-friendly option for TSMC customers in 2023. This is a conservative proposal, not something bleeding edge.

> Now, the biggest showstopper

Barely any stock market reaction this morning on the Taiwan exchange open (ticker: 2330:TT).

>Taiwanese president is vehemently against letting fab tech out of the island

she is pan-green. CCP does not like pan-green anything at all. they especially doesn't like her and favor the pan-blue candidate during the president race. CCP even went as far as bribing Taiwanese in China to go back and vote for pan-blue candidate. this story just break recently.

Taiwan needs U.S. security guarantees. she will let TSMC out of island if it mean it doesn't piss off Uncle Sam and US have been applying a lot of pressures on TSMC for making chip for China.

I'd have to agree.

The US is more "business-friendly" than Europe, but nothing compared to China or Taiwan for people with the right connections.

Occasionally you'll see unscripted interviews between a US reporter and a visiting Chinese business leader.

The US reporter always asks if the visitor will be opening a local factory (because the US is so awesome!), and the visitor starts looking around wildly and demurs, since they're just here to park money.

You might have missed the part about the stranglehold that the U.S. administration is exerting on Taiwan, and especially TSMC, by kicking around the idea of banning U.S. companies to selling components to a supplier of Huawei. It is an insane measure that the hawks have been thinking about for a while now.

A good summary of what's going on: https://asia.nikkei.com/Opinion/Taiwan-s-semiconductor-indus...

So what happens when we get to 2nm? Do we just stop getting faster computers?

CMOS process scaling is already on its last legs. There are various options: more efficient code (ROFL), stacking of various sorts, new material/techs, co-processing (GPU, FPGA, quantum for certain classes of problems), etc.

But there's certainly a school of thought that CMOS process scaling was the big engine of tech over the past few decades and there may be nothing that comes close to replacing it to anything like the same degree.

There are challenges but work is on the way, transitioning to gate-all around FETS (GAA Fets) https://semiengineering.com/making-chips-at-3nm-and-beyond/

We could see a 4x boost in computer speed by scrapping all scripting languages and electron. Also the future seems to be more cores. A 64 core laptop may not be too far away.

Can you please stop using allcaps for emphasis? You've been doing it a lot, and this is in the site guidelines:

Please don't use uppercase for emphasis. If you want to emphasize a word or phrase, put asterisks around it and it will get italicized.



California already has the major semiconductor equipment manufacturers like Applied Materials, Lam Research, Nanometrics which might make money selling equipment to them.

Edit: I used to do R&D on SEM equipment when I was a mechanical engineer back in the 90s. I've reinvented my career many times over. I see the company I worked for still has one of those in major production. Maybe it will go into this fab. Its awesome seeing something you worked on decades ago still in production!

Surely just poaching from Intel's existing Chandler fab. No one fabs chips in the the actual silicon valley anymore.

There's still a lot of chip design and electronic design automation going on in Silicon Valley. The actual manufacturing has moved to lower cost areas.

That's not TSMC's gig though, they make the chips.

We should rename SV to Social Media Valley :)

I've been a long-time fan of "Silly Valley"

+1 for Silly Con Valley.

It's a quite accurate name for its phony startup culture (no offense to real startups).

Superhighway Valley. A good name to show its historical root on the Valley's current iteration of life.

Well I mean the main raw material rely on is basically sand, so can't fault them for choosing Arizona.

Huh, I just realized that I have no idea about two fundamental questions re: semiconductor fabrication. 1) Do fabs make their own silicon wafers? 2) Are silicon crystals for wafers made from sand relatively directly, or is it broken down and then reformed? I can't think of it and my google-fu is failing me, but I know there's a simple compound that reacts with water to create (relatively) pure silicon dioxide, so I'm wondering if that -- or something like it -- is used.

Answering your second question, Silicon wafers are made from from an ingot of monocrystaline silicon[1], as the fault line formed from the boundary between two crystals could interfere with circuit operation. Said ingots are often made with the Czochralski method, wherein a seed crystal is dipped into a crucible of molten silicon and slowly pulled upward. The silicon required for this process is nine-9s purity, so there'd be a number of steps beforehand to convert sand or even relatively pure silicon dioxide to such highly-pure silicon.

Given the carefully-controlled conditions required, it's an expensive process. Whether fabs in-house their wafer production or farm it out to a supporting firm, I don't know. I could see it going both ways. I'd not be surprised if Intel or TSMC had their own in-house wafer or ingot production. But, there's more fabs than just Intel and TSMC, so I'd guess the smaller operations would get their wafers from a company that specializes in creating them.

1. https://en.wikipedia.org/wiki/Monocrystalline_silicon 2. https://en.wikipedia.org/wiki/Czochralski_method

> Do fabs make their own silicon wafers?

(Did you mean ) The answer to that oscillates from "yes" to "no" all the time. Last time I saw, only the highest tech ones did that by themselves.

> Are silicon crystals for wafers made from sand relatively directly, or is it broken down and then reformed?

I don't completely understand that question. But silicon crystals are composed of pure silicon, while sand is composed of silicon oxide. On the cases where silicon is refined from sand (there are many kinds of silicon oxide), people reduce it with carbon in a furnace, like they do with iron. After refined it goes into further purification and in some cases a different process that solidifies it into a single crystal.

So there is a lot of processing.


The sand isn't particularly important. What matters is the huge amount of etchants and dopants used on it. It's more important to have somewhere you can dump a load of used carbon tetrachloride. Quite a lot of the US's old fabs are now Superfund sites.

Silicon refinery and wafer production are highly specialized industrial divisions. Most fabs don't produce their own wafers, although some of the biggest players have their own subsidiaries.

* TSMC / WaferTech subsidiary https://en.wikipedia.org/wiki/TSMC#WaferTech_subsidiary

WaferTech, a subsidiary of TSMC, is a pure-play semiconductor foundry located in Camas, Washington, USA. It is the second largest pure-play foundry in the United States. The facility employs 1100 workers.[citation needed] The largest is GlobalFoundries Fab 8 in Malta, NY, which employes over 3,000 workers with over 278,709 m2 (3,000,000 sq ft) under rooftop.

WaferTech was established in June 1996 as a joint venture with TSMC, Altera, Analog Devices, and ISSI as key partners. The four companies along with minor individual investors invested US$1.2 billion into this venture, which was at the time the single largest startup investment in the state of Washington. The company started production in July 1998 in its 200 mm (8 in) semiconductor fabrication plant. Its first product was a 0.35 micrometer part for Altera.

TSMC bought out the joint venture partners in 2000 and acquired full control, and currently operates it as a fully owned subsidiary.


Even before you start making wafer, you need to have pure silicon, which requires a massive industrial refinery operation. See:

* Silicon Production https://en.wikipedia.org/wiki/Silicon#Production

Silicon of 96–99% purity is made by reducing quartzite or sand with highly pure coke. However, even greater purity is needed for semiconductor applications, and this is produced from the reduction of tetrachlorosilane (silicon tetrachloride) or trichlorosilane. The former is made by chlorinating scrap silicon and the latter is a byproduct of silicone production. These compounds are volatile and hence can be purified by repeated fractional distillation, followed by reduction to elemental silicon with very pure zinc metal as the reducing agent. The spongy pieces of silicon thus produced are melted and then grown to form cylindrical single crystals, before being purified by zone refining. Other routes use the thermal decomposition of silane or tetraiodosilane (SiI4). Another process used is the reduction of sodium hexafluorosilicate, a common waste product of the phosphate fertilizer industry, by metallic sodium: this is highly exothermic and hence requires no outside fuel source. Hyperfine silicon is made at a higher purity than almost every other material: transistor production requires impurity levels in silicon crystals less than 1 part per 1010, and in special cases impurity levels below 1 part per 1012 are needed and attained.


* Czochralski method https://en.wikipedia.org/wiki/Czochralski_method

High-purity, semiconductor-grade silicon (only a few parts per million of impurities) is melted in a crucible at 1,425 °C (2,597 °F; 1,698 K), usually made of quartz. Dopant impurity atoms such as boron or phosphorus can be added to the molten silicon in precise amounts to dope the silicon, thus changing it into p-type or n-type silicon, with different electronic properties. A precisely oriented rod-mounted seed crystal is dipped into the molten silicon. The seed crystal's rod is slowly pulled upwards and rotated simultaneously. By precisely controlling the temperature gradients, rate of pulling and speed of rotation, it is possible to extract a large, single-crystal, cylindrical ingot from the melt. Occurrence of unwanted instabilities in the melt can be avoided by investigating and visualizing the temperature and velocity fields during the crystal growth process. This process is normally performed in an inert atmosphere, such as argon, in an inert chamber, such as quartz.


Silicon wafers are not made of Sande, they are made from Quartz (fact check me). Arizona has 113 quartz mines ..


It's the same thing. A crystal of silicon dioxide is a mineral we call quartz. Crush it up into a powder and we call it silica sand.

I am a little skeptical, but also more convinced by the lack of Trump-hype behind the announcement. If it did accompany it, it would have all the hallmarks of a Foxconn-style announcement that leads to nothing.

Is anyone going to cry forced tech transfer?

TSMC develops the tech often times in partnerships. AMD definitely paid for a lot of the tech and manufacturing research that went into the current state.

Altera too. The regularity of FPGA designs makes them perfect for optimizing process yield.

In which direction?

Isn't this risky regarding industrial espionage? Also, what if the USA sanctioned China? Wouldn't it be bad if TSMC had a factory in the USA?

I question the amount of secret stealing that could effectively occur in the factory itself. Tool parameters can be loaded over encrypted networks directly from secure systems. Reticles at 7nm and lower nodes are hopelessly impossible to reverse-engineer. Tool operator interfaces can be locked down to provide only the most essential information. What exactly are you going to pilfer from these factories? Everyone knows you buy EUV from ASML. Not a whole lot of insight to be gleaned by just observing operations. Sure, you could decide that IP exfiltration is not going to work and resort to sabotage. Causing defects intentionally is not very difficult, but I doubt you would get away with it for very long. The amount of measuring, logging and tracking that occurs throughout is beyond incredible. Everything is a consideration in the hopes of squeezing out more yield. As a consequence, you have the best audit log in any business on earth.

I've worked in the industry and you can't even go inside the lobby of one of these buildings without having epoxy poured into all the USB ports of your various electronics. Security is best described as absolute in these facilities. Getting into the factory itself is yet another security check. You cannot bring any personal items into the clean room last I was told the rules.

TSMC is a Taiwanese company, not a Chinese one.

I know but what if TSMC had to apply possible US sanctions against China? First, TSMC would be a US company, which could give the USA an excuse to tell TSMC to apply sanctions against China. Also, the USA could take the US factory.

The US already has that ability to coerce foreign companies. TSMC buys chip making supplies and equipment from US companies, and there's been discussions around requiring foreign companies using US technology to obtain a license to not supply chips to Huawei or risk getting cut off from the US tech network themselves.

Are you under the impression that Taiwan wouldn’t welcome US sanctions on China with dancing in the streets? Taiwan was bitterly opposed to detente.

TSMC already has some US operations, so this wouldn't change the ability of the Trump Administration to put that kind of squeeze on them.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact