As a construction electrician who worked at TSMC in Phoenix, I can wholeheartedly agree with you that TSMC was dragging their feet in getting the Fabrication building completed. I had "adjustments" to my scope of work due to the designs changing multiple times, resulting in a lot of wasted time and material. It was incredibly frustrating trying to meet deadlines when scope was changed mid-stream on multiple occasions.
Yeah ... is it due to changes, necessary changes? Or have they just dog an ponied building something to get the gov't money? I just don't think they'll be sincere.
As a truck driver I pass by that TSMC facility north of Phoenix regularly. I didn't realize what that project is until just now. Yes, it's been under construction for MANY years. It is MASSIVE, impressive, and a mess. Thank you for triggering my 40 watt bulb to turn on this morning.
Why the delay? The factory has to be built eventually. In addition to the issue of the workers' union, the most important thing is to have President Trump cut the ribbon, rather than having it completed now and counting it as President Biden's achievement.
@@UlyssesMax Cutting a ribbon does NOT signify it is/was trump's achievement. It is a purely ceremonial act. If you're going to give credit to a president, it's Biden, as it was his 'CHIPS' act which funded $280 billion towards bringing semiconductor and other technology back to the U.S. It was part of his $1 trillion package to revamp aging U.S.infrastructure While trump's 2016 campaigned talked about a trillion infrastructure plan, he did ZERO in that area. Once he was president, Trump suggested privatizing highways, making them for-profit toll roads. Biden's softer style helped gain Bipartisan support in Congress to get quite a few infrastructure bills passed. But, there's no doubt trump will claim full credit despite no engagement in it.
I live in Phoenix and work in manufacturing. The thing that hurts our heads is that they are trying to make chips (a very water hungry industry) in a desert city. A desert city that just lost a large chunk of it's water rights after the last negotiations.
@@ace1776 What makes them dumb? And who is going to work in the factory once those "Dumb Desert People" move because they lost their water? We gonna bring in workers on rotation and then bus them to somewhere smart?
I think moving manufacturing to the US will be highly automated. It’s the direction of manufacturing in general and especially makes sense in developed nations.
Peter, the whole process of growing a silicon crystal starts with using a small silicon crystal seed which has the correct FCC (Face-Centered-Cubic) orientation required for semiconductor processing which should be propagated throughout the drawn silicon ingot. The silicon structure of the ingot is sampled early in in the process of drawing the ingot from the pool of liquid silicon and if it doesn't match the required FCC orientation, which I think is 011, then the process is restarted.
This mostly a supplier problem and NOT one that TSMC (or any other semiconductor fab). worries about - it's supply chain qualification/rejection criteria. TSMC only worries about the volume and qualitied supplied regardless of supplier. The failing supplier simply does not get paid for the rejected material. But another supplier will step up and fill the gap. This is the major challenge of being a TSMC supplier/vendor - they are demanding and competing is expensive.
@@Jeff-i8u Yep, there are only 2 or 3 companies at best that even grow the silicon ingots. And Zeihan covered a serious problem in another of his videos that one of the best (in the USA) sources of the silicon used for this had one of their mines flood. And production capacity was severely cut. Maybe they recovered by now. I have read other stories that the TSMC work culture at the Fab is not going over well with US workers. IMHO Intel is on the brink of collapse in terms of fab production, and the Biden CHIPs act is on the Trump chopping block. So this will take time to play out.
I think Peter is misunderstanding what’s really needed for most deep learning model training. The push for sub 10nm processes, or even 100nm, isn’t as important as he makes it sound. Compute power isn’t the bottleneck for AI workloads. The real issue is memory bandwidth and how efficiently vector cores can handle matrix multiplication (MATMUL) operations. If you’re building a superfast GPU for general purpose use, then yeah, smaller feature sizes matter because you’re trying to cram as many transistors as possible onto a silicon chip. But when it comes to AI workloads, like training transformer models or running inference, the priorities are different. You want to load as much of the model into RAM as possible and run massive parallel operations in a single cycle. For training, you also need to calculate and store gradients for backpropagation and gradient descent, but none of this actually requires the newest lithography. What Peter is missing is that AI tasks are super parallel and asynchronous. You don’t need the latest and greatest chip tech for that. You could use something like 100nm lithography to make chips loaded with tons of 16 bit floating point adders and multipliers, with DRAM built right in. It wouldn’t be cutting edge, but it would still handle AI workloads really well and be way more cost effective. The problem with Peter’s argument is that he’s assuming not having access to sub 10nm tech means a country like China is locked out of the AI game. Smaller feature sizes do help with efficiency and performance, but AI workloads don’t actually need the cutting edge to be competitive.
@@dcc70 I think you are very wrong. By the time of the Apollo project,. computers were solid state. No more Univacs. Even the worst of solid state computers made anything vacuum tube laughable for many reasons.
According to IEEE on 12/26/2024, one TSMC FAB in Arizona is producing 4nm transistors. The second FAB is scheduled to be operational by 2028 and produce 2nm transistors. Those 4nm transistors are what power NVIDIA's newest GPUs. Absolutely cutting edge.
The TSMC facilities are the most complex engineering created by mankind, perhaps apart from the Large Hadron Collider in Europe. A fully fledged manufacturing facility akin to the plants in Taiwan, comes with a pricetag of ~$100bn 😳
Put down a layer of e.g. copper, coat the chip with photo resist, Expose to EUV the photo-resist to print the outline of wires on the chip, etch off the areas not exposed, repeat (up to 90x times). Sometimes after printing a layer, you dope (heat & infuse the semiconductor layer with doping materials like germanium) to increase the semi-conducting property, then you remove the entire layer and start over.
Intel and the state of Ohio have given boat loads of money to all the local community colleges here in central Ohio to pump out as many semi-conductor grads as possible in the next 2 years. We’ll see if it works out.
it wont, we no longer do anything on merit, prob end up with a bunch of know nothings that just got passed anyway to make the numbers look good. Who cares if they even wanted to go in that field at all
They'd better also pump boatloads of money into building more houses, because the cost of housing is about to go through the stratosphere around here if they don't. The economic boost is great, but less so if nobody can afford to live here.
1:49 - That's making an ingot, that's sliced into wafers. The ingots and wafers are usually made in another facility. They are often a different company than the one making the chips. Think the companies that make tires and sheet metal, that the car company uses to make their vehicles.
Peter seems to be mixing up totally separate phases of semiconductor manufacturing - from crystal pulling and wafer manufacturing to actual processing of those wafers into chips. Siltronic is a company that specializes in the former, TSMC in the later.
TSMC isn’t developing the technology. They work on the “tail-end” of production with the actual companies that develop the technology. The real high tech is at those hundreds of other companies.
@@chillxxx241 id say the machine maker is the real hero, but what he is talking about is the process he didnt even bring of TSMC, he i just saying I find it amazing that some dude figured out how to grow chips on wafers
@ The core technology that ASML uses was built by the US government bank in the 70s, but none of the US companies wanted it. There are now 800 or more companies that help ASML make those machines at an affordable price. Neither ASML nor TSMC design the semiconductors.
not "someone". Hundreds of thousands (if not millions) of high-tech engineers, over many painful and expensive iterations across what must be 50-60 years. Many have failed and we never hear about them but even the failures were extremely important in the knowledge we eventually solidified into modern chip manufacturing.
I've just finished reading a book called "Chip War" about the battle for supremcy in chip production and it it explains in lay person terms these processes for producing and the machinery involved. All I can say is there are some bloody clever people in this world, it's insane what they have developed, your average consumer is oblivious to the level of genius they have in their hands every time they check their smartphone.
Why the delay? The factory has to be built eventually. In addition to the issue of the workers' union, the most important thing is to have President Trump cut the ribbon, rather than having it completed now and counting it as President Biden's achievement.
So, Kissinger basically decided to move shit to Redchina and now someone said it's time to move things back. Each time we the simple people have to live with the effects.
@@nazcamain Labor-intensive jobs are never coming back but most semiconductor manufacturing is 100% automated. I've spent a lot of time in Taiwan on their fab floors and MOST of the workers you see are vendor contractors keep their equipment working. Actually fab employees are less numerous on the fab floor. So NO high tech today is ever going to create mass quantities of jobs. The world died in the 1960s and 1970s. It's never coming back. Most jobs are technicians "feeding the automation" but that's still going to be labor costs (and thus # of hires) less than 1% of what it was in the labor-intensive manufacturing era of the 1960s. Most money for governments will NOT be employment taxes but other taxes only. There are no business models that hire 100K-500K employees anymore.
Actually it was moved to all of Asia except Red China. Taiwan and Singapore being the closest in terms of Chinese culture ever Communism and top-down control are still incompatible with complex STEM. If you can't fail, you can't innovate. Currently China is largely stuck on planar processes at 15 nm and above. That tried some of the 3D FinFET technologies before 15 nm but they do not have the EUV so they have to make due with the far more complicated larger wavelength older photolith. And these require incredible skills and math/computing intense multi-patterning photolith to make work. Tariffs and export restrictions a nice FU to China: tie both hands behind your back and wear this blindfold. LOL As it turns out 90% of the fabs created in China over the last 15-20 years were sham companies designed for embezzlement rather than semiconductor, and most have now failed. The original SMIC is still around but struggling to make sub-15 nm parts. China is way behind and largely locked out of anything resembling the processor in your iPhone. On the other hand, that market is relatively small - most semiconductor volume is NOT deep nanometer but making things for dish washers, cars, etc that are much larger. This is where the West screwed itself - we still need large geometry IC designs and products. Some companies are working on doing that and small geometry but China has dominated this low margin business. Other the other hand, things is China are going VERY VERY badly economically. Bad enough that a revolution could happen to over-through the CCP/ChiComs. But we will see.
@Jeff-i8u this was more a statement about the general and sudden destructive top down deindustrialization of the rust belt and Europe. I know better chips never came from Redchina. But let's be honest: the same disconnected functional elites that decided now that things are suddenly moving back (again at the cost of consumers, before at the cost of employees and competition) will or are already making any other industry move around globally or simply not being available to Joe Average. The confrontation course of their fellow WEF puppet In the Kremlin was known to them already since Georgia (2009), latest since Donbass and Crimea (2014). It's a top down deindustrialization and down scaling. Hidden behind some not so sudden geopolitics. It's a dirty game. Like it always was.
@Jeff-i8u will the elites sell us more expensive washing machines with super chips or phase us the cannon fodder population generally out during this centrally planned down scaling of markets and deindustrialized under the guise of green policies and instigated international conflicts? I mean: we were encouraged by them to completely allow an integrated global economy and forced to accept the alleged "peace dividend" and consequences of it.
@@jumbam9938 That depends on if trump turns against the US or turns against Russia. If trump protects american interests, he would do the same as nixon and end the war by ramping up ukraine's attacks on russia. If trump supports russia, he will take steps to undermine ukraine. We'll know in a month and a half.
Fr lmao. Figured this out about 4 years ago. Inside China Business with Kevin is far more insightful. Dude actually lives in China since 2012, and his 6-10 min videos are dense with recent news articles and him tying each of them together. Peter doesn’t even speak Chinese nor has been to China yet is somehow an expert, where the Same experts in the US about China have been insanely wrong (like Gordon Chang)
The move toward EUV lithography in TSMC’s U.S. facilities is a fascinating leap. These advancements are reshaping global semiconductor manufacturing, especially in improving efficiency and reducing environmental impact.
@@AdvantestInc let's hope the simple folks will still enjoy the innovation dividend from that! ... before Peter and his friends are crawling into their luxury version of doomsday bunkers while the rest of us and our devices will absorb a few EMPs 😂😆
Having worked directly on several mega construction projects including clean room R&D and chip making facilities I can tell you there is NEVER a "Final" design and you are ALWAYS updating as the project continues. Often I have seen ongoing work on one end of the building while the previous end is being torn out and reconfigured. This isn't so much feet dragging as it is Producer level changes to their delivery contracts. The construction progress is time consumptive and the years it takes to develop a project is the hamstring in producing current tech products. There is indeed a point where construction completeness outweighs further reconstruction for updates, but it is often very late stage. It is only at that point that the "Final" design is really just a decision to stop changes in favor of "good enough" to get products flowing.
Peter, you are wrong. EUV simplified earlier nodes, but for the latest nodes they use multiple patterns just like with DUV. The wavelength of EUV (Extreme Ultraviolet) lithography is 13.5 nm, which is larger than the feature sizes being created in TSMC's 3nm (N3) process. This means that even with EUV, multiple patterning techniques are still required to achieve the desired resolution for such small features. The complexity of the process increases as the feature sizes shrink, making advanced lithography techniques essential for producing these cutting-edge chips.
I think, that the constant reworks are not dragging feet, but an attempt to stay at the cutting edge. Which is a rapidly moving target. It is very hard to be on the extreme edge of anyhigh tech enterprise, because by the time you design and build a facility, you are a generation behind. The theory and small scale modeling is changing so fast, that scaling it up to production size, is a constantly moving target, which moves faster than the facility design and construction process is constantly out pacing you.
Who can blame TSMC for dragging their heels on their US facilities? If I were them - or the Taiwanese government - I'd be VERY worried that once the US has all of the facility up and running, it would no longer consider the survival of Taiwan terribly important and leave it to the tender mercies of China.
Problem is that TSMC keeps the latest and greatest manufacturing process in Taiwan which is currently 2nm. Even Intel, who has their own foundries, used TSMC for their newest generation of chips because they currently can't produce 2nm.
It's partly that but also figuring out how to move all "mission critical TSMC employees" to the US if a Chinese invasion were to happen. It's a politically uncomfortable reality.
@Jeff-i8u Continuous learning is the reality of successful engineering and technology companies so training the people needed in USA is still a win for TSMC.
Intel is building a large Fab in Chandler, AZ also. I believe it was originally going to be a single 4x patterning immersion DUV process, which Intel decided to scrap mid-construction and turned one Fab into three. Intel apparently believed that EUV would not be viable for many more years and in the 2010's spent a lot of time doing R&D on the 4x patterning immersion process that only yielded 10nm. They actually fell behind and these decisions were rather bizarre considering that they had invested hundreds of millions in grant money to Cymer (later bought by ASML) to develop EUV laser technology.
A friend of mine works on the TSMC site in Arizona and he has said that TSMC is stalling the whole thing so badly, it borders on sabotage, literally every hurdle that can be thrown up is thrown up, and every headache that can be caused is caused, to the point where even emails are purposely left unanswered until the very end of the day just so as to cause another full workday of delays. Everyone there knows this is entirely purposeful, but they're doing just enough that you can't call it outright sabotage...
That's because the taiwanese government really doesn't want to establish a factory in some place that isn't Taiwan, they would much rather have the US provide them with weapons. The taiwanese think the US is stealing TSMC. And yes TSMC being a taiwanese company is under their control.
Why the delay? The factory has to be built eventually. In addition to the issue of the workers' union, the most important thing is to have President Trump cut the ribbon, rather than having it completed now and counting it as President Biden's achievement.
Honestly the Chips Act is small potatoes when it comes to leading edge fabs. It is probably good as "an additional incentive" or to build education supply chains for semiconductor.
Hi, I think the TSMC has already bought and plan to received the high-na euv from ASML before the end of this year. So there is the update from Peter’s information.
It's all about wafer output. I doubt it will even match the output of their fabs in Taiwan, so most of the stuff will still be imported. As for EUV vs DUV, I think it's the amount of patterning that requires the many adjustments. EUV though expensive, requires less layers and DUV needs more patterning and complexity (increases the possibility of errors).
IDK, I am just trying to remember something I read once, but isn't the supplier of dopants and other things need to be ridiculously pure as well? So to make locally, it isn't just the plant itself, but the suppliers that need to be up to speed?
The process of converting silicon ingot into doped and etched sheets x90 sounds like the macroscopic folding process that converts a lump of steel into a Japanese katana, the most high tech sword in the world from the mediaeval period of steelworking.
Thank you Peter, for your thorough analysis. Being a Jack of all trades (usually having no knowledge on the subject), you always rely on three independent sources: rumors, speculations, and media hallucinations. But I am captivated by your frequent hiking endeavors.
Did you mean "trades"? Which point do you disagree with? They have produced the A16 chip in Phoenix, it's the one used in the Iphone 15. He gets a lot right but no one bats a thousand.
Only in American, not in English. They say 'drug' and we say 'dragged', exactly as we say 'dived' and they say 'dove'. It's just a dialect artifact of spending most of our histories separated by about 3500 miles of ocean.🤷🏻♀️
I FIGURED OUT OUR POPULATION DECLINE !! Bucket seats. In the back too ! Pure mayhem. I was proudly conceived in the back of a 57 Chevy Rag Top. At a proper out door movie theater. Get back to basics.
I also like to add it is impossible to synchonize high end chip production between Taiwan and the U.S.. The most high end chip is made while the R&D is continuing to its procedure and yield rate. It can only be done with the R&D center at hand as well as with all the supply chains at hand to make quick adjustment. Unless the U.S. can have the same R&D center as tsmc as well as the 1000 supply chains at hand. There will be at least one to two generations gap.
The nm rankings tends to mean less and less lately. The real prize is how many transistors you can stack in the same spot. It's like going from the city of New York back when it was farmland to the Empire State Building. When real estate became scarce people began building up. New York invented the skyscraper for this reason. Now NYC does more per square foot than just about any other city. Same with chips. Instead of packing 10 transistors area-wise where once was one, stack 20 on top each other.
Wow, congratulations on your impressive :nvestment success! Your discipline and focus on delayed gratification is truly inspiring. I'm curious, what are some of the key factors that you consider when making :nvestment decisions? Do you have any tips for those of us who are just starting to dip our toes into the world of :nvesting? Thanks for sharing your story!
i can't blame them for taking their time. chip production is maybe their one biggest leverage on the US, there's a lack of skilled workers to recruit from in that region, and US labor is an expensive/annoying pain the ass to deal with by Taiwanese or Asian standards, in general.
@ the state of our higher education system is a different dumpster fire, and probably has little immediate relevance to the lack of talent that companies struggle with. For massive projects like this, the talent that companies struggle to find are the experienced midlevel-senior engineers and analysts (not newbie college graduates) who have the work experience and really know what they’re doing. Those skilled workers tend to migrate towards big (usually coastal) cities where the highest paying job opportunities are. TSMC not only has to struggle to locate talent in their area, but also compete for workers who expect double or triple what an equivalent engineer in Taiwan would be making
@@holycrapitsjake_ The erosion of education is exactly why companies cannot hire. Their aren't enough educated grads that actually learned skills and did not get an empty diploma. The biggest issue for hiring is you get lots of applications, but no one is qualified. People who do graduate have never worked a job in their life and barely passed. It is all artificial. We did it to ourselves by forcing college to be super expensive and then only offering high interest rate loans which can never be paid off even if you get a decent job.
Not everywhere but pretty much most places. Yes you literally have to have a housewife or housekeeper to make shopping work in Germany or Switzerland. A joy in California is to shop 24x7 for groceries and even boozes. LOL Not all states have that however.
I respect you a lot Peter, but you’ve missed a few things here. 4nm is what nvidia gpus are built on today and likely for at least a couple more years. And TSMC does have access to high NA with the first installation beginning in September. You may like to revisit you sources.
Micron is an interest duck. The key thing is they are RAM/Flash which has a very different business model and set of hurdles - generally not comparable to TSMC or to Intel.
“baking” does not make anything “stick”. The heat treatments make dopant impurities become active in the silicon lattice or are used to grow or deposit masking layers for the next level.
the name discs is only used for silicone slabs. Once an operation occurs (usually a base metal layer) it becomes a wafer. 90 times? Maybe for a 1980s M68000. Most are 200+ operations. And you're right about automation. the wafers are too large to handle by hand, not to mention humans are pretty dirty when you're talking about clean rooms.
Hi Mr. Zeihan, could you please let me know what gear you are using for your videos? Hardware and software? Do you use any teleprompter software for example?. Thank you for your great content.
@@BlackBullRising Maybe you are since both Peter and the commenters have thoroughly explained the foot dragging. Maybe your pretty outside that you see isn't ready inside.
Hello, everyone... just randomly watch this video and was curious about how Americans react to this ... quite interesting here in the comments. i just couldn't help clarifying some misunderstanding. about the opening delay. yea, we could make it faster, but there is an awkward timing...... the election. if we make the fab done during the campaign period, we will be considered to be in favor for Biden/Harris. So it is better to be done after the result comes out. then the second awkward thing is ..... Trump wins. Since Biden/Harris didn't win, we have to wait for the new president for the opening ceremony. It is quite important, from the political and business perspectives, to kiss the new president's ...
The tear down and rebuild is not a delaying tactic. It is the way to correct production problems before opening completely. Honda builds 25% of the plant, runs it. Corrects the bugs, then rebuilds. At 50% Honds runs the plant tears it apart. Corrects the bugs and rebuilds. At 75% Honda runs the plant tears it down. Corrects the bugs and rebuilds. At 100% Honda tears down the plant. Corrects the bugs and rebuilds. Six months after opening Honda tears down the plant. Corrects the bugs and rebuilds. Honda takes 2 and a half years longer then American car companies to build their car. manufacturing plants. Only then is the plant considered completed. This is why Japanese cars are so good. They practice Edward Deming quality principles to perfection.
Since Dutch company ASML makes photolithography machines (which I think is the main input to create the chips, correct?), why can't Europe produce these high end chips?
There's much more to it than the Lithography and the price of a fab plant is $2Bn and up so European finance is not too keen.. ARM is just a design house not an end to end maker.
Intel’s fab near Magdeburg not only will be Europe’s most advanced semiconductor manufacturing facility; according to Intel CEO Pat Gelsinger, it will also be the world’s most advanced fab, with 1.5-nm chip production. This will start operation in 2030. Intel will rather build this high end fab in Europe than in USA because of trade war between US and China.
You are wrong about the EUV high-NA. tsmc has bought one and installed at its R&D center. Original plan was to use it on 1nm. However, ASML wish tsmc can use it on 1.6nm so it has actually sent 400 engineers to tsmc R&D center to help out to speed up the process to put the high-NA on line. You must understand the high-NA is experimental machine, not yet going into mass production of chips. More research needed to be done, tsmc also sent 250 engineers to ASML Netherlands to work on it. I believe the high-NA process tsmc is also ahead of Intel, there is nothing but delay after delay coming out of Intel. tsmc has no rejection to use high-NA as early as possible, high-NA can reduce the cost to make chip.
Another of my reply also disappeared about the most high end chip. It is impossible to synchronize Taiwan and the U.S. chip making. The most high end in making still needs R&D and yield rate improvement. As well as supply chains to make quick adjustment. As long as the U.S. does not have the same R&D center and the complete supply chains, you will be behind for one to two generations.
The high NA machines are in lay-terms EUV with bigger mirrors. So it's not an whole new revolutionary tech, it's even bigger and more expensive machines for a little bit better result.
Dropping bits of tin hundreds of times a second and hitting it mid air with lasers twice repeatedly seems pretty revolutionary to me.... But ion implantation still seems revolutionary so I guess you just never really get used to it
okay, dumb question: the columbus facility is due to (hopefully) start making stuff in 2026...what are the chances that the tech they are currently talking about being used there will already be yesterday's news and a new process will be coming out?
It already is... the top end technology for TSMC will always remain in Taiwan. The Columbus facility will just enable more supply of other commercial demands for general consumer electronics, cars, etc. Not for AI accelerators nor the latest computing and smart devices. It more to deglobalize away from China than to decouple from Taiwan.
American industrial management has been in crisis since the 1940s when US automotive engineering expertise took itself off to Mitsubishi, Japan, because American corporate managers thought they knew better than engineers. Profits are higher than ever because corporate America has rigged the legal system via $$Billions of lobby cash to Congressional politicians. The cost has been covered by suppressing workers' wages... which they've only just begun to realise... and boy, are they pissed!
@@JohnSmith-xi9nd Intel was capable but they rested on their laurels. High turnover of H1B visa staff from India and China to cut costs has been an issue. Ivy League schools (MIT, Stanford) and top engineering schools (Berkeley, Michigan, Georgia Tech, Austin, Ohio) bragged that they are the best but then they could not produce some engineering staff to supply Intel? No wonder the public confidence in U.S. colleges have deteriorated). No wonder the public sentiment is colleges are a scam.
maybe i can offer my answer from Taiwan's point of view. yea, we could make it faster, but there is an awkward timing...... the election. if we make the fab done during the campaign period, we will be considered to be in favor for Biden/Harris. So it is better to be done after the result comes out. then the second awkward thing is ..... Trump wins. Since Biden/Harris didn't win, we have to wait for the new president for the opening ceremony. It is quite important, from the political and business perspectives, to kiss the new president's ...
Ever thought that the extremely highly skilled labor needed to run the fab doesn't want to work there at that location? TSMC knows this, economists know this, local politicians know this (but will never admit it).
Trump has a tenuous relationship with TSMC - as he has trash-talked Taiwan for "stealing" chip business from the US. As well, Trump will likely destroy the Chips Act created during the Biden administration once he is in office, as that was legislation was created by Democrats (recall how he worked to unwind anything Obama had accomplished?) Once that happens, TSMC will slow roll their plans and not expand beyond what little start they have in Arizona, because the Chips Act offered billions to TSMC and without that, what's the point? TSMC might also close or sell off their only other US wafer fab in Washington state - a 200mm fab over 30 years old and likely a distraction from TSMC's core business. I bet either Skywater or ON buys it.
Trump has a tenuous relationship with TSMC - as he has trash-talked Taiwan for "stealing" chip business from the US. As well, Trump will likely destroy the Chips Act created during the Biden administration once he is in office, as that was legislation was created by Democrats (recall how he worked to unwind anything Obama had accomplished?) Once that happens, TSMC will slow roll their plans and not expand beyond what little start they have in Arizona, because the Chips Act offered billions to TSMC and without that, what's the point? TSMC might also close or sell off their only other US wafer fab in Washington state - a 200mm fab over 30 years old and likely a distraction from TSMC's core business. I bet either Skywater or ON buys it.
@@malin5468 They are releasing a graphics card that is better than an Nvidia 4060 in January, at a price point of 250$. They are going to be flying off the shelves. They announced it yesterday.
@@malin5468 Make sure you are diversified. Don't put all your eggs in one basket. Intel's problems started nearly 20 years ago. They were SURE that Desktop computing was all that mattered. When Steve Jobs approached them to supply a processor for the imminent new product called the "iPhone", Intel said no. So Apple embraced ARM processors and started designing their own after the iPhone started to take off. That was when Intel missed the mobile market out of arrogance. I knew a lot of people at Intel back then and there was a political divide between "Desktop" and "Mobile" supporting employees. The former smugly won the argument and smacked down the pro-Mobile camp - they were "canceled" in a Woke sense of it. One thing those latter engineers and managers were interested in was power efficiency. MOST of the latter left Intel at that point creating a brain-drain of engineers they really needed. But Desktop supporters didn't really care as long as Intel processors delivered top performance at any power price to the Desktop. It's not clear if the mistake was ever realized. Maybe now. The iPhone volume and total revenue eclipsed total PC sales volume and revenue around 2012. That was the beginning. There have been additional "Epic Fails" since then. Many avoidable. Some legitimate errors in innovation. There have been several attempts to become a foundry like TSMC. Those failed for entirely predictable reasons. Intel had problems with their 10 nm process. Honestly I don't blame them for this - changing from Copper to Cobalt is a major challenge and other fabs will face the same problem soon enough. Missing mobile has compounded over time as mobile has taken over as surely as microcomputers (Intel's original claim to fame) overtook discrete TTL minicomputers. Basically Intel missed that transition and has had problems figuring out how to catch up or move in another direction.
Those of us working and living in Silicon Valley said the same thing. We were ignored. And everything we said would happen has mostly happened. I moved to Taiwan for a while because that was easier and paid better than changing my career from scratch.
It depends as these places are super anti union and are trying to import asian work conditions and low wages which isn't working well for americans there, and i feel is on purpose as a way to get H1B visas for taiwanese workers
I remeber watching short video that just like stare every step it takes to make a chip and the comment section on that video say it best. We basically just inscribing magical runes onto a rock XD.
I dont think TSMC was dragging their feet. The facility was supposed to produce 5nm nodes and then converted to 3nm nodes. So likely a late change was required. Also, getting ASML machines to Arizona probably took some time with Intel ordering the bulk of it at the beginning. The CEO of TSMC resigned because the facility was not built on time. Had it been built on time, it will likely be a 5nm node instead of 3nm.
Proud North Carolinian here. If 70% of the world's silicon comes from Spruce pines, even my tiny brain can see that being a massive step in a good direction, as far as production costs. That's stuck in my head because of the state. So what's the availability on the rest of the ingredients? Kind of wondering silicon is the majority of the mass and weight that has to be moved around
@@johnrielley5003 And also: Globalization means every little piece of the final product comes from all over the globe. No one place is special or unique. Until globalization collapses. Then it's all up for grabs...
At 4NM you start having issues pushing actually electrons through. Also 4 NM does not always mean better. You get smaller but also require additional high level integrated chips. Moving forcing TMSC is such an American move to not protect interests.
I didn't like your decision to delay your videos a week, I watch much less because of it. That said, I'm still watching. Excellent update on this vital capability the U.S. needs domestically. I know our armed forces are training for war with China, but I would hate to sacrifice so many Americans over Taiwan. Shutting down trade with China is the wisest thing to do. We can still buy low end products, but no more.
As a construction electrician who worked at TSMC in Phoenix, I can wholeheartedly agree with you that TSMC was dragging their feet in getting the Fabrication building completed. I had "adjustments" to my scope of work due to the designs changing multiple times, resulting in a lot of wasted time and material. It was incredibly frustrating trying to meet deadlines when scope was changed mid-stream on multiple occasions.
How much union work was involved in this contract?
Yeah ... is it due to changes, necessary changes? Or have they just dog an ponied building something to get the gov't money? I just don't think they'll be sincere.
Welcome to the chip world. I'm retired, and that was the norm. At any point we would be sent out the gate because a chip was outdated in 6 months.
Having an alternative to buying Taiwanese chips make the US less likely to defend Taiwan when the CHICOMS attack
Meh, doesn’t sound different from the public sector, i.e., normal government contracts. 😅
As a truck driver I pass by that TSMC facility north of Phoenix regularly. I didn't realize what that project is until just now. Yes, it's been under construction for MANY years. It is MASSIVE, impressive, and a mess. Thank you for triggering my 40 watt bulb to turn on this morning.
Lol. I have a 60 watt bulb but i run this thing at 10k when i can!
+10B gigafactories
If that bulb is an LED, 40 Watts is mighty bright!
Why the delay? The factory has to be built eventually.
In addition to the issue of the workers' union, the most important thing is to have President Trump cut the ribbon, rather than having it completed now and counting it as President Biden's achievement.
@@UlyssesMax Cutting a ribbon does NOT signify it is/was trump's achievement. It is a purely ceremonial act. If you're going to give credit to a president, it's Biden, as it was his 'CHIPS' act which funded $280 billion towards bringing semiconductor and other technology back to the U.S. It was part of his $1 trillion package to revamp aging U.S.infrastructure While trump's 2016 campaigned talked about a trillion infrastructure plan, he did ZERO in that area. Once he was president, Trump suggested privatizing highways, making them for-profit toll roads.
Biden's softer style helped gain Bipartisan support in Congress to get quite a few infrastructure bills passed. But, there's no doubt trump will claim full credit despite no engagement in it.
Peter is the absolute king of backdrops. Sometimes I can’t help but watch just to see the backdrop.
I live in Phoenix and work in manufacturing. The thing that hurts our heads is that they are trying to make chips (a very water hungry industry) in a desert city. A desert city that just lost a large chunk of it's water rights after the last negotiations.
Who do you think they will prioritize? Dumb desert people or the computer chips of the future?
@@ace1776 What makes them dumb? And who is going to work in the factory once those "Dumb Desert People" move because they lost their water? We gonna bring in workers on rotation and then bus them to somewhere smart?
A lot of that water can be recycled
At least you'll always have more "rights" than Native Americans on the DeathMarchReservation System.
I bet it uses a lot less water than any of those Scottsdale golf courses.
I think moving manufacturing to the US will be highly automated. It’s the direction of manufacturing in general and especially makes sense in developed nations.
Considering that each wafer is just short of a million dollars, a 4% improvement is actually huge, "not insignificant" .
Peter, the whole process of growing a silicon crystal starts with using a small silicon crystal seed which has the correct FCC (Face-Centered-Cubic) orientation required for semiconductor processing which should be propagated throughout the drawn silicon ingot. The silicon structure of the ingot is sampled early in in the process of drawing the ingot from the pool of liquid silicon and if it doesn't match the required FCC orientation, which I think is 011, then the process is restarted.
This mostly a supplier problem and NOT one that TSMC (or any other semiconductor fab). worries about - it's supply chain qualification/rejection criteria. TSMC only worries about the volume and qualitied supplied regardless of supplier. The failing supplier simply does not get paid for the rejected material. But another supplier will step up and fill the gap. This is the major challenge of being a TSMC supplier/vendor - they are demanding and competing is expensive.
@@Jeff-i8u Yep, there are only 2 or 3 companies at best that even grow the silicon ingots. And Zeihan covered a serious problem in another of his videos that one of the best (in the USA) sources of the silicon used for this had one of their mines flood. And production capacity was severely cut. Maybe they recovered by now. I have read other stories that the TSMC work culture at the Fab is not going over well with US workers. IMHO Intel is on the brink of collapse in terms of fab production, and the Biden CHIPs act is on the Trump chopping block. So this will take time to play out.
The Phoenix factories are completely done on the outside at least. I live down the street, Phoenix is creating an entirely new city around them.
I think Peter is misunderstanding what’s really needed for most deep learning model training. The push for sub 10nm processes, or even 100nm, isn’t as important as he makes it sound. Compute power isn’t the bottleneck for AI workloads. The real issue is memory bandwidth and how efficiently vector cores can handle matrix multiplication (MATMUL) operations.
If you’re building a superfast GPU for general purpose use, then yeah, smaller feature sizes matter because you’re trying to cram as many transistors as possible onto a silicon chip. But when it comes to AI workloads, like training transformer models or running inference, the priorities are different. You want to load as much of the model into RAM as possible and run massive parallel operations in a single cycle. For training, you also need to calculate and store gradients for backpropagation and gradient descent, but none of this actually requires the newest lithography.
What Peter is missing is that AI tasks are super parallel and asynchronous. You don’t need the latest and greatest chip tech for that. You could use something like 100nm lithography to make chips loaded with tons of 16 bit floating point adders and multipliers, with DRAM built right in. It wouldn’t be cutting edge, but it would still handle AI workloads really well and be way more cost effective.
The problem with Peter’s argument is that he’s assuming not having access to sub 10nm tech means a country like China is locked out of the AI game. Smaller feature sizes do help with efficiency and performance, but AI workloads don’t actually need the cutting edge to be competitive.
Thank you for this comment. I hope Peter addresses this.
Thanks for this insight, even if some of it is beyond my grasp.
Vacuum tube computers got us to the moon, so low tech is not necessarily useless, it's just more resource intensive
@@dcc70 I think you are very wrong. By the time of the Apollo project,. computers were solid state. No more Univacs. Even the worst of solid state computers made anything vacuum tube laughable for many reasons.
Agreed
According to IEEE on 12/26/2024, one TSMC FAB in Arizona is producing 4nm transistors. The second FAB is scheduled to be operational by 2028 and produce 2nm transistors. Those 4nm transistors are what power NVIDIA's newest GPUs. Absolutely cutting edge.
Worked at PHX TSMC until Jan ‘24 and I never heard the phrase “take your time” so many times in my life.
The TSMC facilities are the most complex engineering created by mankind, perhaps apart from the Large Hadron Collider in Europe.
A fully fledged manufacturing facility akin to the plants in Taiwan, comes with a pricetag of ~$100bn 😳
100bn only builds a small hospital in Oz.
Put down a layer of e.g. copper, coat the chip with photo resist, Expose to EUV the photo-resist to print the outline of wires on the chip, etch off the areas not exposed, repeat (up to 90x times). Sometimes after printing a layer, you dope (heat & infuse the semiconductor layer with doping materials like germanium) to increase the semi-conducting property, then you remove the entire layer and start over.
Intel and the state of Ohio have given boat loads of money to all the local community colleges here in central Ohio to pump out as many semi-conductor grads as possible in the next 2 years. We’ll see if it works out.
it wont, we no longer do anything on merit, prob end up with a bunch of know nothings that just got passed anyway to make the numbers look good. Who cares if they even wanted to go in that field at all
That's not all they did, Gelsinger really did have a master plan.
commenting so i can come back to this later
They'd better also pump boatloads of money into building more houses, because the cost of housing is about to go through the stratosphere around here if they don't. The economic boost is great, but less so if nobody can afford to live here.
TSMC can't find enough US engineers
1:49 - That's making an ingot, that's sliced into wafers. The ingots and wafers are usually made in another facility. They are often a different company than the one making the chips. Think the companies that make tires and sheet metal, that the car company uses to make their vehicles.
Yes, Zeihan knows next to nothing about this industry, yet he keeps opining as though he’s an expert.
@@TimAZ-ih7yb He has explained these facts for the better part of 2 decades. I’m critical of him when he’s off, but this isn’t one of those spaces.
Thats correct. Silicon wafers (substrates) are made elsewhere.... Supplied by one of the big Silicon wafer suppliers such as Shin Etsu, SUMCO etc..
That’s the terminal I Built!!! Hope you had a good weekend time in Boston
Nice terminal! Thanks for it!
Well done Chad 🫡
Peter seems to be mixing up totally separate phases of semiconductor manufacturing - from crystal pulling and wafer manufacturing to actual processing of those wafers into chips. Siltronic is a company that specializes in the former, TSMC in the later.
Chip production is amazingly high tech. It’s hard to believe someone came up with that process.
TSMC isn’t developing the technology. They work on the “tail-end” of production with the actual companies that develop the technology. The real high tech is at those hundreds of other companies.
@@chillxxx241 id say the machine maker is the real hero, but what he is talking about is the process he didnt even bring of TSMC, he i just saying I find it amazing that some dude figured out how to grow chips on wafers
@ The core technology that ASML uses was built by the US government bank in the 70s, but none of the US companies wanted it. There are now 800 or more companies that help ASML make those machines at an affordable price. Neither ASML nor TSMC design the semiconductors.
not "someone". Hundreds of thousands (if not millions) of high-tech engineers, over many painful and expensive iterations across what must be 50-60 years.
Many have failed and we never hear about them but even the failures were extremely important in the knowledge we eventually solidified into modern chip manufacturing.
I've just finished reading a book called "Chip War" about the battle for supremcy in chip production and it it explains in lay person terms these processes for producing and the machinery involved. All I can say is there are some bloody clever people in this world, it's insane what they have developed, your average consumer is oblivious to the level of genius they have in their hands every time they check their smartphone.
Thank you for this video
Thanks for covering this. The MSM certainly hasn't been.
Why the delay? The factory has to be built eventually.
In addition to the issue of the workers' union, the most important thing is to have President Trump cut the ribbon, rather than having it completed now and counting it as President Biden's achievement.
I was hoping for video of the site, went pass a couple years ago and was amazed at the scope.
So, Kissinger basically decided to move shit to Redchina and now someone said it's time to move things back.
Each time we the simple people have to live with the effects.
Unfortunately, the vast majority of that manufacturing is never coming back. This is a tiny tiny drop in the bucket of what was lost.
@@nazcamain Labor-intensive jobs are never coming back but most semiconductor manufacturing is 100% automated. I've spent a lot of time in Taiwan on their fab floors and MOST of the workers you see are vendor contractors keep their equipment working. Actually fab employees are less numerous on the fab floor. So NO high tech today is ever going to create mass quantities of jobs. The world died in the 1960s and 1970s. It's never coming back. Most jobs are technicians "feeding the automation" but that's still going to be labor costs (and thus # of hires) less than 1% of what it was in the labor-intensive manufacturing era of the 1960s. Most money for governments will NOT be employment taxes but other taxes only. There are no business models that hire 100K-500K employees anymore.
Actually it was moved to all of Asia except Red China. Taiwan and Singapore being the closest in terms of Chinese culture ever Communism and top-down control are still incompatible with complex STEM. If you can't fail, you can't innovate.
Currently China is largely stuck on planar processes at 15 nm and above. That tried some of the 3D FinFET technologies before 15 nm but they do not have the EUV so they have to make due with the far more complicated larger wavelength older photolith. And these require incredible skills and math/computing intense multi-patterning photolith to make work. Tariffs and export restrictions a nice FU to China: tie both hands behind your back and wear this blindfold. LOL
As it turns out 90% of the fabs created in China over the last 15-20 years were sham companies designed for embezzlement rather than semiconductor, and most have now failed. The original SMIC is still around but struggling to make sub-15 nm parts. China is way behind and largely locked out of anything resembling the processor in your iPhone.
On the other hand, that market is relatively small - most semiconductor volume is NOT deep nanometer but making things for dish washers, cars, etc that are much larger. This is where the West screwed itself - we still need large geometry IC designs and products. Some companies are working on doing that and small geometry but China has dominated this low margin business.
Other the other hand, things is China are going VERY VERY badly economically. Bad enough that a revolution could happen to over-through the CCP/ChiComs. But we will see.
@Jeff-i8u this was more a statement about the general and sudden destructive top down deindustrialization of the rust belt and Europe.
I know better chips never came from Redchina.
But let's be honest: the same disconnected functional elites that decided now that things are suddenly moving back (again at the cost of consumers, before at the cost of employees and competition) will or are already making any other industry move around globally or simply not being available to Joe Average.
The confrontation course of their fellow WEF puppet In the Kremlin was known to them already since Georgia (2009), latest since Donbass and Crimea (2014).
It's a top down deindustrialization and down scaling. Hidden behind some not so sudden geopolitics.
It's a dirty game. Like it always was.
@Jeff-i8u will the elites sell us more expensive washing machines with super chips or phase us the cannon fodder population generally out during this centrally planned down scaling of markets and deindustrialized under the guise of green policies and instigated international conflicts?
I mean: we were encouraged by them to completely allow an integrated global economy and forced to accept the alleged "peace dividend" and consequences of it.
Need to do an addendum on how Ukraine is the worlds leading producers on the gases used in the laser that are used in the production of these chips
Problem is most of those minerals n rare earth metals are now in Russian controlled territory
I believe he touched on that during the start of the war
Those resources are everywhere. Just have to be found and more expensive to dig out. But still better than funding an enemy
@@jumbam9938 That depends on if trump turns against the US or turns against Russia. If trump protects american interests, he would do the same as nixon and end the war by ramping up ukraine's attacks on russia.
If trump supports russia, he will take steps to undermine ukraine.
We'll know in a month and a half.
That's neon which really is everywhere. It's in the air you breathe.
Gotta love the confidence of Peter. Almost seems he knows what he is talking about.
Agree, he seems credible until he talks about something you know about.
Fr lmao. Figured this out about 4 years ago. Inside China Business with Kevin is far more insightful. Dude actually lives in China since 2012, and his 6-10 min videos are dense with recent news articles and him tying each of them together. Peter doesn’t even speak Chinese nor has been to China yet is somehow an expert, where the Same experts in the US about China have been insanely wrong (like Gordon Chang)
The move toward EUV lithography in TSMC’s U.S. facilities is a fascinating leap. These advancements are reshaping global semiconductor manufacturing, especially in improving efficiency and reducing environmental impact.
@@AdvantestInc let's hope the simple folks will still enjoy the innovation dividend from that!
... before Peter and his friends are crawling into their luxury version of doomsday bunkers while the rest of us and our devices will absorb a few EMPs 😂😆
Having worked directly on several mega construction projects including clean room R&D and chip making facilities I can tell you there is NEVER a "Final" design and you are ALWAYS updating as the project continues. Often I have seen ongoing work on one end of the building while the previous end is being torn out and reconfigured. This isn't so much feet dragging as it is Producer level changes to their delivery contracts. The construction progress is time consumptive and the years it takes to develop a project is the hamstring in producing current tech products. There is indeed a point where construction completeness outweighs further reconstruction for updates, but it is often very late stage. It is only at that point that the "Final" design is really just a decision to stop changes in favor of "good enough" to get products flowing.
Peter, you are wrong. EUV simplified earlier nodes, but for the latest nodes they use multiple patterns just like with DUV. The wavelength of EUV (Extreme Ultraviolet) lithography is 13.5 nm, which is larger than the feature sizes being created in TSMC's 3nm (N3) process. This means that even with EUV, multiple patterning techniques are still required to achieve the desired resolution for such small features. The complexity of the process increases as the feature sizes shrink, making advanced lithography techniques essential for producing these cutting-edge chips.
4nm is absolutely cutting edge. I don't know what people are complaining about.
It is, my potato gaming machine is still running a 25 nm AMD chip.
@RamonChiNangWong078 lol. My laptop is still something like 32nm. If only it had a TPM that was v2.0 complaint...
@@10100rsn You made me laugh!
1-2nm fabs being built in Taiwan right now
You can’t have true 1 manometer chips due to quantum tunnelling
I think, that the constant reworks are not dragging feet, but an attempt to stay at the cutting edge. Which is a rapidly moving target. It is very hard to be on the extreme edge of anyhigh tech enterprise, because by the time you design and build a facility, you are a generation behind. The theory and small scale modeling is changing so fast, that scaling it up to production size, is a constantly moving target, which moves faster than the facility design and construction process is constantly out pacing you.
I think Taiwan wants to make chips at home so that the USA can't afford to let China wreck the place.
Who can blame TSMC for dragging their heels on their US facilities? If I were them - or the Taiwanese government - I'd be VERY worried that once the US has all of the facility up and running, it would no longer consider the survival of Taiwan terribly important and leave it to the tender mercies of China.
Good point but not likely as CCP having TSMC is very different to TW having TSMC and different to USA depending on TSMC in TW.
Problem is that TSMC keeps the latest and greatest manufacturing process in Taiwan which is currently 2nm. Even Intel, who has their own foundries, used TSMC for their newest generation of chips because they currently can't produce 2nm.
It's partly that but also figuring out how to move all "mission critical TSMC employees" to the US if a Chinese invasion were to happen. It's a politically uncomfortable reality.
@Jeff-i8u Continuous learning is the reality of successful engineering and technology companies so training the people needed in USA is still a win for TSMC.
Intel is building a large Fab in Chandler, AZ also. I believe it was originally going to be a single 4x patterning immersion DUV process, which Intel decided to scrap mid-construction and turned one Fab into three. Intel apparently believed that EUV would not be viable for many more years and in the 2010's spent a lot of time doing R&D on the 4x patterning immersion process that only yielded 10nm. They actually fell behind and these decisions were rather bizarre considering that they had invested hundreds of millions in grant money to Cymer (later bought by ASML) to develop EUV laser technology.
A friend of mine works on the TSMC site in Arizona and he has said that TSMC is stalling the whole thing so badly, it borders on sabotage, literally every hurdle that can be thrown up is thrown up, and every headache that can be caused is caused, to the point where even emails are purposely left unanswered until the very end of the day just so as to cause another full workday of delays. Everyone there knows this is entirely purposeful, but they're doing just enough that you can't call it outright sabotage...
That's because the taiwanese government really doesn't want to establish a factory in some place that isn't Taiwan, they would much rather have the US provide them with weapons. The taiwanese think the US is stealing TSMC. And yes TSMC being a taiwanese company is under their control.
Why the delay? The factory has to be built eventually.
In addition to the issue of the workers' union, the most important thing is to have President Trump cut the ribbon, rather than having it completed now and counting it as President Biden's achievement.
No mention of the chips & science act?
USA is becoming Russia... Peter is no exception. When turmp decide to stay past his second term, he'll say it's fine
Honestly the Chips Act is small potatoes when it comes to leading edge fabs. It is probably good as "an additional incentive" or to build education supply chains for semiconductor.
@@Jeff-i8u it brough about 80% of financing, but it's just "small potatoes" obviously
Hi, I think the TSMC has already bought and plan to received the high-na euv from ASML before the end of this year. So there is the update from Peter’s information.
ture, but not for the 2nm manufacturing.
TSMC bought one for RD department.
High NA EUV is at oregon/USA site FYI Peter
Noticed a timestamp date on this vid. Nice touch.
Now this is a Zihan I can get behind I want to know everything tsmc
Peter shows why I hate red-eye flights.
What’s the point of him being here, then there, ad nauseam?
It's all about wafer output. I doubt it will even match the output of their fabs in Taiwan, so most of the stuff will still be imported. As for EUV vs DUV, I think it's the amount of patterning that requires the many adjustments. EUV though expensive, requires less layers and DUV needs more patterning and complexity (increases the possibility of errors).
Are you on the run? Each video is in a new location. but I love the outdoor hiking videos.
He's real life Where's Waldo.
@Utoko You must be new here
Peter must run the table on Polymarket
Why he is wrong non-stop
@@adairjanney7109 😂😂😂
@@adairjanney7109 He predicted the war in Ukraine, for starters.
Peter theil is evil
@@adairjanney7109 He wasn't wrong on Ukraine, among other things.
IDK, I am just trying to remember something I read once, but isn't the supplier of dopants and other things need to be ridiculously pure as well? So to make locally, it isn't just the plant itself, but the suppliers that need to be up to speed?
The process of converting silicon ingot into doped and etched sheets x90 sounds like the macroscopic folding process that converts a lump of steel into a Japanese katana, the most high tech sword in the world from the mediaeval period of steelworking.
Thank you Peter, for your thorough analysis. Being a Jack of all trades (usually having no knowledge on the subject), you always rely on three independent sources: rumors, speculations, and media hallucinations. But I am captivated by your frequent hiking endeavors.
Did you mean "trades"? Which point do you disagree with? They have produced the A16 chip in Phoenix, it's the one used in the Iphone 15. He gets a lot right but no one bats a thousand.
Thank you Peter
Since when 4nm is not state of the art, currently only TSMC can do it at scale.
"Drug out" is the part of "drag out"?
Only in American, not in English. They say 'drug' and we say 'dragged', exactly as we say 'dived' and they say 'dove'. It's just a dialect artifact of spending most of our histories separated by about 3500 miles of ocean.🤷🏻♀️
Does still grate a bit and live lived here for over 20 years
I FIGURED OUT OUR POPULATION DECLINE !! Bucket seats. In the back too ! Pure mayhem. I was proudly conceived in the back of a 57 Chevy Rag Top. At a proper out door movie theater. Get back to basics.
Dec 2 ‘release’ date? Or is that recorded date? Where released if not here?
He releases to Patreon and then after a week releases it again on RUclips.
Thanks Peter
Everything is easy for Peter
Damn you're in Boston. I'd buy you a drink!!
I also like to add it is impossible to synchonize high end chip production between Taiwan and the U.S.. The most high end chip is made while the R&D is continuing to its procedure and yield rate. It can only be done with the R&D center at hand as well as with all the supply chains at hand to make quick adjustment. Unless the U.S. can have the same R&D center as tsmc as well as the 1000 supply chains at hand. There will be at least one to two generations gap.
The nm rankings tends to mean less and less lately. The real prize is how many transistors you can stack in the same spot.
It's like going from the city of New York back when it was farmland to the Empire State Building. When real estate became scarce people began building up. New York invented the skyscraper for this reason. Now NYC does more per square foot than just about any other city.
Same with chips. Instead of packing 10 transistors area-wise where once was one, stack 20 on top each other.
Yes. Nicely said!
Thank you for recommending Sarah Jennine Davis on one of your videos. I reached out to her and investing with her has been amazing.
Wow, congratulations on your impressive :nvestment success! Your discipline and focus on delayed gratification is truly inspiring. I'm curious, what are some of the key factors that you consider when making :nvestment decisions? Do you have any tips for those of us who are just starting to dip our toes into the world of :nvesting? Thanks for sharing your story!
Do you mind sharing info on the adviser who
assisted you? I'm 39 now and would love to
grow my portfolio and plan my retirement
@@FreyaFreya3 Sarah Jennine Davis is highly recommended
You most likely should get her basic info when you search her on your browser.
@@mayor-o1wHow do I access her ? I really need this
+156
i can't blame them for taking their time. chip production is maybe their one biggest leverage on the US, there's a lack of skilled workers to recruit from in that region, and US labor is an expensive/annoying pain the ass to deal with by Taiwanese or Asian standards, in general.
The issue is that kids stopped going to college due to cost and student loans having credit card levels of interest. We eroded our education system.
@ the state of our higher education system is a different dumpster fire, and probably has little immediate relevance to the lack of talent that companies struggle with. For massive projects like this, the talent that companies struggle to find are the experienced midlevel-senior engineers and analysts (not newbie college graduates) who have the work experience and really know what they’re doing. Those skilled workers tend to migrate towards big (usually coastal) cities where the highest paying job opportunities are. TSMC not only has to struggle to locate talent in their area, but also compete for workers who expect double or triple what an equivalent engineer in Taiwan would be making
@@holycrapitsjake_ The erosion of education is exactly why companies cannot hire. Their aren't enough educated grads that actually learned skills and did not get an empty diploma. The biggest issue for hiring is you get lots of applications, but no one is qualified. People who do graduate have never worked a job in their life and barely passed.
It is all artificial. We did it to ourselves by forcing college to be super expensive and then only offering high interest rate loans which can never be paid off even if you get a decent job.
Are your stores in the US open 24/7? Do we in Switzerland have flights at 5 o'clock morning? I think we don't even have trains that early.
Not everywhere but pretty much most places. Yes you literally have to have a housewife or housekeeper to make shopping work in Germany or Switzerland. A joy in California is to shop 24x7 for groceries and even boozes. LOL Not all states have that however.
Love this topic
I respect you a lot Peter, but you’ve missed a few things here. 4nm is what nvidia gpus are built on today and likely for at least a couple more years. And TSMC does have access to high NA with the first installation beginning in September. You may like to revisit you sources.
What about the new microchip plant being built in Boise? Micron is building a facility at warp speed.
Micron is an interest duck. The key thing is they are RAM/Flash which has a very different business model and set of hurdles - generally not comparable to TSMC or to Intel.
Micron is all RAM, not CPU.
TSMC has been on a hiring spree for process technicians in Arizona.
“baking” does not make anything “stick”. The heat treatments make dopant impurities become active in the silicon lattice or are used to grow or deposit masking layers for the next level.
Hello from Boston.
the name discs is only used for silicone slabs. Once an operation occurs (usually a base metal layer) it becomes a wafer. 90 times? Maybe for a 1980s M68000. Most are 200+ operations. And you're right about automation. the wafers are too large to handle by hand, not to mention humans are pretty dirty when you're talking about clean rooms.
Maybe appoint Randy Quaid to the 'get TMC moving faster' czar position;-)
Hi Mr. Zeihan, could you please let me know what gear you are using for your videos? Hardware and software? Do you use any teleprompter software for example?. Thank you for your great content.
Before I clicked the video I saw the background and assumed Peter was getting a tour of the facility kol
Watching this on 22nm CPU from the previous decade, 4nm sounds good enough just get the damn fab rolling. 1:17
The TSMC facility in Phoenix looks complete to me. It went up pretty quickly.
Not what Peter and many Phoenix commenters say. I think you are wrong.
That's the outside your seeing, the inside is a whole different ball game.
@@frequentlycynical642 I only pass by it once a week. Guess I'm just seeing things.
@@BlackBullRising Maybe you are since both Peter and the commenters have thoroughly explained the foot dragging. Maybe your pretty outside that you see isn't ready inside.
Hello, everyone... just randomly watch this video and was curious about how Americans react to this ...
quite interesting here in the comments.
i just couldn't help clarifying some misunderstanding.
about the opening delay. yea, we could make it faster, but there is an awkward timing...... the election.
if we make the fab done during the campaign period, we will be considered to be in favor for Biden/Harris.
So it is better to be done after the result comes out.
then the second awkward thing is ..... Trump wins.
Since Biden/Harris didn't win, we have to wait for the new president for the opening ceremony.
It is quite important, from the political and business perspectives, to kiss the new president's ...
The tear down and rebuild is not a delaying tactic. It is the way to correct production problems before opening completely. Honda builds 25% of the plant, runs it. Corrects the bugs, then rebuilds. At 50% Honds runs the plant tears it apart. Corrects the bugs and rebuilds. At 75% Honda runs the plant tears it down. Corrects the bugs and rebuilds. At 100% Honda tears down the plant. Corrects the bugs and rebuilds. Six months after opening Honda tears down the plant. Corrects the bugs and rebuilds. Honda takes 2 and a half years longer then American car companies to build their car. manufacturing plants. Only then is the plant considered completed. This is why Japanese cars are so good. They practice Edward Deming quality principles to perfection.
Since Dutch company ASML makes photolithography machines (which I think is the main input to create the chips, correct?), why can't Europe produce these high end chips?
There's much more to it than the Lithography and the price of a fab plant is $2Bn and up so European finance is not too keen..
ARM is just a design house not an end to end maker.
@@lonpfrb I guess I'll have to study what else goes into creating the end product. Thanks.
Intel’s fab near Magdeburg not only will be Europe’s most advanced semiconductor manufacturing facility; according to Intel CEO Pat Gelsinger, it will also be the world’s most advanced fab, with 1.5-nm chip production. This will start operation in 2030. Intel will rather build this high end fab in Europe than in USA because of trade war between US and China.
@@keyto5526 Don't get your hopes high on Intel. They will be lucky if they survive until that date with the US govt bankrolling them.
You are wrong about the EUV high-NA. tsmc has bought one and installed at its R&D center. Original plan was to use it on 1nm. However, ASML wish tsmc can use it on 1.6nm so it has actually sent 400 engineers to tsmc R&D center to help out to speed up the process to put the high-NA on line. You must understand the high-NA is experimental machine, not yet going into mass production of chips. More research needed to be done, tsmc also sent 250 engineers to ASML Netherlands to work on it. I believe the high-NA process tsmc is also ahead of Intel, there is nothing but delay after delay coming out of Intel. tsmc has no rejection to use high-NA as early as possible, high-NA can reduce the cost to make chip.
Another of my reply also disappeared about the most high end chip. It is impossible to synchronize Taiwan and the U.S. chip making. The most high end in making still needs R&D and yield rate improvement. As well as supply chains to make quick adjustment. As long as the U.S. does not have the same R&D center and the complete supply chains, you will be behind for one to two generations.
The high NA machines are in lay-terms EUV with bigger mirrors. So it's not an whole new revolutionary tech, it's even bigger and more expensive machines for a little bit better result.
Dropping bits of tin hundreds of times a second and hitting it mid air with lasers twice repeatedly seems pretty revolutionary to me.... But ion implantation still seems revolutionary so I guess you just never really get used to it
I don't mean that EUV isn't revolutionary, I mean that high NA EUV isn't an revolutionary step from "normal" EUV.
okay, dumb question: the columbus facility is due to (hopefully) start making stuff in 2026...what are the chances that the tech they are currently talking about being used there will already be yesterday's news and a new process will be coming out?
It already is... the top end technology for TSMC will always remain in Taiwan. The Columbus facility will just enable more supply of other commercial demands for general consumer electronics, cars, etc. Not for AI accelerators nor the latest computing and smart devices. It more to deglobalize away from China than to decouple from Taiwan.
I don't know if you are aware, but TSMCs Arizona facility is up and running. Just not at full production.
If only #INTEL could survive the management crisis first.
American industrial management has been in crisis since the 1940s when US automotive engineering expertise took itself off to Mitsubishi, Japan, because American corporate managers thought they knew better than engineers.
Profits are higher than ever because corporate America has rigged the legal system via $$Billions of lobby cash to Congressional politicians. The cost has been covered by suppressing workers' wages... which they've only just begun to realise... and boy, are they pissed!
It is hard to believe Intel has been pulled from the top NASDAQ listings and replaced by Nvidia.
True, but even Intel had to go with TSMC for their recently released gen of chips because their own foundry isn't capable of achieving 2nm.
@@JohnSmith-xi9nd
Intel was capable but they rested on their laurels.
High turnover of H1B visa staff from India and China to cut costs has been an issue.
Ivy League schools (MIT, Stanford) and top engineering schools (Berkeley, Michigan, Georgia Tech, Austin, Ohio) bragged that they are the best but then they could not produce some engineering staff to supply Intel?
No wonder the public confidence in U.S. colleges have deteriorated). No wonder the public sentiment is colleges are a scam.
Indeed. As an Intel alum.
You really need to get up to date on semiconductors. Nvidia H100s are built on TSMC 4N.
Fabs are much cleaner now lol
Why would TSMC intentionally delay completion? You need to explain this.
maybe i can offer my answer from Taiwan's point of view.
yea, we could make it faster, but there is an awkward timing...... the election.
if we make the fab done during the campaign period, we will be considered to be in favor for Biden/Harris.
So it is better to be done after the result comes out.
then the second awkward thing is ..... Trump wins.
Since Biden/Harris didn't win, we have to wait for the new president for the opening ceremony.
It is quite important, from the political and business perspectives, to kiss the new president's ...
Ever thought that the extremely highly skilled labor needed to run the fab doesn't want to work there at that location? TSMC knows this, economists know this, local politicians know this (but will never admit it).
Phoenix is a shit hole.
Where do you get all this info? Where can I read more about the Trump-TSMC relationship?
also researched it and it's invisible to google
to study Chinese and baidu on this edging techno.
Trump has a tenuous relationship with TSMC - as he has trash-talked Taiwan for "stealing" chip business from the US. As well, Trump will likely destroy the Chips Act created during the Biden administration once he is in office, as that was legislation was created by Democrats (recall how he worked to unwind anything Obama had accomplished?) Once that happens, TSMC will slow roll their plans and not expand beyond what little start they have in Arizona, because the Chips Act offered billions to TSMC and without that, what's the point? TSMC might also close or sell off their only other US wafer fab in Washington state - a 200mm fab over 30 years old and likely a distraction from TSMC's core business. I bet either Skywater or ON buys it.
Fox news
Trump has a tenuous relationship with TSMC - as he has trash-talked Taiwan for "stealing" chip business from the US. As well, Trump will likely destroy the Chips Act created during the Biden administration once he is in office, as that was legislation was created by Democrats (recall how he worked to unwind anything Obama had accomplished?) Once that happens, TSMC will slow roll their plans and not expand beyond what little start they have in Arizona, because the Chips Act offered billions to TSMC and without that, what's the point? TSMC might also close or sell off their only other US wafer fab in Washington state - a 200mm fab over 30 years old and likely a distraction from TSMC's core business. I bet either Skywater or ON buys it.
What kind of logistics adjustments are needed for TSMC to build on the other side of the world?
There's vertical integration but still some specialist suppliers.
First you need a big pot of money like the Chips Act.
According to CNN 6 days ago... "Intel ruled tech. Now it’s in deep trouble."
Trouble is a nice way to put it
@@sabero5668 What do you mean? Is it going bust? I hope not for the sake of my shares!
@@malin5468 They are releasing a graphics card that is better than an Nvidia 4060 in January, at a price point of 250$.
They are going to be flying off the shelves. They announced it yesterday.
@@ScienceTeacher92262 I hope you are right!
@@malin5468 Make sure you are diversified. Don't put all your eggs in one basket.
Intel's problems started nearly 20 years ago.
They were SURE that Desktop computing was all that mattered. When Steve Jobs approached them to supply a processor for the imminent new product called the "iPhone", Intel said no. So Apple embraced ARM processors and started designing their own after the iPhone started to take off.
That was when Intel missed the mobile market out of arrogance. I knew a lot of people at Intel back then and there was a political divide between "Desktop" and "Mobile" supporting employees. The former smugly won the argument and smacked down the pro-Mobile camp - they were "canceled" in a Woke sense of it.
One thing those latter engineers and managers were interested in was power efficiency. MOST of the latter left Intel at that point creating a brain-drain of engineers they really needed. But Desktop supporters didn't really care as long as Intel processors delivered top performance at any power price to the Desktop. It's not clear if the mistake was ever realized. Maybe now.
The iPhone volume and total revenue eclipsed total PC sales volume and revenue around 2012.
That was the beginning. There have been additional "Epic Fails" since then. Many avoidable. Some legitimate errors in innovation.
There have been several attempts to become a foundry like TSMC. Those failed for entirely predictable reasons.
Intel had problems with their 10 nm process. Honestly I don't blame them for this - changing from Copper to Cobalt is a major challenge and other fabs will face the same problem soon enough.
Missing mobile has compounded over time as mobile has taken over as surely as microcomputers (Intel's original claim to fame) overtook discrete TTL minicomputers. Basically Intel missed that transition and has had problems figuring out how to catch up or move in another direction.
I would have sworn that was IAD, I guess they have similar food court architecture.
Is drug really the past tense of “to drag” now? Because it wasn’t when I was school.
Why this isn’t our bread and butter is kinda weird…we need to be the best at this.
Those of us working and living in Silicon Valley said the same thing. We were ignored. And everything we said would happen has mostly happened. I moved to Taiwan for a while because that was easier and paid better than changing my career from scratch.
More than the weight of which model VW?
Can Hank who used to work in the now defunct car factory and was able to upgrade his pickup truck every 2 years work in these factories?
It depends as these places are super anti union and are trying to import asian work conditions and low wages which isn't working well for americans there, and i feel is on purpose as a way to get H1B visas for taiwanese workers
Clean room working has zero oily rags so it's up to Hank to change...
Depends on if Hank goes back to school and learns some new skills.
Oh yeah. . . . Peter Ziehan. . . . . .
I remeber watching short video that just like stare every step it takes to make a chip and the comment section on that video say it best. We basically just inscribing magical runes onto a rock XD.
I dont think TSMC was dragging their feet. The facility was supposed to produce 5nm nodes and then converted to 3nm nodes. So likely a late change was required. Also, getting ASML machines to Arizona probably took some time with Intel ordering the bulk of it at the beginning. The CEO of TSMC resigned because the facility was not built on time. Had it been built on time, it will likely be a 5nm node instead of 3nm.
TSMC Phoenix is the factory of the future, and will always be in the future.
4nm is enough to make server farms if you have the money to throw lots of power at it.
I’m building the Columbus facility…. Its BIG
Proud North Carolinian here. If 70% of the world's silicon comes from Spruce pines, even my tiny brain can see that being a massive step in a good direction, as far as production costs. That's stuck in my head because of the state. So what's the availability on the rest of the ingredients? Kind of wondering silicon is the majority of the mass and weight that has to be moved around
NC silicon isn't in the wafers, it's in the crucible that holds the silicon.
@@johnrielley5003 And also: Globalization means every little piece of the final product comes from all over the globe. No one place is special or unique.
Until globalization collapses. Then it's all up for grabs...
@@johnrielley5003 The mine at Spruce Pine, NC produces the quartz needed for the process of making silicone wafers, not the silicone.
At 4NM you start having issues pushing actually electrons through. Also 4 NM does not always mean better. You get smaller but also require additional high level integrated chips.
Moving forcing TMSC is such an American move to not protect interests.
Can you cover the Bank of Canada rate cuts and the coming 25% US tariff on all goods from US. Will the Canadian dollar go lower than the Mexican Peso?
I didn't like your decision to delay your videos a week, I watch much less because of it. That said, I'm still watching. Excellent update on this vital capability the U.S. needs domestically. I know our armed forces are training for war with China, but I would hate to sacrifice so many Americans over Taiwan. Shutting down trade with China is the wisest thing to do. We can still buy low end products, but no more.
Good for America. We need all our chips made here.