As of now, we have a $20 off yearly subscriptions for Nebula if you sign up with my links: nebula.tv/techaltar Interview with Tim Culpan (Bloomberg): nebula.tv/videos/techaltar-tsmc-interview-tim-culpan-bloomberg Interview with Jon Y (Asianometry): nebula.tv/videos/techaltar-tsmc-interview-jon-asianometry Also fun fact, I now have a Nebula exclusive Class and a Podcast as well!
One point i think might have an impact in this space over the next 5 years is AI in chip design. Google has made some good headway in this, but soon we could see positive impact on costs from designs created by AI in a fraction of the time with a fraction of the resources, to be sent to fabs. I guess to be fair AI is a whole separate topic with a lot of potential to change everything once the arms race truly heats up. We could see exponential gains once more powerful/efficient chips get designed, implemented and then lead to more powerful AI designing even greater chips etc.
I think you're missing an important factor. It's not like Intel, Samsung, TSMC or Global just randomly makes engineering decisions. The key reason is TSMC got ahead is talent. The reason they stayed ahead is they grow the talent to make sure the talent pipeline stays full. Before Gelsinger, Intel had non-engineering executives who failed to nurture talent. One of the first things Gelsinger did when he returned was to bring back the old talent. You can get ASML machines, but without talented engineers you won't be able to deliver. If Intel wants to take back the lead, they'll need to make sure they setup a culture that nurtures younger engineers to fill the talent pipeline. Without that, Intel will fall behind again when the old guys retire again.
@@GeorgeWashingtonLaserMusket I moved to the US from Taiwan when I was a kid. One thing we (ie US) have problem with is our culture devalues certain types of jobs. As a result, professions like fabs and EE aren't considered as "sexy" as being in a startup. Working in a fab is grueling tough and requires a certain mindset. There are people willing to work those jobs, but not enough fill the demand. Until we fix the culture issue, I don't think we'll be able to take the lead back. There's also the anti-science sentiment across the US the last 2 decades. That isn't helping draw talent into the fab industry. As much as politicians want to give handouts to Intel, they aren't doing jack to fix the actual problem of declining public education. Who is gonna work the fabs? Are they going to ship engineers from Taiwan over to Arizona?
@@woolfel How can there be anti-science sentiment in the US when all the design is made there? the truth is no US company will ever be able to rival tsmc unless they get the same level of government subsidies as them.
@@maximme The US has already been working closely with TSMC and has been working with the Taiwanese government unofficially and not all that secretly for years. In fact, a part of US's defense budget and R&D is actually dedicated to that. There's definitely some vaults in the US that has all the TSMC's master blueprints, that can be opened in case of crisis. The problem for China is TSMC's factories are of no use without the experts who know how to run them, and, as Ukraine shows, they likely would not be standing any longer anyway if a war broke out. The CCP knows this, so threats are effectively more for power and negotiation. For the US, Taiwan is not just essential for technology, but geographic defense=China at the moment could be completely cut off nearly all resources, and have their export economy, which is actually highly dependent on imports. The island chains from the Malacca Straits all the way to Japan and Korea, China could be cut off from all shipping; bad economy is the quickest way for the CCP to have unrest and lose their power (as the cycle of Chinese history shows.) Nevertheless, who wants anyone to die? Diplomacy and negotiation is always in everyone's best interest.
The story of Morris Chang is impressive. His education, his work experience, then as a CEO, surpassed many of his contemporaries in the industry. It would be a shame if his greatest investment in the US, is scuttled by some politician's misguided vision of "Buy US made". Chang is, after all, one of the greatest engineers to have worked his entire career at Texas Instruments.
It's hard to describe just how value a video like this is to me. Thank you so much. This is exactly the content I live for. You condense so much valuable knowledge into something understandable for everyone. There are so many wrong assumptions and misinformation surrounding a company like TSMC, but you find a way to legitimacy inform people, through your unique understanding of both very tangible technology and the business world. I would so love for more content in the style of this video, for other technology companies - both big and small, but in particular those who are underestimated or highly complex.
The video is valuable because propaganda? It is anything but unbiased. Might as well be a mainstream media video about Chinese balloons being problematic.
Much appreciated video. If I may give a suggestion: A date of reference for the charts would help to give some perspective on the given information, specially on market-related figures. Because your analysis is quite good, when people visit this videos in the future, the original dates of the data help setting it on the current timeframe. Again, thank you for your content.
Alright, you managed to convince me, I'll go to nebula next week and get the subscription, because the interview should be highly educative and very interesting. It's just wonderful to see this very old channel stick to it's roots and always high quality content. Respectful salute to you Sir.
Nebula is the 200,000th most popular site on the internet. In Brazil. The average page visit lasts only one click, roughly 50k page views per day, with the average time spent on site being under five minutes. As much as I'd like to support small creators, this is just beating a dead horse. The future of all content is free.
@@patricko9479 Ad revenue would work fine if youtube stopped being a douche. Subscription services are dying all around right now, some never took off to begin with. Sponsors also work just fine, so long as they aren't sponsoring you to SELL THEIR product only. Though really, yeah I'd prefer it if we all just created for the sake of creation rather than for profit. I've given my art away for free for decades now, and everything I spend money on eventually belongs to something else also. It hasn't made my life any worse off in any way. Contrary to popular belief, you don't actually need money to survive, just for stuff.
Maybe I'm in the wrong here, but it would be better phrased as "TSMC hasn't made a bad decision that hurt them in the long run." They've been on a roll since 16 nm, but before that they invested in 32nm and scrapped it later due to not having enough buyers (Radeon HD 6000 series kinda sucked due to being stuck on 40 nm), and their late jump to FinFET meant that their 20 nm process still used planar technology, and was really bad in terms of leakage (Snapdragon 810 was very much a victim of that, also the reason for Nvidia keeping Maxwell on 28 nm).
TSMC is on a list of companies worldwide that must be protected at all costs, a consideration that China must take seriously. The Arizona Fab, which is under construction at the moment, may find problems at the initial start-up of production, this may take some time before all the glitches and snags have been ironed out, before full-scale production. Taking water from the already depleted aquifer may not provide suitably consistent quality. The osmosis process for returning the used water to the aquifer may have problems attaining a suitable standard of purity, considering the chemicals used in the preparation of the silicone discs.
Why are they building in Arizona? Do they rely on solar power or what? I imagine cooling is gonna consume much energy also. Can something similar to what happened with Texas' power grid and cuts happen in Arizona.
@@Koksn_Todorov Texas power grid is a mess. It doesn't get much help from FERC and as a result, it can't handle emergencies like what happened recently. It's a making of its own doing. I love Texas, but they have made some dumb decisions due to their horrible leadership in the past decades.
@@Koksn_Todorov Arizona's power supply comes from coal (28%), natural gas (24%), and nuclear (29%). You are right, cooling is going to be a problem. From memory. I think that Texas, unlike most other states was not interconnected to out-of-state power supplies. Unfortunately, politics plays a part in the decision to build in Arizona, which TSMC has to work with, considering the fact that the Biden government gave billions in subsidies to TSMC to build the fab in a less desirable location. The main reason to build in the US is that China may decide to invade Taiwan. The loss of ultra-small microchip manufacturing, if China invaded, would seriously affect the Western world economies. To overcome the loss of Taiwan a US fab would maintain the chip supply. Even on a mundane level, Apple would lose their entire supply of microchips. Basically, no other company in the world can make chips that small.
If compute stops being the limiting factor for performance and memory & I/O concerns take over (as is already the case in many AI applications), how would that affect TSMC? I’ve read that Intel has great photonics (for I/O), and other companies may have some sort of advantage as well.
That depends on where the bottleneck is. If there isn't enough memory per GPU/TPU, just add more. This provides more $ to memory manufacturers, but TSMC gets to keep most of their share. If the interconnect is slow, make a faster interconnect (like NVLink). Which probably depends on TSMC silicon. More disk throughput? Add more pcie lanes, and maybe add DPU's. Both probably good for TSMC. Often, though, the main effect of memory and IO limitations is that the ML pipeline that prepares batches for the GPU's/TPU's need to be streamlined. Reading from disk, cleaning, uncompressing and making the datasets available to the GPU/TPU precisely at the right time, at speeds that could be 100s of GB/s requires a lot of work on the software side, but not really any new hardware. Memory latency is helped by 3D stacking, as already mentioned. Longer term, though, Machine Learning may start depending on more memory integrated directly on the compute die (or compute on the memory die). That is going to be a real challenge, of course, since the processes that optimal for compute are not efficient for memory, and vice versa. I'm not sure how they will solve that without huge compromises. Maybe some kind of micro-chiplet-stacking? Or maybe some kind of hybrid lithography?
@@ThirdLife86 If I/O becomes the limiting factor for some application, then Nvidia or whoever might go to the manufacturer of the best I/O and accept a downgrade in compute. If TSMC isn't the best (or at least one of the best) manufacturer for I/O-bound tasks, they could get into trouble. Essentially, it's one (or several) of those important technology-bets. It will be interesting to see how it develops.
16:01 it will occur at some time, because there are physical limits on how small you can make structures. Some estimate it is at 2-3 nm, so we are not far from it. Density might be increased a bit further after that with chip stacking, but this will reach thermal limits at some point.
Around the 15:00 mark you talk about increase cost when shrinking nodes. It is great to hear 6nm is now cheaper than 7nm. Rdna2 refresh gpus should be great for the price.
Why should RDNA2 Refresh cards be cheaper ? AMD wants to increase margins just like any other company. I wouldn't expect a big difference here as its just approximately a 15% die shrink while still having the same performance and power constraints. And the "cheapness" comes from a lot of other factors like being able to reuse designs from N7, also cost saved that you won't see on the finished product. RDNA3 has been more cost optimized than RDNA2 ever was and what do you get ? More expensive cards and higher margins. I don't get why people always think companies will hand out gifts now that they save a tiny fraction on something.😂 Just calculate how much of an impact those 15% wafer area have - its next to fucking nothing. The 6950XT has a 520mm² chip, with a very low defect density of say .075 you can get approximately 71 chips out of a 9350 USD 12" wafer -> 131 USD per chip. Now with a 15% smaller die (452mm²) it would result in about 84 dies per wafer -> 111 USD. If AMD gave you all the savings your 699 USD card will now cost 679 USD. That is really GREAT for prices right ? 😆 Those 20 bucks really make that difference on 700 Dollar (ex 1k+ Dollar) Product. Doesn't change anything for customers and in reality they'd pocket the difference on manufacturing.
Very well presented. I loved every bit of it. As a huge tech fan, I don't know how I'm just discovering this channel and it seems to have been around for a while. New subscriber alert!!!
It will be super interesting to see where this whole thing with TSMC(and ASML , not mentioned in this video) goes with regards to the future of the semiconductor industry and the economics and politics surrounding it. Considering that these foundries keep bring newer nm process nodes online every few years it makes you wonder what they have access to but aren't releasing to consumers just yet
He does NOT move BACK to Taiwan. He moved TO Taiwan. Taiwan was a totally new place for him back then. He mentioned that in his interview with Computer History Museum. He is Chinese American who moved to US at the age of 18.
Great video, Im sitting here watching it on my 5nm M1Air which works like a portable machine always promised but failed to deliver. Truly all day battery life and powerful enough for everything short of video editing in 8K and is barely noticeable carrying it in a bag for the day, now if they can just put all the relevant ports back into them like they used to include so we can forget the dongle bad dream forever and I will be even more excited for a 3nm future.
@@msp5138Nothing about it is a toy, it doesn't even support good games. I bought it second hand for half new price. I bet you end up wasting more money than me on another garbage pc laptop before I need another machine.
I have a HP laptop. I dropped it out of my unzipped back pack 3 times from shoulder height. Still works great! Bought IPAD , fell out of tankbag . Broke screen Replace screen. Broke screen again. Scrap. HP makes good products!
Samsung is its main competitor, 3000 series Nvidia used Samsung, 4000 series use TSMC, If Samsung invests enough they could be competitive. A monopoly is not good.
It's not only business decisions. The US and others would want TSMC fabs on their soil or other leading edge fabs as a matter of national security. In a way I believe Taiwans protection insurance is TSMC. TSMC and the many other IT manufacturers like board manufacturers CAN be successful abroad, but is that really in the national interest of free Taiwan? Probably not. So in the end it is about what is more important money or democratic principles. Depends on the government. But you are going to see nations drop Taiwan like a stone once they have fab needs fullfilled. Sad but true.
Even without TSMC, Taiwan is still an important strategic ally for the US in countering China. What happened here was that they further raised the stakes that just ended up making China more likely to invade and the US more committed to it's defence. It's a delicate balancing act...
Well wrong ..... Taiwan has also a geographic key location to block China advance through the Pacific Ocean...and Taiwan Strait life line of Japan and South Korea
TSMC might be a one trick pony but Taiwan has a mature tech. infrastructure. They offer more than just chips. I can't imagine all chips being producted domestically. That would just cost the average customer way too much unless we are willing to pay lower wages or pay more for the product. And Taiwan is also in a strategic position in terms of military advantages. Taiwan isn't going to be dropped in any way lol
I think TSMC knows it won't be long before the average consumer will stop buying the latest chip because cost per performance have actually lower than previous gen, on par at best. The industry and professional sector will fill that gap, with the rise of AI and also Cloud gaming there will be plenty of business to be done. All they need is scaling it a little bit. That's why they're expanding their old fabs into US just like Samsung did. Long term demands for older chip will be there, since at this point most devices don't even need last two gen chip to perfectly function. Best tech made in their home turf while other countries got the old one with stable demands is pretty good decision.
One small correction: $20B a year is not that much money to build a fab, it's about 60 jumbo jets, with about 400 such jets produced a year. Or it's about the GDP of about 1M Americans, which is a mid-sized US city, typically just a suburb. Fairfax County VA for example has a GDP greater than this. But for a UA-cam video this is fine.
I think the bar chart at 1:19 is misleading. The bars are not in correct proportion to each other. A quick excel plot suggests that no bar does match with other bars. No big deal, but I somehow noticed it
7:12 Ive walked this silver highway thousands of times. Every time I walk down it, I am marveled by the amount of advancement our race has made in technology the past few decades.
In June 2021, TSMC started to provide N5A military-grade 5nm chip manufacturing service for high performance computing and high-end AI chips. That will enables the military chip designer of Taiwan, the US, Europe and Japan to apply and design high-tech AI weapons, or to upgrade the arms to be the AI enabled ones.
Morris Chang is not from Taiwan. He was actually born in China before WWII. Later he went to study in the US. Morris became a high-level manager before moving to Taiwan to run TSMC. He remains a US citizen.
@@Omer1996E.C No first Chinese then American. Has never gotten Taiwan (ROC) citizenship. Ironically Taiwanese trust him more than any other high-level figure in the country. Quite an extraordinary man.
@@JonProphet111 Do you still live at the bottom of the well? Hong Kong Peiying Primary School Chongqing Nankai Middle School Massachusetts Institute of Technology This is his education process
That is some serious brain power and performance results from TSMC!!! Absolutely brilliant!!! The reason why TSMC had the inverted manufacturing curve is because they CAN make chips less than 10 nanometers!!! There aren't too make other chip makers that can do that!!!
The TSMC Arizona fab will begin producing 5nm chips in 2024. Another fab is being constructed in Arizona to produce the 3nm chip in 2026. In the meantime, the successor to the 3nm chip will be made at a fab in Taiwan. So, the TSMC Arizona fab will not take over from their Taiwan fabs when it comes to the latest chips.
@@maximme lol, U.S. is not an authoritarian BS of a government like China, even an obvious spyware like Tiktok gets to run in U.S. market. Nice try, bot.
@@1ycan Doesn't really matter all that much - intel already has access to 4nm tech, the actual systemic efficiency is what you want to _steal_ from tsmc, so more than just the people that culture of the company as well
The only fab in Arizona capable of building 4nm chips, currently is Intel. The TSMC fab being built only has the capacity for 7nm chips, and the 4nm fabs for TSMC is still currently in Taiwan. Unless Apple is talking about sourcing chips from Intel (Intel's 4nm is up and running, and Apple hasn't named from WHO they were sourcing it from), or talking about building non-CPU chips (which doesn't need as advanced of a process node), it doesn't make sense for them to use a worse fabrication node currently.
This is a new start-up fintech advisory company and we are having hard time finding private investors because they think it will fail or it do not meet expectation. Above video is absolutely correct for tsmc history.
Most people will agree that most of the computing power that the newest Apple silicon pack goes unutilized for normal users. I think in the future we won't need powerful chips every single year. The need for a more powerful chip will be felt only after a gap of a few years. Also cloud computing etc will reduce the need for powerful chips in most devices.
There's no point to chips smaller than 10 nm except some specialty applications with super constrained space and power. For snything like a laptop or phone it makes no difference.
hmm, perhaps TSMC is comparable to AWS, they are both a near monopoly service company that extracts a significant proportion of the value from an entire industry, and both by providing access to a technology that their customers would maybe rather do in house, but would require a lot more capital investment, and both also command a generous margin by doing so!
Tech altar amazing mind blowing tech research analysis .Huge respect to your informative video. Eagerly waiting for your new video .Kudos to you & to your team.
1:15 they were different in every possible way. 2:00 they broke one of the common graphs in all business school. 2:20 the manufacturing is a low margins and lots of competitors. 2:45 TSMC's net income is higher than any of their clients: Apple, AMD, ... . 3:15 pulling further ahead from the competition. ... could their luck even run out? 4:50 the rest of the country did for other industries: contract manufacturing now this is called, a pure play foundry 5:20 key employees: frequently taiwanese american engineers. 5:35 he could take it from there and catch it up. 7:30 the business is a brutal treadmill that keeps getting faster with each generation. 8:10 design arm and manufacturing arm. 8:50 generous government backing. 9:10 hiring ground for engineers. 9:30 seen as trustworthy from the start as they don't compete with their clients. 9:55 hard forks in the road. 11:30 3 distinct scenarios that might hurt them. 1. politics: chip assets. export their model abroad. onshore their chip manufacturing. 2. messing up: no garantee 3. cost: 12:55 revenue from which part: N america.
The market is driven by cost, power cunsumption and speed. The idea of loss of cost advantage of smaller features is not a realistic view of the market.
When I built my first PC(not off the shield all in one) back in 2013, Intel was so dominant, and practically a monopoly. It was two years ahead of everryone in term of fabrication. Now Intel is more an also-run, thanks to those two bean counting CEOs before Pat Gelsinger. The chip industry evolves so fast. I won't bet my money on TSMC still the biggest mover in five years.
Intel was donminant only in one thing. Everyone only thinks about gaming, instead of all the practical devices we use everyday, especially smartphones. Everything has a chip in it. People are always surprised to realize that Intel hasn't been the dominant manufacturer for a long time because gaming is like the most least important part of computing.
@@MichaelSidneyTimpson You forget CPUs for big server farms and data centers. It's a huge market as well. Intel fails to develop smartphone SoC to compete with Qualcom. Which is its biggest failure. And of course, theu stuck with 14nm for centuries has no one but themselves to blame.
@@spektrumB Yeah, actually there is a CPU in everything now, even fridge's and cars, everything that used to mechanical in fact has a computer in it, and I don't mean the "smart" variety, but the plain old regular type. Many servers are using TSMC chips now. I only replied because I was surprised to realize all this too. Our Microwaves are more power than the computer in the Apollo landing. I remember Intel's SOC failures...I still had some of those tablets and netbooks, in fact. We also forget that the full dominance of x86 was really not for that long. IBMS's PowerPC cpu had been a major competitor (albeit maybe strong 2nd place) in the Server market through 90s and mid-2000s and other manufacturers, like Motorola before that (not just in Macs, as many probably think). Funny, I actually remember when a few cars started having microchips in them decades ago and we thought it was such a big deal, whereas it would be considered impossible now. Gaming is, of course, the first thing everyone thinks of, because it is the more obvious and immediate thing that we associate with CPU power, and certainly a heck of a lot more interesting to demonstrate in reviews (other than video editing, I suppose). Of course, most people in the general public didn't notice ARM's importance until Apple brought it to an obvious forefront (even though it had long been already, even with apple's portable devices.)
@@MichaelSidneyTimpson You keep mention gaming, which I couldn't care less. I think back in 2013, console gaming was bigger than PC. Yeah, everything has a CPU in it now, even a microwave oven and frig. But they mostly are ARM base 50 cents a piece, and no need for the latest from TSMC. When the pandemic hit two years ago, the car industry found themselves short of microchips. The media reported because the demand for PC was high(people work at home). That ate up the supply of microchips for the car industry. Which is 100% rubbish. Contracts to foundries are made quarters ago. If they found themselves run out of CPU, that means they don't book enough a year before. Moreover, those microchips used in cars aren't using the latest node from TSMC and Intel, unlike smartphone and PC. X86 is old enough for me. IBM's PowerPC is way before I know anything about PC.
The thing I don’t get is how TSMC is on top of everything with the latest nodes especially when ASML makes all the fabs. Like why doesn’t any other foundry get the latest fav technology. And if it’s just good engineering how does no other fab manufacturer poach employees.
This is a very interesting question. Few people know that manufacturing chips is not just a technology of using machines. A lot of scientific research must be carried out in the early stage to determine the structure of the transistor and the material of the insulating layer. The 2nm chip published by IBM belongs to this type of research. However, IBM does not have a fab. They transferred the research results to Samsung to assist Samsung in developing mass-production technology. At this stage, ASML's EUV is needed. In Taiwan, National Applied Research Laboratories (NARLabs) is responsible for the preliminary research work. NARLabs is a huge force hidden behind TSMC. In addition, there is also the Industrial Technology Research Institute (ITRI) that assists TSMC in developing equipment. ITRI overcomes all equipment obstacles for TSMC, such as X-ray measurement equipment used for 2nm GAA, or equipment used to detect nano-level suspended particles. The TSMC you see is just the top of the iceberg that floats out of the water, hidden behind TSMC Yes, it is Taiwan’s national power and private enterprise power. Even for the washing of dustproof clothes in clean rooms, there are professional private enterprises and TSMC for joint research and development. Even the United States cannot replicate this huge and sophisticated semiconductor ecology. It cannot be solved by purchasing equipment or poaching a few engineers. You can buy ice cubes, but you cannot buy icebergs.
News flash: Qualcomm and android makers with it already are avoiding 3nm for now since it’s too expensive. Apple drives full steam ahead towards 3nm, with all of their devices. Not a huge surprise with Apple being the biggest TSMC investor, and biggest client. So I don’t see your argument coming true for Apple stopping to use new fab process.
While apple is big in the states it's not number one across the world in every single country. Apple has a track record of using others tech until they decide to go in house and that could be the case here. Android services more consumers because Android is an open OS. Like iOS is and OS . Smartphone manufacturers tend to look at production vs returns and not all phone models can afford the increase in cost. 3mn will be used more when it is easier to make rather that be from TSMC finding was to make production more affordable or one of there competitors doing the same
I wouldn't say qcom are avoiding 3nm. Apple had almost exclusive 5nm capacity forn1~2 year before others begin adopting. Maybe apple is getting preferential treatment being TSMC's biggest client and leading edge node collaborator. Maybe apple is paying more for chip until yield and price get better for competitor. I think 3 nm will be a similar story where qcom/mediatek adopt the node 1~2 years later
tsmc adds a new fab, Fab 21, in US since most of their customers are in US. I only see expanding, diverse production, not moving. tsmc fabs expand at much bigger scale in Taiwan now.
Both TSMC and ASML were both co founded by Philips Electronics. Unfortunately for them they sold off all their shares and are worth just 14 billion today.
I know even in Europe they are trying to get fabs up and running from the mayor names. Intel, AMD,Nvidia etc. It might even be more stressful for ASML because transport would probably be 100% via road. Where when you ship to Taiwan/US it's freight plane for 90% of the ride. Some people will sweat only at the thought of this. Road transport is brutal
15:00 Now that A.I. is ramping up, there will be another 30+ years of high demand for high-end chips. Tesla is already working with TSMC to have them make Tesla's Dojo A.I. chips and so is Nvidia for theirs.
I was going to watch this on nebula but I haven't really used that platform ever since I subscribed. I've used it for a total of 5 hours sinc I bought it in December, for context my daily average for UA-cam is 3 hours 🤦
As of now, we have a $20 off yearly subscriptions for Nebula if you sign up with my links: nebula.tv/techaltar
Interview with Tim Culpan (Bloomberg): nebula.tv/videos/techaltar-tsmc-interview-tim-culpan-bloomberg
Interview with Jon Y (Asianometry): nebula.tv/videos/techaltar-tsmc-interview-jon-asianometry
Also fun fact, I now have a Nebula exclusive Class and a Podcast as well!
One point i think might have an impact in this space over the next 5 years is AI in chip design. Google has made some good headway in this, but soon we could see positive impact on costs from designs created by AI in a fraction of the time with a fraction of the resources, to be sent to fabs. I guess to be fair AI is a whole separate topic with a lot of potential to change everything once the arms race truly heats up. We could see exponential gains once more powerful/efficient chips get designed, implemented and then lead to more powerful AI designing even greater chips etc.
Everything is bad about Asian according to wh!73 europeans is it?
I think you're missing an important factor. It's not like Intel, Samsung, TSMC or Global just randomly makes engineering decisions. The key reason is TSMC got ahead is talent. The reason they stayed ahead is they grow the talent to make sure the talent pipeline stays full. Before Gelsinger, Intel had non-engineering executives who failed to nurture talent. One of the first things Gelsinger did when he returned was to bring back the old talent. You can get ASML machines, but without talented engineers you won't be able to deliver.
If Intel wants to take back the lead, they'll need to make sure they setup a culture that nurtures younger engineers to fill the talent pipeline. Without that, Intel will fall behind again when the old guys retire again.
@@GeorgeWashingtonLaserMusket I moved to the US from Taiwan when I was a kid. One thing we (ie US) have problem with is our culture devalues certain types of jobs. As a result, professions like fabs and EE aren't considered as "sexy" as being in a startup. Working in a fab is grueling tough and requires a certain mindset. There are people willing to work those jobs, but not enough fill the demand. Until we fix the culture issue, I don't think we'll be able to take the lead back.
There's also the anti-science sentiment across the US the last 2 decades. That isn't helping draw talent into the fab industry. As much as politicians want to give handouts to Intel, they aren't doing jack to fix the actual problem of declining public education. Who is gonna work the fabs? Are they going to ship engineers from Taiwan over to Arizona?
@@woolfel America is full of dumb conspiracy theorists.
@@TykoBrian7 I wish that wasn't the case, but sadly it's true.
@@woolfel How can there be anti-science sentiment in the US when all the design is made there? the truth is no US company will ever be able to rival tsmc unless they get the same level of government subsidies as them.
@@GeorgeWashingtonLaserMusket Thank you for the daily cringe
morris is an elderly person who is respected by all Taiwanese.Taiwanese people are very grateful to him.
Because he’s simply a genius.
US will "nationalise" the factory and its tech by saying its NATIONAL SECURITY interest is at stake.
TSMC will not have any option.
@@maximme The US has already been working closely with TSMC and has been working with the Taiwanese government unofficially and not all that secretly for years. In fact, a part of US's defense budget and R&D is actually dedicated to that. There's definitely some vaults in the US that has all the TSMC's master blueprints, that can be opened in case of crisis. The problem for China is TSMC's factories are of no use without the experts who know how to run them, and, as Ukraine shows, they likely would not be standing any longer anyway if a war broke out. The CCP knows this, so threats are effectively more for power and negotiation. For the US, Taiwan is not just essential for technology, but geographic defense=China at the moment could be completely cut off nearly all resources, and have their export economy, which is actually highly dependent on imports. The island chains from the Malacca Straits all the way to Japan and Korea, China could be cut off from all shipping; bad economy is the quickest way for the CCP to have unrest and lose their power (as the cycle of Chinese history shows.) Nevertheless, who wants anyone to die? Diplomacy and negotiation is always in everyone's best interest.
@@maximme You're confusing China with the US.
@@ThePowerLover lol US is the aggressor here not China you blind fool
The story of Morris Chang is impressive. His education, his work experience, then as a CEO, surpassed many of his contemporaries in the industry. It would be a shame if his greatest investment in the US, is scuttled by some politician's misguided vision of "Buy US made". Chang is, after all, one of the greatest engineers to have worked his entire career at Texas Instruments.
It's hard to describe just how value a video like this is to me. Thank you so much. This is exactly the content I live for. You condense so much valuable knowledge into something understandable for everyone. There are so many wrong assumptions and misinformation surrounding a company like TSMC, but you find a way to legitimacy inform people, through your unique understanding of both very tangible technology and the business world.
I would so love for more content in the style of this video, for other technology companies - both big and small, but in particular those who are underestimated or highly complex.
i support free speech and all but your pfp should be illegal
The video is valuable because propaganda? It is anything but unbiased. Might as well be a mainstream media video about Chinese balloons being problematic.
Much appreciated video. If I may give a suggestion: A date of reference for the charts would help to give some perspective on the given information, specially on market-related figures. Because your analysis is quite good, when people visit this videos in the future, the original dates of the data help setting it on the current timeframe.
Again, thank you for your content.
Alright, you managed to convince me, I'll go to nebula next week and get the subscription, because the interview should be highly educative and very interesting.
It's just wonderful to see this very old channel stick to it's roots and always high quality content. Respectful salute to you Sir.
Happy to hear!
Nebula is the 200,000th most popular site on the internet.
In Brazil.
The average page visit lasts only one click, roughly 50k page views per day, with the average time spent on site being under five minutes.
As much as I'd like to support small creators, this is just beating a dead horse.
The future of all content is free.
@@jeffbrownstain And how do you think creators should be payed? Not at all?
@@patricko9479 Ad revenue would work fine if youtube stopped being a douche. Subscription services are dying all around right now, some never took off to begin with.
Sponsors also work just fine, so long as they aren't sponsoring you to SELL THEIR product only.
Though really, yeah I'd prefer it if we all just created for the sake of creation rather than for profit.
I've given my art away for free for decades now, and everything I spend money on eventually belongs to something else also.
It hasn't made my life any worse off in any way.
Contrary to popular belief, you don't actually need money to survive, just for stuff.
@@jeffbrownstain Right we only need air no need for mooney.
USA does have big reason to protect Taiwan at all cost. The whole world is almost revovled around by TSMC and ASML.
Dont forget military tech too.
You're delusional. Taiwan is and was part of China. The US is trouble maker, stupid politicians
@@Anomize23Lol can you imagine how much fewer Palestinians civilians would die if military chips would stop being produced?
USA has no right to do that
Maybe I'm in the wrong here, but it would be better phrased as "TSMC hasn't made a bad decision that hurt them in the long run." They've been on a roll since 16 nm, but before that they invested in 32nm and scrapped it later due to not having enough buyers (Radeon HD 6000 series kinda sucked due to being stuck on 40 nm), and their late jump to FinFET meant that their 20 nm process still used planar technology, and was really bad in terms of leakage (Snapdragon 810 was very much a victim of that, also the reason for Nvidia keeping Maxwell on 28 nm).
AMD Bulldozer was also on 32nm. Well it didnt sell though :D
So did these have less of an effect because they weren't as important or because TSMC handled their mistakes better?
I wish I had channels like these in my youth to get me interested in topics that seemed so 'mundane' to me when I was studying.
I just want the best for tsmc’s ceo, and the best for Taiwan 🇹🇼
TSMC is on a list of companies worldwide that must be protected at all costs, a consideration that China must take seriously. The Arizona Fab, which is under construction at the moment, may find problems at the initial start-up of production, this may take some time before all the glitches and snags have been ironed out, before full-scale production. Taking water from the already depleted aquifer may not provide suitably consistent quality. The osmosis process for returning the used water to the aquifer may have problems attaining a suitable standard of purity, considering the chemicals used in the preparation of the silicone discs.
They say water is 99% treated and recycled back to the plant.
Why are they building in Arizona? Do they rely on solar power or what? I imagine cooling is gonna consume much energy also. Can something similar to what happened with Texas' power grid and cuts happen in Arizona.
@@Koksn_Todorov Texas power grid is a mess. It doesn't get much help from FERC and as a result, it can't handle emergencies like what happened recently. It's a making of its own doing. I love Texas, but they have made some dumb decisions due to their horrible leadership in the past decades.
@@Koksn_Todorov Arizona's power supply comes from coal (28%), natural gas (24%), and nuclear (29%). You are right, cooling is going to be a problem. From memory. I think that Texas, unlike most other states was not interconnected to out-of-state power supplies.
Unfortunately, politics plays a part in the decision to build in Arizona, which TSMC has to work with, considering the fact that the Biden government gave billions in subsidies to TSMC to build the fab in a less desirable location.
The main reason to build in the US is that China may decide to invade Taiwan. The loss of ultra-small microchip manufacturing, if China invaded, would seriously affect the Western world economies. To overcome the loss of Taiwan a US fab would maintain the chip supply. Even on a mundane level, Apple would lose their entire supply of microchips.
Basically, no other company in the world can make chips that small.
If compute stops being the limiting factor for performance and memory & I/O concerns take over (as is already the case in many AI applications), how would that affect TSMC?
I’ve read that Intel has great photonics (for I/O), and other companies may have some sort of advantage as well.
3d stacking is one way to introduce more interconnects
TSMC is also working on packaging. Look at the chiplets stuff that AMD is doing, for instance - that's done in TSMC fabs.
Why would it affect TSMC ? You still need a chip to compute regardless of your other architectural issues like Memory.
That depends on where the bottleneck is. If there isn't enough memory per GPU/TPU, just add more. This provides more $ to memory manufacturers, but TSMC gets to keep most of their share.
If the interconnect is slow, make a faster interconnect (like NVLink). Which probably depends on TSMC silicon. More disk throughput? Add more pcie lanes, and maybe add DPU's. Both probably good for TSMC.
Often, though, the main effect of memory and IO limitations is that the ML pipeline that prepares batches for the GPU's/TPU's need to be streamlined. Reading from disk, cleaning, uncompressing and making the datasets available to the GPU/TPU precisely at the right time, at speeds that could be 100s of GB/s requires a lot of work on the software side, but not really any new hardware.
Memory latency is helped by 3D stacking, as already mentioned. Longer term, though, Machine Learning may start depending on more memory integrated directly on the compute die (or compute on the memory die). That is going to be a real challenge, of course, since the processes that optimal for compute are not efficient for memory, and vice versa.
I'm not sure how they will solve that without huge compromises. Maybe some kind of micro-chiplet-stacking? Or maybe some kind of hybrid lithography?
@@ThirdLife86 If I/O becomes the limiting factor for some application, then Nvidia or whoever might go to the manufacturer of the best I/O and accept a downgrade in compute. If TSMC isn't the best (or at least one of the best) manufacturer for I/O-bound tasks, they could get into trouble.
Essentially, it's one (or several) of those important technology-bets. It will be interesting to see how it develops.
16:01 it will occur at some time, because there are physical limits on how small you can make structures. Some estimate it is at 2-3 nm, so we are not far from it. Density might be increased a bit further after that with chip stacking, but this will reach thermal limits at some point.
Thanks for this! Was looking all over for info on this, and nobody did it like you do!
Don’t forget that the head of nVidia and AMD are Taiwanese as well， the engineering foundation for the island is incredible
AMD and the spin-off foundry business Global Foundries are not Taiwanese.
@James Sun Nvidia and AAMD CEOs were raised in America and are apparently related to each other haha
Awesome, highly informative, and educational as usual!
Going sub nanometer is the greatest roadblock ahead. Glad you mentioned it.
Who would've known that TSMC might have turned that smile curve into a Frown curve!
1000 IQ move by TSMC
6:43 is that Fab 42???? I recognize that bay, it looks like the upper 800s. Ah I miss 42
Just found your channel. Great stuff. Thanks for the video!
Great content! This video really provided a lot more clarity on why Fabs are so important....
I don't know why, but I find the short intro music and lighting to your videos very cool
The graph shown at the 2:30 is so subtle yet fruitful, the statistics and representation is just next level.
Around the 15:00 mark you talk about increase cost when shrinking nodes. It is great to hear 6nm is now cheaper than 7nm. Rdna2 refresh gpus should be great for the price.
Why should RDNA2 Refresh cards be cheaper ? AMD wants to increase margins just like any other company. I wouldn't expect a big difference here as its just approximately a 15% die shrink while still having the same performance and power constraints.
And the "cheapness" comes from a lot of other factors like being able to reuse designs from N7, also cost saved that you won't see on the finished product. RDNA3 has been more cost optimized than RDNA2 ever was and what do you get ? More expensive cards and higher margins.
I don't get why people always think companies will hand out gifts now that they save a tiny fraction on something.😂
Just calculate how much of an impact those 15% wafer area have - its next to fucking nothing. The 6950XT has a 520mm² chip, with a very low defect density of say .075 you can get approximately 71 chips out of a 9350 USD 12" wafer -> 131 USD per chip. Now with a 15% smaller die (452mm²) it would result in about 84 dies per wafer -> 111 USD. If AMD gave you all the savings your 699 USD card will now cost 679 USD.
That is really GREAT for prices right ? 😆 Those 20 bucks really make that difference on 700 Dollar (ex 1k+ Dollar) Product.
Doesn't change anything for customers and in reality they'd pocket the difference on manufacturing.
Your mic setup has very clear audio. Do you mind sharing the brand or the buy link ?? thanks.👍
Very well presented. I loved every bit of it. As a huge tech fan, I don't know how I'm just discovering this channel and it seems to have been around for a while. New subscriber alert!!!
Great overview. Clear, concise and provides link to more information. Thanks
It will be super interesting to see where this whole thing with TSMC(and ASML , not mentioned in this video) goes with regards to the future of the semiconductor industry and the economics and politics surrounding it. Considering that these foundries keep bring newer nm process nodes online every few years it makes you wonder what they have access to but aren't releasing to consumers just yet
Beautiful video, very informative. Thank you!!!
Glad you enjoyed it!
@@TechAltar Fantastic Video
This episode was very educational. Thank you!
I just recently stumbled upon your channel. I really enjoy your saturday chillout podcast. I like both your voices, they are nice to listen to
The guy who founded TSMC really helped taiwan keeping Taiwanese independence by starting that sillicon sheild.
He does NOT move BACK to Taiwan. He moved TO Taiwan. Taiwan was a totally new place for him back then. He mentioned that in his interview with Computer History Museum. He is Chinese American who moved to US at the age of 18.
Great video, Im sitting here watching it on my 5nm M1Air which works like a portable machine always promised but failed to deliver. Truly all day battery life and powerful enough for everything short of video editing in 8K and is barely noticeable carrying it in a bag for the day, now if they can just put all the relevant ports back into them like they used to include so we can forget the dongle bad dream forever and I will be even more excited for a 3nm future.
So basically you wasted your money on a overpriced toy.
@@msp5138Nothing about it is a toy, it doesn't even support good games. I bought it second hand for half new price. I bet you end up wasting more money than me on another garbage pc laptop before I need another machine.
I have a HP laptop. I dropped it out of my unzipped back pack 3 times from shoulder height.
Still works great!
Bought IPAD , fell out of tankbag .
Broke screen
Replace screen.
Broke screen again.
Scrap.
HP makes good products!
Samsung is its main competitor, 3000 series Nvidia used Samsung, 4000 series use TSMC, If Samsung invests enough they could be competitive. A monopoly is not good.
It's not only business decisions. The US and others would want TSMC fabs on their soil or other leading edge fabs as a matter of national security. In a way I believe Taiwans protection insurance is TSMC. TSMC and the many other IT manufacturers like board manufacturers CAN be successful abroad, but is that really in the national interest of free Taiwan? Probably not. So in the end it is about what is more important money or democratic principles. Depends on the government. But you are going to see nations drop Taiwan like a stone once they have fab needs fullfilled. Sad but true.
very true....i dont understand why TSCM is giving up so easily, they are literally irreplaceable! 92%market share on top nodes its de facto monopoly
Even without TSMC, Taiwan is still an important strategic ally for the US in countering China. What happened here was that they further raised the stakes that just ended up making China more likely to invade and the US more committed to it's defence. It's a delicate balancing act...
Well wrong ..... Taiwan has also a geographic key location to block China advance through the Pacific Ocean...and Taiwan Strait life line of Japan and South Korea
TSMC might be a one trick pony but Taiwan has a mature tech. infrastructure. They offer more than just chips. I can't imagine all chips being producted domestically. That would just cost the average customer way too much unless we are willing to pay lower wages or pay more for the product. And Taiwan is also in a strategic position in terms of military advantages. Taiwan isn't going to be dropped in any way lol
TSMC is co-owner of those factories. They will have influence through them.
I think TSMC knows it won't be long before the average consumer will stop buying the latest chip because cost per performance have actually lower than previous gen, on par at best.
The industry and professional sector will fill that gap, with the rise of AI and also Cloud gaming there will be plenty of business to be done.
All they need is scaling it a little bit. That's why they're expanding their old fabs into US just like Samsung did. Long term demands for older chip will be there, since at this point most devices don't even need last two gen chip to perfectly function.
Best tech made in their home turf while other countries got the old one with stable demands is pretty good decision.
This is a fantastic video. Thank you!!
2:29 the curve would be wrong if you look at how much TSMC earns from R&D and design services.
16:02 due to co-design, a price sensitive customer won't redesign a low end product just to get the same performance elsewhere
Always like this kind of deep dive explanation video like this🙂. Thank you for creating it👍🙏
One small correction: $20B a year is not that much money to build a fab, it's about 60 jumbo jets, with about 400 such jets produced a year. Or it's about the GDP of about 1M Americans, which is a mid-sized US city, typically just a suburb. Fairfax County VA for example has a GDP greater than this. But for a UA-cam video this is fine.
That's insane amount of money - equivalent to cost of huge nuclear power plant ( Barakah ) or more than Ford class carrier.
As long as you are way ahead of your enemy and keep on moving forward while they play catching up.
I think the bar chart at 1:19 is misleading. The bars are not in correct proportion to each other. A quick excel plot suggests that no bar does match with other bars. No big deal, but I somehow noticed it
AMD CEO: Real Men have fabs
Lisa Su: bet!
ASML, NXP Semiconductors, and STMicroelectronics lead the Dutch chip industry and are also major players in the global arena.
7:12 Ive walked this silver highway thousands of times. Every time I walk down it, I am marveled by the amount of advancement our race has made in technology the past few decades.
ASIANOMETRY!! One of the best channels on this platform, hands down.
Would you also consider making a video about fabs and the markets around other semiconductor technologies like GaN or SiGe?
Where is SiGe used? I already own some GaN PSUs.
Transistors are close to their end. New tech is needed. This means that depending on who comes up with it may lead the next cycle.
What is theorized to be next?
TSMC didn't break the smiling curve, their biggest advantage is actually R&D
In June 2021, TSMC started to provide N5A military-grade 5nm chip manufacturing service for high performance computing and high-end AI chips. That will enables the military chip designer of Taiwan, the US, Europe and Japan to apply and design high-tech AI weapons, or to upgrade the arms to be the AI enabled ones.
Next do ASML please
Morris Chang is not from Taiwan. He was actually born in China before WWII. Later he went to study in the US. Morris became a high-level manager before moving to Taiwan to run TSMC. He remains a US citizen.
Intereresting considering the question of US defence of Taiwan
He's Chinese, then Taiwanese, then American?
@@Omer1996E.C No first Chinese then American. Has never gotten Taiwan (ROC) citizenship. Ironically Taiwanese trust him more than any other high-level figure in the country. Quite an extraordinary man.
@@JonProphet111 Do you still live at the bottom of the well? Hong Kong Peiying Primary School Chongqing Nankai Middle School Massachusetts Institute of Technology This is his education process
@@Omer1996E.C In his autobiography, although holding a U.S. passport, he was still a Chinese
China exist:
USA exist:
TSMC: *akwardly sweating*
This felt like a high production value Asianometry video. *checks description* Yeah, makes sense.
Great analysis, I've enjoyed listening.
That is some serious brain power and performance results from TSMC!!! Absolutely brilliant!!! The reason why TSMC had the inverted manufacturing curve is because they CAN make chips less than 10 nanometers!!! There aren't too make other chip makers that can do that!!!
isnt basically every conflict between two or more humans an armed conflict, considering how most humans have two arms?
The TSMC Arizona fab will begin producing 5nm chips in 2024. Another fab is being constructed in Arizona to produce the 3nm chip in 2026. In the meantime, the successor to the 3nm chip will be made at a fab in Taiwan. So, the TSMC Arizona fab will not take over from their Taiwan fabs when it comes to the latest chips.
US will "nationalise" the factory and its tech by saying its NATIONAL SECURITY interest is at stake.
TSMC will not have any option.
@@maximme That is certainly a possibility. But the Arizona fab will not be producing the smallest chips. Those will be produced in Taiwan.
@@maximme lol, U.S. is not an authoritarian BS of a government like China, even an obvious spyware like Tiktok gets to run in U.S. market. Nice try, bot.
@@1ycan Doesn't really matter all that much - intel already has access to 4nm tech, the actual systemic efficiency is what you want to _steal_ from tsmc, so more than just the people that culture of the company as well
The only fab in Arizona capable of building 4nm chips, currently is Intel. The TSMC fab being built only has the capacity for 7nm chips, and the 4nm fabs for TSMC is still currently in Taiwan. Unless Apple is talking about sourcing chips from Intel (Intel's 4nm is up and running, and Apple hasn't named from WHO they were sourcing it from), or talking about building non-CPU chips (which doesn't need as advanced of a process node), it doesn't make sense for them to use a worse fabrication node currently.
Great video man, but that dirty lens effect is just so annoying. Who thought that was good idea?
The phrase was said by Jerry Sander, not Lisa Su whom photo you showed
thanks for such great video you made to bring it to us
This is a new start-up fintech advisory company and we are having hard time finding private investors because they think it will fail or it do not meet expectation. Above video is absolutely correct for tsmc history.
Our company is called noscam initiative
@@NOSCAM Totally not a scam 💀
As usual, another great vid!
great video as always! would it be possible to share the source of the charts shown in 15:07? i m very eager to learn more about these trends, thx!
Most people will agree that most of the computing power that the newest Apple silicon pack goes unutilized for normal users. I think in the future we won't need powerful chips every single year. The need for a more powerful chip will be felt only after a gap of a few years. Also cloud computing etc will reduce the need for powerful chips in most devices.
Finally someone with half a brain...
I don't agree it.
one word....... ChatGPT
There's no point to chips smaller than 10 nm except some specialty applications with super constrained space and power. For snything like a laptop or phone it makes no difference.
How about asml, photomask providers, Carl zeiss and some other tools?
16:10 The magical solution to more advanced chips for less cost will be the AIs taking over production and design of the chips.
hmm, perhaps TSMC is comparable to AWS, they are both a near monopoly service company that extracts a significant proportion of the value from an entire industry, and both by providing access to a technology that their customers would maybe rather do in house, but would require a lot more capital investment, and both also command a generous margin by doing so!
TSMC has high margin because they basically the only game in town of scale. Specialization is key to profitability
Tech altar amazing mind blowing tech research analysis .Huge respect to your informative video. Eagerly waiting for your new video .Kudos to you & to your team.
Very interesting video, thanks!
1:15 they were different in every possible way.
2:00 they broke one of the common graphs in all business school.
2:20 the manufacturing is a low margins and lots of competitors.
2:45 TSMC's net income is higher than any of their clients: Apple, AMD, ... .
3:15 pulling further ahead from the competition. ... could their luck even run out?
4:50 the rest of the country did for other industries: contract manufacturing
now this is called, a pure play foundry
5:20 key employees: frequently taiwanese american engineers.
5:35 he could take it from there and catch it up.
7:30 the business is a brutal treadmill that keeps getting faster with each generation.
8:10 design arm and manufacturing arm.
8:50 generous government backing.
9:10 hiring ground for engineers.
9:30 seen as trustworthy from the start as they don't compete with their clients.
9:55 hard forks in the road.
11:30 3 distinct scenarios that might hurt them.
1. politics: chip assets. export their model abroad. onshore their chip manufacturing.
2. messing up: no garantee
3. cost:
12:55 revenue from which part: N america.
New thumbnail got my attention
Wow. Thank you. Great video. Helps explain why Buffett is making an investment in this company. Maybe me too.
Well, an iteresting video to watch. Good day from Taiwan
The market is driven by cost, power cunsumption and speed. The idea of loss of cost advantage of smaller features is not a realistic view of the market.
When I built my first PC(not off the shield all in one) back in 2013, Intel was so dominant, and practically a monopoly. It was two years ahead of everryone in term of fabrication. Now Intel is more an also-run, thanks to those two bean counting CEOs before Pat Gelsinger. The chip industry evolves so fast. I won't bet my money on TSMC still the biggest mover in five years.
Intel was donminant only in one thing. Everyone only thinks about gaming, instead of all the practical devices we use everyday, especially smartphones. Everything has a chip in it. People are always surprised to realize that Intel hasn't been the dominant manufacturer for a long time because gaming is like the most least important part of computing.
@@MichaelSidneyTimpson You forget CPUs for big server farms and data centers. It's a huge market as well.
Intel fails to develop smartphone SoC to compete with Qualcom. Which is its biggest failure. And of course, theu stuck with 14nm for centuries has no one but themselves to blame.
@@spektrumB Yeah, actually there is a CPU in everything now, even fridge's and cars, everything that used to mechanical in fact has a computer in it, and I don't mean the "smart" variety, but the plain old regular type. Many servers are using TSMC chips now. I only replied because I was surprised to realize all this too. Our Microwaves are more power than the computer in the Apollo landing. I remember Intel's SOC failures...I still had some of those tablets and netbooks, in fact. We also forget that the full dominance of x86 was really not for that long. IBMS's PowerPC cpu had been a major competitor (albeit maybe strong 2nd place) in the Server market through 90s and mid-2000s and other manufacturers, like Motorola before that (not just in Macs, as many probably think). Funny, I actually remember when a few cars started having microchips in them decades ago and we thought it was such a big deal, whereas it would be considered impossible now. Gaming is, of course, the first thing everyone thinks of, because it is the more obvious and immediate thing that we associate with CPU power, and certainly a heck of a lot more interesting to demonstrate in reviews (other than video editing, I suppose). Of course, most people in the general public didn't notice ARM's importance until Apple brought it to an obvious forefront (even though it had long been already, even with apple's portable devices.)
@@MichaelSidneyTimpson You keep mention gaming, which I couldn't care less. I think back in 2013, console gaming was bigger than PC. Yeah, everything has a CPU in it now, even a microwave oven and frig. But they mostly are ARM base 50 cents a piece, and no need for the latest from TSMC. When the pandemic hit two years ago, the car industry found themselves short of microchips. The media reported because the demand for PC was high(people work at home). That ate up the supply of microchips for the car industry. Which is 100% rubbish. Contracts to foundries are made quarters ago. If they found themselves run out of CPU, that means they don't book enough a year before. Moreover, those microchips used in cars aren't using the latest node from TSMC and Intel, unlike smartphone and PC.
X86 is old enough for me. IBM's PowerPC is way before I know anything about PC.
@@spektrumB thanks, interesting
Not only FAANG but also dozens of software companies took most of the top talents from Intel. It's the trend that Pat will not be able to change.
The thing I don’t get is how TSMC is on top of everything with the latest nodes especially when ASML makes all the fabs. Like why doesn’t any other foundry get the latest fav technology. And if it’s just good engineering how does no other fab manufacturer poach employees.
This is a very interesting question.
Few people know that manufacturing chips is not just a technology of using machines. A lot of scientific research must be carried out in the early stage to determine the structure of the transistor and the material of the insulating layer. The 2nm chip published by IBM belongs to this type of research.
However, IBM does not have a fab. They transferred the research results to Samsung to assist Samsung in developing mass-production technology. At this stage, ASML's EUV is needed.
In Taiwan, National Applied Research Laboratories (NARLabs) is responsible for the preliminary research work. NARLabs is a huge force hidden behind TSMC. In addition, there is also the Industrial Technology Research Institute (ITRI) that assists TSMC in developing equipment.
ITRI overcomes all equipment obstacles for TSMC, such as X-ray measurement equipment used for 2nm GAA, or equipment used to detect nano-level suspended particles. The TSMC you see is just the top of the iceberg that floats out of the water, hidden behind TSMC Yes, it is Taiwan’s national power and private enterprise power. Even for the washing of dustproof clothes in clean rooms, there are professional private enterprises and TSMC for joint research and development. Even the United States cannot replicate this huge and sophisticated semiconductor ecology. It cannot be solved by purchasing equipment or poaching a few engineers.
You can buy ice cubes, but you cannot buy icebergs.
Hold tight, since 97', Mediatek is catching up fast
Far far behind
if you are all right, then why is Buffet taking back his $ 5 bn bet on tsmc
apple already skipped whole series soc upgrades. they only upgraded the soc on the 14 pro max from last yr
Apple botched their ray tracing design, and delayed 3nm until this is fixed.
Do you have the permission to copy and upload Asianometry's video to start with?
Great content as usual
I like the monopoly perspective on global value chain smiles😂
TSMC is a case of too many critical eggs being concentrated in a single basket.
News flash: Qualcomm and android makers with it already are avoiding 3nm for now since it’s too expensive. Apple drives full steam ahead towards 3nm, with all of their devices.
Not a huge surprise with Apple being the biggest TSMC investor, and biggest client. So I don’t see your argument coming true for Apple stopping to use new fab process.
While apple is big in the states it's not number one across the world in every single country. Apple has a track record of using others tech until they decide to go in house and that could be the case here. Android services more consumers because Android is an open OS. Like iOS is and OS . Smartphone manufacturers tend to look at production vs returns and not all phone models can afford the increase in cost. 3mn will be used more when it is easier to make rather that be from TSMC finding was to make production more affordable or one of there competitors doing the same
I wouldn't say qcom are avoiding 3nm. Apple had almost exclusive 5nm capacity forn1~2 year before others begin adopting.
Maybe apple is getting preferential treatment being TSMC's biggest client and leading edge node collaborator. Maybe apple is paying more for chip until yield and price get better for competitor. I think 3 nm will be a similar story where qcom/mediatek adopt the node 1~2 years later
@@doctorwilly
Becos Samsung failed in their poor yield for 5nm chip.
Interesting Apple gets mentioned in the profit curve since they also get their stuff produced at TSMC :D
The opposite of Apple? Value adding over marketing fashion
Great analyze, but my opinion, we'll see TSMC slowly fall and slowdown after moving to US and make in US...
tsmc adds a new fab, Fab 21, in US since most of their customers are in US. I only see expanding, diverse production, not moving. tsmc fabs expand at much bigger scale in Taiwan now.
Both TSMC and ASML were both co founded by Philips Electronics. Unfortunately for them they sold off all their shares and are worth just 14 billion today.
“just 14Bs” 😂
@Crocuta Phillips also used to own what became NXP semiconductor
I know even in Europe they are trying to get fabs up and running from the mayor names.
Intel, AMD,Nvidia etc.
It might even be more stressful for ASML because transport would probably be 100% via road. Where when you ship to Taiwan/US it's freight plane for 90% of the ride.
Some people will sweat only at the thought of this. Road transport is brutal
AMD, Nvidia have no fabs. They use TSMC.
@@ttb1513 and they would use the EU counterpart here. You knew exactly what i was meaning. Point is there will be fabs for them in the EU and USA
@@ttb1513 GOOD POINT,
even after watching the video, some people still don't have a clue...
Rivers, trains and planes exist.
mayors of which cities?
15:00 Now that A.I. is ramping up, there will be another 30+ years of high demand for high-end chips. Tesla is already working with TSMC to have them make Tesla's Dojo A.I. chips and so is Nvidia for theirs.
2:40 It's almost like economics textbooks are wrong 🤔
Nothing lasts forever, my man.
I was going to watch this on nebula but I haven't really used that platform ever since I subscribed. I've used it for a total of 5 hours sinc I bought it in December, for context my daily average for UA-cam is 3 hours 🤦