TSMC Outlines 2nm Plans: N2P Brings Backside Power Delivery in 2026, N2X Added | Hacker News Reader

TSMC Outlines 2nm Plans: N2P Brings Backside Power Delivery in 2026, N2X Added

anandtech.com

83 pointsrbanffy67 comments

America and Europe need to start making these kinds of advanced chips themselves as soon as possible. Taiwan is now the only place where this is possible.

If we have to go back to 4 or 5nm processors for a while, it won't be the end of the world.

Europe doesn't have 4 and 5 nm fabs on its soil nor European owned fabs aboard. GloFo Dresden goes down to 22nm.

Smallest probably 7nm Intel in Ireland. Not sure it is online yet though

Intel Ireland (Fab 24/34) will be producing Intel 4 and 3, which is an EUV based node

So potentially they could even go to finer architectures?

Intel has stopped using "nm" - it's meaningless anyway; hasn't been meaningful since 1994 (https://en.wikipedia.org/wiki/Semiconductor_device_fabricati...).

Now it's just Intel 7 (currently in use), Intel 4 (successor to Intel 7 for internal use) and Intel 3 (based off Intel 4 for their foundry services).

Intels Fab 24 in Ireland does 14nm at least. But yea, Europe really missed the boat hard on chip manufacturing even though they have ASML sitting in the Netherlands. Actually they already were in the game with several fabs. And then there was ARM. Feels like they not just missed the boat but mismanaged and then sold it for scraps while they could have had a realistic shot of being at the top of the game. Sigh.

Wonder why that is. Is it too easy to move abroad where the money is way better?

ASML who makes the EUV machines and Carl Zeiss who make the optics for the ASML machines are both in Europe?

Exactly, so why are the leading foundries also not there?

America and Asia have cheaper labor and energy although I don't know how much of an advantage that really is for such a highly complex product.

Taiwan made the foundry business a strategic priority for multiple reasons. Intel at least for the US was always a business with national security impact. Europe never realized the importance and didn't make it anywhere near priority. They did not provide enough on-going incentives nor did they protect their companies from being sold to foreign ones. Instead the companies jumped hard on the outsourcing train and pushed as much manufacturing abroad as possible. This was a very shortsighted move that temporarily increased profits but traded the future for it. When nation-states put their weight behind something then it's nearly impossible to compete unless your own nation-state puts their weight behind you as well.

There is still some chance for Europe to come back into it but I have doubts they'll find the will to do it.

There is a really good video about it: https://www.youtube.com/watch?v=5ZdmS-EAbHo

Modern fab is being built in Germany by Intel iirc

Maybe

Isn't that why only Taiwan has them?

Because they have vastly more incentive to have them made on the island, then anyone else would have on their territory.

And the Taiwanese employees have more external motivation then anyone could feasibly have in the U.S. or Europe.

If I understand you correctly you are saying their unique geopolitical position has an effective on incentives to make semiconductors. That’s really interesting and once I think about it explicitly it makes sense. Excellent point!

> 護國神山 "Sacred Mountain protecting the Country" = TSMC

https://www.reddit.com/r/hardware/comments/wzqtd1/chinese_it...

that's Media's b.s. No country based their national security on a private company. its ridiculous.

Apple make iPhone that everyone love. Apple is the Shield that protect USA or maybe Google, Facebook? how about Ukraine is the food basket of Europe or perhaps even the world since Ukraine is a big exporter of grains. surely no one will attack Ukraine because the food price will soar and possible world wide famine.

They are a small island off the coast of a superpower that is pre-inclined to want to conquer them. They’ve been dealt a unique and very difficult hand, they are playing it was well as they can.

Did you reply to the wrong post? I didn't make any claims whatsoever and yet somehow you managed to build a strawman out of mine.

Can you please elaborate. Feels to me that the US trying to ban advanced semi in China combined with the fact they're produced an arm's distance away in Taiwan is a huge incentive for China to invade Taiwan.

Part of Taiwan’s defense strategy is that they produce a valuable enough piece of the puzzle that it is at least plausible that the US would step in and defend them if China attacked. This strategy is widely reported, it even has a name, the “Silicon Shield.”

Some might consider it unlikely that we actually would show up, but they are a small island off the shore of a superpower, so they are playing the hand they were dealt as well as they can. And, a small chance of the US showing up is still a pretty big disincentive to starting an invasion.

This is not fact checked. But I have heard that TSMC's factories have explosives installed in the buildings to self destroy if China invades.

China needs to deactivate or sabotage the explosives before any invasion. Even in that case, I am sure there is a plan B to nuke the factories.

Yeah, there is _zero_ chance of them being captured intact. It's a clean room! It's very easy to ruin! And both the Taiwanese and American governments have strong incentives to not let it be captured, and to signal that in advance.

Not only that, it's very dependent on the skills and hard work of its staff. Which is one of the things that makes TSMC so hard to replicate.

The history of Intel's decline, TSMC's rise and how the US lost the lead in manufacturing is one of the most relevant and poorly explored that I can recall.

If you speak to anyone not generally interested in the field, they are oblivious despite being profoundly impacted by it.

Here is a brilliant video on Why Europe Lost Semiconductors: https://www.youtube.com/watch?v=5ZdmS-EAbHo

It boils down to education and economical structure differences with US and JP.

To be on the cutting edge requires extreme scale and multidisciplinary talent. Europe, despite being in the front lines of the innovation, falls off as the European industries and military don't create the demand at scale required to keep up with the Americans.

It's probably a similar story with the US today. China is 5X the population and they do all the manufacturing, they have amazing talent which vanished from the west with de-industrialisation.

It was way more profitable to design stuff in California and make it in China. If everything is made in China and you demand fat margins, why would you have large scale cutting edge chip manufacturing in EU or USA?

This become problem only after the break down of the supply chains and increase in political conflicts.

But as the rest of the world catches up with IP and R&D, the manufacturing is bound to come back to the west. When you can do so much with a laptop and brains and China and India having so much from those, the West will have to start doing the low margin "uncool" stuff like producing physical goods.

Given the context when you say China, I assume you mean Taiwan. China plays a role too, but TSMC is not Chinese.

Given the lithography equipment used by TSMC is built by a single European company, I think it’s disingenuous to exclude Europe from the production chain.

I think the reality is fairly complex and the entire world is involved in the end to end production. Further, TSMC is working to diversify their production facilities to meet some of the onshorinf demands rising from the pandemic chip shortage fiasco. Additionally the instability between China and Taiwan is driving regional diversification. China has made itself such an undesirable business partner there’s enormous pressure top to bottom across all industries to find alternatives for sourcing. Margins in China have gotten worse due to a rising labor movement and improved general prosperity. Finally, there are other semiconductor manufacturers (global foundries, etc) that are striving to catch up - and it’s the nature of things that someone will. So, I agree, the world will start diversifying production out of China for many reasons. This will be better for everyone, even, IMO, China.

Politically Taiwan is not China, sure. Also it's true that the TSMC also has a global supply chain but this way or another the magic happens there.

We like it or not, the world is global. Still, overspecialisation is proving dangerous when not all parties are open democracies because they can act in economically unwise but ideologically coherent ways to keep the political class in power.

If you break it down, ASML is dependent on a US based company called CYMER(owned by ASML) without cymer EUV light source exports ASML is fucked will have a hard time creating EUV machine.

The US also makes a lot of the chemicals and EDA software packages. The US pretty much controls and owns advance chips. What they needed from taiwan was just a highly skilled sweat shop workers. Just look at the difference between the workload between TSMC workers in the US and on Taiwan.

That is why US sanctions are so effective they are like an octopus and have their tentacles all over the place controlling key nodes which can disrupt or destroy complete supply lines.

This is what following the tech war has kind of thought me.

ASML's EUV tech also came in part from the US's EUV LLC (public/private partnership from the 90s), and that's one reason we were able to control their exports to China.

I am so terrified for the effects on technological and economic progress a US/China military conflict over Taiwan would bring (not to mention the obvious other tragic and heinous consequences of war).

TSMC used to be 30% owned by Phillips, the same company that started ASML, the company that makes the machines that produce the chips for ASML. So the machines used to make these chips are Dutch and TSMC that uses these machines was originally Dutch. I think Europe could easily make these chips if they did not outsource so much to save money.

I wonder if a company that didn’t outsource to save money on these sorts of chips could survive in the Europe/US markets…

“Can” is always a funny phrase in these sorts of conversations. I mean it is clearly possible to make these chips because they make them in Taiwan. We’d just need a fundamentally different economy and long term investments, including a different labor force, probably different education system to get it…

I like that we are moving away from (fake) nanometers and giving names to process nodes instead; comparing different technologies by Xnm was completely broken.

At least you could meaningfully compare using "<" and ">". You can't do that with codenames.

How was it meaningful? What would the info "This one is called 5nm and this one is called 8nm" give you? Because it didn't really say anything about performance or watt or performance per watt.

I have an old ATI Radeon HD4850 GPU, I bet it beats your Nvidia RTX 4090 since 4850>4090 ;)

That is true, but just calling it the TSMC 2024 Process also allows comparing. But they don't like using dates because then if they stay on the process for a few years it is more obvious that it is "old".

Technically TSMC doesn't use nm.

It's the N2 node, not the 2nm node, but tech press is the way it is.

Unfortunately TSMC does indeed use nm extensively https://www.tsmc.com/english/dedicatedFoundry/technology/log...

well or else you will get intel naming scheme of using 10nm ++++++

The node names have numbers in them that are clearly intended to be compared. It's no coincidence that "Intel 3" and "N3" both have 3 in them.

Ideally it could be a measurement of density - how many transistors can dance on the head of a pin.

I wish we could get a measure like transistor density for some reference designs or something. (I mean it would be nice for us if every transistor was the same size, but just like we provide benchmarks for different instruction mixes, I guess we would want density benchmarks for low power/high chip flavors).

Worth noting GAA was suppose to come in N3, and Backside Power Delivery was originally scheduled for N2. So both feature got delayed and we end up with a fine tuned N3E. ( We could safely ignore N3.)

As with TSMC tradition we should see a fine tuned, cost reduction of N2P in 2027/2028. That is when all the mainstream products could enjoy those features.

> GAA was suppose to come in N3

any source on that? as far as i know there was never any official roadmap or announcement by tsmc of gaafet on n3.

this article seems to be from the n3 announcement (april edit: 2020) (https://www.techpowerup.com/265991/tsmc-3nm-process-packs-25...) and it was announced as finfet then.

another article from n3 announcement in april 2020: https://seekingalpha.com/article/4338843-taiwan-semiconducto... ("Taiwan Semiconductor Manufacturing Company Losing Its Process Leadership To Intel")

question: isn't there a bottom transistor size that it's physically impossible to go below?

So far any prediction of the bottom has not really come true. There must be some limit, but these engineers are pretty crafty. I bet they won’t beat one transistor per atom! Any other prediction, I’m suspicious about.

Physically impossible? I wouldn't say that. 2nm is a marketing term, depending on how you define it, it is actually 20nm or 45nm here.

It's probably very hard to shrink further than a very low number of atoms but we're far from that. The "2nm" marketing name has no relation to the actual physical size of the transistors. The physical features of the "3nm" process node are something like 24nm and 48nm. The Van der Waals radius of a silicon atom is about 0.2nm. Transistors made from a single molecule have been demostrated. Maybe we could even shrink to elementary particles but that's in the realm of SciFi so far.

The absolute hard minimum has to be one silicon atom wide, or 0.2nm. But note that "3nm" doesn't quite have any features that are that size, it's a non-precise marketing term.

Do we know what size these structures really are?

I dont know about 3nm but for 5nm they still have feature size ranging from 5x nm to 2x nm.

That would be surprisingly accurate

The smallest transistor is bounded by the atomic radius; a recent paper boasts 0.34nm for a graphene transistor. Scaling that up, to borrow a phrase from quantum computing, is "just an engineering problem." But for now we can only produce things on the atomic scale a handful of atoms at a time, and making a trillion-transistor processor would take somewhere between thousands and billions of years with today's tech.

Yes, in addition to the limits set by atomic radius, you also have the issue of electrons tunneling through the tiny energy barriers that exist when the the transistors are so close together.

So is Apple planning to use the N3E this year (not the N3)?

According to some reports it might be both. N3 (aka N3B) for A17 and N3E for M3. N3E is probably too late to produce the numbers needed for the A17 which should hit the market in September.

2 years behind Intel, as they release 20A with backside power delivery in 2024.

We'll see if that -

1) Is actually released in 2024 2) In volume 3) Plays a meaningful role in getting Intel back into striking distance to the performance crown in perf per mm per watt. Intel has pulled every shortcut lever (larger dies, crank up the power) to continue to be relevant. They're running out of shortcuts.

Intel 20A is scheduled for mass production in 2H24 and TSMC N2 in 2H25, so 1 year according to current roadmaps.

However, fabs get judged on what they ship, not what the promise. TSMC is shipping N3 in volume, and Intel 4 is still not shipping, and sounds like it will be a cut down limited volume release (https://www.tomshardware.com/news/intel-meteor-lake-cpus-for...) most likely because the yields are not there.

Based on prior delivery it is highly unlikely that Intel will ship meaningful volume on 20A before TSMC does on N2, but miracles happen.

>Intel 20A is scheduled for mass production in 2H24 and TSMC N2 in 2H25, so 1 year according to current roadmaps.

TSMC N2P ( Backside Power Delivery ) is 2026. So that is what OP meant by 2 years.

you are right. but the claim that tsmc is "2 year behind intel" based on those roadmaps is assuming a lot.

intel has consistently failed to deliver on their prior roadmaps and is currently behind tsmc in what they are shipping, so it requires a bit of magical thinking to believe that they will suddenly get a 2 year lead on tsmc.

intel is still not shipping 4, which is their first part to use euv. based on what they are shipping, they cant do it at volume yet. the idea that they are going to scale euv and then jump to gaafet and bspd in the next 18 months is kind of beyond belief.

when intel was failing to deliver 10nm (now called 7) in volume, they were saying the exact same thing, that 10nm was delayed but 7nm (now called 4) was on schedule (https://www.anandtech.com/show/15217/intels-manufacturing-ro...). back then intel 4 was scheduled for 2021. the magic leap didnt happen and it is 2 years late so really no reason to think that everything else wont get pushed back the same this time.

tsmc has already been shipping euv products in volume for years and they have a roadmap for introducing gaafet and bspd incrementally which looks a lot more realistic.

Intel's scheduled release dates tend to be... optimistic.

TSMC was never about being first, despite somehow they are now the leading edge Fab.

TSMC has always been conservative with feature set. GAA was supposed to be 3nm but ultimately pushed to 2nm. Same with Backside delivery.

TSMC is very clear about what their date does and means. Intel's 2024 ( and it is late 2024 ) will only be starting to producing and shipping to customer in 2025.

The volume difference between Intel's 20A and TSMC is also not comparable. This is a similar case with Samsung where they announce their latest nm but you dont see actual product shipping with it.

That being said I still think Intel would catch up by 2025. The question is whether they could find enough customers and volume.

What makes you think they'll catch up so soon?

Reply on news.ycombinator.com