TSMC Could Delay 2 nm Mass Production to 2026

[AleksandarK]

According to TechNews.tw, TSMC could postpone its 2 nm semiconductor manufacturing node for 2026. If the rumors about TSMC's delayed 2 nm production schedule are accurate, the implications could reverberate throughout the semiconductor industry. TSMC's alleged hesitancy could be driven by multiple factors, including the architectural shift from FinFET to Gate-All-Around (GAA) and potential challenges related to scaling down to 2 nm. The company is a crucial player in this space, and a delay could offer opportunities for competitors like Samsung, which has already transitioned to GAA transistor architecture for its 3 nm chips. Given the massive demand for advanced nodes due to the rise of AI, IoT, and other next-gen technologies, it is surprising to hear "sluggish" demand reports.

However, it's also possible that it's too early for customers to make firm commitments for 2025 and beyond. TSMC has dismissed these rumors, stating that construction is progressing according to plan, which includes having 2 nm pilot run in 2024, and mass production in the second half of 2025.. Despite this, any delay in TSMC's roadmap could serve as a catalyst for shifts in market dynamics. Companies that rely heavily on TSMC's advanced nodes might need to reassess their timelines and strategies. Moreover, if Samsung can capitalize on this opportunity, it could somewhat level the playing field. As of now, though, it's essential to approach these rumors with caution until more concrete information becomes available.



View at TechPowerUp Main Site | Source

 

[AnotherReader]

It took them 3 years to move to N3 from N5. Like Intel before them, they could be slowing down.

 

[bonehead123]

It took them 3 years to move to N3 from N5. Like Intel before them, they could be slowing down.

The tech world, like Time, "waits for no one"

They can slow down production ramps as much as they need to, as long as my dividend checks don't slow down or get smaller

 

[bug]

When demand is all but guaranteed, you don't need to keep migrating to newer nodes to secure your orders.
But, if confirmed, this can be a blessing in disguise: the longer 3nm is their main workhorse, the longer the time they have to recoup initial investment and make a profit. This could mean lower prices for 3nm parts. Hypothetically, at least.

 

[ARF]

When demand is all but guaranteed, you don't need to keep migrating to newer nodes to secure your orders.
But, if confirmed, this can be a blessing in disguise: the longer 3nm is their main workhorse, the longer the time they have to recoup initial investment and make a profit. This could mean lower prices for 3nm parts. Hypothetically, at least.

For ROI they need more of your orders. Not only iphones 15 pro max, but everything - CPUs, GPUs, consoles, chips 3nm everywhere.
The thing is that the 3nm production is limited to Apple only but it looks also limited in volumes. That's why they can't sell wafers to anyone else, including nvidia which is also capable to pay as much as asked for a wafer.

The bad thing is for AMD - AMD will stay on 5nm till 2025, and then in 2030 it will move to 3nm... and maybe by 2035 they will move to 2nm

 

[Prima.Vera]

Is 2nm actual size, or is just a marketing name, like O2, or something?
What is the real size and how does it compare to competition I wonder? (Such as Intel's 18A, for example)

 

[AnotherReader]

For ROI they need more of your orders. Not only iphones 15 pro max, but everything - CPUs, GPUs, consoles, chips 3nm everywhere.
The thing is that the 3nm production is limited to Apple only but it looks also limited in volumes. That's why they can't sell wafers to anyone else, including nvidia which is also capable to pay as much as asked for a wafer.

The bad thing is for AMD - AMD will stay on 5nm till 2025, and then in 2030 it will move to 3nm... and maybe by 2035 they will move to 2nm

Apple is expected to be the only customer for N3B. AMD's Ryzen CPUs are very small and as such, cheap to manufacture. While they will wait for N3E to be made available and mature, they won't take until 2030 to do so. Nvidia hasn't moved to N3B because it is an immature process and Nvidia has avoided new processes for a while now as their chips are very large and require mature processes to yield well.

 

[ARF]

Apple is expected to be the only customer for N3B. AMD's Ryzen CPUs are very small and as such, cheap to manufacture. While they will wait for N3E to be made available and mature, they won't take until 2030 to do so. Nvidia hasn't moved to N3B because it is an immature process and Nvidia has avoided new processes for a while now as their chips are very large and require mature processes to yield well.

The Apple A17 Pro, the 3nm chip is said to be around 100-110 mm^2, which makes it around the same die size as nvidia's AD107.

Apple has a stranglehold on the TSMC 3nm chip supply in 2023

Apple has reserved nearly all of TSMC's 3-nanometer process chip production for upcoming iPhone, Mac, and iPad models with the A17 Bionic and Apple Silicon M3 chips.

appleinsider.com

 

[AnotherReader]

The Apple A17 Pro, the 3nm chip is said to be around 100-110 mm^2, which makes it around the same die size as nvidia's AD107.

Apple has a stranglehold on the TSMC 3nm chip supply in 2023

Apple has reserved nearly all of TSMC's 3-nanometer process chip production for upcoming iPhone, Mac, and iPad models with the A17 Bionic and Apple Silicon M3 chips.

appleinsider.com

With new nodes being so expensive, Nvidia would never give peasants, i.e. gamers, GPUs built on the latest nodes. I'm talking about huge dies like the GH100. Those can afford to be on the best node, but Nvidia is so far ahead of its competitors in the AI space that they don't need to use the latest node.

 

[ARF]

With new nodes being so expensive, Nvidia would never give peasants, i.e. gamers, GPUs built on the latest nodes. I'm talking about huge dies like the GH100. Those can afford to be on the best node, but Nvidia is so far ahead of its competitors in the AI space that they don't need to use the latest node.

This I agree. Historically, it was AMD who used the state-of-the-art processes first. But it was 10 years ago, now it's not longer possible...

 

[AnotherReader]

This I agree. Historically, it was AMD who used the state-of-the-art processes first. But it was 10 years ago, now it's not longer possible...

AMD migrated to N7 when it was fairly new with the Radeon VII following the iPhone XS by only four months. After that, they decided to follow Nvidia and migrate to processes when they are mature. For CPUs, they don't have to do that as their chiplets are very small, but they must have their reasons besides cost for that decision.

 

[ARF]

AMD migrated to N7 when it was fairly new with the Radeon VII following the iPhone XS by only four months. After that, they decided to follow Nvidia and migrate to processes when they are mature. For CPUs, they don't have to do that as their chiplets are very small, but they must have their reasons besides cost for that decision.

I remember the 7nm endeavours but the VII was a pale shadow of such masterpieces like Radeon HD 3870, 4870 and 5870 which I was writing about.

 

[AnotherReader]

I remember the 7nm endeavours but the VII was a pale shadow of such masterpieces like Radeon HD 3870, 4870 and 5870 which I was writing about.

Of course, both the Vega 64 and its successor, the Radeon VII, failed to challenge Nvidia, but it was the last time that AMD jumped to a new node.

 

[dicobalt]

TSMC's too fast roadmap not going as planned, I could have never predicted that. Some 3nm products have been delayed also. TSMC did too much posturing towards investors and customers to inflate their importance and solidify contracts. Meanwhile Nvidia has turned to Intel for some products.

 

[Wirko]

Is 2nm actual size, or is just a marketing name, like O2, or something?
What is the real size and how does it compare to competition I wonder? (Such as Intel's 18A, for example)

Just add thirty. Works for all new nodes.

 

[kondamin]

if that means they can do a quicker ramp up than 3nm I guess that's still OK.

 

[alwayssts]

TSMC's too fast roadmap not going as planned, I could have never predicted that. Some 3nm products have been delayed also. TSMC did too much posturing towards investors and customers to inflate their importance and solidify contracts. Meanwhile Nvidia has turned to Intel for some products.

While I had not heard that's necessarily the case, it certainly appears possible (and agree it has been hinted); it absolutely has been said nVIDIA is aware of Intel's 18A performance. While I don't have the quote in front of me, and by no means am I assuming it won't perform well, from what I remember it sounded like nVIDIA posturing; in reality putting TSMC (and conceivably Samsung) in check and making them aware they have another bargaining chip for lower prices on future high-end nodes. It was with that level of shrewdness they made their knowledge of Intel's node known; as is Huang's way, just has he also openly chastized TSMC many times in the past to gain leverage (or a scapegoat). The fact nVIDIA used Samsung 8nm for Ampere also grants them a huge bargaining chip, as they have proven they will crunch the numbers and go where they will make the greatest profit, not the greatest product, because sometimes they don't have to due to factors like their architecture's perceived strengths, software stack, and sheer defacto placement as THE gpu vendor for many people/companies. While many have appeared leery to use Samsung for copious reasons (like their design tools/yields), nVIDIA may have a notable advantage with their designs being largely aided by an AI supercomputer which may help largely mitigate the former, and a CEO able to negotiate like almost no-other for pricing concerning the latter.

The important thing I wanted to respond to though is exactly about this giving Intel a more likely entry into the space as a notable supplier, which is a huge deal. While I think many of us were skeptical they could keep to their timeline for 18A (especially for external customers) in any realistic manner given their 10nm woes, that appears more and more conceivably plausible. Showing Lunar Lake was a master class in stating they have early confidence in the 18A node more-so than anything else. They clearly are open for business and are optimistic about their future node advancements (backside power etc) and timeline, while TSMC appears to be slowing down and their future appears less-certain post their success in finfet; one could argue the latest news of their 2nm 'strike force' could be perceived as panic; perhaps they are aware of what Intel is going to be able to offer and are struggling to deliver similar advancements in any kind of similar time-frame to customers. It's almost starting to reek of the troublesome period when they switched from SiON to HKMG, but with less worry of a broken product and more-so of one that is late or less performant/advanced than is expected or the competition can deliver.

Samsung remains a huge wildcard. Can they get their design tools in order? Will their TTM with whatever they call their node actually be competative with what TSMC/Intel are able to get out the door? Will their yields ever improve to a level that they (re)gain the confidence of new or even former partners? I don't know. Maybe. Or maybe it will just be relatively much cheaper than the competition. If there is anyone that has been willing to play ball in that arena, it has been them.

Truly the space is interesting again, though, and it hasn't been for a very long time (ever since GF gave up the 7nm ghost). We likely will see true competition again, which is wonderful wrt not only different performance potential(s), but pricing competition where-as TSMC's position as the sole purveyor of leading-edge nodes appeared to see it spiraling out of control. With the perpetual power of Apple money fueling TSMC's advancement and dominance, it's been far too long since we've seen competition from the likes of IBM/GF/Samsung; or even farther back UMC.

While the immediate future for most-everything is probably still N3E, beyond that appears nebulous...and to me that's actually extremely exciting.

In a time where advancements appear to be slowing, what better time to see these companies feet put to the fire to deliver their best? This will only lead to stronger products and potentially pricing than has been the trend, and is especially important when you think of the timeframe of important products like a future console generation (which will likely have their pick of a mature version of whatever Samsung/Intel/TSMC can cobble together post-3nm). In that respect that's great news. This competition may in fact be the spur to the ass that keeps some remnant of Moore's Law alive in a time it has otherwise appeared very much dead.

 

[bug]

Is 2nm actual size, or is just a marketing name, like O2, or something?
What is the real size and how does it compare to competition I wonder? (Such as Intel's 18A, for example)

At this point, you'd need a PhD to figure that out. There's no feature that physically 2nm in size and I wouldn't know where to begin trying to compare to Intel or Samsung.
The only sane metric that still stands is transistor density.

 

[Fourstaff]

No surprises here, everyone is slowing down and TSMC is finally slowing down.

 

[AnotherReader]

At this point, you'd need a PhD to figure that out. There's no feature that physically 2nm in size and I wouldn't know where to begin trying to compare to Intel or Samsung.
The only sane metric that still stands is transistor density.

CPP, i.e. contacted gate pitch, and minimum metal pitch are still the right metrics.

 

[watzupken]

It took them 3 years to move to N3 from N5. Like Intel before them, they could be slowing down.

Slow down is inevitable when it gets exponentially harder to shrink the transistors. So it will not apply to just Intel for sure. I do think that TSMC probably became complacent due to their success over the years, and starting to trip over themselves. For example, the current N3 used for Apple's A17 SOC don't seem to be doing well. May well turn out to be another TSMC's costly mistake with their 20nm, which the SOCs on 20nm draws a lot of power and overheats.

While I had not heard that's necessarily the case, it certainly appears possible (and agree it has been hinted); it absolutely has been said nVIDIA is aware of Intel's 18A performance. While I don't have the quote in front of me, and by no means am I assuming it won't perform well, from what I remember it sounded like nVIDIA posturing; in reality putting TSMC (and conceivably Samsung) in check and making them aware they have another bargaining chip for lower prices on future high-end nodes. It was with that level of shrewdness they made their knowledge of Intel's node known; as is Huang's way, just has he also openly chastized TSMC many times in the past to gain leverage (or a scapegoat). The fact nVIDIA used Samsung 8nm for Ampere also grants them a huge bargaining chip, as they have proven they will crunch the numbers and go where they will make the greatest profit, not the greatest product, because sometimes they don't have to due to factors like their architecture's perceived strengths, software stack, and sheer defacto placement as THE gpu vendor for many people/companies. While many have appeared leery to use Samsung for copious reasons (like their design tools/yields), nVIDIA may have a notable advantage with their designs being largely aided by an AI supercomputer which may help largely mitigate the former, and a CEO able to negotiate like almost no-other for pricing concerning the latter.

The important thing I wanted to respond to though is exactly about this giving Intel a more likely entry into the space as a notable supplier, which is a huge deal. While I think many of us were skeptical they could keep to their timeline for 18A (especially for external customers) in any realistic manner given their 10nm woes, that appears more and more conceivably plausible. Showing Lunar Lake was a master class in stating they have early confidence in the 18A node more-so than anything else. They clearly are open for business and are optimistic about their future node advancements (backside power etc) and timeline, while TSMC appears to be slowing down and their future appears less-certain post their success in finfet; one could argue the latest news of their 2nm 'strike force' could be perceived as panic; perhaps they are aware of what Intel is going to be able to offer and are struggling to deliver similar advancements in any kind of similar time-frame to customers. It's almost starting to reek of the troublesome period when they switched from SiON to HKMG, but with less worry of a broken product and more-so of one that is late or less performant/advanced than is expected or the competition can deliver.

Samsung remains a huge wildcard. Can they get their design tools in order? Will their TTM with whatever they call their node actually be competative with what TSMC/Intel are able to get out the door? Will their yields ever improve to a level that they (re)gain the confidence of new or even former partners? I don't know. Maybe. Or maybe it will just be relatively much cheaper than the competition. If there is anyone that has been willing to play ball in that arena, it has been them.

Truly the space is interesting again, though, and it hasn't been for a very long time (ever since GF gave up the 7nm ghost). We likely will see true competition again, which is wonderful wrt not only different performance potential(s), but pricing competition where-as TSMC's position as the sole purveyor of leading-edge nodes appeared to see it spiraling out of control. With the perpetual power of Apple money fueling TSMC's advancement and dominance, it's been far too long since we've seen competition from the likes of IBM/GF/Samsung; or even farther back UMC.

While the immediate future for most-everything is probably still N3E, beyond that appears nebulous...and to me that's actually extremely exciting.

In a time where advancements appear to be slowing, what better time to see these companies feet put to the fire to deliver their best? This will only lead to stronger products and potentially pricing than has been the trend, and is especially important when you think of the timeframe of important products like a future console generation (which will likely have their pick of a mature version of whatever Samsung/Intel/TSMC can cobble together post-3nm). In that respect that's great news. This competition may in fact be the spur to the ass that keeps some remnant of Moore's Law alive in a time it has otherwise appeared very much dead.

In my opinion, Intel may have a better chance if TSMC trips over. But in the longer term, I don't believe Intel as a fab is going to be cost effective, and hence, attractive. As far as I am aware, most of Intel's fabs are not in cost friendly locations, including the new ones. I suspect most are using Intel now to,
1. For diversification amidst all the geopolitical tension,
2. Use Intel fab as a bargaining tool for better pricing with TSMC/ Samsung. Yes Intel likely cost more, but it is still a viable alternative. So if TSMC tries to increase price, the likes of Nvidia cans start moving some business to Intel.

 

[Wirko]

No surprises here, everyone is slowing down and TSMC is finally slowing down.

And we're not yet hitting the limits of physics. We're about to crash into the combined limits of physics and economics.

Transistors made at new and upcoming nodes can still be cheaper than the previous generation but only if their number keeps growing exponentially. To make this possible, consumers need to multiply much faster, invade Mars, then an exponentially growing number of planets. [Maybe that's what aliens did, that's why they arrived as far as to the Earth, brought us the transistor, and said in a language we didn't understand: Here you are, and please do repeat our mistakes, we'll watch from afar.]

At this point, you'd need a PhD to figure that out. There's no feature that physically 2nm in size and I wouldn't know where to begin trying to compare to Intel or Samsung.

To be exact (too exact), and optimistic too (too optimistic), IBM published images of their experimental 2nm chip a couple years ago and there are insulation or passivation layers visible that are about 2nm thick. Yes, I understand those aren't characteristic dimensions of a node.

The only sane metric that still stands is transistor density.

Yes but even that is just approximate. The Zen 4 and 4c have the same number of transistors and are made on the same node, yet one is pretty much smaller. Same with certain Arm big-mid-little designs (Snapdragon?) where two of the three sizes are the same type of Arm core, but with different performance. Lower performance transistors are smaller. Beside that, Intel and others can't agree on how to measure density. Intel uses a certain mix of logic gates and SRAM cells as a sample, others have other methods.

it's been far too long since we've seen competition from the likes of IBM/GF/Samsung

I occasionally look for news about the former AMD-GF-IBM-Samsung alliance but never find any. Are any remains of it still active?

But IBM keeps developing advanced stuff, like the 2nm chip I mentioned. They need to sell that tech to someone ... who's that? Intel, lately, maybe?

 

[Vayra86]

It took them 3 years to move to N3 from N5. Like Intel before them, they could be slowing down.

I don't know if I've ever read about a node that arrived on time to be honest.

This isn't news

 

[Wirko]

I don't know if I've ever read about a node that arrived on time to be honest.

This isn't news

Yes it is news, you just have to put it into context and read it as "2nm Delayed More Than Everybody Expected".

 

[Vayra86]

Yes it is news, you just have to put it into context and read it as "2nm Delayed More Than Everybody Expected".

Still not news lol.
Did you miss Intel 10nm? TSMC 7? EUV?

The context is that every node ever has been a delayed/staggered launch and in practice it means every time you get access to it far later than anyone 'roadmapped' for.