Support NeoGAF

[tusharngf]

Could start volume production of 1.4nm chips in 2027 - 2028.

Development of TSMC's 1.4nm-class manufacturing technology is well underway, the company revealed during the Future of Logic panel during the IEEE International Electron Devices Meeting (IEDM). TSMC also re-emphasized that mass production using its 2nm-class fabrication process is on track for 2025.

TSMC's 1.4nm production node is officially called A14, according to a slide published by Dylan Patel from SemiAnalysis. For now, TSMC has not disclosed when it plans to start high-volume manufacturing (HVM) on A14 and its specifications, but with N2 scheduled for late 2025 and N2P set for late 2026, it is reasonable to guess that A14 is coming after that (2027 - 2028).

When it comes to features, A14 is unlikely to adopt vertically-stacked complementary field effect transistors (CFETs), although TSMC is exploring the technology. Therefore, A14 will probably rely on the company's 2nd or 3rd generation gate-all-around FETs (GAAFETs) — just like N2 nodes.

Nodes such as N2 and A14 will require system-level co-optimization to really make a difference and enable new levels of performance, power, and features.

What remains to be seen is whether TSMC plans to adopt High-NA EUV lithography tools for its A14 process technology in the 2027 - 2028 timeframe. Given the fact that by that time Intel (and possibly other chipmakers) will have adopted and perfected next-generation EUV litho machines with a 0.55 numerical aperture, it should be fairly easy for the contract maker of chips to use them. However, because High-NA EUV lithography tools halve reticle size, its usage will bring some additional challenges to both chip designers and chipmakers.

Source: https://www.tomshardware.com/tech-i...-for-the-first-time-says-2nm-remains-on-track

TLDR : Next gen machines will be power-efficient and expensive

 

[StreetsofBeige]

I remember forums where armchair techies would claim stuff like "it cant get smaller than 8nm due to physics and overheating"

6nm... 5nm....4nm..... 1.4nm.

 

[March Climber]

I remember forums where armchair techies would claim stuff like "it cant get smaller than 8nm due to physics and overheating"

6nm... 5nm....4nm..... 1.4nm.

I saw it happen here too. I don't know enough about this topic to take a stance on it though. I'm just glad that progress is still being made.

 

[CrustyBritches]

Can’t wait for 0nm. That shit will be lit.

 

[James Sawyer Ford]

Yet MS wants to start next gen in 2026?!

 

[StreetsofBeige]

I saw it happen here too. I don't know enough about this topic to take a stance on it though. I'm just glad that progress is still being made.

Amen to that. I'm no techie, but I love technological progress.

Almost all my career is based in traditional manufacturing company stuff (the shit you see at grocery stores and Walmart). Even though I'm sure the R&D guys do their magic making things better and cheaper, there's only so much progress household supplies or edibles can progress. It's kind of the same shit, where the biggest innovation half the time are new flavours, scents or packaging art. Some of the shit we make hasn't changed, since I joined the company over 10 years ago except the artwork refreshes every 5 years.

 

[Little Chicken]

Stop. Let me save you a few thousand posts.

This will have nothing to do with the next PlayStation or Xbox.

 

[Deerock71]

I remember forums where armchair techies would claim stuff like "it cant get smaller than 8nm due to physics and overheating"

6nm... 5nm....4nm..... 1.4nm.

It's similar to approaching the speed of light.

 

[Codeblew]

I remember forums where armchair techies would claim stuff like "it cant get smaller than 8nm due to physics and overheating"

6nm... 5nm....4nm..... 1.4nm.

True, but eventually they will hit the limits of physics. Can't go smaller than an atom wide.

 

[Embearded]

I remember forums where armchair techies would claim stuff like "it cant get smaller than 8nm due to physics and overheating"

6nm... 5nm....4nm..... 1.4nm.

I don't have much knowledge on the topic but i understand most of what i read about it due to my profession.

Truth is those people were somehow right. The nm is supposed to be the length of the transistors gate and there is a hard limit that you cannot cross without quantum effects kick in.
They started using different techniques, materials and processes to work around that but gate length hasn't really shortened and nm numbers are mostly marketing now.
We do get better products of course, it's just that the nm they announce do not really have the same meaning as they used to.

 

[magnumpy]

ah, seems this will come too late for PS6 etc. not that this level of technology would be in the budget anyway...

 

[Silver Wattle]

I bet this is just the 2nm+, what 4nm is to the 5nm process, and the 6nm is to the 7nm process.

 

[draliko]

This is like intel 10nm++++++++ tech, please separate marketing and underlying tech... Things are getting smaller but not this smaller, we're getting smarter with design but we still haven't overthrown physics laws

 

[SolarFry]

I predict yearly (or 2) increase of performance by 1-5% and you'll have the privilege to buy it for a lot of money. 1.4+, 1.4++, 1.4+++ etc. They can stretch it into infinity. 1.3, 1.2, 1.1++++++++++++++++++++++++++++++

 

[Tsaki]

Nah, volume productions would mean small chips, like SoCs for phones. Add 2 years after that date to see this tech in consoles, at best. We'll be lucky enough to get consoles with the N2P node, and hopefully it will have great yields by that point in time (2027-2028).

 

[Xyphie]

Nah, volume productions would mean small chips, like SoCs for phones. Add 2 years after that date to see this tech in consoles, at best. We'll be lucky enough to get consoles with the N2P node, and hopefully it will have great yields by that point in time (2027-2028).

Yep. Apple basically buys the entirety of TSMC's leading edge node for at least a year with high performance variants only following later. E.g. Apple A14 was the first 5nm chip back in 2020, only two years later in 2022 did nVidia/AMD adopt it. We're seeing the same thing now with 3nm, with nVidia looking to ship 3nm in late 2024 at best.

 

[KuraiShidosha]

I remember forums where armchair techies would claim stuff like "it cant get smaller than 8nm due to physics and overheating"

6nm... 5nm....4nm..... 1.4nm.

Wait until this guy learns that all node sizes since 14nm are bullshit marketing terms and they're still not actually sub 8nm.

 

[Zathalus]

Yet MS wants to start next gen in 2026?!

1.4nm will have no impact on next-gen. Volume production in 2028 means the soonest a console will have it is by 2030. If Microsoft does choose end of 2026 they will miss out on 2nm though.

 

[James Sawyer Ford]

1.4nm will have no impact on next-gen. Volume production in 2028 means the soonest a console will have it is by 2030. If Microsoft does choose end of 2026 they will miss out on 2nm though.

Missing out on 2nm means they miss out on the leap required for true next gen

 

[SABRE220]

Thats good news for the next gen consoles let Sony wait out until 2029/30 to get a real upgrade with the pro adding life to the lineup. It looked a bit grim with how we were forced to go towards giant chips to get generational gains but with mcm and good progress with the die shrinks next gen might be bright. That being said next step quantum computing baby lol.

 

[crackajack]

I remember forums where armchair techies would claim stuff like "it cant get smaller than 8nm due to physics and overheating"

6nm... 5nm....4nm..... 1.4nm.

"due to physics" is correct though. I think 4nm is the gap given for physics not working anymore as intended, which is the number arm chair techies just cite, and this isn't anymore the same thing as the marketing nm number which is just a label to give an estimate off how the actual gaps would need to be with an old chip design to represent the same efficiency. I can't remember if there is one reference all use and if it is even remotely accurate or just ever shriveling, straight fantasy bs, Intel might be comparing to their Pentium 1 and AMD to their K7 or whatever. The whole Finfet, superfin, nanoribbon whatever tech inclusion changes and the nm number should reflect that. I think there are several different important gaps that could be used, and their values differ between Intel and Amd, but none actually are the nm given.