TSMC schedules 1.4nm mass production for 2028, targets sub-1nm pilots in 2029

Posted by sr_local@reddit | hardware | View on Reddit | 53 comments

[-]

_hlvnhlv@reddit

I wonder what naming scheme will they use once they can't keep making shit up.

For those that don't know, the "nm" in the process node used to indicate the gate pitch or something, and when they weren't able to keep shrinking it, they just started making it up

An example: a "3nm" node has a gate pitch of 40 something nanometers, not 3

Basically nothing uses features that small, and the improvements are usually in other places besides "smaller goes brrrrrr"

[-]

Process naming has traditionally referred to the effective resolution of the lithographic process, the limits of what the optics can reliably pattern. It roughly tracking the minimum gate length of an ideal planar transistor was more of a historical accident, that held only from the ’70s through the ’90s.

For its intended audience, node naming is still a useful shorthand for the generation and capabilities of a given process.

If you’re not steeped in the EE side, it’s easy to misinterpret those labels. And when people don’t understand something, they often fill in the gaps with conspiratorial speculation that misses the mark.

[-]

account312@reddit

They’re not even remotely close to running out of SI prefixes. Though they’re kinda running low on distances that aren’t smaller than a silicon atom.

[-]

superkickstart@reddit

6.187142e+25 planck length process

[-]

gorion@reddit

They already use Angstroms. This node is called: TSMC A14

Anyways, TSMC, Intel, Samsung already dont use "nm" in their node prooces names, marketing people and pseudo journalists does.
Eg. "3nm", "5nm", "7nm"

[-]

gumol@reddit

Anyways, TSMC, Intel, Samsung already dont use "nm" in their node prooces names, marketing people and pseudo journalists does.

TSMC website: "TSMC’s 2nm (N2) technology has started volume production in 4Q25 as planned."

https://www.tsmc.com/english/dedicatedFoundry/technology/logic/l_2nm

[-]

jmlinden7@reddit

The official name of the process node is N2, their marketing people decided to call it 2nm on the website for whatever reason

[-]

gumol@reddit

do TSMC marketing people not work at TSMC?

[-]

DerpSenpai@reddit

Eventually they will use the density per mm^2

[-]

ComplexEntertainer13@reddit

Of what? Not all transistors or libraries are the same.

SRAM density used to be a reasonable metric. But with scaling breaking down and transistor types scaling differently depending on what you are using them for, there are no perfect metrics.

[-]

DerpSenpai@reddit

most likely a mixture of logic and SRAM, the industry will have to decide, current naming just doesn't scale much longer.

[-]

_hlvnhlv@reddit

That's a win to me.

But yeah, it's just too complicated really, it cannot be resumed in just a magic number

[-]

Different_Lab_813@reddit

Blame journalist not TSMC, they haven't used nanometers in their roadmaps for a long time.

[-]

gumol@reddit

Blame journalist not TSMC, they haven't used nanometers in their roadmaps for a long time.

https://www.tsmc.com/english/dedicatedFoundry/technology/future_rd

"Summary of TSMC's Major Future R&D Projects"

"2nm logic technology platform and applications"

[-]

Seanspeed@reddit

We dont need to 'blame' anybody, as it's not actually a real problem in the first place.

TSMC's customers aren't you and I, they're major chip design companies who are not choosing a manufacturing process based purely on its name! lol They know what they're doing. It's not an issue.

[-]

Seanspeed@reddit

What is going to stop them 'making shit up'? :/

It's just a naming scheme.

[-]

III-V@reddit

I sure hope so. We've got a generation of pedantic redditors coming into power once these boomers finally croak. Then it's all over for these lawbreakers.

[-]

jmlinden7@reddit

Its supposed to be a 2x improvement in performance per generation.

[-]

J_KBF@reddit

Angstroms then elections maybe lol

[-]

mrybczyn@reddit

they should start making units.

next node 1.21 jigawats!

[-]

Marco-YES@reddit

Jiga isn't a made up SI unit. It's a mispronunciation of Giga.

[-]

szakee@reddit

can't wait to browse instagram faster!

[-]

PoemPuzzleheaded8651@reddit

Slower.

Companies will use the faster chips to justify pushing shitty features with shittier code to slow down apps.

[-]

amidoes@reddit

It already happens with games

Crysis from 2007 looks better than most of nowadays games who need shitty AI and upscaling with a RTX 5080 to not run and look like shit

[-]

Brickman759@reddit

Go back and look at Crysis again. it was mind-blowing at the time and still looks great. But it was easily surpassed by games that were coming out over a decade ago.

It's also built like garbage. Super inefficient, can only use one CPU core. It ran under 30fps on pretty much all hardware for at least 3-4 years after. Release.

[-]

Gronfir@reddit

The high end also moved from 1080p@30+ to 4k@120+ pushing 16 times the pixels.

[-]

Seanspeed@reddit

This reminds me of the stupid takes people have where GPU's get more powerful and gamers claim it'll just mean devs get lazier and stop optimizing.

Like if we all just still used Geforce 256's today, we'd totally still have games like Resident Evil Requiem and they'd totally run so much better!

[-]

BlueSiriusStar@reddit

I dont think devs are getting any lazier though. Its the corporate machine which looks at this tech as a panacea to all their problems, shrinking timelines and such.

[-]

Seanspeed@reddit

But timelines aren't shrinking. Game development timelines are WAAAY longer than they've ever been!

The reason games have issues or dont always run well(and they've never all run well) is largely because they're just seriously complicated pieces of software and games are getting ever more advanced and complex. Then on PC specifically, you need the game to still interface with the operating system and so there can be issues there as well(DX12 has notoriously been a source of many modern issues).

Almost never is 'more powerful hardware' a cause of these problems.

[-]

gumol@reddit

Then on PC specifically, you need the game to still interface with the operating system

on consoles you also need to interface with the OS

[-]

TheFaithlessFaithful@reddit

Sure, but there's simply way less variety. It's a single (or a few highly similar) hardware/software combos.

In contrast to PCs, you have people running a huge variety of hardware, on a huge variety of OSs (many of which are behind on updates to drivers and the OS).

[-]

BlueSiriusStar@reddit

I mean the game engine also could be an issue here. UE5 could be notoriously hard to run with nanite and such. I just think that game engine aren't optimised for the workloads we require today.

I mean on every PC you still have to interface with the OS and the driver. Complicated doesn't mean slower games automatically.

[-]

lazyhustlermusic@reddit

Your take is shortsighted.

[-]

fatong1@reddit

Sorry not to be pedantic, but is it though? Looking at the current software trend + over reliance on LLM generated code I'd say no.

[-]

lazyhustlermusic@reddit

Would you expect the same code volume in 2026 as we had in 2006?

[-]

fatong1@reddit

Yeah i know, but you have to agree it's a bit funny that modern software is just becoming more and more bloated with "features" no one wants. In a perfect world the code base growth should decrease not increase, while still implementing the same amount of features.

[-]