TSMC Reveals 2nm Node: 30% More Performance by 2025 | Hacker News Reader

TSMC Reveals 2nm Node: 30% More Performance by 2025

tomshardware.com

20 pointspinewurst21 comments

Nice to see another confirmation that N3 is expected to be in high volume production by the end of the year, even if delivery to customers is still pushed out until early next year.

>TSMC's first 3 nm-class node is called N3 and this one is on track to start high volume manufacturing (HVM) in the second half of this year. Actual chips are set to be delivered to customers in early 2023.

https://www.anandtech.com/show/17452/tsmc-readies-five-3nm-p...

Who else but Apples hands will these end up in early 2023? From what I understand Apple basically buys all their cutting edge volume.

I think Intel is already making GPUs on TSMC 5nm, and they certainly have the money to buy their way to the front of the line for 3nm.

I have a feeling that Apple's Mac Pro will be on a new family of chips that were designed for TSMC 3nm.

...Assuming that TSMC is still around, independent, and manufacturing in 2025.

50% more quantity?

What happens when they get down to zero nanometers? Will they start using a different fake measurement, or go into the negatives?

real numbers, dude... 0.9nm, 0.8nm, 0.6nm ...

This nomenclature is gonna get awkward soon.

0.2 nm is the diameter is a Si atom.

Somewhere around 10^-35 they're going to hit the plank constant which is the smallest discrete measure of length.

Just a nitpick. The plank length is not the smallest discrete measure of length. It's not even known whether our universe has discrete or continuous space at all.

It seems to me like maybe we should be referring to the resolution in atoms at this point. If and when lithography is accurate to the individual atom, that's the absolute end of miniaturization, right? (Unless maybe we start super-cooling chips to make them contract..? but that's not lithography.)

Current "nm" values do not measure anything. No part of your "5nm" chip is 5nm in length/width/depth/etc. It's been a decade since that, why start now?

Intel seems to want to use angstroms as a unit (instead of intel 2(nm), it will be intel 20a, don’t know if an angstrom is a tenth of a nm.

My understanding (from undergrad a while back) is that they already jimmy the numbers. It used to be that X nm meant that the gate width was X nm. Now it (usually) means that the smallest feature you can find is maybe X nm. Closer to a marketing term like '3G'.

Not that this isn't a real technical and industrial achievement, you just can't take the X nm at face value.

Maybe you are joking but nanometer is just a unit. We can easily go below 1 nm. For example 500 picometers. In fact that's exactly what the semiconductor industry did when they moved below 1 micrometer.

Intel have talked about future processes in terms of Angstroms (0.1µm), but for some time it's not been a real measure of size, just a marketing term. Intel has actually dropped the units and now calls their process nodes Intel 7, 4, 3. Apparently they plan to use 20A after that in reference to the Angstrom, although as before, not because they are actually measuring anything to that specific size.

Not possible, obviously.

But in a hypothetical scenario where process shrinking continued unabated by considerations such a particle size and uncertainty, a negated logarithmic scale would do for naming conventions.

    gen = offset + log(1/feature_size)

This might be the best for some time until they get to -1 or Windows style version numbers. Notice how all those hype headlines publish little or no technical info ? This looks more like P(entium)R(ate) numbers.

Can someone translate this?

Small chip go 1/3rd faster.

Depends on a lot of factors. Every trace is a capacitor, an inductor and a resistor. And they can also form parasitic diodes.

As transistors get smaller, you can put more in a given area. More transistors = more work can be done.

Reply on news.ycombinator.com