IBM introduces 2 nm node chip

Get your science fix here: research, quackery, activism and all the rest
Post Reply
User avatar
shpalman
Light of Blast
Posts: 4485
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond

IBM introduces 2 nm node chip

Post by shpalman » Fri May 07, 2021 6:43 am

IBM Introduces the World’s First 2-nm Node Chip

Image

Well if I were the referee I'd point out there's no scale bar on that. But this helps explain what we're looking at:

Image

(from The Nanosheet Transistor Is the Next (and Maybe Last) Step in Moore’s Law)

I noted that the process uses silicon germanium, but only because there are chemicals which etch SiGe and not Si, so you can etch the SiGe layers away from between the Si to leave these free-standing Si ribbons.
molto tricky

User avatar
jaap
Fuzzable
Posts: 333
Joined: Mon Nov 11, 2019 2:05 pm
Location: Netherlands
Contact:

Re: IBM introduces 2 nm node chip

Post by jaap » Fri May 07, 2021 7:54 am

Regarding the scale, I read that the "2nm" is no longer a real size measure but a kind of proxy equivalent. The transistors are now stacked up in several layers, but if you were able to squeeze them all in a single layer, then you would need a 2nm resolution to make it. The resolution used to produce these chips is actually less than that, but it is damned impressive nevertheless.

User avatar
basementer
Catbabel
Posts: 870
Joined: Mon Nov 11, 2019 1:03 pm
Location: Aotearoa
Contact:

Re: IBM introduces 2 nm node chip

Post by basementer » Fri May 07, 2021 4:30 pm

There's something analogous going on in the time domain. The speculative instruction execution of current processors is so accurate that a fixed point add operation takes, on average, less than one clock cycle. (Mentioned in one of Lex Fridman's YouTube interviews, I'll try to find the source tomorrow.)
Money is just a substitute for luck anyway. - Tom Siddell

User avatar
basementer
Catbabel
Posts: 870
Joined: Mon Nov 11, 2019 1:03 pm
Location: Aotearoa
Contact:

Re: IBM introduces 2 nm node chip

Post by basementer » Sat May 08, 2021 12:47 am

basementer wrote:
Fri May 07, 2021 4:30 pm
There's something analogous going on in the time domain. The speculative instruction execution of current processors is so accurate that a fixed point add operation takes, on average, less than one clock cycle. (Mentioned in one of Lex Fridman's YouTube interviews, I'll try to find the source tomorrow.)
it was in this conversation with Jim Keller:
https://www.youtube.com/watch?v=Nb2tebYAaOA
Money is just a substitute for luck anyway. - Tom Siddell

Post Reply