IBM says it's built the world's first 2nm semiconductor chips

IBM Research says it's made the world’s first 2nm process node chips, squeezing 50 billion transistors onto each of the fingernail-sized dies. Big Blue today claimed this is “a breakthrough in semiconductor design” using nanosheet technology. That involves layering three sheets of material to form a stack containing an NMOS …

[JassMan]

Sounds like it is time for a new standard

I (and I am sure many others) hadn't realised the quoted size no longer had any direct relationship to any physical characteristic of the chips. I was going to propose that since this is the case and they will shortly run out of integers, that they need to have a new measure. My suggrstion is the number of transistors which can be shoved up a flea's arse. That way as packing density gets higher so does the number. Thusly you might have say 100TPFA. (txsistors/fleas arse), or 1kTPFA if you increase the density 10 fold.

[Spherical Cow]

Re: Sounds like it is time for a new standard

African or European flea?

[Fred Goldstein]

Re: Sounds like it is time for a new standard

Spanish flea. With music by Herb Alpert.

[Korev]

Re: Sounds like it is time for a new standard

If this carries on then I'm going to flea the area...

[Ken Shabby]

Re: Sounds like it is time for a new standard

They'll be starting from scratch then?

[SuperGeek]

Re: Sounds like it is time for a new standard

Thank you flea much for the best laugh I've had in a while! :)

[Anonymous Coward]

Re: Sounds like it is time for a new standard

Gnatt's chuff Shirley?

[Fruit and Nutcase]

Re: Sounds like it is time for a new standard

Indeed. We already have a "Gnat's Crotchet" or a "Midge's Minim" for minuscule time measurements, courtesy of Humphrey Lyttleton, so it would be logical to base this standard on one those two benchmark creatures

[Ken Moorhouse]

Re: the number of transistors which can be shoved up a flea's arse

Surely that will increase the number of bugs?

No flea in the ear for JassMan's suggestion. Have one of these instead--->

[the spectacularly refined chap]

Re: Sounds like it is time for a new standard

I was going to propose that since this is the case and they will shortly run out of integers, that they need to have a new measure.

Won't stop, either step down a unit or go decimal. Both have already happened at least once. I remember when at uni fabs would proudly boast of their 0.18μm processes. And the howls of "you cannae break the laws of physics" at the prospect of going beneath 100nm at around the same time for that matter.

[Dazed and Confused]

Re: Sounds like it is time for a new standard

I remember when at uni fabs would proudly boast of their 0.18μm processes. And the howls of "you cannae break the laws of physics" at the prospect of going beneath 100nm at around the same time for that matter.

Back in the early 80s the lab I was working in had got their 5um process into production and were well on the way to getting production ready at 3.5um. HP had just come out with their 1um NMOS process for their HP9000 series 500 systems. I went to talk at the RI given by a guy from IBM's labs talking about how their were using X-Ray diffraction gratings to do the focusing as optical wave lengths too long for the line sizes they were researching. Can't remember the line sizes they were quoting but it was orders of magnitude better than we were managing.

[Boothy]

Re: Sounds like it is time for a new standard

For anyone curious, there's a bit of history on WikiChip on when nm no longer actual meant a real size anymore (around 32/28nm era, approx. 10 years ago).

https://en.wikichip.org/wiki/technology_node#Meaning_lost

[DS999]

They won't run out of integers

They didn't run out when sizes used to be in microns. They went to decimals like .35 microns etc. but when at .09 microns they decided to start counting in nanometers.

So they can either start counting in fractions of a nanometer, or count in picometers. Since they no longer reflect actual transistor dimensions they are more of a placeholder for when you get another shrink with somewhat less than 2x as many transistors per area as the previous one.

[John Brown (no body)]

Re: They won't run out of integers

"So they can either start counting in fractions of a nanometer, or count in picometers."

So, picometres it is then. Marketing won't stand for numbers meant to be getting smaller suddenly getting bigger, even if the signifier means it's smaller because marketing won't believe the users are clever enough to understand (because marketing don't understand, and that's their measure of consumers intelligence)

[DS999]

Re: They won't run out of integers

You don't really think marketing enters into designating a process as "7nm" or whatever, do you? Who are they marketing to? Certainly not end consumers, 99% of whom couldn't tell you what process the SoC in their phone or CPU in their PC is built on.

[Raithmir]

Re: Sounds like it is time for a new standard

What is the average diameter of a flea's anus?

[swm]

Maybe now they can put a chip in the COVID vaccine.

[Korev]

Yeah, chips have transistor GATES and we all know GATES is Bill's surname? WAKE UP SHEEPIES!

[Anomalous Cowshed]

Chip in vaccine

People used to say I had a chip on my shoulder, now Bill Gates wants to put it inside my arm? Is that what they call embedded technology?

[Pascal]

"the size of the process node is more of a naming convention"

... "No part of it is 2nm in size" ... "The actual transistor gate length is 12nm" ...

I went from "YAY, 2 nm!" to "fuck you, marketers" in 3 sentences flat.

[redpawn]

My...

dick's smaller than your.... I could drill a hole through a Higg's boson.

[seven of five]

Meanwhile, in managment...

"Sooo, with these smaller transistors, do we need lees people to make and operate then?"

[Zebo-the-Fat]

Re: Meanwhile, in managment...

No, you just need (much) smaller people!

[Pete 2]

Making a splash

> IBM says it can fit 50 billion transistors onto a 150mm2 die

The BC108 transistors in my components drawer are cylinders 5mm tall and 5mm in diameter. Assuming rows and columns stacked (not close packed) that would permit 8 million transistors per cubic metre. Or 2.5 olympic sized swimming pools to hold 50 billion of IBM's new product.

As for the power consumption? I hate to think.

[Mage]

Re: power consumption?

The power consumption claims for a phone a bit disingenuous given RF transmit power, screen lighting and screen driving.

Also the best use for most of those transistors is RAM, or maybe Flash, but Flash is rather different in construction. 1G RAM per core and little shared RAM to reduce bottlenecks, like Transputers.

Also the BC108 chips are maybe 1 mm x 1 mm by maybe 0.1mm to dissipate power. But you can run them at 1.2V supply and a 1 or 1 uA per device. You need more then because gain is low. I did once make a matchbox radio for MW/LW with a soldered in AAA cell that lasted nearly 6 months and no power switch. Ceramic earphone.

[hoola]

Re: power consumption?

When I was a kid my Father lashed up this long wire from the house to the end of the garden. There was a coil on a ferrite rod and one of those huge air gapped capacitors that old radios had.

A germanium diode, a crystal (I assume similar to ceramic) and hey presto, a radio. I picked up all softs of interesting stuff.

[My-Handle]

Re: power consumption?

I had a similar kit as well, it used the cold water pipes as a ground and a long length of wire as the aerial.

I just wish the manufacturer actually told me how the damned thing was supposed to work. Just shoving a bunch of components at me and telling me how to put them together only teaches me how to assemble something that's already been designed, not how to design and build something like that myself.

Of course, now we have the internet. Give me a sec, I'm going to go read up on crystal radios...

[John Brown (no body)]

Re: power consumption?

"Give me a sec, I'm going to go read up on crystal radios..."

Better be quick before AM transmitters disappear! I don't think an FM crystal is possible. And FM might disappear too. Before long, there won't be the possibility of teaching kids the bare basic of radio with home-made components. They'll have to buy a "black box" digital radio kit.

[Yet Another Anonymous coward]

Re: power consumption?

>teaches me how to assemble something that's already been designed

That's when we had proper education.

Now you would just learn how to harvest the data from people looking at ads for the product that has already been assembled by someone else

[Jan 0]

Re: power consumption?

Those radios were known to their makers/owners as "crystal sets". The air gaps were to allow the plates to move to alter the capacitance and thus the resonant frequency of the circuit. As a boy, I did try using a cat's whisker and a Selenium? crystal as the detector, but found a new fangled germanium diode to be far easier to use. From my bedroom in South London, I could listen to Russian jamming signals. Radio Luxembourg and cheerful people on Radio Sofia (Bulgaria). My aerial ran from a fire escape to a sycamore tree about 40 metres away.

[Mike 137]

NOT a new standard then

"No part of it is 2nm in size. The actual transistor gate length is 12nm. For comparison, Intel's 14nm has an 20nm gate length, and Samsung's 7nm node has an 8-10nm length."

Obviously Samsung have the edge in the physical domain. Intel just have it in the promo domain.

Furthermore I doubt if real geometries can get much smaller. High frequencies and small inter-conductor distances result in crosstalk that's ultimately limiting - that's physics and immutable.

[DS999]

Re: NOT a new standard then

Furthermore I doubt if real geometries can get much smaller. High frequencies and small inter-conductor distances result in crosstalk that's ultimately limiting - that's physics and immutable.

That's why they keep changing how transistors are made. They used planar until leakage currents become a problem due to shrinking dimensions. Then they were FinFET and extensions of that starting about a decade ago and now those only have another generation or two left. So now IBM's 2nm (and TSMC 2nm and Samsung 3nm) transistors are made using GAA nanosheets.

I'm sure there has been active research for years in the next step beyond that when those run out of gas, and the researchers involved probably have a pretty good idea what the next step will be.

[zuckzuckgo]

Re: NOT a new standard then

Smaller feature size would imply shorter interconnects which could reduce cross talk. Smaller transistors usually mean faster switching and higher frequency operation or lower power consumption when operated at lower frequencies.