The Lemon Fool

IBM 2nm chip breakthrough claims more power with less energy

It's significant I think. But probably more so is the closing comment on the article.

And Intel's chief executive has announced a $20bn (£14.6bn) investment in two new plants in the US, telling the BBC that having 80% of the world chip supply in Asia is not a good idea.

AiY

In a science thread "power", "energy", ... should be treated as terms of art and used carefully.

At the same time this news both changes everything and changes nothing

An article here about the hoops various manufacturers are jumping through to acheive sub-5 nm devices.

https://semiengineering.com/new-transistor-structures-at-3nm-2nm/

9873210 wrote:In a science thread "power", "energy", ... should be treated as terms of art and used carefully.

I've given up moaning about sloppy use of dimensions. I regularly see "amps per hour", "kiloWatts per day" and similar nonsense written by people who really ought to know better. Using the terms "power" and "energy" as synonyms also drives me demented, power being the rate at which energy is being used (loosely!)

For other silicon geeks out there - there is more information in the following IBM blog post.
https://www.ibm.com/blogs/research/2021/05/2-nm-chip/

Still very little information on leakage and variance, Personally I would like to see some figures.

When I started out in the semiconductor industry in the 1970s, were were talking about 10 micron (10,000nm) features on 2 or 3 inch wafers. Clock speeds of a MegaHertz or two. And now here we are with 2nm, 12 inch wafers and GigaHertz clock speeds. I worked in semiconductors for 35 years, but I still find the state of current technology utterly boggling.

--kiloran

kiloran wrote:When I started out in the semiconductor industry in the 1970s, were were talking about 10 micron (10,000nm) features on 2 or 3 inch wafers. Clock speeds of a MegaHertz or two. And now here we are with 2nm, 12 inch wafers and GigaHertz clock speeds. I worked in semiconductors for 35 years, but I still find the state of current technology utterly boggling.

--kiloran

Completely agree, I'm 35+ years in the industry and still consult for a few days a week (following early retirement that didn't quite go to plan). The idea that transistors less than 10 silicon atoms wide are considered to be viable is astonishing. The idea that you can attempt accurately model the operation of these devices is even more unbelievable.

25 to 20 years ago I was using 68HC11 micro controllers. For development we used the EEPROM version, which had a window so you could see the die. We had three generations of these in use. On the oldest the die filled the window, about 1cm across. The newest (still ancient by todays standard) was about 2mm across, a small dot in the middle of the window.

We also found a few bugs in our circuit boards. They did not work at the temperature extremes with the newer chips. We traced these to missing pull-up resistors. On the older, larger chips the intrinsic pull-ups could sink enough current to hide our design errors; the smaller chips were not so forgiving.

9873210 wrote:For development we used the EEPROM version, which had a window so you could see the die.

Small technical point, they would be EPROM (one E).

UV light (short wavelength that can give you skin cancer and cataracts) was used to erase them. There are a few devices still available that use EPROM such as some of the PIC microcontrollers (The ones with a C in their names) but they do not have windows and are, therefore, not eraseable. Most PICs available now use FLASH which can be reprogrammed many times, in most cases, using a cheap Chinese programmer off eBay.


Julian F. G. W.

jfgw wrote:

9873210 wrote:For development we used the EEPROM version, which had a window so you could see the die.

Small technical point, they would be EPROM (one E).

UV light (short wavelength that can give you skin cancer and cataracts) was used to erase them. There are a few devices still available that use EPROM such as some of the PIC microcontrollers (The ones with a C in their names) but they do not have windows and are, therefore, not eraseable. Most PICs available now use FLASH which can be reprogrammed many times, in most cases, using a cheap Chinese programmer off eBay.


Julian F. G. W.

Yes one E, the details get foggy after 20 years.

The HC11 family pulled the trick you mention with the PIC. The PROM version (one time programable) we used for production were the same die, but packaged without the quartz window so they could not be erased. Rumor had it they could be erased with soft x-rays which would go through the ceramic packaging, but we never got that desperate.

The FLASH chips I've used have all be programmed in circuit without any off board programmer. That saves the cost of a socket, which is sometimes more expensive than the chip.

9873210 wrote:The PROM version (one time programable) we used for production were the same die, but packaged without the quartz window so they could not be erased. Rumor had it they could be erased with soft x-rays which would go through the ceramic packaging, but we never got that desperate.

I had not heard of that but it makes sense since X-rays are higher energy than UV. I had read that if an EPROM didn't have a foil sticker over the window and someone photographed it using a flash gun, the data could be corrupted.


Julian F. G. W.

Intel is struggling to keep third or fourth ranking, never mind the #1 it was only a few years back. It blew things when it failed to mass-produce 10nm chips and its 7nm ones are a year or so away from production.
Taiwanese TSMC who make the new chips for Apple that replace Intel’s have also taken the unusual step of offering fabrication facilities to other chip designers, something Intel has now said it will also do at its planned hugely-expensive fabrication factories in Arizona. Intel has no real choice: it is either 'bet the company' or die.

So huge congrats to IBM for making a 2nm chip although there is a long long way – at least five years - from a lab chip to mass production. (IBM also once 'bet the company' with System 360: totally new hardware and operating system that at a stroke obsoleted its current range. A huge $$risk - but they won.)

Chip fabrication is currently driven by TSMC, US-based Global Foundries and Samsung, with Chinese SMIC some way behind (one reason perhaps why China wants Taiwan ‘back’!)

AsleepInYorkshire wrote:It's significant I think. But probably more so is the closing comment on the article.

And Intel's chief executive has announced a $20bn (£14.6bn) investment in two new plants in the US, telling the BBC that having 80% of the world chip supply in Asia is not a good idea.

AiY

Wouldn't that just be a bit of corporate pique about being marginalised in this century's growth areas - particularly the last decade - by the ARM ecosystem?

UncleEbenezer wrote:

AsleepInYorkshire wrote:It's significant I think. But probably more so is the closing comment on the article.

And Intel's chief executive has announced a $20bn (£14.6bn) investment in two new plants in the US, telling the BBC that having 80% of the world chip supply in Asia is not a good idea.

AiY

Wouldn't that just be a bit of corporate pique about being marginalised in this century's growth areas - particularly the last decade - by the ARM ecosystem?

It's not the ARM ecosystem on it's own, it's ARM in combination with the TSMC 5nm process at least in Apple's case.
https://www.theverge.com/2020/11/10/215 ... -processor

AMD are also eating Intel's lunch with TSMC's fabrication technology plus they have x86/x64 compatablility.

Disclosure, I hold TSMC and AMD shares.

Mike4 wrote:

9873210 wrote:In a science thread "power", "energy", ... should be treated as terms of art and used carefully.

I've given up moaning about sloppy use of dimensions. I regularly see "amps per hour", "kiloWatts per day" and similar nonsense written by people who really ought to know better. Using the terms "power" and "energy" as synonyms also drives me demented, power being the rate at which energy is being used (loosely!)

Not sure I see the problem here (unless quotation marks should have been used) as it is surely, in a sense, a play on words (possibly unintentionally):

"It is also more energy efficient - using 75% less energy to match current performance, IBM said."

"power" = Computational 'power' (performance)

"energy" = Electrical energy used

Sorcery wrote:

UncleEbenezer wrote:

AsleepInYorkshire wrote:It's significant I think. But probably more so is the closing comment on the article.

And Intel's chief executive has announced a $20bn (£14.6bn) investment in two new plants in the US, telling the BBC that having 80% of the world chip supply in Asia is not a good idea.

AiY

Wouldn't that just be a bit of corporate pique about being marginalised in this century's growth areas - particularly the last decade - by the ARM ecosystem?

It's not the ARM ecosystem on it's own, it's ARM in combination with the TSMC 5nm process at least in Apple's case.
https://www.theverge.com/2020/11/10/215 ... -processor

AMD are also eating Intel's lunch with TSMC's fabrication technology plus they have x86/x64 compatablility.

Disclosure, I hold TSMC and AMD shares.

I'd never heard about TSMC 5nm process until I looked it up after reading your post. It amazes me that chip manufactures are able to continue to make such improvements. Long may Moore's law continue. My son is much more in the know about computer chips than me and he says that in the PC market AMD have a significantly better CPU line-up v Intel. The Apple M1 chip looks pretty outstanding.

RC

ReformedCharacter wrote:

Sorcery wrote:

UncleEbenezer wrote:Wouldn't that just be a bit of corporate pique about being marginalised in this century's growth areas - particularly the last decade - by the ARM ecosystem?

It's not the ARM ecosystem on it's own, it's ARM in combination with the TSMC 5nm process at least in Apple's case.
https://www.theverge.com/2020/11/10/215 ... -processor

AMD are also eating Intel's lunch with TSMC's fabrication technology plus they have x86/x64 compatablility.

Disclosure, I hold TSMC and AMD shares.

I'd never heard about TSMC 5nm process until I looked it up after reading your post. It amazes me that chip manufactures are able to continue to make such improvements. Long may Moore's law continue. My son is much more in the know about computer chips than me and he says that in the PC market AMD have a significantly better CPU line-up v Intel. The Apple M1 chip looks pretty outstanding.

RC

TSMC claim to have 3nm ready, at least for an ARM variant that I think they used as a test of principle. Apple are fairly certain to use it in their next product iteration. https://asia.nikkei.com/Business/Tech/S ... -chip-tech

Not sure about Intel using it in that article, TSMC are hardly likely to tell Intel their secrets unless protected by patent. It's possible Intel will source chips from TSMC, since Intel seem to have not kept up with TSMC.

XFool wrote:

Mike4 wrote:

9873210 wrote:In a science thread "power", "energy", ... should be treated as terms of art and used carefully.

I've given up moaning about sloppy use of dimensions. I regularly see "amps per hour", "kiloWatts per day" and similar nonsense written by people who really ought to know better. Using the terms "power" and "energy" as synonyms also drives me demented, power being the rate at which energy is being used (loosely!)

Not sure I see the problem here (unless quotation marks should have been used) as it is surely, in a sense, a play on words (possibly unintentionally):

"It is also more energy efficient - using 75% less energy to match current performance, IBM said."

"power" = Computational 'power' (performance)

"energy" = Electrical energy used

There is always a problem with bad puns.

Even in terms of "computational power" the units are wrong. Computer power is measured per unit time, e.g. in "Floating Point Operations Per Second", not in "Floating Point Operations". The easiest way to get more power for less energy is get a powerful CPU and turn it off, which is why I spent several months of my life getting sleep mode to work.

"More calculations for less energy" would be correct.