News Intel puts 1nm process (10A) on the roadmap for 2027, aiming for fully AI-automated factories with 'Cobots

[Admin]

Intel shared an updated production roadmap that has its 1nm (10A) process arriving in late 2027. The company also outlined its intention to fully automate its fabs with AI 'cobots'

Intel puts 1nm process (10A) on the roadmap for 2027, aiming for fully AI-automated factories with 'Cobots : Read more

 

[2Be_or_Not2Be]

Could this finally be Factorio in real life?

 

[abufrejoval]

Well I guess it's natural for Intel to insource the packaging when that is where much if not most of the value is actually created when you Lego commodity chiplets in order to create bespoke products.

And of course they need to have packaging follow similar (ideally better) scale and cost curves as the chip(lets), or risk not being able to sell after they've produced 'their' parts.

And again, the reason these packaging jobs are in lowest-cost locations to day is that they are far too manual to be affordable next to the foundries today, so without robotics and automation they'd run into endless trouble.

Just too bad, that all these people currently doing these jobs, will soon no longer be able to afford Intel products, nor those from Apple or other chip giants...

 

[btmedic04]

Wasn't Intel 4 (7nm) their first single digit process node?

 

[George³]

Wasn't Intel 4 (7nm) their first single digit process node?

These are not actual nanometers, but trade names.

 

[usertests]

I want to see how 14A and 10A compare to the classic 14nm+++++++++. Transistor density scaling is worse than ever but you should still get what, 10x the transistors between "14nm" and "1.4nm"? Instead of the 64x if it was doubling per major node, 24x if it was +70%.

14nm - 10nm - 7nm - 5nm - 3nm - 2nm - 1.4nm

 

[pointa2b]

Its nice to see Intel fighting to regain dominance. This is a win-win for the consumer, regardless of who you like.

 

[Giroro]

The 'A' in 10A does not refer to Angstroms. It doesn't refer to anything.,

 

[vijosef]

I want to see how 14A and 10A compare to the classic 14nm+++++++++. Transistor density scaling is worse than ever but you should still get what, 10x the transistors between "14nm" and "1.4nm"? Instead of the 64x if it was doubling per major node, 24x if it was +70%.

14nm - 10nm - 7nm - 5nm - 3nm - 2nm - 1.4nm

If nm were physical distances,the scaling from 14nm to 1.4nm would be 100X

(14/1.4)²=10²=100

 

[setx]

The 'A' in 10A does not refer to Angstroms. It doesn't refer to anything.,

This.
Like in "Intel 7" the '7' is not '7nm' it's just abstract 7 unrelated to anything.

 

[PlutoDelic]

It feels rather good to see that nobody trusts Intel's game on the names. This 1A will end up being 4nm.

 

[rtoaht]

This.
Like in "Intel 7" the '7' is not '7nm' it's just abstract 7 unrelated to anything.

Same for TSMC 7nm also known as N7. That 7 doesn’t represent any “real nanometers” or unit of any physical distance. It’s just a marketing name for the node from TSMC.

 

[jddigitalchaos]

The 'A' in 10A does not refer to Angstroms. It doesn't refer to anything.,

Yes, it does though. Intel has specifically called out 20A as "entering the Angstrom era". If you're referring to it not representing any transistor dimension, then yes you are correct. Process names haven't actually been associated with any transistor dimension for decades (even for TSMC or Samsung). Theses are public names that really just associate with "equivalent density" for an older transistor dimensioning scheme that is no longer valid. I think we'll see process names start to approach the size of an atom in our lifetimes, but transistors are unlikely to get anywhere that close.

 

[jddigitalchaos]

It feels rather good to see that nobody trusts Intel's game on the names. This 1A will end up being 4nm.

Hate to break it to you, but Intel's new naming convention really better matches what TSMC and Samsung have done. Intel was just dead set for too long on each process node giving a density improvement of <70% from the previous generation, but TSMC didn't get anywhere near that from 10->7nm (can't seem to post link to article showing this, but you can search it).

 

[Mad_2]

Seems author thinks he is very clever to conclude 10A "production" in 2027 based on only this chart. Did anyone from intel actually say this?

This article incorrectly states that intel did not previously disclose date of 14A until this talk, but it was in the chart from earlier Tom's article on this event, shown clearly by 2027. Impossible that 10 arrives same year! Does anyone even think about this before writing such a ridiculous headline?

Same chart shows similar volume of 20A/18A already in 2023! By author's logic, headline should therefore be "Intel 18A secretly in production since 2023". Not!

Nowhere on this chart does it say this is "production" volume. It must include development wafer starts, and likely minimum of 2 years from significant development starts to production.

 

[Mad_2]

.. Seems author is paying attention. All instances of "production" just changed to "production/development".

If I was editor, I'd just change to "early wafer volume" or "development wafer starts". "Production/development" is just word salad without any meaning.

 

[DavidC1]

It seems quite irresponsible for a company the size of Intel to claim large language models will fully automate factories, but hey, they can put whatever they want on powerpoint.

Seems author thinks he is very clever to conclude 10A "production" in 2027 based on only this chart. Did anyone from intel actually say this?

He also got this part wrong:

(For example, the difference between Intel 7 and Intel 4 was a 15% improvement.)

It was 20%, not 15%. But when you aren't striving for accuracy rather quantity over quality as with all modern press, what can you expect?

 

[rtoaht]

It seems quite irresponsible for a company the size of Intel to claim large language models will fully automate factories, but hey, they can put whatever they want on powerpoint.

He also got this part wrong:

It was 20%, not 15%. But when you aren't striving for accuracy rather quantity over quality as with all modern press, what can you expect?

Intel didn’t claim LLM would automate their factories.

Cutting edge semiconductor fabs are already almost fully automated. Robots are heavily used and humans don’t “touch” the wafers. All Intel claim was to expand it further by introducing more AI making the robots more collaborative and cohesive.

 

[KraakBal]

We were at single digit nanometres at 8nm already, subtitle

 

[PaulAlcorn]

Seems author thinks he is very clever to conclude 10A "production" in 2027 based on only this chart. Did anyone from intel actually say this?

This article incorrectly states that intel did not previously disclose date of 14A until this talk, but it was in the chart from earlier Tom's article on this event, shown clearly by 2027. Impossible that 10 arrives same year! Does anyone even think about this before writing such a ridiculous headline?

Same chart shows similar volume of 20A/18A already in 2023! By author's logic, headline should therefore be "Intel 18A secretly in production since 2023". Not!

Nowhere on this chart does it say this is "production" volume. It must include development wafer starts, and likely minimum of 2 years from significant development starts to production.

Intel refused to disclose the start date of 14A, as discussed in the original article. They did disclose that 14A-E would fall in the 2027 time frame, as indicated in the chart.

 

[PaulAlcorn]

Intel didn’t claim LLM would automate their factories.

Cutting edge semiconductor fabs are already almost fully automated. Robots are heavily used and humans don’t “touch” the wafers. All Intel claim was to expand it further by introducing more AI making the robots more collaborative and cohesive.

You're right, they never said an LLM would run the factory. However, they are aiming for full autonomy. Read the slide: 10X moonshot of 'Fully Autonomous Factories."

And yes, they are already heavily automated, as noted in the article, but definitely not fully -- I have been in multiple and can attest to the hundreds of workers in these facilities.

 

[PaulAlcorn]

It seems quite irresponsible for a company the size of Intel to claim large language models will fully automate factories, but hey, they can put whatever they want on powerpoint.

He also got this part wrong:

It was 20%, not 15%. But when you aren't striving for accuracy rather quantity over quality as with all modern press, what can you expect?

Gelsinger was asked to define what a node means in a Q and A with press. I based the statement on his reply. His exact words, taken from my recording:

"Generally, when we say a node, it's at least double-digit power performance improvements per node. You know, I think our cutoff is 14%, 15%. So between seven [Intel 7] and four [Intel 4] was at least 15%, four [Intel 4] and three [Intel 3] was at least 14%, 15%." -- Gelsinger.

 

[vertuallinsanity]

"Cobots"..

I guess we're still working on getting the whole planet on the cob, then. Eh-corn? Good luck with that..

This is not a real story and Intel doesn't backslash their node names because... Logic!

 

[michaelmantion]

I can put a lot of stops on a Map, doesn't mean I will ever get there.

The only certainty with Pat is he will disappoint.

 

[michaelmantion]

It seems quite irresponsible for a company the size of Intel to claim large language models will fully automate factories, but hey, they can put whatever they want on powerpoint.

He also got this part wrong:

It was 20%, not 15%. But when you aren't striving for accuracy rather quantity over quality as with all modern press, what can you expect?

I have been focused on Machine learning, AI, GANs, LLM for over a decade.

I am certain that AI can replace workers in almost every field

I am also certain that people will use the wrong term, especially when attacking future technologies.

The good news for you, is there are tools that allow you to re-format your comments to hide your ignorance of the topic.