Intel 20A Node Cancelled

[erek]

“The 20A node, while now cancelled for Arrow Lake, has played a crucial role in Intel's journey towards 18A. It served as a testbed for new techniques, materials, and transistor architectures essential for advancing Moore's Law. The 20A node successfully integrated both RibbonFET gate-all-around transistor architecture and PowerVia backside power delivery for the first time, providing valuable insights that directly informed the development of 18A. Intel's decision to focus on 18A is also driven by economic factors. With the current 18A defect density already at D0 <0.40, the company sees an opportunity to optimize its engineering investments by transitioning now. However, challenges remain, as evidenced by recent reports of Broadcom's disappointment in the 18A node. Despite these hurdles, Intel remains optimistic about the future of its foundry services and the potential of its advanced manufacturing processes. The coming months will be crucial as the company works to demonstrate the capabilities of its 18A node and secure more partners for its foundry business.”

Source: https://www.techpowerup.com/326289/...ers-arrow-lake-mainly-manufactured-externally

 

[noko]

Let see, they were supposed to be making part of Arrowlake with 20a, if 18a is ready or very close then why is it not on 18a? Yet it went from 20a to some other fab. Seems like Intel was more ready to ditch their own fabs and process and use someone else’s. Who in their right mind would not want to do the same and skip using Intel Fabs? See another nose dive coming for Intels stock.

 

[Armenius]

Let see, they were supposed to be making part of Arrowlake with 20a, if 18a is ready or very close then why is it not on 18a? Yet it went from 20a to some other fab. Seems like Intel was more ready to ditch their own fabs and process and use someone else’s. Who in their right mind would not want to do the same and skip using Intel Fabs? See another nose dive coming for Intels stock.

18A isn't taped out yet. Not until next year. 20A was already taped out. Both processes are 5nm. The difference is 18A is on Intel's second generation RibbonFET while 20A is on the first generation.

 

[ChadD]

Someone needs to get Intel into a meeting.... a Gamblers Anonymous meeting.

Not sure if this helps or hurts their stock. On one hand they are cutting the fat... on the other hand. The process that just publicly got a third party shit kicking is what they went all in on. Well lets go now, make it work or find out what being de listing feels like. I'm sure this makes sense in the long run the 20a less then process wasn't attracting any big investor soothing fab clients. I'm not convinced 18a will either, but at least they seem committed to trying to stop the bleeding.

 

[OKC Yeakey Trentadue]

18A isn't taped out yet. Not until next year. 20A was already taped out. Both processes are 5nm. The difference is 18A is on Intel's second generation RibbonFET while 20A is on the first generation.

What process size are we actually on? I heard 14nm was true 14nm, but 7nm is actually 10nm and now you are saying 20A (2nm) is actually 5nm...

 

[ChadD]

What process size are we actually on? I heard 14nm was true 14nm, but 7nm is actually 10nm and now you are saying 20A (2nm) is actually 5nm...

None of them release anything concrete anymore so who actually knows how anything compares to what anymore.
7nm was 7nm... it got confusing cause Intel tried convincing the industry their 10nm was = to the competitions 7mm. Which was horse shit. Intel pointed to one metric amongst many that make the process the process and said see we are = or better. Then they when they couldn't make that work they dialed back those numbers anyway to ship their 10nm parts which 100% were 10nm.

With this 20A 18A. I do believe Intels 18A is at least as good as what TMSC is cooking... on paper maybe even a little better. The issue is Intel hasn't had one process turn out yields the industry finds acceptable for a long time. Intel has been dealing with high defect rates no matter what they say... which you can get away with if they are your own chips and you have OEMs you can push off a bunch of cast off parts too for cheap. If your trying to sell that fab space however... you gotta get that defect rate down. Broadcomm was disappointed with 18a... if I had to guess I'm assuming that Broadcomm has no need of neutered parts with chunks fused off. Intel has always had the luxury of being able to fuse off half the cache a handful of cores or the iGPU and use those chips for lower end or mobile parts. IMO that has made fixing their fabs much harder. They have gotten mulitple processes to the good enough for Intel point for over a decade now. If you want to sell fab space however, you want your customers to be fusing fully functional parts to simply fill the market if need be not to salvage silicon out of nesesity as half the chip is defective. (Think back years now to old AMD GPUs were a bios flash would re enable cores that worked perfectly fine... those were off a fab with good yield they got fused or firmware killed to fill the market not becuase the silicon was defective. That is what a good foundry can do for you.)

 

[Armenius]

What process size are we actually on? I heard 14nm was true 14nm, but 7nm is actually 10nm and now you are saying 20A (2nm) is actually 5nm...

There is no industry standard. Intel starting naming their processes based on how they think it competes with TSMC. Intel 7, for example, is their 10nm process, but has transistor density comparable to TSMC 7nm. In this case, Intel is saying that their 5nm+ process would be equivalent to TSMC's 1.8nm process. Intel 4 and Intel 3 are both 7nm by Intel's definition.

It is quite annoying to read how many sites keep calling 20A and 18A a 20 angstrom/2nm and 18 angstrom/1.8nm process.

 

[ChadD]

There is no industry standard. Intel starting naming their processes based on how they think it competes with TSMC. Intel 7, for example, is their 10nm process, but has transistor density comparable to TSMC 7nm. In this case, Intel is saying that their 5nm+ process would be equivalent to TSMC's 1.8nm process. Intel 4 and Intel 3 are both 7nm by Intel's definition.

It is quite annoying to read how many sites keep calling 20A and 18A a 20 angstrom/2nm and 18 angstrom/1.8nm process.

The biggest issue for Intel... is other companies that buy fab space get detailed specs, and understand them.

 

[Lakados]

What process size are we actually on? I heard 14nm was true 14nm, but 7nm is actually 10nm and now you are saying 20A (2nm) is actually 5nm...

None of them have been to scale since 130nm, everything after that was a marketing pitch.

The ASML NXE 3600D which is the hardware TSMC uses for their 5 and 3nm nodes is a 13.5nm wavelength with a 13nm resolution, and the new ASML EXE 5000 series Intel has installed in their new fabs for 20A, and 18A is an 8nm resolution with a 13.5nm wavelength.

But despite the same wavelength being used the EXE 5000 series allows for objects on a single layer to be almost half the size of that from the 3600D, which results in nearly 3x the density.
Anyway, what it comes down to is these machines burn human souls to generate a laser that carves intricate runes into a rock to perform dark Voodoo rituals that make the world go round. At this point, the numbers are arbitrary because there is a near-infinite means of measuring them to generate almost any value you want that could be justified on a sheet of paper with a convincing enough argument that a marketing team could run with it.

But take this here: https://en.wikipedia.org/wiki/International_Roadmap_for_Devices_and_Systems

Using the "International Standards" they define 3nm as anything with a contacted gate pitch of at least 48nm and a metal pitch of roughly 24nm... Intel only recently bought into these around 2020 or so, which is when they renamed all their stuff, until then everybody was calling it whatever they wanted.

So in reality the names are made up and the nodes don't matter!

Only results matter.

 

[Lakados]

Steve Zinsser, has admitted its going to 'skip over productising' the 20A node to save cash, effectively cancelling its use and making it little more than a test node for its backside power and gate-all-around technologies.

Based on other reports, 18A is already doing better than a 60% yield, not at 80% (production ready) though, but 20A was still sub 60%, it pointed out critical flaws that they fixed for 18A, and rather than try to make them work they are just scrapping it. Nobody wants a repeat of early 10nm ...
As 18A is just 20A++, they don't need both nodes to coexist, scrapping one to get the other up and at a finished production volume faster is a smart move.

 

[Absalom]

What process size are we actually on? I heard 14nm was true 14nm, but 7nm is actually 10nm and now you are saying 20A (2nm) is actually 5nm...

They are marketing terms at this point. It's literally whatever TSMC calls their node vs whatever Intel calls theirs. You'd need a chart to figure it all out.

 

[erek]

Hmm


View: https://youtu.be/1jlQ9Q-VPgs?si=xj5icOFcnHvFrnGH

 

[idiomatic]

Intel has survived more company killing strategic errors than I can count, including having a generation of hamburger managers in there for a while.

Boeings hamburger management generation has left them unable to execute at all. Intel, seems to have reset from that at least.

but writing something off as a test node.... ouch.

 

[HockeyJon]

Intel has survived more company killing strategic errors than I can count, including having a generation of hamburger managers in there for a while.

Boeings hamburger management generation has left them unable to execute at all. Intel, seems to have reset from that at least.

but writing something off as a test node.... ouch.

Intel owns a lot of the relationships with strategic PC OEMs like Dell, who service a significant amount of the commercial and consumer market. The sales teams of their partners have done a fantastic job of convincing people they’re the “premium brand” and they always lead with Intel. It would take a lot for any competitor to make inroads at that point, and it covers Intel for a lot of what would otherwise be catastrophic failures.

 

[toast0]

As 18A is just 20A++, they don't need both nodes to coexist, scrapping one to get the other up and at a finished production volume faster is a smart move.

It's a smart move if they've done what they claimed to make their designs more portable to other nodes. Intel 10nm was such a disaster because their fab didn't work and all the in progress designs were tied to the specs for the fab. Thus, an epic pipeline stall. If they have designs in the pipeline that required 20A, another failed node just reinvokes the pipeline stall (but yall said they were able to move arrow lake to other nodes).

Still, Intel's dominance through 14nm was consistent execution on die shrinks and redesigns, until they get back on a cadence of fab improvements roughly on schedule, their traditional design pipeline doesn't work well. And if Intel isn't running well, AMD doesn't stay motivated.

 

[GoodBoy]

I get the feeling that yields were not good, so saying it is cancelled (because it isn't viable anyway: costs too much to use due to low yields) buys them time to get shit working on 18A.

If Intel can actually get these advanced nodes working and start making their chips on these smaller processes, would save the company. This feels like a last ditch hail mary.