News Intel announces cancellation of 20A process node for Arrow Lake, goes with external nodes instead, likely TSMC [Updated]

[technologyiskiller]

Intel outsourcing to TSMC is not a good look. Normally you would use 20A to showcase to potential customers that your advanced nodes are yielding well, sufficient volume, and cost-efficient. Instead, customers see that even Intel is not confident in it's own internal nodes, preferring TSMC instead. So why would any customer abandon TSMC unless the US gov't forces it at gunpoint?

They have already outsourced for Lunar Lake. Even ARL also primarily was for TSMC. I don't think 20A would have been competitive enough to TSMC and HVM on this would have further delayed 18A. So instead of spending time and money on something which is not competitive it makes sense to move to the one which will be used for actual products.

 

[technologyiskiller]

Intel is not moving Arrow lake to 18A. 18A wont ramp until late 2025. Intel chose TSMC over 20A. 20A and 18A capacity are the same tool set.

Wait till you see what arrow lake refresh looks like in 2025. Hint its not 18A either LOL

the problem is that 20A isnt s good as N3..... Its not as good as Intel 3 either, apparently

What makes you say it's not as good as Intel 3?

 

[cpdx7]

Since Intel moved to half-nodes, 20A to 18A is like 14+ to 14+++. Nearly the same toolset, slight differences. There's no wasted effort here; can't get to 18A without having gone through 20A. As for LNL and ARL, these decisions were made many years ago and committed to TSMC before Intel 20A/18A started looking viable.

 

[Maltrer]

So, why did Intel even bother developing and spending so much money on the 20A process node, when they have already come so far with the more advanced 18A process? Just a couple of days ago it was reported that the 18A process node is doing remarkably well and has low defect rates, considering its development stage.

It's the same node.

 

[Pierce2623]

Do not try to to find negativity everywhere. From the beginning Intel said that 20A was never for mass production. They had multiple new technologies which were coming in like BSPDN and GAA so putting all of that in just one node is very risky. So 20A was used mostly as a stepping stone for 18A.

There is no point in commenting if you have no idea how developing a new process node works.

Intel DID NOT say from the beginning that 20a was never for production. They’ve obviously only recently decided that Arrow Lake can’t use it.

 

[Pierce2623]

Since Intel moved to half-nodes, 20A to 18A is like 14+ to 14+++. Nearly the same toolset, slight differences. There's no wasted effort here; can't get to 18A without having gone through 20A. As for LNL and ARL, these decisions were made many years ago and committed to TSMC before Intel 20A/18A started looking viable.

So why not just develop one node in the first place? There’s literally hundreds of millions of dollars in 20a if not billions.

 

[phead128]

It's very likely that 20A is unusable and 18A will face the same yield issues. Par for the course for Intel ever since debacles at 14nm and 10nm nearly a decade ago.

 

[NinoPino]

No it doesn't. Its just the way the press these days has made it.

If you don't think this is a negative fact maybe you think that abandoning a process node in the roadmap by years is positive, right ?

Most of the tech writers are incompetent and have absolutely zero idea about this (naturally because most of them have no idea about physics). So they just want to create a populist sensation by creating negative reports.

Of what tech writers are you talking ? And what are the wrong affirmations made in the article ?

 

[NinoPino]

It's the same node.

It is different in Intel's opinion and marketing material.

 

[technologyiskiller]

Intel DID NOT say from the beginning that 20a was never for production. They’ve obviously only recently decided that Arrow Lake can’t use it.

That's not possible. Whether they said it or not you can't just change a design from one node to another in a matter of days (or even months). If they change the design to a different node now, it will take a long time to redesign (technically called porting) the product on this node.

 

[technologyiskiller]

If you don't think this is a negative fact maybe you think that abandoning a process node in the roadmap by years is positive, right ?


Of what tech writers are you talking ? And what are the wrong affirmations made in the article ?

Most of the tech writers have no idea about how a new node design works (that's obvious because they are not engineers). But problem happens when they just assume something based on their half baked knowledge.

In this case the write up just assumed that 20A was not needed but it might have been needed because it was a stepping stone.

 

[technologyiskiller]

It's very likely that 20A is unusable and 18A will face the same yield issues. Par for the course for Intel ever since debacles at 14nm and 10nm nearly a decade ago.

Intel has publicly mentioned that 18A D0 is < 0.4 (if you understand D0). That is a quite good number at this stage of a new node.

Now I have no idea about the performance and power draw but at least from yield point of view it looks ok for now.

"Very likely" and similar words are your personal bias. They are not facts.

 

[Pierce2623]

That's not possible. Whether they said it or not you can't just change a design from one node to another in a matter of days (or even months). If they change the design to a different node now, it will take a long time to redesign (technically called porting) the product on this node.

Yeah and they’re having to port the architecture to TSMC n3 because it was originally designed for Intel nodes. They tried as hard as they could to stay on an Intel node as they have SIGNIFICANTLY less profit on chips fabbed at other foundries. When they make them in the US, and do the package in the US it saves them ungodly amounts of money compared to having them fabbed and made into a package at TSMC.

 

[technologyiskiller]

Yeah and they’re having to port the architecture to TSMC n3 because it was originally designed for Intel nodes. They tried as hard as they could to stay on an Intel node as they have SIGNIFICANTLY less profit on chips fabbed at other foundries. When they make them in the US, and do the package in the US it saves them ungodly amounts of money compared to having them fabbed and made into a package at TSMC.

It's not the architecture, it's the physical implementation which is dependent on the node. And that takes months to do for a given node.

Essentially all I am saying is the design was already targeted for external process and this announcement doesn't change anything.

 

[Co BIY]

It's not the architecture, it's the physical implementation which is dependent on the node. And that takes months to do for a given node.

Essentially all I am saying is the design was already targeted for external process and this announcement doesn't change anything.

I think the automated design tools in use now may make this porting process much easier and shorter than in the past. Especially if the design tools and node were created with a foundry process in mind.