Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Intel proposes to catch up on process nodes by renaming them
https://www.anandtech.com/show/16823/intel-accelerated-offensive-process-roadmap-updates-to-10nm-7nm-4nm-3nm-20a-18a-packaging-foundry-emib-foveros
Their old 10 nm Enhanced SuperFin (or perhaps 10+++ or whatever) will now be called "7 nm". It's not an actual die shrink, but calling it 10 nm when TSMC and Samsung are on to 5 nm is getting embarrassing. Meanwhile, their old "7 nm" will now be called "4 nm". It probably would have been a lot better than Samsung's 7 nm, and likely also better than TSMC's 7 nm. But with Intel joining the other foundries with blatantly exaggerated process node names, this eliminates the last vestige of a connection between process node names and any physical dimension.
To some degree, it didn't really matter what Intel called their process nodes in the past. For decades, they wouldn't let anyone else use their fabs at all, and when they finally did, it was only a handful of very restricted customers. If Global Foundries had come up with creative node names to make it look like Intel was behind on process nodes, that wouldn't have convinced people to prefer a Bulldozer CPU over the much faster Sandy Bridge.
But now, Intel wants to be a major foundry like Samsung that creates chips for a lot of other customers, in addition to their own. Intel will probably never be a pure play foundry (i.e., one that doesn't design their own chips to sell, but only fabricates chips designed by others) like TSMC, so I wouldn't expect to see AMD CPUs fabricated by Intel anytime soon. But if you're trying to convince, say, Qualcomm to have their chips fabricated by you instead of Samsung or TSMC, there's no advantage to looking like you're way behind on process node nomenclature. Qualcomm, Apple, Nvidia, Bitmain and so forth aren't going to be fooled by creative process node names, but there's still no advantage to making it look like you're behind.
Intel will now do what TSMC and Samsung started doing around 12 nm or so: new nodes will get smaller numbers, even if it's not a true shrink. So there will be 4 nm, then 3 nm, then 2 nm. But there aren't very many positive integers smaller than 2, so instead of "2 nm", Intel is calling it "20 A", as in Angstroms. We'll have to see how long it takes before someone starts advertising their new 50 pm node, even though that's smaller than the radius of a silicon atom.
Their old 10 nm Enhanced SuperFin (or perhaps 10+++ or whatever) will now be called "7 nm". It's not an actual die shrink, but calling it 10 nm when TSMC and Samsung are on to 5 nm is getting embarrassing. Meanwhile, their old "7 nm" will now be called "4 nm". It probably would have been a lot better than Samsung's 7 nm, and likely also better than TSMC's 7 nm. But with Intel joining the other foundries with blatantly exaggerated process node names, this eliminates the last vestige of a connection between process node names and any physical dimension.
To some degree, it didn't really matter what Intel called their process nodes in the past. For decades, they wouldn't let anyone else use their fabs at all, and when they finally did, it was only a handful of very restricted customers. If Global Foundries had come up with creative node names to make it look like Intel was behind on process nodes, that wouldn't have convinced people to prefer a Bulldozer CPU over the much faster Sandy Bridge.
But now, Intel wants to be a major foundry like Samsung that creates chips for a lot of other customers, in addition to their own. Intel will probably never be a pure play foundry (i.e., one that doesn't design their own chips to sell, but only fabricates chips designed by others) like TSMC, so I wouldn't expect to see AMD CPUs fabricated by Intel anytime soon. But if you're trying to convince, say, Qualcomm to have their chips fabricated by you instead of Samsung or TSMC, there's no advantage to looking like you're way behind on process node nomenclature. Qualcomm, Apple, Nvidia, Bitmain and so forth aren't going to be fooled by creative process node names, but there's still no advantage to making it look like you're behind.
Intel will now do what TSMC and Samsung started doing around 12 nm or so: new nodes will get smaller numbers, even if it's not a true shrink. So there will be 4 nm, then 3 nm, then 2 nm. But there aren't very many positive integers smaller than 2, so instead of "2 nm", Intel is calling it "20 A", as in Angstroms. We'll have to see how long it takes before someone starts advertising their new 50 pm node, even though that's smaller than the radius of a silicon atom.
![[Deleted User]](http://w8.vanillicon.com/v2/8187bdfebcad0c6fe557549abc3ef848.svg)
Comments
------------
2024: 47 years on the Net.
Logic, my dear, merely enables one to be wrong with great authority.
Really it comes down to cost and performance - the size is tangential to that, and the marketing name for it less so. But the foundarioes aren't the only ones who want to have a snazzy marketing name - the companies that use them would like for it too - so they can differentiate from the competition. I'd almost rather them pick random nouns or made up names to name it after, rather than numbers which get confusing, deceptive and misleading.
After all, I'm sure AMD has certainly benefited from the fact that TSMC markets their process as 7nm, and 7 is a better number than 14+++++, 10, or 8. So much so that they even named a GPU release after it.
The reason AMD sold their foundries wasn't that they couldn't make new nodes anymore. It was that it was too expensive to do so, and AMD simply didn't have the volume to justify it.
That is ultimately the same reason why Intel is opening up their own fabs to other customers. Even though Intel's volume is massively higher than AMD's was a decade ago, they still don't have the volume to justify the cost of developing their own fabs without spreading the cost among other customers.
And it's also the reason why Global Foundries decided to abandon the cutting edge. Even though they bought AMD's fabs and IBM's fabs and also some others, and also had both AMD and IBM as customers as well as a number of others, even then, their volume still wasn't large enough to justify the cost of developing new nodes.
For that matter, it's the reason why dozens of other companies that used to have fabs dropped them and started hiring others to fabricate their chips. Every new process node costs more than the previous, and whatever your volume is, at some point, it's not enough to justify the cost of keeping your own fabs.