It's actually TSMC trying to catch up with Intel. Just fyi, they've both been in this business for close to 3 to 4 decades. And Intel had always been the leader. In 2017, Intel lost footing and started doing the unpleasant 14+++ thing. Coupled with the 10nm delays, their foundry lost a total of 4 years doing nothing.
Then came Pat Gelsinger.
He ditched the existing 7nm node. Added EUV & multi-patterning and turned it into a 4nm node. Intel 4nm is actually on par with current TSMC 3nm. And is already ahead of schedule by 6 months!
He then introduced Intel 3nm which is also ahead of schedule by 6 months and will be on par with the upcoming TSMC N3 variants (e/p/s/x).
And then, the most unexpected & unimaginable thing happened. Intel announced that it's 20A & 18A are also very much ahead of schedule! Meaning, Intel 2nm will volume ramp in 2024 while TSMC 2nm is expected to be production ready ONLY IN 2026! This is hard to believe but true.
I wouldn't be so sure of Intel 18A yield as it relies very heavily on multi-patterning for increase in MTr/mm2. Ring any bells? Might be a repeat of 10++++. That's why they've confined it to a niche product: A 15W LP cpu. Just a moonshot/sideshow. Might not make it on time.
In essence, they have now progressed much better than TSMC. EUV, Tiles, 3D Stacking, Disaggregated IP, GAA, BPD. The amount of progress made by Intel has TSMC worried already. But there's only so much tsmc can do. Their ROI is 3 years minimum i think for a new node running at full capacity. Cos, they have to have different types of nodes, libraries, etc to support all types of customers.
Intel on the other hand, Intel doesn't have that many generic nodes (except Intel 3). Most nodes are designed to service very specific needs. So, the nodes remain lean & mean & nimble and a lot cheaper than a comparable TSMC node. For an custom Intel node, the ROI is roughly half that of TSMC.
So, TSMC is now trying to catch up with Intel Innovation. Otherwise, starting 2024, TSMC will lose leadership is all aspects.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
9 Comments
Back to Article
Exotica - Thursday, October 27, 2022 - link
How does this 3DBlox initiative compare and/or compete with UCIe?thestryker - Thursday, October 27, 2022 - link
The way I read it 3DBlox should incorporate UCIe as it seems to have rather broad scope.Threska - Thursday, October 27, 2022 - link
Well Backside Power Delivery is what's really needed.https://www.fabricatedknowledge.com/p/backside-pow...
nandnandnand - Thursday, October 27, 2022 - link
It's supposed to lower power consumption, right? I didn't see that in the article or maybe I skimmed over it.Lakados - Saturday, October 29, 2022 - link
Is this TSMC trying to put together an alternative to Intels tiles?evanh - Tuesday, November 22, 2022 - link
Nope, this is TSMC generalising for regular contract work what was custom for AMD for the past four years.SiliconFly - Thursday, December 8, 2022 - link
It's actually TSMC trying to catch up with Intel. Just fyi, they've both been in this business for close to 3 to 4 decades. And Intel had always been the leader. In 2017, Intel lost footing and started doing the unpleasant 14+++ thing. Coupled with the 10nm delays, their foundry lost a total of 4 years doing nothing.Then came Pat Gelsinger.
He ditched the existing 7nm node. Added EUV & multi-patterning and turned it into a 4nm node. Intel 4nm is actually on par with current TSMC 3nm. And is already ahead of schedule by 6 months!
He then introduced Intel 3nm which is also ahead of schedule by 6 months and will be on par with the upcoming TSMC N3 variants (e/p/s/x).
And then, the most unexpected & unimaginable thing happened. Intel announced that it's 20A & 18A are also very much ahead of schedule! Meaning, Intel 2nm will volume ramp in 2024 while TSMC 2nm is expected to be production ready ONLY IN 2026! This is hard to believe but true.
I wouldn't be so sure of Intel 18A yield as it relies very heavily on multi-patterning for increase in MTr/mm2. Ring any bells? Might be a repeat of 10++++. That's why they've confined it to a niche product: A 15W LP cpu. Just a moonshot/sideshow. Might not make it on time.
In essence, they have now progressed much better than TSMC. EUV, Tiles, 3D Stacking, Disaggregated IP, GAA, BPD. The amount of progress made by Intel has TSMC worried already. But there's only so much tsmc can do. Their ROI is 3 years minimum i think for a new node running at full capacity. Cos, they have to have different types of nodes, libraries, etc to support all types of customers.
Intel on the other hand, Intel doesn't have that many generic nodes (except Intel 3). Most nodes are designed to service very specific needs. So, the nodes remain lean & mean & nimble and a lot cheaper than a comparable TSMC node. For an custom Intel node, the ROI is roughly half that of TSMC.
So, TSMC is now trying to catch up with Intel Innovation. Otherwise, starting 2024, TSMC will lose leadership is all aspects.
SarahKerrigan - Wednesday, December 14, 2022 - link
Hi Pat! How's the weather in Santa Clara?Zoolook - Monday, December 26, 2022 - link
LOL even Intels PR department couldn't come up with this, I think you should add an /s otherwise some people will think you are serious!