You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Depends on how you mean. Assuming that we are talking revenue, and we aren't talking about maybe Intel consistently out-executing TSMC for the next 20-30 years. Then TSMC needs to have at least 3 10 nanometers in a row. The most recent 2 TSMC nodes have for the past little while have been around half of TSMC's revenue. So Intel needs to win 100% of 2 nodes in a row to be around the size of TSMC. But if we are talking technology as long as TSMC and Intel's roadmaps hold, the current roadmap from TSMC looks weak to me (3 years between full nodes versus Intel's 2 years with the Intel half nodes being bigger steps and the TSMC full nodes being similar sized steps). 18A is more advanced than N2 in some ways, N2 is more advanced than 18A in some ways. The wind seems to be blowing towards N2 being on average better. Foundry products come around a similar time frame, and Intel 18A products lead Apple N2 by about a year. 14A is easily lining up to be better than N2, and if Intel is serious about 2 year cadence, we are talking internal products beating first A14 products to market by 2 years and intel foundry 14A products should lead by around 1 year. Then by the time Apple launches their A14 chips and starts A14 production for everyone who isn't Apple, we should be seeing 10A Intel chips and 10A foundry production start. Even if we assume TSMC is sandbagging by 5% on all of the numbers TSMC showed for A14, I just cannot comprehend how that could be competitive with whatever Intel has for 10A unless 10A is the biggest nothingburger of a node in semiconductor history. Intel even has the advantage of their BSPDN application, allowing them to be 1-2 nodes behind TSMC on litho scaling while maintaining the same density.
While we are now getting waaaaaaaaaaay into too far out for anyone to really known anything, but just for fun (and also to illustrate just how bad a 3 year cadence looks to me). I know Dan Hutcheson believes that from the research progress he has seen and Intel's demonstrated process development capabilities that he thinks Intel will be first to CFETs sometime during the 1st half of the 2030s. For their part IMEC believes CFET will be around A7 generation in 2031. If that is what bears out, then Intel 7A CFET products on shelves with foundry customers starting first wafers 6 years after 18A in 2031 versus A14P/A14P+BSPDN products on shelves and A10 forksheets only just entering production. Now FWIW, I don't think TSMC will just completely roll over. But without working on more processes in parallel, or having each full node be a big enough improvement to overcome the slower cadence, it is not possible to maintain process leadership. I feel like TSMC has to have something big so Intel doesn't just gradually open up another multi-year process lead. We just don't know what it is at the moment. In a sense, this is the same dilemma the industry encountered when process complexity was increasing exponentially in the heavy multipatterning era. Intel chose to extend development times but have more aggressive shrinks to compensate. On paper the CAGR was greater than what TSMC was doing, but it was too hard to execute and Intel fell behind. TSMC responded with keeping doing more modest process node definitions with gaps not to exceed 2 years starting with 16FF->10FF until the 3 year gap between N5->N3 (and with 10FF/N7 they even had a 1 year gap between process nodes).
Depends on how you mean. Assuming that we are talking revenue, and we aren't talking about maybe Intel consistently out-executing TSMC for the next 20-30 years. Then TSMC needs to have at least 3 10 nanometers in a row.
Let's not forget TSMC did not enter the world of FinFETs gracefully either. 20nm was not good and 16nm was late enough to lose half the Apple business for the iPhone 6+.
Please say "Intel just needs to out execute TSMC for the next 10 years" like that's an easy thing to do. First, TSMC is very good at executing, and execution comes down to having the right org structure and culture. Second, executing on the wrong strategy is worse than not executing at all, and Intel has the wrong strategy.
Please say "Intel just needs to out execute TSMC for the next 10 years" like that's an easy thing to do. First, TSMC is very good at executing, and execution comes down to having the right org structure and culture. Second, executing on the wrong strategy is worse than not executing at all, and Intel has the wrong strategy.
IMO IDM 2.0 strategy is not bad otherwise Lip Bu would have given us a new strategy.
Intel killed/ sold off many businesses in last 4 years that were bad but foundry is not one of them
IMO IDM 2.0 strategy is not bad otherwise Lip Bu would have given us a new strategy.
Intel killed/ sold off many businesses in last 4 years that were bad but foundry is not one of them
It was many people's opinion that the IDM business model was still viable for the last 10 years and they have all been proven wrong. If Lip Bu does not realize this soon he will be another person on the scrap heap of former Intel CEOs (honestly not a bad place to be since most of these guys collect tens if not hundreds of millions on their way out).
It was many people's opinion that the IDM business model was still viable for the last 10 years and they have all been proven wrong. If Lip Bu does not realize this soon he will be another person on the scrap heap of former Intel CEOs (honestly not a bad place to be since most of these guys collect tens if not hundreds of millions on their way out).
The last point doesn't get enough attention for Intel's foundry roadmap. If Intel Products is the vast majority of revenue until at least 2027, Foundry still needs Products to do well. If for some reason design can't fulfill their end of the bargain, the resulting volume and margin impact makes the foundry side more unsustainable which creates an ugly feedback loop.
The last point doesn't get enough attention for Intel's foundry roadmap. If Intel Products is the vast majority of revenue until at least 2027, Foundry still needs Products to do well. If for some reason design can't fulfill their end of the bargain, the resulting volume and margin impact makes the foundry side more unsustainable which creates an ugly feedback loop.
Intel is a relatively small percentage of TSMC revenue, and if Intel misses a product cycle, chances are people will be buying chips from AMD or someone else who can pick up the capacity.
Intel is a relatively small percentage of TSMC revenue, and if Intel misses a product cycle, chances are people will be buying chips from AMD or someone else who can pick up the capacity.
Can you clarify what you mean here? Are you saying Intel is not a relatively small percentage of TSMC revenue? Or something else. Because I would guess than Intel is less than 20% of TSMC revenue vs 90%+ of Intel Foundry revenue. Who is going to have a bigger problem if Intel product volumes drop?
Can you clarify what you mean here? Are you saying Intel is not a relatively small percentage of TSMC revenue? Or something else. Because I would guess than Intel is less than 20% of TSMC revenue vs 90%+ of Intel Foundry revenue. Who is going to have a bigger problem if Intel product volumes drop?
i mean in terms of the amount not a percentage of TSMC revenue can't find the exact number but it's a big amount that is spent elsewhere vs in your biz to offset the cost due to poor planning.
Can you clarify what you mean here? Are you saying Intel is not a relatively small percentage of TSMC revenue? Or something else. Because I would guess than Intel is less than 20% of TSMC revenue vs 90%+ of Intel Foundry revenue. Who is going to have a bigger problem if Intel product volumes drop?
Based on the information from 2024 Intel Foundry day and the quarterly earnings filings, I believe the way the accounting is being done is as follows:
1. Intel Products pays Intel Foundry "market price" for a comparable wafer (e.g. Intel 7 wafer pricing = n7 market rate, Intel 3 = n3 rate etc.).
2. Intel Products signs up to buy a guaranteed # of wafers from Intel Foundry.
3. Currently Intel Foundry's volume is dominated by Intel 7 whose market price is way lower compared to Intel 3 and Intel 18A. David Zinsner keeps referring to ASP for Intel 3 and Intel 18A wafers being ~3x Intel 7.
4. As Intel's product mix moves to more Intel 3 and 18A, foundry losses should narrow.
5. If Intel Product can't make competitive products and can't sell the volume, the corresponding losses will be attributed to Intel Products, with overall Intel financials taking a hit, but not necessarily Intel Foundry. The earnings statement clearly separates Products and Foundry financials with Inter-segment reconciliation for overall Intel financials.
Therefore, Intel Products volume dropping may be a problem for Intel Products and Intel, but may not be an issue for Intel Foundry financials, if wafers are pre-bought in guaranteed quantity.
Let's not forget TSMC did not enter the world of FinFETs gracefully either. 20nm was not good and 16nm was late enough to lose half the Apple business for the iPhone 6+.
TSMC process RnD is never an "all-in", TSMC always develops both a more, and less conservative version of the node, some times without a key node feature at the same time.
At worst, they will have a half-node ready in time to offer to clients under the guise of a full node shrink. That was certainly the case with copper interconnects, and HKMG.
