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!
"Intel Foundry has an embarrassment of riches: amid a shortage of advanced-packaging capacity, the company has excess supply.
The manufacturing unit of the largest U.S. chip company is welcoming TSMC customers to transfer their designs directly from TSMC’s CoWoS to Intel’s Foveros advanced packaging. Demand for 3D wafer-level packaging from Nvidia and a handful of companies making AI chips has exceeded the ability of top foundry TSMC to meet demand since last year. "
"Intel Foundry has an embarrassment of riches: amid a shortage of advanced-packaging capacity, the company has excess supply.
The manufacturing unit of the largest U.S. chip company is welcoming TSMC customers to transfer their designs directly from TSMC’s CoWoS to Intel’s Foveros advanced packaging. Demand for 3D wafer-level packaging from Nvidia and a handful of companies making AI chips has exceeded the ability of top foundry TSMC to meet demand since last year. "
Is it just because Intel didn't do enough marketing, as mentioned in the interview, or is there something else that caused Intel's advanced packaging capacity to remain idle for external customers?
Is it just because Intel didn't do enough marketing, as mentioned in the interview, or is there something else that caused Intel's advanced packaging capacity to remain idle for external customers?
Do potential customers trust Intel to execute ? (18A, back-side power, 2.5D, 3D, EMIB, ...)
Ponte Vecchio was a big black eye for Intel execution. (years late, canceled immediately after partial delivering for 1 data center, $300M write-off for not delivering)
Do potential customers trust Intel to execute ? (18A, back-side power, 2.5D, 3D, EMIB, ...)
Ponte Vecchio was a big black eye for Intel execution. (years late, canceled immediately after partial delivering for 1 data center, $300M write-off for not delivering)
Well that is true for new technologies but not for technologies that exists EMIB is not new and neither is foveros.
They have shipped 40Million CPUs with foveros last year for client and has millions of Xeons to DC for few years containing EMIM(SPR/EMR/GNR/SRF).
I'm not convinced people will buy wafers from one foundry and package them at another on a regular basis. TSMC is building packaging facilities at a rapid pace. They are much easier to build than a fab for sure. I had heard TSMC is more than doubling packaging capacity every year. We will get an update next month at the TSMC Symposium and the TSMC investor call. My guess is that they will say CoWos is booked up but not on allocation. Never will TSMC say "we have more CoWoS capacity than we know what to do with.
TSMC North America customers are very loyal and want to tell their investors and customers that their chips are made here in the good old USA. I see that happening in AZ over the next 5 years, absolutely.
Well that is true for new technologies but not for technologies that exists EMIB is not new and neither is foveros.
They have shipped 40Million CPUs with foveros last year for client and has millions of Xeons to DC for few years containing EMIM(SPR/EMR/GNR/SRF).
I'm not convinced people will buy wafers from one foundry and package them at another on a regular basis. TSMC is building packaging facilities at a rapid pace. They are much easier to build than a fab for sure. I had heard TSMC is more than doubling packaging capacity every year. We will get an update next month at the TSMC Symposium and the TSMC investor call. My guess is that they will say CoWos is booked up but not on allocation. Never will TSMC say "we have more CoWoS capacity than we know what to do with.
TSMC North America customers are very loyal and want to tell their investors and customers that their chips are made here in the good old USA. I see that happening in AZ over the next 5 years, absolutely.
Can Intel license the CoWoS process from TSMC and begin receiving orders from TSMC's customers? Amkor is already doing this, so why can't Intel take the same approach instead of asking TSMC's CoWoS customers to transition to Intel's technologies? In the contract manufacturing business, the foundry should prioritize accommodating customers, not the other way around.
Granted, TSMC may be reluctant to agree, but this presents an opportunity for Intel's new CEO Lip-Bu Tan to demonstrate his persuasive abilities and his great relationship with TSMC.
Can Intel license the CoWoS process from TSMC and begin receiving orders from TSMC's customers? Amkor is already doing this, so why can't Intel take the same approach instead of asking TSMC's CoWoS customers to transition to Intel's technologies? In the contract manufacturing business, the foundry should prioritize accommodating customers, not the other way around.
Granted, TSMC may be reluctant to agree, but this presents an opportunity for Intel's new CEO Lip-Bu Tan to demonstrate his persuasive abilities and his great relationship with TSMC.
I doubt TSMC would do that. The relationship between Pat Gelsinger and CC Wei was not good. Maybe Lip-Bu can fix that. Lip-Bu and CC Wei are well acquainted. Packaging is not a high margin business but it is another reason for customers to stay with TSMC.
Maybe there is some form of bundling of services together. If a customer were to fab at tsmc and not package there, they may need to pay higher fees or lower in priority. Not sure if that actually happened but I see that being done in many other industries even if that can be deemed as monopolistic practices
Is there such a big difference between Intel and TSMC in terms of packing?
Who leads who it's not clear for me at least.
Intel dishes out slides they have more advanced packing than TSMC but I doubt it. It may be true
Everyone says they have the most advanced advanced packaging. Be it ASE, TSMC, SK Hynix, Intel, Samsung, JCET, or AMKOR. Marketing slides are there to market the product.
For Cu-Cu hybrid bonding TSMC leads. Intel doesn't launch their first product until early 2026 and TSMC did their first in like 2022. Soon (like next year if memory serves) TSMC is going from 9 to 4 micron bump pitch so even when Foveros direct finally makes an appearance TSMC will still lead Intel. For 2.5D solutions I would say that big picture Intel leads TSMC. Si bridge technology is simply better than Si interposers from a scalability and cost perspective. Hence why TSMC is moving towards CoWoS bridge to replace their interposers. But like Intel with Cu hybrid bonding TSMC is simply behind 2017's Intel EMIB. Intel also has better full stack integration than TSMC with their turn key singulated die test/sort/package test assembly/test service s. As opposed to TSMC customers which need to go through an OSAT while also protecting TSMC IP from their OSAT competitors. Intel also should have scale advantages. I doubt TSMC makes more CoWoS/SoIC units per year than Intel makes core ultra plus all Xeons in a year.
Is it just because Intel didn't do enough marketing, as mentioned in the interview, or is there something else that caused Intel's advanced packaging capacity to remain idle for external customers?
I mean Intel cut client unit shipments in half of what they projected to investors back in 2022 and back when Intel announced fab push outs they also paused Malaysia start up until NM was filled. So it isn't exactly news that Intel over built their advanced packaging network expecting strong demand from Intel products. Advanced packaging while growing rapidly is a naccient low volume market compared to wafer making. NVIDIA and Intel products effectively are the 2.5D market. There really isn't that much in the grand scheme of things to fight over or win IMO. Back in like mid 2024 Intel said they have 9 18A customers and even more advanced packaging wins. While 10+ customers sounds significant I don't think even with something crazy like 30 customer wins (if there are even 30 customers to win) would make up for the floor falling from under Intel products' feet.
Why do that if Intel made Foveros and recently EMIB transparent with some customers not even needing any redesign work? Especially since EMIB is a suppior technology, and TSMC would have no interest in giving away their hot new technology (that and I can't think of TSMC ever licencing any process technology). The closest thing is SSMC (which TSMC partially owns) and AMKOR (but they just do the test portion and the organic stuff after the TSMC bit is done). Licencing would also reduce diferentation and forces a price war that would hurt both of them.
