TSMC plans to expand production with a 400 billion silver bullet and buys EUV. These Taiwanese manufacturers are extremely excited.

[Daniel Nenni]

2nm advanced process is the key to mass production! A total of more than 60 units will be delivered this year and next, with a total investment estimated to exceed 400 billion.

TSMC's 2nm advanced process capacity will be mass-produced in 2025, and equipment factories are in full swing, especially the EUV (extreme ultraviolet exposure machine) used in the advanced process, which is critical. A total of more than 60 EUV units will be delivered this year and next. The total investment amount exceeds 400 billion yuan. With the continuous expansion of production capacity, ASML's delivery volume will increase by more than 30% in 2025, and the supply chain of Taiwanese manufacturers will benefit. Among them, Jiadeng is actively working with ASML to invest in the research and development of next-generation High-NA EUV. In addition, Fanxuan, Itex, Gongzhun, Jingding and Xiangming are expected to benefit simultaneously.

Equipment industry insiders revealed that the supply of EUV machines is tight and the delivery time is as long as 16 to 20 months. Therefore, most of the orders in 2024 will be delivered the year after next. According to legal estimates, TSMC’s EUV orders will reach 30 units this year and 35 units next year. However, There will be slight adjustments due to capital expenditures. Apart from the R&D center, High-NA EUV has no plans for mass production equipment next year.

ASML planned new production capacity last year in response to customer demand, and the growth in delivery volume next year is clear. The total number of deliveries is expected to reach 53 units this year, and more than 72 units next year.

The supply chain pointed out that ASML's 2025 annual production capacity plan, the goals of 90 EUV units, 600 DUV units, and 20 High-NA EUV units have not changed. The 2024 Investor Day will be held on November 14 this year, and the latest blueprint will be proposed for the next five years.

TSMC’s advanced process production capacity is gradually opening up. For example, the 3-nanometer Tainan plant will enter mass production in the third quarter. Next year, the P8 plant will also gradually introduce EUV machines. Hsinchu Baoshan 2-nanometer has continued its EUV sales momentum for three years. Qiangqiang and Kaohsiung 2nm are also in progress simultaneously.

The supply chain also revealed that the additional funds for TSMC's U.S. P1A project are expected to be received in the third quarter, and the factory has entered the final stage of construction. In addition, TSMC has construction needs for other domestic and overseas factories, such as Baoshan, Hsinchu, Kumamoto, Japan, Nanzi, Kaohsiung, and packaging and testing. Factories, etc., CSP data center construction needs, equipment demand is not optimistic.

More EUV machines have driven the simultaneous growth in the use of EUV mask boxes. The legal person believes that Jiaden will be the most benefited among them. FOUP (front-opening wafer transfer box) continues to capture market share. Compared with competition As a competitor, Jiadeng builds its own hot air extraction environment to remove plastic particles or pollutant particles and sends them directly to the customer's factory for use. At the same time, ASML High-NA EUV Jiadeng has also invested in cooperative research and development.

The production environment is better than the clean rooms of large manufacturers. Jiaden is confident to provide customers with better product quality and services. With the continuous expansion of production capacity in the second half of the year, it will gradually catch up with customer demand next year and drive operational growth.

https://www.ctee.com.tw/news/20240628700065-439901

 

[Daniel Nenni]

This makes sense. TSMC N2 is again taking all of the big design wins while Samsung gets the crumbs. Intel 18A has internal Intel design wins to get started and I would bet that Intel 18A wafers by far outnumber what Samsung will produce at 2nm.

 

[fansink]

This makes sense. TSMC N2 is again taking all of the big design wins while Samsung gets the crumbs. Intel 18A has internal Intel design wins to get started and I would bet that Intel 18A wafers by far outnumber what Samsung will produce at 2nm.

Has anyone heard of any "external" 20A or 18A design wins, rumored or otherwise?

 

[Daniel Nenni]

Has anyone heard of any "external" 20A or 18A design wins, rumored or otherwise?

I don’t think 20A is being offer as a foundry node. 18A certainly is and I’m sure we will hear about it on the next investor call.

 

[Jumper]

I like how Intel thought they would just outspend TSMC. But it will be hard or even impossible to outspend or outinnovate them in upcoming years.

 

[osnium]

I like how Intel thought they would just outspend TSMC. But it will be hard or even impossible to outspend or outinnovate them in upcoming years.

Do you think Intel’s attempts to go into the foundry business is an exercise in futility? If you can’t outspend and innovate and if you have intrinsically higher cost structures, what’s the point?

 

[Jumper]

Do you think Intel’s attempts to go into the foundry business is an exercise in futility? If you can’t outspend and innovate and if you have intrinsically higher cost structures, what’s the point?

That's exactly what I think. If they had plethora of customers already it would be different. This seems to be trying to use war chest that's drying out. Meanwhile TSMC has war chest that's filling up.

 

[Fred Chen]

This article looks like it is for promoting the awareness of some Taiwanese companies involved in supporting fab equipment. Its estimates on EUV units sold haven't matched up with the sales record.

 

[siliconbruh999]

Catching upto tsmc is not possible in near future unless something like 10nm happens to TSMC which I doubt

 

[MKWVentures]

Catching upto tsmc is not possible in near future unless something like 10nm happens to TSMC which I doubt

No one is catching up to TSMC but you are 100% correct on the scenario IMO. As I have mentioned before, for Intel to gain significant market share in next 10 years:
1) have strong technology and demonstrated delivery success with Internal ramp and foundry
2) pricing to match TSMC (probably needs to be lower than TSMC)
3) TSMC has to have a problem (business or technical) that Intel jumps on. TSMC has taken major customers from Samsung like this.

 

[Daniel Nenni]

Do you think Intel’s attempts to go into the foundry business is an exercise in futility? If you can’t outspend and innovate and if you have intrinsically higher cost structures, what’s the point?

I think Intel out innovates TSMC on the IDM side all of the time, they have in the past (HKMG, FinFET, BSPD, etc...). In regards to the foundry side it is much more difficult since customers must adopt the technology as well. TSMC is the master of customer relations and risk adverse technology adoption. It will take years for Intel to have this level of foundry customer experience.

At the Design Automation Conference I discussed BSPD with many companies in the ecosystem. It was even discussed on the panel I moderated on 3D IC with AMD, Intel, QCOM, and Imec. The bottom line is the the foundry ecosystem is not ready for BSPD yet (there are thermal issues). The TSMC version is BSPD light which was spec'd out through customer collaboration for TSMC A14. I'm sure Intel will have BPSD in production with their internal products but all of the big customer design starts I am familiar with are using TSMC N2 which does not support BSPD.

 

[prime007]

No one is catching up to TSMC but you are 100% correct on the scenario IMO. As I have mentioned before, for Intel to gain significant market share in next 10 years:
1) have strong technology and demonstrated delivery success with Internal ramp and foundry
2) pricing to match TSMC (probably needs to be lower than TSMC)
3) TSMC has to have a problem (business or technical) that Intel jumps on. TSMC has taken major customers from Samsung like this.

It doesn't even seem possible for Intel right now? Coming from a non-semi pro point-of-view...

1. 1) The current (and unresolved) debacle with Intel's 13th and 14th gen CPUs (i9-13900K and i914900K) I imagine isn't going to inspire much confidence in their foundry. The unfortunate timing that their next chip (15th gen Arrow Lake) utilizes TSMC rather than Intel Foundry means that they can't "prove" that their process tech is reliable until their 18A node. If Daniel says that BSPD isn't ready, then what "strong technology" does Intel have left to show?
   
   2) Has Intel (Foundry) handicapped itself when it partnered with Apollo to build Fab 34 in Ireland (Thank you Mooredaddy for the original idea/post)? Maybe not since TSMC keeps raising their prices on the advanced nodes? That said having to 'do more with less' against TSMC doesn't sound like a winning strategy in the long run.

3) There is that ever-present threat from China. But outside of that, waiting/hoping for TSMC to fail isn't viable for either Intel/Samsung.

By the way, anyone wonder how long can Samsung continue to sustain its foundry business? If Intel is going to take their non-TSMC customers, then when is keeping up with TSMC going to be even too much for even Samsung?! It really sounds like China's memory and display companies are starting to become extremely fierce competitors to Samsung/LG/SK Hynix.

 

[MKWVentures]

It doesn't even seem possible for Intel right now? Coming from a non-semi pro point-of-view...

Its possible. If Intel out executes others and focuses on next 18 months rather than 2030. Also Intel has this whole "foundry to the foundry" business (UMC/Tower/others) that is pretty interesting. I wrote about the "Reasons Intel doesnt need to beat TSMC to win" in a previous blog.

IMHO Samsung Foundry has attacks from all sides. they have lots of plans and announcements, they need to focus on executing (Like Intel)

 

[MKWVentures]

I think Intel out innovates TSMC on the IDM side all of the time, they have in the past (HKMG, FinFET, BSPD, etc...). In regards to the foundry side it is much more difficult since customers must adopt the technology as well. TSMC is the master of customer relations and risk adverse technology adoption. It will take years for Intel to have this level of foundry customer experience.

At the Design Automation Conference I discussed BSPD with many companies in the ecosystem. It was even discussed on the panel I moderated on 3D IC with AMD, Intel, QCOM, and Imec. The bottom line is the the foundry ecosystem is not ready for BSPD yet (there are thermal issues). The TSMC version is BSPD light which was spec'd out through customer collaboration for TSMC A14. I'm sure Intel will have BPSD in production with their internal products but all of the big customer design starts I am familiar with are using TSMC N2 which does not support BSPD.

When will Intel ship a product with BSPD? How will foundry customers deal with 18A... or is there a non-BSPD version?

 

[samwilde]

Intel just by itself can no longer use the full capacity that is offered with the latest nodes. They tried to mitigate this with their Altera acquisition, which would increase fab utilization by also making FPGAs, but that did not go so well. Of course they have to attract 3rd party users to their fabs.

I doubt Intel's Foundry business will be able to capture the largest players without more formally breaking ties with Intel. I think the foundry business needs to be made independent long term.

 

[VCT]

At first, Intel came out as new and an additional competitors that could hurt TSMC. However it turn out
Intel and Samsung now have to compete for a limited pool of customers and orders. This has reduced the economic scale of the second-largest foundry compared to before. which hurt their learning curve.