hist78
Well-known member
"Japanese startup foundry Rapidus is set to open the company’s 2-nm pilot fab in April 2025, CEO Atsuyoshi Koike told EE Times in an interview. "
Rapidus Set to Open 2-nm Pilot Fab, CEO Says
- Thread starter hist78
- Start date
It looks like Samsung will lose yet another foundry customer….. IBM.
"Koike is counting on the support of IBM and global R&D organization imec to start commercial production of the world’s most advanced silicon just two years behind industry heavyweight TSMC. Koike’s single-wafer processing innovation will provide Rapidus a competitive advantage over TSMC and Samsung by accelerating cycle time, he says."
It looks more likely Rapidus buy double insurances of 2nm technologies from IBM and imec.
hist78
Well-known member
In 2015 IBM gave its semiconductor manufacturing division and $1.5 billion to GlobalFoundries but keeps the semiconductor research organization inhouse. I guess IBM semiconductor Research probably has less than 1,000 people working on the advanced semiconductor manufacturing. It's not enough to help Rapidus to go to high volume production in a reasonable timeframe. IMEC's capabilities and experience will definitely help.
Last edited:
IBM Semiconductor is now a patent factory. Patents do not get you into semiconductor HVM however.
hist78
Well-known member
It may not be good for Samsung but financially it's probably insignificant due to IBM's low volume orders.
Should Intel be worried about Rapidus? Intel is also new to the foundry business just like Rapidus.
Mooredaddy
Well-known member
No one should be worried about Rapidus. This venture is so just unbelievably goofy in its goals and ambitions
hist78
Well-known member
I think Rapidus is probably in the very last section of the list that Intel needs to worry about.
MKWVentures
Moderator
Intels foundry plans are challenging. They seem 100 times more credible than Rapidus. it all starts with "New company, no history, aggressive timeline, based on mythical IBM technology"
I will change my opinion IF
1) Rapidus ships lower volume products from pilot line in 2025 that are publicly displayed (demo server with Rapidus based CPU)
2) Rapidus partners with Samsung, TSMC, Intel or even Global on foundry
3) Rapidus hires me to do business development
I personally think 2 will happen...
hist78
Well-known member
Intels foundry plans are challenging. They seem 100 times more credible than Rapidus. it all starts with "New company, no history, aggressive timeline, based on mythical IBM technology"
I will change my opinion IF
1) Rapidus ships lower volume products from pilot line in 2025 that are publicly displayed (demo server with Rapidus based CPU)
2) Rapidus partners with Samsung, TSMC, Intel or even Global on foundry
3) Rapidus hires me to do business development
I personally think 2 will happen...
I think #3 is probably more likely.
#2 scenario has some big challenges:
1. Japanese government subsidies won't be used on helping Samsung, a Korean company.
2. TSMC has no such need.
3. Intel 5 nodes 4 years strategy and IDM 2.0 plan make the partnership with Rapidus unnecessary and less profitable.
4. The grievance and lawsuits between GlobalFoundries and IBM need to be resolved first.
I think #3 is probably more likely.
#2 scenario has some big challenges:
1. Japanese government subsidies won't be used on helping Samsung, a Korean company.
2. TSMC has no such need.
3. Intel 5 nodes 4 years strategy and IDM 2.0 plan make the partnership with Rapidus unnecessary and less profitable.
4. The grievance and lawsuits between GlobalFoundries and IBM need to be resolved first.
Rapidus will be for Japan customers in my opinion. I do not see them being able to replicate the required ecosystem to be competitive in the general foundry market. I do not know however if Japan is a big enough customer to keep Rapidus in business. We shall see. Japan is very big on mature nodes but what are they doing at the leading edge?
MKWVentures
Moderator
that is true for demand. and if they were planning 7-16nm I would agree with you. but Single wafer processing with a 2nm process from IBM (or IMEC) in 2026 volume production? get your popcorn, this will be interesting.
When you decide at government level that you do not want to depend for some critical components from TSMC, than the Japanese way, even if a nearly impossible one, is the only one. And I would not completely discount Koike-san and Higashi-san capabilities in driving this venture. Is not like they are announcing they will run TSMC out of business and develop 5 generations in 4 years. The fab is a small one that looks for very special business, the one that would probably not even be accepted by TSMC, and that makes customers a lot more comfortable in dealing with a Japanese company in Japan than with a Taiwanese in Taiwan. And the single wafer choice, if they can make it work, is quite right for the target they aim for: small lots, fast turnaround, small volumes. If someone can fix an IBM process is a Japanese engineer and they have the advantage of starting their fab without any need of compatibility with the existent, knowing all the pitfalls and the points where to improve the production flow. They do not need to be profitable in the short term if they keep the Japanese Government happy and showing working chips will be the hard test, not price or capacity. Hey, US was ready to pay 10k$/unit for a specific toilet seat for the military, imagine what some chips may fetch for "critical" applications if someone is there to take the orders and the customer can be there at every step of the way checking what happens to his design.
jorgequinonez
Member
The Japanese have some defense law or rule about producing some things in Japan (e.g., high price ticket items in the defense industry like tanks, jet fighters in the past, transport aircraft, submarines). They do this even if the product costs double or triple if it had been imported. I wonder if the Rapidus venture falls under that, like Intel does for the USA.