Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/ibm-eyes-deeper-partnership-with-rapidus-for-sub-1nm-chip-development.23156/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

IBM eyes deeper partnership with Rapidus for sub-1nm chip development

Daniel Nenni

Daniel Nenni

Admin
Staff member
1_b.jpg

Credit: AFP

IBM is seeking a long-term partnership with Japan's Rapidus to develop sub-1nm chips, according to Yomiuri Shimbun. Building on their 2nm collaboration, IBM has deployed engineers to Rapidus's Hokkaido site, signaling deeper ties as both companies pursue next-generation semiconductor production and Japan ramps up investment in chip innovation.

According to the report, Mukesh Khare, General Manager of IBM's semiconductor R&D division, expressed IBM's intention to build a long-term partnership with Rapidus to advance next-generation semiconductor technologies. In addition to ongoing collaboration on 2nm chip mass production, IBM hopes to continue working with Rapidus on even more advanced process nodes.

The report noted that IBM aims to develop sub-1nm semiconductor technology within the next few years, and Rapidus may take on the task of mass production for these chips in the future. Khare confirmed that IBM has dispatched about 10 engineers to Rapidus's facility in Hokkaido and emphasized the company's full support to help Rapidus achieve successful 2nm chip production by 2027.

IBM and Japanese semiconductor firm Rapidus have expanded their collaboration to develop mass-production technologies, with a focus on 2nm-generation semiconductors. The partnership, announced in June 2024, supports Japan's New Energy and Industrial Technology Development Organization (NEDO) project aimed at advancing chiplet and package design for next-generation semiconductors.

By December 2024, the collaboration reached a notable milestone with the creation of a new chip construction method called selective layer reductions. This process enables the consistent production of nanosheet gate-all-around transistors with multiple threshold voltages (multi-Vt). The innovation facilitates more energy-efficient and complex computations, which are crucial for scaling 2nm transistors to mass production levels.

www.digitimes.com

IBM eyes deeper partnership with Rapidus for sub-1nm chip development

IBM is seeking a long-term partnership with Japan's Rapidus to develop sub-1nm chips, according to . Building on their 2nm collaboration, IBM has deployed engineers to Rapidus's Hokkaido site, signaling deeper ties as both companies pursue next-generation semiconductor production and Japan ramps...
www.digitimes.com
 
How is that pilot line coming. It started a couple months ago. is it putting out wafers now for customer evaluations? what does the PDK look like ?

Or are they too preparing a "we are going to work on 1.4nm and focus on the future technologies and customers" announcement.
 
MKWVentures said:
How is that pilot line coming. It started a couple months ago. is it putting out wafers now for customer evaluations? what does the PDK look like ?

Or are they too preparing a "we are going to work on 1.4nm and focus on the future technologies and customers" announcement.
Click to expand...

Since you asked, from what I hear inside the ecosystem, Rapidus is paying IP vendors big money to port IP over. Why you ask? Because Rapidus will be lucky to ship 25k wafers per month in 2026 and no big volumes means there will be limited profits for the ecosystem unless they get paid upfront. The good news is that the PDK is good and the initial test silicon is as expected. Broadcom is rumored to be a customer and several Japan companies so demand could outpace supply if there are no viable 2nm alternatives.

The other bad news is that the ecosystem is busy with TSMC N2 and Rapidus 2nm so they are not spending time with Intel 18A and Samsung 2nm. Unless of course there is some big upfront payment or a big customer requesting it which I cannot confirm.
 
Daniel Nenni said:
The other bad news is that the ecosystem is busy with TSMC N2 and Rapidus 2nm so they are not spending time with Intel 18A and Samsung 2nm. Unless of course there is some big upfront payment or a big customer requesting it which I cannot confirm.
Click to expand...

Why would they prioritize Rapidus over Intel or Samsung?
 
Xebec said:
Why would they prioritize Rapidus over Intel or Samsung?
Click to expand...

Because Rapidus is paying IP vendors millions of dollars up front to make up for the lack of volume. The other foundries do not do that. Most IP is royalty based so if there are no customers it is a loss for the IP vendor. IP companies generally start with TSMC since that is where the customers are. Next they port to other foundries based on customer demand.

This is why TSMC N2 has such momentum. All of the IP vendors have TSMC N3 IP so it was easy for customers to move to N2 because they knew the same IP would be made available at N2 before the other foundries. It is the snowball effect.
 
Daniel Nenni said:
Because Rapidus is paying IP vendors millions of dollars up front to make up for the lack of volume. The other foundries do not do that. Most IP is royalty based so if there are no customers it is a loss for the IP vendor. IP companies generally start with TSMC since that is where the customers are. Next they port to other foundries based on customer demand.

This is why TSMC N2 has such momentum. All of the IP vendors have TSMC N3 IP so it was easy for customers to move to N2 because they knew the same IP would be made available at N2 before the other foundries. It is the snowball effect.
Click to expand...

Do IP companies need to pay anything to TSMC, or vice versa? Or do they not pay anything to each other?
 
hist78 said:
Do IP companies need to pay anything to TSMC, or vice versa? Or do they not pay anything to each other?
Click to expand...
I believe TSMC is the only adv foundry that does not pay IP vendors. Their dominate market position is the key reason. In fact, it's hard for a IP startup to even get into their eco-system, without customers (what tsmc think is worthy customers) that ask for those IP. So a catch 22 for those IP vendors. Same thing for EDA startup too
 
swka said:
I believe TSMC is the only adv foundry that does not pay IP vendors. Their dominate market position is the key reason. In fact, it's hard for a IP startup to even get into their eco-system, without customers (what tsmc think is worthy customers) that ask for those IP. So a catch 22 for those IP vendors. Same thing for EDA startup too
Click to expand...
And when the dominant IP vendors are also the dominant EDA vendors who command $10k-$100k per license for some 180nm tech that was developed 25 years ago, well then shucks I wonder where all the semiconductor startups went.
 
Daniel Nenni said:
Since you asked, from what I hear inside the ecosystem, Rapidus is paying IP vendors big money to port IP over. Why you ask? Because Rapidus will be lucky to ship 25k wafers per month in 2026 and no big volumes means there will be limited profits for the ecosystem unless they get paid upfront. The good news is that the PDK is good and the initial test silicon is as expected. Broadcom is rumored to be a customer and several Japan companies so demand could outpace supply if there are no viable 2nm alternatives.

The other bad news is that the ecosystem is busy with TSMC N2 and Rapidus 2nm so they are not spending time with Intel 18A and Samsung 2nm. Unless of course there is some big upfront payment or a big customer requesting it which I cannot confirm.
Click to expand...
Great input. Intel will definitely not ship 25K wafers of external foundry per month in 2026. That is clear from Intel.

If Rapidus has more wafer revenue external wafer starts than Intel foundry then .... well ... I'm at a loss for words. I did not have that on my Intel foundry scenario BINGO card
 
MKWVentures said:
Great input. Intel will definitely not ship 25K wafers of external foundry per month in 2026. That is clear from Intel.
Click to expand...
Intel's priority should be filling fabs no matter the customer Internal+External ratio
MKWVentures said:
If Rapidus has more wafer revenue external wafer starts than Intel foundry then .... well ... I'm at a loss for words. I did not have that on my Intel foundry scenario BINGO card
Click to expand...
well Rapidus Public Timeline is 2027 and i am still more pessimistic about Rapidus regarding Tech but not customers and for Intel it's the opposite should have paired with each other imo.
 
MKWVentures said:
Great input. Intel will definitely not ship 25K wafers of external foundry per month in 2026. That is clear from Intel.

If Rapidus has more wafer revenue external wafer starts than Intel foundry then .... well ... I'm at a loss for words. I did not have that on my Intel foundry scenario BINGO card
Click to expand...

As I may have mentioned, the NOT TSMC market is alive and well. The other advantage you get when paying the top IP companies to port over is that they start pushing Rapidus as an alternative to TSMC and people listen to IP companies.

Fab builds in Japan are faster and cheaper than the US. Rapidus could scale up faster than people think, as long as customers come. It will not be an Intel/Samsung Field of Dreams scenario, building fabs without customers.

I do agree that the Rapidus technology is not HVM ready and may never be if IBM history repeats itself. Exciting times in the semiconductor industry, absolutely.
 
Daniel Nenni said:
As I may have mentioned, the NOT TSMC market is alive and well. The other advantage you get when paying the top IP companies to port over is that they start pushing Rapidus as an alternative to TSMC and people listen to IP companies.

Fab builds in Japan are faster and cheaper than the US. Rapidus could scale up faster than people think, as long as customers come. It will not be an Intel/Samsung Field of Dreams scenario, building fabs without customers.

I do agree that the Rapidus technology is not HVM ready and may never be if IBM history repeats itself. Exciting times in the semiconductor industry, absolutely.
Click to expand...
Intel partially built it and no one came. Turn Ohio into a Baseball park with Corn field around it. Maybe Shoeless Joe can get someone to buy wafers from Intel. (Great movie BTW).

The Not TSMC business you have highlighted is clear (I quote this and give you the credit all the time). I just didnt ever expect people to go with Rapidus over Intel. SMH

Lets see how this plays out.

18A has some roadmap changes coming .... internal as well as external.... before the year is out.
 
MKWVentures said:
They will paper launch something. But the bigger issue is that 18A doesnt make financial sense. So the volume needs to be changed for internal products as well to minimize cash flow and losses... LBT is getting pretty good at tough decisions.
Click to expand...

The layoffs are happening this week and they are deep. The next investor call will be telling. The foundry group definitely got hit.
 
You must log in or register to reply here.
Back
Top