Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/tsmc-to-make-3nm-chips-for-intel-sources-claim.13645/
        )

    [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] => 2021370
            [XFI] => 1050270
        )

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

TSMC to make 3nm chips for Intel, sources claim?

Daniel Nenni

Admin
Staff member
Well, they almost got it right:


Now, another major development related to the company has surfaced online. As per the report from Taiwanese media, Intel has given an outsourcing contract to Taiwan Semiconductor Manufacturing Company (TSMC) for manufacturing 3nm chips.

The report claims that Intel’s upcoming chips designed with a 3nm process will go into mass production in the second half of 2022. With this, Intel will become the second-largest customer for TSMC’s 3nm process after Apple.

I highly doubt Intel will be in 3nm mass production in second half 2022 at the same time as Apple. First half of 2023 is more like it. But this certainly explains the big CAPEX bump for 2021.
 
Last edited:
It doesn't make sense.

If this new CPU based on TSMC 3nm, according to the original Digitimes article, is coming out in 2022 first, that can greatly diminish the marketability of 7nm CPU manufactured by Intel itself and scheduled to be released in 2023.

Or people may guess the 2022 CPU will be a low-end model and 2023 CPUs will be more high-end. But how can Intel use more advanced and more expensive outside foundry's technology to make supposed to be cheaper and low-end CPU?
 
Last edited:
I would say server chips will stay at Intel and PC chips will be outsourced but that would mean the "majority" of the chips would be outsourced. So maybe they will insource high performance chips and outsource the rest? It will be interesting to see how Intel spins this on investor calls. Either way it will be 2023 not 2022 and Intel will not have to choose between making a certain chip first to ramp the fabs, they can flood the market with all of the above with the help of TSMC.
 
I would say server chips will stay at Intel and PC chips will be outsourced but that would mean the "majority" of the chips would be outsourced. So maybe they will insource high performance chips and outsource the rest? It will be interesting to see how Intel spins this on investor calls. Either way it will be 2023 not 2022 and Intel will not have to choose between making a certain chip first to ramp the fabs, they can flood the market with all of the above with the help of TSMC.
Yes, this "majority" part will not be consistent with what Intel has said.

It's kind of strange that Intel's high performance/high price servers CPU will use 7nm process while desktop CPU will use more advanced 3nm technology. If Intel does go down this path, it can push more Intel's customers to find alternative routes.
 
Yes, this "majority" part will not be consistent with what Intel has said.

It's kind of strange that Intel's high performance/high price servers CPU will use 7nm process while desktop CPU will use more advanced 3nm technology. If Intel does go down this path, it can push more Intel's customers to find alternative routes.

How much more advanced will TSMC 3nm be in comparison to Intel 7nm? Both will have full EUV and I would hope Intel is hyper focused on performance. In regards to density/cost TSMC 3nm will have the edge for sure. If Intel changes the name from 7nm to 3nm that would help. :cool:
 
How much more advanced will TSMC 3nm be in comparison to Intel 7nm? Both will have full EUV and I would hope Intel is hyper focused on performance. In regards to density/cost TSMC 3nm will have the edge for sure. If Intel changes the name from 7nm to 3nm that would help. :cool:

This is a very good question for Mr. Guru Scotten Jones.
 
Maybe we are all wrong!
Pat said that he is confident that the majority of 2023 products are made internally. I read that "number of products", but could later also explained as "sales value" of these products.
Keep in mind that current Intel CPU products cover an ASP range of ~15 to 10k USD. Maybe they just do a few of the very expensive ones at TSMC just to keep the majority (reads number) of products internally. This would also fit the Scotten latest blog here where I understood that TSMC N3 is slightly better than Intel 7nm.
 
Here is the translation of the Digitimes article:

Intel is set to become TSMC's second largest customer, next to Apple. The cooperation between the two will continue through the 2nm generation. Intel would not comment on market rumors. However, a few days ago, Pat Gelsinger, who is about to take over as Intel’s CEO, gave a rather vague statement about Intel's outsourcing plans, saying only that: "the majority of 2023 products will be manufactured internally, but the proportion of outsourcing will still increase."

This response is somewhat different from the market's expectation that a large order will be released, leaving outsiders confused as to where TSMC's confidence comes from.

With regards to this, sources from the semiconductor industry claim that Intel had already signed a new outsourcing contract with TSMC in 2020. The largest order is for CPUs that will use TSMC’s 3nm process slated to enter mass production in the second half of 2022. Intel’s technology manufacturing team has already visited Southern Taiwan Science Park several times in the past 2 years to understand the progress. Barring the unforeseen, Intel will adopt a dual-track manufacturing strategy. High volume production for new, core products will be handled by TSMC with Intel will also simultaneously produce them internally, but at a smaller proportion.

For Intel, this allows them to focus on R&D to advance process technology and reduce mass production risks. The crisis of shortages and process delays will be fully resolved by then.

For TSMC, 3nm has already secured numerous customer commitments even before entering mass production. Besides Apple’s core orders, Intel's large orders are also in the pocket. Plus, almost all of the orders for advanced manufacturing processes from AMD, Mediatek and other major chip companies, as second and third wave mass production customers. Coinciding with the arrival of the 5G and AI era, order visibility is "cloudless". TSMC is of course super optimistic about the next 5 years.

TSMC has already begun initial site preparation and cooperation with Apple in technology research and development for 2nm, and has also secured Intel's commitment for continued cooperation; its confidence is well-founded. TSMC would not respond to customer and order-related queries.

It is worth noting that TSMC is ASML's largest customer for EUV lithography equipment. Cumulative purchases are estimated to exceed 30 units by the end of 2020. The 5/3nm production base in Southern Taiwan Science Park alone currently has more than 20 units. In 2021, purchases will be further expanded, with an estimated 18-20 units. Cumulatively, that is roughly the sum total of Intel and Samsung's purchases combined.

Although Intel has established an outsourcing plan, it is understood that its own research and development for advanced EUV processes has not slowed down. In the past six months, it has continued to purchase EUV-related equipment according to previously established plans. Intel's direction to conduct its own research and development and manufacturing [aka IDM] remains unchanged. The purpose of outsourcing large orders to TSMC is to focus on intensifying R&D and reduce mass production risk. Recent rumors about spinning off fabs are not in Intel's plan.

 
It was mentioned that Intel will be doubling production soon so I assumed this is what they meant.

Intel's I5 and I7 at 10 nm are still the best and plan to buy one.

Intel is stagnant at $20 billion a quarter and that's what it's all about. Nvidia and amd will take market share but lack the killer instinct to take Intel out.
 
Last edited by a moderator:
This is what rumour mill says if somebody here has Digitimes subscription.

 
It's sometimes bend not break. 2 or 3 nm is different than 350, 90 or 45 nm.
Quite an opposite of the "Don't copy, and don't touch" tradition on TSMC.

They were always going all in on a certain process, and one really big fab from RnD start to mature production.

Enormous effort is made to bruteforce problems of the process by having thousands of PhDs, and postdocs keep banging their head against the wall, looking for solutions through exhaustive search, and when anything marginally usable comes out of that, big "do not touch" banners are slapped onto everything.

Two retired TSMC researchers were kind of informal mentors to me when I was still aiming into career in process 13 years ago.
 
TSMC can only let Intel see their 5nm or 3nm PDK if Intel signs the agreement to stop development of (Intel's) 7nm and more advanced nodes. Apple and Qualcomm do not have their own fabs, so Intel is not like just another TSMC customer. Even Samsung might be more acceptable for 3nm; at least we know Samsung uses nanosheets at 3nm while TSMC stays with FinFETs.
 
Last edited:
Computer espionage is more popular than ever. It's so complex that it has to be on a computer and competitors (samsung and intel) probably already have an idea how it's done
 
Back
Top