Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intel-renames-process-nodes-to-better-align-with-tsmc.14454/
        )

    [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
)

Intel renames process nodes to better align with TSMC

Daniel Nenni

Admin
Staff member
We will be blogging about this more but the most interesting thing said today at the Intel Accelerated event was the renaming of nodes. Something which Scott Jones and I have been actively pushing for:

Intel Process rename 2021.jpg



Instead of using the ++ nomenclature, Intel will call 10nm ++ 7 and will call the original 7nm process 4. An improved 4 version will be called 3. Intel GAA implementation will be called 20A (angstrom). An angstrom in physics is .1nm. I seriously doubt it will catch on so we may see another naming division at that time. Please note that Intel is dropping the industry standard "nm" nomenclature.

All-in-all good news in my opinion. This tells me that Intel is listening.

Intel SemiWiki 2021.jpg
 
I think it’s good for the industry. Despite the fact that they’ve been trailing since 2018, their marketing kept portraying this false image - oh it’s the naming scheme, they’re really not behind. they’d better be confident that they can keep up with TSM, or else their shareholders would push them to stay away from advanced nodes because they have nowhere to hide now
 
Hi Daniel,
Intel 4..........2022
Intel 3 ........2023
Intel 20A (is that 2nm?).......Ramp in 2024

Is that a realistic goal?

In last earning call, Intel said 10nm wafer had 47% cost improvement. That means last year (2020) 10nm yeild was really bad. Intel launch 10nm in the end of 2018. After 2 years the yield was still very bad. 3rd year finally had great improvement.
 
This is a very interesting change. I wish the numbers were still more than just marketing (like back in 65+nm days), but at least the marketing will now be consistent.

I'm curious how long this will stay synchronized between Intel and the other foundries. As I understand it, there have already been two examples of node naming changes -- after about 65nm or so era the manufacturers switched to equivalents and no longer exact measurements. Then around 20nm I believe it was TSMC that first decremented the node number without a corresponding decrease in transistor size. If innovations only change node naming (like we're seeing now from Intel, as well as TSMC and Samsung), how do you keep those numbers honest for density?
 
Rearranging the chairs on the sinking titanic. 4 consecutive quarters of declines the mediatek 8192 will change things cost less and if you just need some word processing the wintel product really isn't worth the $1200. Changing the names won't solve the problems.
 
Hi Daniel,
Intel 4..........2022
Intel 3 ........2023
Intel 20A (is that 2nm?).......Ramp in 2024

Is that a realistic goal?

In last earning call, Intel said 10nm wafer had 47% cost improvement. That means last year (2020) 10nm yield was really bad. Intel launch 10nm in the end of 2018. After 2 years the yield was still very bad. 3rd year finally had great improvement.

Intel 4 in 2022 will be a stretch in my opinion. I'm sure they will release wafers out of the OR R&D fab in 2022 to save face but the true test is when the regular Intel fabs hit HVM and that will be 2023, hopefully.

Meanwhile TSMC 4nm is in HVM in 2021 (iPhone13) and 3nm in 2022 (iPhone14) so Intel is still 1-2 years behind.

The only problem I had with the presentation is this:

“Building on Intel’s unquestioned leadership in advanced packaging, we are accelerating our innovation roadmap to ensure we are on a clear path to process performance leadership by 2025,” Intel's new CEO Pat Gelsinger (above) said during the "Intel Accelerated" livestream today.

Intel's packaging leadership has been questioned by TSMC and for Intel to not acknowledge that puts them in peril. Intel said the same thing about their process leadership and here they are, behind both TSMC and Samsung.
 
Intel 4 in 2022 will be a stretch in my opinion. I'm sure they will release wafers out of the OR R&D fab in 2022 to save face but the true test is when the regular Intel fabs hit HVM and that will be 2023, hopefully.

Meanwhile TSMC 4nm is in HVM in 2021 (iPhone13) and 3nm in 2022 (iPhone14) so Intel is still 1-2 years behind.

The only problem I had with the presentation is this:

“Building on Intel’s unquestioned leadership in advanced packaging, we are accelerating our innovation roadmap to ensure we are on a clear path to process performance leadership by 2025,” Intel's new CEO Pat Gelsinger (above) said during the "Intel Accelerated" livestream today.

Intel's packaging leadership has been questioned by TSMC and for Intel to not acknowledge that puts them in peril. Intel said the same thing about their process leadership and here they are, behind both TSMC and Samsung.
What is the difference between Intel's "Foveros Direct" and TSMC's "CopperHybridBonding" ?

AMD has already shown silicon with TSMC's approach ... and said they will be shipping a product by end-of-year 2021.
 
What is the difference between Intel's "Foveros Direct" and TSMC's "CopperHybridBonding" ? AMD has already shown silicon with TSMC's approach ... and said they will be shipping a product by end-of-year 2021.

Not much really and certainly not in the "unquestioned leadership" territory. Tom Dillinger will write more about this shortly. In my opinion AMD and TSMC will match or exceed Intel's packaging abilities for CPU/GPU. And today TSMC's InFos is the "unquestioned leader" in SoC packaging.
 
Not much really and certainly not in the "unquestioned leadership" territory. Tom Dillinger will write more about this shortly. In my opinion AMD and TSMC will match or exceed Intel's packaging abilities for CPU/GPU. And today TSMC's InFos is the "unquestioned leader" in SoC packaging.
I looked at Intel's roadmap again. They showed they only have 50micro ubump pitch until 3Q/2022 -- then it is 36. They said they would have HybridCopperBonding at mid-2023 -- at least 2 years from now.
 
I looked at Intel's roadmap again. They showed they only have 50micro ubump pitch until 3Q/2022 -- then it is 36. They said they would have HybridCopperBonding at mid-2023 -- at least 2 years from now.

We had a call with Intel this morning for clarification. Tom will get something up on Monday.
 
It's interesting in the note that intel said unquestioned leadership in performance per watt and not performance per dollar. Even if achievable, road map seems expensive...
 
It's interesting in the note that intel said unquestioned leadership in performance per watt and not performance per dollar. Even if achievable, road map seems expensive...
Rumors have been that Intel's base costs (not price -- cost) has already been higher than a foundry. They did unique wiring & ground rules -- and were willing to accept +20% cost because they were driven by CPU performance. Also, foundry customers were annoyed by restrictive design rules which made layout collaboration and IP integration more difficult.

As they enter foundry market -- it will be interesting to watch if they do away with the CPU baggage. (another option would be to sell it to IBM & companies that want to squeeze the maximum performance out of the design)
 
It's interesting in the note that intel said unquestioned leadership in performance per watt and not performance per dollar. Even if achievable, road map seems expensive...
It will be very expensive. If AMD manages to snag even more market share from intel then Intel will have no other option then to start selling it's least profitable businesses to finance the research.
 
Last edited:
Back
Top