Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intel-plots-1nm-silicon-for-2027-but-are-the-wheels-coming-off-its-existing-roadmap-2n6y.19720/
        )

    [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 plots 1nm silicon for 2027 but are the wheels coming off its existing roadmap? 2N6Y?

fansink

fansink

Active member
Well, Intel 10/7 was 1N5Y so 2N5Y (Intel 3 and 18A) is double the productivity so that should be celebrated. Here is the problem with PR spin (5N4Y): First of all the semiconductor professionals that matter know it is BS. Second, 5N4Y is impossible to sustain so history will show that Intel made it all up. My opinion.
 
Daniel Nenni said:
Well, Intel 10/7 was 1N5Y so 2N5Y (Intel 3 and 18A) is double the productivity so that should be celebrated. Here is the problem with PR spin (5N4Y): First of all the semiconductor professionals that matter know it is BS. Second, 5N4Y is impossible to sustain so history will show that Intel made it all up. My opinion.
Click to expand...

Thank you for your professional knowledge-based comments 👍
 
yes there are half nodes but IF they execute.... even that is good for Intel. That said Intel 4/3 is very low volume today (we will have update volume numbers by node in a couple weeks) and 20/18 doesn't have any products planned until end of year.

If they launch (sell products) Intel 3 in Q2 24.... and Intel 20A in Q4 2024.... and 18A in Q2 2025, that will be a huge accomplishment and I will be the first to say they turned it around completely and executed well.
fansink said:
This article seems to ~mirror some comments on a recent thread below.

The article more concisely spells out the case for 2N6Y, as opposed to Gelsinger’s 5N4Y.

Intel plots 1nm silicon for 2027 but are the wheels coming off its existing roadmap?

https://www.pcgamer.com/intel-plots...e-the-wheels-coming-off-its-existing-roadmap/

Gelsinger 5 Nodes in 4 years; Success or Failure?

https://semiwiki.com/forum/index.php?threads/gelsinger-5-nodes-in-4-years-success-or-failure.19695/
Click to expand...
 
Seems to me like the article lacks a little critical thinking. Given how long it takes from a design start on a node to HVM, Intel isn't going to require additional leading edge chip capacity until 18A. Given what we've seen, they are more likely shunt off older, fully depreciated nodes to substrates for packaging, UMC-like deals, and mature node foundry work for markets like automotive, while building scale for the nodes where they are finding leading edge customers.
 
That chart is very interesting. It shows 20A/18A in production right now, even though 20A products only are released around end of 2024. Then Intel 4/3 does not increase until 2Q 2024... The only products using Intel 3/4 are meteor lake and Granite Rapids. That would mean that Granite Rapids is delayed, or at least not getting much adoption until next year.

image_2024-03-01_015732471.png
 
lefty said:
That chart is very interesting. It shows 20A/18A in production right now, even though 20A products only are released around end of 2024. Then Intel 4/3 does not increase until 2Q 2024... The only products using Intel 3/4 are meteor lake and Granite Rapids. That would mean that Granite Rapids is delayed, or at least not getting much adoption until next year.
Click to expand...

It seems that Gelsinger pushes PR spin to the wishes/BS end of the spectrum.

If wishes were horses, beggars would ride.
 
fansink said:
It seems that Gelsinger pushes PR spin to the wishes/BS end of the spectrum.

If wishes were horses, beggars would ride.
Click to expand...
it does not show 18A and 20A in production. they have small capacity in development. The will add some to Oregon site for low volume production and then ramp in AZ in late 2025. There are a lot of details on the Intel Future in these slides.
 
MKWVentures said:
it does not show 18A and 20A in production. they have small capacity in development. The will add some to Oregon site for low volume production and then ramp in AZ in late 2025. There are a lot of details on the Intel Future in these slides.
Click to expand...
Well, there's clearly a green band showing 20A production from 2023. Maybe it's just that the graph isn't accurate
 
lefty said:
That chart is very interesting. It shows 20A/18A in production right now, even though 20A products only are released around end of 2024. Then Intel 4/3 does not increase until 2Q 2024... The only products using Intel 3/4 are meteor lake and Granite Rapids. That would mean that Granite Rapids is delayed, or at least not getting much adoption until next year.

View attachment 1726
Click to expand...

It also shows that Intel 4/3 and 20/18A are half nodes. 😂
 
I still struggle a little with us picking on Intel so much. We should be poking at the marketing for all three bleeding edge foundries:

“TSMC N7” , “N6”, “N5”, “N4”, N3” — that reads as “Node 7”, “Node 6”, leading us to think, ‘one node per year’. TSMC 12nm was less dense than Intel 14nm..

And then there’s Samsung, 3nm with “perfect yields” with the density of TSMC N4/N5, and plenty of other density marketing games.

Intel was actually last to play the node name marketing games IMO.

Was Intel 5N4Y a success? Let’s see if they deliver amazing products in 2025 in 18A - if yes, then yes.. We’ll know which way the wind is blowing with 20A products in < 12 months.
 
Xebec said:
I still struggle a little with us picking on Intel so much. We should be poking at the marketing for all three bleeding edge foundries:
“TSMC N7” , “N6”, “N5”, “N4”, N3” — that reads as “Node 7”, “Node 6”, leading us to think, ‘one node per year’. TSMC 12nm was less dense than Intel 14nm..
And then there’s Samsung, 3nm with “perfect yields” with the density of TSMC N4/N5, and plenty of other density marketing games.
Intel was actually last to play the node name marketing games IMO.
Was Intel 5N4Y a success? Let’s see if they deliver amazing products in 2025 in 18A - if yes, then yes.. We’ll know which way the wind is blowing with 20A products in < 12 months.
Click to expand...

Intel 20A products will be mostly TSMC die since it is a chiplet only node like Intel 4. I want to see a full chip, not chiplets, if you are comparing it to TSMC. That is why I call them half nodes. Call me old school but if you can't do a full chip it is not a full node. Nobody in the ecosystem I talked to at the Intel event is buying the 5N4Y schtick. It's embarrassing and it is hard to believe that no one in Pat's circle is telling him he has spinach in his teeth.
 
Daniel Nenni said:
Intel 20A products will be mostly TSMC die since it is a chiplet only node like Intel 4. I want to see a full chip, not chiplets, if you are comparing it to TSMC. That is why I call them half nodes. Call me old school but if you can't do a full chip it is not a full node. Nobody in the ecosystem I talked to at the Intel event is buying the 5N4Y schtick. It's embarrassing and it is hard to believe that no one in Pat's circle is telling him he has spinach in his teeth.
Click to expand...
I love old school btw :), but how do we define a full chip vs. chiplet exactly?

An Intel Pentium III Coppermine on 180nm was ~ 95mm2 (later versions), not much larger than the Compute die for Meteor Lake (70mm2). Apple A14 was the lead N5 chip at 88mm2, and A17 leads N3B at ~ 100mm2. Yes I recognize the MTL Compute die is not a full chip by definition, but it's still demonstrating >5GHz frequency for a large # of transistors on an advanced node - is that worth less than an Apple SoC that is ~ only 20-30% larger?

For Intel 3, the compute dies for Granite Rapids and Sierra Forest are certainly going to be a lot larger, and we don't yet know how large Arrow Lake's compute die is going to be. Clearwater Forest on 18A will certainly have large dies, though we don't yet know when precisely.

I'm not saying I agree with Pat at all, but I'm not sure the competition's marketing gives him much choice. The competition started this "change the number/declare a new node even when the dimensions are the same" business. And that's on top of the numbers becoming less meaningful generations before FinFet was even in production.
 
Xebec said:
I still struggle a little with us picking on Intel so much. We should be poking at the marketing for all three bleeding edge foundries:

“TSMC N7” , “N6”, “N5”, “N4”, N3” — that reads as “Node 7”, “Node 6”, leading us to think, ‘one node per year’. TSMC 12nm was less dense than Intel 14nm..

And then there’s Samsung, 3nm with “perfect yields” with the density of TSMC N4/N5, and plenty of other density marketing games.

Intel was actually last to play the node name marketing games IMO.

Was Intel 5N4Y a success? Let’s see if they deliver amazing products in 2025 in 18A - if yes, then yes.. We’ll know which way the wind is blowing with 20A products in < 12 months.
Click to expand...

I start thinking this funny clock. We might need one for the semiconductor business.

1709345553422.png
 
lefty said:
That chart is very interesting. It shows 20A/18A in production right now, even though 20A products only are released around end of 2024. Then Intel 4/3 does not increase until 2Q 2024... The only products using Intel 3/4 are meteor lake and Granite Rapids. That would mean that Granite Rapids is delayed, or at least not getting much adoption until next year.

View attachment 1726
Click to expand...

This graph tells us a lot(Assuming that it's accurate). Look at the wafer capacity of Intel 4/3 which already got mass production chips(Meteor lake). Comparing its sheer volume to Intel 7(Which also has mass production chips), we can say that their manufacturing is not really recoverd. Maybe Intel 4/3 is somewhat broken node. It's not recovering even with Granite rapids and Sierra forest(estimated in 2025, BIG chips) launch. Difference in volume between Intel 7 and Intel 4 has gone to TSMC.

Funny thing here is Intel 20A/18A. They expect its volume will surpass that of Intel 4/3 in 2025 already. So they're telling us that they can get Arrow lake + Foundry customers in Intel 20A/18A, even if Intel 4/3 is not working well.

Also even with all conundrum, Intel expects Intel 4/3 be a BIG node in 2028. Maybe Intel expects it to be base nodes for legacies and advanced packaging...? Maybe im thinking too much with vague graph..
 
diediealldie said:
This graph tells us a lot(Assuming that it's accurate). Look at the wafer capacity of Intel 4/3 which already got mass production chips(Meteor lake). Comparing its sheer volume to Intel 7(Which also has mass production chips), we can say that their manufacturing is not really recoverd. Maybe Intel 4/3 is somewhat broken node. It's not recovering even with Granite rapids and Sierra forest(estimated in 2025, BIG chips) launch. Difference in volume between Intel 7 and Intel 4 has gone to TSMC.

Funny thing here is Intel 20A/18A. They expect its volume will surpass that of Intel 4/3 in 2025 already. So they're telling us that they can get Arrow lake + Foundry customers in Intel 20A/18A, even if Intel 4/3 is not working well.

Also even with all conundrum, Intel expects Intel 4/3 be a BIG node in 2028. Maybe Intel expects it to be base nodes for legacies and advanced packaging...? Maybe im thinking too much with vague graph..
Click to expand...

I think Intel simply scaled back investment In building capacity for 4/3 to focus on 20A/18A. At least from a chip performance perspective, Intel 4 already seems better than Intel 10nm/7 was during the Ice Lake era. Intel is basically dumping 12th and some 13th gen chips for mobile (see slickdeals.net for some impressive laptop deals), and Meteor Lake is running at high clocks and seems available in volume. Pat/Intel also inherited a lot of pre negotiated TSMC N3 capacity from his predecessor.

This is definitely different than how Intel usually does it. I hope it all pays off for Intel.
 
Xebec said:
I love old school btw :), but how do we define a full chip vs. chiplet exactly?
Click to expand...

A chiplet is a subset of a full chip.

In the old school design world we used to call them blocks. Multiple teams did digital blocks and assembled them to tape out a full chip. Now chiplets actually are die level blocks that you can stack together to make a chip.

How about this, a full fab does full chips and a fablet does chiplets?
 
Daniel Nenni said:
A chiplet is a subset of a full chip.

In the old school design world we used to call them blocks. Multiple teams did digital blocks and assembled them to tape out a full chip. Now chiplets actually are die level blocks that you can stack together to make a chip.

How about this, a full fab does full chips and a fablet does chiplets?
Click to expand...

Accordingly, on a scale of 1 to 5, where 1 is a fablet and 5 is a full fab, where is Intel currently vs. where are they heading? Like currently 4 and heading towards 2?
 
You must log in or register to reply here.
Back
Top