[SemiAccurate] Intel kills off the 10nm process

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
Intel added two cores to pretty much every chip in their lineup, finally giving Avg. Joe a solid reason to replace his Sandy Bridge dual-core laptop!

I think that's part of it. I also think a lot of people on Haswell and Broadwell-gen Xeons had active incentive to upgrade early to Skylake-SP due to Meltdown/Spectre (some of the worst performance hits, if I recall correctly, were on older platforms). I think the modems were part of it. I think the chipset shift was part of it. The picture isn't so simple as to say, "Hey Intel outperformed on total revenues, so clearly demand for 14nm is up". It is not that simple. If I had their unit numbers for client group in front of me, maybe I could add that to the picture.
 

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
In a year where their server products got hit hard by design flaws that needed fixing, Intel still managed to get higher revenues from server customers. In other words, they profited from their choice to completely ignore security in their design choices.
 
  • Like
Reactions: scannall

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
In a year where their server products got hit hard by design flaws that needed fixing, Intel still managed to get higher revenues from server customers. In other words, they profited from their choice to completely ignore security in their design choices.

That's possible. Without having more data on who replaced hardware to get away from older platforms that took bigger hits from Spectre/Meltdown, it's hard to say exactly how much money they earned from poor design decisions.


Informative, but it still leaves some questions. I guess they get to hide their ODM numbers behind percentages (without knowing how many units they shipped in Q3 2017, we can't use the 15% increase to show how many units they shipped in Q3 2018). I still find it interesting that the two numbers under client group (notebook/laptop and desktop) fall below the revenue gain for the client computing group. 8%/1% vs. 16%. Intel definitely benefited from higher ASPs moreso than increased demand. Same for their server products.
 

jpiniero

Lifer
Oct 1, 2010
14,513
5,160
136
https://semiaccurate.com/2019/01/25/why-semiaccurate-called-10nm-wrong/

Charlie's got a new article regarding 10 nm. He agrees with me now that Intel instead of killing 10 nm off completely will just ship some relatively low volume parts and will just have to be creative with binning.

Intel IMO needs to go to 7 ASAP (ie: Products out in 1H 2021)

When we wrote the original piece there were four fabs slated to transition to 10nm. One of these has been backported to 14nm, something which can’t be undone in a time relevant to the 10nm transition. Two of the remaining fabs installed lots of EUV tools which are meant for the 7nm process, not the 10nm process. This effectively precludes these facilities from producing 10nm.

This left one fab which was slated for 10nm, and try as we might we couldn’t get definitive information on. Meanwhile we had several sources confirming the information about the three other fabs and telling us that 10nm was unquestionably dead. As it turns out they, and SemiAccurate, were wrong, it is coming out in some form in Q4/2019.
 

moinmoin

Diamond Member
Jun 1, 2017
4,935
7,621
136
https://semiaccurate.com/2019/01/25/why-semiaccurate-called-10nm-wrong/

Charlie's got a new article regarding 10 nm. He agrees with me now that Intel instead of killing 10 nm off completely will just ship some relatively low volume parts and will just have to be creative with binning.

Intel IMO needs to go to 7 ASAP (ie: Products out in 1H 2021)
I thought they were working on some fixed 10nm+ or 10nm++ with better yield of which we will see products this and next year? Shipping low volume parts to fill the time needed until some 7nm EUV process is ready won't cut it in the current increasingly competitive market environment. To me this reads like more bad news for Intel.

The only potentially good news from the article is that apparently of four 10nm fabs Intel moved one to 14nm and two to 7nm EUV, and the remaining 10nm fab is primarily used to resolve issues that would affect 7nm EUV as well.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
Charlie makes the wild assumption EUV will only be used on 7nm. Both Samsung and TSMC have a 7nm node (=10nm Intel) using EUV and Intel had 10nm+ and 10nm++ planed for respectively this year and next.

If Charlie read up a bit more about Intel 7nm he'd know it's a MASSIVE jump from 10nm equal to TSMC 3nm. Intel skipping 10nm volume like he claims is about ~0 percent chance. What was killed is first gen 10nm for desktop and likely regular mobile.

What will happen is that first gen 10nm will only roll out in ultra low power versions as these hit lower clocks and use smaller dies. 10nm+ with EUV is where volume will be for mobile, desktop and server.

7nm will be out in 2021 the very soonest, but 2022 seems more likely.
 

jpiniero

Lifer
Oct 1, 2010
14,513
5,160
136
I theorized that Intel might do a in-between node on EUV, but so far there haven't even been any rumors of that sort.

I thought they were working on some fixed 10nm+ or 10nm++ with better yield of which we will see products this and next year?

There was talk that Intel would reduce the density to fix the 10 nm problems, but that appears to not have happened. Icelake-U is pretty small; from the screenshot that I saw it doesn't appear that it's much bigger than Cannonlake-U.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
I theorized that Intel might do a in-between node on EUV, but so far there haven't even been any rumors of that sort.

They never specified the process improvements for 10nm+ and 10nm++. EUV seems all but certain.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
The information that I have compiled.

2018 -> 10nm risk
2020 -> 7nm risk
2022 -> 5nm risk
2025 -> 3nm risk
Products can ship in risk production do to processor development starting at the development stage of the node.

193i -> 10nm
EUVL -> 7nm
DSA+NIL -> 5nm
DWPL -> 3nm

It is R&D in a way that previous and forward lithographies can be used.

If EUV is not up to par Intel can skip it DSA+NIL or stay on 193i.
If DSA+NIL doesn't make it, they can skip it to direct write plasmonic lithography or stay on EUV and move to High-NA machines.

Intel is lithographically buffered, so the processes will come out on time.
 
Last edited:

jpiniero

Lifer
Oct 1, 2010
14,513
5,160
136
They never specified the process improvements for 10nm+ and 10nm++. EUV seems all but certain.

While they didn't give that much info, BK way back did say that the original 10+ and 10++ nodes were not using EUV and that they were targeting 7 for moving to EUV. Could they have looked into a 10 nmish process using EUV in the meantime as a backup plan? I suppose but it doesn't appear that they would be ready to take on any EUV process until 20/21 anyway.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
Intel has previously stated that 7nm is when they’ll first insert EUV.

They did? That'd be pretty damning since they went with sub 40nm metal pitch that needs quad paterning with no backup whatsoever. Going that route would not just be management but complete engineering failure.

In that case 10nm+ needs to move above 40nm metal pitch.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
I would not be naive enough to believe Intel claims about process before they really deliver.

You're confusing PR claims about when a product launches with what is known within the industry with regards to the transistor design (MP, CPP, fin height etc.). They are two very different things.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
But if 7nm is more complex than 10nm what are the chances Intel will have issues with it?

There’s no real guarantee that difficulty with one node means the next will be even harder. It’s pretty obvious that Intel made a few missteps with the way that built their 10nm process that didn’t work out and have been difficult to fix.

Starting fresh with 7nm means they can avoid some of the issues they’ve had entirely. The mistakes with 10nm have hopefully given them some valuable insights that make 7nm easier.
 

linkgoron

Platinum Member
Mar 9, 2005
2,286
810
136
There’s no real guarantee that difficulty with one node means the next will be even harder. It’s pretty obvious that Intel made a few missteps with the way that built their 10nm process that didn’t work out and have been difficult to fix.

Starting fresh with 7nm means they can avoid some of the issues they’ve had entirely. The mistakes with 10nm have hopefully given them some valuable insights that make 7nm easier.

If I recall correctly, that's exactly what people said after 14nm was delayed as well (regarding 10nm).

https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-and-core-i3-8121u-deep-dive-review/2

As part of those presentations, 10nm was a key part of it, with Intel stating that while 10nm was set to have more masking layers than 14nm, the company expected that the delays that bogged down 14nm would not be present in bringing 10nm to market.

“We were told that Intel has learned that the increase in development complexity of 14nm required more internal testing stages and masking implementations was a major reason for the delay, as well as requiring sufficient yields to go ahead with the launch. As a result, Intel is improving the efficiency testing at each stage and expediting the transfer of wafers with their testing protocols in order to avoid delays. Intel tells us that that their 10nm pilot lines are operating 50% faster than 14nm was as a result of these adjustments. So while the additional masking steps at 10nm which ultimately increases fixed costs, Intel is still quoting that their methods results in a reducing in terms of cost per transistor without needing a completely new patterning process.”

In this, the key part is that Intel had identified where it went wrong in 14nm, and was ready to remove those bottlenecks in its development of 10nm. Intel stated that 10nm would come with innovation, however going beyond to 7nm will require new materials and processes that Intel would introduce progressively.
 
  • Like
Reactions: krumme

moinmoin

Diamond Member
Jun 1, 2017
4,935
7,621
136
Since for 7nm Intel will be using EUV tools and those will be already in use for some time at TSMC and Samsung Intel may be able to avoid early pitfalls. That is if they don't deviate from common industry practices and try to bite off more than they can chew again.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
I would not be naive enough to believe Intel claims about process before they really deliver.
While I agree it’s wise to be cautious based on recent experience, EUV gives Intel the opportunity to avoid quad pattering. That should considerably improve Intel’s chance of success with 7nm (much better yields). We’ll just have to wait and see what Intel does with materials, geometries and chemistry and whether or not those possibilities create process problems. I think Intel has likely lost some confidence in their process engineering team, so more aggressive approaches will likely be rejected.
 

Dolan

Junior Member
Dec 25, 2017
14
10
51
Do you think that TSMC/Samsung will share their know-how with Intel?

Why there are not companies offering 14/10/7nm in China, Russia or Middle east? They have capital to buy whatever they want. Why it is Exclusive only in Taiwan/Korea? Maybe because it is not just that simple as buying 7nm-ready tool from ASML and pressing power button.

Also you are too optimistic about Intel's 7nm readiness/progress. There is lot difficult parts (single patterning EUV will not be sufficient for critical layers so they will have to go above SAQP to achieve that crazy high transistor density you probably heard about...). I asked bit about it and it will be more difficult than 10nm. Definitely harder. Just forget about dreams like "10 was delayed by stupid mistakes but 7 will yield ahead of schedule".


Look. Lately Intel is talking lot about future. Like 4+ years into future for process, architectures, "new" memory type... Jesus, they publicly talked about dedicated GPUs when they just started development (with nothing in hands and barely something on paper). So how you can be so confident about their 7nm? It can be well over 5-6 years in future until we will be able to buy something reasonable.
 

stockolicious

Member
Jun 5, 2017
80
59
61
Demand has been going up for Intel chips. They're guiding to $71.2B in sales this year, up from their original projection of $65B.

The sheer number of extra chips that Intel has to produce to meet that $6.2B extra in demand is nontrivial.

Be careful when assuming the increase in revenues was due to demand and volumes - if you look at the report much of the revenue growth was due to pricing. ASP's were higher for desktop and data center.