Thursday, May 9th 2019

Intel Switches Gears to 7nm Post 10nm, First Node Live in 2021

Intel's semiconductor manufacturing business has had a terrible past 5 years as it struggled to execute its 10 nanometer roadmap forcing the company's processor designers to re-hash the "Skylake" microarchitecture for 5 generations of Core processors, including the upcoming "Comet Lake." Its truly next-generation microarchitecture, codenamed "Ice Lake," which features a new CPU core design called "Sunny Cove," comes out toward the end of 2019, with desktop rollouts expected 2020. It turns out that the 10 nm process it's designed for, will have a rather short reign at Intel's fabs. Speaking at an investor's summit on Wednesday, Intel put out its silicon fabrication roadmap that sees an accelerated roll-out of Intel's own 7 nm process.

When it goes live and fit for mass production some time in 2021, Intel's 7 nm process will be a staggering 3 years behind TSMC, which fired up its 7 nm node in 2018. AMD is already mass-producing CPUs and GPUs on this node. Unlike TSMC, Intel will implement EUV (extreme ultraviolet) lithography straightaway. TSMC began 7 nm with DUV (deep ultraviolet) in 2018, and its EUV node went live in March. Samsung's 7 nm EUV node went up last October. Intel's roadmap doesn't show a leap from its current 10 nm node to 7 nm EUV, though. Intel will refine the 10 nm node to squeeze out energy-efficiency, with a refreshed 10 nm+ node that goes live some time in 2020.
The transition from 10 nm+ to 7 nm EUV will significantly increase transistor-densities. Intel is also making the manufacturing process efficient, by reducing "design rules" by 4 times, giving chip-designers more flexibility and creative freedom in the way they design nanoscopic circuitry. The process will also be optimized for heterogeneous chip designs, Foveros packaging (a very advanced form of MCM), and EMIB (reduced footprint interposer).

The 7 nm EUV node will receive two major updates in quick succession. The 7 nm+ node is slated for 2022, and succeeding 7 nm++ node in 2023. Intel did not detail the two besides illustrating performance/Watt gains by almost as much as the transition from 10 nm+ to 7 nm. Elsewhere in the market, the early 2020s could see TSMC 6 nm EUV take center-stage, and Samsung implement its 5 nm EUV node.
Intel will build an Xe enterprise GPGPU based on 7 nm EUV for market release in 2021. The company was specific in mentioning that an "enterprise GP-GPU" will be built on it, and not its entire Xe lineup that includes client-segment, professional, and cloud GPUs. The Xe discrete GPU team, led by Raja Koduri, is probably making a ropewalk, by giving Intel "something" to build on its own fabs, while seeking out Samsung's cutting-edge 5 nm EUV node for the rest of its lineup. Intel confirmed that the its first 7 nm product will be a GPGPU, followed closely by a server CPU.
Source: AnandTech
Add your own comment

23 Comments on Intel Switches Gears to 7nm Post 10nm, First Node Live in 2021

#1
TheGuruStud
More lies. This is basically 100% baloney to stop investors from suing, but it's just gonna give them ammo if you ask me. What are they gonna do? Paper launch 6mo ahead for a crappy mobile chip? Release a low core count xeon? It's just sad.
Posted on Reply
#2
Xzibit
They said their 7nm would be the first with EUV? So no EUV for 10nm. Thought that was a big portion for the delay.
Posted on Reply
#3
ZoneDymo
I find the claim "relentless innovation continues" quite funny coming from Intel tbh.
Posted on Reply
#4
btarunr
Editor & Senior Moderator
TheGuruStudMore lies. This is basically 100% baloney to stop investors from suing, but it's just gonna give them ammo if you ask me. What are they gonna do? Paper launch 6mo ahead for a crappy mobile chip? Release a low core count xeon? It's just sad.
Investors will bail regardless. 2021 is too late for 7 nm EUV. By then TSMC and Samsung will have cornered the 6 nm EUV market with a mature node, and Samsung will begin 5 nm EUV.

Even Raja doesn't trust Intel foundry, so he's designing a token Xe product (a Larrabee-like GPGPU accelerator) for Intel 7 nm EUV, while every other Xe GPU will be built on Samsung 5 nm EUV.
Posted on Reply
#5
londiste
btarunr2021 is too late for 7 nm EUV. By then TSMC and Samsung will have cornered the 6 nm EUV market with a mature node, and Samsung will begin 5 nm EUV.
The inconsistencies with node naming remain. Intel 7nm is roughly the same as TSMC/Samsung 5nm.
TSMC intends their 5nm to start volume production in the end of 2020.

Edit:
btarunrWhen it goes live and fit for mass production some time in 2021, Intel's 7 nm process will be a staggering 3 years behind TSMC, which fired up its 7 nm node in 2018. AMD is already mass-producing CPUs and GPUs on this node. Unlike TSMC, Intel will implement EUV (extreme ultraviolet) lithography straightaway. TSMC began 7 nm with DUV (deep ultraviolet) in 2018, and its EUV node went live in March. Samsung's 7 nm EUV node went up last October. Intel's roadmap doesn't show a leap from its current 10 nm node to 7 nm EUV, though. Intel will refine the 10 nm node to squeeze out energy-efficiency, with a refreshed 10 nm+ node that goes live some time in 2020.
So, point by point:
1. Intel 7nm (mass production 2021) is comparable to TSMC 5nm (volume production end of 2020). That is on 4th slide as well.
2. Intel has no choice when it comes to EUV due to feature sizes on a small node. 7nm (and Intel's 10nm) are right on the line where not using EUV is still feasible but barely. Both TSMC and Samsung are starting to partially use EUV on 7nm (7nm+ for TSMC and Samsung right away). EUV has some teething problems, technical side of things has been mostly worked out but tooling availability is still a problem and things just take time.
3. Intel's roadmap is curious, both in terms of foundry and CPUs. Looks like there will be very limited Intel 10nm and despite fast 10nm iterations they intend to largely skip it and shoot for 7nm in 2021. This makes perfect sense both financially as well as technically. Setting up foundries for a node is expensive (billions, with a B) and time consuming. If the intended lifetime of a node is about a year before the next one is ready, it would not make much sense to widely adopt the node.
Posted on Reply
#6
PhantomTaco
btarunrInvestors will bail regardless. 2021 is too late for 7 nm EUV. By then TSMC and Samsung will have cornered the 6 nm EUV market with a mature node, and Samsung will begin 5 nm EUV.

Even Raja doesn't trust Intel foundry, so he's designing a token Xe product (a Larrabee-like GPGPU accelerator) for Intel 7 nm EUV, while every other Xe GPU will be built on Samsung 5 nm EUV.
What TSMC is offering as "7nm" isn't 7nm at all. The fact that they're using DUV is proof of that. DUV is a fancy word for slightly more advanced current lithography methodologies (using the 193nm excimer laser). NOW that they just got their EUV machines from ASML (which use 13.5nm EUV as a source) a few months back they could be doing proper 7nm, but even then they're only using it for "four non-critical layers" of the wafer process. That in itself is indication that their naming nomenclature doesn't really stack up. Their proper 7nm is likely to be the 7nm+ or 6nm that they've been discussing, where they will use full on EUV for the lithography process.
Posted on Reply
#7
londiste
PhantomTacoWhat TSMC is offering as "7nm" isn't 7nm at all. The fact that they're using DUV is proof of that. DUV is a fancy word for slightly more advanced current lithography methodologies (using the 193nm excimer laser). NOW that they just got their EUV machines from ASML (which use 13.5nm EUV as a source) a few months back they could be doing proper 7nm, but even then they're only using it for "four non-critical layers" of the wafer process. That in itself is indication that their naming nomenclature doesn't really stack up. Their proper 7nm is likely to be the 7nm+ or 6nm that they've been discussing, where they will use full on EUV for the lithography process.
Naming nomenclatures do not stack up in general. This went all the way to hell several nodes ago and today the name does not match up to any particular feature size. Different foundries naming the nodes differently is not new either.

TSMC is using EUV on 4 non-critical layers on their 7nm+ node. They have not been very specific but 5nm is going to use - at least mostly, if not completely - EUV. The intermediate 6nm seems to be in large part just from extended use of EUV on otherwise 7nm node. Samsung is following largely the same path except they skipped doing 7nm without any EUV.
Posted on Reply
#8
PhantomTaco
londisteNaming nomenclatures do not stack up in general. This went all the way to hell several nodes ago and today the name does not match up to any particular feature size. Different foundries naming the nodes differently is not new either.

TSMC is using EUV on 4 non-critical layers on their 7nm+ node. They have not been very specific but 5nm is going to use - at least mostly, if not completely - EUV. The intermediate 6nm seems to be in large part just from extended use of EUV on otherwise 7nm node. Samsung is following largely the same path except they skipped doing 7nm without any EUV.
Problem with that is 193nm excimers you aren't able to actually get to ITRS spec for 7nm. You can't have enough refractory lenses set to do it efficiently and immersion lithography isn't enough to get you to that point properly either last i checked. My whole point was that what they're calling 7nm is anything but.
Posted on Reply
#9
londiste
PhantomTacoProblem with that is 193nm excimers you aren't able to actually get to ITRS spec for 7nm. You can't have enough refractory lenses set to do it efficiently and immersion lithography isn't enough to get you to that point properly either last i checked. My whole point was that what they're calling 7nm is anything but.
SAQP? Definitely cannot be done efficiently but it is kind of possible. I believe the understanding is that EUV becomes necessary when going to under 40nm metal pitch. Of course, this is a vast simplification but seems to be a very good indicator. This, incidentally, is where TSMC stayed at while Samsung and Intel went below (36nm for both if my memory serves right).

By the way, what is ITRS spec for 7nm? Quick Google didn't give me a proper answer, hoping you have a link at hand :)
Posted on Reply
#10
TheGuruStud
PhantomTacoWhat TSMC is offering as "7nm" isn't 7nm at all. The fact that they're using DUV is proof of that. DUV is a fancy word for slightly more advanced current lithography methodologies (using the 193nm excimer laser). NOW that they just got their EUV machines from ASML (which use 13.5nm EUV as a source) a few months back they could be doing proper 7nm, but even then they're only using it for "four non-critical layers" of the wafer process. That in itself is indication that their naming nomenclature doesn't really stack up. Their proper 7nm is likely to be the 7nm+ or 6nm that they've been discussing, where they will use full on EUV for the lithography process.
But it doesn't matter when it's actually profitable with high yields...unlike the 500 lb baby monkey flailing around.
Posted on Reply
#11
Gungar
btarunrInvestors will bail regardless. 2021 is too late for 7 nm EUV. By then TSMC and Samsung will have cornered the 6 nm EUV market with a mature node, and Samsung will begin 5 nm EUV.

Even Raja doesn't trust Intel foundry, so he's designing a token Xe product (a Larrabee-like GPGPU accelerator) for Intel 7 nm EUV, while every other Xe GPU will be built on Samsung 5 nm EUV.
Wtf ? you don't know that Intel process have always been smaller than other manufacturers??? I am really disappointed to see this kinda of comment on Techpowerup!
Posted on Reply
#12
Assimilator
Seems like 7nm is finally the ticket to Intel getting back on track. They're only dedicating resources to 10nm because they have designs for it that they don't have time to redo for 7nm.

Reading more and more about their 10nm fiasco, it seems a lot of the issue was that Intel needed EUV to make it actually work, but the technology literally wasn't available. So they were forced to attempt to implement the far-more-complex multi-patterning... which has finally come right, just in time for EUV to be practical. Well, that's the risk of being a tech leader; sometimes you are forced to take detours, and sometimes those detours end up being dead ends.
Posted on Reply
#13
SL2
TheGuruStudPaper launch 6mo ahead for a crappy mobile chip?
What makes it crappy?

Don't think this qualifies as a paper launch.
Posted on Reply
#14
TheGuruStud
MatsWhat makes it crappy?

Don't think this qualifies as a paper launch.
Zero details, just vauge references.... Hint: It's worse than the current ones and apparently nowhere near actually making it into products.
Posted on Reply
#15
Rahnak
Yeah.. I'll believe it when I see it.
Posted on Reply
#16
BorgOvermind
What intel 10nm ?

For now, intel states that "is nearly ready to ship 10-nanometer processors in large volumes "... in June.
Posted on Reply
#17
TheGuruStud
BorgOvermindWhat intel 10nm ?

For now, intel states that "is nearly ready to ship 10-nanometer processors in large volumes "... in June.
So ready they only have "simulated" performance numbers for em. Intel should write a book, "How to become the laughing stock of the tech industry in these 10 easy steps!"
Posted on Reply
#18
SL2
Intel had a test system running last year, possibly running at 1 GHz for all we know, but my point is that it's not all slides at this point.

www.techpowerup.com/250549/intel-demoes-sunnycove-high-performance-core

If you WANT it to be crappy, that's a whole different thing, but it's not much to make an assumption out of. :D If anything, I'm a AMD fan myself, and I'd like to see AMD take some of the CPU market share, but this isn't a paper launch. You need model numbers, price, launch dates, performance tests and maybe even reviews for that.

It may still be crap, but we have zero evidence that it is. The fact that Intel stayed at 4 core and 14 nm forever, or changes the socket too often, or sells multiplier locked CPU's, or calls the 9900K a 95 W SKU isn't evidence for Ice Lake being crap.
Posted on Reply
#19
efikkan
Intel's 7 nm process will be a staggering 3 years behind TSMC, which fired up its 7 nm node in 2018. AMD is already mass-producing CPUs and GPUs on this node.
You should know better than to compare TSMC 7nm with Intel 7nm when it's actually comparable to Intel's 10nm.

For the record;
Intel was first to ship a product on this generation of nodes in April 2018 followed by AMD in ~December 2018, but neither have shipped volumes yet. But at least AMD have shipped the largest chip so far, and both expect to ship "volumes" in Q3 2019.
Posted on Reply
#20
Assimilator
efikkanYou should know better than to compare TSMC 7nm with Intel 7nm when it's actually comparable to Intel's 10nm.
Oh come now, are you really expecting bta to write a 2019 "article" that isn't sensationalist clickbait?
Posted on Reply
#21
HD64G
Totally expected move from any rational person following the tech industry and news. Point is, it isn't too certain they will succeed in 7nm either but the possibilities are better for sure. And the main question now is, how far behind Zen arch will the get until they present both a new arch and a feasible and efficient process to battle the one that other fab owners will have then. Imho, the fall of the giant has already begun. Let's hope that competition will exist in a year or so in the consumer CPU market.
Posted on Reply
#22
WikiFM
AssimilatorThey're only dedicating resources to 10nm because they have designs for it that they don't have time to redo for 7nm.

Reading more and more about their 10nm fiasco, it seems a lot of the issue was that Intel needed EUV to make it actually work, but the technology literally wasn't available. So they were forced to attempt to implement the far-more-complex multi-patterning...
I totally agree, Intel's 10 nm is too dense for DUV, that is why it will never be a good process, Intel was too ambitious. TSMC's 7 nm worked out in DUV cause is less denser than Intel's 10 nm. By comparison an Apple A12 has around 82.86 MTr/mm2 and Intel Cannon Lake around 90.78 MTr/mm2.

I think that desktops will never see a 10 nm product, cause Comet Lake will launch in Q1 2020 and Rocket Lake in Q1 2021 both using 14 nm, Golden Cove would launch at the end of 2021 in 7 nm for desktop. So desktops will skip both Ice Lake and Tiger Lake in 10 nm (and Cannon Lake which by now is more than certain that is dropped).
londisteBy the way, what is ITRS spec for 7nm? Quick Google didn't give me a proper answer, hoping you have a link at hand :)
irds.ieee.org/images/files/pdf/2017/2017IRDS_MM.pdf page 12
Posted on Reply
Add your own comment
Dec 8th, 2024 21:55 CST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts