• Hey there Guest! This holiday season you have a chance to win some cool AnandTech swag! Check out “The Holiday Hotlist” giveaway, which you can find here.

[SemiAccurate] Intel kills off the 10nm process

mikk

Platinum Member
May 15, 2012
2,376
4
126
Intel confirmed they are making further node improvements to 14 nm. Maybe Comet will have further improvements.

Further 14nm++ improvements, doesn't mean there is a 14nm+++
 
Apr 27, 2000
10,192
112
126
https://twitter.com/david_schor/status/1064922837197209600

David Schor is suggesting that the 8160 (5G modem) is going to be fabbed at TSMC. Doesn't say what node. Makes sense, given they have to make Apple's deadline or they will lose the contract forever and presumably at the time of the decision couldn't have trusted that the New 10 nm node would have worked out.

Intel making modems at TSMC = lulz

You would think they could fab something as small/simple as a modem on their "failed" 10nm, unless they're already in the process of retooling those fabs for something else.
 

maddie

Platinum Member
Jul 18, 2010
2,200
63
136
Intel making modems at TSMC = lulz

You would think they could fab something as small/simple as a modem on their "failed" 10nm, unless they're already in the process of retooling those fabs for something else.
It's worse than that.

Apple wants modem from Intel who gets TSMC to make it.

Apple asks, why am I paying you?
 
Apr 27, 2000
10,192
112
126
To design the chip and get it produced?
Yeah but Intel only has the business so that Apple can spite Qualcomm. If Apple could bury the hatcher, they'd get a better modem design from Qualcomm, and hey that would be fabbed at TSMC as well. Probably.
 

scannall

Golden Member
Jan 1, 2012
1,250
47
136
Yeah but Intel only has the business so that Apple can spite Qualcomm. If Apple could bury the hatcher, they'd get a better modem design from Qualcomm, and hey that would be fabbed at TSMC as well. Probably.
I have to wonder how long it will be until Apple starts doing their own modems as well. They can't be happy with the current state of affairs. Plus integrating them into their SOC's would be a nice added bonus.
 

jpiniero

Diamond Member
Oct 1, 2010
5,805
42
126
Further 14nm++ improvements, doesn't mean there is a 14nm+++
Intel is moving away from the + marketing so even if they don't call it 14+++++ it's pretty much the same thing.
 
Mar 11, 2004
17,409
56
126
I have to wonder how long it will be until Apple starts doing their own modems as well. They can't be happy with the current state of affairs. Plus integrating them into their SOC's would be a nice added bonus.
I think patents were the issue with 4G/LTE stuff (Intel bought Infineon's wireless division in 2010). That should change some for 5G (I assume for the better), so that'd be my guess. They'll probably still need a 4G/LTE and possibly 3G modem though so it'll probably be a decade before we get Apple chips with purely Apple modem. Well unless they work a deal to license the 4G stuff, which should be possible (I think Qualcomm's board was pushing for them to move to licensing as their major income, although after they got bought by Broadcom I'm not sure if that changed).
 

Mopetar

Diamond Member
Jan 31, 2011
4,193
24
126
Intel making modems at TSMC = lulz

You would think they could fab something as small/simple as a modem on their "failed" 10nm, unless they're already in the process of retooling those fabs for something else.
Could also be a case that without the 10 nm node online yet, Intel doesn't have enough capacity to handle everything. We've already seen stories that they're having a hard time keeping all of the chip markets supplied which is leading to shortages.

They probably don't make as much selling modems to Apple as they do from their other lines, which makes it a good candidate for moving to TSMC. I'll agree that it's not a good look for them, but it's better than trying to squeeze it in on their own, already limited, production capacity.
 
Apr 27, 2000
10,192
112
126
(I think Qualcomm's board was pushing for them to move to licensing as their major income, although after they got bought by Broadcom I'm not sure if that changed).
The Broadcom sale was canned by the Feds. But it is possible that Qualcomm prepared several changes to accommodate the sale and followed through on them anyway.

Could also be a case that without the 10 nm node online yet, Intel doesn't have enough capacity to handle everything. We've already seen stories that they're having a hard time keeping all of the chip markets supplied which is leading to shortages.

They probably don't make as much selling modems to Apple as they do from their other lines, which makes it a good candidate for moving to TSMC. I'll agree that it's not a good look for them, but it's better than trying to squeeze it in on their own, already limited, production capacity.
We'll probably never know how much 10nm fab capacity came online before Intel effectively pulled the plug on it. They did make a few 2C 10nm mobile chips (that were terrible), so obviously that have/had some capacity.
 

Mopetar

Diamond Member
Jan 31, 2011
4,193
24
126
We'll probably never know how much 10nm fab capacity came online before Intel effectively pulled the plug on it. They did make a few 2C 10nm mobile chips (that were terrible), so obviously that have/had some capacity.
It's hard to say exactly how much they had, but even Intel admitted that 14+++ was going to have superior performance characteristics. Maybe they would have sufficient wafers to dedicate most of the production to making modems for Apple (and baseband chips tend not to be overly large, so even poor yields wouldn't hurt too much) but if the performance wasn't up to snuff or there were other reliability issues (the one Cannon Lake CPU they did make didn't even have the GPU enabled and had a lower boost than the Kaby Lake processor it was supposed to replace) they might have backed off from that plan.

Even if they are on the way to fixing 10nm and getting it running, there's probably too much uncertainty surrounding it for them to use for something they're contracted to deliver to another company, especially one like Apple.
 

Spartak

Senior member
Jul 4, 2015
203
30
86
The wild speculation / theories here are quite entertaining to read.

First of all the Qualcomm-Apple dispute has literally zero relevancy to the fab producing the modem. I wouldnt be surprised if Apple would actually prefer TSMC as it is their long standing partner, as long as the rumored TSMC alternative matches the power/performance characteristics agreed upon in the modem contract. 10nm TSMC seems the most likely alternative acceptable to Apple but depending on performance Intel might have to pony up some damages.

Second, I'm reading everywhere on this forum Intel 10 nm is dead yet some suggest Intel would give Apple a free ride on a more expensive and either superior or inferior (pick your choice) process when there are strict contractual obligations.

I'm not even sure where to start on how ridiculous these theories are. I guess people have no idea how multibillion contracts work.
 
Apr 27, 2000
10,192
112
126
First of all the Qualcomm-Apple dispute has literally zero relevancy to the fab producing the modem.
That just isn't true, at all. Had Apple continued buying modems from Qualcomm, it's all but guaranteed that Qualcomm would have used TSMC as the fab. Apple choosing Intel was supposed to guarantee that Intel would use its own fabs for said modems; and yet, here we are.

Second, I'm reading everywhere on this forum Intel 10 nm is dead yet some suggest Intel would give Apple a free ride on a more expensive and either superior or inferior (pick your choice) process when there are strict contractual obligations.
Who said anything about a free ride?

Intel had planned to build the 8160s on their own 10nm. They had the 10nm fabs up and running. They were fabbing 10nm chips, albeit bad ones. That "more expensive" node was exactly what Intel had planned for their next stage of modem production. Due to the small size and relatively low clockspeed/low power of the modems in question, it's rational to conclude that Intel could have punted on all their previous 10nm projects and just used whatever capacity they had on modems. That way the 10nm process wouldn't be a total loss. And yet, here we are . . .

I'm not even sure where to start on how ridiculous these theories are. I guess people have no idea how multibillion contracts work.
Well, for starters, if Intel told Apple "hey guys, we're going to make your 5G modems on our 10nm process", that Intel would be obligated to use their own 10nm process, and might have to pay a penalty for delays that might be incurred moving them to a TSMC process instead.
 
Feb 23, 2017
306
58
96
It's ok; he's simply in denial.
Intel will have gone from a huge process lead to a process deficit, and over a time frame where they had been expected to increase that process lead.
Intel are kind of stuck, but as long as they get things sorted by the time DDR5 is abou then they'll be perfectly fine. The reason being that AMD simply would not be able to ramp production enough in that time frame to even be in a position to dominate.
 

jpiniero

Diamond Member
Oct 1, 2010
5,805
42
126
What is your source?
Ashraf, plus there was an interview with an Intel exec (forget who) which basically downplayed the whole 14+ vs 14++ marketing comparison.
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,188
9
106
7nm @ Intel:
~27-nm Fin Pitch
~47-nm Gate Pitch
~36-nm Metal-1 Pitch
Aggressive SAQP/SADP BEOL pitch.

Performance-only (no power)
Intel 14nm => 1x
Intel 14nm+ => 1.12x
Intel 14nm++ => 1.26x
Intel 10nm => 1.25x
Intel 10nm+ => 1.4x
Intel 7nm => 1.8x

From what I can further gather is that "future" i7/i9/Xeon will be first to used 7-nm. Since, it is aimed at high performance first. Intel is throwing the meteor down and its going for >80-core Chiplet design with full native AVX512 bandwidth.

7nm+ is more aggressive on performance, but the numbers aren't out yet.

Also, 7-nm contains the fixes which were guaranteed with the new 193i tools. So, cumulative yields are significantly higher than 10-nm and maybe even on par with 14-nm.

7-nm(10-nm with fixes and faster FEOL) sooner than Icelake.
5-nm(old 7-nm, but is no longer Tri-gate) is 2022.
 
Last edited:
Feb 23, 2017
306
58
96
It's good to know that their 7nm will be out earlier than their supposedly defunct 10nm process. Still, that doesn't mean we'd be seeing it any time this side of armageddon.
43% increase in performance over their current process? I'm not convinced. AMD aren't looking at anywhere near that despite starting from a worse 14nm base. I don't see how Intel will skip 10nm after issues and then be more successful at 7nm than those that went directly to 7nm in the first place.
 

TheGiant

Senior member
Jun 12, 2017
251
8
76
7nm @ Intel:
~27-nm Fin Pitch
~47-nm Gate Pitch
~36-nm Metal-1 Pitch
Aggressive SAQP/SADP BEOL pitch.

Performance-only (no power)
Intel 14nm => 1x
Intel 14nm+ => 1.12x
Intel 14nm++ => 1.26x
Intel 10nm => 1.25x
Intel 10nm+ => 1.4x
Intel 7nm => 1.8x

From what I can further gather is that "future" i7/i9/Xeon will be first to used 7-nm. Since, it is aimed at high performance first. Intel is throwing the meteor down and its going for >80-core Chiplet design with full native AVX512 bandwidth.

7nm+ is more aggressive on performance, but the numbers aren't out yet.

Also, 7-nm contains the fixes which were guaranteed with the new 193i tools. So, cumulative yields are significantly higher than 10-nm and maybe even on par with 14-nm.

7-nm(10-nm with fixes and faster FEOL) sooner than Icelake.
5-nm(old 7-nm, but is no longer Tri-gate) is 2022.
where did you get those informations?

so that means 14nm more+ until late 2020 ?
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,188
9
106
where did you get those informations?

so that means 14nm more+ until late 2020 ?
1. Various sources.
2. 7-nm is late 2019. <-- Not EUV
7-nm+ is late 2020. <-- 200 wph EUV or 200 wph JFIL // wph => wafer per hour.
Which ever doesn't have a back-order and meets the goal.
43% increase in performance over their current process?
Hyperscaling in 10nm => Area
Hyperscaling in 7nm => Performance, Power, Area, Cost, Yield.

Other than that, Intel will need to show and tell. For the exacts what they changed from 10-nm problem to the 7-nm non-problem. Development for 7-nm ASICs started from last year around or before September 2017.
 
Mar 10, 2004
27,788
45
126
7nm @ Intel:
~27-nm Fin Pitch
~47-nm Gate Pitch
~36-nm Metal-1 Pitch
Aggressive SAQP/SADP BEOL pitch.

Performance-only (no power)
Intel 14nm => 1x
Intel 14nm+ => 1.12x
Intel 14nm++ => 1.26x
Intel 10nm => 1.25x
Intel 10nm+ => 1.4x
Intel 7nm => 1.8x

From what I can further gather is that "future" i7/i9/Xeon will be first to used 7-nm. Since, it is aimed at high performance first. Intel is throwing the meteor down and its going for >80-core Chiplet design with full native AVX512 bandwidth.

7nm+ is more aggressive on performance, but the numbers aren't out yet.

Also, 7-nm contains the fixes which were guaranteed with the new 193i tools. So, cumulative yields are significantly higher than 10-nm and maybe even on par with 14-nm.

7-nm(10-nm with fixes and faster FEOL) sooner than Icelake.
5-nm(old 7-nm, but is no longer Tri-gate) is 2022.
Can I have ice cream on that pie?
 
Sep 9, 2017
27
0
41
Intel is throwing the meteor down and its going for >80-core Chiplet design with full native AVX512 bandwidth.
So, that's Intel's "Core 2 Duo" response for 2019, I assume.

Like others have stated, can you specify your sources, please?
 

jpiniero

Diamond Member
Oct 1, 2010
5,805
42
126
Ah, Nosta being Nosta. There were rumors that Intel had thought about renaming the 10 nm node to 7 because of the foundries; but that doesn't appear to be happening. What definitely won't happen is Intel hitting the specs (or even denser, as Nosta is suggesting) of the original 10 nm without EUV being involved.

My prediction is that Icelake Client is based upon a modified version of the 14 nm SoC process they call 10 nm; and maybe they will throw in the FPGAs on it. But that's it for 19 and 20.
 

ASK THE COMMUNITY