Intel 10nm and GF 7nm at IEDM 2017

Discussion in 'CPUs and Overclocking' started by raghu78, Oct 19, 2017.

  1. Dayman1225

    Dayman1225 Senior member

    Joined:
    Aug 14, 2017
    Messages:
    351
    Likes Received:
    137
    First time for what? Cobalt? Yes, though it seems Intel decided to use it quite a bit more which will effect yields negatively. AFAIK SAQP is new to both GloFo and Intel, but I think yet again Intel used it in more areas than GloFo, yet again effecting yields negatively.
     
    CatMerc likes this.
  2. maddie

    maddie Golden Member

    Joined:
    Jul 18, 2010
    Messages:
    1,745
    Likes Received:
    378
    This isn't the warp drive from out of nowhere, or like how the EM drive arrived. Everyone in the industry is aware of the potential benefits of Cobalt.
     
    CatMerc likes this.
  3. maddie

    maddie Golden Member

    Joined:
    Jul 18, 2010
    Messages:
    1,745
    Likes Received:
    378
    Are you saying this a good decision or a poorly thought out one? Seems like overreach to me. For certain, they're paying a heavy price.
     
  4. raghu78

    raghu78 Diamond Member

    Joined:
    Aug 23, 2012
    Messages:
    3,819
    Likes Received:
    1,000
    Yeah it looks like Intel bit off a bit more than they could chew.

    1. Cobalt contacts, metal layers M0,M1 and cobalt cap for copper interconnect M2-M5
    2. SAQP for lowest metal layers.
    3. Single Dummy gate
    4. Contact over Active Gate.

    Anyway this experience at 10nm will make them be a bit more conservative at future nodes.
     
    CatMerc and Dayman1225 like this.
  5. raghu78

    raghu78 Diamond Member

    Joined:
    Aug 23, 2012
    Messages:
    3,819
    Likes Received:
    1,000
    yeah in hindsight it looks to have been a poor decision. GF saying sticking with copper/low k for metal interconnect layers brings reliability benefits in the form of reduced complexity and yield risk seems to indicate that the aggressive use of cobalt is probably causing yield challenges at Intel 10nm.
     
    Dayman1225 likes this.
  6. Dayman1225

    Dayman1225 Senior member

    Joined:
    Aug 14, 2017
    Messages:
    351
    Likes Received:
    137
    As they say, go big or go home :)
     
  7. raghu78

    raghu78 Diamond Member

    Joined:
    Aug 23, 2012
    Messages:
    3,819
    Likes Received:
    1,000
    Thats a poor line of thinking especially when Intel had the process node lead at 14nm. They just needed to deliver a decent bump in perf and density with a good time to market lead over foundry 7nm( atleast 1 yr). Intel could then have moved to their next node aggressively using EUV by 2020.
     
  8. stuff_me_good

    stuff_me_good Member

    Joined:
    Nov 2, 2013
    Messages:
    191
    Likes Received:
    13
    Yeah, there is a time go big or go home and time when not.
     
  9. CatMerc

    CatMerc Senior member

    Joined:
    Jul 16, 2016
    Messages:
    778
    Likes Received:
    732
    Basically Intel was more aggressive. The aggression didn't pay off.
     
  10. LTC8K6

    LTC8K6 Lifer

    Joined:
    Mar 10, 2004
    Messages:
    25,754
    Likes Received:
    868
    Well, Intel is technically trying to do 7nm with cobalt now, since their 10nm is like everyone else's 7nm.

    When Intel goes to what they call 7nm, it will be like everyone else's 5nm, I think.

    You have to try if you want to advance.
     
  11. LTC8K6

    LTC8K6 Lifer

    Joined:
    Mar 10, 2004
    Messages:
    25,754
    Likes Received:
    868
    I don't think we know yet if it will pay off.
     
  12. raghu78

    raghu78 Diamond Member

    Joined:
    Aug 23, 2012
    Messages:
    3,819
    Likes Received:
    1,000
    The delays to 10nm are going to hurt Intel a lot in the short term till 2020/2021. Long term we have to wait and see how things play out.

    Sent from my SM-G935V using Tapatalk
     
  13. CatMerc

    CatMerc Senior member

    Joined:
    Jul 16, 2016
    Messages:
    778
    Likes Received:
    732
    Intel went from a few years lead from other foundries (not in names but in actual density and performance) to being on par with others. 10nm already lost them that. By the time we will see HVM 10nm products, TSMC will already be in HVM of their equally dense 7nm.

    Whatever damage it might do or if it will be successful doesn't change the massive loss in time to market lead.
     
    moinmoin likes this.
  14. LTC8K6

    LTC8K6 Lifer

    Joined:
    Mar 10, 2004
    Messages:
    25,754
    Likes Received:
    868
    I never thought Intel really had the 3 year lead it claimed to have, or whatever it was, exactly.
     
  15. CatMerc

    CatMerc Senior member

    Joined:
    Jul 16, 2016
    Messages:
    778
    Likes Received:
    732
    Intel's 14nm Skylake released in late 2015. Other foundries 10nm just caught up with its density using their 10nm. The first product with 10nm was Galaxy S8 released in April 2017. That's a very large lead whatever way you slice it.

    I don't buy into the 3 years lead marketing either, but there was a lead.
     
  16. maddie

    maddie Golden Member

    Joined:
    Jul 18, 2010
    Messages:
    1,745
    Likes Received:
    378
    I agree.

    The point is not that eventually it will work, but what was the price. Losing the process lead and apparently still falling behind relative to the competition is not a win.
     
    CatMerc likes this.
  17. raghu78

    raghu78 Diamond Member

    Joined:
    Aug 23, 2012
    Messages:
    3,819
    Likes Received:
    1,000
    Yes Intel atleast had a density lead with Intel 14nm vs foundry 16/14nm. The foundries caught up in H1 2017 with their 10nm process. TSMC is now going to beat Intel 10nm in density next year with their N7 6T low power process. TSMC will most likely beat Intel in terms of time to market and HVM yield ramp. GF 7LP 6T (7SoC) too beats Intel 10nm on density. Intel went from a significant density lead in 2015-2016 to lagging in 2018-2019. Thats how bad Intel messed up.

    https://www.semiwiki.com/forum/cont...alfoundries-discloses-7nm-process-detail.html

    Intel 10nm is broken according to charlie of semiaccurate. We need to wait and see when 10+ arrives and if it fixes the issues at Intel 10. Intel's recent slides with 10++ bringing a rearchitected metal stack lends further evidence that something was badly messed at 10. It could be the aggressive use of cobalt, SAQP for lowest metal layers or just the sheer complexity of multiple first time technologies at Intel (Contact Over Active Gate, Single dummy gate along with the above mentioned features) not working well together or being very hard to yield.
     
    #68 raghu78, Dec 8, 2017
    Last edited: Dec 8, 2017
  18. Ajay

    Ajay Diamond Member

    Joined:
    Jan 8, 2001
    Messages:
    4,376
    Likes Received:
    1,151
    Yeah, I'd bet that it was a combination of issues given the amount of time Intel is taking to resolve the problem. I'm concerned at this point that even IceLake may have relatively low yields (historically) for decent clock rates - expecting something like the CoffeeLake rollout (for different reasons, obviously). I also wonder how much the 10nm debacle is hurting 7nm deployment. Intel has great depth, but it seems like a lot of process engineers are likely being pulled in for long hours on 10nm (just a guess).
     
    raghu78 likes this.
  19. Bondrewd

    Bondrewd Senior member

    Joined:
    Oct 16, 2017
    Messages:
    358
    Likes Received:
    114
    I think Intel went all-in with their 10nm node, thinking they have enough time to spare should anything go wrong.
    Well, and here we have it.
     
  20. Ajay

    Ajay Diamond Member

    Joined:
    Jan 8, 2001
    Messages:
    4,376
    Likes Received:
    1,151
    Only if the Cobalt depo process is immature. From the materials Hans provided, it looks like it should be more reliable (better wetting for lower defect rates).
     
  21. Ajay

    Ajay Diamond Member

    Joined:
    Jan 8, 2001
    Messages:
    4,376
    Likes Received:
    1,151
    Well, it looks like they did everything they could to best the competition. They will probably still be tops in electrostatics, but lose most of their density lead due competitors catching up (not that x86 CPUs are optimal for comparing density).
     
    #72 Ajay, Dec 8, 2017
    Last edited: Dec 9, 2017
  22. Arachnotronic

    Joined:
    Mar 10, 2006
    Messages:
    11,475
    Likes Received:
    1,805
    Business as usual for them, then, after a brief density lead.
     
    Ajay likes this.
  23. Ajay

    Ajay Diamond Member

    Joined:
    Jan 8, 2001
    Messages:
    4,376
    Likes Received:
    1,151
    Good point. Sadly, most of the process focus is on density. While important, the other side of the coin is xtor performance and perf/watt.
     
  24. Bondrewd

    Bondrewd Senior member

    Joined:
    Oct 16, 2017
    Messages:
    358
    Likes Received:
    114
    You forgot time to market.
    It's equally, if not more, important as other metrics.
     
    Ajay and raghu78 like this.