TSMC 28nm: major changes?

Article: Process Technology at IEDM 2008
By: David Kanter (dkanter.delete@this.realworldtech.com), August 5, 2009 12:04 pm
Room: Moderated Discussions
anon (anon@anons.com) on 8/5/09 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 8/5/09 wrote:
>---------------------------
>>One Anonymous (unknown@unknown.com) on 8/5/09 wrote:
>>---------------------------
>>>Great article David, congratulations. I just thought I'd comment on the TSMC part
>>>because it seems there's something big you didn't catch. Note I'm not being particularly
>>>bullish on their HKMG execution prospects even now, although that's not really the point.
>>>
>>>The 28nm process TSMC presented is outdated. That variant will never go into production: TSMC has decided to switch from gate-first HKMG to gate-last HKMG. This was confirmed in the 2Q09 conference call where Morris Chang himself made sure the question was answered. This is clear in the SeekingAlpha transcript or in this blog post: http://blog.shrinkingviolence.com/2009/07/tsmc-explains-40nm-yield-probl.html
>>>
>>>This will probably improve the performance numbers a bit - but I don't know by
>>>how much. Assuming they don't change most other things (at least it'll probably
>>>still have the same density as before), it should be a very interesting comparison
>>>point for gate-first versus gate-last.
>>>
>>>Such a fundamental change means a delay, of course. LaPedus has covered it here although some of his numbers don't seem right: http://www.eetimes.eu/germany/219000163
>>
>>>
>>>- in reality the 28HP process has slipped from 2Q10 for 'risk production' to 4Q10 for 'tape-outs'. Interestingly, this is fully aligned with GlobalFoundries' updated bulk roadmap: despite all the hype, 32nm Bulk HKMG will never go into production at GF and 28nm Bulk HKMG tape-outs are also scheduled for 4Q10: http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=218500937&pgno=2
>>>
>>>Of course it's clear that TSMC's struggles are with high-k, not 28nm. 28LPT/28LP is still basically on track for 1Q10. Early customers taping out in Q1 are likely to be Qualcomm for Snapdragon3 and NVIDIA for the Tegra3, plus maybe some companies in the CE world. Compared with GlobalFoundries' 40LP ramp in 2Q10 and tape-outs for 28LP with HKMG in 1Q11, it's pretty clear TSMC remains at the top of the wireless & CE game. FWIW, there's some decent info on some other aspects of the 28nm processes in this brochure, probably made before the gate-last HKMG change: http://www.tsmc.com/download/english/a05_literature/00_28_Nanometer_Process_Technology.pdf
>>
>>>
>>>>
>>
>>You're right I did miss that and I appreciate you pointing it out. The article
>>was originally written from December through June and polished in mid-July...so
>>even had I noticed, I'm not sure I would have revised that much.
>>
>>I'm planning on having some analysis a little later, and I think that would be a great place to include an update.
>>
>>David
>
>
>The EETimes analysis of TSMC call contain an interesting quote:
>"Another wild theory: Intel recently licensed its Atom IP to TSMC. Intel and
>TSMC denied the deal involved high-k. Now, I'm really beginning to wonder if high-k is part of that deal."

>
>Talk about my enemy's enemy is a friend...

Not quite that simple.

You can't take a design optimized for HKMG and just run it on SiON. The beta ratio (NFET/PFET drive) is way better for HKMG, which is one of the things that enables a move away from dynamic logic towards static (if beta=1, there's very little or no need for dynamic logic).

Migrating the design could turn up all sorts of nasty speed paths, which would really impact performance.

For Intel - having TSMC succeed is essential, since they will enable Intel to reach new markets.

There is also the enemy of my enemy (GF) is a friend, but I don't know how much of a motivator that is.


David
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Article on 32nm process tech from IEDM 08 and VLSI 09 onlineDavid Kanter2009/08/03 01:22 AM
  Article on 32nm process tech from IEDM 08 and VLSI 09 onlineHans de Vries2009/08/03 02:28 AM
    Article on 32nm process tech from IEDM 08 and VLSI 09 onlineDavid Kanter2009/08/03 02:47 PM
      Article on 32nm process tech from IEDM 08 and VLSI 09 onlineHans de Vries2009/08/03 04:51 PM
        Article on 32nm process tech from IEDM 08 and VLSI 09 onlineDavid Kanter2009/08/03 10:09 PM
          Article on 32nm process tech from IEDM 08 and VLSI 09 onlineHans de Vries2009/08/04 11:56 AM
            Article on 32nm process tech from IEDM 08 and VLSI 09 onlineHans de Vries2009/08/04 12:31 PM
        Article on 32nm process tech from IEDM 08 and VLSI 09 onlineslacker2009/08/04 10:37 AM
          Article on 32nm process tech from IEDM 08 and VLSI 09 onlineAlexander Tomkins2009/08/04 11:36 AM
            Article on 32nm process tech from IEDM 08 and VLSI 09 onlineslacker2009/08/04 03:10 PM
              Article on 32nm process tech from IEDM 08 and VLSI 09 onlineAlexander Tomkins2009/08/04 07:12 PM
        Thanks for spotting thatDavid Kanter2009/08/05 01:30 AM
  Article on 32nm process tech from IEDM 08 and VLSI 09 onlinea reader2009/08/03 09:57 AM
    Article on 32nm process tech from IEDM 08 and VLSI 09 onlineAlexander Tomkins2009/08/04 11:30 AM
    smaller but better I_on/I_offMoritz2009/08/09 03:20 AM
      smaller but better I_on/I_offDavid Kanter2009/08/09 10:15 AM
  Article on 32nm process tech from IEDM 08 and VLSI 09 onlineWilliam Campbell2009/08/03 05:37 PM
  Article on 32nm process tech from IEDM 08 and VLSI 09 onlineBryan Catanzaro2009/08/04 07:04 AM
  Intel 32 nm PMOSjokerman2009/08/04 12:09 PM
    Intel 32 nm PMOSDavid Kanter2009/08/04 12:57 PM
    Good eyeDavid Kanter2009/08/05 01:30 AM
  TSMC 28nm: major changes?One Anonymous2009/08/05 01:31 AM
    TSMC 28nm: major changes?David Kanter2009/08/05 02:34 AM
      TSMC 28nm: major changes?anon2009/08/05 08:20 AM
        TSMC 28nm: major changes?David Kanter2009/08/05 12:04 PM
  Article on 32nm process tech from IEDM 08 and VLSI 09 onlineanon2009/08/07 01:48 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?