Replies to post #136521 on Intel Corp (INTC)

[mas]

For mad Nenni that was almost a sane post ;-). I have never bought the idea that Apple would dual-source its A chips. It is too much trouble to try and fit different processes just for one year's iteration. You would expect TSMC to get A9 as 16FF is an incremental step from their 20nm. Coincidentally I think Intel has a good chance of getting the successor A10 with its 10nm as that will be coming a year later than 16FF while the foundries twiddle their thumbs in 2016 just rolling out 14/16FF. Intel's 10nm will allow Apple to jump to quad-core and still improve single-thread performance over 16F.

Broxton (or whatever it's called now) is not cancelled which is a foolish thing to say as it is the next-gen Android performance successor to Merrifield but then again he could not be relied upon to make a completely sane post.

[Unkwn]

....this guy says Intel to equip mothballed fab in 2015 for 14 nm.
....also says the TSMC 16FF+ process, which Morris Chang said would be delayed until 2H 2015, has been pulled into 1H 2015
....I could see the first one since it will take the better part of a year to reach production.
.....claims TSMC back in the running to fab Apple A9s next year......adds some wild speculation about Broxton

Yep. That would basically be the first process generation that has been announced to be delayed and then successfully brought out earlier than predicted. But hey, TSMC can do the unexpected. I still doubt 14nm FinFet by any foundry, other than Intel, by next year and I even have high doubts for TSMC bringing 16nm FinFets then. We'll see.

Core M is filling one additional 14nm fab from Intel? Doubt it. It either has to do with very good PC processor sales or some high volume foundry customer is on its way - think Apple. Maybe due to noone else bringing Finfets by end of next year?

Brian Krzanich didn't mention Broxton lately and that's why Intel seems to be cancelling its whole mobile strategy? Am, yeah, that's what is called journalism these days I guess.

[wbmw]

....also says the TSMC 16FF+ process, which Morris Chang said would be delayed until 2H 2015, has been pulled into 1H 2015

He may be misquoting. This is the reference "out of Taiwan" that he is likely referring to.

http://www.digitimes.com/news/a20140825PB201.html

It mentions 16nm, but not 16FF+. According to this blog post, 16FF+ is a 2nd generation FinFET process.

http://www.cadence.com/Community/blogs/ii/archive/2014/04/28/2014-tsmc-technology-symposium-full-speed-ahead-for-16nm-finfet-plus-10nm-and-7nm.aspx

No doubt TSMC is now very aggressive with their process roadmap. Here's what I've been able to mine online.

 
 20nm: Claimed vs. 28nm: 1.2x perf, 1.3x power redux, 1.9x density 
    - Risk production Q2'13, Volume production Q1'14, Qualcomm modem Q2'14, Apple A8 Q3'14 
   
 16nmFF: Claims 1.4x perf and ISO-area vs. 20nm 
    - Risk production Q4'13, Expected volume production Q1'15, Expected Apple A9 Q3'15...? 
   
 16nmFF+: 1.6x perf and 0.85 die size reduction vs. 20nm 
    - Design kits ready April, with additional roll-out in May, July.  No specifics on risk or volume production 
    - (my comment) Can't see this being all that popular, unless it refreshes a few designs in H1'16, or if 10nm is delayed 
   
 10nm: Claimed vs. 16nmFF+: 1.25x perf, 1.45x power redux, 2.2x density 
    - Vs. 28nm: ~2x perf, 0.5x power, ~5x density 
    - Average CAGR per node (out of 3 nodes): 1.25x perf, 0.8x power, 1.7x density 
    - Product tapeouts and risk production end of 2015, goal of customer product launches in 2016 
    
 7nm: No data on benefits 
    - Targeting mid-2017 risk production, 6 months earlier than 2-year cycle 

What this says is that TSMC plans to provide 3.5 process nodes in 3 years (20/16/16+/10 in '14/'15/'16), and 4.5 process nodes in 5 years (add 7nm and '18).

This is a roadmap that ought to catch up to Intel's cadence. So it should be as easy as a walk in the sunshine for them.