tarlinian (tarlinian.delete@this.gmail.com) on August 17, 2014 3:08 pm wrote:
> juanrga (nospam.delete@this.juanrga.com) on August 17, 2014 2:29 pm wrote:
> > tarlinian (tarlinian.delete@this.gmail.com) on August 16, 2014 1:56 pm wrote:
> > > juanrga (nospam.delete@this.juanrga.com) on August 16, 2014 12:01 pm wrote:
> > > > Doug S (foo.delete@this.bar.bar) on August 16, 2014 10:28 am wrote:
> > > > > Maynard Handley (name99.delete@this.name99.org) on August 16, 2014 9:56 am wrote:
> > > > > > I wouldn't trust anything in that www.electronicsweekly.com story. (It lost all credibility
> > > > > > when it went off on some strange tangent about providing Apple with an A57 variant.) BUT
> > > > > > I think you're being too optimistic about Intel's 14nm. Let's read the
> > > > > > Intel press release closely (ie like a lawyer). Here's what they say:
> > > > > > "
> > > > > > • The first systems based on the Intel® Core™ M processor will be on shelves for the
> > > > > > holiday selling season followed by broader OEM availability in the first half of 2015.
> > > > > > • Additional products based on the Broadwell microarchitecture and
> > > > > > 14nm process technology will be introduced in the coming months.
> > > > > > "
> > > > > >
> > > > > > So what will be available 1H15 is more Broadwell-Y parts --- ie the parts that may
> > > > > > be of interest to Surface Pro 4 and similar devices, but are probably not of interest
> > > > > > to even MacBook Air level devices, let alone larger laptops and desktops.
> > > > > > REAL Broadwell devices get the very specific due date of "coming months" which doesn't
> > > > > > indicate much confidence on the part of Intel as to when they might ship.
> > > > > >
> > > > > > Charlie at SemiAccurate has stated that dual-core parts will arrive in six months, and quad-core in eleven
> > > > > > months. You can have whatever opinion you like about his accuracy, but everything Intel has said, HOW
> > > > > > they have said it, and what they have NOT said, fits with such an (extremely delayed) schedule.
> > > > >
> > > > >
> > > > > While I don't disagree with any of that, foundries can call whatever they like "14nm" or "16nm" or "10nm".
> > > > > There is no UK advertising standards group that will fine
> > > > > them if they're stretching the truth. The process
> > > > > name being the same or similar doesn't mean they're necessarily comparable to Intel's 14nm.
> > > > >
> > > > > The whole idea of saying "because Intel promises 14nm on date x" and "these foundries have promised
> > > > > 16nm, 14nm and 10nm on these dates" tells you anything about who is ahead is ludicrous. Intel's
> > > > > 14nm has been delayed, and may be further delayed for higher power Broadwell parts - but maybe it
> > > > > was always planned to be skipped entirely. Maybe the rumor from a year or two ago that Intel would
> > > > > not sell socketed Broadwell parts was true, but not for the reasons that concerned people.
> > > > >
> > > > > The one thing I saw that suggested foundries weren't so far behind Intel after all was
> > > > > the density of Apple's A7 on Samsung's 28nm process, which was a lot closer to Intel's
> > > > > 22nm density than their 28nm density (though granted comparing a mobile ARM SoC versus
> > > > > a mobile x86 SoC based on transistors and die size is far from foolproof)
> > > > >
> > > > > But that doesn't mean that what Samsung calls "14nm" will be 4x as dense as what they call "28nm".
> > > > > Based on what I've read that does at least appear to be true for Intel. And power and performance
> > > > > are a whole other matter, as improving those will typically require sacrifices in density.
> > > >
> > > > Well what Intel decided to call 22nm is what rest of industry calls ~26nm. However, rest of industries
> > > > have decided to redefine the nanometer and play with Intel to the same silly marketing game.
> > > >
> > > > But as you say the node name is irrelevant, what is relevant are its parameters. According
> > > > to the ITRS, the key metric of progress is the M1 HP. Intel 14nm node has a M1 HP of 26nm.
> > > > TSMC 16nm has a M1 HP of 32nm. Thus Intel 14nm has an advantage of about 23% over TSMC 16nm.
> > > > That is about half node, which implies that Intel traditional lead has been reduced. It
> > > > is expected that the lead will vanish at the 10nm node, somewhere in 2016--2017.
> > >
> > > Come on, I've been reading your posts and giving you the benefit of the doubt on microarchitecture,
> > > but are you really going to play the silly node naming game as well? No one expects nodes to
> > > be comparable across competitors. (But I will note that Intel didn't decide to arbitrarily call
> > > 32/22 nm 28/20 nm. They freely admit that the same generation foundry processes have better
> > > density. Each node has scaled quite well from the previous one. 14/16 nm will have no effective
> > > shrink outside of what you can wring from having better performing devices.)
> > >
> >
> > Even former Intel people agrees on that Intel relabeled nodes and reinvented the nanometer for marketing
> > purposes. Hans (Chip architect) has explained this silly marketing game much better than I could do:
> >
> >

No, What TSMC/IBM/Samsung/GF said is:
> >
> > We call these processes 14nm / 16nm because:
> > Intel deviated from the official ITRS process naming rules and Intel
> > marketing/shareholders always abuse this against us to claim they
> > are many, many years ahead.
> >
> > Indeed, based on BEOL density Intel's processes should have been called
> > differently according to the official ITRS naming standards:
> >
> > Intel's 65nm should have been called 80nm instead.
> > Intel's 45nm should have been called 55nm instead.
> > Intel's 32nm should have been called 39nm instead.
> >
> > You can see this in the graph below:
> >
> > -The blue squares show the official metal 1 half pitches according to the ITRS.
> > -The purple circles and squares are the actual BEOL metal 1 half pitches of Intel's processes.
> >
> > You can see that going from 90nm to 65nm Intel improved the BEOL MT1
> > by only 5%. They did exactly the same then as TSMC/IBM/Samsung/GF
> > are doing now.
> >
> > It is due to the ignorance of the general public, which has no knowledge
> > of the official process naming rules, that TSMC/IBM/Samsung/GF are
> > forced to deviate from the official naming rules in the same way as Intel
> > did.

> >
> > > You're also ignoring the fact that no one has even announced ship dates for any 14/16 nm products. Until
> > > now we've seen maybe 1 HVM 20 nm parts, the Qualcomm modem. No one other than Intel has started HVM of any
> > > 14/16 nm parts. Forget even just one low power part like
> > > Intel. TSMC claims risk production of 16 nm started
> > > in late 2013. We aren't going to see any 16 nm parts in products until 2H 2015 at the earliest. That's a
> > > 2 year delta. And you think that 10 nm is just going to show up the next year. The foundries might have
> > > decided what their 10 nm process is going to look like by then, forget actually making anything.
> >
> > Broadwell-EP is expected for Q4 2015. Several ARM servers are expected for 2016, including AMD K12.
> > Nobody will be announcing ship dates this early, but the products are expected for that time.
>
> If you continue to believe that M1 half-pitch is all that matters when comparing process technologies,
> I don't suppose there's any point in discussing this with you. That might have been true 20 years
> ago when everyone's transistors were identical, wire delay was meaningless, leakage didn't matter
> and no one even bothered considering design rules, but it certainly isn't true now.

What I said is again different to what you interpret. Moreover, this is a slide from a talk given by Andrew B Kahng the past year (it is not 20 years ago)