Purana Archer (ancientarcher.delete@this.gmail.com) on October 4, 2013 6:38 am wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on October 2, 2013 10:45 am wrote:
> > > I was not talking about Samsung's total capex, just the capex on Logic which is expected to be around
> > > US$7-8bn. Samsung's total capex for 2013 is expected to come at US$23bn (far far higher than that
> > > Intel) includes its capex on DRAM, NAND, consumer electronics and displays. On just the logic capex,
> > > the point I was making was that this is the first time in 25 years, since the start of the microprocessor
> > > industry really, that a competitor in Logic is able to match Intel's capex. That has huge implications.
> > > Assuming the other guys ,like Samsung, TSMC and GloFo are not entirely stupid and don't literally
> > > piss away the money, you are going to see a narrowing of the tech lead that Intel enjoys today -
> > > which is the direct result of its outsized capex spend of previous years.
> >
> > I don't know what numbers you are using, but they sound wrong:
> > http://www.eetimes.com/document.asp?doc_id=1280675
> >
> > I don't care about samsung's spending on displays or CE. That's not relevant to logic.
> >
> > Also, capex != R&D.
> >
> sure, TSMC said that along with Alliance partners, it spends much more than Intel on R&D.
>
> "Unlike Intel and Samsung Electronics' vertical integration, TSMC is working closely with EDA, IP, software
> IP, systems software and design services providers – which the foundry has dubbed a "Grand Alliance."
> The TSMC-led alliance allocated a combined US$13.56 billion to R&D in 2012, the sources estimated."
> http://www.digitimes.com/news/a20131003PD215.html
>
> The Common Platform (Samsung, IBM and GloFo) also spend as much, I would guess.

None of that is related to actual process technology development. Moreover, if you've ever talked with Synopsys, you'd know that they do far more cutting edge work with Intel than anyone else because Intel is far ahead of everyone else on process.

In fact, Intel uses that to get advantageous pricing from tools vendors, because the vendors are effectively learning how to work with things like FinFETs.

Also, digitimes isn't a particularly reliable source, it's basically a mouthpiece for Taiwanese companies.

> > > Well, Intel introduced HKMG in 2007, Samsung in 2012 and
> > > TSMC in 2013. Finfet - Intel introduced in 2011 and
> > > TSMC and Samsung in 2014.
> >
> > When making these sorts of comparisons, you should be very careful. Intel starts shipping products
> > in massive volumes...typically the first production at TSMC is FPGAs, and it takes a while for
> > them to really reach the same level of maturity as Intel. Ditto for any foundry really.
> >
> > Samsung and TSMC have committed to early 2014 customer tape-outs. That means actual production in 2015.
> >
> > Just think about it for a minute. TSMC and Samsung aren't going to even ship any 20nm
> > parts in production this year, because their yields are terrible according to Digitimes.
> > Why would you believe that in ~1 year they can fix the 20nm problems, ramp FinFETs and
> > handle the yield issues there, and then ramp volume? That isn't plausible.
> >
> > In my mind, the foundries have a huge credibility problem about any claims they make because they
> > consistently fail to deliver, and many of their claims are based on non-comparable timelines.
> >
>
> you are right. 14/16nm finfet is going be done only in 2015.

So you believe we will see 16nm FinFET SoCs (not FPGAs) in high volume in 2015?

>Remember however that the foundries will be
> using the same front end as the 20nm scalar, so the 2 node jump in 2 years, doesn't >look too ambitious.

You are getting FEOL and BEOL confused. BEOL is metallization, FEOL is the devices. 16nm uses a 20nm metal stack, which is why the density doesn't improve.

Moreover, nobody in their right mind would ever consider TSMC 28nm --> 16nm to be two nodes, that's just BS marketing from TSMC. Just look at the improvement in terms of performance and density. The density of 16nm is one node better than 28nm...which was mentioned earlier in this thread.

> > >So, the gap is already narrowing.
> >
> > I will believe it when I see actual evidence.
> >
> > >My point is, if you don't assume the others are
> > > chipmunks, they will invest the capex spend relatively wisely,
> >
> > Like spending on gate-first HKMG? Or FD-SOI? etc. etc.
> >
> > >then there is no way that Intel can maintain
> > > that 3-4 years edge in tech when spending the same or less.
> > > It is mathematical really, the edge is a function
> > > of the cumulative capex and r&d spend. when the spend gap narrows, the tech difference >should also narrow.
> >
> > First of all, Intel is working under very different constraints than Samsung or TSMC. Foundries need
> > to support an ecosystem and are limited by that. Moving to FinFETs is hard because all your IP needs
> > to change. OTOH, Intel basically controls the IP development and can really co-optimize.
> >
> > I agree you have a valid point that as TSMC or Samsung becomes larger, they have
> > the potential to narrow the gap. But that doesn't seem to be happening.
> >
>
> The gap is narrowing.

You have no data to support his claim and moreover, there is no indication that the foundries are actually catching up to Intel...except from foundry marketing departments, who cannot be trusted.

>Just give some time for the r&d and capex dollars that the ARM camp has invested
> to flow through.

So why don't you tell me when you expect TSMC/Samsung to catch up with Intel on contacted gate pitch, transistor performance, etc.?

Moreover, you still fail to understand that 'the ARM camp' is a bunch of folks competing with each other and very few actually do much on process tech. When we talk about manufacturing, the only companies that are relevant are TSMC, Samsung, GF, and IBM. It doesn't matter how much money Rockchip or Mediatek makes.

This is exactly like the people who talk about "ARM vs. Intel" without realizing that those two companies are not really competitors (well perhaps with Quark they are, but they aren't competitors in any meaningful sense).

>After all, as I said, if you earn more, you can invest more and thereby stay ahead and
> earn more.

Except that the foundries have to 'catch up' before they can 'get ahead'. Also, some of them have yet to master that whole 'earn money' trick.

>Intel benefited from this virtuous cycle in x86, thereby crushing AMD. The unfortunate thing
> for Intel is that the ARM guys got money from somewhere else that was not dominated by >Intel.

Yet again...it doesn't matter how much money Apple makes on iPhones if they aren't spending anything on process technology development. Please stop making specious arguments.

> Sooner or
> later they will invest the money in capex and r&d to narrow the tech gap and maybe >overtake Intel.
>That
> will happen irrespective.

Tell me when. Do you think it's 2 years? 4 years? 7 years?

David