Samsung and GLOBALFOUNDRIES Forge Strategic Collaboration to Deliver 14nm FinFET

[Khato]

and the very detailed report from BNP Parisbas that he linked - that I'm not done with

Not done with it either, and right now I'm not certain if I'll manage as I can't stop laughing about the process technology roadmap on page 10 - in the best kept secret ever, apparently Global Foundries has had a 20nm process available for the last year and a half! (Note that said roadmap also claims that GF introduced 28nm half a year before TSMC.) Regardless, I'll try to continue reading for other such gems of amusement, but sadly for the mot part it's nothing more than regurgitated information and educated guesses.

 

[mavere]

I recall GloFo getting 20nm tapeouts for ARM designs around the same time TSMC did. Both of those dates were way before the dates shown in BNP's roadmap, so it seems like the graphic used some criteria separate from both tapeouts and production ramp-up.

While I don't give BNP Paribas the benefit of all doubts, I think they'd have enough interns and financial monkeys to get old dates right. Of course, the (undetailed) criteria is suspect.

 

[witeken]

Ok, so assuming Apple makes Samsung/GF push out 14 nm in 2015, when can we expect AMD to get access to 14 nm? Early 2016?

That would be for production, I suppose, not product availability.

 

[Fjodor2001]

That would be for production, I suppose, not product availability.

Ok, well I meant product availability (sorry that my question wasn't so clear on that). So when do you think that'll be? And why do you think AMD should be so much later to the market with 14 nm compared to Apple? Both are using Samsung/GF 14 nm...

 

[Qwertilot]

Apple have both gargantuan volume demands and absolutely absurd quantities of cash to deploy when needed

Like with 20nm they'll likely book up (nearly) all the early capacity for a good chunk of time.

 

[witeken]

I expect H2'16. Why? Because I don't see signs that there will be a massive amount of FF production in (early) 2015 and because companies supposedly want some return on 20nm investment, and AMD doesn't seem to be fast with 20nm either. Fourthly because the price/transistor of FinFET is higher than 28nm.

 

[AtenRa]

Fourthly because the price/transistor of FinFET is higher than 28nm.

You mean over 20nm.

 

[witeken]

You mean over 20nm.

No.

 

[AtenRa]

No.

20nm has 1.9x density over 28nm, in order for the 20nm transistor/$ to have the same value as 28nm it means that 20nm is 2x more expensive than 28nm which is incorrect.

 

[ShintaiDK]

20nm has 1.9x density over 28nm, in order for the 20nm transistor/$ to have the same value as 28nm it means that 20nm is 2x more expensive than 28nm which is incorrect.

I dont see you supply the documentation on that.

And its known that more and more companies either complain about transistor price or says 28nm is their last node due to cost.

nVidia called it a few years ago. And it turned out to be even worse than their projection:

The question is what companies can afford a lower node. For 16/14nm FF aka 20nm with FF only seems to have Qualcomm and Apple on the train now.

While 20nm only got Apple, Qualcomm and MediaTek on the current list of confirmed mass consumer products for the fabless companies.

 

[witeken]

So ARM is wrong, too?

Edit:

And Nvidia.

 

[Ajay]

I recall GloFo getting 20nm tapeouts for ARM designs around the same time TSMC did. Both of those dates were way before the dates shown in BNP's roadmap, so it seems like the graphic used some criteria separate from both tapeouts and production ramp-up.

That and Arm A7 core test shuttles don't really mean that much - it's just part of the PR battle.

 

[Ajay]

Not done with it either, and right now I'm not certain if I'll manage as I can't stop laughing about the process technology roadmap on page 10 - in the best kept secret ever, apparently Global Foundries has had a 20nm process available for the last year and a half! (Note that said roadmap also claims that GF introduced 28nm half a year before TSMC.) Regardless, I'll try to continue reading for other such gems of amusement, but sadly for the mot part it's nothing more than regurgitated information and educated guesses.

Yeah, seeing all the technical data in there gave me a false sense of hope

 

[MisterMac]

Very Interesting Article from Handel.


It's kinda all coming together now ain't it ?
Even in technology you can't really stop Old Money.

Sure we have new companies sprout up - and disrupt but if Old Money wants to muscle in...it will.

There's nothing real to indicate it - but wouldn't it be funny if TSMC started slowly loosing to the pockets of the koreans while GF\Mubadala just accepts and tags along with hopes of becoming 2nd?

 

[AtenRa]

So he says that 28nm wafer cost is at $2600, 20nm wafer cost is at $3200 (23% higher) and 16nm wafer cost is $4000 ( ~54% higher than 28nm).

With 1.9x higher density over 28nm, transistor/$ is lower at 20nm.

To have a little example, lets assume you have 1000 transistors at 28nm per wafer. Then each transistor costs $2.6

At 20nm you have 1900 transistors per wafer, then each transistor costs $1,68

At 16/14nm you have ~2185 transistors (+15% density), then each transistor costs $1,83

Hes words not mine

Handel also sees uncertainly in the use of FinFET devices due to higher wafer cost. “We see quite a few new designs. The problem again is the cost per wafer. For 28nm, we have about $2600 and for 20nm we have about $3200 and for 16/14 we have about $4000.

 

[witeken]

Don't forget yields.

 

[ShintaiDK]

Current 20nm wafers is certainly not 3200$. Its 2-3x higher due to Apple. And 28nm wafers are actually cheaper.

In projected growth graphs, 28nm will be the highest revenue for a forseeable future. Foundries are simply losing customers fast that can go 20nm and below due to cost. Also why we see more and more new 28nm products to replace 28nm products, including GPUs.

 

[AtenRa]

Don't forget yields.

What about yields ??

 

[witeken]

Here's the math behind the graph:

 

[AtenRa]

Current 20nm wafers is certainly not 3200$. Its 2-3-x higher. And 28nm wafers are actually cheaper.

Then according to your sources ( any links about wafer costs will be appreciated) what he says is wrong.

 

[witeken]

What about yields ??

Lower yields = higher price/transistor.

 

[AtenRa]

Here's the math behind the graph:

Any higher size than that ?? i cant see the numbers

 

[AtenRa]

Lower yields = higher price/transistor.

Obviously we are assuming the same yields, otherwise 40nm was cheaper than 28nm 3 years ago.

 

[ShintaiDK]

Then according to your sources ( any links about wafer costs will be appreciated) what he says is wrong.

Well you did skip a few things when making your conclusion:

Handel said the gate utilization is an issue because of limitations of the design tools and parasitic effects. “The other factor is parametric yields, which are strictly tied into leakage control for the 20nm and of course for the 16nm FinFETs,” he said. “You can break this. Intel has shown that it can be broken and of course that’s an excellent achievement. But, it’s based on very high design costs, potentially $1 billion per design, so you need $10 billion in revenue. It also takes a number of years,” he said. He noted that, in the smartphone market, designs move very fast. “You can’t make that kind of investments in designs.”

This is essentially why companies like Broadcom, nVidia and AMD cant afford it.

Also when you look on the growth chart, its quite obvious that 28nm is where the game ends for most companies. And that newer nodes simply shrink in revenue from there on. One has to wonder if companies like Samsung and TSMC can even afford below 10nm without ending with a negative ROI.

 

[AtenRa]

Well you did skip a few things when making your conclusionmpanies like Broadcom, nVidia and AMD cant afford it.

You said wafer cost is 2x 3x more than what HE said, i dont see the links to support that.