Intel team of 1,000 people working on iPhone modem, foundry SoC

[stingerman]

Very true. If they pick Intel for baseband and then foundry at 10nm there is no going back to TSM and Samsung at which point Intel will surely pitch x86 mobile and the benefits of a unified code for the entire Apple ecosystem. I know unified software sounds terrible to Apple fans but it does make writing software and maintenance a lot simpler. Imagine being able to run your apps on the Desktop or vice versa terrible I know.

Apple avoids single sourcing. They went as far to implement two different A9 designs to use both IBM's 14 NM tech at Samsung and TSMC hybrid 16NM process. By the way, that means when Global Foundries 14NM comes online, Apple can use their IBM 14NM based process as well, since it is similar to Samsung.

IMO, Intel will take care of Intel's needs rather than becoming a low margin manufacturer to Apple's designs. Intel hasn't delivered in quantity a performant level processor at such low voltages as smartphones limit. Notice that the iPad Pro claims to be faster than 80% of the processors shipped amongst the volume of Notebook computers.

 

[ShintaiDK]

Apples dual sourcing with the A9 wasn't because they want dual sourcing. Its because the foundries cant deliver more than that. TSMC wasn't even to make the A9 originally.

 

[kpkp]

Apples dual sourcing with the A9 wasn't because they want dual sourcing. Its because the foundries cant deliver more than that. TSMC wasn't even to make the A9 originally.

Regardless, it's in Apples interest to not be stuck with one foundry.

Look at what Intel was able to do with 22nm moorefield on the zenfone2

Nothing all that impressive, cost aside as we do not know at what prices ASUS got those.

 

[Arachnotronic]

Have you confirmed that it's FinFet and not FinFet+ somewhere? All indicators are that they are on TSMC's 2nd generation.

It's FinFET+.

 

[stingerman]

Apples dual sourcing with the A9 wasn't because they want dual sourcing. Its because the foundries cant deliver more than that. TSMC wasn't even to make the A9 originally.

That's overly simplistic. Multiple factors underly their strategy. Not least of which is flexibility and risk management. They're not going to repeat the PowerPC strategy.

Furthermore, Apple prefers to capture their outsourced manufacturing. No way Intel will allow submit to such arduous terms. And no way will Apple surrender their future by being beholden to one manufacturer for both their x86 and ARM needs. Very low chance Intel will manufacture processors for Apple and be beholden to Apple to give them orders year after year, competing with TSMC, Samsung and Global Foundries.

As attractive as an Apple sized order may seem, it would be long-term suicide for Intel to put their name on an Apple processor.

 

[Systems analyst]

Intel are now an ARM partner. Who knows? - they may even develop a custom ARM processor.

 

[ShintaiDK]

Intel are now an ARM partner. Who knows? - they may even develop a custom ARM processor.

They have been that for ages. And Intel for example developed Xscale using ARM.

 

[readers]

Really bad news for just about everyone else...

 

[witeken]

Caught up? That's laughable they haven't even launched the node yet and you are claiming they caught up not to mention Intel is not standing still either.

http://i.imgur.com/bDlN7Lm.png
.

Nice find .

 

[Systems analyst]

They have been that for ages. And Intel for example developed Xscale using ARM.

XSCALE is ancient history. Intel bought Altera, who had an ARM based product being produced in Intel's fabs. They also bought AXXIA, who also have an ARM based product. But there may be something new going on, in addition to these.

 

[ShintaiDK]

The custom foundry got its own portal as well:
http://www.intel.com/content/www/us/en/foundry/overview.html

Business is open for anyone willing to pay.

The introduction from 2014 in case anyone forgot:
http://www.semiwiki.com/forum/files/2Sunit Rikhi_Intel.pdf

 

[Nothingness]

Altera was willing to pay, still nothing to see, while Xilinx is shipping test chips in 16nm. Guess to benefit really from Intel foundry they had to be bought by Intel :biggrin:

 

[ShintaiDK]

Altera was willing to pay, still nothing to see, while Xilinx is shipping test chips in 16nm. Guess to benefit really from Intel foundry they had to be bought by Intel :biggrin:

You dont think Altera is shipping test chips? And by "16nm" you do mean 20nm FF right?

https://www.youtube.com/watch?v=dwQnWFvwv4Q

 

[NTMBK]

Look at what Intel was able to do with 22nm moorefield on the zenfone2. Great performance good battery life in a $300 phone.

If Moorefield were a genuinely competitive offering, we would have seen it in more phones. The only manufacturers who used it were PC OEMs who have a long friendly relationship with Intel (like Dell and Asus)- Samsung, HTC, Motorola, Lenovo, Sony, they've all ignored Moorefield and gone for a variety of Qualcomm and Mediatek parts.

 

[Dresdenboy]

Semiwiki isnt exactly a neutral site to put it gently. So I dont give anything for it and their bias. Semiwiki and TSMC is pretty much the same.

You're right:

Companies that currently subscribe to SemiWiki or have a professional connection with a SemiWiki Blogger include:

Aldec, Atmel, Andes Technology, ANSYS, ARM, Arteris, Cadence, CASPA, CEVA, ClioSoft, CLKDA, Concept Engineering, Coventor, Design Automation Conference, Dassault Systemes, eSilicon, Fractal Technologies, GlobalFoundries, IC Knowledge, ICScape, Intel Custom Foundry, Kilopass, Mentor Graphics, Methodics, MunEDA, NetSpeed, S2C, Samsung Foundry, Semiconductor Intelligence, SEMI, Sidense, Silvaco, SmartDV, Solido DA, Sonics, STMicro, Synopsys, TSMC, and Xilinx.

 

[ShintaiDK]

You're right:

Nice try, but you know better. Daniel Nenni is TSMC gospel. Even a SA Admin says Danni Nenni is a paid PR guy

 

[ShintaiDK]

You should provide source for that data (which looks correct).

https://www.semiwiki.com/forum/content/3884-who-will-lead-10nm.html

 

[dealcorn]

If they pick Intel for baseband and then foundry at 10nm there is no going back to TSM and Samsung at which point Intel will surely pitch x86 mobile and the benefits of a unified code for the entire Apple ecosystem.

Your comment sounds good, but I think it is not in Intel's best interests to kill ARM at Apple. From a strategic perspective I suspect Intel's highest, existential priorities are to (1) shut down advanced process nodes at competing fabs before they become a viable competitive threat (e.g., to DCG), and (2) not run afoul of anti trust authorities. Intel has multiple promising technologies and if most or all work, they increase Intel's profile to anti trust authorities. If Intel fabs ARM for Apple, the SoC market looks more competitive and Intel lowers it's antitrust profile. As long as Intel receives a reasonable foundry margin from Apple, the benefits of a competitive SoC market may be greater than the additional SoC margin Intel would make from Apple upon conversion. If SoFIA and Broxton are reasonably well received in the high volume low end and middle markets, it becomes important that Intel maintain competition in the high end SoC market. The value of advanced process node domination is far greater than high end mobile market share.

 

[Nothingness]

You dont think Altera is shipping test chips? And by "16nm" you do mean 20nm FF right?

No I mean 16nm, UltraScale+, the 20nm UltraScale has been available for months.

https://www.youtube.com/watch?v=dwQnWFvwv4Q

Ha nice! I had missed that information, which I couldn't find on Altera site, thanks

Looks like finally Altera and Xilinx are close. Now let's get back on topic...

 

[Khato]

Have you confirmed that it's FinFet and not FinFet+ somewhere? All indicators are that they are on TSMC's 2nd generation.

How exactly can you call it 2nd generation FinFET when the "1st generation" never made it out of the fab? By that logic, who knows what generation of FinFET we're getting from Intel?

I'm looking at top level performance at a mobile level voltage. This is not speaking to what happens when you up the voltage, if the material can handle it, and put a better cooling solution on it.

Your point being? Intel's had the exact same manner of "x% increased frequency at same power" claims as TSMC, but we don't have processors that are that much faster. It's always the low level which sees the gain. So sure, you'll see what was a 1.2 GHz chip jump up to 1.6 GHz at same power, but that high performance 2.3 GHz chip might hit 2.5 GHz if they're lucky?

The low end is what they all love making those claims with respect to for two reasons. First, they actually get the FinFET voltage scaling on their side. Second, the lower frequencies don't use up all of the transistor gains on increased parasitics due to the tighter pitch on interconnects.

 

[witeken]

So here's a crazy hypothesis.

Maybe Intel delayed Broxton and SoFIA in favor of putting 1000 people on putting out a 14nm modem for Apple.

It seemed a bit unlikely to me that Intel would release both XMM 7630 and the FinFET successor in 2016, but given that SoFIA 14nm is being launched in H2'16, it might be possible.

 

[Arachnotronic]

So here's a crazy hypothesis.

Maybe Intel delayed Broxton and SoFIA in favor of putting 1000 people on putting out a 14nm modem for Apple.

Doubt it. 14nm modem won't launch until late 2016 and probably won't start showing up in customer designs until 2017.

It seemed a bit unlikely to me that Intel would release both XMM 7630 and the FinFET successor in 2016, but given that SoFIA 14nm is being launched in H2'16, it might be possible.

SoFIA 14nm is being launched in "H2'16" but I wonder what Intel's definition of "launch" and "H2'16" is. Don't expect commercially available devices using 14nm SoFIA until 1H 2017 at the earliest.

 

[Thala]

So here's a crazy hypothesis.

Maybe Intel delayed Broxton and SoFIA in favor of putting 1000 people on putting out a 14nm modem for Apple.

It seemed a bit unlikely to me that Intel would release both XMM 7630 and the FinFET successor in 2016, but given that SoFIA 14nm is being launched in H2'16, it might be possible.

Yes thats crazy. You have a much too unrealistic view on the situation

 

[witeken]

Yes thats crazy. You have a much too unrealistic view on the situation

1,000 people is not nothing, the Infineon acquisition was years ago, and I'm sure Intel would rather offer Apples its latest and greatest 14nm modem than a 28nm outsourced one. The timing isn't too unrealistic, so we'll see. It was only an idea.

 

[ShintaiDK]

Seems LG is joining Intel as a foundry customer for their SoCs.

http://english.hankyung.com/news/apps/news.view?popup=0&nid=0&c1=03&newscate=1&nkey=201510210644401