[Motley Fool] Apple dodged an Intel-shaped bullet

[NTMBK]

There used to be a lot of speculation that Apple (NASDAQ:AAPL) would one day move at least some of its A-series applications processor production to Intel (NASDAQ:INTC). Apple originally relied on Samsung (NASDAQOTH:SSNLF) to build its custom-designed chips, but has, in recent product generations, moved over to contract chip-manufacturing specialist Taiwan Semiconductor Manufacturing Company (NYSE:TSM).

IMAGE SOURCE: APPLE.

Indeed, TSMC manufactures Apple's current A10 Fusion chip and is expected to manufacture Apple's upcoming A11 Fusion chip, as well.

In hindsight, Apple seems to have dodged a rather serious bullet by not relying on Intel to build its chips. Here's why.

https://www.fool.com/investing/2017...a-serious-bullet.aspx?source=iaasitlnk0000003

Does Intel still have any serious foundry customers? LG's phone chips are completely MIA. Spreadtrum's chip is massively delayed and running an ISA nobody wants... oh, and they're moving to TSMC for 12nm. The flagship customer was Altera, who Intel then acquired (hence they aren't foundry customers any more).

 

[moinmoin]

Honestly wasn't aware Intel made any serious effort getting foundry customers outside of Altera. Spreadtrum is surprisingly recent as well considering Intel's now long running issues with still upcoming process nodes. Thanks for the links!

 

[Qwertilot]

I'm honestly finding the whole thing baffling.

Even with the companies they've purchased it seems to be taking them forever to get their stuff ported over to their own processes.

 

[arandomguy]

Intel comparisons to TSMC to some extent are more apples to oranges than apples to apples.

For Intel to commit to volume manufacturing of a new node it has to make economic sense and higher yield maturity. If 10nm yields poorly they don't have much incentive to ramp it at low yields just to push a 10nm product if can improve performance metrics via 14nm variants or just a plain bigger chip.

For TSMC to commit to volume manufacturing it only needs a customer in this case Apple. If the new node is economically worse Apple will still be willing to commit if they feel that can pass the cost down the chain, and there are already rumors of massive price increases for the new iphones.

Also from what it appears Apple tends to factor in longer outlooks and things beyond just immediate pricing/performance in what they actually do, see no further than the situation surrounding Qualcomm and Intel radios.

 

[raghu78]

Intel foundry is basically non existent and its only getting worse at 10nm.

https://semiaccurate.com/2017/09/06/intel-foundry-customer-bails/

 

[IntelUser2000]

Does Intel still have any serious foundry customers? LG's phone chips are completely MIA. Spreadtrum's chip is massively delayed and running an ISA nobody wants... oh, and they're moving to TSMC for 12nm. The flagship customer was Altera, who Intel then acquired (hence they aren't foundry customers any more).

I speculated its bad, but its even worse.

This ties in with the secrecy surrounding their products now. They refuse to reveal details that they did readily just 5 years ago. They keep putting out powerpoint presentations about how they have a process lead.

It's all for show. If the details were released then the whole world would know they are nowhere near ahead as they claim. Imagine how bad would it be for their reputation! The public already knows about years of delay for their process. Now if the details tell us that they don't even have the lead they claim to have(like in density). It would be the worst PR disaster in decades.

 

[IntelUser2000]

Ashraf adds fuel to the fire: https://www.fool.com/investing/2017...king-another-big.aspx?source=iaasitlnk0000003

From the article:

Intel says that its 14nm technology offers a density of about 37.5 million transistors per square millimeter.

TSMC's 10nm technology -- which is in production now and will be powering the tens of millions of new iPhones that'll ship this year alone -- seems to be capable of cramming 55 million transistors into a single millimeter.

Not to mention, there doesn't exist a single Intel product that meets the density claims of 37.5 million transistors per square mm on the 14nm process.

55 mil tr/mm2 is based on Anand's Huawei article.

 

[Phynaz]

I'm honestly finding the whole thing baffling.

Even with the companies they've purchased it seems to be taking them forever to get their stuff ported over to their own processes.

Basically because there's no such thing as porting a CPU design. You basically have to do about half the work over again to meet the design rules of new manufacturing process. So you are looking at two years minimum.

 

[R0H1T]

I speculated its bad, but its even worse.

This ties in with the secrecy surrounding their products now. They refuse to reveal details that they did readily just 5 years ago. They keep putting out powerpoint presentations about how they have a process lead.

It's all for show. If the details were released then the whole world would know they are nowhere near ahead as they claim. Imagine how bad would it be for their reputation! The public already knows about years of delay for their process. Now if the details tell us that they don't even have the lead they claim to have(like in density). It would be the worst PR disaster in decades.

Anyone taking PR slides on face value should need to get their hards checked, also it's pretty evident that Intel's foundry business has been slipping for quite a few years now. The delays, sometimes worse performance, than projected via PR slides, & of course deadlines slipping all point to the same thing. The extent of which can only be speculated at this point in time.

 

[Lodix]

I speculated its bad, but its even worse.

This ties in with the secrecy surrounding their products now. They refuse to reveal details that they did readily just 5 years ago. They keep putting out powerpoint presentations about how they have a process lead.

It's all for show. If the details were released then the whole world would know they are nowhere near ahead as they claim. Imagine how bad would it be for their reputation! The public already knows about years of delay for their process. Now if the details tell us that they don't even have the lead they claim to have(like in density). It would be the worst PR disaster in decades.

What process lead ? Even when they had a theoretical process lead other foundries products were more dense. Current 10nm from the competition is ahead in density and they have very aggressive roadmaps for the upcoming years.

 

[witeken]

Ashraf adds fuel to the fire: https://www.fool.com/investing/2017...king-another-big.aspx?source=iaasitlnk0000003

From the article:

Not to mention, there doesn't exist a single Intel product that meets the density claims of 37.5 million transistors per square mm on the 14nm process.

55 mil tr/mm2 is based on Anand's Huawei article.

FYI

Intel's SRAM has a density of 100 million transistors per square mm. SoCs contain a lot of SRAM.

 

[maddie]

FYI

Intel's SRAM has a density of 100 million transistors per square mm. SoCs contain a lot of SRAM.

Relevant how?

Simple present fact. Intel appears to have lost a process density advantage, irrespective of what it's named. The important point is what this means for the future.

 

[IntelUser2000]

FYI

Intel's SRAM has a density of 100 million transistors per square mm. SoCs contain a lot of SRAM.

I know. I even have a topic in the video cards section.

But I'm not sure how long we can excuse them for such corner case scenarios. I said myself that the lack of density is probably due to high performance circuits. But they have practically stopped advancing there.

On Foundry they have no one's business. They could have said: Gazillion mil tr/mm2 and it wouldn't have mattered. The SRAM is so small, so why are the sizes staying stagnant? What is the point of 100 mil tr/mm2?

 

[WhoBeDaPlaya]

Basically because there's no such thing as porting a CPU design. You basically have to do about half the work over again to meet the design rules of new manufacturing process. So you are looking at two years minimum.

If one could just automagically port RTL into a new process, they wouldn't pay me 6 figures a year
<-- Works in physical design

 

[R0H1T]

If one could just automagically port RTL into a new process, they wouldn't pay me 6 figures a year
<-- Work in physical design

You should ask for a raise

 

[WhoBeDaPlaya]

You should ask for a raise

But my job is already so fulfilling. Erection-inducing even!

 

[NTMBK]

I'm honestly finding the whole thing baffling.

Even with the companies they've purchased it seems to be taking them forever to get their stuff ported over to their own processes.

I can only presume that the 10nm delay has to be a factor. I'm sure Infineon would love to be able to offer a 10nm modem.

 

[moinmoin]

Intel comparisons to TSMC to some extent are more apples to oranges than apples to apples.

For Intel to commit to volume manufacturing of a new node it has to make economic sense and higher yield maturity. If 10nm yields poorly they don't have much incentive to ramp it at low yields just to push a 10nm product if can improve performance metrics via 14nm variants or just a plain bigger chip.

For TSMC to commit to volume manufacturing it only needs a customer in this case Apple. If the new node is economically worse Apple will still be willing to commit if they feel that can pass the cost down the chain, and there are already rumors of massive price increases for the new iphones.

Also from what it appears Apple tends to factor in longer outlooks and things beyond just immediate pricing/performance in what they actually do, see no further than the situation surrounding Qualcomm and Intel radios.

All you did is explain (very well) how Intel managed to become so complacent in their core area of expertise. They have no serious foundry customers so no pressure from that direction. And internally their process node was good enough™ for far too long to put more effort into improving yield maturity more timely. Now they have no foundry customers and competition in their core market is suddenly heating up. Bad timing all around.

 

[Arachnotronic]

You should ask for a raise

6 figures a year could mean anywhere from $100,000 to $999,999.

 

[R0H1T]

6 figures a year could mean anywhere from $100,000 to $999,999.

In which case he should aim for 7.

 

[oak8292]

Very humble opinion. I think it is a lot more complex than Ashraf or many here are making it. Altera is a pretty low volume customer and they can be fit into Intel's manufacturing schedule pretty easily. Apple is at volumes where they need most of the wafers from a fab. TSMC is almost their only option. I think they were 'stuck' contractually at Samsung probably longer than they wanted as the iPhone ramped faster than expected. Samsung Austin was built about the time that Apple wanted production and Apple probably gave somewhere around a four year commitment to wafers, e.g. AMD wafer agreement for the fab construction. However, Apple essentially outgrew Austin. More than likely they gave TSMC a commitment for a number of years. TSMC started the 16/20nm node with 60K wafer starts per month for Qualcomm and Apple and grew to 100K wafer starts per month on the third year that Apple was using them on that node while also pushing Qualcomm over to Samsung. At Intel the equipment has a five year depreciation and you can't have a 40K wafer start per month customer come in for a single year. If Apple moves to Intel it will be for a number of years. Apple has moved to the TSMC 7/10nm node and I will guess that they have at least another two years at TSMC locked in.

Intel's process advantage is immaterial if it is unavailable. Intel needs a sizable level of second tier customers to utilize capacity after they are done with a node so they can move on. They appear to be more serious about establishing foundry services giving both ARM and the EDA vendors earlier access to a node, however TSMC has an established customer advantage just like Intel has an x86 software advantage. Their is a lot of inertia and energy required to move. Apple at 200+ million die has the most money to make a move but they can also destroy capital allocation if they jump between foundries.

 

[FIVR]

Their "process advantage" is more or less theoretical at this point. They ship products that are less dense, perform the same or worse, use more power, and produce more heat than the competition.


I guess it is a testament to the skill of their marketing team that people still believe they have a process advantage.

 

[WhoBeDaPlaya]

In which case he should aim for 7.

While I would love to have that (and screw supermodels every day of the week), I have to live within the sad reality of this world

 

[ksec]

One of the reason, if not the sole reason why Intel wanted to do Custom Fabs were to fill up their capacity. But they now have a new chips order to fill the capacity they will no longer need to work on their Custom Fabs. ( Short Sighted perhaps )

And the new chip is the Intel 14nm XMM 7560 Modem for Apple.

There is an Interplay between all these, Intel could have moved to 10nm, but then they will not have time to recoup the cost of 14nm investment. In a ideal scenario, their flagship could stay in 10nm leading node while their custom fab could work with 3rd parties on 14nm.

Given how Intel has modeled the PC industry continuing down trend, they will sell less chip over time, and Custom Fab was one way to help.

Intel badly wanted to be in 5G, ( they were late or wrong on 4G ), so the baseband modem on iPhone makes the perfect partnership. 200M+ unit per year, while the die size being much smaller, it is roughly the same annual unit shipped in the whole PC industry.

With these kind of volume, and Intel likely playing into 5G chipset as well, they will no longer Custom Fabs, for now.

Once the baseband ship with next iPhone, ( not the coming one on tuesday ). Intel will have moved out of its capacity deadlock.

So I dont think 10nm was delayed or what, the whole plan was to fill the fabs as cost efficiently as possible. What changes the matter was AMD suddenly became a decent competitor, and all the flaws from Intel are now being challenged.

 

[Phynaz]

I'm honestly finding the whole thing baffling.

Even with the companies they've purchased it seems to be taking them forever to get their stuff ported over to their own processes.

Here ya go...

https://newsroom.intel.com/newsroom...con-mesa-intel-tmg-china-event-fact-sheet.pdf