How TSMC Won Back Exclusivity With Apple for the A10 Chip in iPhone 7

[SMIDG3T]

Not fair that we have to wait 13 months for the next Apple phone. Can we sue anyone?

I can sue you for being, well, I guess stupid. iPhone 7 will be a huge success even if it does "look the same".

 

[Gudi]

I disagree. A better chip could mean better battery life for iPhone and better performance for iPad. A9X is still a bit inferior to Core M brodwell according to anandtech, so IMHO ...

Core M isn't the benchmark, Intel chips are good enough for most of a decade now. And battery life improvements are eaten up with bigger high-density screens and thinner lighter devices. Or in case of the iPhone Apple adds a Taptic Engine, which steals space from the battery and battery life stays the same even with more efficient CPUs. So if you're able to live without 3D Touch there is little reason to upgrade.

 

[CmdrLaForge]

Thanks to the Macrumors team for a really good technical article !

 

[lchlch]

I disagree. A better chip could mean better battery life for iPhone and better performance for iPad. A9x is still a bit inferior to core m brodwell according to anandtech, so IMHO improvement is needed for the iPad to level up (or for use in mac... I wonder what can be done with a 4 core A10x)

In terms of CPU performance the a9x is inferior to the core m series processors. However in gpu tasks the a9x has about same power as a Xbox 360. It's pretty much a tie. So it depends on what you want, better cpu or better gpu?

In regards to TSMC 16nm and samsung 14nm, from the sources i read the TSMC is the better one. This is because the TSMC chips has lower leakages that leads to better power efficiency and thermals. In addition, the Samsung part had to use increased voltages to hit the desired clock speed which resulted even worse efficiency.
[source: http://www.eetimes.com/author.asp?section_id=36&doc_id=1328041 ]

 

[S.B.G]

Interesting read. I'm quite amazed at how much technology is packaged in such minute things and small devices like iPhones. The amount of things it can do with such a small size is simply amazing.

 

[TallManNY]

Okay Apple, put this in a new Apple TV and get some real games in your App Store.

Yes, I know that is not going to happen. We probably won't even see an A9x Apple TV this Fall.

 

[cdmoore74]

This chip is going to scream performance.

Yes but my iPhone 6S was never slow. Actually that's the best feature. I now want a total redesign to go with the speed. I don't think it's much to ask for after 2 years.
On a motorcycle there's not much difference going 150mph vs 200mph. Fast is fast. With this cpu update we will be lucky to a second or two improvement in most apps.

 

[SMIDG3T]

Yes but my iPhone 6S was never slow. Actually that's the best feature. I now want a total redesign to go with the speed. I don't think it's much to ask for after 2 years.
On a motorcycle there's not much difference going 150mph vs 200mph. Fast is fast. With this cpu update we will be lucky to a second or two improvement in most apps.

You don't miss what you don't have. Who knows if the 6s is fast compared to the 7?

The chip will also improve efficiency by a lot, with the larger batteries we could see 60-120 extra minutes of battery life. Could.

 

[Piggie]

In terms of CPU performance the a9x is inferior to the core m series processors. However in gpu tasks the a9x has about same power as a Xbox 360. It's pretty much a tie. So it depends on what you want, better cpu or better gpu?
]

I keep hearing such claims, but I've yet to see a load of new amazing 3D games for these mobile devices, that are as good as titles on an XBox360

We have painfully embarrassing titles like crossy road and such like.

If it's genuinely as good as a 360, why are people not hooking them up to TV's and playing loads of high end games via their Apple phones/tablets ?

 

[Neodym]

Great gif!

Tsmc really needs to update their logo.

Man TSMC really need to update that logo.

Why should they? No need for posing or style updates when you're not dealing with end customers. The money for a professional redesign and all the implicit consequences (necessary brand transfer activities, update of presentation headers, website etc.) is better invested in R&D and improved products.

 

[Zaft]

Great gif!




Why should they? No need for posing or style updates when you're not dealing with end customers. The money for a professional redesign and all the implicit consequences (necessary brand transfer activities, update of presentation headers, website etc.) is better invested in R&D and improved products.

You're probably right but that logo just makes me feel like im in 1982 or something.
[doublepost=1470919262][/doublepost]

I keep hearing such claims, but I've yet to see a load of new amazing 3D games for these mobile devices, that are as good as titles on an XBox360

We have painfully embarrassing titles like crossy road and such like.

If it's genuinely as good as a 360, why are people not hooking them up to TV's and playing loads of high end games via their Apple phones/tablets ?

Making a game like on a traditional console can be expensive and take time, developers just dont care when they can make money of simpler games and charge a couple bucks as apposed to $40-$60.

That whole thing a couple years ago about the death of consoles never came to fruition. Consoles are here to stay for real games. I love me some Rocket League on my xbox one.

 

[lchlch]

I keep hearing such claims, but I've yet to see a load of new amazing 3D games for these mobile devices, that are as good as titles on an XBox360

We have painfully embarrassing titles like crossy road and such like.

If it's genuinely as good as a 360, why are people not hooking them up to TV's and playing loads of high end games via their Apple phones/tablets ?

The hardware is definitely there.

The issue I believe is two fold.

1) the App Store model encourages a downward pressure in price. Making it hard for developer to recoup their investment. This problem is further exaggerated by the huge budget required for AAA games.

2) failure of the App Store to allow controller only games. Some games just don't work that well with touch and gyro.

I sincerely believe that Apple should either make AAA games themselves for the iPad or pay someone else to port their game to the iPad. For example I would love to run civilization vi on the iPad.

 

[Crzyrio]

The truth is, A-series chips are good enough already. If you own an A8/A9 device, there is little reason to upgrade.

That is like saying a C2D is adequate and no reason to upgrade. For the average consumer that might be true but software is starting to catch up to hardware very quickly. I can easily occupy 50% of the CPU for real time sound analysis. We also have deep learning on our phones now which can always benefit from processing power.

 

[Gudi]

That is like saying a C2D is adequate and no reason to upgrade.

Coincidentally I haven't bought a new notebook since the "Penryn" C2D MBP.

 

[spacemnspiff]

I can foresee that Samsung would consider buying TSMC for the business that it lost.

 

[Crzyrio]

Coincidentally I haven't bought a new notebook since the "Penryn" C2D MBP.

Yes and like I mentioned for most people that will be enough.

But there is a fair amount of the population that very easily can run their i7 upto 100%

 

[truthertech]

Steve Jobs would've never attached the silicon die directly to the PCB to minimize height and reduce the lengths of interconnects between components.

Good one. These type of articles really tie the trollers into knots because they struggle to be able to bash Apple over the news.

 

[9081094]

Did Samsung stop making Qualcomm's high-end SOCs?

Samsung Semi's 2Q sales are up 6% (YoY) and 8% (QoQ), so despite Apple's recent dramatic decline, the company is doing well -- a sign that the division doesn't depend on Apple sales.

It would be far worse for Apple if Samsung dicided not to make any parts for Apple. Because if you like it or not... iPhone is almost a rebatched Samsung.

 

[grkm3]

Got any data to back that up? If you're talking about Samsung's 14LPE (early) being first gen and 14LPP (production) being second gen, I believe the A9 was already on the 14LPP process, and real world showed that was slightly inferior in performance to chips manufactured on TSMC N16ff+ (which was *their* 2nd gen of 16nm)

since in your own words say you believe it was on 14lpp can you back that up with a source that it was?

 

[iBluetooth]

Yes but my iPhone 6S was never slow. Actually that's the best feature. I now want a total redesign to go with the speed. I don't think it's much to ask for after 2 years.
On a motorcycle there's not much difference going 150mph vs 200mph. Fast is fast. With this cpu update we will be lucky to a second or two improvement in most apps.

You get a total redesign of the motorcycle every two years?

 

[astroboy888]

So this does have a number of cool advantages, but I have to wonder if it's really going to give them much of a leg up on Samsung. The Galaxy S7 uses a 14nm process CPU already, whereas this is a 16nm process. Still has other advantages, but I don't expect the improvements to be that significant.

TSMC's 16nm and Samsung 14nm are consider the same generation process. Now almost 2 years into mass production we find that there is no advantage with 14nm and Samsung's transistors suffered from higher power, which means it drains battery faster while giving no appreciable difference in performance. Worse of all, in early 2015 Samsung's 14nm process yield is considerable lower, which means it is much more expensive to produce. These were well documented comparing two Apple A9 version of the chip in the infamous "chipgate" incident last year. These are all factors which prompted Apple to go exclusive with TSMC for A10 and A11.
[doublepost=1470956623][/doublepost]INFO-WLP is an exclusive TSMC technology it has an advantage over other WLP technology such as Samsung's upcoming FO-WLP in that, it makes the chip package even thinner. So your entire device can be built thinner.

The upcoming iPhone 7 should be thinner than current generation iPhone6S, this made possible by INFO-WLP. The new iPhone will be the first electronic product in history using this new technology.
[doublepost=1470957309][/doublepost]

Apple moving away from them is still a fairly significant chuck of business they are losing.

Apple pays TSMC $3.6-3.8B USD per year. About ~15-18% of TSMC's total revenue. So to be exact $3.6-3.8B USD revenue or about roughly ~5% of total Samsung corporation revenue. It is a significant chunk.

But Samsung makes some of it back by Apple buying Samsung memories, SSD, and LCD panels.

 

[catportal]

It's the most significant SoC advance in a decade. The overall performance of the system will be many fold greater. We're looking at a huge reduction in power between major components with a similar large increase in both communications speed and bandwidth. There will be absolutely no way Apple will be able to not infringe upon Mactel performance and promises a major shift in computing to mobile, obsoleting the need for Intel Processors in most of Apple's devices, dare I say even Mac...

I've lurked this forum for years, read many, many idiotic comments. This comment led me to make an account because this amount of stupidity cannot go posted unchallenged. First, this is not the most significant SoC advancement in a decade, FinFET technology has a drastically greater impact. First of all, do you even know what a SoC is? The reason SoC's can exist today is because process technologies have reduced the area requirements for the CPU CORES, allowing other "SYSTEMS" that were previously separate chips, to be integrated on the same die. Apart from the cost savings, it greatly reduces latencies between interconnects AND power requirements.

Next, performance increase vs Intel processors. Where to even begin, process technologies for silicon will propagate to EVERYONE, this would include AMD, Intel, Freescale (PowerPC), Samsung, etc. The likeliness of Apple switching from Intel ISA to ARM ISA is so slim, it is non existent. There is a reason why Intel performance is unmatched right now, and it is not because of process technology, rather architecture design. Now, before you spew BS about CISC and RISC, know that there is no such thing in today's modern processors. ARM ISA is "RISC" but it's microops and decoder stages work like any other CISC design. Intel's x86/AMD64/add any new ISA, has "CISC" instructions, but are implemented using RISC micro ops. The reason Apple won't switch from x86/64 to ARM for their Mac line is because ARM chips are still terrible for any applications that make use of highly intensive superscalar instructions, various virtualization methods, and multicore scalability. Even the strongest ARM cores would be terrible in comparison. Not to mention, millions in the B2B environment relying on software that are x86 only and will never be ported to ARM. Remember Windows RT? That's what happens when you try to put ARM in a traditional PC, closed app store, no compatibility, bad user experience and eventual shutdown.

Next time you are on this website, please consider pressing the log out button before browsing. You don't know jack **** about computer architecture.

 

[chrmjenkins]

Maybe not but it is better than some of the process shrinks Intwl went through. In any event to me that isn't the important thing, the important thing is that we are getting improvements without process shrinks. Realistically we only have a few, maybe even a couple, of process shrinks left for the current technology approaches. So gains like this between process shrinks are very welcomed indeed.

Combine a new process like this with additional architecture improvements from Apple and we might see a real world 20% increase in CPU performance. Why only 20%, mainly because things like improved caching has a negative impact on thermals, not all of the thermal improvements in the packaging system will translate directly into gains in performance. It really depends upon how aggressive Apple will be in this spin of the processor architecture. If there is room for thermal improvement, in the SOC architecture, we could see them going beyond 30% but Apple chips are already pretty aggressive thermally.

Interestingly what I see happening is that greater bandwidth to RAM will benefit the GPU substantially. On most of the so called APU designs out there the biggest performance problem isn't the GPU cores but their ability to get to RAM. People may be in for a surprise in this respect.

Mobile designers are already well aware of RAM bandwidth limitations due to overall system power. They can't just throw power and wider interfaces at it like the desktop guys do (well, now they can).

I tried to visualize the solder bumps because I couldn't figure out how the CPU would be attached to the PCB if they were eliminated. As as mentioned in the OP, it sounded like the connection method was NOT going to be with solder bumps. After looking at your visualization, it seem that the connection to the PCB will still be via solder bumps, and the major change here is the connection between the different components that comprises the actual chip.

1) What is the material used for connection of the various chip components? Is it wire?

2) If my interpretation is correct, which it may not be, the original article should be updated to reflect that it is not connection to the PCB that has changed. Additionally, if it is wire that connects the components of the chip, that needs to be updated too.

1) It depends. INfO supports POP style packages, meaning the DRAM would sit on top of the CPU and have its I/O wirebonded to the edge of the package, just like previous devices. However, INfO supports multiple dies, in which case multiple flip chip style die could be mated through the wafer level package. The image you quoted shows this possibility.

2) Die with large fanout such as CPUs will always be soldered bumps, unless intended to be socketed in which case they'll have columns. The connection to the PCB has changed because there is no longer a substrate between the die and the PCB. That more direct connection is the point of INfO and why it has those performance advantages.

 

[smalm]

The Galaxy S7 uses a 14nm process CPU already, whereas this is a 16nm process.

These are only different names for what is technically more or less the same process node with Transistor Fin Pitch, Contacted Gate Pitch, and Interconnect Pitch having the same length on both nodes, 14FF LPP and 16FF+ and being significantly shorter on Intel's 14FF process.

Samsung Semi's 2Q sales are up 6% (YoY) and 8% (QoQ), so despite Apple's recent dramatic decline, the company is doing well -- a sign that the division doesn't depend on Apple sales.

Samsung Semi's 2Q operating profits are down 22% (YoY) so I'm sure they are very happy to have substituted Apple orders with Qualcomm orders.

ARM is tailored for max performance at low watts while Intel's chips are for pure performance. There's a reason why Intel has been having a hard time getting their chips to perform really well at low wattages. Apple would have the same problem trying to scale their low power designs up to high wattage performance.

Apple ist faster in cranking up the speed of their design than Intel is in reducing power of theirs.

I believe the A9 was already on the 14LPP process

Mass production for LPP startet January this year. The A9 APL0898 is produced in LPE.
Funny thing is while 16FF+ is slightly smaller than 14FF LPE the APL0898 is slightly smaller than the APL1022.

On a motorcycle there's not much difference going 150mph vs 200mph. Fast is fast.

∆ speed = 33%
∆ air resistance = 78%
Faster is not just faster. I have some doubts you ever went over 150mph on a motorcycle.

Apple pays TSMC $3.6-3.8B USD per year.

I think for the next two product cycles all SoCs will be built by TSMC, and we will not see any APL0898 anymore. Also the S2 SiP will be produced by TSMC. That will add a few dollars to what will not be payed to Samsung but to TSMC instead...

 

[0lf]

Core M isn't the benchmark, Intel chips are good enough for most of a decade now.

Core M is one of my benchmark ... because I have a 2016 macbook. I know what it can do easily, what is OK but perfectible, and what I would not run on it. For example, lightroom is OK 99% of the time, but struggle when before/after view is in use (although it is not worse than with my surface pro 2 with i5 4200u).
I think iPad can be a really great platform for pro level applications, but for this it need a bit more juice.