Possibility of a Cortex A15 based chip in iPhone 5...

[SR71]

So, people keep saying how it's impossible for Apple to even use an A15 based processor in their next iPhone, but I was thinking... What if they have secretely been developing it over the past year or so? The next iPhone will more than likely have a 32nm Cortex A9 based chip, but is it really "impossible" as some claim, for Apple to somehow have developed their own A15 chip with the help of a chipmaker that they might've partnered with?

Think about it... Apple has stated many times publicly that they have over $100 BILLION and don't know what to do with it. It's only reasonable that they'd want this iPhone (the last one that Steve Jobs worked on and had the most say in how it was made) to be the greatest iPhone they've ever built. They easily could've spent maybe $1-3 billion on R&D on their own A15 chip after the iPhone 4S's release and have it ready in time for their next iPhone and kept it secretive by only letting a select few trustworthy executives/employees know about the project. Although many think this may be impossible, don't forget that everyone though a device like the iPhone was impossible back in 2007, but look how that turned out.

Does anyone think this could be a possibility? I think this could be a great discussion, and yes I realize there are other threads with similar topics, but I think I've brought up some good points that could be discussed on both sides. So, with that... Discuss!

 

[Big.Mac.Daddy]

Nothing is impossible.

 

[M87]

They could be using potato chips for all I know.

 

[ixodes]

Remember that when the discussion turns to specs, if Apples aren't the best, the default response is "specs don't matter".

 

[VulchR]

Remember that when the discussion turns to specs, if Apples aren't the best, the default response is "specs don't matter".

Correct. User experience matters. However, match a fast chip with optimized software and the user experience will get better.

 

[yow.]

Some things can't be accelerated, no matter how much money you have. I'm not sure if this applies to A15 or not... it depends on the reasons for the delay.

Does anyone know why it takes so long to get new silicon IP into consumer products? Naively, I would think that once they had the IP, they would need only design the supporting components, get it to fabs (assuming they have sufficient capacity at that process size), and then physical manufacturing, assembling, packaging, shipping, building up sufficient inventory.

I feel the physical tasks would only take weeks/a month or so; and the bulk of the delay is in getting the design right, interactions between components, various bugs and unanticipated teething troubles.

I feel like it should be much quicker than it is - but lots of clever people with tons of resources and motivation for being first to market not doing it faster means that I simply don't understand what is involved. Anybody know?

If we knew what took the time, we could better determine if throwing billions of dollars at it could speed it up.

 

[SR71]

Nothing is impossible.

Agreed.

They could be using potato chips for all I know.

You're making me hungry...

Remember that when the discussion turns to specs, if Apples aren't the best, the default response is "specs don't matter".

Haha, too true! I feel as if Apple is very conservative with their specs and one of their main priorities is battery life, which is another reason why they might go for an A15 based chip -- they offer significant gains in battery life, power, and their GPU's are MUCH faster, which is something Apple loves to tout. Specs are great, but sometimes you have too look past specs (i.e. the 2010 MacBook Air's -- they seemed to perform VERY quickly when in reality they used old Cord 2 Duo processors), while on the other hand, specs are very important (i.e. when deciding between what brand to get for an Android phone -- Motorola phones offer decent specs when compared to Samsung and in turn, the Samsung Galaxy S line of phones performs much quicker, although this could also be related to manufacturer's optimizing their skins better).

So, while fanboys of Apple can proclaim "specs don't matter", they do. That's like saying what if the next iPhone shipped with a single core Cortex A8 at 500MHz. It'd run terribly slow. It needs good specs in order to perform at the level of speed we have become accustomed too.

I'm liking this discussion, keep it going, guys!

 

[Tom8]

I thought that the main problem was producing enough A15 chips to satisfy the demand for the new iPhone and avoid shortages? I could be wrong on that.

 

[SR71]

Some things can't be accelerated, no matter how much money you have. I'm not sure if this applies to A15 or not... it depends on the reasons for the delay.

Does anyone know why it takes so long to get new silicon IP into consumer products? Naively, I would think that once they had the IP, they would need only design the supporting components, get it to fabs (assuming they have sufficient capacity at that process size), and then physical manufacturing, assembling, packaging, shipping, building up sufficient inventory.

I feel the physical tasks would only take weeks/a month or so; and the bulk of the delay is in getting the design right, interactions between components, various bugs and unanticipated teething troubles.

I feel like it should be much quicker than it is - but lots of clever people with tons of resources and motivation for being first to market not doing it faster means that I simply don't understand what is involved. Anybody know?

If we knew what took the time, we could better determine if throwing billions of dollars at it could speed it up.

What makes you say that (referring to your first statement about money)? More money spent on more researchers = faster development. More money spent on hiring chip designers, on getting it to fabs, and lastly money spent on partnering with some manufacturers will also speed up the process. Like you said, the physical aspects of it might not take too long, it's the designing process, bug testing, getting voltages at the right level, getting the supporting parts ready, etc. But that also brings up the question of how long? They've had a little less than a year now, possibly more if they started looking into this before the release of the iPhone 4S.

Knowing Apple, they've had this in the pipeline for a while (look at their Project Purple design that was started in 2005 and only released as the iPhone 4 in 2010), so I don't think they've only recently started this development.

You bring up good points, and for all I know, I could be talking out of my ass lol... I don't know much, if anything about developing a new chip architecture.

----------

I thought that the main problem was producing enough A15 chips to satisfy the demand for the new iPhone and avoid shortages? I could be wrong on that.

This is true, but again, this can be solved with spending more money on getting more manufacturers to help build the new chips needed for the iPhone's. This is why I really think that an A15 chip is possible, for the most part.

 

[CosmoPilot]

Wirelessly posted

The true answer is the "right" specs w/the "right" hardware/software. It's a balancing act of size (form factor, pocketability), performance (ability to run today's software efficiently), and usability (battery life, radio capability, cameras, etc.).

A truly good phone maker (think Apple) doesn't put stock in one of the above categories over the other. They balance the entire spectrum to provide an end user experience second to none. Apple could easily make a power hungry device in a small form factor, or they could make a large phone with a huge battery at the expense of pocketability.

In the end, Apple has not proven to me any track record that suggests the next gen iPhone wil not outperform the next. The 4S is still a powerhouse device. So the iP5 should be amazing!

 

[Wuiffi]

I'm with you on this one. I know the A5 is insanely fast and still is one of the fastest chips on the market. If they clock it with 1 Ghz (which should be no problem, when moving to 32nm) it's even faster. I guess it could perform as good as the quad core in the Galaxy S3. But we should not forget: as soon as the A15 chips are ready (which is expected Q4 12-Q1 13) a lot of phones will move to dual core A15 and even quad core A15. those chips will easily outperform the A5. If Apple keeps it's Sept-Oct release, this means, that even for about half a year most smartphones on the market will outperform the iPhone

 

[yow.]

What makes you say that (referring to your first statement about money)? More money spent on more researchers = faster development.

The old example is that a baby takes nine months... you can't speed it up to 1 month with 9 mums. More formally, there's Amdahl's Law. In project management, it's called critical path analysis - it's when you can't do part B of the project until you've done part A. You can't do them in parallel by hiring more researchers. There can be many such dependencies - a long chain - so it can take a long time.

How does this apply to silicon IP in general and Cortex A15 in particular? I don't know.

I can imagine: if there's a lot of tweaking needed, and each experiment takes time. You could do many experiments in parallel, but if you only have a few experimental apparatuses because they are so expensive (e.g. fabs) then that limits you. Also, if having the results of the previous experiment really helps guide your search, each experiment will be less effective if you proceed in parallel (because you lack that guidance).

I've heard that at Intel, a lot of careful tweaking goes on at several complex stages of production, when they move to a smaller process node - it still takes them a couple of years for each loop of their tick-tock cycle. They have billions of dollars, they're *only* doing the CPU (not a whole device), and when they work on the integrated circuit (the silicon chip), we must remember that they invented it.

One of Apple's hallmarks is making integrated computers. They design the software, the hardware, the SoC etc - this means they can make them work especially well with each other, so they interface efficiently without waste. As a contrast, PCs are modular and use commodity components, with standarized interfaces - the advantage of standards is you can quickly mix-and-match components (plug-and-play) to design a new computer, and it all works. This was actually an express goal of IBM when they made the first PC (it only took them a year, IIRC, because they just used COTS components). The disadvantage of components is the same as one-size-fits all - it doesn't *really* fit, but it sorta kinda works well enough for the job. It's adequate.

So, this "integrated" aspect is like a bespoke suit, with each part perfectly matched to the rest. That takes great skill and a long time. Maybe there also just aren't that many people with the very specific skills, at the very high standard required? That would also limit how much you could parallelize.

Of course, Apple isn't the only one making Cortex A15 cores - there are 5 or 6 other designers, who aren't so integrated. But the chip itself - the "system on a chip" - is integrated, with many different components apart from the CPU. So maybe that delays everyone.

But I'm just guessing (apart from some Intel, Apple, IBM and SoC facts), I don't actually know.

They've had a little less than a year now, possibly more if they started looking into this before the release of the iPhone 4S.

I think they were working on it before the iPhone 4S release, because of the two serial numbers embedded in iOS before the iPhone 4S came out - sparking rumours of an A6 way back then.

*EDIT* AnandTech thinks just a plain A5, shrunk and upclocked http://www.anandtech.com/show/6196/preparing-for-the-iphone-next-rumors-analyzed/2

 

[SR71]

The old example is that a baby takes nine months... you can't speed it up to 1 month with 9 mums. More formally, there's Amdahl's Law. In project management, it's called critical path analysis - it's when you can't do part B of the project until you've done part A. You can't do them in parallel by hiring more researchers. There can be many such dependencies - a long chain - so it can take a long time.

How does this apply to silicon IP in general and Cortex A15 in particular? I don't know.

I can imagine: if there's a lot of tweaking needed, and each experiment takes time. You could do many experiments in parallel, but if you only have a few experimental apparatuses because they are so expensive (e.g. fabs) then that limits you. Also, if having the results of the previous experiment really helps guide your search, each experiment will be less effective if you proceed in parallel (because you lack that guidance).

I've heard that at Intel, a lot of careful tweaking goes on at several complex stages of production, when they move to a smaller process node - it still takes them a couple of years for each loop of their tick-tock cycle. They have billions of dollars, they're *only* doing the CPU (not a whole device), and when they work on the integrated circuit (the silicon chip), we must remember that they invented it.

One of Apple's hallmarks is making integrated computers. They design the software, the hardware, the SoC etc - this means they can make them work especially well with each other, so they interface efficiently without waste. As a contrast, PCs are modular and use commodity components, with standarized interfaces - the advantage of standards is you can quickly mix-and-match components (plug-and-play) to design a new computer, and it all works. This was actually an express goal of IBM when they made the first PC (it only took them a year, IIRC, because they just used COTS components). The disadvantage of components is the same as one-size-fits all - it doesn't *really* fit, but it sorta kinda works well enough for the job. It's adequate.

So, this "integrated" aspect is like a bespoke suit, with each part perfectly matched to the rest. That takes great skill and a long time. Maybe there also just aren't that many people with the very specific skills, at the very high standard required? That would also limit how much you could parallelize.

Of course, Apple isn't the only one making Cortex A15 cores - there are 5 or 6 other designers, who aren't so integrated. But the chip itself - the "system on a chip" - is integrated, with many different components apart from the CPU. So maybe that delays everyone.

But I'm just guessing (apart from some Intel, Apple, IBM and SoC facts), I don't actually know.



I think they were working on it before the iPhone 4S release, because of the two serial numbers embedded in iOS before the iPhone 4S came out - sparking rumours of an A6 way back then.

*EDIT* AnandTech thinks just a plain A5, shrunk and upclocked http://www.anandtech.com/show/6196/preparing-for-the-iphone-next-rumors-analyzed/2

Although it may not be possible to work on Part B until you've finished Part A, you can still hire more and more researchers to work on Part A, thus speeding up the process. So, in the end it still boils down to how much Apple would want to spend on fabs, researchers, etc. Like you said, they've most likely been working on this for well over a year. I think an dual-core Cortex A15 based chip is definitely a possibility. We'll have to wait till next week. You brought up very good points. Great discussion, thanks all for participating.

 

[yow.]

an A15 based processor in their next iPhone

Supporting your idea: it's not enough for the iPhone 5 to reach parity with current gen phones; it must retain competitiveness with A15 phones of early 2013. They'd wait til it's ready.

But I'm betting on an A5X, shrunk + GPU downclocked to free up power for CPU upclock (relative to iPhone 4S), because I'd rather be pleasantly surprised than disappointed again. :.(

 

[Fernandez21]

I think the hold up is in the manufacturing, not in the chip design. A15 is 28nm chip, currently only the snap dragon s4 has and they are struggling to keep up with demand.

 

[yow.]

I think the hold up is in the manufacturing, not in the chip design. A15 is 28nm chip, currently only the snap dragon s4 has and they are struggling to keep up with demand.

The Cortex A15 isn't tied to a particular process node. For example, Samsung's Exynos 5 has dual A15's at 32mn. Apple already has capacity at 32nm, since the A5 SoC in the iPad 2 moved to 32nm (when the iPad 3 was released). It seems likely they'll stay with that, instead of switching to 28nm. Of course, there may be a hold up even so - there's often limited capacity at new process nodes. You could be right.

BTW: according to wikipedia, snapdragon s4 doesn't use the cortex A15, but a different CPU core design with similar performance called "Krait".

 

[midtownhd]

I think the hold up is in the manufacturing, not in the chip design. A15 is 28nm chip, currently only the snap dragon s4 has and they are struggling to keep up with demand.

The snapdragon s4 are not quite A15 yet, they're more like 28nm A9+. Trends do show that they're conservative with these kinds of decisions: 3G, LTE, NFC