Apple’s Homemade Application Processor
It is frequently believed that Samsung designed the iPhone's application processor, but that's not completely accurate. It is definitely manufactured by Samsung, and they did have a hand in designing it, but according to our information a significant part of the design at the Register Transfer Level (RTL) was created in-house by Apple. The physical synthesis, however, seems to have been done entirely by Samsung.
This methodology is sometimes known as an ASIC flow manufacturing model, in opposition to COT (Customer Owned Tooling) where the design is brought to completion by a single company with no direct assistance from the fab. It has to be said this terminology is rather ambiguous, given the usage of the former term for other things than manufacturing models. Amusingly, Apple's previous supplier for the iPod (PortalPlayer, now part of NVIDIA) also used an ASIC flow for all the chips they supplied to Apple, so everything they used to do has now been integrated firmly in-house.
Apple's strategy of designing only the application processor is actually the exact opposite of certain other manufacturers, such as Sony Ericsson (and Nokia until very recently), which design the baseband in-house and use a third-party application processor either as a standalone chip or integrated in a joint single-chip solution. There's a cost advantage to that, and maybe a little bit of lead time, but nothing drastic and no real power consumption advantage.
The advantages of Apple’s approach, however, are very compelling: they are one of the only manufacturers using their own full-featured operating system, which makes tighter software/hardware integration especially interesting. Very importantly, this allows for much better simultaneous development of the phone’s software and hardware; this drastically reduces the lead time compared to the traditional alternative where the phone manufacturer can only start part of the software development process once the chip is back from the fab and the design win has been finalised.
On the other hand, it's only really worth doing something in-house if your solution can still be highly competitive with those designed by more traditional industry players who possibly have up to a decade of experience in the field. In the original iPhone, they likely benefited from Samsung's expertise in multimedia processing and embedded memory (which the iPhone uses); there’s another thing Apple seems to be doing as of this generation though: pick-and-choose intellectual property cores if they’re better than what they can do in-house.
With all due respect to the application processor/wireless companies out there, many have a ‘Not Invented Here’ mentality in terms of hardware. It doesn’t matter if it's provably suboptimal; who cares as long as it’s good enough, right? That’s a loser's strategy, of course, and it should be easy to see that using carefully selected third-party intellectual property (especially for multimedia) is a better approach unless you are absurdly unit-cost constrained (which Apple is clearly not; the manufacturing cost of the application processor is likely negligible from their point of view).
The Specifications
In August 2007, Imagination Technologies’ PowerVR group announced a licensing deal with a mystery company for their handheld 3D and video decode cores. Much more recently, it has also been revealed that Samsung would be the manufacturer of the chips using that license, clearly implying that Apple was the mystery company. We believe the cores in question are the SGX 520 and VXD 330 for the 3G iPhone, and the SGX 530 (or 540) and the VXD 380 for a possible 2009 iPhone.
Furthermore, based on the 3G iPhone's timeframe, it is extremely safe to claim that it will be manufactured on a 65nm process and it will likely still sport an ARM11 core (but probably clocked noticeably higher); it’s too early for a Cortex-A9, and we have a variety of reasons why we believe it’s not an A8. As for video encode or image signal processing (for the camera), there’s a slight chance Apple might have designed those in-house, but we wouldn’t be surprised at all if they haven't.
While this is not a technical discussion of the merits of different video decode architectures, we are extremely confident that VXD 330 and 370/380 are the best, or at least within the very best, video decode cores available on the market. The VXD 370 claims to deliver (and actually does in Intel’s Poulsbo chipset for their UMPC/MID Atom-based platform) power numbers for 1080p High Profile H.264 that are comparable to those of many competitors for their VGA resolution decode. Clearly, that’s a huge advantage, and given what we’ve examined of their architecture based on public information, we’re ready to believe it.
We believe it is more likely that Apple will go with the VXD 330 core, which does not support HD video resolutions, for two main reasons: first of all, the VXD 370 is still relatively big on 65nm and requires fairly high memory bandwidth. Secondly, Apple has never been on the cutting edge for this kind of functionality and neither HDTV/HDMI penetration nor NAND storage capacities are high enough yet to truly justify such a feature on a 2008 mass-market device.
Furthermore, it is interesting to note that the availability date of the PowerVR VXD 330 and PowerVR SGX 520 match quite closely. The former was, in Q1 2007, available for ‘lead partners this quarter and for general licensing next quarter’ while the latter was also made available in 1H 2007. While it may seem aggressive to release in June 2008 a product based on RTL IP that was only completed slightly more than a year earlier, we believe this is highly plausible given Apple’s unique position and history of short lead times.
The consequences of the VXD 330 core are likely to be simple: better battery life. As for the SGX 520, it will help the iPhone’s entry in the gaming category by delivering more than twice the performance of the MBX Lite in the original iPhone. Furthermore, some additional OpenGL ES 2.0 features may also come in handy to further improve the iPhone’s user interface in the future.