Get email delivery of the Cadence blog featured here
In case you may have missed it, Apple recently launched a new iPhone. As per the iPhone launch tradition, it brings with it a lot of excitement over the latest capabilities. Of course we don't know everything until it is actually available, but this latest incarnation has broken all kinds of records for pre-sales, so it must be groundbreaking. Right?
There are a lot of comparisons out there between the iPhone 5 and the 4S, but in reading through all of them it seems on the surface like the 5 is basically delivering a lot of incremental improvements. From a software point of view, it comes with iOS 6, but the existing iDevices will be upgraded to that as well, and there are no iOS 6 features exclusive to the iPhone 5. The new "Lightning" connector is probably a wash from a customer benefits point of view -- its smaller footprint is a good thing, but having to replace or buy adapters for all your chargers, docking stations, etc, quickly makes you think twice about upgrading.
This iPhone finally has LTE so you'll get faster data speeds, but that consumes more power. The display is larger and higher resolution, but that consumes more power. They tout big improvements in performance, which means a faster CPU which must consume more power, right? The device is thinner and lighter, thanks in large part to a thinner battery. With all this extra power consumption and a smaller battery, how can they possibly be touting longer talk, browsing, and standby times?
This is where Anand Shimpi is such a valuable resource. He has done some excellent investigation, which you can read here:
Basically his conclusion is that this phone sports a custom Apple-designed processor that uses the ARMv7 instruction set. Apparently Apple licensed the instruction set from ARM shortly after their PA Semi acquisition. For more on that:
The key piece of that article is the last paragraph:
At this point, Apple has spent about $400 million to buy PA Semi and Intrinsity, tens of millions for a license to design its own ARM CPUs, and probably north of $100 million to support its CPU design efforts over the past four years. It appears that the end result will be that Apple ships a Cortex-A15-class CPU about three months before arch-enemy Samsung does. These three months happen to come during the big holiday buying season, during which the iPhone 5 could generate $25 billion in revenue. So that half billion dollars could be money well spent.
I'm sure they didn't just have a 3 month time-to-market advantage in mind when they spent a half-billion dollars and they could not have forecasted the timing of it. They must see this capability as something they can leverage to build sustainable differentiation by being able to optimize the hardware for the needs of the system. In this case it's simply the raw performance vs. power tradeoff. But this is just the first time through. Now they have the ability to implement in hardware what they could previously only implement as software. An analogy is what they did when they integrated Audience's noise reduction IP into their A5 SoC to enable the Siri capability.
Even though Siri looked like a software-only feature, Apple felt it would not have worked smoothly enough without integrating the Audience processor IP directly into their SoC. It was not only the capability, but also the user experience, and that required software and hardware. Of course Audience is a separate company and if you go to their web page you can see that their processors are available in a wide variety of devices, so others could quickly build copycats, at least on the hardware side. I don't know what other capabilities Apple could add to an iPhone in the future, but now they can design both the software and hardware aspects if they feel they can do better than off-the-shelf.
So if you're one of those device makers listed on Audience's site, how do you create differentiation? They all run a 3rd party OS, most of them Android, so the ecosystems are the same too. They all use commercially-available processors. Most of them even have the same rectangle-with-rounded-corners shape that has been so discussed in the news lately. This is why so many incredibly powerful smartphones end up just becoming low-cost alternatives.
Samsung has made the most inroads in the smartphone market segment, and their Galaxy S3 is has been a formidable opponent on many fronts. Now, Samsung designs a very good processor, but they also sell it to other device makers (including Apple until now). They could gain advantages if their system designers work closely with the processor design group, by better optimizing design on both sides for a better overall Samsung device and gaining some time-to-market advantage over the device makers to whom they sell their processors.
Is that enough though? The Galaxy S3 has a Siri-like capability that uses a standalone Audience processor, so likely the best they can do there is catch up. But the S3 also has features that help it interface more readily with televisions and other home entertainment devices -- this is an area where they are leveraging their strength to add differentiated capability if it works smoothly. The latter part of that sentence is the key, and where implementing the capability in hardware would most likely provide the best experience.
But for device makers without the financial resources of Apple, investing a half-billion dollars just in chip design and implementation engineers is a large barrier. Fortunately, higher abstraction design can offer a more efficient path to hardware without the need for an army of engineers.
Just as the move to RTL democratized chip design in the early 90's, moving to the transaction level (TLM) can enable more systems companies to add their own flavor of differentiation to existing SoC platforms. In fact, some of our existing C-to-Silicon Compiler customers are doing just that. Instead of buying the same off-the-shelf processor that their competitors are and implementing their value-add in software, they can now add their own hardware IP to an SoC platform. For instance we have seen facial tracking, on-the-fly data encryption, video codecs, and other compute-intensive features get implemented as hardware IP instead of taxing the processor as application software.
So the iPhone 5 may be the beginning of a hardware design Renaissance. Though it's really about designing a product and being able to optimize the software vs. hardware and off-the-shelf vs. secret sauce tradeoffs. Apple of course can afford to invest heavily to be able to do this. The easiest way for other device makers to play this game is to move their design and verification up in abstraction to TLM.