Apple’s A4 dissected, discussed…and tantalizing
The Apple A4 discussion kicked off in earnest when Steve Jobs introduced the iPad on January 27. In an infamous series of slides tracing the roots of Apple back to the two Steves in his garage, Jobs introduced the Apple A4 processor that powers the iPad, operating at a clock frequency of 1GHz.
The first question was whether the A4 could be the first offspring spawned by Apple's April 2008 acquisition of PA Semi. However, it seemed there was not enough time for a brand new design. Jon Stokesprovided an interesting early insight into the CPU-GPU combination and the potential role of PA Semi.
By the end of March the iPad was in the wild, allowing the teardown and reverse engineering houses to begin their work. The CPU core debate converged to the ARM Cortex-A8 once onlya single core was found and more particularly to the “Hummingbird” implementation of this designed by Intrinsity. First silicon of this core wasreported in July 2009 andannounced one month later.
Once the acquisition was confirmed in April 2010, Intrinsity's name become much more prominent in the A4 discussion promptingMark Anderson to take a look back at some earlier coverage of Intrinsity.
So where have the news, opinion and analysis taken us so far? Can we say that the A4 is truly differentiating hardware, or is it more of an evolution of those that came before it?
Recent discussions of the design of the A4 have largely focused on the identity and capabilities of the CPU implemented in this SoC and the possible integration of design from two of Apple's major acquisitions over the last two years, PA Semi and Intrinsity. The short answer is that the A4 is heavily influenced by Apple's long established relationship with Samsung and represents an evolution rather than a revolution in circuit design. To test this hypothesis it is instructive to look at physical evidence provided by two processors closely related to the A4, the Samsung S5PC110 and the iPhone 3GS application processor (AP).
UBM TechInsights publishedside-by-side die images of the Samsung S5PC110 and the A4, illustrating that the CPU cores are indistinguishable. They then point to aSamsung press release indicating use of an ARM Cortex-A8 CPU core. Based on this work it would be difficult to argue against the theory that the CPU core implementations on each SoC are identical. In other words, the size and arrangement of the circuits within the Cortex-A8 cores on both chips were designed by the same people.
Some blogs have described Samsung's S5PC110 and its role in the Samsung Galaxy phone and tablets. TheSamsung release does not specifically mention the Hummingbird design, but the published specifications, to be conservative, point in that direction. However, is there physical evidence, apart from clock frequency, of the Intrinsity design? To this end we took a closer look at the A4.
 |
Going beyond what was previously available to the general public, we have produced high-resolution microphotography of the A4 architecture using infrared illumination and image acquisition through the backside of the intact die. This new information allows us to see that the standard cells in the A4's Cortex core are arranged in rows half as high as the SoC's other logic blocks. This is a significant distinction in the design philosophy employed in the two regions considering the same 45-nm layout rules must be applied across the entire die. While this depends on certain assumptions, it is certainly suggestive of a design built with something other than Samsung's standard cell library found elsewhere on the die. Is this the mark of Intrinsity? If so, it means the S5PC110 also incorporates Intrinsity IP.
TheUBM TechInsights comparison images provides the most reliable identification of the ARM Cortex generation so far, pinning it down to the A8 based on Samsung specifications for their S5PC110. But this is an SoC. While the CPU is central (literally) to the design, it represents only one of many circuit blocks assembled on the chip. A quick look at the A4 and the S5PC110 side by side suggests the S5PC110 and the A4 are similar, but there are certainly differences. On the digital side, there are the same number of blocks, at least it appears so from the detail offered by the UBM TechInsights die images. On the analog side, some blocks are drawn identically on both dies. In one particular case, Samsung employs two instances of the circuit while Apple's A4 contains only one, proving that block level customization took place.
A block level comparison is an important part of understanding the A4 and its capabilities, but what does it reveal about the argument for SoC customization by Apple? One of the best reasons for Apple to create its own design rather than simply buy a standard product is that it allows Apple to eliminate many of the blocks typically deployed by IC vendors anticipating the needs of a broad range of OEMs. Why go with one-size-fits-all when you can get something tailor-made that reduces the chip complexity, footprint and cost? To be clear, we are not talking about full custom circuit design, yet. This is about choosing from essentially the same catalog of individual IP building blocks but selecting fewer of them. By now the comparison of Apple to Samsung on their respective Cortex-A8 generation SoCs should convince us that Apple has created a custom design and named it A4.
With its earlier release, the iPad was central to much of the A4 discussion. Apart from a prototype left hanging around in a bar, the appearance of the A4 in the iPhone 4 was officially announced at the WWDC keynote on June 7. A comparison can now be made between similar platforms, the upcoming iPhone 4 and its predecessor, the 3GS AP. For the chip level comparison, we used two pieces of information that have been available publicly for quite some time. UBM TechInsights released a die image of the 3GS deprocessed to the diffusion or active area level. This 3GS image was analyzed and compared to a Chipworks' annotated die micrograph that was the first to reveal useful details of the A4 floorplan.
The simplest and most striking observation is just how little discernible change there is to the number or type of circuit block. Both devices have a relatively high percentage of the die consumed by an ARM CPU core containing a large L2 SRAM cache memory along with 10 additional blocks of digital logic. The A4 die is smaller, 51.8 mm2 versus 72.2 mm2 , but this says little about the design since A4 is manufactured with 45-nm technology.
Although the architecture appears to be very similar, there are significant changes in the allocation of real estate between the Apple A4 and iPhone 3GS AP. Looking at the contribution of specific regions as a percentage of the total die area, several observations can be made:
 |
From a circuit design perspective, the changes in the subdivisions of chip real estate are relatively minor. A lot can be attributed to transitioning manufacturing to 45-nm. To summarize the block level comparison with the two “reference” designs, there were no wholesale changes to the floorplan. Yes, the A4 is different, but not by more than one or two blocks. However, even if only one or two circuit blocks in Samsung's Cortex-A8 generation of SoC are missing from Apple's, we should be convinced the A4 is a custom design. It is also reasonable to describe it as evolutionary compared to the references.
At the 27:30 mark of his January keynote, Steve Jobs introduced the A4, but he actually suggests it isnot the first custom Apple design. “We have an incredible group that does custom silicon at Apple.” He goes on to say that the A4 is, “our most advanced chip we've ever done.” That's a pretty good hint they've been at it for a while.
Getting back to the Apple acquisitions, the preponderance of evidence points to an Intrinsity designed ARM Cortex-A8 core. This had more to do with the existing relationship between Samsung and Intrinsityjust look at the timing. The acquisition of Intrinsity does, however, draw a line in the sand that may well be ahead of any Intrinsity re-designed Cortex-A9 appearing, leaving only Apple with access to Intrinsity for the multi-core generation.
What about PA Semi? The A4 certainly does not appear to have much influence from PA Semi, at least at the block level. Some designers were no doubt involved in the A4, but definitely not to the point of a wholesale CPU design. The team was likely involved in other blocks or just applying their expertise to help pull together a complex SoC design. Steve Jobs described the PA Semi acquisition as part of Apple's strategy of product differentiation. But if a truly unique CPU core is forthcoming, it will be a long road. This is certainly echoed in areview at AnandTech of Apple's history of acquiring chip design companies. To wrap up this line of thinking, it's also worth goingback to the first A4 article we referenced where a number of possible design directions were considered.
Despite offering only an optimized version of a standard CPU, the A4 is becoming increasingly important to Apple's strategy with it appearing now in the iPhone and surely in iPod touches to be released in Septembernot to mention any future iOS product lines. In all the discussion of CPU identity it would be easy to lose sight of the fact that the A4 is not the offspring of either an integrated semiconductor house or a fabless designer. It is from a mobile devices OEM (sorry Mac). Apple is producing the A4 for its own consumption, where the revenue of the final OEM device is the prize. A full analysis requires thinking about the chip in the context of the whole device, including the OS, which is known to be quite efficient. Consideration should also be given to the broader business and IP implications of the possible expanding role of circuit design in Apple's strategy and the arguments for introducing a new product with incremental advances in technology rather than wholesale design changes. Not that Apple has ever done that before.
About the authors
Paul Boldt is currently a Principal atned, maude, todd & rod inc., an Ottawa based technology consulting company. Prior to this he spent six years drafting and prosecuting patent applications at one of Canada's largest IP law firms and three years reverse engineering integrated circuits. Paul holds a Ph.D. from the Department of Materials Science at McMaster University.
Don Scansen spent ten years as a reverse engineer at Semiconductor Insights where he held senior roles in engineering and product management. Seeking a more holistic view of technology, he foundedsemiconDr Inc. Always keen to share his opinions on technology, Don blogs regularly at semiconDr.com and makes frequent contributions to EE Times. He holds a Ph.D. in Electrical Engineering from the University of Saskatchewan and is licensed as a professional engineer in the Province of Ontario.
Tim Whibley is a technology consultant specializing in IP management, infringement detection, and patent analysis. For the past nine years, he worked as a reverse engineering analyst and as a successful patent broker before recently foundingTen Yard Technology.
Acknowledgments
The authors would like to thank the contributions of product teardown and market analystsFomalhaut Techno Solutions for providing iPad component details as well as the A4 circuit board which allowed us to analyze the SoC architecture of Apple's first ever integrated circuit. Our understanding of the floorplan of the A4 was facilitated by backside infrared imaging undertaken byMuAnalysis.
Leave a ReplyCancel reply
This site uses Akismet to reduce spam.Learn how your comment data is processed.
Advertisement
