The A6 has landed - sort of

Speculation around the design of the A6 reached a fever pitch in the last few days, and all that I have seen focuses on the particular variants of the CPU and GPU. One might even say these cores are the whole discussion. It is a widely held notion that the A(x) processors are simply a CPU and GPU (now each having multiple cores) tied together with a bit of transistor glue that makes everything work. That is a shame.

It was back in January 2010, before the iPad was presented to the world I recounted a comment Steve Jobs made at the time of the PA Semi acquisition in April 2008. To paraphrase, Steve said, “We want to further differentiate our products through IC design.” I have always taken this to heart making it the working premise of our analyses of the A4 and A5. Today I will extend this to the A5X and forthcoming A6.

The A5X saw the light of day back in March of this year with the iPad3. As is apparent, it is -- or at least Apple considers it -- a variant of the A5. It was a whopping die indeed. Chipworks even used the phrase “Big is Beautiful” in the title of their article comparing the A4, A5 and A5X die. One can quickly compare the A5 and A5X die photos and see the majority of the increase in die size is related to the addition of two GPU cores. I say majority because there is more to it than that.

More specifically, with the help of a trusted HP-15C, the GPUs on the A5 consumed approximately 30 mm², or 24% of the total die area. This number increases to 62 mm² , or 40% of the total die area for the A5X. In both cases the dual ARM cores consume approximately 18 mm².

Does this account for all of the increase in die size? The short answer is no. With a die size of 165 mm² the A5X is approximately 43 mm² larger than the A5. After the 32 mm² increase in GPU real estate there are 11 mm² remaining. This does not sound like too much for the A5X, but it is 25 % of the total A4 die area. So there is additional design here. One can further count 15 digital blocks outside of the CPU/GPU combo in the A5X compared to the 12 of the A5. Beyond this are what appear to be two arrays on the A5X that are absent in the A5. These “arrays” have been highlighted on a partial A5X die photo. The RE houses have probably already confirmed what these are, but it is fun to speculate that they might be ROM containing microcode.

What is the stuff outside of the CPU and GPU cores? It is a mixed bag. There is certainly somewhat mundane stuff to make the whole thing operate such as memory bridges, specific interface circuitry, other peripherals and the associated overhead for these. Then there are the various IP blocks that may well include the anobit (now Apple) Flash controller. There are also likely blocks that perform dedicated tasks of the processing. We also discussed the possibility of hardware based encoder blocks in our A5 article. The general theme of “extra” area was picked up by Microprocessor Report which commented that 33 mm² can not be accounted. This was very similar to our earlier estimate of 34 mm². Back to the A5X, the extra 11 mm² falls into this group. I really think this “unaccounted” area could include circuitry that breathes life into Steve’s “further differentiation” comment.

We find ourselves in the neutral zone at the moment. It has been less then a week since the keynote that officially introduced the A6 to the world and less than a week from it being in the hands of consumers and, more to our point, the reverse engineering houses. Until the die photos begin to appear this weekend, we can only go on bits of available information and the A(x) processor lineage. As pointed out in an arstechnica article, there was scant information about the A6 in last week’s keynote. No real surprise there. Apple provided one key piece of information in that the A6 is 22% smaller than the A5. This gives a die size of 96 mm². At this point, the arstechnica article brings in the A5X and the reduction in die area associated with the pretty much certain process shrink from 45 nm to 32 nm. With an A5X die area of 165 mm² and a 41% reduction in area with the shrink one gets 96 mm². Yes it is the same number, but I think this is a red herring. I do not think any interpretation of GPU/CPU cores can be extracted from this number. Again it is not simply an accounting of the CPU/GPU area. There is some really interesting circuitry outside this combo. Apple is doing more than just re-arranging the furniture.

Then came last Saturday’s article in Anandtech. It really blew the doors off the A6 discussion. The well reasoned article claims and provides evidence that Apple actually designed its own ARM cores for the A6. Here we are drawn back to the PA Semi and Intrinsity acquisitions. With the A6 we might see a fuller picture of the CPU work that has taken place at Apple since their buying spree started. Recent, and widely quoted, geekbench results suggest the iPhone 5 is scoring very well, hinting of some serious design here.

As soon as people talk about custom design and/or reduced die area they often fall into the economics trap. This is not about increased margins or reduced silicon costs. The A5X killed that notion. It is all about the end user experience. Apple will always choose to fight on this experience rather than on final cost. In his recent review of the Kindle Fire HD, Walt Mossberg commented, “The Fire HD isn't as polished, fluid or versatile as the iPad ... the Fire HD showed signs of latency—apps and content displayed delays in launching. This latency disappeared after a reboot.” Apple will consume extra silicon and design it with intimate knowledge of the OS to make sure theirs is a fluid experience. Remember. Apple’s silicon is about further differentiation of the iOS device. Nothing more. Certainly nothing less.