posted on June 07, 2012 |
This is going back a bit but the appearance of A5 “single core” in the "new" 32 nm process has always been intriguing. At its root there are two questions: Why is the part marketed as “single core” and why expend the effort of a process shrink for a now older part that is showing up in low volume products i.e Apple TV 3 and iPad 2? There is also a feeling that we could get closer to the answers. If you will indulge us for a moment, follow along while we suggest that there is, in fact, one answer to both questions.
Chipworks posted their reverse engineering
. It was a nice find to say the least. Well done. In their article on the A5 single core, they present a hypothesis for this terminology used by Apple. The idea of “binning” is put forward, where binning refers the practice of sorting chips according to their performance during testing. This is a common practice in the industry. Intel for example may design to one target speed of a given processor. Then, after the parts are tested they “bin” the processors according to performance and voila one has say 2.0, 2.2 or 2.4 GHz parts for the sales team to address multiple segments of the market. What is unclear is whether this practice is applied to cases where a whole core may not work. This seems less likely. What is more likely?
Let’s start with what we have been told.
Apple uses the phrase “A5 Single Core” on the current tech specs page
for the Apple TV. Before attempting to draw conclusions from any marketing message, one should note what is said definitively and what is open to interpretation. Apple is no slouch in the marketing department so the cautious approach definitely applies to this phraseology. The use of A5 is definite, so one can probably anticipate essentially the A5 processor, including the major logic blocks. But, “single core” is indefinite. The term core could apply to several different blocks, including the CPU, GPU or hardware encoders. It seems everyone that has commented on this chip, including Chipworks, assumes Apple is referring to the CPU. They identify the two ARM 9 cores in the die photo, reproduced here. But there are other “dual” cores on the A5, shrink or no shrink.
Subsequent to publishing our article on Apple's A5
last year, one of the authors had discussions with a colleague who designs video application processors. The A5 die photo showed 3 sets of dual logic blocks. One set is the ARM 9 CPU, the second is the GPU and the third was postulated to be twin H.264 encoders. This agrees with earlier rumours we brought forward in our article on the A4
. It was further thought these encoders on the A5 were designed for up to 1080P (or full HD) video. Since we have not done the circuit level reverse engineering or even logical block analysis, we will call the existence of the H.264 encoders a hypothesis. To this end it is worth noting that such encoders will provide the most efficient processing in terms of power and speed for HD video. It is also noted that our hypothesis does not imply the blocks will be used to the fullest in every product in which they appear. Rather there will be one or more products in the lifecycle of the applications processor that implement this power.
The last remaining device implementing the A4 is the 5th generation iPod Touch
. It sports a single 720P high resolution camera. Although the iPod Touch is still a dual camera device, the front-facing imager is only low pixel count VGA. Going back many years now for VGA processing, there has been little need for hardware acceleration for 0.3 megapixels. The advantages started to appear as camera pixel counts grew, and consumers demanded higher frame rates. For the full 30 frames-per-second 720P video offered in the Touch, it is not difficult to imagine that a dedicated encoder would come in handy for processing three times the pixels in each frame of video.
Moving forward to the A5 we move up to two H.264 encoders that can each handle 1080P. Going with our feeling that the A5 is definitive, the A5 single core should have these twin encoders. What does the die photo say? We have reproduced the Chipworks die photo and added our own annotation. We have identified the ARM cores as A9-1 and A9-2. These cores consume approximately 16% of the total die area, compared to the 14% we found for the A5. So far so good. We have labelled 3 dual blocks X 1-2, Y 1-2, and Z 1-2. It is not clear if the Ω 1-2 blocks are symmetrical from the resolution of the photo, however it appears they are. Now there has to be more than the dual core GPU and any supporting circuitry here. These eight blocks consumed approximately 21% of the die area compared to 25% for the four blocks above the ARM cores on the A5. Going with our working hypothesis the dual H.264 encoders are here.
Now move forward again to the Apple TV 3. It presumably only has to encode/ decode one HD video stream. The term single core might then refer to the use of only one the H.264 encoders, and not the dual CPU cores. Now, why would you go to the effort of a shrink on the A5 when it has been superseded by the A5x and likely soon by the A6. Any followers of Intel's tick-tock approach to microprocessor product development will tell you that there is too much risk in rolling out a new design on a new process technology node. The better practice is to shrink existing designs in one cycle and rollout new designs on mature process technology platforms in the alternate cycle. Apple may be flush with cash, but this is no leisure time activity. Real money needs to be spent on these tasks.
The Apple TV has always been described as a hobby, a bit of a learning experience. However, the much rumoured HDTV from Apple would certainly not be a learning experience. It would most definitely be a high volume socket for the processor that lands therein. That would certainly make a shrink worth the effort. It is also much rumoured that the Apple TV would have some smarts, Siri and maybe some sort of visual recognition. Now the latter of these would need a camera, probably HD to boot. Once again voila. The second H.264 core that is not being used in Apple TV3 would be used in a complete panel television set. So, it might be the case that this part is destined for a full TV. Time will tell.