This Technology Blog will be live updated to include some cross sectional and polysilicon shots of the Apple A5. If you are seeing this message then check back regularly or follow us on Twitter @chipworks to get up to the minute updates.
Apple TV is a curious beast. It is undeniably a good product, but one that is treated almost like an accessory, or even an appliance, by Apple. New iterations come out with no fanfare and software features are held consistent across older and newer variants. The people paying the closest attention are those who are interested in cracking the device (which has eluded the hacker market since the Apple TV 2 days)…and, of course, us.
Despite the lack of commercial attention, there has been much ballyhoo out there with the latest version to hit the store shelves with the part number MD199C/A. Again, mostly indistinguishable from the prior version, except this time with a brand new A5 processor with the part number APL7498. And again all the questions about who is the foundry, what are the differences, what is Apple up to this time around.
First up a look inside the “downstream part” because Apple has made a few notable changes in this ATV version we tore down.
What is unusual is that the device does not use the package-on-package configuration for the DRAM/App-Processor that we have come to expect for the A5 (and not the A5x). So does this mean it is some kind of A5X or a hotter running chip or is this just a lower cost way for them to manufacture the device?
The wireless module supplier for the ATV continues to be USI (versus Murata for the phones) with the part number 339S0203. The board shows a nice antenna path for those interested. Based on specifications, inside the core functionality is provided by the Broadcom BCM4334.
Notable by its absence is the lack of an Apple branded audio chip (by Cirrus Logic). Apple may be following a trend we have seen with other applications processors by including this functionality on the die. Once the polysilicon photo is available we’ll be able to look closer.
Correction – as has been pointed out by our readers, there was no discrete audio codec in earlier generations of the Apple TV, by Cirrus or anyone else – our bad for getting confused in all these Apple teardowns!
Other notable chips cataloged (so far):
- SMSC LAN8742 Ethernet Controller
- Hynix H2JTCG8T22MBR NAND Flash
- Elpida B4432BABH DRAM
- Apple 338S1127 Power Management Chip, likely another of the girls from Dialog Semi
- Texas Instruments 54218 PMIC
Let us now look closer at the Apple A5 itself
Is this new A5 a pipe-cleaner for TSMC in a lower volume (and risk) device? Is this new A5 a lower cost variant that we will see in new lower end phones from Apple? Is it a shrink from Samsung’s 32-nm process to their 28-nm? All were speculated upon, and this time around we can deny the first rumor by confirming that the chip has die markings consistent with the continued use of Samsung as the foundry partner. With respect to the second rumor, we can provide our $.02 that we do expect to see this chip in future Apple devices (such as a possible phone or iPod) because that behavior would be consistent with what they have done in the past. As for the die shrink, the math tells us that the size reduction is more than a simple shrink – some functionality has changed too. We’ll have to wait for the cross-section to see if it is fabbed on the smaller process. (Update – see below.)
The new device measures 6.1 x 6.2 mm when compared to the prior generation A5 measuring 37.8 mm2 down from the prior 32-nm variant at 69 mm2.
Now we have the cross-section, we can say from the transistor spacing that it is still Samsung’s 32-nm process – no shrink here, definitely a new design.
Update – here is our first guess at the functionality using our die photo of the new A5 chip de-layered to the transistor level.
If we compare this with the earlier 32-nm APL2498 part, it looks as though we now have a single-core ARM A9 CPU, together with a dual-core GPU as in the APL2498. The Apple TV has always been 1-core, but the APL2498 had one core disabled – maybe Apple now thinks that sales will be enough to justify a dedicated part, or maybe we are going to see another single-core device in a different product line.
To get the almost 50% die size reduction that we have, though, there has to be more than removal of one core – other changes have been made. As yet we haven’t quantified them.
We’ve also looked at more of our cross-sectional pictures, and we now think that this part uses a mixed-signal version of the 32-nm process that allows extra passive components such as resistors, capacitors, and inductors, that is much more suited to analog circuitry. So it is possible that the analog sections have been re-designed, always a work in progress when we get this small, since analog circuitry does not shrink anywhere near as predictably as digital.
We’ll keep you posted if and when we find out more!