The incredible popularity of Apple's iPhones means that it often has to diversify its supplier base in order to ensure that it has enough components to fulfill tens of millions of orders. For example, Apple might employ one or more display suppliers to deliver enough panels. In most cases, this will typically go unnoticed by users, even if there are inevitably minor differences in performance. Intriguingly, Chipworks
some iPhone 6s models to its yearly teardown of Apple's iPhone SoC package, and determined that not only is the A9 fabricated by two suppliers (TSMC and Samsung), but that Samsung's more advanced fab process has yielded an A9 chip that is 10 percent smaller than the TSMC-fabbed A9. Based on some preliminary testing that MacNN
has conducted, there is some evidence that this results in a slight performance differential between the two.
Samsung uses a 14nm FinFET process -- which it has recently been convicted of stealing from TSMC
-- to produce its current iteration of the Apple A9 SoC, whereas TSMC uses a 16nm FinFET process. Anyone familiar with the basics of fabrication processes will be aware that, even when chip architecture remains the same, the move to a smaller, more advanced, fabrication process will result in both performance and energy efficiency gains.
For example, Intel's current pattern is to create a revised architecture in one year (a "tick"), followed by a die shrink of the same architecture the following year (a "tock"). Typically, this results in varying degrees of performance and energy efficiency gains without necessitating any revisions of the underlying architecture.
The reason for the performance and efficiency gains largely has to with the fact that the smaller the die, the closer the electrons are packed together. This brings with it inherent performance advantages. Therefore, even though the difference in fabrication processes between the Samsung fabricated A9 SoC (APL08980) and the TSMC fabricated A9 SoC (APL1022) is relatively slight (and not anywhere near the scale of Intel's tick-tock patterns), there will be some inherent performance advantages in Samsung's iteration of the A9.
It is not known, however, exactly what the proportion of Samsung-fabricated A9 SoCs is in relation to the TSMC-fabricated A9 SoCs, but it is the case that these are randomly distributed through the iPhone 6s and 6s Plus production line. MacNN
has put three iPhone 6s models and one iPhone 6s Plus model through the Geekbench 3 test. By chance, there was an even split in performance of 50/50 between the four iPhones, albeit relatively minor.
As you can see from the Geekbench 3 benchmark results, the iPhone 6s unit 2 and the iPhone 6s unit 3 look to have a slight performance advantage of around three percent over the iPhone 6s unit 1 and the iPhone 6s Plus. While our sample size is clearly limited, it is nonetheless interesting to see that there could be a minor performance difference between iPhone 6s models.
The only way to be absolutely certain about the result would be to x-ray the devices to confirm which A9 SoC is in which iPhone. However, given that there is a known variation between A9 SoCs' likely performance differentials as a result, it seems plausible that, based on our results here, that the iPhone 6s units 2 and 3 are fitted with the Samsung A9 SoC. Although not tested here, it is also possible, that these models might have a slight edge in battery life as well.
Does this matter? In real world, everyday use, probably not -- both are blisteringly fast and power efficient -- which is why Apple has proceeded with this approach, even if it wasn't entirely ideal. As for whether you end up with a TSMC-fabricated A9 or Samsung-fabricated A9, that is purely luck of the draw -- but who would ever have thought that you would've ended hoping that your iPhone 6s is the one with (albeit stolen) Samsung technology in it?
- Sanjiv Sathiah