A9 Is TSMC 16nm FinFET and Samsung Fabbed
chipworks.comFTA:
> From a benchmarking point of view, the smaller die size shows a leadership in technology scaling for Samsung.
Almost certainly true.
> On the other hand, for Apple to go through all the trouble of dual-sourcing a custom designed part and launching on day one with both parts, suggests major sourcing problems.
Well... maybe. Or alternatively this is Apple throwing their cash around for long term leverage. They certainly can pay to have the initial design done on two different processes. And having done so, it puts TSMC and Samsung (and GF, who shares Samsung's process and would be an obvious third source) into a terrible bind: Apple can squeeze them with the constant (and credible!) threat of flight to a competitor.
See previous coverage of NVIDIA's rocky relationship with TSMC for an example of something Apple is probably trying to avoid.
Another way to look at it is that using multiple suppliers and multiple designs allows them to shorten the production time window before launch. That allows them to use more time for design while curtailing their suppliers' ability to leak info about the parts.
Another alternative is that Apple wants both foundries to stay reasonably healthy.
What implications will this have on performance? Will one phone with TSMC's Chip out/under perform the other phone with Samsung's Chip?
It shouldn't have any implications for the customer, performance/power consumption/reliability are much more defined during the design phase than the fabrication phase.
The end user shouldn't worry, or even think, in what foundry was the chip fabricated in because it doesn't really matter if you are not the engineering team.
Manufacturing process definitely impacts transistor performance including critical things like leakage and speed/voltage trade-offs. I'd be quite surprised if benchmarks that included power consumption couldn't distinguish between the two sources. As to whether it's going to be noticeable in practice - who knows, but the fact that apple's risking it suggests they're confident enough.
Similarly, I'm not sure how you expect reliability not to be affected by process; of course it'll be. But how often have you heard of a chip failing after it passes initial validation (something presumably apple does have a hand in)? It's not going to matter.
Just a note before anything else: it's not a "manufacturing process". Nobody is laying out the transistor by hand, it's a fabrication process. (I apologize for nitpicking in the wording, but it's important)
Apple does these kind of things for two main reasons: Because they have the money and because they want to test both fabrication processes for future products
The differences between a 14nm process and a 16nm process are quite minimal mostly because one process can offer some advantages over the other one. For example: it's expected that the smaller process has bigger leakage current, increasing power consumption, while it's expected for the bigger process to produce more heat.
In the end, you could say that if you sum the advantages and disadvantages of both, you will not reach any conclusion if you are not the engineering team looking for extreme optimization and with Apple's resources at your disposal.
It's just hard to come to a concrete conclusion from a consumer's point of view.
It's not just 14nm vs. 16nm though - two 16nm processes can still have significant differences due to materials
And even "16nm" something of a marketing term - something tiny feature in there is at 16nm resolution, but lots of other things are larger. And which things are how large, of course, also matters...
You can't have one process consume more power and the other process produce more heat. Power is converted to heat at 100% efficiency so more heat must mean more power.
That's... sort of exactly wrong. The process design rules (and transistor models) define the envelope within which the design must exist. The days of simple scalable design rules that port between fabs are long gone. Every process has its own crazy rules about how to draw a transistor.
Now... it's true that all the media reporting so far seems to treat TSMC's 14nm and Samsung/GlobalFoundries's 16nm processes as "essentially the same from a design perspective" (though both seem to lag Intel's 14nm in density). But until parts reach the market we won't actually know.
I didn't say that the fabrication process doesn't influence the characteristics of the final chip, I was stating that the design itself takes the bigger role of the chip's optimization and that from the end user's point of view, it doesn't matter if the A9 is fabricated in TSMC or in Samsung's fabs.
Actually, if they are dual sourcing the same chip from a 14nm process and from a 16nm process as well, it's very likely that they had to use different analog designs, so the 14nm chip probably has advantages that the 16nm chip has and vice-versa.
It's pointless for the end user to nitpick differences of the same chip in 2 equivalent processes when the software is going to mask everything out.
And I'm not saying this because I read it in the media, I affirm this from experience in the semiconductor industry.
At the very least it's going to affect power consumption. I agree that they probably went for iso-performance.
Analogy: if you build one thing out of steel and one thing out of aluminum, you can get the same weight or the same strength, but not both.
All manufacturing processes have a certain amount of variability. Binning is the process of testing how fast a given chip can go and sorting them into categories based on that and possibly selling them for different prices based on the category. Apple generally has very few categories but maybe they've got Apple TV rated A9s, iPad mini A9s, and iPhone A9s. There's probably a difference in how many of each manufacturer's chips end up in each bin but unless you're into overclocking it's not something you'll ever have to worry about as a consumer. And I've never heard of someone overclocking an iPhone.
I would guess that Apple has been careful to keep performance the same.
I'm guessing the frequencies (performance) are the same, but it's possible the Samsung process is able to run the same frequencies at lower voltages.