Inside the iPhone 6s

[Daniel Nenni]

Look at the 2015 TSMC monthly revenues with YoY Change. They are all over the map:

[table] cellpadding="4" style="width: 100%"
|-
| align="center" | Jan.
| align="center" | 87,120
| align="center" | 69.4%
|-
| align="center" | Feb.
| align="center" | 62,645
| align="center" | 33.8%
|-
| align="center" | Mar.
| align="center" | 72,269
| align="center" | 44.7%
|-
| align="center" | Apr.
| align="center" | 75,330
| align="center" | 21.7%
|-
| align="center" | May
| align="center" | 70,155
| align="center" | 15.4%
|-
| align="center" | Jun.
| align="center" | 59,955
| align="center" | -0.6%
|-
| align="center" | Jul.
| align="center" | 80,953
| align="center" | 24.7%
|-
| align="center" | Aug.
| align="center" | 67,038
| align="center" | -3.2%
|-
[/table]


Where are the A9 wafers? Compare that to 2014 where you can clearly see the A8 wafers kick in:
[table] cellpadding="4" style="width: 100%"
|-
| align="center" | Jan.
| align="center" | 51,430
| align="center" | 8.4%
|-
| align="center" | Feb.
| align="center" | 46,829
| align="center" | 13.7%
|-
| align="center" | Mar.
| align="center" | 49,956
| align="center" | 13.2%
|-
| align="center" | Apr.
| align="center" | 61,887
| align="center" | 23.6%
|-
| align="center" | May
| align="center" | 60,789
| align="center" | 17.4%
|-
| align="center" | Jun.
| align="center" | 60,344
| align="center" | 11.7%
|-
| align="center" | Jul.
| align="center" | 64,925
| align="center" | 24.6%
|-
| align="center" | Aug.
| align="center" | 69,279
| align="center" | 25.8%
|-
| align="center" | Sept.
| align="center" | 74,846
| align="center" | 35.1%
|-
| align="center" | Oct.
| align="center" | 80,736
| align="center" | 55.9%
|-
| align="center" | Nov.
| align="center" | 72,275
| align="center" | 63.0%
|-
| align="center" | Dec.
| align="center" | 69,510
| align="center" | 39.9%
|-
| align="center" | Total
| align="center" | 762,806
| align="center" | 27.8%
|-
[/table]

Again, we will know the truth from the Q3 conference call.

 

[Daniel Nenni]

From IBD:

<With reports that Samsung has backed out of the A10, Arcuri sees Apple tapping Intel (NASDAQ:INTC) for a future processor.

"Intel desperately wants the Apple foundry business," he wrote. "This could be a window."


Ten nanometers won't be available for the A10, but Intel's 14-nanometer chips would offer a "meaningful improvement" in performance and power consumption over Taiwan's 16 nanometer chip and Samsung's 14 nanometer chip," Arcuri wrote.
"The Intel process is sufficiently different/better that if Apple goes with Intel as a foundry, it would result in a single source for Intel SKUs," he wrote.>

Hahahahaha......... Not gonna happen and I will bet on that one. Any takers?

 

[prime007]

Just saw an article regarding the "good" (TSMC) & the "bad" (Samsung) A9 chip.
Does your iPhone have a good or bad A9 CPU?

Apparently, the TSMC A9 lasts up to 2 hours(!) longer than the Samsung's A9. While the performance of both chips were essentially identical, it was assumed that Samsung's (smaller) chip would be more efficient. It appears the opposite is true. If correct...I'm just FLOORED that TSMC's chip can last up to 2 hours longer. Perhaps THIS is the reason why Apple choose to use TSMC's chips more than Samsung's?

I may want to download the diagnostic to see which A9 chip my wife's iPhone will have. If her iPhone has Samsung's A9...its going to be returned!

 

[khaki]

Don't want to defend Samsung or TSMC. But any benchmark just based on two samples is flawed. Any chip has a performance/leakage distribution. And by the way, the battery life has little to do with the processor power dissipation unless you play games or run benchmarks non-stop. Time will tell if there are in fact returns due to inferior battery life.

 

[astilo]

Don't want to defend Samsung or TSMC. But any benchmark just based on two samples is flawed. Any chip has a performance/leakage distribution. And by the way, the battery life has little to do with the processor power dissipation unless you play games or run benchmarks non-stop. Time will tell if there are in fact returns due to inferior battery life.

I agree. A 2 hours difference is just too huge. Anyway, if this result should be confirmed and verified on a bigger sample size, it is going to be very embarassing for Apple (I would say more than for Samsung).

 

[Daniel Nenni]

I agree. A 2 hours difference is just too huge. Anyway, if this result should be confirmed and verified on a bigger sample size, it is going to be very embarassing for Apple (I would say more than for Samsung).

If the phones are identical other than the SoC we are probably talking about a power leakage problem and that is a process level issue. Especially when the SoC that has the better power usage is also a larger die size.

It really did shock me that Apple put different SoCs in the same phone for this exact reason. Why wouldn't they split it between the 6s and 6s+? Or at least split them amongst different countries?

If true, someone is going to lose their job...

 

[Daniel Nenni]

From Forbes:

Did Apple Sell You A 'Good Or Bad' iPhone 6S?

The A9 chips used by Apple appear to offer two different levels of performance depending on the manufacturer of the component. Those manufactured by Samsung appear to be drawing more power and offering a shorter battery life than those manufactured by TSMC.

This could be a VERY disruptive story! Stay tuned for more details....

 

[kingmouf]

Dont you think that if the two chips are based on the same design, two hours is just too big of a difference for kinda similar processes (where Samsung should theoretically have an edge)? From the screen shots (https://www.reddit.com/r/iphone/com...esults_iphone_6s_samsung_14nm_vs_tsmc/cvpns61), I get to see that the TSMC phone is supposed to last almost 8 hours at the benchmark while the Samsung one a little bit more than 6 hours. Now that means that the TSMC has almost 30% more energy efficiency. Danni you suggest that this could be a leakage problem in the process, but again two hours seems like there are issues, severe ones. I am wondering, with everyone excited on the Exynos 14nm chips would this have gone unnoticed all those months if there was such as serious problem with the energy efficiency of the process? This 30% is almost generational improvement if not more!

I am also wondering about a few other things... if Samsung was earlier in the game with their 14nm process than TSMC with their 16nm one, could it be that the two chips are not identical (minus the process) and there are issues with the initial A9 design that went into production first? Could it be that the TSMC design is a respin? I am also wondering, if Apple did the work on the Samsung A9 chip and as you suggested in a previous post (https://www.semiwiki.com/forum/f296/inside-iphone-6s-6599-2.html#post24518) TSMC did the porting to the -of course- TSMC A9 chip, has it done any better job at low level stuff that reflects to such differences?

Other companies are using double sources for some of their chips, I think Qualcomm is one of them. Have you seen such big differences in the same SoCs when they are made at different manufacturers (comparable processes, e.g. 28nm bulk)? Or am I mistaken and they do not fabricate the same SoC at different fabs and they just use different fabs for different SoCs of their lineup?

 

[astilo]

https://www.semiwiki.com/forum/cont...-core-rotten.html?postid=24712#comments_24712

If we look at the official geekbench page, the average runtime value for the battery test of the iphone 6s plus is 07:32:47, quite close to the score of the phone with the TSMC chip inside and far away from the Samsung performance. But this is the average value of a huge database (based on all the benchmarks that people have done so far), not just a single and unverified test! Now, we also know, that inside the iphone 6s plus, there are more Samsung A9 chips than TSMC ones (for the iphone 6s is instead the opposite).
I think it is already clear that even assuming a 50%/50% SoC split between the two foundries, we should have expected an official battery runtime value around the 7 hours.
That´s to me already enough to conclude that the above test is not accurate at all.
Either a faulty screen, battery, or SoC, but still an exception and not the rule.
Even the test is not 100% fare, since for the samsung chip, the iphone had no sim inside.

 

[Andrei]

A Forbes article suggests that the Snapdragon 810-overheating rumors were started by Samsung.

Qualcomm PR was spreading the rumour that Samsung was spreading the rumour. I know of some people in the media that really got taken on a wild ride and Daniel is likely one of them.

Qualcomm attributed its Snapdragon 810 overheating problems to early silicon. Their actual words were "pre-production", although some OEMs went to production with those versions.

2.1 was just a band-aid fix that solved the memory controller but didn't really change the issue of power-consumption.

So QCOM moved development to Samsung for 14nm (Samsung being their largest customer) only to get hit in the face with Exynos for the Samsung S6 products. Now I'm told QCOM is back at TSMC but that could change if Apple gets another sweetheart deal at 10nm.

In regards to the 810, this is a TSMC 20nm part but it does not use the QCOM custom architecture. It uses standard ARM 64-bit cores just like Exynos. .... So yes it probably was power challenged but I was told by friends at QCOM that is was pre production silicon and has since been fixed.

They hit themselves in the face, Samsung went with pure Exynos because QC's alternatives were absolutely unacceptable. I wrote about this back in January and it ended up exactly as I reported. S810 and S808's problems are not related to the architecture or the process, it was a large fuckup in Qualcomm's physical design implementation. The problem was not fixed and you should not believe your "friends".