During the Apple (AAPL) product introduction last Tuesday, Tim cook mentioned that the heart of the new iPhone 5S was the A7 chip, which contained a billion transistors. The size of the chip grew from 96 sq. mm for the A6 to 102 sq. mm for the A7.
Experts at Chipworks who claim to be familiar with the Apple "A" chip part numbering system suggest that the early peeks at the A7 chip indicate that it was made at a foundry other than Samsung (SSNLF.PK).
Also, Samsung introduced its new Octa chip on September 10, apparently to steal some thunder from Apple's A7 chip announcement. Is that a smart thing to do if you are the foundry for the A7?
It appears that to get 1 billion transistors on a 102 sq. mm chip would require something much better than the 28nm processes available at either TSMC (TSM) or Samsung. Some are speculating that the A7 is being built by TSMC on a 20 nm planar process. That would be all well and fine except that TSMC claims that its 20nm process will not be ready for volume production until early 2014.
What do we know about relative chip size from 32nm to 22nm processes?
A Sandy Bridge EP-4 is made on a 32nm process and had 1.27 billion transistors with a chip size of 294 sq. mm, so a 1 billion transistor logic chip on 32nm should be around 232sq. mm. Even giving a 25% improvement for a 32 to 28nm transition, the chip size should be about 174 sq. mm.
The Ivy Bridge HE-4 is made on the Intel (INTC) 22nm Trigate process and has 1.4 billion transistors with a chip size of 160 sq. mm, so a billion transistor logic chip on 22nm should be around 114 sq. mm. "Hand packing" the A7 design might give another 10% size improvement, which would make the A7 chip almost exactly 102 sq. mm. on the Intel 22 nm process.
I will leave it for you to come to your own conclusions, but if there is another semiconductor manufacturer that is capable of the A7 volume at anything close to 22nm, Intel and I are in deep yogurt.
If the size of the chip grew only about 7% (96 to 102), that would easily be accounted for by the reduction in process gate lengths from 32nm to 28nm while allowing for a small increase in transistor count. Your area per transistor is approximately 75% (SQR(28/32)) of what it used to be so you can pack in 30% more transistors in the same area. If you follow this guy's logic, the previous chip must have been produced at about 24nm. It seems Intel's chips are not packed as densely as Apple's.
Interesting comparison, but there are so many good reasons why Intel should not make an ARM chip for Apple or anyone else. The one slight possibility I might consider is if Apple already agreed to migrate to x86 and a 64 bit A7 is only an interim solution.