256 MB of RAM. Same as iPhone 3GS.

    This is the SDRAM inside the A4. 
    Yes, that's a Samsung logo. No, 
    that doesn't mean Samsung designed 
    the A4—just the RAM.*

http://s1.guide-images.ifixit.com/igi/i41Jh6tuFo6RFBVE.large

There's also a great side-cutaway shot, showing the 2 RAM dies.

http://www.ifixit.com/Misc/iphone_processor_crossection.jpg

Wow only 256MB RAM? Anyone else surprised?

Not at all. This thing runs a slim Unix-based embedded OS. Why would it need more memory?

I use a netbook with 1.5 gigs and a non-slim Unix-like OS with X and Gnome on top of it and it rarely hits swap.

this doesn't actually tell you anything about the processor, just that it's in the package, under two ram dies, and that it's single core "so must be an a8".

It tells you more than that: it's not an A9 and does not contain a GPU. So doesn't it dispel some rumours? What did you expect?

No, it just asserts that because there's only one core it has to be an A8 (superscalar, no provision for SMP) not an A9 (out-of-order speculative issue superscalar, single core and 1-4 cores SMP versions).

well as far as i understand things, an arm chip is a collection of modules that perform different functions. so it could have described which modules are present (you'd need to grind horizontal sections).

also, although i guess more difficult, it could perhaps have tried to look at what makes this chip different. arm chips seem to be pretty common; what was so special about this that apple needed to do their own?

I'd love to see a full layer-by-layer teardown of the actual die. Multi-die packages are nothing special. Getting past metal8, even it it means going straight to silicon, would let you pick out individual blocks.

Why are wafers round? Is there a part of the process that requires spinning the wafer?

I believe its because the wafers are actually sliced of a single large crystal of silicon, which grows in a circular cross-section.

http://en.wikipedia.org/wiki/Wafer_(electronics)

Yes, wafers are spun during processing. Saw a video long time ago. Liquid photoresist is dropped on the wafer. It's then spun to even out the layer.

I bet reverse engineering is against the license agreement.

anyone know how much this process costs?

Enough that they didn't pay to have the metal layers "peeled off" one by one (or just all of them as Arubis suggests) so that we could start to guess what it really is....

The consumables involved in stripping off metal and oxide layers have a negligible price point. Most of the cost is engineer time, though how much depends on what equipment/techniques are used (liquid chemistry, manual polishing, plasma processing, etc.).

This isn't the case with all circuit forensics techniques: once you start loading parts into more expensive equipment (even electron microscopes) you're paying for tool time to offset training and tool value depreciation.

I know that's not really quantifying things, but that's something you'd have to get from the lab that actually performed the work.

EDIT: if it helps, the actual process of removing a metal layer and an oxide layer varies with the process, but would generally take under an hour (and usually well under). There exist techniques to skip all that and go straight to silicon significantly faster, though as these are all destructive processes, you lose the ability to go back and see what was there before.

I would not be surprised if Chipworks wasn't paid for this article. It might have been a PR project. At work we sometime do this to get our company in the news.