With the exception of law enforcement, I am not aware of anyone storing a full finger print image for verification. All that would be stored would be a short term image and 3 megabytes seems rather large.
For security, all that should be stored is a fingerprint signature. This is more than adequate for verification and removing false positives, but makes it pretty much impossible to recreate the fingerprint.
Apple patented the fingerdata verification techinique for mobiles in very recently infact in 2012 they filed for the patent. I do not know where they store this information but 3MB looks rather too much for just few fingerprint.
Looking at the layout, Apple appears to have used 1 DDR channel to feed the CPU and another channel for GPU data. The large SRAM above the GPU appears to be the embedded cache for graphics memory, which is similar to approach used by Intel's IRIS Pro and XBOX One
To store finger print data? You need to use NVM, not SRAM or you have to keep feeding power to the chip, which is not possible
Also for the CPU highlighted by chipworks, it appears that the complex includes 2 memory controllers from layout, so the actual area for the CPU complex (excluding the memory controller) is likely smaller than 17% or ~17 sq mm
It makes more sense to me this large on-chip RAM is L-3 Cache for DRAM controller directly next to it. If I am Apple, I will never store direct finger print data in anywhere. Because wherever you store, people will figure out. A good choice is to apply special imaging hashing function to it and only store hashing value. Not sure this technology exist today.
I doubt google will jump in at this level any time soon. they've typically partnered with a hardware vendor for this kind of stuff, like microsoft. I do know they have custom hardware made for their datacenters though.
Who knows with google, they don't tend to follow a narrow path forward, so they might decide to experiment with SoC development of their own.
Soc wouldn't be a good investment for Google, for that amount of money (few hundred millions at least) they can achieve amazing feasts in many areas, while a soc gets only a small relative and temporary advantage.
And anyways, they managed to get unique features in their soc in the motox phone by working with others, so it's seems good enough.
That would be madness IMO. Google should focus on what they are good at (software, search technology) and partner with others in other areas. Buying Motorola Mobile was a mistake in my opinion, so to go for the SoC design business would be madness for me....
But they bought MMI for the patents more so than for the hardware. Still, now that they find themselves in the hardware business, an SoC should not be automatically ruled out. There is always a cost/benefit analysis to be done, product differentiation, etc. It really depends on Google's strategic plan with that business, and how the relationship with its Android partners factors into that plan.
Apple and tsmc rescue Apple iPhone 5S fingerprint identification chip.
The article was published in DigiTimes on August 12, 2013 in Chinese and the partial abstract is translated in English and listed below for your reference:
1) Fingerprint ID chip production was not ready until July, 2013 by assembly issues in Xintec. The original mass production schedule of the ID chip was in May, 2013.
2) Apple sent technical people to Taiwan and work with tsmc engineers to solve assembly and test issues in Xintec (wholly own OSAT by tsmc).
3) Apple iPhone 5S fingerprint ID chip gets 30,000 8" wafer per month capacity from tsmc.
4) The schedule delay may be related to a) the incompatibility between IOS7 software and ID chip hardware; b) ID chip is interfered by Sapphire blue paint and c) the new assembly methodology that results in very low yield.
5) Both Apple and tsmc have confidence that the low yield problem will be solved and the mass production will be started by the end of August.
This may imply that the 3MB SRAM is not for storing fingerprint data and the separated fingerprint ID chip may have EEPROM on 8" wafer.
Fully in line...look for NVM not for SRAM.
so rather look into M7 (which has both large quantity of embedded flash, and encryption capability).
PS: it would be worth to have floor plan analysis of M7 as well, and get evaluation of technology used and embedded flash size.
Blog Doing Math in FPGAs Tom Burke 24 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...