Preliminary look at Apple's new A7 processor reveals new architecture, quad-core GPU

Posted:
in iPhone edited January 2014
The team at Chipworks on Friday posted a tentative look at the iPhone 5s' A7 system-on-a-chip, finding a few changes from the previous A6 processor, including the addition of a mysterious SRAM cell never seen in A-series silicon.

A7
Source: Chipworks


Continuing its comprehensive investigation of the new A7 SoC, which on Tuesday confirmed a 28nm manufacturing process, silicon experts at Chipworks are digging deeper into the processor's major structures and overall design.

The firm notes in the report that the latest findings are currently just "best guesses," as circuit extraction testing has not yet been performed to yield conclusive data.

Taking the caveat into consideration, one of the more interesting findings would be the "secure enclave" used for Touch ID fingerprint data storage that Apple discussed during the iPhone 5s reveal earlier this month.

Chipworks was unable to identify the area in today's image, however, but guesses that a new SRAM cell located above the GPUs is a prime suspect as the rather large storage space is a new addition to the A-series lineup. Extrapolating the size of the memory block and comparing it with metrics of known technology, the firm estimates the SRAM module to have a density of about 3MB.

Moving to the CPU, the A7's dual-core unit and accompanying memory, estimated at 1MB for L2 cache and 256KB for L1, takes up about 17 percent of the die area. Located across from the CPU setup is a quad-core GPU with shared logic, which accounts for 22 percent of on-die space.

A side-by-side comparison of the A7 and last year's A6 show the chips' CPUs do not share the same layout. The new A7 appears to be closer to a conventional automated design, which would be a departure from Apple's most recent custom-architected A-series SoCs.

Aside from the above discoveries, the rest of the A7's design appears to be carried over from the A6. Of note, the USB, LCD and camera interfaces are identical to those found in Apple's A5, A6 and A6X chips.

Comments

  • Reply 1 of 12
    Where's the Image Signal Processor, the Audience (replacement) audio processor, and what's the other unlabeled half of the A7 doing?

    Kind of funny Chipwork's sleuthing is like brain experiments trying to figure out how it works, and also an archeological dig attempting to describe dinosaurs by their bone fragments.
  • Reply 2 of 12
    How much space was saved by offloading some tasks to the M7 since "the A7 SoC no longer services the accelerometer, gyroscope and compass".
  • Reply 3 of 12
    22july2013 wrote: »
    How much space was saved by offloading some tasks to the M7 since "the A7 SoC no longer services the accelerometer, gyroscope and compass".

    It's hard to say since those components aren't the same as before, but if you look at the M7 it's not very large. The reason for offloading it was likely so that devices that won't ever get something as sophisticated as the A7 can also use the M7; and iWatch. It surely doesn't appear to be about space or the 3MB for the presumed Secure Enclave may not have been there or so large.
  • Reply 4 of 12
    Quote:
    Originally Posted by 22July2013 View Post



    How much space was saved by offloading some tasks to the M7 since "the A7 SoC no longer services the accelerometer, gyroscope and compass".

     

    Not so much I think.

     

    All sensors was and are still outside of the A6/M7 chips, I believe (can be wrong here) the task was handle by software and depending on which bus was used for the CPU to communicate with the sensors no special hardware was put on the A6.  The M7 is merely a controller a lightweight and dedicated arm CPU that handle his task independently of the A7. 

  • Reply 5 of 12
    wizard69wizard69 Posts: 12,719member
    The guys at Chipworks are a bit slow if you ask me. 3MB is just the right size for a GPU frame buffer for the iPhone. If this is the case the memory array could be significant in power saving as the GPU would not need to go off chip nor trash the cache, simply to keep the screen updated.

    Of course we may not know for sure for a very long time. I understand Apples need to protect itself from the competition but this is very interesting technology.
  • Reply 6 of 12
    mstonemstone Posts: 11,510member
    Quote:

    Originally Posted by Corrections View Post



    Where's the Image Signal Processor, the Audience (replacement) audio processor, and what's the other unlabeled half of the A7 doing?



    Kind of funny Chipwork's sleuthing is like brain experiments trying to figure out how it works, and also an archeological dig attempting to describe dinosaurs by their bone fragments.

     

    Only Samsung knows.

  • Reply 7 of 12
    Quote:



    Originally Posted by wizard69 View Post



    The guys at Chipworks are a bit slow if you ask me. 3MB is just the right size for a GPU frame buffer for the iPhone. If this is the case the memory array could be significant in power saving as the GPU would not need to go off chip nor trash the cache, simply to keep the screen updated.

     

    Exactly. Fingerprint data would be slow and infrequently accessed, there's a whole lot smaller and lower power technologies for that purpose, like 1T-SRAM or eDRAM. However, FB accesses shouldn't trash the cache, the memory is simply marked uncacheable, like they do in desktop integrated graphics.

     

     

    Quote:
    Originally Posted by 22July2013 View Post



    How much space was saved by offloading some tasks to the M7 since "the A7 SoC no longer services the accelerometer, gyroscope and compass".


     


    None. The sensors in the last generation were I2C bus devices. Apple still probably needs the I2C bus for various chips, and in any case is a very simple (small) bus.

  • Reply 8 of 12
    Does anybody know the difference on graphics processing between A6X and A7, since there both quad core?
  • Reply 9 of 12
    wizard69wizard69 Posts: 12,719member
    konqerror wrote: »
    Exactly. Fingerprint data would be slow and infrequently accessed, there's a whole lot smaller and lower power technologies for that purpose, like 1T-SRAM or eDRAM.
    It would most likely be some sort of flash to survive a dead battery.
     <span style="line-height:1.4em;">However, FB accesses shouldn't trash the cache, the memory is simply marked uncacheable, like they do in desktop integrated 
    </span>
    Possibly but by not using the cache the processor has to go to ram which is a high power approach. It is a bit different than a desktop where run a frame buffer out of RAM doesn't mean much because a bigger heat sink and power supply can be used.
    <span style="line-height:1.4em;">graphics.</span>



    It is unfortunate that Apple needs to keep this design secret, we could have a very interesting discussion about what is in the chip and why. Unfortunately we are left to guess.

    My theory is that an on chip frame buffer would save a huge amount of power by lowering memory bus usage. In other words the more you stay on the silicon the better.
  • Reply 10 of 12
    wizard69wizard69 Posts: 12,719member
    Does anybody know the difference on graphics processing between A6X and A7, since there both quad core?

    Initial guesses are that the A 7 uses an entirely different GPU possibly a new Imagination core or a core tweaked by Apple. Generally performance is vastly improved if you believe initial testing. I believe Anandtech did an in depth analysis of what he thinks is in A7. There are a couple of short comings in synthetic tests but in real world app it looks awesome. The processor and the GPU in many cases are 2X faster than A6.

    As far as a comparison to A6X, that is jumping the gun a bit. Most likely there will be an A7X variant. If so that would be a worthwhile comparison.
  • Reply 11 of 12
    Quote:

    Originally Posted by wizard69 View Post


     


    It is unfortunate that Apple needs to keep this design secret, we could have a very interesting discussion about what is in the chip and why. Unfortunately we are left to guess


     

    I really don't think Apple would just come out and say: "We just spent (X) millions of dollars and (X) man-months designing our new chip... Here are all the design details. Please enjoy and discuss this among yourselves!"

  • Reply 12 of 12
    The GPU doesn't consist of 4 cores rather it Is comprised of 4 clusters. This points to a Rogue 64xx design with some possible Apple reworking of the core design.
Sign In or Register to comment.