Android is winning… Is it?

  • Posted: 07 April 2011 08:50 AM #46

    Analyst: Apple qualifying TSMC

    To me it looks like more FUD could some chip expert tell us ?

    Some are doubtful of Apple Ax processor technology.. way behind Apple may decide to buy from somebody else

    http://www.eetimes.com/electronics-news/4214849/Analyst—Apple-qualifying-TSMC?

    ‘‘Apple’s success in iPad/iPhone products is obviously not based solely on its internal Ax series apps processors; however, from a purely semiconductor perspective, we believe the long-term viability of the Ax series requires a significant redesign or the use of a merchant apps processor vendor (why bother if it is cheaper and better to buy it from somebody else),’’ he said.

    ‘‘A more interesting dynamic for us as we look at structural issues in the apps processor world is the massive size of the Apple A5 dual core coming in at approximately 122-mm(square) (45-nm at Samsung). The A5 is a honker of a chip and compares to the A4 at 53 mm(square) (45-nm at Samsung),’’ he said.

    ‘‘In comparison, Nvidia’s Tegra2 dual core is a svelte 49mm2 (40-nm at TSMC). We strongly suspect the upcoming Tegra3 quad core this fall will be smaller than the A5, with industry observers pegging its size at 80mm2 (40-nm at TSMC),’’ he said.

    ‘‘Notwithstanding how the A5 is going to fit into the upcoming iPhone5 (run it at very low speeds to conserve battery life), from a semiconductor perspective, Apple is falling behind?way behind,’’ he said. ‘‘The A5 is within striking distance in terms of size to Intel’s dual core Sandy Bridge at 140mm2, which addresses mainstream PC and server applications.’‘

         
  • Avatar

    Posted: 07 April 2011 11:14 AM #47

    Hamourabi - 07 April 2011 11:50 AM

    Analyst: Apple qualifying TSMC

    To me it looks like more FUD could some chip expert tell us ?

    Some are doubtful of Apple Ax processor technology.. way behind Apple may decide to buy from somebody else

    http://www.eetimes.com/electronics-news/4214849/Analyst—Apple-qualifying-TSMC?
    ...

    Here is an interesting reply: Apple, Nvidia: A5 Chip Has Reason To Be Big, Says RBS

    One reason for the discrepancy is that Apple?s chip was built in a 45-nanometer process by Samsung (SSNLF), while Nvidia used Taiwan Semiconductor?s (TSM) 40-nanometer process, so features are larger on the Apple chip overall.
    ...
    This was, then, a deliberate move, Scemama thinks, on Apple?s part, to make a larger chip, in order to have substantial graphics performance. And here, he cites OpenGL benchmark results that suggest, ?that the A5 beat Tegra2 handsomely in third party tests.?

    All that means that Nvidia is somewhat hamstrung, he thinks, by the need to keep chips in the $15 to $20 range for customers, while Apple can pay Samsung $25 per chip.

         
  • Avatar

    Posted: 07 April 2011 12:30 PM #48

    Hamourabi - 07 April 2011 11:50 AM

    Analyst: Apple qualifying TSMC

    To me it looks like more FUD could some chip expert tell us ?

    Some are doubtful of Apple Ax processor technology.. way behind Apple may decide to buy from somebody else

    http://www.eetimes.com/electronics-news/4214849/Analyst—Apple-qualifying-TSMC?

    ‘‘Apple’s success in iPad/iPhone products is obviously not based solely on its internal Ax series apps processors; however, from a purely semiconductor perspective, we believe the long-term viability of the Ax series requires a significant redesign or the use of a merchant apps processor vendor (why bother if it is cheaper and better to buy it from somebody else),’’ he said.

    ‘‘A more interesting dynamic for us as we look at structural issues in the apps processor world is the massive size of the Apple A5 dual core coming in at approximately 122-mm(square) (45-nm at Samsung). The A5 is a honker of a chip and compares to the A4 at 53 mm(square) (45-nm at Samsung),’’ he said.

    ‘‘In comparison, Nvidia’s Tegra2 dual core is a svelte 49mm2 (40-nm at TSMC). We strongly suspect the upcoming Tegra3 quad core this fall will be smaller than the A5, with industry observers pegging its size at 80mm2 (40-nm at TSMC),’’ he said.

    ‘‘Notwithstanding how the A5 is going to fit into the upcoming iPhone5 (run it at very low speeds to conserve battery life), from a semiconductor perspective, Apple is falling behind?way behind,’’ he said. ‘‘The A5 is within striking distance in terms of size to Intel’s dual core Sandy Bridge at 140mm2, which addresses mainstream PC and server applications.’‘

    I think this analysis is very superficial and may miss out on some of the advantages of an in-house design.  SJ claimed Apple was first to scale a dual core in a commercial design with the A5.  This could be debated since folks were delivering Nvidia Tegra 2 designs but probably not in the quantity of the iPad 2.  The jump from 53mm2 to 122mm2 is a big jump but the chip dimensions will still allow Apple to cram the design into a new iPhone 5.  As far as what they are doing with the extra transistors.  The only die photo I’ve seen are from Chipworks and unfortunately reading a die is not something I do for a living.  From what I’ve read there are two major differences between Apple’s design and the stock ARM A9.  The first is a 64bit memory bus vs the 32bit bus in the Tegra.  It also appears Apple has implemented Dynamic Voltage and Frequency Scaling(DVFS).  This improvement requires additional transistors to implement fine grain clock gating.  Which allows for shutting off portions of the chip at a much finer grain level.  Resulting in power savings or improved battery life.  Additionally Apple has added a second GPU when they went from the Power VR 535 to the Power VR 543MP.  From the benchmarks I’ve seen the A5 outperforms the Tegra 2 design and is more power efficient.  That pretty much is the cat’s meow from an engineering perspective as long as it fits in your design.  The cost thing is misrepresented.  Not sure of the contract deal Apple has with the Samsung Foundry.  But lets say they pay $5000 for each good wafer.  The wafer is processed and then cut into multiple dice.  The A5 is 12.1mm x 10.1mm and should yield about 480 dice per wafer at a cost of 10.41.  The Tegra 2 would yield about 1240 dice for a cost of 4.03.  The dice are packaged in POP for Apple with a memory stacked on top of the SOC.  Nvidia also offers a POP configuration as well as a design where the memory is separated from the SOC. I would assume Nvidia wants to recover the cost of their design work so the are not charging their customers their raw cost.  An analysis I saw in reference to TI 5430 core estimated there is $10M of IP costs in their design and then you have to pay your design team which might run another 30-50M per year.  Depending on the number of units and wether the company is packaging themselves or having Nvidia subcontract the cost per unit will vary but in the overall BOM the larger amount of silicon is costing Apple $ 5 and the design cost is spread across the whole iOS product line so cost per unit is pretty small if we assume 100M units and 60M design cost.

    As far as qualifying TSMC as a suppier, my guess if true it would be for the A6 which I hope is a 28NM design.  It is not a simple process to switch foundries.  Samsung is using gate first and TSMC gate last   Integrating high-k /metal gates: gate-first or gate-last?

    It will be interesting to see if Apple can tie up TSMC wafer production at 28NM by buying up the capacity and preventing the likes of Nvidia or Qualcomm from ramping their design faster then Apple.  Nvidia’s Tegra 3 is a 40NM design so the chip will be larger then Tegra 2 as they add more transistors for the additional cores.  If Apple goes to 28NM in 2012 they can either shrink the die or add more cores or maybe some of both and maintain their lead in 2012.  They also have the possibility of switching to the next generation IP from ARM the A15 and Imagination the PowerVR 6 series Rogue.  For those looking for technical details on the next generation Application processor excluding Apple’s design.