POST A COMMENT

24 Comments

Back to Article

  • jeffrey - Wednesday, April 28, 2010 - link

    Articles like this are outstanding. Articles providing insight into the semiconductor industry players always seem to be my favorite. Visits to Nvidia's offices, graphics architecture chats with AMD/ATI, and Intrinsity's acquisition are very much appreciated.

    I will now go click on some ads as a reward!
    Reply
  • deputc26 - Sunday, May 02, 2010 - link

    So now we know what the "A4" is. A "fastmathed" A8, but on what node? I've heard both 45 nm and 65nm... Reply
  • ganeshts - Sunday, May 02, 2010 - link

    It is most definitely 45nm! I think ChipWorks already has the A4 decapped (XRayed and compared with a Xilinx 45nm chip) to confirm that. Reply
  • greylica - Wednesday, April 28, 2010 - link

    Its' very interesting that the industry change it's own about performance. Now, we are in the times of ''performance per watt'' , but it seems that the progress in microprocessor architeture is making the industry and the consumer loose it's way to understand what is REAL performance. Proprietary Instruction set optimizations, and performance diferences between processors that use the same Instruction set are getting to a point that we all start to get confused. BUMP.... Reply
  • dagamer34 - Wednesday, April 28, 2010 - link

    Anand, this part doesn't make much sense: "On the other hand, if the aim was to prevent the competition from getting access to this technology, it may succeed to quite an extent."

    Just about everyone is already working on Cortex A9 designs, with most of them shipping this year. Playing keep away with souped up A8 technology is nice, but I doubt anyone would really bother with it now that nVidia is offering it's Tegra 2 series of chips for OEMs to use by the end of the year.
    Reply
  • ltcommanderdata - Wednesday, April 28, 2010 - link

    I believe they are referring to the domino logic technology which will in theory allow future Apple versions of the ARM Cortex A9 to scale to higher clock speeds and be faster than regular reference ARM Cortex A9 designs. Reply
  • ganeshts - Wednesday, April 28, 2010 - link

    dagamer34, Thanks for your observation. Maybe, the context wasn't clear in the article.

    My intent was to suggest that Apple might not want other companies like TI / Nvidia / Broadcom etc. to be able to utilize Intrinsity's Fast14 technology (not Cortex-A8s or Cortex-A9s in general).

    Fast14 could potentially speed up the implementations of the core CPU in app processors which could be seen as a competition to Apple's own A4 lineup. (if the companies desigining them were to license the FastCore versions)
    Reply
  • faydrus - Thursday, April 29, 2010 - link

    Does domino logic really matter anyway? The Intel Nehalem processor is pure static CMOS, no more domino logic there. If a high performance processor has no use for domino logic, why would an embedded processor want it? Reply
  • metafor - Thursday, April 29, 2010 - link

    That doesn't necessarily mean domino logic isn't useful anymore. Nehalem had simply gotten to a point of complexity where hand-designing domino circuits (especially since they moved to an internal IP-based chip assembly flow) was too cumbersome considering the benefits (and I suspect their 32nm node was a bit too variable).

    In much smaller processors like the A9, it could indeed benefit. Now, there are significant trade-offs to domino logic -- area and power will suffer -- but if strategically used, it can definitely give a competitive edge in A9 class processors.

    That's not to say Apple isn't micro-architecting their own ARM processor anyway. They aren't hiring a bunch of chip micro-architects for no reason :)
    Reply
  • Mike1111 - Wednesday, April 28, 2010 - link

    I only heard that some of PA Semi's engineers left Apple, what's the source for "MOST of the PA Semi engineers have since moved on"? And I wouldn't be a 100% sure that Apple is gonna stick with Samsung for 32nm/28nm, GlobalFoundries would be a good alternative. Reply
  • ganeshts - Wednesday, April 28, 2010 - link

    @Mike1111, Thanks for your inputs.

    When I mentioned that 'MOST' of the PA Semi engineers had moved on, I referred to the top brass consisting of PA Semi's CEO / CTO / COO etc. A quick Google search reveals that all these guys had indeed moved on to Agnilux. As for the rest of the PA Semi employees who stayed on at Apple, I am sure they are no different from any other Silicon Valley VLSI engineers that Apple could hire since Apple doesn't seem to be using any IP / specific knowhow of PA Semi in their products.

    As for GlobalFoundries, yes, Apple could always shift foundries, but the acquisition of Intrinsity lessens this probability, in my opinion. Considering that Intrinsity engineers are very familiar with the Samsung process, it would probably increase development time if they were to shift to another vendor. Note that Intrinsity's technology is closely tied to the process and foundry. So, it is unlikely that Apple would take the risk of schedule slip. Of course, this is just a conjecture, and stranger things have happened in the industry. If I have left off some specific angle which would make Apple shift foundry allegiance, please do let us know :) I am very open to corrections.
    Reply
  • MySchizoBuddy - Thursday, April 29, 2010 - link

    Correction. Apple isn't using any PA.Semi specific IP "that we know off". Apple is too secretive to draw conclusion that PA.Semi was a lost acquisition. Reply
  • ganeshts - Thursday, April 29, 2010 - link

    Well, this is the same thing that people were talking about when no Raycer Graphics products appeared under Apple's name even after 3 - 4 years of the acquisition.

    Many professional analysts ( and I am not one :) ) have long since beaten this to death, and I will put it down here again : "PA Semi's IP at acquisition time was a 25W ARM processor based chip. There is no product line that Apple currently has, which could be using this type of chip. If Apple indeed had planned on using PA Semi IP of some sort in any of their products, the best bet would have been in the iPad -- of which, rumors abounded towards the beginning of this year. This didn't turn out to be true. There has been no tangible benefit (except for engineering resources) to Apple from the PA Semi purchase yet -- even after 2 years of the acquisition"

    I don't find anything which would make me disagree with the analysts' conclusions.
    Reply
  • Mike1111 - Thursday, April 29, 2010 - link

    If Apple is sticking with Samsung, any information out there what's Samsung's LP roadmap for 28nm (if they do half-node) and 22nm/20nm looks like? I couldn't find anything concrete (only that 32nm and 28nm will be rolled out this year - is that HP(L) or LP, risk or volume?). GlobalFoundries seems to be on an aggressive schedule, as is TSMC (although their problems with 40nm make me skeptical that they can really deliver 20LP in volume production in the third quarter of 2013 as announced).

    As for PA Semi: If Apple is really trying to develop its own CPU core (like Scorpion), then it's no wonder that they weren't involved in the A4. A custom ARMv7 architecture won't find its way into products until 2012 at the earliest. 2012 would also be the year when I expect a dual-core Cortex-A9 (or similar custom architecture) to come to the iPhone, so that would fit (because I don't think dual-cores make much sense for smartphones before available in 32nm/28nm).
    Reply
  • ganeshts - Thursday, April 29, 2010 - link

    Except for TSMC's public roadmap (which, in my opinion, is something to save face while their 40nm fiasco continues), foundries usually put their roadmaps under NDA. So, it is not surprising that we are unable to find out information about Samsung's plan beyond 28nm.

    GlobalFoundries looks aggressive and acts aggressive because they want to attract customers (Remember that they are starting out completely new!).

    @ Mike1111, Yes, I agree with you on PA Semi possibly working on ARMv7 architecture (assuming Apple does have the architecture license). This custom architecture could be further accelerated using Intrinsity's technology too. This is exactly what Tom Halfhill suggests in Microprocessor Report dated April 26th, 2010.

    Note that Qualcomm licensed ARMv7 in 2005, and it took them till 2009 end to start shipping products based on this. If one assumes PA Semi team started this in 2008, it would take them the timeframe that you mention to get designed into a working phone or tablet. However, loss of the top guys from PA Semi would have definitely pushed this behind schedule. As I noted in the article, Intrinsity tech, if applied to this custom architecture, is only bound to delay this further.

    Just look at how the A4 Hummingbird designed in 6 - 8 months by Intrinsity seems to be better than the Snapdragon Scorpion which took more than 4 years! When Apple comes out with their version, I am sure a stand alone processor from ARM itself (like the off-the-shelf Cortex-A9's successor) would have better performance; That way, I am not sure Apple would be taking full advantage of Intrinsity's tech.
    Reply
  • Mike1111 - Thursday, April 29, 2010 - link

    Thanks for the response!

    Regarding Samsung's foundry business: I couldn't even find detailed information on Samsung's 32nm and 28nm process, not just the stuff beyond 28nm. Is that information out there? I would be really interested in knowing when the 28LP volume production is scheduled to start. Early enough for the 2012 iPhone (AFAIK that would mean Q3/2011 or earlier)?

    Halfhill's Intrinsity article sounds good, but since I don't have a subscription I can't really access it. But in the part I could access he says that Intrinsity has already put more than 1 year into a Fast14 implementation of ARM’s Cortex-A9 dual-core processor. Interesting.
    Reply
  • ganeshts - Thursday, April 29, 2010 - link

    Mike, an article appeared in ElectronicsWeekly just today:

    http://www.electronicsweekly.com/Articles/2010/04/...

    Summary from the above link: Looks like Samsung will offer 32, 28 and 22nm ; Q3 2010 is when 32nm volume production might start ; No hard dates provided for other nodes.
    Reply
  • Mike1111 - Thursday, April 29, 2010 - link

    Thanks for the link!

    Q3/2010 for 32LP volume production is very good. Just risk production or 32HP in Q3/2010 wouldn't have suprised me.
    That means that a 32nm SoC in 2011's iPhone could be theoretically possible. Although, going from 65nm to 32nm in 2 years in a smartphone sounds too good to be true. And if Apple wants to continue to use the same SoC in the iPhone (Q2/2010) as in the iPad (Q1), it gets even more unlikely.
    Reply
  • Mike1111 - Thursday, April 29, 2010 - link

    Correction: And if Apple wants to continue to use the same SoC in the iPhone (Q2/2011) as in the iPad (Q1/2011), it gets even more unlikely. Reply
  • Mat3zz - Friday, April 30, 2010 - link

    This article seems to imply that ATI used Fast14 in their GPU's. I'm pretty sure they never did. When ATI released the R600 series in 2007, it was asked of Eric Demers in a Beyond3D interview about Fast14 and he made it clear it wasn't in their product. Radeons haven't increased in clock speed much since then. If ATI paid Intrinsity, it was probably for nothing. Where's my Ghz GPU!? Reply
  • ganeshts - Friday, April 30, 2010 - link

    Thanks for your comment and bringing the interview to my attention.

    When ATI first struck a deal with Intrinsity, analysts predicted that Fast14 would be used for ATI GPUs. Unfortunately, it never came to pass. It is possible that Fast14 technology is not amenable to GPU designs but more to conventional microprocessor architectures, but ATI realized this only later.

    It would be interesting to find out the duration of the licensing agreement that ATI stuck with Intrinsity, because this would actually give light on whether ATI would get affected at all by this purchase in any way. AMD / ATI could still use this technology in future process nodes for some other products, if they still ahve a valid license.
    Reply
  • toyotabedzrock - Saturday, May 08, 2010 - link

    Did you just suggest that Nvidia would use TSMC's 28nm node right after it came out? Reply
  • ven - Friday, September 23, 2011 - link

    Iam a Novice in hardware, so if my question seems silly then don't consider it.

    ARM provides architectural design details and fab company's like samsung or TSMC will implement that in the silicon for their customers like (Apple),then what role does apple do in their A5 or A4 chip?. . Just printing their Apple White Logo on top of the chip package?..

    so,please someone clarify my doubt.
    Reply
  • MAKEMESMILE - Sunday, May 13, 2012 - link

    I am currently writing an "essay" about Apple's acquisitions and couldn't find an article explaining the whole story of Intrinsity's purchase.
    This text is amazing. It gives an holistic view of the acquisition, provides accurate informations and some tracks for further digging.

    Thanks for sharing ! You made my day.
    Reply

Log in

Don't have an account? Sign up now