The 3G Baseband Conundrum

So we managed to take about 2500 words to talk about a bunch of aspects of the 3G iPhone that nobody really focused on previously. So surely now we’ll take about 25000 for the baseband part of the equation, which has been by far the most discussed in the industry, right? Well, we’d love to, but sadly there’s not just enough information to write anything of that magnitude. Either way, let’s quickly look at what we do know and the different alternatives.

The most likely possibility by far is that Apple stuck to Infineon, as indicated by the ‘SGOLD3’ string present in a recent firmware version and a wide number of rumours and speculative tidbits. It is worth pointing out, however, that this only specifies the generation: not the standard (there’s a 2G SGOLD3 and that’s a likely candidate for any iPhone Nano) or the specific product. The only publicly known chip from this generation that provides a 3G baseband, the PMB8878, also embeds a primitive application processor that would do little but increase costs and perhaps even power consumption slightly (depending on the existence of power islands) in Apple’s case.

We therefore believe that a custom chip from Infineon is just as plausible as the PMB8878. However, Infineon isn’t the only possibility. In rough order of likelihood, there’s Icera, Qualcomm, Qualcomm, Broadcom and InterDigital. Marvell is incredibly unlikely: they have little prior expertise in 3G Basebands and no market penetration outside of Research in Motion. As for InterDigital, it doesn’t make a lot of sense for Apple to go straight to them instead of Infineon which uses their IP.

Icera is a very interesting possibility, however. They currently share the performance leadership with Infineon/InterDigital, with Qualcomm following closely behind according to third-party studies. Their architecture is unique and highly interesting (we’ve got an article in the wings examining it, by the way, so we won’t focus on it now) and overall they are very competitive in terms of power consumption and seem to have one of the smallest if not the smallest die size and cost in the industry.

A variety of anecdotal evidence is what makes Icera plausible. First of all, they are the only ones with a significant new product (a shrink of their previous 90nm chip along with a new RF chip) between June 2007 and June 2008, which are the release dates of the 2G iPhone & 3G respectively. Secondly, the operators interested in or using their gear coincide strangely with some of Apple’s key partners: AT&T, Orange and Softbank Mobile.

The latter was Icera’s first major customer and has recently announced that they would sell the 3G iPhone in Japan; given the apparent lack of relative interest in the iPhone in Japan, it would make sense that only a carrier that has already certified the baseband would be fully willing to sell the product. One major negative, it would seem, is that Icera only announced their 65nm chip in early May. However, when examined further, this is actually a positive.

Icera publicly claimed that their 65nm samples would be back from the fab in late 2007. Clearly this late announcement would imply that’s not the case, right? Wrong: one of their product briefs has a clear shot of their chip’s package, and it says: ‘074110’. The 07 stands for 2007, while the 41 for the 41st Week of the year which is in late October. That date likely corresponds very closely to the time they got chips back from TSMC. So why was there no announcement? One theory is that they needed a respin or had a variety of other technical difficulties. Another is that they had a lead customer willing to pay to have it first.

While we do not believe Icera is quite as likely as Infineon, it does seem sufficiently plausible to us that investing in Infineon or InterDigital as a way to capitalise on the 3G iPhone's success becomes quite unattractive. And of course, Qualcomm or Broadcom are also not completely out of the question, although Apple has never dealt with the former and the latter would likely imply a custom chip and possibly lower performance. Furthermore, Qualcomm's process advantage (they’re basically the first to 45nm) wouldn’t help here because there is no evidence that those parts have started mass production yet.

As a last note, a recent leak from engadget claims that the iPhone 3G uses an "Infineon PMB6952 / S-GOLD3 six-band UMTS / HSDPA transceiver". Assuming Infineon's PMB6952 is indeed used, which is actually a WCDMA/EDGE RF chip, it’s extremely unlikely for Broadcom or Qualcomm to have the baseband design slot. On the other hand, Icera has relationships with both Infineon and Skyworks, so they’re not out of the picture even then but it does make them less likely. It should finally be pointed out that yet another possibility is for the PMB6952/SGOLD-3 combo to actually be for a new '2G' model.