"So, the Enterprise has had its maiden voyage, has it? She is one well-endowed lady! I'd like to get my hands on her ample nacelles if you'll pardon the engineering parlance." - Montgomery Scott, Star Trek
At Intel's (INTC) investor meeting, the company finally bit the bullet and admitted that the 6331 part - one that I had reported had been killed - was dead as a doornail. It's no secret that integrating a cellular baseband into the same piece of silicon as the applications processor has been challenging for Intel, particularly as the company inherited a modem team with a full pipeline of TSMC (TSM)-built modems. However, it seems that Intel is finally approaching the problem in a sensible fashion that, in my humble opinion, will turn out to be the "right" way to approach it.
SoFIA is Intel's integrated applications processor, cellular baseband, and connectivity chip. The first edition (scheduled for launch in 2H 2014) will have Wi-Fi, Bluetoooth, GPS, and 3G cellular connectivity. The next one will sport LTE. It will feature Intel's "Silvermont" processing core (it's not clear if there will be one or two cores, but Silvermont cores seem to be designed to come in pairs). The shocker, however, is that this chip will be built at TSMC.
What the heck? Why is Intel building the chips that it intends to address the majority of the smartphone and low end tablet market with at rival TSMC when it has fabs to fill? It's simple. The problem here is that Intel's XMM 7260 discrete modem is built on TSMC's 28nm process and porting it over is difficult. Now, that's not to say that the engineers at Intel couldn't build a modem on the 22nm or 14nm processes, but rebuilding the modem on a new process would very likely require that the company go through the entire carrier certification process all over again. Porting over the "Silvermont" CPU cores or the Imagination Technologies (OTCPK:IGNMF) GPU core probably isn't anywhere near as difficult since there isn't any need for carrier certification of the other IP blocks.
Now, eventually (apparently in late 2015), this family off products will be ported to Intel's 14 nanometer process (which implies that Intel will move its entire cellular baseband line to its own fabs by the 2016 timeframe). At this point, the product will go from yet-another-mobile-SoC to something that can truly be ahead of its competition. And that's when things get interesting.
SoFIA Isn't A Looker Today, But By 2016 She'll Be A Bombshell
I'll be honest - SoFIA doesn't look like it'll be anything special. Sure, it'll do well against the quad Cortex A53s and the dual core Kraits that it will be fighting in the low end of the market when it finally shows up in devices, but it won't be particularly well differentiated. Further, given that Intel will be building these at an external foundry, it's not clear if Intel will have much (if any) of a cost structure edge (in fact it will probably not have the volumes that Qualcomm (QCOM) and MediaTek have, which means that its cost structure could be worse).
So, these parts will essentially be proof that Intel can do an integrated part and it'll probably find homes in a number of handsets and low end tablets (the latter especially thanks to all of the sockets Intel is buying this year with Bay Trail). However, the real fun will happen when this part is moved to the 14 nanometer process. At that point, I expect a move to the "Airmont" CPU cores (14nm shrink of Silvermont), a fairly beefy GPU, and a solid connectivity/cellular solution - all at an excellent cost structure.
At that point, I expect that Intel will be extremely competitive with the market leader - Qualcomm - which will probably just be moving its chips to TSMC's "16nm" FinFET process (Intel's 14nm process is 35% smaller and the transistors are likely much higher performance, given that the performance characteristics disclosed by TSMC imply rough performance parity with Intel's 22nm process). This implies a significant cost and performance advantage for Intel (especially since Intel doesn't have to hand over the foundry margins as Qualcomm will).
In short, by early 2016, Intel will likely have a top-to-bottom competitive product line with the best-of-the-best from the market leader in each segment. This is a long-term game, but it's tough to see Intel not being competitive in this space by the 14 nanometer generation.
It's disappointing that it's going to be a while before Intel is leveraging its full might to attack the low end/mid range of the smartphone space, but SoFIA built on TSMC's process should at least be competitive with the low end parts available at the time. For the short term trader, don't expect Intel to dominate the low end of the smartphone market overnight, but for the long term investor this is a compelling game plan - if the company can finally execute in this space.
Additional disclosure: I may go long IGNMF.PK at any time.