In the world of semiconductor technology, there are two critical technical challenges that define the performance and power characteristics of a given product: design engineering and the underlying semiconductor manufacturing technology. Just about every mobile player builds their chips at Taiwan Semiconductor (TSM), Samsung (GM: OTC:SSNLF), or Global Foundries - all of whom are roughly on the same technology cadence in low power chip manufacturing (TSMC is the leader, Samsung is second, and Global Foundries comes in last - but they're all very close). Qualcomm (QCOM), the leading vendor of mobile processors and cellular modems, leads the pack when it comes to adopting new semiconductor technologies in mobile chips - but will that be enough to counter the impending threat from Intel (INTC)?
Understanding the very real threat from Intel
It is no secret that Intel is generally ahead of its competitors with respect to semiconductor manufacturing. While Intel has been shipping FinFETs (this is a new transistor structure that dramatically improves performance and lowers power, but is very tricky to figure out how to make in high volume) since late 2011, the rest of the semiconductor foundry ecosystem - Samsung, TSMC and Global Foundries - are still working through some real pains in getting their first generation FinFET processes for production readiness.
Indeed, while TSMC will ramp its 20 nanometer process (not FinFETs, but the older structures known as "planar transistors") throughout 2014 and into 2015, Intel will ramp its second generation 14 nanometer FinFET process for its PC products and, later in the year, will see commercial availability of its first 14 nanometer designs. From a process technology standpoint, the gap between Intel and the rest of the fabless players will continue to be rather wide which means that if Intel can get the right designs out, it will have natural performance-per-watt and cost advantages.
Even if TSMC gets FinFETs ready for 2015, they'll be at a disadvantage to Intel's
At Intel's investor meeting, the company presented the following slide comparing the density of its 14 nanometer 2nd generation FinFET process to what TSMC/Samsung/Global Foundries have indicated that they would have available in their 16nm/14nm processes:
In the graph, Intel claims that at the 14/16nm generation it will have 35% better density than its competitors at process nodes of the same name. This means that for a given chip of fixed area, Intel will be able to - if its designers keep density in mind - pack much more functionality in a given space than its competitors can while still remaining economically viable.
On top of that, given the claims that TSMC is making about the performance and power characteristics of its transistors vis-à-vis its 28nm process, ...
(click to enlarge)
it is likely that TSMC's 16nm process will have similar performance characteristics to Intel's 22nm process. Given that Intel believes that it will be seeing a rather substantial performance increase from its 22nm node to its 14nm node (even larger than TSMC claims for its 28nm -> 16nm), it is likely that on a per-transistor level, Intel's process will be dramatically superior which, if the design teams at Intel can utilize it properly, should mean that on performance and power, Intel will have no equal.
When will Qualcomm have FinFETs?
The question, now, is just when Qualcomm will have FinFET based products available in the marketplace. Given that Qualcomm announced that its first 20nm modem will be sampling in 1H 2014 for launch in 2H 2014, and given that Qualcomm has not (yet) announced a Snapdragon on 20nm, it seems likely that it will not be until 2015 that Qualcomm is shipping 20nm Snapdragon parts. If this is true then this suggests a late 2015/early 2016 launch for 16nm FinFET based Snapdragon SoCs.
Interestingly enough, I am compelled to further analyze the following slide given at a TSMC event back in October.
In this slide, we see that the first tape-outs (that is, completion of designs) on 20nm happened in 1H 2013. The first commercially available products (not samples) appear to be on track for early 2014 with production quantities available in 2H 2014 (think Xilinx (XLNX) FPGAs and Qualcomm's 20nm Gobi Modem).
My expectation is that Qualcomm will be one of the first to tape out on 16nm FinFET, which means that if they're really lucky - and the first tape-out is the one that actually goes to production (very unlikely) - Qualcomm could conceivably see a 1H 2015 launch of 16nm Snapdragon SoCs. I wouldn't hold my breath on that one, though. I expect Qualcomm's first FinFET products to hit in 2H 2015 or 1H 2016.
Intel now has a window of opportunity, but it's still tough
Intel, yet again, has a window of opportunity. If it can really deliver on the promise of its next generation "Broxton" system-on-chip for tablets and phones in "mid-2015" (as promised at the Investor Day), then Intel could capture a meaningful amount of the high end smartphone share currently held by Qualcomm:
Of course, Intel will have the better transistors, but it will also need a better design, too. As Intel showed with Bay Trail and as it will show with Merrifield/Moorefield, better transistors aren't enough. The best transistors aren't worth squat if your design isn't up to snuff. This means having leadership across all IP blocks from Image Signal Processing to CPU to graphics, as well as having a leading edge cellular modem solution. Without these, a fast SoC is nice but ultimately useless.
And that's my biggest "problem" with Intel's mobile strategy. It shows so much promise and has so much potential, but time and again, Qualcomm has shown that it can tweak its designs faster, build the right products for a wide swath of markets, and has leadership IP blocks across the board. Intel needs to basically become Qualcomm with a manufacturing edge - that's the only way it can win this war. Design and methodology matter and Qualcomm, at least in mobile chips, has Intel beaten here as of today.
Despite my extreme frustration with Intel's execution in mobile, I can't help but be excited for Broxton (or as Rob Tanner calls it, "next chip forever").
Intel has an amazing process technology advantage over the rest of the industry but it has squandered it for several generations in a row, largely thanks to lackluster SoC designs. I'm more than happy as an investor to write this off as the "price of entry", but after Medfield, Clover Trail, Clover Trail+, Bay Trail, and Merrifield all missed the mark, it's time to gun for leadership and win decisively. If Intel can't do it with Broxton, I'm not sure they'll ever be able to do it.