Qualcomm (QCOM +1.36%) is, by far, the leading mobile system-on-chip vendor. A big part of this, as the company routinely points out, is due to its "multi-tiered chipset strategy." In other words, Qualcomm develops chips that have the correct feature sets and, perhaps most important, cost structure for each price segment of the market.
During its 2014 analyst day, Qualcomm posted the following slide laying out its product lineup in each tier through, presumably, the majority of 2015:
Source: Qualcomm.
Note that the Snapdragon 808/810 chips will be manufactured on a 20-nanometer process, while the remaining Snapdragon parts are slated to be built on a 28-nanometer "LP" process. Keep in mind that Taiwan Semiconductor (TSM +2.82%) offers multiple flavors of 28 nanometer today, with the two most popular known as 28-nanometer LP and 28-nanometer HPM. The latter is higher performing than the former, but it's also more expensive.
The question I'd like to explore in this article is: When Qualcomm advances the Snapdragon 600/400/200 family of processors, what manufacturing technology will it use?
28-nanometer HPM is probably out
According to industry expert Handel Jones of International Business Strategies, in moving from the 28-nanometer node(s) to the more advanced 20-nanometer node, the cost per transistor is expected to move up. This is a problem because any given functionality takes X transistors to build, so the cost of the chip rises when you add more functionality with a process that has a higher transistor cost.
However, according to an article by Chris Edwards, Qualcomm and Taiwan Semiconductor worked together to make the 20-nanometer process more cost-effective than the initial definition had called for. According to the piece, Qualcomm claims the 20-nanometer node is now more cost-effective than 28-nanometer HPM and about on par with the cost of 28-nanometer LP.
Given the density advantage of 20 nanometer, and the improved transistor performance and power characteristics, Qualcomm would seemingly be right to move its 2016 low-end system-on-chip lineup to 20 nanometers, rather than stick to 28-nanometer LP or shift to 28-nanometer HPM.
However, there's one more potential option for Qualcomm.
28-nanometer HPC looks like a good option
In September, Taiwan Semiconductor announced yet another flavor of its 28-nanometer manufacturing technology, known as 28-nanometer HPC, or high performance compact. The company said it is targeted at cost-sensitive applications and offers significant performance improvement and power reduction from the 28-nanometer LP.
This might be the optimal process from a cost-per-transistor perspective in the 2016 time frame. Furthermore, given the increased costs that doing a design on the more advanced 20-nanometer node reportedly requires (per Handel Jones), even if 28-nm HPC is not cheaper than 20 nanometer on a cost-per-transistor basis, it could lead to overall cheaper products as design costs might conceivably fall.
Some additional thoughts
In its press release on 28-nanometer HPC, Taiwan Semiconductor claimed that "a comprehensive 28HPC IP ecosystem is also established and seamlessly applicable to 28HPM designs, accelerating time-to-market for customers."
Qualcomm's first 20-nanometer modem is the MDM9235, a category 6 LTE-Advanced cellular baseband. To move the 2016 low-cost product stack to 20 nanometer, the company would also likely need to include this baseband, which might be overkill for those product segments.
It could make more sense to bring the MDM9625 LTE-Advanced baseband that Qualcomm integrated on the Snapdragon 800/801 to this market for the purpose of proper product segmentation. The argument against this, though, is that if competitors move more aggressively than I expect in this tier of the market, holding back on process and modem technology could put Qualcomm at a competitive disadvantage.
