Contrary to first impressions of many analysts and encouraged by Qualcomm’s common practice of conflating processors, chipsets, and platforms, the new Snapdragon 865 for 5G smartphones is not a baseband-applications processor. It does not combine 4G, 5G and applications processing in one SoC, as many had anticipated. Nope.
Qualcomm pulled out all the stops for the applications processor, but in so doing, decided to remove the 4G communications processor, as used in the Snapdragon 855, and put it in another package along with the 5G communications processor. Instead of the expected one integrated circuit processor package for next year’s flagship smartphones, Qualcomm announced that OEMs will need two:
This looks very much like the previous Qualcomm Snapdragon 855 + X50 modem, which also consists of two IC packages, except for the location of the 4G baseband processor.
On the downside, the approach used by Qualcomm for its latest 5G flagship smartphone processor offering:
- Increases PCB real estate and production cost.
- Increases power consumption where common silicon hardware block shared between the apps processor and baseband have been eliminated.
On the upside:
- This approach allows Qualcomm to maximize the capabilities of the apps processor.
- The two-chip approach also gives the company time to design and optimize a fully integrated processor for flagship smartphones, with production put off until 2021.
Like Qualcomm, Samsung has also opted for a stand-alone applications processor, the Exynos 990, plus stand-alone multi-mode 5G modem, the Exynos 5123, for its 2020 platform for flagship 5G smartphones. Integration in 2020 will be limited to lower-capability SoCs for mid-range 5G smartphones, for example the HiSilicon Kirin 990, the Samsung Exynos 980, and the MediaTek Dimensity 1000.
Qualcomm will also ship a multimode baseband-applications processor for mid-range 5G smartphones in 2020, the Qualcomm Snapdragon 765. This has lower applications performance than the Qualcomm Snapdragon 865. When Qualcomm announces a fully integrated flagship SoC, probably in late 2020, it will resemble the Snapdragon 765, but with more applications capabilities than the Snapdragon 865.
Capabilities of the flagship Snapdragon 865 processor include a GPU core with 25 percent faster rendering and 35 percent better power efficiency than the GPU in the Snapdragon 855. Qualcomm also includes an AI engine with 15 TOPS (tera operations per second) and more inferences per watt than the competition. The image signal processor (ISP) core in the 865 is capable of 64 megapixels and 30 frames per second, with 2 gigapixel/second and 200 megapixel camera support.
Separating the modem (baseband processing) from the applications processor moves us back to the way cellular processors were implemented in the first years of the transitions from 3G to 4G. Only Apple still uses this approach in cellphones, leading us to speculate that the Qualcomm X55 will be well-suited to work with Apple’s processor in the first Apple 5G smartphone, expected in September of 2020.
Qualcomm’s approach for 2020 mirrors its solution from 2011, during the second year of 4G smartphones. At that time the company moved from a two-chip solution to a different two-chip solution with discrete modem and separate apps processor functionality, before finally integrating the modem and AP into a single SoC in the third generation of its 4G offerings. In 2012, Qualcomm was openly critical of its competitors’ two-chip 4G solutions, claiming they provided lower performance and consumed more power, so it’s surprising and somewhat ironic that Qualcomm has followed the same path in 5G.
To help make up for splitting the 5G solution into two chips, Qualcomm announced its “5G Modular Platform,” which will combine 40+ Qualcomm components into two or three modules.
The company did not specify exactly how it would partition the radio chips into these modules, but we suppose that the two processors and sub-6 transceivers could go into one module, with most of the cellular RF front-end in another, and connectivity into a third. Big modules like this can make RF layout, design and manufacturing of the PCB easier for OEMs, but also limits flexibility, for example making a particular phone unsuitable for a specific region or price target.
Companies have attempted to put the entire cellular radio including the RF front end into a single module in the past, but did not have the scale to do this economically. Essentially, a company offering large, complete modules for smartphones is betting that they can produce most of the phone PCB at lower cost than the high-volume ODMs and assembly houses. The closest to this we have today is probably low-cost cellular M2M modules, which usually contain relatively low-rate radios with limited band support, for example for LTE Cat. M and NB1.
If Qualcomm succeeds with its 5G Modular Platform, this will put more pressure on its chipset competitors (MediaTek, HiSilicon etc.) that lack Qualcomm’s RF front end capabilities, and the RF front end suppliers (Skyworks, Qorvo etc.) that do not offer chipsets. And of course, this could drive further consolidation of the cellular chipset (baseband, transceiver, PMIC) companies with the RF front end (PA, filter, switch) companies.
To track the progress of the cellular processor suppliers and 5G, you can read more in the following reports (subscription may be required):
Written with contributions from Sravan Kundojjala and Stuart Robinson.