Recently, I have seen a lot of automotive chips, and I think it is necessary to sort out several manufacturers of core chips in the United States, starting with Qualcomm first.
Qualcomm's fiscal 2021 revenue was $33.6 billion, up 43% year-on-year, and net profit was $9 billion, up 74% year-on-year. The hardware part earns $27 billion, and the patent revenue is $6.3 billion (this part is really unlearnable). Automotive revenue in 2021 was $1.019 billion, up 43.7% year-over-year from $709 million in 2020, with automotive revenue of $240 million in the latest quarter, up 40% year-over-year. If you compare with Mobileye, you have worked hard for many years, and with intel's blessing, it is now the difference of $100 million per quarter.
Figure 1 At the current rate, Qualcomm will surpass Mobileye in 2 years
Part 1
Qualcomm's orders
With annual automotive revenue set to grow to $3.5 billion over the next five years and $8 billion over the next 10 years, Qualcomm has partnered with more than 25 automakers with cockpit and autonomous driving design platforms, with a total automotive business order valuation of more than $13 billion.
Figure 2: The development of Qualcomm's corporate business
From a logical point of view, the basic point of Qualcomm's automotive business revolves around the vehicle network chip and digital cockpit in 2021, and the business volume estimate for 2025 is that the automatic assisted driving > the car networking > cockpit, this path we can do some combing.
The digital cockpit SoC main chip integrates CPU, NPU, GPU, DPU and various peripherals, mainly responsible for the data operation processing work in the infotainment system including camera video, neural network accelerator NPU, audio processing, voice and multiple display image rendering and output (GPU, DPU), in-car Bluetooth WiFi interconnection and Ethernet data interaction of other major ECUs in the car. For Qualcomm:
●First generation (28nm)
Snapdragon 620A practice and temptation
●Second generation (14nm)
Snapdragon 820-820A
●Third generation (7nm)
Snapdragon SA6155, SA8155, SA6155 entry-level cockpit, SA8155 mainstream high-end to flagship-level smart cockpit
●Fourth generation (5nm)
Snapdragon 888-SA8295
The inflection point in the cockpit field is the Snapdragon SA8155 series, using the 7nm process to bring the large computing power (100k DIMPS, GPU 1100 GFLOPS), compared with the traditional car factory (Renesas, NXP, Texas Instruments) the computing power has formed a very large differentiation, through the mobile phone technology migration to ignite the intelligent cockpit market of automobile companies, at this time a large number of luxury car companies began to turn to Qualcomm, and a large number of domestic auto companies began to import this SOC chip at this time.
Figure 3: Qualcomm's cockpit iteration
At present, Qualcomm's iteration rhythm is very fast, to the fourth generation of cross-domain fusion design, according to the centralized publicity of an OEM, the SA8295 CPU computing power created by the 5nm process process exceeds 200KDMIPS, the GPU computing power exceeds 3000GFLOPS, supports WiFi6 and Bluetooth 5.2, and the AI computing power of NPU reaches 30Tops. Information security, visual processing accelerators, isps of visual perception processing and fusion (for in-cabin monitoring) are also added here.
Figure 4: Qualcomm's fourth-generation cockpit platform
At this time, we can fully see that Qualcomm and TI, MTK gameplay is not the same, and it is also different from Renesas and NXP gameplay, becoming an opponent with NVIDIA in terms of cross-domain integration (NVIDIA is still partial to the big market of autonomous driving).
Figure 5:Qualcomm's cockpit has become a starting point for cross-domain convergence in 2022
Part 2
Self-driving Ride Vision SoC
In terms of intelligent driving, Qualcomm came up with this Snapdragon Ride Vision System, which supports multiple cameras (high-resolution 8MP cameras), soC is a processor using 4nm process, including visual AI processors, ordinary data CPUs, visual accelerators for map crowdsourcing, ISPs, interfaces and vehicle safety-related processors, and so on.
Figure 6: Qualcomm's self-driving Ride Vision SoC
The intelligent driving platform contains three core processors, the visual SoC is used to process information from the camera, the ADAS Soc series, which supports L1-L3 automatic driving assistance, and then supports L3/L4 autonomous driving requires a third accelerator chip. This piece of ecological support includes:
● Qualcomm internal support for Operating System and Virtual Machine can be selected for Qualcomm, QNX, Linux Redhat middleware.
● Driver monitoring series prefabricated, self-developed by car companies or Seeingmachines.
●Parking is made by Valeo, Bosch, Zhongmu or car companies.
●The map can be treated with Tomtom and Here visual fusion.
● Driving decisions, Arriver or OEM homemade.
Figure 7:Complete three SoC chips
Summary: That is to say, according to Qualcomm's planning, a smart connected car needs 1 communication chip (V2X, 5G and Wifi), 1 cockpit chip and 2 intelligent driving chips (Vision and ADAS SoC), and the way to play to a higher process is actually more affordable than Mobileye, and the most important thing is to produce more, from the perspective of big logic, traditional car chip companies have no way to burn money with this big family.