New Entrants to Autonomous Vehicle Solutions Will Drive the Market in the Coming Decade

Autonomous driving is one of the most important applications for semiconductors in the next five years. The artificial intelligent (AI) chip in the autonomous driving system is the most critical component and also a new opportunity for SoC designers. Driven by Tesla, many vendors have joined the race to design chips for autonomous driving over the past few years. Both start-up companies (such as Cruise) that are good at artificial intelligence algorithms and giant chip vendors (such as Google, NVIDIA, Intel) want to take this chance to demonstrate their capability. However, the production of autonomous driving chips also requires huge investment and needs the understanding of chip design and manufacturing, so the companies that specialize in both have a greater advantage.

Figure 1: Design process for autonomous driving chips

Counterpoint Design process for autonomous driving chips

The camera is a critical sensor for autonomous driving systems. The images/videos taken by the camera must be processed by artificial intelligence chips to identify and interpret the environment around the vehicle and then make decisions. Examples like the LKA (Lane Keeping System) in level 2 systems, can detect line markings, and control the vehicle to automatically drive centrally in the lane without human intervention. The processing power of the chip and the associated algorithms will determine the ability of the car to drive autonomously.

The autonomous driving system above level 3, uses more sensors (such as cameras, lidars, and radars) to reconstruct the 3D model around the car. The system automatically takes the corresponding action based on instructions from the system. When combined with data from all other cars in a fleet, the collected information can be used to continuously improve the algorithms, with updates being pushed to all cars that use the same drive systems.

Thanks to the improvement of semiconductor manufacturing technology, 40nm technology has been widely used to produce automatic level 2 autonomous vehicle chips. In order to meet the high-speed demand for level 2 and above, mainstream autonomous vehicle chips are also moving below 7nm which will also bring improvements both in cost and performance. The number of units per system is expected to increase with autonomy level because of the requirement for system redundancy and higher performance.

Figure 2:  Autonomous vehicle chips unit sales forecast

Counterpoint autonomous vehicle chips unit sales forecastSource: Counterpoint, Updated in January 2020 

Solutions from Major vendors


NVIDIA announced the next generation of self-driving car platform DRIVE AGX Orin at GTC China 2019. It delivers 200 TOPS performance, seven times that of the previous generation of Xavier. DRIVE AGX Orin is expected to begin mass production in 2022 on Samsung’s 8nm LPP process which will be quite mature in 2022. According to NVIDIA, the platform will use two DRIVE AGX Orin and two GPUs to achieve Level 5 autonomy. This solution can achieve 2000 TOPS, which is the highest of all solutions announced, but at the cost of 750W power consumption, enough to significantly dent the range of electric vehicles. NVIDIA is leading at computer vision, and it also believes level 5 can be achieved by using cameras as the principal sensor; although NVIDIA’s solution also supports lidar and other sensors.


Mobileye was acquired by Intel in March 2017 and is currently the company that sells the most autonomous driving chips. Mobileye launched the fifth-generation SoC “EyeQ5” for fully autonomous driving at CES 2018. EyeQ5 is expected to begin mass production on a 7nm process in 2021. EyeQ5 can achieve 24 TOPS at a power consumption of 10 W, supporting autonomous levels 4 and 5. Intel plans to combine EyeQ5 with the “Intel Atom” processor and develop its AI computing platform for autonomous driving. Although Mobileye is adamant that the camera can achieve full self-driving, EyeQ5 can also support radar and lidar. EyeQ5 can also perform sensor fusion to process the data from various sensors. Intel and Mobileye believe that two EyeQ5 SoCs and one Intel Atom processor are enough to achieve level 5 autonomous driving.


Following in NVIDIA’s and Intel’s (Mobileye) footsteps, Qualcomm has launched its solution for autonomous driving, Snapdragon Ride in CES 2020. Qualcomm has long been supplying solutions for infotainment but is now applying its capability to the autonomous vehicle systems. However, the deployment of the autonomous system requires long-term data collection, cooperation with Tier 1 suppliers, sensor company support, and data training capabilities. NVIDIA and Intel (Mobileye) have a head start. Qualcomm has a long way to go but we’re still early enough in the development of autonomous systems that most of the market is still to play for.

This is a summary of Counterpoint’s research for autonomous vehicles SoC solutions in 2020. Please contact Counterpoint Research for the detail.