Autonomous vehicles (AVs) of the future will require different types of storage—and plenty of it—to collect data from LiDAR, radar, cameras and other sensors as well as in-vehicle infotainment, navigation systems and maintenance data.
Depending on the type of vehicle, the storage capacity per vehicle will range from 1 TB to nearly 11TB by 2030, according to a new forecast by Counterpoint Research conducted in collaboration with major storage producer Western Digital.
Robo-taxis will require the most storage because they will operate all day. A fully autonomous Level 5 robo-taxi will require nearly 10.9 TB of storage in 2030. By comparison, a passenger vehicle could function at Level 5 with 842 GB, according to Counterpoint.
The brains of AVs will also demand different types of storage to support split-second decisions. Counterpoint predicts in-vehicle systems will move from Single Level Cell flash NAND and Multi Level Cell NAND to faster Universal Flash Storage and embedded Solid State Drive technology to keep pace with performance requirements.
Counterpoint estimated that about 30% of cars sold globally in 2025 will support Level 2 autonomy capabilities, which already started appearing in 2015. Level 2 offers Advanced Driver Assistance Systems (ADAS) such as adaptive cruise control, lane centering and autosteering, but the driver basically performs all driving tasks. About 8% of cars sold in 2020 will have such capabilities.
By 2030, about half of all vehicles sold will have Level 2 capabilities.
Fully autonomous vehicles will take much longer to catch on, with about 10% of all vehicles at that level by 2030, Counterpoint said. Fully autonomous, or Level 5, means the vehicle can perform all driver functions in most conditions. Counterpoint said a Level 5 vehicle will have as many as 28 sensors, including nine ultrasonic sensors, six radars, seven cameras, four LiDARs, and two infrared sensors.
By comparison, a Level 2 vehicle will have 13 sensors. Counterpoint said that businesses using robo-taxis will most likely be the drivers of Level 5. Robo-taxis will follow predictable routes, usually, such as moving from airports to downtowns. Level 5 will also be used for industrial and agricultural purposes, often driven in controlled environments.
Counterpoint said radar devices per car will generate up to 0.4 GB/hour, while LiDARs on a car will generate much more—up to 252 GB/hour. All the cameras on a car could generate up to 300 GB/hour. As a comparison, a typical smartphone user in the U.S. could consume up to 30 GB of data per month on Wi-Fi and cellular.
One factor that could result in a surplus of storage in an AV is how long it takes to improve machine learning models. At first, carmakers are expected to store more data than less on cars as they work to find ways to upload data to the cloud to strengthen the machine learning. As systems become more sophisticated, the amount of data stored locally on cars could level off or possibly decline, Counterpoint said in its research.
Even though 5G cellular with low latency at less than 10 milliseconds will help cars communicate with the cloud, it will take years for 5G to roll out in all markets with broad coverage, Counterpoint said. “Until then, systems will be based on 4G LTE networks and will have to rely on locally generated, stored and processed data,” Counterproint said in its research.
“We can’t count on the connectivity,” said Russell Ruben, director of marketing for auto at Western Digital. In an interview, he said that the coming AV storage demands are “not a huge surprise to us.” Testing on AVs has already shown such vehicles could generate massive amounts of data.
Western Digital competes with Samsung, SK Hynix, Kioxia and Micron, according to Neil Shah, an analyst at Counterpoint. Carmakers and their suppliers use Western Digital’s automotive grade iNAND eMMC, iNAND UFS and SD cards. Also SSDs and hard drives from Western Digital can be used in data centers, like those that carmakers may develop for analyzing vehicle data. In May, Western Digital introduced the first 3D NAND e.MMC.