Memory // Storage

Micron’s Automotive Memory: Enabling Black Box Capabilities

By Gil Golov - 2020-05-21

A vehicle’s black box records sensor data that is intended to be used to reconstruct and determine the root cause of an accident. Autonomous vehicle Levels 3 through 5 are increasingly requiring black box recording capabilities, with requirements that are being defined by authorities and which tend to vary between countries. The U.S. Department of Motor Vehicles (DMV) published a paper (excerpt below) with requirements stating that autonomous vehicles store sensor data for at least 30 seconds prior to the event of a collision.

The autonomous vehicle has a separate mechanism, in addition to, and separate from, any other mechanism required by law, to capture and store the autonomous technology sensor data for at least 30 seconds before a collision occurs between the autonomous vehicle and another vehicle, object, or natural person while the vehicle is operating in autonomous mode. The autonomous technology sensor data shall be captured and stored in a read-only format by the mechanism so that the data is retained until extracted from the mechanism by an external device capable of downloading and storing the data. The data shall be preserved for three years after the date of the collision.

Figure 1: Block diagram showing the processing of data in a black box recorder

Figure 1 shows a block diagram of a black box recorder, where a continuous stream of data that is aggregated from multiple sensors is sent. For cost and security reasons, some systems first compress the data and then encrypt it. The main contributor of sensor data bandwidth is typically image sensors. An autonomous vehicle can contain up to 12 image sensors, where long-range cameras can reach up to 8M pixel resolution at 60 frames per second. The resulting data stream can reach up to 20 GB/s.

For cost-sensitive applications, H.265 compression offers up to a 50% reduction in the resulting bits, leading to overall lower storage requirements. H.265/HVEC is a lossy compression algorithm that removes parts of the data — where the human vision system is less sensitive. However, some artificial intelligence (AI) algorithms might be sensitive to such data distortion and such compression techniques might distort the operation of the AI algorithm when trying to reproduce the root cause of the accident based on the recording. For this reason, some systems, especially those used in robotaxis, tend to avoid the use of data compression or use a very low compression ratio. Consumer vehicles tend to be more tolerant to the use of compression, especially at the lower levels of autonomous driving.

To have a record of the period just before the accident (such as the last 30 seconds), a cyclic buffer is used. The cyclic buffer is memory that can be based on DRAM or flash and needs to have sufficient capacity to store the required buffer length. For example, to capture 30 seconds worth of data prior to an accident, assuming a 1GB/s uncompressed sensor data rate, the cyclic buffer needs to have 30GB of storage capacity.

Flash is commonly used to implement the cyclic buffer as it is designed not to lose data if power is lost. DRAM, on the other hand, requires backup power to ensure the collected data isn’t lost if the main battery loses a connection. The challenge associated with flash technology is overall endurance. In the extreme case of a robotaxi that is on for 24 hours for almost five years, nearly 45,000 hours of continuous operation would result. Assuming an extreme case of 1GB/s sustained-sensor data stream, the required endurance would be in the range of 150PB. This level of endurance would be challenging if not impractical to achieve with today’s flash technology.

The NVM storage directly following the cyclic buffer in Figure 1 provides a long-term storage location for the 30-second data snapshots associated with either an accident or a potential accident. The system relies on analysis from an accelerometer (G-sensor) and AI sensors to determine when an accident or potential accident could have occurred. These sensors flag when the data in the cyclic buffer should be written into this long-term NVM storage. This long-term storage device is typically based on flash memory, and unlike the cyclic buffer, it has much more modest endurance requirements.

Micron system architects work closely with our tier-1 customers and OEMs to architect system-level solutions such as the black box to ensure optimal cost, performance, and power tradeoffs that meet the most demanding applications. They also want to confirm that the system integrator designs the application in a way that keeps any potential memory failure from causing harm to data, people or property.

With over 28 years of continued focus on the automotive market and the broadest automotive memory portfolio, it is no wonder Micron is the top automotive memory supplier.

Gil Golov

Gil Golov

Gil Golov, senior manager of Automotive System Architecture & Strategic Marketing, is responsible for Micron’s autonomous driving system architecture and solutions. Prior to working for Micron, Gil spent more than 15 years in various R&D roles. He holds a Bachelor of Science in electronics from Tel Aviv University and a Master of Science in microelectronics from Brunel University (in the U.K.). Gil holds 26 patents.