Memory for Autonomous Vehicle Black-Box Recorders

By Robert Bielby - 2018-07-25

Authorities mandate car OEMs to install integrated “black box” recorders to record video and sensor data. The black box not only stores the video that comes from various cameras already installed in the vehicle, but also records data coming from LIDAR and RADAR sensors. The black box is required to record the last 30 seconds of video and sensor data before a potential accident happens. This information is intended to be used to help determine the root cause of an accident.

Typically, a black box is based on 2 memory subsystems. The first memory subsystem consists of a 30 second temporary memory buffer that typically is based on either DRAM or a Non-Volatile-Memory. The second memory subsystem intended for long-term data storage is based on a Non-Volatile-Memory technology, such a SSD (Micron Automotive).

When the vehicle operates in an autonomous mode, the black box starts to record sensor data into the temporary cyclic buffer. When the buffer is full, the new data will overwrite the old data. In the case of an accident or if there is a high probability of an accident event, e.g. the Emergency Braking System was activated, the car will send a signal to the black box to copy the temporary cycle buffer into the long-term NVM storage device. The data that is ultimately written to the long-term storage device reflects sensor and camera data which was captured 30 seconds prior to the actual triggering event.

One of the main challenges in designing such a system is the design and selection of the memory that is used for the cyclic buffer. Considering that the sensors can generate data at a sustained bandwidth of 1GB/s, storing 30 seconds requires 30GB storage capacity (Micron Automotive). Beyond the sheer storage capacity challenge, there is also stringent endurance requirements over the device’s lifetime. For instance, recording 8000 hours over the lifetime of a car would require Total-Byte-Write (TBW) capability of ~29 Peta-Bytes. One option to address the high TBW requirement would be to implement the cyclic buffer in a DRAM device, such LPDDR4. However, this solution may suffer from data loss if a power cut happens, and additional backup power would be required. Using a backup power is uncommon in such applications due to the lack of technology that can support the ISO26262 ASIL-D reliability standard. Another option would be to use an NVM device as the cyclic buffer. The required endurance makes such a solution challenging. However, Micron’s NVM 3D storage technology supports an extended SLC mode that helps to extend the device’s endurance for such applications.

Fleet vehicles typically record raw sensor data as described above. Consumer car OEMs, that try to keep the system’s cost down, may decide not to record raw sensor data, but to use a video data compression mechanism such h.264 or h.265. These are lossy compression algorithms that offer high compression ratio for video, but cannot reproduce the data up to the bit level. After an accident, the recording is good enough for visual inspection of the accident, but insufficient for feeding at the lab the autonomous vehicles sensors fusion computer for algorithm’s behavior reproduction. Employing an algorithm with a heavy compression ratio, a single NVM device as an eMMC, UFS or SSD, may be able to handle both tasks of data buffering and long-term data storage.

The black box is required to support temperature ranges from -40C up to +105C, and should have data retention capabilities for at least one month with power off. Micron’s automotive NVM and DRAM devices are fully compatible with these requirements.

Micron’s system architecture team works closely with automotive customers and helps them to optimize the system and choose the right memory devices for such black box applications. Please contact Micron for further details.

Rob Peglar

Robert Bielby

Robert Bielby, senior director of Automotive Systems Architecture and Segment Marketing, is responsible for the strategy, marketing and product definition for Micron’s Automotive Division business group. Before joining Micron, Robert spent more than 30 years in systems, semiconductor and solutions businesses holding various engineering and executive roles at Kodak, Altera, LSI Logic, and Xilinx. Robert brings a wealth of experience at the system level in architecture, strategy, vertical marketing and product planning. Robert has authored multiple articles on broad industry topics and holds more than 40 patents in the areas of channel coding, digital signal processing, and programmable logic devices.