42 The demands of highperformance sensors, coupled with artificial intelligence (AI) and machine learning (ML), are driving up the requirements for data storage in uncrewed systems, both during the development phase and in operation. For driverless cars and trucks in particular, the storage requirements are running into terabytes of data, taking streams of information from radar, Lidar and cameras, as well as large ML models. Whether this storage is used and then discarded, stored locally on the vehicle or sent to the cloud is a key architectural decision. The architecture of the autonomous vehicle (AV) is moving towards a centralised computing engine with a software-defined architecture, requiring significantly more memory for the code that runs the vehicle. It must be secure, reliable and able to be updated over the air (OTA), which places key requirements on memory storage. Various memory technologies and standards are being adopted. Multilevel NAND flash provides high-density storage for data from sensors and local AI frameworks, while NOR flash is used for millions of lines of code. With software footprints expected to expand from 100 million lines of code in today’s high-end vehicle to something in the order of one billion in future, automotive memory has become a key consideration. Autonomous vehicles need a safe and secure, centralised computing engine with sufficient memory storage, writes Nick Flaherty Data, data everywhere The first quad-port automotive SSD (Image courtesy of Micron Technology) June/July 2024 | Uncrewed Systems Technology The demands on memory and storage performance to achieve autonomous operation are driving the need for higher memory bus widths at 256 bits and 512 bits, which operate at an input/ output (I/O) signalling rate of 8.5 Gbit/s. This has to be achieved with functional safety to accommodate random and systematic fault coverage. This safety framework provides guidance when considering breakthrough automotive memory solutions and support. There are significant advantages to improving data storage with centralised architectures in AVs. This can reduce the weight of the wiring and therefore the weight of the vehicle by 20 kg, leading to a potential increase in range of up to 4,000 miles per year. Developmental challenges The major challenge for data storage is during the development phase. AV algorithm development requires data from many vehicle sensors, including more than 12 cameras, infrared, Lidars, radars, ultrasonic sensors and other parameters, which create over 100 TB of data per car per day. This data is recorded and transferred in a variety of ways to remote data centres for algorithm development – from consumer-grade storage drives in large bays within the vehicle to integrated data-loggers and storage devices. In both instances, security, scalability and transfer costs are exceedingly expensive and challenging. Solutions can cost more than $200,000 per vehicle to support the capacity required. For a fleet of 100 vehicles, that is more than $20m in upfront costs to secure the required data storage. Additionally, these solutions require new capex purchases of storage components as technology continues to push boundaries towards fully automated vehicles. AV developers need a more cost-effective and scalable solution to gain access to more data faster. The collected data needs to flow from vehicle to data centre, where
RkJQdWJsaXNoZXIy MjI2Mzk4