Hardware Accelerators for Autonomous Vehicles

An autonomous vehicle (AV) is a massive system composed of sensors, electrical/electronic (E/E) hardware devices, and software technologies to improve driving performance, prevent accidents, provide optimized driving conditions to drivers, or automate driving. Nowadays, AV technologies leverage rich sensor information and promising performance by AI/DL-based methods within actual vehicles. The vehicle E/E architecture should inevitably be able to manage a large amount of raw data from various sensors and complex computation for algorithms to serve autonomous driving. Building design strategies for AVs have been challenging because the process should also ensure power efficiency, performance accuracy, mobility, cost, safety, and many other things. Embedded processing systems for AVs are required to digest enormous data in real time so that the autonomous driving tasks operate in purposed performances. For operations safety issues of AI accelerators, the hardware accelerators must ensure the achievement of requirements for functional safety and safety of the intended functionality (SOTIF), so it is mandatory to comply with international standards during the development process. Under the guidance of those standards, development parties, including OEMs and Tiers, decide the specification of AI accelerators at their discretion. Manufacturers in the automotive industry are designing and developing their hardware form factors with GPUs, FPGAs, and ASICs to satisfy requirements and to implement well-performing algorithms and applications of AVs within the given conditions. Companies such as Tesla, Nvidia, and Intel have been developing AI accelerators for edge computing–based autonomous driving systems, and the market is expected to grow further as the commercialization of Level 3 or higher autonomous vehicles approaches.