Unmanned Systems Technology 042 | Mayflower Autonomous Ship | Embedded Computing | ElevonX Sierra VTOL | UUVs insight | Flygas Engineering GAS418S | Ocean Business 2021 report | Electric motors | Priva Kompano

10 Platform one Mobileye has developed a single- chip high-performance controller for consumer autonomous vehicles at Level 4 (writes Nick Flaherty). The EyeQ Ultra provides 176 TOPS of performance. First silicon for the device is expected in late 2023, with full automotive-grade production in 2025. Mobileye designed the chip after having first built an autonomous vehicle to understand exactly what it needs to operate at a very high mean time between failures. This approach enables the optimum balance of performance across different accelerators and general-purpose processors, which are implemented as ‘chiplets’ in the package. There are four different accelerators in the package that take data from two sensing subsystems – one camera-only system and the other radar and Lidar combined – as well as the vehicle’s central computing system, the high- definition map and driving policy software. There are 64 accelerators, built in a 5 nm silicon process technology, each paired with a CPU, image signal processors (ISPs) or GPUs. The 12-core CPU uses a customised version of the RISC-V open source instruction set architecture, with each barrel-based core handling two threads and switching between the threads on a cycle-by-cycle basis. Using RISC-V has allowed Mobileye to optimise the design of the processors for autonomous vehicle operation to handle large amounts of data using a single- instruction, multiple-data architecture, and have the low latency using a barrel- threaded design. The previous Eye5 and Eye6 chips used the 64-bit MIPS architecture and an ISP based on the ARM instruction set. The ISP in the Ultra has 256 cores, handling 2.4 GP/s for two channels of 4K video encoding using the H.265 encoding standard at 60 frame/s. Mobileye is part of Intel, which has its own graphics processing technology that can be used for image recognition using neural network machine learning. This uses a course-grained reconfigurable architecture for flexibility in processing a deep neural network. The accelerators are separate chips in the package, linked by a high- speed interconnection fabric designed to ensure that the data is coherent across all the different processors with low-power DDR4 DRAM memory in the package. This reduces the power consumed for data transfers between the accelerators and brings the overall power consumption down to less than 100 W for the 176 TOPS. By contrast, the current state of the art for a central controller, the Orin processor from Nvidia, is built on an 8 nm process, provides a performance of 254 TOPS and consumes around 80 W. This is being used in some six vehicle models launching this year, although several of them use four of the Orin chips. Mobileye points to the efficiency of the Ultra chip, which is optimised for a combination of camera, radar and Lidar sensors so that only the one chip is needed for Level 4 autonomous operation. In the same 2023 time frame as the Ultra chip, Nvidia’s Atlan processor is aiming at 1000 TOPS of performance, with sampling in 2023 for production vehicles in 2025. Its specification, power consumption and process technology have yet to be announced though. Mobileye’s move to the RISC-V architecture has also meant a shift in the development tools, and the company has launched a software development kit for developers who can also use the OpenCL standard. The design of the Ultra builds on the existing chips used to process data from cameras, and the Ultra is the equivalent performance of 10 of the current EyeQ5 devices. Mobileye is currently working with Chinese firm Zeekr on a Level 4 driverless car design that uses six EyeQ5 devices for sensing as well as handling the Responsibility- Sensitive Safety-based driving policy, based on the IEEE 2846 standard that is expected at the end of March. Level 4 chip controller Driverless cars Xyxxyxxxy February/March 2022 | Unmanned Systems Technology The EyeQ Ultra chip is aimed at consumer autonomous vehicles