Xilinx flexes its muscles in the ADAS arena
For its newly introduced 16nm Ultrascale+ family of FPGAs and MPSoCs (multipcrocessor SoC) Xilinx claims significant performance benefits. Also the performance per watt, an important criterion in power-constrained environments such as ADAS and other automotive applications, has been increased considerably. The architecture combines an ARM Cortex R53 32-bit dual-core processor for real-time tasks with a Cortex A53 64-bit quad-core application processor and a high-end graphics-processing unit Mali-400MP, also from ARM. Also integrated are a power management unit that supports power gating and power islands, a broad selection of FPGA logic including RAM, interfaces and video codecs. For deployment in the connected car which involves high security and safety requirements, the device family also contains an integrated security unit with functions such as information assurance and key management. Lockstep options enable the deployment of the device in environments where functional safety as defined in ISO26262 is an issue.
In specific applications, Xilinx significant performance/power improvements – for example in an Automotive multi-camera driver assist system, the company claims the new Zync Ultrascale MPSoc quintuples the performance per watt, compared to an earlier configuration with double Zynq-7000 AP SoCs. Despite consuming 50% less power, the Ultrascale implementation has 2,5 times more application performance.
At Embedded World, the FPGA vendor showcased a number of applications for the new chips, targeting next-generation automotive environments. Among them were a camera-based 3D reconstruction and position tracking system for inhouse navigation (such as autonomous driving in parking garages or similar use cases). The system included a highly parallel hardware implementation of an algorithm from augmented reality software company Metaio, explained Stefan Janouch, Xilinx senior manager EMEA Automotive. The performance of the prototype was high enough to track up to 2000 features in real-time. This algorithm, called Simultaneous Localisation and Mapping (SLAM) is regarded as a key building block to implement, for instance, self-parking cars.
Another exhibit was a logic board that allows the connection of up to six cameras, enabling the development of surround-view systems for parking assistant systems. Unlike most competing systems, the logiADAK v3.0 (as Xilinx has baptised the development board) is capable of on-target real-time calibration. With varying numbers of cameras connected, the system could be used as platform for advanced driver assistance systems such as blind spot detection, pedestrian detection and range estimation, or lane departure warning. The exhibited system ran the so called Semi-Global Matching algorithm that allows designers to process image data and to devise 3D spatial information out of the stereo image. The algorithm offers high exactness even at large distances, Janouch said.
Related articles:
Why Designing an FPGA makes a Difference for Automotive Engineers
Unified hardware/software debug with embedded virtual prototyping
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
