Korean researchers devise single-chip brains for autonomous cars
The Processor Research Group at ETRI has announced the development of Aldebaran5, a processor for self-driving cars. Aldebaran5 features nine cores to provide high computing power as well as safety mechanisms to ensure reliability for automotive applications, while exhibiting industry leading power efficiency (~1 watt) and a tiny form factor (smaller than the size of a nail head.)
Codenamed Aldebaran5 after the brightest star in the constellation of Taurus, the chip is a significantly upgraded version of its predecessor. Four of its nine cores are dynamically programmable to operate for performance or for reliability. Safety features adhering to international standards ensure reliability with 99% detection coverage of faults and malfunctions and includes recovery mechanisms to mitigate potential negative effects. While the creators of the chip did not mention it explicitly, this probably refers to the ISO standard 26262 which describes the features and development process of safety-relevant components and software.
Dedicated custom-designed acceleration circuitry for image recognition and HEVC (high efficiency video coding) as well as CAN bus communication hardware and an image signal processor (ISP) are integrated into the SoC, all combining to provide a robust and streamlined computing platform designed for autonomous vehicle applications.
A wide range of applications have been demonstrated on Aldebaran5 attesting to a powerful, efficient, robust, yet reliable computing platform tailor-made for autonomous vehicle applications. These include real time pedestrian and vehicle recognition, lane detection/warning/keeping system, ultra high definition black box functionality (dash cam), interfacing and processing of various sensor data (radar, GPS, etc) and vehicle maneuvering via CAN bus control, among others.
Research and development to integrate neural network acceleration hardware in order to provide artificial intelligence capabilities are near completion and expected to be announced within the year. With such a feature, the Aldebaran could possibly compete with Nvidia’s Xavier processor family which is currently the centerpiece of many development processes for automate driving, possessing quite a dominant position.
Related articles:
AI computing platform targets level 5 autonomy
Volkswagen and Nvidia plan to make cars smarter
Nvidia introduces compact single-chip AI platform
Toyota’s autonomous cars get electronic brain from Renesas
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
