In these days, Soft-in was in Ulm University at THU Workshop: Programmable Processing for the Autonomous / Connected Vehicle. Our Technical director Andrea Tomatis and Detlef Teichner of Soft-in Embedded, together with Elmar Cochlovius, spoke about the “Design Challenges for a Zone-based Automotive Network Architecture”.
Design Challenges for a Zone-based Automotive Network Architecture
Authors: Prof. Dr. Elmar Cochlovius (FH Furtwangen, Germany), Andrea Tomatis (Ph.D. Soft-in S.r.L, Torino, Italy) and Dr. Detlef Teichner (Soft-in Embedded GmbH, Eltville, Germany)
Current generation vehicles are using a Domain-based network architecture. Dedicated busses are applied to route data from data sources (e.g. sensors) to domain-specific ECUs and to route processed data from ECUs to data sinks (e.g. sound system). Gateways are needed for the exchange of information between different domains (e.g. body control, infotainment) and for routing control and diagnostic signals.
Next-generation ADAS and autonomous driving functionalities require many more signal links and much higher data rates to be transmitted – with severe latency demands. New system architectures are based on local ‘zone’ computing to aggregate sensor data and high performance central computing units to perform central decision making and control, all related to service oriented communication.
In order to make zone-based architectures become a reality, various design challenges have to be addressed:
- A zonal network requires data aggregation at outer skirts of the network.
- Multiple network protocols have to be multiplexed and de-multiplexed in order to access and utilize a uniform high-bandwidth communication backbone.
- Most of the network traffic has to be deterministic with low latency.
Zone controllers are used to address these challenges. Each physical zone is controlled by a separate high-performance ECU, which provides the following core functionalities:
- Data aggregation of all sensors of its zone
- Filtering and preprocessing of local data before further transmission
- Multiplexing and upstreaming data to one or more central compute units, including traffic shaping where required
Each of these core functionalities is a design and implementation challenge on its own. Current and next generation FPGAs play a vital role here and provide attractive and efficient solutions. A multitude of design choices opens up and must be carefully evaluated. Examples include:
- Load-balancing between central and zone computing: what are the trade-offs of preprocessing local data (e.g. object detection in LIDAR / Radar data) inside the zone computer in terms of bandwidth saved versus additional delay?
- Low-latency protocol converters inside the zone computer: How can FPGAs help to reduce processing load from the standard Soc?
- Preserving time-correlation between data samples of different sensors: timed data of different sensors arriving in parallel need to be correlated efficiently before being serialized.
In this presentation, we will highlight various challenges of next-generation automotive network architectures and how FPGAs can help to address them. We will focus on characterizing the design space and related constraints.