EMDrive

Future vehicles spread their computing work over a network of sensors, control units and central computers. EMDrive (platform concept for distributed heterogeneous automotive real-time computing network architectures) developed a flexibly scalable platform concept for this, designed for the highest safety level, ASIL D.

• Co-design of AI algorithms and electronics and dynamic allocation of processor elements in the vehicle
• Pre-processing close to the sensor to reduce data volumes early
• Integrated monitoring and diagnostics for safety and reliability

Our contribution

We developed a virtual prototype of a neural network accelerator: a SystemC/TLM model of a systolic array coupled to a processor. It lets design alternatives be explored early and safety and security be assessed before any hardware exists. We also trained AI models, above all convolutional neural networks for processing radar data. The project's architecture is described in EMDRIVE Architecture: Embedded Distributed Computing and Diagnostics from Sensor to Edge at DATE 2024.

Partners and funding

Coordinated by Infineon Technologies AG, with Continental, Elektrobit, Siemens, ZF, Kernkonzept, Ambrosys, KIT, FZI, TU Munich, Fraunhofer ENAS and the University of Lübeck. Funded by the German Federal Ministry of Research, Technology and Space (then BMBF) in the MANNHEIM programme of the Zukunftsfonds Automobilindustrie.