TrEKISS - Trusted Edge AI for Multi-sensor systems
Multi-sensor systems are increasingly being used in safety-critical applications such as utility infrastructure, automation technology or supply chains - in some cases under operating conditions that cannot be controlled. This increases the risk of malfunctions and manipulation affecting the function of the overall system. The processing of measured values close to the sensor using AI algorithms and the derivation of decisions on site (so-called edge computing) increase performance and safety in equal measure. A prerequisite for this is tamper-proof, reliable operation of the integrated sensor system. Therefore, the TrEKISS project develops secure, trustworthy multi-sensor systems with high local computing power for AI-based sensor signal processing, data fusion and decision support.
The project will develop design methods that specifically consider AI algorithms and hardware tailored for them, as well as test and analyze trustworthiness. Fraunhofer IPMS is focusing on providing energy-efficient power supply electronics for the multi-sensor electronics, as well as designing and implementing interfaces, software libraries and AI algorithms for the cluster core extension of a RISCV-based edge device. Furthermore, the creation of a toolchain for firmware development, the research and testing of resource-optimized algorithms including the training of neural networks for sensor data analysis, as well as AI-based self-test strategies are part of the project.
The system architecture of the sensor systems will additionally be extended by hardware and software security features to enable self-diagnosis and thereby ensure the demonstrably secure execution of code. For effective protection, a security concept will be created that makes spying by thermal or electromagnetic methods more difficult. Physical intrusion attempts are to be made more difficult by using existing or additionally applied structures as well as suitable material combinations.
In order to test the developed sensor system, application requirements from the field of critical infrastructures (e.g. public utilities, electricity, district heating and gas networks) will be analyzed in the project using a realistic scenario. For the evaluation, the regular use of the system as well as attack scenarios are simulated in test setups and security concepts, e.g. the side channel resistance of the packaging, are evaluated with artificially generated test data.