Main content
Top content
Research Areas
In research, the Computer Engineering group focuses on
Adaptive heterogeneous architectures for embedded systems and cognitive edge computing.
Adaptive Heterogeneous Architectures
On architectural level, we are looking at multiprocessor systems and graphics processing units (GPUs) as well as dynamically reconfigurable systems, which can adapt their internal hardware structure to changing environmental conditions at runtime. Main goal of this research field is to provide design methods for efficient utilization of dynamically reconfigurable architectures in embedded systems in order to increase resource efficiency and reliability. In addition to modeling and theoretical analysis of architectures and applications, we always target practical implementations and demonstrations of the developed concepts by means of field-programmable gate arrays (FPGAs). For this purpose, both commercial platforms and self-developed hardware-software systems are used.
The methods for realizing adaptive heterogeneous architectures developed in the group are used, for example, in image processing, Industry 4.0, smart home applications, and autonomous robots with high demands on energy efficiency. Thereby, our approaches are not limited to single components but are especially suited for use in networked embedded systems. Our goal is to develop intelligent embedded network components that automatically adapt their computing and communication capabilities to the changing requirements in dynamically networked systems.
Cognitive Edge Computing
Today, mobile devices and IoT components achieve their performance primarily by outsourcing application processing to centralized cloud infrastructures. In particular, applications in the field of machine learning are currently depending on data centers with very high computing power accompanied by high energy requirements. However, if decisions have to be made under real-time conditions within milliseconds, new approaches are required.
Cognitive edge computing relocates machine learning algorithms into the "edge", i.e., closer to the end devices or even directly into the sensor-actuator systems. Only in this way applications such as autonomous driving, industry 4.0 or human-machine interaction can be realized. Sensitive data is no longer simply transferred to the cloud, but processed locally in order to meet, e.g., the important requirements for privacy and security in the smart home sector.
Hardware-software platforms for edge computing are subject to severe resource constraints and application developers usually have to compromise on the accuracy of the realized calculations. Here, the Computer Engineering group focuses on scalable heterogeneous systems that combine domain-specific architectures (DSAs) with GPUs and FPGAs. At runtime, these systems can adapt to changing requirements and thus optimize, e.g., power dissipation, performance or quality of the results.