Research

The Department of Computer Engineering researches various ways of using approximations to significantly save energy and resources. Our technology-independent approach is based on the use of hardware description languages (HDL). In addition to established languages such as VHDL and Verilog, we are increasingly relying on more modern paradigms such as Chisel and Clash. These languages, which are based on established general-purpose programming languages, enable us to use the latest techniques from software engineering in hardware development. This allows us to work consistently from the bit level through to complete design space exploration within the same language. A major advantage of these languages is direct access to the hardware abstraction, which enables efficient analysis without time-consuming simulations and the rapid implementation of prototypes (rapid prototyping).

In the field of logic synthesis, we deal with the fundamental processes of translating abstract hardware descriptions into efficient physical implementations. To verify and optimize these designs, we use formal methods, including SAT and SMT solvers, to check the correctness and properties of circuits with mathematical precision and to detect design errors at an early stage.

Another central and highly topical area of research is artificial intelligence, particularly in the context of neural networks. We deal with the many open research questions in this area, including the choice of the optimal neuron model, the concrete activation function and the efficient number representation. A particular focus is on spiking neural networks (SNNs), which aim to model the human neuron in a more biologically accurate way. It has been shown that certain neuron models can be implemented particularly energy-efficiently in hardware.

We are actively involved in the field of open hardware, particularly in connection with the RISC-V architecture. This enables us to contribute to the development of open and flexible processor architectures and to promote accessibility and innovation in hardware research.

The efficiency of the designs we develop is extensively evaluated on FPGA boards. These platforms allow us to implement different circuits quickly and easily and to precisely measure and optimize their energy consumption.