Annina Gutermann, M.Sc.
- Member of Scientific Staff
- Group: Prof. Becker
- Room: 125.3
CS 30.10 - Phone: +49 721 608-41321
- annina gutermann ∂does-not-exist.kit edu
- Engesserstraße 5
76131 Karlsruhe
Multi-valued logic in modern circuits
The end of Moore's Law is in sight and thus there is a need for alternative ways to increase chip performance. One innovative approach is Multi-valued Logic (MVL), in which digital signals are no longer represented in binary but in ternary or quaternary form. In this way, chip area can be minimized and performance increased. In addition to technologies that enable this behavior, potential fields of application also need to be explored. For example, novel FPGA architectures, memory technologies and the use in heterogeneous circuits.
Energy-efficient AI hardware accelerators
AI algorithms are already limited by hardware limitations such as energy consumption, computing power or memory bandwidth. Especially when using AI in the embedded area, there are limitations due to limited resources. Dedicated AI accelerators can be used to achieve more efficient computation. A co-design of the data flows and the hardware architectures must be carried out in order to optimize performance.
In-memory Computing
In classical Von-Neumann computer architectures, CPU and memory are separate units connected by a bus. Loading data from memory is a comparatively slow operation, which leads to the so-called "Von-Neumann bottleneck" in today's highly data-driven world. To circumvent this, calculations can also be performed directly in the memory, which is then referred to as in-memory computing (IMC). This requires new storage technologies and architectures and approaches for integration into the overall system.
| Title | |||
|---|---|---|---|
| Design & Implementation of Embedded Hardware Acceleration for Transformers |