Neuromorphic Computing deals with hardware systems and architectures specially adapted to solve complex problems using neural networks. The goal is to create systems, that mimic the behavior of neurons more closely and realize neural network tasks in a fast and energy efficient manner. While conventional computer systems, implementing a von Neumann architecture, can be applied for a neural network they often do not yield a good result. Massive parallelism and intelligent data reuse in a custom designed architecture can lead to huge improvements in both runtime and energy efficiency.
At our chair we mainly focus on designing novel architecture concepts and integrating state of the art technology like Resistive RAMs (RRAMs) or High-Bandwidth-Memory (HBM). Thereby we can further optimize systems and adapt them for different application areas, ranging from server-based batch processing down to embedded applications.
We work in cooperation with Fraunhofer Institute for Integrated Circuits IIS to build a competence center based around Neuromorphic Computing here in Erlangen. More information about our partner can be found here.