Hardware-Support for Spiking Neural Networks - a look at BrainScaleS-2 and its usage

Seminar paper written in the Bachelor's Seminar on Computer Architecture at the FernUniversität in Hagen. The paper introduces LIF and AdEx as basic neuron models. Afterwards, the architecture, signal processing and training of SNNs is presented. The main part focuses on the architecture of BrainScaleS-2 (BSS) as neuromorphic hardware. It is shown how the neuron models are translated to (analog) circuits and how spike events are processed on the hardware. Finally, simple code examples for SNNs on BSS and challenges arising from the analog nature of BSS are presented.