This thesis presents methods to improve the usability of a neuromorphic hardware de-
vice. The utilized chip physically implements a network of spiking neuron models. It is
operated with a high acceleration compared to biological real-time and is designed for
the investigation of computational principles inspired by the brain. Its application is
hindered by characteristics of the implemented units, as emulation results reflect inho-
mogeneities within the utilized substrate. In a first step, various sources of imperfection
are identified, specified and, if possible, counterbalanced by calibration routines. In or-
der to further increase the homogeneity of the substrate, balancing approaches on the
network level are sought. Extensive software simulation studies prepare the adoption
and successful application of biologically inspired self-stabilizing architectures to the
hardware system. It turns out that the application of short term synaptic plasticity is
vital for achieving a foundation the research on brain-like computing with neuromorphic
hardware can build upon. |
