|title||Phase-Locking on Neuromorphic Hardware|
A barn owl’s auditory system is remarkable because it can locate sounds with a very high azimuthal precision. This requires that the auditory neurons resolve interaural time differences of a magnitude smaller than the time constants of the involved neurons. In a publication of 1996, Gerstner and others presented a spiking neural network model which is capable of resolving these time differences.
The key process leading to such a precision is termed phase-locking. Phase-locked neurons exhibit a very precise temporal spiking behavior, allowing even small time differences to be distinguished in further processing steps. The aim of this thesis is to show phase-locking on the Spikey chip. This requires several modifications of the originally proposed neuron and synapse models and their respective parameters, because neural networks to be emulated on this chip are restricted to its inherent neuron and synapse models as well as limited parameter ranges. Preliminary simulations with a hardware-inspired software model of the network confirm that phase-locking works well with the modified models and parameters. In the further course of this study, many of these parameters on hardware are measured and adjusted as well as possible to the parameters of the hardware-inspired software model. The adapted network is emulated on the Spikey chip and its performance is analyzed. It is shown that phase-locking can be achieved, which is an important step towards sound localization on neuromorphic hardware.
|Datei||Bachelorarbeit Anne-Christine Scherzer|