Evaluating Neuron Models through Application-Hardware Co-Design

A key issue in the field of neuromorphic computing is how much inspiration to take from the brain in the im-plementation of neuromorphic hardware systems. For example, neuron implementations in neuromorphic hardware have varied from non-spiking McCulloch-Pitts style neurons to extremely biologically-detailed Hodgkin Huxley style neurons. In this work, we examine a variety of biologically-inspired features to include in neuron models for neuromorphic systems. We evaluate features such as leak, absolute refractory period, and relative refractory period. We evaluate these features in terms of their impact on algorithm and application performance.