Fig. 5: Implementation of the somatosensory spiking network on neuromorphic chip.
From: Bioinspired spiking architecture enables energy constrained touch encoding
a Design of the tactile system implemented on the DYNAP-SE processor: FBG signals are converted to spike trains and injected into the chip. b Raster plot of the activities of the neurons integrated on the chip. c, d Raw and filtered activities of the neurons in the output layer during indentation. e Cumulative distribution function of the performed localization error for the implementation on the neuromorphic chip with (orange) and without (blue) filter application, and for the software implementation with (green) and without (red) weight quantization. Shaded regions represent standard deviations of the CDF estimated via blocked bootstrap. f Histogram of the power consumption of the network during inference and relative contribution of spike generation, transmission and post-synaptic integration mechanisms to the total power consumption of the chip.
