Live demonstration: A scaled-down version of the BrainScaleS wafer-scale neuromorphic system

This demonstration is based on the wafer-scale neuromophic system presented in [1], [2] and [3]. One wafer of this system contains 384 analog network chips (ANC) for a total of 40 M synapses and 200k neurons. 12 FPGA boards hosting 48 digital network chips (DNC) [4] connect the wafer to outside stimuli by 768 2 Gb/s serial links. Since it is not feasible to bring a wafer module to the conference, a scaled-down demonstrator system was prepared, containing one FPGA board, one DNC and eight ANC chips (Fig. 1). The FPGA connects to a host PC via Gbit Ethernet and an oscilloscope will be used to visualize the analog signals within the network chips. The total setup will contain nearly one million individually programmable synapses and 4k neurons. Each neuron can be separately configured by 24 analog parameters. The whole system can emulate biological neural networks up to 10 5 times faster than biological real time.