The impact of neural model resolution on hardware spiking neural network behaviour

摘要

This paper contributes to the development of the proposed EMBRACE mixed-signal, reconfigurable, Network-on-Chip based hardware Spiking Neural Network. EMBRACE-FPGA is an FPGA-based prototype of the proposed EMBRACE architecture. Results from successful evolution of an EMBRACE-FPGA SNN robotics controller are presented. Noise in best fitness plots for a range of evolved EMBRACE-FPGA based SNN applications, including the robotics controller, have been observed. This paper investigates the sources of neural noise, and considers their impact in evolving digital-based hardware SNNs. The paper considers the expected performance benefits of the EMBRACE analogue neural cell.