Space-and-Cost-Efficient Neural Control /Sensory Element Using an Analog FPGA

As neural networks are applied to control and sensory, software neuron models cannot sometimes fulfill speed requirement as well as simple add-and-sigmoid is not enough for functionality. This paper proposes a small and inexpensive hardware neuron based on FitzHugh-Nagumo model. By making full use of chip property and maximum circuit packing through placement search, it surpasses the previous implementation by factors of 4 for space and 15 for cost.