Digital pulse mode neuron with robust nonlinear activation function

This paper proposes a new type of digital synchronous pulse mode neuron with robust nonlinear activation function. Proposed neuron employs additive Gaussian random noise and exponential averaging for input signals to have stable internal potential. Theoretical analysis and simulations are carried out to verify the feasibility of the neuron. The results show that the activation function is invariable against the change of the number of inputs to the neuron. Experimental multilayer neural network is fabricated with the proposed neurons to perform binary classification problems. Experimental results show that the proposed neuron has comparable performance to other neurons while requiring less hardware and providing faster operation.