Analogue circuit realization of a programmable sigmoidal function and its derivative for on-chip BP learning

In an on-chip Back-Propagation (BP) algorithm learning neuron, both the sigmoidal activation function and its derivative are needed. A novel analog circuit is proposed, which can realize both functions. The neuron can be adapted to various environments by programming the threshold and the gain factor of the sigmoidal function. The nonlinear partition problem is used to verify the operation of the proposed circuit.