An alternative polynomial approximation for the activation sigmoid function is developed here. It can considerably simplify the input/output operations of a neural network. The recursive algorithm is found for Chebyshev expansion of all constituting polynomials.