Hardware Implementation and Optimization of the Sigmoid Activation Function in CNNs Using Taylor Series Approximation

The activation function is crucial in artificial neural networks for transforming inputs into outputs, introducing nonlinearity necessary for learning intricate patterns and making precise predictions. This study focuses on the hardware implementation of an optimized sigmoid activation function in convolutional neural networks (CNNs). The sigmoid function, commonly used in classification tasks, provides a smooth and bounded output. We propose a hardware optimization approach for the sigmoid function using Taylor series approximation to address challenges related to computational efficiency and accuracy. Our results, obtained using ModelSim software, demonstrate that the proposed high-precision sigmoid circuit closely replicates the ideal sigmoid, achieving a maximum error of 1.05% and an average error of 0.18%. Future research will aim to further enhance accuracy and efficiency for real-time CNN applications.