Improved Cmos-Based Noise-Immune Sigmoid Activation Function for Neural Networks

This work discusses an improved CMOS-based, noise-immune nonlinear sigmoid activation function designed for enhanced neural network performance. The proposed architecture provides a precise, variabilityresistant approximation of the sigmoid function, significantly boosting network efficiency. Implemented using the UMC 180 nm technology node PDK, this compact design offers substantial performance improvements for neural networks, achieving both noise immunity and compactness with a trade-off of only a minimal 1 % increase in power consumption.