A Survey on Efficient Realization of Activation Functions of Artificial Neural Network

Now a day's most of the applications utilize services of Artificial Neural Networks (ANN). Due to overwhelming adoption of ANN based technologies in practical applications, a lot of research has been done for implementation of ANNs on Hardware. Implementing an ANN on hardware is considered as a complex and challenging task. One of these challenges is the design of the neuron's activation function module. Several researchers of the domain have proposed various methods for implementation of activation function (AF). Generally, hardware implementation of a nonlinear AF is done by one of the approximation or LUT based methods. The present article briefly discuss both of major approaches and subsequent developments in methods employed for hardware implementation of such AF. Also, a discussion related to the hardware architecture and associated Hardware, time complexities has been done.