ANN in Hardware with Floating Point and Activation Function Using Hybrid Methods

Artificial neural networks are bio-inspired models used mainly in the problems solution with nonlinear behavior. Reconfigurable devices (FPGA) are widely employed in the implementation of artificial neural networks. The contributions of this work is in the implementation of a Multilayer ANN with four neurons, one hidden layer and hyperbolic tangent for activation function. A nonlinear function approximation performed for system validation of artificial neural network. This paper makes analysis of nine scenarios with several methods to implementation of the activation function in ANN. Two hybrid methods are used in the approximations related to PWL method, Combinational and RALUT. Results are compared between distinct scenarios and the activation function with recent literature. The results were examined by performance, FPGA area and errors absolute and relative. Activation function has achieved the best average error in 10−4 and the ANN in 10−3. ANN system is suitable for target application and with portability to others platforms.