FPGA Realization of Hyperbolic Function

Abstract

Machine Learning (ML) is applied in many real world applications. Artificial Neuron (AN) is the main building block of Artificial Neural Networks (ANN) which is the backbone of ML. ML algorithms for different applications like image and speech recognition etc., speech with large data are implemented on workstations with GPUs. However, the power of ML techniques for real time applications like monitoring, control and diagnostics, typical in power and process plant is still in nascent stage. Most of these applications require FPGA realization of ANNs. Realization of AN with both sigmoid and hyperbolic tangent activation functions along with results is reported in this paper. Design of building blocks consisting of Floating Point Unit (FPU) and Function approximation (FA) are presented. Implementation using Verilog HDL along with results of accuracy with simulated inputs and gate count and timing diagrams are presented.