Hyperbolic tangent activation function integrated circuit implementation for perceptrons

Abstract

Activation function refers to the mathematical function implemented in multilayer perceptrons to limit the output signal of neurons. The objective of this project is to design a hyperbolic tangent activation function (TANH) for integrated circuit implementation using digital design approach. This paper compares the performances of area, speed and power between ring architecture and tree architecture of activation function using repetitive instantiation of multipliers and adders. TANH function. The project scope covers digital design using Verilog hardware description language followed by functional simulation and synthesis using Silterra 0.18 μm technology files. The activation function performed the TANH function mathematical operation with ±0.02% accuracy. It is shown that the tree architecture performs better than ring architecture with lower propagation delay. On the other hand, ring architecture achieves lower power dissipation.