Hyperbolic Tangent Activation Function Integrated Circuit Design / Fadzilatul Husna Adnan

Activation function is a mathematical function implemented used in neural network to limit the output signal. The objective of this project is design a hyperbolic tangent (TANH) activation function on digital design implementation for integrated circuit (IC). This thesis compares the performances of area, speed and power between ring and tree architecture of activation function. Multiplier, adder and subtraction are designed as submodules in the TANH activation function structure. TANH activation function is approximated up to three orders terms in Taylor series expansion since direct instantiation code is not possible. The project scope is to approach the digital design with hardware description language (HDL) and functioned simulation. The design is synthesized to produce the gate netlist allowing layout implementation process. The activation function successfully performed the TANH mathematical operation with ±0.02 percent accuracy. Tree architecture performs better than ring architecture in terms of speed since it have lower propagation delay by 22 percent and use less hardware by 23.34 percent in the design. Ring architecture has low power dissipation than tree architecture only by 0.005 percent.