Implementation of PWL and LUT based approximation for hyperbolic tangent activation function in VLSI
S. Saranya, B. Elango · 2014
Hardware implementation of neural networks plays a major role in many applications like digital signal processing, pattern recognition, testing of analog circuits etc. The major building blocks for implementation are adder, multiplier and non-linear activation function. A major challenge is faced in the implementation of activation function.Efficiency of hardware realisation of neural network lies in the activation function. Good approximation scheme used results in efficient implementation of activation function. The efficiency of proposed approximation scheme can be analysed in terms of delay, area and the number of gates used. This hybrid approximation is coded in verilog and is simulated and analysed in cadence digital environment (i.e.) NC launch.