On the Implementation of Fixed-Point Exponential Function for Machine Learning and Signal- Processing Accelerators

The natural exponential function appears in a broad range of conventional and emerging applications, including the activation functions used in deep learning applications. This article presents an optimized implementation of exponential function for variable precision fixed-point negative inputs. Hence, this design can boost the performance and energy efficiency of a vast amount of hardware accelerators and systems.

Method overview

The exponential of sum of two numbers can be written as the product of exponential of the individual numbers, that is, \begin{equation*} e^{x+y}=e^{x} \times e^{y}.\tag{2}\end{equation*}