Currently, neural and fuzzy systems are methods which have found wide application area. Implementation of these systems on a hardware platform providing their own features is important. With parallel data streaming and processing features, FPGAs have become a preferable hardware platform for implementing of neural and fuzzy systems. In hardware implementation of these systems, the cell activation function, which is the most important unit, is of key importance. In this study, implementation mathematical approximations of commonly used logarithmic sigmoid, hyperbolic tangent sigmoid and Gaussian activation functions on FPGA using single-precision floating-point number format is investigated. For the each function, suitable approach to implement on FPGAs is comparatively given and obtained the actual synthesis results from the Xilinx Virtex 5 xc5vlx110- 3ff1153 FPGA are presented. Obtained experimental results show that proposed implementation approaches have consumed very little hardware resources. Using these proposed approaches, neural and fuzzy systems in various structures can be implemented.

-