Design Methodology of a Fully Parallelized Neural Network on a FPGA

Abstract

In this work a methodology of a parallelized neural network has been designed. It explains a way to design a Neural Network using Mathworks and Xilinx Tools. Initially, the floating point algorithm was evaluated using MatlabNeural Network Toolbox. Afterwards, the fixed point algorithm was designed on a Field Programmable Gate Array (FPGA).The architecture was fully parallelized. The design tool used is System Generator of Xilinx, which works over Simulink. Finally the System Generator design is compiled for Xilinx Integrated System Environment (ISE).