High Speed and Flexible Configuration Neural Network ☆

The research presents the neural cell design with supervised learning method adapting to many algorithms which require high speed and accuracy in this paper. Based on the supervised learning method and real neural structure, we built an artificial neural architecture which can process real numbers. This architecture easily increases the speed by expanding the floor numbers modeled on pipeline structure. To ensure high speed and accuracy we try to optimize a part of a processing and training architecture. The architecture easily extends and controls many applications by configuring network parameters on the FPGA. The research group’s results are very positive when the the network has 30 cells with 89379 LUTs and 92761 registers based on 28nm technology. Operating frequency can be reached to 214 Mhz.