Field Programmable Neural Array for feed-forward neural networks
Marek Bohrn, Lukáš Fujčík, Radimı́r Vrba · 2013
This paper focuses on a novel circuit suitable for implementation of feed-forward artificial neural networks. Structure of the circuit is derived from the concept of Field Programmable Neural Array - Field Programmable Neural Network. The aim of the paper is to describe an innovative promising approach to fast hardware implementations of feed-forward artificial neural networks and introduce some special techniques that was used in the design of this circuit and its optimization. The proposed circuit facilitates implementations of all common used topologies of feed-forward neural networks and covers a wide spectrum of applications of neural networks. The circuit is optimized to be implemented into new FPGAs from Xilinx, primarily families 5 and 6 and it also can be implemented into ASIC.