A NEW METHOD FOR FPGA IMPLEMENTATION OF ARTIFICIAL NEURAL NETWORK USED IN SMART DEVICES Stefan Oniga Lecturer Eng Nprth University o/Baia Mare, Romania Smart devices development with leaming capabilities and adaptive behavior is a need of these days. The implementation of such devices is possible using artificial neural networks (ANN). The present work shows a new, efficient and rapid method to design, train and implement in FPGA neural networks. lSystem Generator tool for Simulink is for ANN design using neural networks specific blocks, created by author. The Matlab is used to perform the off-chip leaming task. System Generator also allows the easy generation of hardware Description Language (HDL) code from the system represented in Simulink. The VHDL design can then be synthesized for implementation in the XiliIL'<. family ofFPGA devices. Keywords: smart, neural network, adaptive, learning, FPGA, VHDL INTRODUCTION Nowadays the development of intelligent and more natural smart devices, without need of knowledge for parameters setting activity, is attracting the interest of many research groups worldwide. The need to have learning capacity and adaptive behavior for such smart devices can be satisfied using neural networks and FPGA implementation is an easy an attractive way for hardware implementation Among possible applications are intelligent computer peripherals enabling people with any kind of handicap to use computer and communicate, as any kind of industrial or domestic device with leaming and adaptive capabilities. The goal of this work was to develop hardware-software codesign platfol1n enabling the fast development of smart interfaces using: - Applieation specific sensors, - hardware modules that ean be easily connectcd - VHDL modules that can manage sensors, - Artificial Neural Networks (ANN) used for example for featurcs extraction, pattern recognition, etc. Using this framework development of new smart devices needs only design anel synthesis of new VHDL drivers for the new sensors and new application-specitic ANNs. THEMETHOD The integrated hardware-software environment represents a new framework for hardware implementation of the Artificial Neural Networks and could be used for example for: 1. Development of ANN specific blocks in Simulink Xilinx blockset 31