A controller design tool development for automatically mapping neural networks onto FPGAs

Artificial Neural Networks (ANNs) are small copies of human brain and are used in several different areas efficiently. Hardware implementations of the ANNs are preferred when software implementations do not provide desired performance. ANNs can show their real performance on parallel hardware architectures and using FPGAs (Field Programable Gate Array) is a suitable implementation choice for parallel applications. As a result, FPGAs are good candidate for implementing ANNs hardware. On the other hand, implementing ANNs on FPGAs is time consuming process and requires expert personal. Moreover, debugging is required for the error happens during the implementations process. In this study, ANNCONT (Artificial Neural Network Controller), a controller design tool, and ANNSYS (Artificial Neural Network System), a top level ANN system design tool were developed for automatically mapping ANNs to FPGAs. The purpose of the study was to eliminate the debugging state of the design and implementation process and to eliminate the need for the expert personal and to shorten design and implementatıon time of ANNs on FPGAs by automating the whole mapping process. A couple of test cases were developed for testing ANNCONT and ANNSYS design tools. The tools were applied to the test cases and VHDL (Very High Speed Integrated Circuit HDL (Hardware Description Language)) codes were produced in seconds. The effectiveness and correctness of the tools were proved by synthesizing the automatically produced VHDL codes using through Xilinx?s ISE tool.