A Data Path Design Tool for Automatically Mapping Artificial Neural Networks on to FPGA-Based Systems

Abstract

-Artificial Neural Networks (ANNs) are usually implemented as software running on general purpose computers. On the other hand, when software implementations do not provide sufficient performance, ANNs are implemented as hardware on FPGA based systems for performance enhancement. Mapping ANNs to FPGAs is a time consuming and error prune process. In this study, a novel data path design tool, ANNGEN, has been proposed to help automate mapping ANNs to FPGA based systems. ANNGEN accepts ANN definitions in a NetList form. First, it parses and analyzes given NetList. Second, it checks the availability of the neurons. If all the neurons required by the NetList are available in its neuron Library, ANGENN performs the design procedure and produces VHDL code for the given NetList. ANNGEN has been tested with several different test cases, and it is observed that it is able to successfully generate VHDL codes for given ANN NetLists. Our practice with ANNGEN has showed that it effectively shortens the time required for implementing ANNs on FPGAs. It also eliminates the need for expert people. Additionally, ANNGEN produces error free code; thus, the debugging stage is also eliminated.