Two Novel Generic, Reconfigurable Neural Network FPGA Architectures

Two novel generic, scalable, and reconfigurable neural network architectures implemented using field programmable gate arrays (FPGAs) are presented in this paper. Previous Implementations of feed-forward Neural Networks face two major issues: 1) Limited resources available on the FPGA compared to the large number of multiplications required by Neural-Networks, 2) Limited reusability of the design when applied to the Neural-Network applications with different architectures. Our proposed implementations circumvent both issues. The designs' scalability allows the user to program and implement different applications with variable number of neurons, starting from one neuron to the maximum number of neurons in any layer, this is performed with programming-like ease and flexibility. A GUI was implemented to allow automatic configuration of the processors for different applications. Finally, Propositions for future work are outlined.