Skip to Main Content
Neural networks play a key role in many electronic applications, we can find them from industrial control applications to predictive models. They are mainly implemented as software entities because they require a great amount of complex mathematical operations. With the increasing power and capabilities of current FPGAs, now it is possible to translate them into hardware. This hardware implementations increase both the speed and usefulness of this neural networks. This paper presents a hardware implementation of a particular neural network, the general regression neural network. This network is able to approximate functions and it is used in control, prediction, fault diagnosis, engine management and many others. The paper describes an implementation of this neural network using different hardware platforms and using different implementation for each hardware target. This paper also presents an integrated development environment to produce the final hardware description in VHDL code from the Matlab generated neural network. The paper also describes a simulation scheme to test if the assumptions made to increase the performance of the network have a negative impact on the precision for the particular implementation.