The present work aims at the implementation of a hybrid NPU. For this purpose, an analogue artificial neuron was developed by using electronic components. Then, the control of the neuron s inputs was implemented, by electronic switching circuits, and a circuit for the storage of its outputs values was developed, in order to reuse it as a single analogue neuron, capable of simulate several neurons of a given neural network. Next, it was implemented a digital controller for the number of inputs and the number of neurons per layer, for a network with two hidden layers and one output layer. It is also highlighted that it was developed a dedicated hardware, optimized by genetic algorithms, to implement an activation function that satisfactorily fits a sigmoid. Computational simulations of the developed electronic topology have been carried out. A MLP neural network was optimized by software and its obtained weights were applied to the developed NPU, in order to evaluate how close the network obtained by software is modeled by the proposed NPU. Additionally, the NPU performance was evaluated and possible future improvements are suggested.