FULLY DIGITAL ARTIFICIAL NEURON HARDWARE

Analysis of artificial neuron models is performed for hardware implementation. A method of synthesis polynomial threshold gate is suggested, which gives optimal values of weights and threshold in terms of hardware implementation. New paradigm of understanding artificial neuron as a logic gate is developed, which helps to implement artificial neuron hardware by multiplexor. This kind of model for artificial neuron has a lot of advantages over the known ones in terms of hardware implementation. Supervised learning algorithm is developed for multiplexor based neuron, which is simpler compared with the known learning algorithms, and also conversion from threshold neuron model to multiplexor neuron model is developed, which helps to use the known learning algorithms for developed neuron model. Finally multiplexors with 3 different structures is developed in 28nm technology and simulated over PVT by HSPICE, as well as comparison between them is performed.