Mixed-Mode Artificial Neuron for CMOS Integration

A new approach to designing artificial neurons in CMOS technology is proposed in this paper. Design and simulation results of the basic building blocks are presented. Programmable weights are obtained using a current-mode mixed-signal four-quadrant multiplier, whereas the non-linear output function is implemented with a specifically designed class AB current conveyor. Starting from circuit simulation results, the behaviour of the proposed neuron was modeled. A multilayer perceptron network implemented with the new artificial neuron structure was trained to tackle linearization of a giant magneto-resistive sensor. Simulation results show the efficiency of the new implementation