Ferroelectric neuron circuits with adaptive-learning function

A neon circuit which consists of nonvolatile metal-ferroelectric-semiconductor field effect transistors (MFSFETs) and a uni-junction transistor (UJT) has been proposed. In the proposed circuit MFSFETs act as analog memories to store the synaptic weights which can be changed by the adaptive-learning process during the operations. In this paper, we first simulate the operation of the ferroelectric neuron circuit using a circuit simulator, SPICE. It is shown that the output frequency of the proposed neuron circuit can be changed after it processes a certain number of input pulses. Then, we report the fabrication of UJTs and UJT pulse oscillation circuits using silicon-on-insulator (SOI) substrates. It is found that the output frequency increases with decreasing the charging time of the capacitor in the circuit and that the operation at higher frequencies is possible for integrated UJT oscillation circuits. Finally, we demonstrate the memory and learning properties of n-channel ferroelectric-gate FETs using (Pb,La)(Zr,Ti)O₃ (PLZT) films. It is shown that the drain current of the PLZT/SrTiO₃/Si FETs can be controlled by a “write” pulse before the measurements.