Adaptive-Learning Synaptic Devices Using Ferroelectric-Gate Field-Effect Transistors for Neuromorphic Applications

Abstract

An adaptive-learning ferroelectric neuron circuit is proposed and fabricated on a silicon-on-insulator structure, which is composed of a metal-ferroelectric-semiconductor field-effect transistor (MFSFET) and an oscillation circuit as an artificial synapse and neuron devices, respectively. Typical oxide ferroelectric SrBi₂Ta₂O₉ thin film is selected as a ferroelectric gate insulator for the MFSFET. The synapse MFSFET shows good memory operations and gradual learning effect. The drain current is gradually modulated with increasing the number of input pulses with a sufficiently short duration. The output pulse frequency of the fabricated neuron circuit is also confirmed to gradually increase as the number of input pulses increased. The weighted-sum operation is realized by constructing the synapse array composed of the MFSFETs. The output pulse performance including the pulse amplitude and time-dependent stability are improved by employing Schmitt-trigger oscillator and metal-ferroelectric-metal-oxide-semiconductor gate stack structure, respectively.