Nanocrystal-Embedded-Insulator (NEI) Ferroelectric Field-Effect Transistor Featuring Low Operating Voltages and Improved Synaptic Behavior

Abstract

A novel nanocrystal-embedded-insulator (NEI) ferroelectric field-effect transistor (FeFET) is demonstrated to function as a synaptic device for analog neural network (NN) applications. The NEI layer (down to 3.6 nm in thickness) comprises ferroelectric nanocrystals embedded in amorphous Al2O3, resulting in reduced operating voltages and depolarization effects as compared to conventional doped-HfO2 films. With fixed-amplitude 100 ns potentiation/depression pulses, an NEI FeFET synapse achieves weight update with small non-linearity ( $\alpha _{p}= {0.12}$ , $\alpha _{\text {d}}= -{0.09}$ ) and asymmetry factors, advantageous for analog-style NNs with online training. A convolutional NN is designed and emulated for an MNIST dataset, projecting an online training accuracy of 92%.