Back-End-Of-Line Integration of 2-Terminal Ferroelectric Hafnia Synaptic Weights on Commercial CMOS Process

Abstract

Functionalising the back-end-of-line (BEOL) of CMOS with emerging materials is a key route to neural network accelerators and neuromorphic systems. In this work, we demonstrate for the first time the integration of 2-terminal ferroelectric synaptic weights in the BEOL of as-delivered commercial 180 nm CMOS dies. A novel material stack is proposed, in which a WOx/[HfO2-ZrO2] bilayer is sandwiched between TiN and Mo electrodes. For devices annealed at 450 °C, programming below 1.8 V is achieved with a cycle-to-cycle variation below 3 %. Synaptic weights annealed at 500 °C exhibit more gradual weight updates, while the tested underlying NMOS transistors remain functional after the complete integration process. The trade-off between thermal budget and on-device training performance is illustrated using hardware-aware neural network simulations.