Ferroelectric domain walls

: new 2D functional materials for conventional and neuromorphic computing

Abstract

Neuromorphic computing seeks to perform data-centric computational tasks in an energy-efficient manner by mimicking the way living systems process information. Artificial synaptic devices are key building blocks for this emerging technology. Many efforts have been made to realise biological synaptic behaviour using different device technologies. Among these, ferroelectric tunnel junction-based devices have been prominent, partly due to their purely electrical switching mechanism, without oxygen vacancy/metal ion migration or Joule heating. This has made them attractive in terms of device reliability and power consumption.

An alternative approach to generate synaptic behaviour using ferroelectrics (that does not rely on tunnel currents) is by using a system that exhibits enhanced electrical conductance along domain walls compared to the bulk domains. Recently, it has been shown that a thin film LiNbO₃ (LNO) memristor device, with conductance levels spanning almost 12 orders of magnitude, could be realised by altering the domain wall microstructure with different applied electric fields. Different from ferroelectric tunnel junctions, the memristive response in these simple ferroelectric capacitor structures was controlled by altering the total number of conducting domain walls that straddle the interelectrode gap and by altering their inclination angle with respect to the polarisation axis. Early signs of conduction plasticity have been seen in these devices, in which there is a gradual increase in conductivity, induced by applying multiple voltage pulses of equal magnitude. These findings led us to investigate neuromorphic aspects of domain wall transport in this ferroelectric system.

Building on the knowledge established in the earlier reports, this thesis explores the neuromorphic properties of domain wall conductivity in 500 nm thick LNO capacitors and investigates how neuromorphic functions and operating voltages change as ferroelectric film thickness decreases. Two-probe current-voltage measurements supported by atomic force microscopy (AFM) characterisation have shown that the domain wall conductivity in LNO thin films is a complex history-dependent function of electric field and time. Field-time kinetics of microstructural change in this system can replicate important features of synaptic response such as potentiation & depression, spike-rate-dependent plasticity and spike-time-dependent plasticity. Measurements on samples prepared by tomographic AFM (TAFM) and dimple grinding have demonstrated that operating voltages scale almost linearly with film thickness in the 50-500 nm regime. Similar neuromorphic behaviours have been observed in ≈100 nm thick capacitors at about five times lower voltages. Local variations in wall inclination angle resulted in diode-like behaviour in the dc conductance response, enabling half-wave rectification at low frequencies. Boolean "AND" and "OR" logic gates have been experimentally demonstrated using domain wall diodes and resistors. Results show that ferroelectric domain walls in lithium niobate thin films might be useful not only for neuromorphic but also for conventional computing systems.

Date of Award	Jul 2024
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	EU Horizon 2020 Marie Sklodowska-Curie ITN Programme
Supervisor	Marty Gregg (Supervisor) & Amit Kumar (Supervisor)