Chalcogenide and oxide-based memristor for advanced memory and neuromorphic computing applications

Abstract

Under thewave of artificial intelligence, howto process vast data more efficiently is an emerging issue of information technology. Information processing currently centres on memory units and computing units, following the classical von Neumann architecture. The data transport between memory units and processor units reduces data analysis efficiency and increases energy consumption. To address the requirements for efficient information processing, a brain-inspired architecture, neuromorphic computing has been proposed. The neuromorphic computing architecture relies on stable hardware units with high reliability and low power consumption. To mimic the ability of the brain, neuromorphic computing can perform effective processing for complicated tasks. Memristor devices meet the hardware requirements of neuromorphic computing and also have the characteristic of a simple structure. Among memristive devices, esistive memories have certain advantages in terms of scalability and controllability, providing potential for hardware implementation of neuromorphic computing.
To fulfil the computational requirements in practical environments, more demands are proposed on the switching of resistive memories in aspects such as stable endurance, higher ON/OFF windows, synaptic switching characteristics and learning capability. To achieve these requirements, how to control the switching performance of the resistive memory from material and structural engineering aspects is the focus of this thesis. In this thesis, the control of materials science mainly manifests in double-layer structures and novel porous materials. These methods are utilised to regulate the filament structure of memristors, thereby improving the performance to meet the needs of neuromorphic computing.
An Al2O3/MoO3 bilayer resistive memory is fabricated via an atomic layer deposition process. The bilayer resistive memory exhibits bipolar non-volatile switching and the conductance mechanisms can be explained by space charge limited conduction. The formation and rupture of conductive oxygen vacancy-based filaments are illustrated as the mechanism for the observed resistive switching behaviours. The virtual electrode formed in the Al2O3 helps improve the stability of filaments. Compared to the MoO3 single layer-based resistive memory, the ON/OFF ratio and variability are improved.
Similarly, a hybrid organic-inorganic material-based memristor device is made using chalcogenide material, GeSbTe. The memristor is designed by a fast and simple electrochemical fabrication method. These hybrid memristors require no electroforming process and exhibit reliable and reproducible bipolar resistive switching at low switching voltages. Multistate switching behaviour can also be achieved by controlling the compliance current. The formation and rupture of conductive Ag filaments are explored within the hybrid PMMA/GeSbTe matrix. The PMMA structure serves as a barrier between GeSbTe and the top electrode, reducing the penetration of Ag atoms into the voids within the GeSbTe thin film. Another chalcogenide-based resistive memory is based on the Sb2S3 thin film. The Sb2S3-based resistive memory utilises a single-precursor in-situ solvothermal deposition method, enabling typical non-volatile resistive switching characteristics. The Sb2S3-based resistive memory shows low operating voltage and bipolar resistive switching behaviour. The switching behaviours betweenONand OFF states are based on the formation and rupture of Sb filaments.
A series of mesoporous silica-based memristors use porous materials to investigate how the structure of the electrolyte affects memristor switching performance. These memristors are fabricated via the dip-coating method within a sol-gel process which enables the creation of films with controllable porosities. These films can serve as electrolyte layers in the diffusive memristors and lead to tunable neuromorphic switching dynamics. The mesoporous silica based memristors demonstrate short-term plasticity which is essential for temporal signal processing. As porosity increases, significant changes in operating currents, facilitation ratios, and relaxation times are observed. The underlying mechanism of such systematic control is investigated and attributed to the modulation of hydrogen-bonded networks within the porous structure of the silica layer which significantly influences both anodic oxidation and ion migration processes during switching processes.

More information

Identifiers

ORCID for Ruomeng Huang: orcid.org/0000-0003-1185-635X

ORCID for Ioannis Zeimpekis: orcid.org/0000-0002-7455-1599

ORCID for Kees De Groot: orcid.org/0000-0002-3850-7101

Catalogue record

Export record