Y2O3-based memristors: From model devices and arrays toward predictive synaptic computing

• Applicant: Professor Dr. Lambert Alff
• Subject Area: Electronic Semiconductors, Components and Circuits, Integrated Systems, Sensor Technology, Theoretical Electrical Engineering ; Synthesis and Properties of Functional Materials
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 546570457

Oxide-based RRAM devices have great potential in artificial neural networks (ANNs) as they are able to imitate the electrical behavior of biological synapses. In order to fully realize their potential, however, a deep understanding of the underlying physical processes and their relation to device characteristics is necessary. This can only be achieved in a synergistic approach, where materials engineering, advanced electrical characterization techniques, as well as a device- and system-level simulations meet. The main goal of the proposed project is the creation of compact models to be implemented for neuromorphic computing, based on model devices and arrays fabricated in a dedicated way to facilitate the modeling with step-wise increasing complexity. With this, we aim to fill an existing void in the simulation of neural networks, where most of the time characteristics from only a few memristors are taken as the basis of software-driven ANN accuracy modelling. A key feature of our approach is to cross over from a physical to a behavioral model as we go from individual devices to an array description. Note that the suggested modelling allows a correlation of device characteristics to materials properties such as oxygen vacancy concentrations. It will be therefore a milestone to allow feedback from array modelling to materials properties, thereby, providing important information to material optimization and understanding of the functionality of the materials. Such materials-related behavioral predictions based on measurements from real arrays are indeed worthy of investigation, as such studies are almost completely absent in the field as of the time of writing. We will perform this study using Y2O3 as functional oxide as it offers - due to intrinsic oxygen vacancies - a large number of intermediate resistive states, necessary for analog computing. We will deal with the accurate selection, control, and material characterization required for the fabrication of individual and combined devices, as well as the corresponding study of their behavioral aspects such as stability, variability, endurance, switching mechanisms, resistance window, connectivity properties, etc. The characterization and modelling will focus in collaboration with our Taiwanese partners on the ac and sequential pulsed measurement with different amplitudes and time scales. Based on these large data sets gained at the National Cheng Kung University in Taiwan we will develop a predictive model for arbitrary synaptic functions. The predictive power of the model, in turn, will be validated in suggested array structures. At the same time, the modelling will give feedback on the required materials properties based on measurements on systematic sample series. This proposal will therefore contribute to establish a methodology to correlate directly materials properties and neuromorphic behavior which can be extended to other materials and memristive technologies.

• DFG Programme: Research Grants
• International Connection: Taiwan
• Partner Organisation: National Science and Technology Council (NSTC)
• Cooperation Partner: Professorin Dr. Jen-Sue Chen