MACHINE LEARNING GUIDED DEFECT STUDY OF ELECTROCHEMICAL NANOSTRUCTURED MATERIALS FOR THE NEUROMORPHIC APPLICATION
Abstract
Energy storage materials for thin film batteries (e.g. amorphous lithium lanthanum titanate, lithium titanate) have high ionic conductivity for the neuromorphic memory devices. The structure and properties of these materials under electrochemical conditions are different and these characteristics offer great potential for stabler, safer, more efficient memristive devices which consume less power. Meanwhile, defects in these materials play a critical role in resistive switching properties of solidstate devices. Mesoscopic characterization and defect engineering on these materials are critical to the mechanistic understanding, especially when defects in these electrochemical energy materials are prone to vary under electrical bias, it is thus important to perform in situ characterization on defect properties to determine the property accurately. Neuromorphic device performances including stability and switching voltage will be also predicted based on the predicted binding energy and defect properties. A combination of advanced mesoscopic characterization tools and machine learning methodology will provide the pathway to energy materials for the neuromorphic computing application. Three key aspects will be enfolded: 1) Probing defect states in neuromorphic materials at both the atomic and mesoscopic scales; 2) Machine learning methodology for optimizing neuromorphic materials; 3) Neuromorphic performance evaluation via fabrication of electrochemically smart devices. We aim to discover new materials suitable for the neuromorphic devices based on the electrochemical energy materials. By combining in-situ mesoscopic characterization tools with machine learning guided methodology, we will predict the properties required for the neuromorphic application (e.g., defect states, structural configurations, phases, etc.). Performance of neuromorphic devices such as stability and switching voltage using newly discovered materials will be predicted by density function theory calculation and will be evaluated in device implementation. Our combined advanced characterization/computational methodology will guide and accelerate these novel materials to the neuromorphic application.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 10, 2022
- Source ID
- FA23862114055XX0
Entities
People
- Ying Meng
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of California, San Diego