The rapid integration of artificial intelligence (AI) into various aspects of daily life has significantly increased energy consumption, largely due to the inefficiencies of the traditional von Neumann architecture that requires constant data shuttling between memory and processing units. In response, this dissertation explores brain-inspired analog and in-memory computing, aiming to mimic the human brain's efficiency by merging computing and memory and thus empowering sustainable and energy-efficient AI. 

Aim 1 examines the physics of adaptive oxide synaptic devices through a compact model that effectively emulates experimental findings and studies the effect of titanium doping on the resistance change mechanisms of these devices. 

Aim 2 focuses on improving the performance of adaptive oxide synaptic devices. Abrupt resistance changes during operation are addressed by a thin barrier layer that controls oxygen ion movement. Furthermore, MAX phase materials are utilized in these devices for enhanced endurance, reduced off-state current, higher on-off ratio, and multi-state capability.

Aim 3 extends the application of these findings to optimize system-level performance in an analog brain-inspired chip for deep learning. It led to the development of a novel electrical biasing technique, ReSta, which effectively restores the performance of analog AI hardware post-fatigue and the demonstration of transfer learning on analog AI hardware.

Overall, this dissertation aims to advance brain-inspired computing, offering new insights and methodologies at the material, device, and system levels, thus paving the way for more robust, efficient, and sustainable AI.