Fabia Farlin Athena

Research Abstract:

In the contemporary data-driven landscape, Artificial Intelligence (AI) has emerged as a pivotal technology in shaping our daily lives. However, its increasing utilization raises concerns regarding sustainability due to rising energy requirements that far exceed the energy efficiency gains from standard computing architecture. My research centers on understanding and enhancing brain-inspired AI, particularly nanoscale synaptic devices, as a low-power alternative that integrates memory and processing, thereby mimicking the brain’s efficiency. By exploring the mechanisms and applications of these devices in deep neural networks (DNNs), I aim to contribute to the development of more sustainable AI technologies. My work involves optimizing materials, device fabrication, material, and electrical characterization, and modeling of adaptive oxide-based synapses, especially Resistive Random Access Memories (ReRAMs). I focus on their behavior under various conditions, such as the impact of doping. Through experimentation, I have identified optimal doping conditions that lead to low-power synapses with reduced forming voltage improving energy efficiency, and I have conducted extensive electrical characterization to understand their analog synaptic performance. Delving deeper into the workings of adaptive oxide synapses, I have developed a compact computational model, C-STAO, to elucidate their mechanisms under analog pulses which guided the design of a barrier layer at a critical location in the device stack for enhanced performance. Further experimentation showed the viability of the barrier layer addition and it showed improvement in analog resistance change. I have also proposed the utilization of an emerging material with unique thermal and electrical properties, to improve critical device properties such as off-state current and dramatically reduce energy consumption. My research also extends to practical applications of these technologies, as demonstrated during my internships and collaborations with IBM TJ Watson Research Center, where I studied the integration of adaptive oxide synapses into DNNs. There, I developed a novel electrical biasing technique, ReSta, to recover the deep learning performance of these adaptive oxide devices over repeated usage, and demonstrated transfer learning in an analog brain-inspired AI chip. Looking ahead, my goal is to work toward emerging nanoelectronics for low-power, sustainable AI solutions, focusing on reducing energy usage, broadening access, and generating positive societal impacts.

Bio:

Fabia Farlin Athena is an Electrical and Computer Engineering (ECE) Ph.D. candidate at Georgia Institute of Technology, advised by Professor Eric M. Vogel. She also received her MS in ECE from Georgia Tech. Fabia's Ph.D. research focuses on studying innovative materials and methodologies that promote energy-efficient computing. During her Ph.D., she has also interned twice at IBM T.J. Watson Research Center. Before starting at Georgia Tech, Fabia worked at Purdue University for two semesters as a graduate researcher and collaborated with Idaho National Lab on nuclear materials for next-generation energy. Fabia completed her BS in Materials Science and Engineering at the Bangladesh University of Engineering and Technology, graduating second in her class. Her research has been recognized with the Georgia Tech ECE Ph.D. Fellowship, Cadence Diversity in Technology Scholarship, MRS Graduate Student Award Finalist, Colonel Oscar P. Cleaver Award for the most outstanding Ph.D. dissertation proposal in Georgia Tech ECE, and the IBM Ph.D. Fellowship.