Virgil Watkins
Research Mentor(s): Yiyang Li
Research Mentor School/College/Department: Materials Science and Engineering
Presentation Date: 08/03/2022
Presentation Type: Poster
Poster Number: 50
Session: Session II: 1:30 – 2:20pm
Room: League Ballroom
Authors: Virgil Watkins, Laszlo Cline, Yiyang Li, Diana Kim, Jingxian Li
Abstract
In today’s world of big data current methods of computing are nearing their physical limits and the energy used to do so is increasing at a staggering rate. Alternatives exist, such as analog memory based computing which can be more energy efficient for the data intensive tasks, we see current machine learning algorithms face. However, this type of computing has been limited by the device’s inability to accurately change its resistiveness and henceforth the device’s inability to accurately change its memory. This computing has also been limited by its volatility, the resistiveness of the device drifts over time and therefore the memory stored on the device changes as well. This project addresses both of these concerns. Bulk diffusion is a statistically deterministic method for moving ions within a solid. Due to this we used the bulk diffusion of ions to change the device’s resistiveness as a means to address the issue of unpredictable and inaccurate switching between resistive states. In order to create a nonvolatile device that can retain its memory for a long time we began to use tantalum oxide. By using tantalum oxide within our device, we are able to take advantage of the thermodynamically driven phase separation of two phases within the material, allowing us to achieve a nonvolatile memory. Through experimentation we have been able to create and demonstrate a nonvolatile analog memory based computing device that switches accurately and predictably. Furthermore, on a more abstract scale we have demonstrated that bulk diffusion and phase separation can be used to engineer nanodevices with predictable and stable behavior.
