About the seminar:

This proposed project aims to develop a new form of memory resistor (memristor) using the spinodal decomposition of V1-xTixO2. Memristors are poised to reduce the energy consumption of artificial intelligence models by modeling neurons in living things and bringing memory onto the CPU. This reduces the need to transfer information between

the RAM and the CPU and allows the computer to permanently learn. Memristors rely on materials with variable resistance to create a nonvolatile memory that mimics neurons by requiring different amounts of input for signals to pass. Many of these memristors are based on oxide materials like TiO2 and VO2 which have issues of memory volatility and temperature control as they base their resistivity on the migration of oxygen vacancies and the

metal-insulator transition (MIT). The oxygen vacancies can spontaneously migrate back into the bulk, as a high concentration of vacancies is metastable, and the MIT of VO2 is at 68 °C, which is difficult to compensate for by cooling under heavy computational loads. By relying on a structural phase transition like spinodal decomposition which is more permanent and is not as sensitive to temperature fluctuations, a longer lasting and more robust memristor could be produced using the same materials. The central question is: How can the spinodal decomposition of V1-XTixO2 be controlled and then used in a memristor? We hypothesize that it is possible to control the spinodal decomposition by utilizing strain induced phase preference at the interface of an electrode and verifying the device works in the predicted mechanism by SEM-EDS, TEM-EELS and modeling. These methods would show if the system

was instead operating via the oxygen vacancy migration. Modeling and variable temperature testing would show if the system was operating on the MIT. This may make a device based on spinodal decomposition that can function as a memristor repeatably and more permanent, by relying on a phase change in the oxide.