TaO_x-Based Memristor with Nanostructured Electrode towards Neuromorphic Computing

In-memory computing technology has garnered significant attention in recent years due to its integration of memory and central processing within a single unit cell, along with its ability to efficiently process tasks in parallel like the human brain. This technology is anticipated to subvert conventional computer architectures.<b> </b>Memristors, a type of passive two-terminal electronic component that exhibits a unique resistance change in response to the time history of the applied voltage or current, play a crucial role in the development of in-memory computing technology due to their non-volatile characteristic, low power consumption, fast switching, and high-density integration. In this research, tantalum oxide (TaO_x) is applied as the switching layer of our memristors because of its controllable resistive switching characteristics. The challenge of the memristor we trying to overcome in this research is the random growth of filaments, which is the main cause of the large variations observed in different cycles or different devices. This performance variation is one of the major obstacles to hindering the advancement of memristor in practical applications. In this presentation, we report on two approaches to mitigate the variation issues. We first focus on the engineering of the oxygen vacancy defect density in the TaO_x switching layer by controlling the oxygen partial pressure in the reactive sputter gas. XPS was utilized to examine the oxygen vacancy concentration of TaO_x switching layers. Next, the electrode’s structure was rationally modified to further reduce the performance variation. The TiN bottom electrode was well defined as nano pyramid pattern. We propose that these purposely introduced nanoscale features can induce periodic higher electric fields across the device, which in turn significantly reduce the randomness of filament growth. Finite element analysis was employed to theoretically validate the hypothesis about the nanostructure effect on the electric field and filament growth. Electrochemical impedance spectroscopy was also used to verify the filamentary switching mechanism of our TaO_x-based memristors. In addition, through pulse measurements, we demonstrate synapse behaviors, i. e. multi-conductance levels of our TaO_x-based memristors. Finally, a neural network simulation was conducted utilizing the attributes of TaO_x-based memristors, demonstrating their capability of image recognition abilities when applied to the Fashion Modified National Institute of Standards and Technology database.