The study led by the University of Michigan has shed light on how memristors retain information without a power source using phase separation and oxygen diffusion. Despite previous discrepancies between experiments and models, the researchers found that nonvolatile memory can be achieved through the combination of these two processes. By focusing on resistive random access memory (RRAM), specifically a filament-type valence change memory, the researchers were able to uncover the mechanisms behind nonvolatile memristor memory, which is crucial for energy-efficient artificial intelligence applications.
In the RRAM studied by the researchers, a conductive filament forms between two platinum electrodes separated by an insulating tantalum oxide layer. By applying different voltages, the cell can be switched between low resistance (representing a “1” in binary code) and high resistance (representing a “0”) states. Contrary to previous beliefs that oxygen diffusion was the key factor in retaining information over time, the study revealed that phase separation plays a crucial role. Oxygen ions prefer to stay away from the filament, ensuring that it will not diffuse back even after an extended period, similar to how oil and water do not mix due to lower energy in a de-mixed state.
To test retention time, the researchers accelerated experiments by increasing the temperature, finding that one hour at 250°C is equivalent to about 100 years at 85°C, the typical temperature of a computer chip. Using atomic force microscopy, they were able to visualize the nanometer-sized filaments within the RRAM device and observed that different filament sizes exhibited different retention behaviors. Filaments smaller than 5 nanometers dissolved over time, while those larger than 5 nanometers strengthened, a phenomenon that cannot be explained solely by diffusion.
Experimental results and models incorporating thermodynamic principles demonstrated that the formation and stability of conductive filaments in RRAM are influenced by phase separation. By leveraging phase separation, the research team was able to extend memory retention from one day to over 10 years in a radiation-hard memory chip built for space exploration. The implications of this research extend beyond memristors, with potential applications in in-memory computing for energy-efficient AI, electronic skin for prosthetics or wearable devices, and tactile sensing for robots performing delicate tasks.
The study was a collaborative effort involving researchers from Ford Research, Oak Ridge National Laboratory, University at Albany, NY CREATES, Sandia National Laboratories, and Arizona State University. The device was built at the Lurie Nanofabrication Facility and studied at the Michigan Center for Materials Characterization, with primary funding from the National Science Foundation. The findings from this study have the potential to inspire new ways of utilizing phase separation in the design and development of information storage devices for various applications.
