Phase Separation Is Key to Energy-Saving AI

Reviewed by Danielle Ellis, B.Sc.Sep 11, 2024

According to a recently published study led by the University of Michigan MaterialPhase separation, the phenomenon where molecules separate like oil and water, works in conjunction with oxygen diffusion to help memristors — electrical components that store data using electrical resistance — retain information even when the power is turned off.

Until now, theories about how memristors store information without a power source, known as non-volatile memory, were not fully understood because models and studies were inconsistent.

While experiments have shown that devices can store information for more than 10 years, models used in the community show that information can only be stored for a few hours.

Jingxian Li, lead author of the study and a PhD graduate from the Department of Materials Science and Engineering at the University of Michigan

The researchers focused on a device known as resistive random-access memory, or RRAM, which is an alternative to the volatile RAM used in classical computers and holds particular promise for energy-efficient AI applications. Their goal was to understand the phenomenon underlying the operation of nonvolatile memristor memory.

In the RRAM under study, a filament-type valence-change memory (VCM), two platinum electrodes are sandwiched between an insulating layer of tantalum oxide. The cell enters a low-resistance state, designated by the binary number “1,” when a specific voltage is applied to the platinum electrodes, creating a tantalum ion bridge that passes through the insulator to the electrodes and allows current to flow.

Another voltage will cause the filament to dissolve as the tantalum ions and returning oxygen atoms react, causing the conductive bridge to “rust” and return to a high resistance state, which corresponds to the binary number “0”.

It was once thought that RRAM stored information over time because oxygen took too long to diffuse. However, a number of studies have shown that earlier models overlooked the importance of phase separation.

In these devices, the oxygen ions prefer to be away from the fiber and never diffuse back, even after an indefinite period of time. This process is analogous to how a mixture of water and oil will not mix, no matter how long we wait, because they have lower energy in the unmixed state.

Yiyang Li, Senior Study Author and Assistant Professor, Department of Materials Science and Engineering, University of Michigan

To estimate retention time, the researchers sped up their experiments by raising the temperature. One hour at 250°C is comparable to about 100 years at 85°C, the average temperature of a computer chip.

The researchers used the extremely high imaging resolution of atomic force microscopy to observe fibers only about five nanometers (20 atoms) wide that formed in the one-micron-wide RRAM array.

“We were surprised that we could find the filament in the device. It’s like finding a needle in a haystack“- added Li.

The research team found that different fiber sizes resulted in variable retention behavior. Fibers smaller than about 5 nanometers disintegrated, while fibers larger than 5 nanometers strengthened over time. Diffusion alone cannot explain the large discrepancy.

Experimental observations and simulations based on thermodynamic principles have shown that the production and stability of conductive fibers depend on the phase separation.

Scientists have used phase separation to extend memory retention from a day to more than a decade in a radiation-resistant memory chip that can survive radiation exposure and could be used in space exploration.

Other possibilities include in-memory processing for more energy-efficient AI applications and memory devices for electronic skin, which is a flexible electronic interface that mimics the sensory characteristics of human skin. This material, also known as e-skin, has the potential to provide sensory feedback for prosthetic limbs, produce new wearable fitness trackers and help robots develop tactile sensitivity for delicate tasks.

Li added: “We hope that our findings will inspire new ways of using phase separation to create information storage devices..”

The study was conducted by scientists from Ford Research in Dearborn, Oak Ridge National Laboratory, the University of Albany, NY CREATES, Sandia National Laboratories and Arizona State University in Tempe.

The device was created at the Lurie Nanofabrication Facility and tested at the Michigan Center for Materials Characterization. The National Science Foundation (ECCS-2106225) provided primary funding for the University of Michigan research.

Magazine reference:

Li, J., and others. (2024) Thermodynamic origin of nonvolatility in resistive memory. Material. doi.org/10.1016/j.matt.2024.07.018