Engineering more power-efficient phase change memory devices

Researchers are constantly looking for new and better ways of storing the 0’s and 1’s of computer memory. One idea has been to represent bits as different atomic structures in a material.

This “phase change memory” would not need constant power to store information, but could switch back and forth much faster than other kinds of memory that rely on physical traits. This kind of “nonvolatile” memory could enable it to be used as the basis of computation, not just storage.

Researchers at the School of Engineering and Applied Science, led by Ritesh Agarwal, a professor in the Department of Materials Science and Engineering, have made a step forward in this effort in a recent paper. Along with Pavan Nukala, one of his students, Agarwal collaborated with fellow materials science professor Russell Composto to make model phase change memory devices that are more power-efficient than ever before.  

Phase change memory relies on the fact that different phases in certain materials give rise to different levels of electrical resistance. Carefully ordered, crystalline lattices of atoms allow electrons to flow freely, while disordered, amorphous regions act like insulators. Finding the right materials and methods to switch back and forth from these phases is the key to making usable memory devices.

Prior to earlier research by the Agarwal group, this phase change process was only thought to be possible through rapidly melting and cooling the material. This process has several drawbacks, however. Getting the material hot enough to partially melt requires a lot of energy, and, being densely packed on a chip, the ambient heat from one device could also accidently switch neighboring devices as well. Most critically, constant melting and recrystallizing eventually causes devices to degrade so much as to become useless. These issues have precluded the widespread application of phase change materials in nonvolatile memory applications.

Agarwal’s alternative was to use short, rapid pulses of electricity to more gradually induce disorder. Tested in germanium telluride nanowire devices, this method showed promise solving the problems associated with melting, but still required more energy than was feasible to use in computers.  

His group’s new collaboration with Composto increases the efficiency of this method by adding some disorder to the nanowires before they are fashioned into memory devices.

“We thought if we could get the nanowires right on the edge of stability, it would only take a little nudge to get them to amorphize,” Agarwal says.

By bombarding phase change material with helium ions from the Laboratory for Research on the Structure of Matter’s particle accelerator, the researchers were able to introduce a carefully controlled amount of disorder throughout the nanowires. This reduced the amount of electric power needed to switch from crystalline to amorphous phases.

“Another electrical pulse can get the amorphous region to recrystallize, but there is still a disordered template surrounding it,” Agarwal says. “That means we can switch it back and forth, as demonstrated in our work.”

The team’s experiments showed orders-of-magnitude increases in efficiency over tens of thousands of cycles, traits needed to get such phase chance memory devices out of the lab and into practical settings. 

Next up for the group is moving their experiments from nanowires to thin films, which is how such memory devices would eventually be mass-produced.

Phase change memory