A joint collaboration between Yale and IBM is investigating how novel computer memory devices may offer optimal information storage.

The team, which published its findings in the journal Advanced Materials on Jan. 12, focused on phase-change memory transistors, a class of devices that stores information differently than the Flash transistors typical to common electronic devices. The team’s research focuses on optimizing transistor function.

While Flash transistors depend on floating electron gates to encode binary information, a phase-change memory transistor employs a crystalline structure. By using a crystal whose structure changes significantly when heated, researchers can create transistors able to store information even more effectively than traditional Flash transistors.

“It’s faster and consumes less power,” said first author Terry Xie GRD ’20, a member of professor Judy Cha’s lab. Because crystals can have varying levels of resistance, a single phase-change memory transistor could potentially store more states than just one and zero, which would improve storage capacities significantly, he noted.

Cha, the paper’s corresponding author, said that most computers rely on two types of data storage: fast-acting but volatile memory and slow but secure storage. Because computing is becoming more data-intensive, though, the high price of upgrading memory presents a barrier to improving computing power, she added. Transistors using phase-change memory, or PCM, are attractive candidates for those hoping to create a third tier of computer-data retention beyond memory and storage, called storage-class memory, she said.

But while phase-change memory has promise, it remains a developing technology, Cha said. One of the main issues with current PCM is crystals’ tendency to form ion voids after repeated use, she added. The team’s most recent paper focused on studying this problem.

When a current is applied to a crystal, positive elements of the crystal travel to the anode and negative elements to the cathode, Xie explained. He added that a problem with the specific crystal used in phase-change memory is that different ions move to opposite sides of the crystal at varying speeds, creating voids. Because these voids significantly affect the efficacy of the cells, it is important to understand and avoid them, Cha said.

Yerin Kim ’18, an undergraduate researcher who automated the instrumentation for the project, wrote in a report sent to the News that “TEM allows for a qualitative analysis of cells that would be otherwise impossible.”

Phase-changing crystals tend to form ion voids much more quickly than anticipated, the team found. The in-situ TEM analysis also led the team to discover a way to remedy void formation. By running a current in the opposite direction, the kind of migration that causes voids to form is reversed, neutralizing any imbalance. Currents can be run backwards when the particular transistor is not in use, Xie said.

PCM memory’s impact on the computing field is not limited to storage improvements, Xie said. While many engineers have used a software-based approach in the search to create robust neural networks, a hardware-based approach utilizing PCM is promising, as well, Xie added, noting that scientists at IBM have focused on this approach.

Traditional methods of data storage have come up against physical limits, like the problem of quantum tunneling and minimum architecture constraints, Cha said. If the progress predicted by Moore’s law — which posits that the number of transistors in a circuit doubles every two years — is to continue, it will be necessary to develop new methods of data storage like PCM. Cha said she believes that PCM may become viable in the next decade; still, she noted that the industry’s reliance on flash electron-gate memory may slow progress.

“People spent the past 20 to 30 years creating hardware structures based on Flash,” she said. “It might be difficult to change all at once.”

Josh Purtell | josh.purtell@yale.edu