Battery-like computer memory keeps working above 1,000°F

Computer memory could one day withstand the blazing temperatures in fusion reactors, jet engines, geothermal wells and sweltering planets using a new solid-state memory device developed by a team of engineers led by the University of Michigan.

Unlike conventional silicon-based memory, the new device can store and rewrite information at temperatures over 1,100°F (600°C)—hotter than the surface of Venus and the melting temperature of lead. It was developed in collaboration with researchers at Sandia National Laboratory.

“It could enable electronic devices that didn’t exist for high-temperature applications before,” said Yiyang Li, an assistant professor of materials science and engineering and the senior corresponding author of the study published today in the journal Device.

“So far, we’ve built a device that holds one bit, on par with other high-temperature computer memory demonstrations,” Li said. “With more development and investment, it could in theory hold megabytes or gigabytes of data.”

There’s a trade off, however, for devices that aren’t at extreme temperatures full time: new information can be written on the device only above 500°F (250°C). Still, the researchers suggest a heater could solve the problem for devices that must also work at lower temperatures.

The heat-tolerant memory comes from moving negatively charged oxygen atoms rather than electrons. When heated above 300°F (150°C), conventional, silicon-based semiconductors start conducting uncontrollable levels of current. Because electronics are precisely manufactured to specific levels of current, high temperatures can wipe information from a device’s memory. But the oxygen ions inside the researchers’ device aren’t bothered by the heat.

They move between two layers in the memory—the semiconductor tantalum oxide and the metal tantalum—through a solid electrolyte that acts like a barrier by keeping other charges from moving between the layers. The oxygen ions are guided by a series of three platinum electrodes that control whether the oxygen is drawn into the tantalum oxide or pushed out of it. The entire process is similar to how a battery charges and discharges; however, instead of storing energy, this electrochemical process is used to store information.

Once the oxygen atoms leave the tantalum oxide layer, a small region of metallic tantalum is left behind. At the same time, a tantalum oxide layer similarly caps the tantalum metal layer on the opposite side of the barrier. The tantalum and tantalum oxide layers do not mix, similar to oil and water, so these new layers will not revert back to the original state until the voltage is switched.

Depending on the oxygen content of the tantalum oxide, it can act as either an insulator or a conductor—enabling the material to switch between two different voltage states that represent the digital 0s and 1s. Finer control of the oxygen gradient could enable computing inside the memory, with more than 100 resistance states rather than a simple binary. This approach could help reduce power demand.

“There’s a lot of interest in using AI to improve monitoring in these extreme settings, but they require beefy processor chips that run on a lot of power, and a lot of these extreme settings also have strict power budgets,” said Alec Talin, a senior scientist in the Chemistry, Combustion and Materials Science Department at Sandia National Laboratories and a co-author of the study.

“In-memory computing chips could help process some of that data before it reaches the AI chips and reduce the device’s overall power use,” Talin said.

The information states can be stored above 1,100°F for more than 24 hours. While that level of heat tolerance is comparable to other materials that have been developed for re-writable, high-temperature memory, the new device comes with other benefits. It can run at lower voltages than some of the leading alternatives—namely, ferroelectric memory and polycrystalline platinum electrode nanogaps—and can provide more analog states for in-memory computing.