The University of Southern California’s Viterbi School of Engineering has announced a breakthrough in high-temperature electronics with the development of a graphene-based memory chip capable of operating at 700 degrees Celsius, far exceeding the long-standing thermal limits of conventional semiconductor devices.
The research, published in Science on March 26, 2026, demonstrates a new type of memristor device that remains stable and functional at temperatures higher than those of molten lava. Traditional electronics typically fail at around 200 degrees Celsius, making this advancement a significant leap for applications in extreme environments such as space exploration, geothermal drilling, and nuclear energy systems.
The device is built using a nanoscale “sandwich” structure composed of tungsten as the top electrode, hafnium oxide as the insulating middle layer, and a single atomic layer of graphene at the base. Tungsten provides durability due to its high melting point, hafnium oxide ensures stability as a ceramic insulator, and graphene prevents failure by stopping metal-atom migration that typically leads to short circuits at high temperatures.
In testing, the memristor retained data for more than 50 hours at 700 degrees Celsius without requiring refresh cycles, endured over one billion switching events, and operated at low voltage with nanosecond-level speeds. Researchers noted that the device did not reach its failure threshold during experiments, with testing equipment limitations capping the temperature range.
The discovery was made unintentionally while researchers were experimenting with graphene-based designs for other applications. Subsequent analysis using advanced microscopy, spectroscopy, and quantum simulations revealed that graphene’s unique chemical properties prevent tungsten atoms from bonding and migrating, eliminating a key failure mechanism in high-temperature electronics.
Beyond durability, the memristor also has implications for artificial intelligence computing. Unlike traditional chips that perform matrix multiplication sequentially, memristors can execute these operations directly through physical electrical processes governed by Ohm’s Law. This enables significantly faster and more energy-efficient computation, potentially transforming AI hardware architectures.
The research team has already begun exploring commercialization pathways through a startup, TetraMem, which is focused on developing memristor-based AI chips for practical applications. While the current high-temperature device represents an early-stage prototype, researchers emphasized that further work is needed to integrate logic circuits and scale manufacturing for real-world deployment.
Two of the materials used in the device, tungsten and hafnium oxide, are already widely used in semiconductor fabrication, while graphene production is advancing toward industrial-scale adoption. The research was conducted as part of USC’s Center for Neuromorphic Computing under Extreme Environments (CONCRETE), with support from the Air Force Office of Scientific Research and the Air Force Research Laboratory.
The breakthrough highlights a major step toward enabling electronics that can operate in some of the harshest environments known, potentially allowing spacecraft, industrial systems, and energy infrastructure to process data directly on-site rather than relying on remote control systems.
KEY QUOTE:
“Over 92 percent of the computing in AI systems like ChatGPT is nothing but matrix multiplication. This type of device can perform that in the most efficient way, orders of magnitude faster and at lower energy. This is the first step. It’s still a long way to go. But logically, you can see: now it makes it possible. The missing component has been made.”
Joshua Yang, Arthur B. Freeman Chair Professor, Ming Hsieh Department of Electrical and Computer Engineering, USC Viterbi School of Engineering
