Researchers at the University of Southern California have unveiled a groundbreaking non-volatile memory chip, a memristor, capable of operating at an astonishing 700°C. This far surpasses the reliability limits of conventional electronics, which typically degrade around 200°C. This remarkable feat opens doors for electronic applications in extreme environments, potentially including missions to Venus.
The key innovation lies in the materials used and the device’s architecture. The memristor is constructed with a stack consisting of tungsten as the top electrode, hafnium oxide as the switching layer, and graphene as the bottom electrode. This configuration allows the device to function reliably at temperatures equivalent to lava, far beyond the capabilities of standard silicon-based electronics.
Published in the journal Science, this research reports the highest operating temperature to date for this type of non-volatile resistive memory. Jian Zhao, the lead author, fabricated devices with dimensions ranging from 200 nm by 1 μm to 1 μm by 1 μm. During testing, these memristors maintained an ON/OFF ratio exceeding three orders of magnitude from room temperature up to 700°C. Furthermore, they retained their memory states for over 50 hours without refresh, with an average retention of approximately 145 hours across 30 devices. The memristors also endured over a billion switching cycles at 700°C, with applied voltages around 1.5V and pulses as short as 30 ns.
A critical factor preventing device failure at high temperatures is the physical mechanism employed. Unlike conventional designs where heat can cause atoms from the top electrode to diffuse through the oxide layer and create a permanent short circuit (stuck in the ON state), this graphene-based memristor behaves differently. Investigations using high-resolution electron microscopy, EDS, EELS, and first-principles simulations revealed that tungsten adheres and stabilizes poorly on graphene compared to platinum (specifically, Pt(111)). This significantly limits the migration that would otherwise destroy the device. Calculations indicate that the surface diffusion of tungsten on platinum is approximately 1,000 times greater than on graphene.
The study also demonstrates the potential for scaling and in-memory computing. At 700°C, the memristors were able to program 32 distinct resistance states, maintaining current-voltage responses with correlation coefficients above 0.995 across 16 representative states between 0 and 0.5V. Additionally, the researchers successfully fabricated a 32×32 crossbar array, a 1K configuration, using a two-terminal setup.
This advancement has significant implications for artificial intelligence (AI) applications, as over 92% of computation in systems like ChatGPT relies on matrix multiplication. While the memristor itself is not yet a market-ready product, requiring the development of high-temperature logic circuits and manual fabrication at a sub-micrometer scale, its potential is immense.
The true value of this research lies in its application in extreme environments. Temperatures above 500°C are encountered in scenarios such as Venus exploration, deep geothermal drilling, nuclear systems, fusion research, and advanced industrial sensing, where conventional silicon electronics are unviable. This memristor is poised to fill that gap, serving as a non-volatile memory and in-memory computing component designed to operate where current electronics cannot.
