SAN JOSE, Calif.: #AIHardware--TetraMem Inc. today announced that its academic and research collaborators have demonstrated RRAM (memristor) devices capable of reliable operation at temperatures up to 700°C, marking a major advance in non-volatile memory for extreme-environment computing.

Published in Science under the title “High-temperature memristors enabled by interfacial engineering,” the work demonstrates a graphene-enabled RRAM architecture that maintains fast switching performance, long data retention, and great endurance under extreme thermal stress. Surprisingly, at 700 °C the devices require less than one-third of the current and half the voltage needed at room temperature to switch, significantly reducing energy consumption while sustaining over one billion switching cycles. The result advances RRAM’s potential for industrial systems, aerospace, and future deep-space AI computing.

The milestone follows TetraMem’s prior demonstration of high-density, high-precision RRAM with 2,048 conductance levels, equivalent to 11 bits per cell, in fully integrated 256 × 256 RRAM arrays monolithically integrated on CMOS, published in Nature in 2023 as “Thousands of conductance levels in memristors integrated on CMOS.” Together with TetraMem’s commercial production of its analog in-memory computing Neural Process Unit (NPU) through commercial foundries, the results demonstrate RRAM’s advantages as an emerging, non-volatile, high-density, high-precision memory technology compatible with standard CMOS manufacturing.

Through advanced interfacial engineering, the research team showed that graphene suppresses metal diffusion and structural degradation, two major failure mechanisms at high temperature. Transmission electron microscopy and first-principles modeling confirmed stable device operation under severe thermal stress.

This breakthrough was achieved through a multi-institution collaboration involving researchers from the University of Southern California, University of Massachusetts Amherst, TetraMem and other research partners. The study was led by Prof. J. Joshua Yang and Prof. Qiangfei Xia, both professors at USC and UMass, respectively, and co-founders of TetraMem, highlighting the strong synergy between academic research and industrial innovation.

“This breakthrough extends the proven advantages of RRAM into a new operating regime,” said Dr. Glenn Ge, Co-founder and CEO of TetraMem. “TetraMem has already demonstrated high-density, high-precision RRAM and in-memory computing silicon manufactured through standard semiconductor processes. This work demonstrates that RRAM can offer exceptional robustness under extreme thermal conditions, making it a strong candidate for future AI systems operating far beyond conventional data center environments. Such capability not only enables reliable operation in harsh conditions but also has the potential to significantly improve cooling efficiency.”

In space, heat dissipation is fundamentally constrained because systems cannot rely on air or liquid convection; heat must be removed primarily through radiation, which increases with the fourth power of chip temperature. RRAM devices that maintain stable operation at extreme temperatures offer a promising foundation for future space-grade AI systems requiring thermal resilience, radiation tolerance, energy efficiency, and long mission lifetimes.

RRAM is also central to analog in-memory computing because it stores multi-level conductance states and performs computation where data is stored. By integrating memory and computation, analog in-memory architectures reduce data movement, one of the largest sources of energy consumption and latency in modern AI systems. The result expands the operating envelope for RRAM-based in-memory computing, from efficient AI acceleration to future systems designed for extreme environments.

About TetraMem

TetraMem is an RRAM semiconductor company building analog in-memory computing fabric for the AI era. Its multi-level RRAM technology integrates memory and compute to reduce data movement, improve energy efficiency, and scale AI processing from edge devices to data centers and future high-performance systems.

Fonte: Business Wire