Add BrightSurf on Google Email

Taming heat at the nanoscale to satiate energy-hungry artificial intelligence

09.21.26 | International Journal of Extreme Manufacturing

Every search query, generated image, and conversational response produced by today's artificial intelligence comes with a hidden cost: an unsustainable surge in electrical power and heat. At the heart of this problem lies computer memory, which struggles to shuffle vast oceans of data without burning excessive energy.

Phase-change memory has long stood out as a promising heir to conventional silicon chips, storing information by rapidly melting and freezing tiny pockets of material between an orderly, conductive crystal and a disordered, insulating glass. Yet, packing multiple bits of information into these microscopic cells has proved notoriously difficult. The violent bursts of heat needed to reset the memory tend to blast atoms out of place, causing the material to degrade and the stored data to drift out of alignment over time.

Prof. Tae Geun Kim at Korea University and his co-workers have dismantled this long-standing barrier by learning how to sculpt heat within a nanometer-thin sandwich of materials. Publishing in the International Journal of Extreme Manufacturing , the team engineered a "double-confined" structure that stacks alternating slices of antimony telluride (Sb 2 Te 3 ) with two distinct helper compounds: highly conductive nickel ditelluride (NiTe 2 ) and heat-blocking molybdenum ditelluride (MoTe 2 ).

Instead of allowing heat to flood the entire cell at once, this structure works like a two-stage thermal dam. A gentle voltage pulse heat is confined to the conduction region below the thermal barrier, melting a tiny and contained pocket of material to write a first distinct data state. Turn the voltage up slightly, and the thermal energy crests over the molybdenum barrier into the upper chamber to melt a second region, cleanly locking in another distinct data state without disturbing the foundation below.

This layered geometry neatly untangles a stubborn trade-off between electrical speed and material durability. Previous attempts had to choose between fast-conducting materials whose weak chemical bonds quickly unraveled under high heat, and tough and heat-resistant materials whose high electrical resistance demanded power-hungry voltages. By pairing the easy electrical flow of NiTe 2 with the rigid and heat-scattering crystal lattice of MoTe 2 , the new architecture achieves rapid switching while keeping long-range atomic diffusion suppressed.

The resulting efficiency gains register at extreme dimensions. The cell completes its multi-level switching transitions in just 50 nanoseconds while cutting reset energy demand by up to 76%, down to 115 pJ. The tendency for stored data to wander over time dropped by more than thirtyfold compared to standard commercial phase-change alloys, allowing the device to endure over one million write cycles without physical breakdown.

Crucially for chipmakers, these ultra-thin layers can be deposited using standard sputtering equipment already installed in modern semiconductor foundries, bypassing the need for expensive cleanroom overhauls. For massive data centers straining against power grids, this design offers a practical blueprint for packing dense and multi-level memory into compact chips that run cool. The researchers are now refining the thickness of each individual slice and testing related alloys to extend device lifespans even further under heavy industrial computing loads.

International Journal of Extreme Manufacturing (IJEM, IF: 25.1 ) is devoted to publishing articles of the highest quality and significance to pushing the limits of scales, precision, performance and environments in manufacturing.

Visit our webpage , like us on Facebook , and follow us on Twitter and LinkedIn .

International Journal of Extreme Manufacturing

10.1088/2631-7990/ae848b

Double-confined phase-change heterostructure memory for ultra-stable low-power multi-level operation

15-Jul-2026

Keywords

Article Information

Contact Information

Yue YAO
International Journal of Extreme Manufacturing
yueyao@ijem.org

Source

This article is based on a news release from International Journal of Extreme Manufacturing. BrightSurf curates and republishes science news from research institutions worldwide; the original release is linked below.

How to Cite This Article

APA:
International Journal of Extreme Manufacturing. (2026, September 21). Taming heat at the nanoscale to satiate energy-hungry artificial intelligence. Brightsurf News. https://www.brightsurf.com/news/14747DG1/taming-heat-at-the-nanoscale-to-satiate-energy-hungry-artificial-intelligence.html
MLA:
"Taming heat at the nanoscale to satiate energy-hungry artificial intelligence." Brightsurf News, Sep. 21 2026, https://www.brightsurf.com/news/14747DG1/taming-heat-at-the-nanoscale-to-satiate-energy-hungry-artificial-intelligence.html.