South Korean media reports that a next-generation AI phase-change memory capable of controlling heat during storage operations and reducing power consumption by up to 76% compared to current levels has been successfully developed. Phase-change memory utilizes the difference in electrical resistance between crystalline and amorphous states of materials to store information. It is a type of non-volatile memory that offers high-speed operation and the ability to store multiple pieces of information in a single cell. It is being considered as the next-generation storage solution for the fields of artificial intelligence and computing-in-memory.
On September 11, the National Research Foundation of Korea announced that a team led by Professor Kim Geon-su from the Department of Electrical and Electronic Engineering at Korea University has successfully developed a "dual-isolated phase-change heterostructure memory." By combining two materials with distinct characteristics, the team has gradually controlled the heat diffusion and phase-change regions within the memory. The results of this research have been published online in the international academic journal International Journal of Extreme Manufacturing.
New Phase-Change Memory Technology Significantly Reduces Energy Consumption
With the surge in data brought about by generative AI and large language models, phase-change memory is gaining widespread attention as an alternative technology to current storage systems. However, traditional phase-change memory has an inherent bottleneck: the heat generated during data writing spreads to surrounding areas, wasting energy and making it difficult to precisely control the phase-change region. More challenging is the fact that materials that operate at low drive voltages often have poor heat resistance, while heat-resistant materials require high voltage to operate, creating a difficult trade-off.
The research team introduced a new structure that arranges two transition metal chalcogenides—NiTe₂ and MoTe₂—which can maintain a crystalline state during operation—within the device. NiTe₂ acts as a heat conduction layer, efficiently generating and transferring heat even at lower voltages, while MoTe₂ serves as a heat barrier layer, inhibiting the outward diffusion of heat. With this design, the team was able to autonomously plan the heat movement path within the memory, achieving precise control over the location and range of phase changes.
Experimental results show that this structure allows for the on-demand design of heat flow paths within the memory, ultimately reducing operating energy consumption by 76% compared to current solutions, while also increasing device durability (i.e., the number of possible drive cycles) by 110%.
Item Specification Energy Consumption Reduction 76% Durability Increase 110%