2024-08-16 04:00:10
Researchers at the University of Minnesota have developed an innovative device that could reduce the energy consumption of artificial intelligence (AI) by a factor of at least 1,000. The breakthrough addresses growing concerns about theenvironmental impact AI technologies, as their global energy demand is expected to double by 2026, according to the International Energy Agency.Energy.
Current AI systems require large amounts of data transfer between memory and processors, a very energy-intensive process. The new device, called RAM Computational Random Access Memory (CRAM) allows data to be processed directly within memory, without having to need to transfer them. This method, first tested by the team of theUniversité of Minnesota, could drastically reduce the energy consumption of AI applications. Yang Lv, a postdoctoral researcher at the University of Minnesota and first author of the study, explains that CRAM allows data to be processed directly in the memory array, eliminating energy-intensive transfer steps. This technologywhich is based on the use of junctions tunnel magnetic resonance imaging (MTJ) technology allows information to be stored and processed much more efficiently than current methods based on transistors.
Jian-Ping Wang, a professor at the University of Minnesota and co-author of the study, points out that this technology is the result of more than 20 years of research and collaboration. interdisciplinary. Initially considered a crazy idea, CRAM has now proven its effectiveness and is ready to be integrated into existing technologies. According to Jian-Ping Wang, the first results show a reduction in energy consumption by a factor of 2,500 compared to traditional systems.
CRAM could represent a sustainable solution for the development of AI, offering unprecedented energy efficiency. This technological advance opens up promising prospects for reducing the environmental footprint of AI systems, while maintaining high performance. The researchers are now considering collaborating with leaders in the semiconductor industry to develop this technology on a large scale.
1723836817
#technique #consume #times #energy
