Researchers at the University of Minnesota have unveiled a groundbreaking hardware device known as CRAM, designed to significantly reduce energy consumption in artificial intelligence applications by at least 1,000 times.

The CRAM chip has demonstrated the potential for energy savings of up to 2,500 times compared to traditional computing methods.

This innovative technology utilizes spintronic devices called magnetic tunnel junctions (MTJs) to perform computations directly within memory, effectively eliminating the need for power-hungry data transfers.

By removing the bottleneck between computation and memory inherent in the von Neumann architecture, CRAM allows for more efficient energy usage.

This advancement is expected to make AI technologies significantly more energy-efficient, addressing the escalating energy demands associated with AI.

The International Energy Agency forecasts that global energy consumption for AI will more than double from 460 terawatt-hours in 2022 to over 1,000 terawatt-hours by 2026, highlighting the urgent need for such innovations.

Jian-Ping Wang, the senior author of the study, emphasized the interdisciplinary collaboration that made this technology possible, involving experts from various fields.

The research team included contributions from multiple researchers at both the University of Minnesota and Arizona University, supported by various U.S. government agencies.

The foundational research for CRAM has been in development for over two decades, with initial concepts dating back 20 years.

The findings of this research were published in the peer-reviewed journal npj Unconventional Computing, and the researchers hold multiple patents related to the technology.

Looking ahead, the team plans to collaborate with semiconductor industry leaders in Minnesota to demonstrate and produce hardware that enhances AI capabilities.

This collaboration aims to address challenges related to scalability, manufacturing, and integration with existing technologies.