New CRAM Technology Slashes AI Energy Use by 1,000 Times

University of Minnesota engineers introduce a breakthrough in-memory computing design, promising unprecedented efficiency for AI applications.

CRAM (Computational Random-Access Memory) processes data entirely within memory, eliminating energy-intensive transfers.
The technology uses spintronic devices called Magnetic Tunnel Junctions (MTJs) to enhance speed and resilience.
Initial tests show CRAM can be up to 2,500 times more energy-efficient and 1,700 times faster than current systems.
Global AI energy consumption is projected to double by 2026, making CRAM's efficiency gains crucial.
Researchers plan to collaborate with industry leaders to scale up and integrate CRAM into mainstream applications.