Compute-in-Memory acts as both processor and non-volatile memory making AI faster, more energy-efficient in rapid analysis of large data sets

https://techxplore.com/news/2025-10-ai-efficiency-advances-spintronic-memory.html

"earlier designs relied on analog computing, which while energy-efficient is less precise than digital... Spin-Transfer Torque Magnetic-Random Access Memory spintronic chip stores data within magnetic tunnel junction... different magnetic alignments create varying electrical resistances corresponding to binary states... 7.4–29.6 ns computation latencies, 7.02–112.3 TOPS/W energy efficiencies for fully parallel lossless matrix–vector multiplications across precision configurations ranging from 4 to 16 bits"