An analog-based chip that enables processing "in memory." Invented at the Peking University in 2025 for matrix equation solving, resistive RAM (RRAM) speeds up the massive amounts of multiply-and-add computations normally handled by GPUs for AI, scientific computing and other "data heavy" tasks.
Instead of shuttling numbers back and forth from a GPU's memory to its computing cores, a voltage is applied to the RRAM that causes an entire array to perform matrix calculations at once. Each memory cell holds values as resistance levels, and the array performs multiply-and-add calculations a thousand times faster than traditional digital methods using a hundred times less energy.
The U.S. Gave China the Incentive
Since 2022, the U.S. has banned the export of advanced GPUs to China, thus giving the country the incentive to do its own AI research. For example, NVIDIA's H100 and A100 GPUs were banned, but the sanctions only served to stimulate China's research. See
RRAM.
6G Towers and AI at the Edge
The major applications for these chips are 6G communications and AI computing at the edge. Towers for 6G require simultaneous processing of hundreds of signals and AI training demands massive amounts of matrix multiplications. With RRAM, AI models can theoretically be trained in days instead of months, and AI at the edge means any computing device can perform training and inference, even a smartphone. Pundits claim that while the West is optimizing old technology, China is inventing tomorrow's technology. Stay tuned!