Peking University team reports sub-10 ms memristor chip for neural dynamics

Researchers led by Yuchao Yang of Peking University, with collaborators at the Shanghai Institute of Microsystem and Information Technology of the Chinese Academy of Sciences, report a phase-change memristor chip that runs neural dynamical calculations in under 10 milliseconds, SciTechDaily wrote, citing a Science paper. SciTechDaily said the chip uses a 40-nm process, packs in-memory computing and conductance-drift arrays into 0.28 square millimeters, runs at 50 MHz, and completes each integration step through nine pipeline stages. In neural dynamics tests it was 3.82 times to 36.27 times faster than compared ASICs and used 11.75 times to 24.73 times less power. For cortical surface reconstruction it reached speedups of as much as 478.18 times versus an NVIDIA A100 GPU, the write-up said. The team reconstructed white- and gray-matter boundaries and 3D cortical surface meshes in real time while preserving folds, with strong scores on average symmetric surface distance and Hausdorff distance metrics. The Science paper is dated 2 July 2026 and carries DOI 10.1126/science.aee6277. Possible uses listed include brain-computer interfaces, surgical navigation, medical imaging, and neurodegenerative disease research tools.