Stock Image
Scientists in China have developed a memristor-based computing chip capable of reconstructing highly detailed 3D models of the brain in under 10 milliseconds, a speed comparable to the pace of biological brain activity. The breakthrough could significantly accelerate applications such as brain-computer interfaces, surgical navigation, and digital brain modeling.
The research was led by Yuchao Yang of Peking University in collaboration with the Shanghai Institute of Microsystem and Information Technology at the Chinese Academy of Sciences. The team’s findings were published in the journal Science, where researchers demonstrated a neural dynamical system powered by phase-change memristors that dramatically outperforms conventional computing hardware in both speed and energy efficiency.
Neural dynamical systems combine artificial neural networks with mathematical models to simulate how complex systems evolve over time. They are widely used for medical imaging, physical simulations, and three-dimensional brain reconstruction, but the calculations typically require enormous computing power and frequent data transfers between processors and memory, slowing performance and increasing energy consumption.
The new chip eliminates much of this bottleneck by performing computations directly where the data is stored. Fabricated using a 40-nanometer manufacturing process, the device integrates in-memory computing with conductance-drift arrays across an area of just 0.28 square millimeters. Operating at 50 MHz, it completes each computational step through a nine-stage processing pipeline.
In testing, the chip completed neural dynamics calculations between 3.82 and 36.27 times faster than leading application-specific integrated circuits while consuming up to 24.73 times less power. During cortical surface reconstruction, it achieved speeds up to 478 times faster than NVIDIA’s A100 GPU without sacrificing accuracy.
Researchers successfully reconstructed the brain’s white and gray matter boundaries in real time while preserving the cortex’s complex folded structure. The resulting 3D models remained smooth, topologically consistent, and highly accurate based on established surface comparison metrics.
The team believes the technology could eventually enable real-time brain-computer interfaces, digital brain twins, advanced surgical guidance systems, and new research tools for neurological disorders including Alzheimer’s and Parkinson’s disease.
