Chinese team’s brain-mimicking chip ‘up to 478 times faster than Nvidia GPU’ - South China Morning Post

Chinese scientists have developed a brain-mimicking chip that is up to 478 times faster than Nvidia's A100 GPU, according to a study published in Science. Developed by researchers from Peking Universi... Read original →

Despite using only a 40nm mature node, this neuromorphic chip achieves 478x the speed of NVIDIA’s A100 by integrating memory and computation—exposing GPU architecture’s inefficiency in brain-inspired workloads. Technically, it will accelerate co-development of EDA tools, advanced packaging, and spiking neural networks, benefiting China’s RISC-V and in-memory computing IP players. From a compliance angle, medical AI applications leveraging this chip could face dual-use export controls in the U.S. and EU, raising regulatory overhead. NVIDIA is likely to respond by acquiring neuromorphic startups or extending CUDA into heterogeneous architectures. Over the next 12–24 months, 'digital brain twin' applications may move from labs to clinical trials—but mass adoption hinges on replicating this efficiency at sub-7nm nodes, a challenge constrained by China’s limited access to EUV lithography.