Chinese researchers have unveiled a specialised chip designed for neuroscience research that they say can significantly accelerate brain modelling. In benchmarks focused on cortical surface reconstruction, the chip reportedly delivered performance gains ranging from 50× to 478× over Nvidia’s A100 GPU on these specialised workloads.
The research was led by Peking University’s School of Information Engineering in collaboration with the Chinese Academy of Sciences’ Shanghai Institute of Microsystem and Information Technology. The findings were published in the journal Science.
The chip uses a phase-change memristor-based architecture with an in-memory computing design, allowing data storage and computation to occur within the same memory array instead of being transferred between separate memory and processing units.
According to the researchers, the chip can reconstruct the brain’s folded cortical surface in less than half a second, with a single-step computation latency of 2.12 milliseconds. Fabricated using a 40-nanometre process, it is designed as a domain-specific accelerator for neuroscience applications rather than a general-purpose replacement for GPUs.
The researchers say the chip could deliver significant speed and efficiency gains for specialised brain modelling workloads. Potential future applications include research into neurological disorders such as Alzheimer’s disease, brain-computer interfaces, surgical navigation, and long-term efforts to develop personalised digital brain twins.
Read Article: Swiggy Moves Closer to IOCC Status as Domestic Shareholding Crosses 50%
