What if a memory chip could process data where it is stored? A new brain inspired design shows how that could transform real time computing.
Researchers from Peking University and the Chinese Academy of Sciences have developed a brain inspired computing chip that performs data storage and computation within the same memory array, significantly accelerating the modelling of complex brain structures. Detailed in the journal Science, the 40 nm chip integrates an artificial neural network and reconstructs cortical brain surfaces in under half a second.
The development addresses a longstanding limitation of conventional computing architectures, where memory and processors are physically separated, creating delays and increasing power consumption as data moves between them. By adopting a computing in memory design, the new chip reduces latency and enables real time neural computation, making it suitable for applications such as brain computer interfaces, medical imaging, and neurosurgical guidance.
A key feature of the chip is its use of phase change memristors, a next generation memory technology. Instead of treating conductance drift, a common source of data instability, as a drawback, the researchers used the phenomenon as part of the computing process. This approach allowed the chip to model complex brain topologies with high fidelity and millisecond scale latency. According to the research team, the hardware achieved performance ranging from 50 to 478 times faster than systems based on the Nvidia A100 GPU for this specific workload.
The researchers believe the architecture could enable personalised digital brain models, early screening for neurological disorders such as Alzheimer’s disease, and faster clinical decision making. Beyond healthcare, the low latency computing platform may also support robotics and embodied AI systems that require rapid neural dynamical computation.
“This breakthrough opens up new possibilities for brain computer interfaces and the diagnosis and treatment of brain diseases. In the future, personalised and dynamic digital brain twins will become possible,” says Yang Yuchao, Professor at Peking University’s School of Integrated Circuits and Deputy Dean of the School of Electronic and Computer Engineering.
