Abstract:Abstract To fulfill complex human-machine interactions, a brain-computer interface (BCI) must not only decipher brain signals but also dynamically adapt to brain fluctuations, ultimately co-evolving with the brain. This necessitates a novel decoder capable of flexible updates with energy-efficient decoding capabilities. In this work, we designed a co-evolutional BCI with a neuromorphic decoder enabled by a 128k-cell memristor chip. By interacting with the brain, the decoder continuously updates its parameters, leading to the successful real-time control of a drone in 4 degrees of freedom (4-DOF) and enabling it to navigate around obstacles. Our approach featured a hardware-efficient one-step memristor decoding strategy, enabling the neuromorphic chip-equipped BCI to achieve decoding performance equivalent to software-based methods. Notably, it accomplished this at three orders of magnitude lower energy consumption and two orders of magnitude higher normalized speed than a central processing unit (CPU). Moreover, employing an interactive update framework, we showed the co-evolution of the brain-memristor decoder over an extended interaction task involving ten subjects. This resulted in a remarkable enhancement of BCI performance by nearly 20%, showcasing the substantial potential of memristor decoders in advancing BCIs. The study results also showed that the decoder initially played a dominant role in the co-evolution, but the brain learned as the process progressed. Eventually, a dynamic balance between the two emerged for decision-making. These findings lay the groundwork for developing future human-centric hybrid intelligence systems.

Memristor Chip-Enabled Adaptive Neuromorphic Decoder for Co-Evolutional Brain-Computer Interfaces