Neural signal analysis with memristor arrays towards high-efficiency brain-machine interfaces
Zhengwu Liu1, Jianshi Tang2,3, Bin Gao1,4
1Institute of Microelectronics, Beijing Innovation Center for Future Chips (ICFC), Tsinghua University, Beijing, 100084, China.
Nature Communications
|August 27, 2020
Summary
This study introduces a novel memristor-based system for analyzing neural signals, significantly improving efficiency for brain-machine interfaces. This technology offers a pathway to more effective restoration of motor functions and brain research.
Area of Science:
- Neuroscience
- Materials Science
- Computer Engineering
Background:
- Brain-machine interfaces (BMIs) are crucial for restoring motor functions and understanding brain mechanisms.
- Increasing electrode counts strain current BMI signal processing capabilities due to conventional digital architectures.
- The von Neumann architecture's digital computation differs fundamentally from the brain's analog processing.
Purpose of the Study:
- To develop a highly efficient neural signal analysis system for advanced BMIs.
- To leverage memristor's bio-plausible analog computing for signal processing.
- To demonstrate the system's efficacy in filtering and identifying epilepsy-related neural signals.
Main Methods:
- Utilized memristor arrays for analog domain neural signal analysis.
- Implemented filtering and identification of epilepsy-related neural signals as a proof-of-concept.
- Compared the memristor-based system's power efficiency against state-of-the-art CMOS systems.
Main Results:
- Achieved 93.46% accuracy in filtering and identifying epilepsy-related neural signals.
- Demonstrated nearly 400x improvement in power efficiency compared to conventional systems.
- Validated the feasibility of memristor-based analog computation for neural signal processing.
Conclusions:
- Memristor-based systems offer a viable, high-performance solution for next-generation BMIs.
- Analog signal analysis using memristors significantly enhances power efficiency in neural signal processing.
- This approach addresses the bottleneck in current BMI signal processing capabilities.
