Electroencephalogram signal analysis with 1T1R arrays toward high-efficiency brain computer interface

Abstract

Brain computer interface (BCI) is a promising way for automatic driving and exploring brain functions. As the number of electrodes for electroencephalogram (EEG) acquisition continues to grow, the signal processing capabilities of BCI are facing challenges. Considering the bottlenecks of the Von Neumann architecture, it is increasingly difficult for the traditional digital computing pattern to meet the requirements of the EEG signal processing in terms of power consumption and efficiency. Here, we propose a 1T1R array-based EEG signal analysis system in which the biological likelihood of the memristor is used to efficiently analyze signals in the simulated domain. The identification and classification of EEG signals are achieved experimentally using the memristor array with an average recognition rate of 89.83%. The support vector machine classification implemented by the memristor crossbar array provides a 34.4 times improvement in power efficiency compared to the complementary metal oxide semiconductor-based support vector machine classifier. This work provides new ideas for the application of memristors in BCI.