RRAM-based Multimodal BCI
Developed heterogeneous processing system for multimodal brain-computer interfaces using RRAM technology
Overview
Developed a RRAM-based heterogeneous processing system for multimodal brain-computer interfaces, focusing on P300 signal recognition and integration with other BCI modalities.
P300 Signal Characteristics
Typical P300 ERP waveform showing the characteristic positive deflection approximately 300ms after stimulus onset.
EEGNet Architecture
Original EEGNet architecture showing the temporal and spatial convolution layers for EEG signal processing.
RRAM-based Classification
RRAM-based P300 signal classification paradigm showing the compute-in-memory approach for efficient neural processing.
Modified Architecture
Left: Modified EEGNet architecture optimized for P300 signal processing. Right: Multimodal EEG signal classification system combining P300 and MI signals using RRAM chips.
Technical Details
- Simulated representative heterogeneous processing paradigm of P300 signal recognition using resistive random-access memory (RRAM)
- Implemented and evaluated fixed parameter disturbance training (FDT) techniques
- Combined multiple datasets:
- BCIC IV IIa dataset
- P300 RSVP dataset
- Designed an RRAM-based multimodal recognizer integrating:
- Pre-trained EEGNet
- Common Spatial Pattern (CSP)
- Modality-fused classifier
Results & Achievements
- Achieved 2.83% higher accuracy using multimodal BCI with FDT compared to without FDT
- Significantly outperformed single-modality approaches:
- 8.19% improvement over MI-alone results
- 13.20% improvement over P300-alone results
Skills & Technologies
- Brain-Computer Interfaces
- RRAM Technology
- Signal Processing
- Machine Learning
- Python Programming
- Neural Networks
Project Advisor
Dr. Zhengwu Liu, The University of Hong Kong