Ferroelectric Synapses Enable Adaptive EEG Decoding
Researchers demonstrate that spiking neural networks (SNNs) deployed on ferroelectric memristive synaptic devices can achieve adaptive EEG-based motor imagery decoding. The work addresses challenges in brain-computer interfaces (BCIs) caused by non-stationary neural signals that vary across sessions and individuals. Programmable memristive hardware offers a substrate for post-deployment adaptation, but practical realization is limited by weight resolution, device variability, nonlinear programming dynamics, and finite endurance. The team fabricated, characterized, and modeled weight updates in ferroelectric synapses, then evaluated deployment under realistic device constraints. Classification performance was comparable to software-based SNNs. The study appears on arXiv (2601.00020v3).
Key facts
- Spiking neural networks deployed on ferroelectric memristive synapses for EEG decoding
- Addresses non-stationary neural signals in brain-computer interfaces
- Programmable memristive hardware enables post-deployment adaptation
- Challenges include limited weight resolution, device variability, nonlinear dynamics, finite endurance
- Fabrication, characterization, and modeling of ferroelectric synapses performed
- Classification performance comparable to software-based SNNs
- Study published on arXiv (2601.00020v3)
Entities
Institutions
- arXiv