Neuromorphic model-based neural decoders for brain-computer interfaces: a comparative study

Brain-computer interfaces that convert neural signals into control commands for external devices have demonstrated great potential in neural rehabilitation. Most recently, the growing demand for developing large-scale brain signal recording devices has highlighted the need for power-efficient neural signal processing. Neuromorphic computing models, implemented with chips, can provide ultra-low-cost computation of neural signals, facilitating brain-implantable devices. In our work, we evaluated the neural decoding ability of neuromorphic models in comparison with traditional neural decoding approaches. We found that neuromorphic model-based neural decoders showed high accuracy comparable to existing decoders while with significantly lower computational costs. Our results suggest the potential of using neuromorphic models for highly efficient computing with large-scale brain data.