An Efficient Brain-Switch for Asynchronous Brain-Computer Interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Engineering

Background:

• Intracortical brain-computer interfaces (iBCIs) traditionally use action potentials (spikes) for signal processing via application-specific integrated circuits (ASICs).
• Asynchronous iBCIs rely on "brain-switches" to detect user engagement based on spiking activity, which is computationally intensive.
• Local field potentials (LFPs) offer a lower-power alternative to spikes for in vivo signal processing in ASICs.

Purpose of the Study:

• To present the first design and implementation of an LFP-based brain-switch for asynchronous iBCIs.
• To develop a more practical and power-efficient solution for in vivo signal processing in iBCIs.
• To demonstrate the feasibility of using gated recurrent neural networks (RNNs) for LFP-based brain-switch functionality.

Main Methods:

• Design and implementation of an LFP-based brain-switch using gated recurrent neural networks (RNNs).
• Development of an in vivo LFP-based feature extraction unit synthesized on a 180-nm CMOS process.
• Evaluation of power consumption and silicon area of the synthesized ASIC.

Main Results:

• The LFP-based brain-switch requires no exhaustive learning phase for channel or frequency band selection, improving practical applicability.
• The synthesized ASIC occupies a small silicon area (0.09 mm²) and consumes minimal power (91.87 nW) at 2 kHz.
• The proposed design achieves state estimation performance comparable to traditional spike-based brain-switches while consuming significantly less power.

Conclusions:

• LFP-based brain-switches offer a highly power-efficient alternative to spike-based systems for asynchronous iBCIs.
• The developed RNN-based design simplifies practical implementation by eliminating complex learning phases.
• This advancement paves the way for more accessible and sustainable in vivo BCI technologies.