NeuroGator: A Low-Power Gating System for Asynchronous BCI Based on LFP Brain State Estimation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Implantable brain-computer interface (BCI) systems generate large data volumes, straining hardware resources, especially in power-limited wireless applications.
• Continuous data processing in BCI poses a significant bottleneck, hindering the efficiency and practicality of implantable devices.

Purpose of the Study:

• To introduce NeuroGator, an asynchronous gating system designed to mitigate the data handling bottleneck in implantable BCI systems.
• To develop a resource-efficient architecture that reduces data size and power consumption for BCI devices.

Main Methods:

• NeuroGator employs hierarchical state classification with a two-stage approach: a low-power silence detector and a Dual-Resolution Gate Recurrent Unit (GRU) model.
• The silence detector filters non-active signals, reducing data size by ~69.4%, while the GRU model analyzes Local Field Potential (LFP) data at varying precisions for activity confirmation.
• The system was implemented on an Application-Specific Integrated Circuit (ASIC) using a 180 nm CMOS process.

Main Results:

• NeuroGator achieved an 82% reduction in overall data throughput while maintaining a high F1-Score of 0.95.
• The system enables implantable BCI devices to operate in an ultra-low-power state for over 85% of the time.
• The ASIC implementation demonstrated minimal silicon area (0.006mm²) and ultra-low power consumption (51 nW).

Conclusions:

• NeuroGator effectively resolves the resource efficiency challenges in implantable BCI systems.
• The proposed asynchronous gating system offers a robust paradigm for next-generation, power-efficient implantable BCI devices.
• This approach significantly enhances the feasibility of advanced BCI functionalities in resource-constrained environments.