NeuralTree: A 256-Channel 0.227-μJ/Class Versatile Neural Activity Classification and Closed-Loop Neuromodulation SoC

Area of Science:

• Neuroscience
• Biomedical Engineering
• Electrical Engineering

Background:

• Closed-loop neural interfaces are crucial for treating neurological disorders but face scalability limitations.
• Existing systems struggle with low channel counts, hindering accurate disease detection and therapeutic efficacy.

Purpose of the Study:

• To develop a highly scalable and versatile closed-loop neural interface System-on-Chip (SoC).
• To enable accurate, real-time detection and potential suppression of neurological disease symptoms.
• To restore lost functions in patients with paralysis.

Main Methods:

• A 256-channel time-division multiplexed (TDM) front-end with a mixed-signal DC servo loop (DSL) for high-resolution neural recording.
• A tree-structured neural network (NeuralTree) classifier for patient- and disease-specific biomarker extraction.
• An energy-aware learning algorithm for optimizing the energy-accuracy trade-off in classification.
• A 16-channel high-voltage (HV) neurostimulator for closed-loop therapy delivery.

Main Results:

• The SoC, fabricated in 65nm CMOS, achieved 0.227μJ/class energy efficiency and 0.014mm²/channel area.
• Achieved 95.6%/94% sensitivity and 96.8%/96.9% specificity on human epilepsy EEG/iEEG datasets.
• Demonstrated in-vivo recordings and biomarker extraction in a rat epilepsy model.
• First on-chip classification of Parkinson's disease tremor from human local field potentials (LFPs).

Conclusions:

• The proposed SoC offers a scalable and versatile platform for closed-loop neural interfaces.
• The NeuralTree classifier provides efficient and accurate disease-state detection.
• This technology holds promise for advancing therapeutic interventions in neurological disorders.