RISC-V CNN Coprocessor for Real-Time Epilepsy Detection in Wearable Application

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Neurology

Background:

• Epilepsy diagnosis relies on electroencephalography (EEG) signals.
• Current AI-based epilepsy detection methods face computational challenges for real-time embedded monitoring.
• Instant seizure detection is crucial for timely treatment within the therapeutic window.

Purpose of the Study:

• To develop a real-time epilepsy detection algorithm framework.
• To design and implement a dedicated coprocessor chip for accelerating AI-based epilepsy detection.
• To create a custom interface for coprocessor reconfiguration and data transfer using RISC-V instructions.

Main Methods:

• An 11-layer convolutional neural network (CNN) was developed for EEG signal analysis.
• A dedicated CNN coprocessor was fabricated using TSMC 0.18-μm CMOS technology.
• The system was validated through animal experiments.

Main Results:

• The epilepsy detection algorithm achieved 97.8% accuracy for floating-point and 93.5% for fixed-point operations.
• The coprocessor demonstrated low energy consumption: 51 nJ/class for data transfer and 0.9 µJ/class for inference.
• The detection latency on the chip was measured at 0.012 seconds.

Conclusions:

• The proposed framework and hardware coprocessor enable efficient, real-time epilepsy detection.
• This integrated solution addresses the limitations of software-only and hardware-only approaches.
• AI-driven epilepsy monitoring on embedded devices is now feasible, improving patient care.