Abstract

We present a 16-channel seizure detection system-on-chip (SoC) with 0.92μW/channel power dissipation in a total area of 1.1mm² including a closed-loop neural stimulator. A set of four features are extracted from the spatially filtered neural data to achieve a high detection accuracy at minimal hardware cost. The performance is demonstrated both by early detection and termination of kainic acid-induced seizures in freely moving rats and by offline evaluation on human intracranial EEG (iEEG) data. Our design improves upon previous works by over 40× reduction in power-area product per channel. This improved energy-area efficiency is a key step towards new designs with higher spatiotemporal resolution, larger array size, and therefore, better seizure detection accuracy.

Details

Actions