A One-Shot Learning, Online-Tuning, Closed-Loop Epilepsy Management SoC with 0.97μJ/Classification and 97.8% Vector-Based Sensitivity

Abstract

We propose a patient-specific closed-loop epilepsy tracking and real-time suppression SoC with the first-in-literature one-shot learning and online tuning. The entire SoC consumes the lowest energy reported to date of 0.97μJ/class. and occupies the smallest area of 0.13mm2/Ch. Verified with CHB-MIT database and a local hospital patient, the 9.8b ENOB 2-Cycle AFE combined with the GTCA-SVM DBE achieves vector-based sensitivity, specificity, and latency of 97.8%, 99.5%, and <1s.