Physical Reservoir Computing for Real-Time Electrocardiogram Arrhythmia Detection Through Controlled Ion Dynamics in Electrochemical Random-Access Memory

Abstract

From the perspective of developing low-power mobile healthcare devices capable of real-time electrogram diagnosis, memristor-based physical reservoir computing (PRC) offers a promising alternative to conventional deep neural network (DNN)-based systems. Here, real-time electrocardiogram (ECG) monitoring and arrhythmia detection are demonstrated using electrochemical random-access memory (ECRAM)-based PRC. ECRAM devices provide the millisecond-range temporal resolution required for bio-potential signals like ECG. Through material and process engineering, it is identified that higher ionic conductivity (σion) in the electrolyte layer and lower ionic diffusivity (Dion) in the channel layer are crucial for achieving non-linear dynamics and fading memory characteristics. In addition, LaF3/WOx-based ECRAM exhibits low-power operation (≈300 pW spike−1) with minimal cycle-to-cycle (CTC) variation (<10%). Arrhythmia detection tests confirmed the feasibility of real-time ECG monitoring, achieving a high classification accuracy of 93.04% with a 50-fold reduction in training parameters compared to DNN-based systems. Therefore, the developed LaF3/WOx-based ECRAM with engineering guidelines of ion dynamics makes a significant contribution to mobile healthcare systems for electrogram diagnosis.