A thermally engineered NbOx memristor with CdTe2 interlayers for high-accuracy ECG arrhythmia detection

Abstract

Niobium oxide (NbO_x)-based threshold switching memristors (TSMs) demonstrate significant promise for hardware implementation of neuromorphic computing, but their threshold stability's susceptibility to external environmental variations remains unclear. This work elucidates the switching mechanism in an NbO_x-TSM incorporating a low-thermal-conductivity CdTe₂ interlayer, which operates via the Poole–Frenkel (PF) emission model. Our investigation reveals that a double interlayer structure yields the highest effective thermal resistance, thereby most effectively reducing the threshold voltage. By implementing this structure, we enhanced the switching stability by 30.9%. Furthermore, increasing the thickness of the double-sided interlayers from 3 nm to 9 nm improved the stability by an additional 24.7% while simultaneously lowering the threshold voltage. The impact of the interlayer thickness on oscillatory behavior was systematically analyzed within a leaky integrate-and-fire (LIF) neuron circuit, where the observed frequency saturation phenomenon provides critical guidance for thermal engineering design. Capitalizing on these findings, we developed a multimodal, integrated memristor-based system for electrocardiogram (ECG) arrhythmia detection that leverages device thickness and temperature characteristics to achieve a classification accuracy rate of 90.0%. This work underscores the significant value of such physically interpretable devices for the hardware realization of neuromorphic computing.