SnNb₂O₆-Based Capacitive Memristive Synapses with Intrinsic LIF Dynamics for Neuromorphic Epilepsy Detection.

Abstract

Integrating tunable leaky-integrate-and-fire (LIF) dynamics with a synaptic memristor holds profound promise in the early diagnosis of neurological disorders. Here, a SnNb₂O₆-based capacitive memristor is presented, which emulates a biosynaptic neuron through an LIF-type excitatory postsynaptic current (EPSC) for epileptic seizure monitoring. The device functions with stable coupled bipolar switching and capacitive response under both DC and pulsed operation over 500 consecutive cycles; in particular, single-pulse stimulus triggers a highly reproducible LIF-type three-stage EPSC feature. Various stimulation protocols enable diverse synaptic behaviors, and the device response can transition from transient to persistent states by tuning readout-delay. Mechanistic analysis attributes the capacitive memristor behavior to the coupling effect between interfacial polarization/displacement current and dynamic Schottky barrier modulation. By employing measured LIF-type EPSC as convolution kernels for raw electroencephalogram (EEG) signals in a lightweight one-dimensional convolutional neural network (1D-CNN), a 95.8% classification accuracy for three epilepsy-related states is achieved on the Bonn dataset. The results highlight the potential of the capacitive memristor for on-chip epileptic seizure monitoring.