Humidity-Gated Memristive Dynamics Enabling Near-Sensor Spiking Computation for Wind Direction and Noise-Resilient Speech Recognition.

Abstract

Spiking in-sensor systems rely on discrete input transitions to enable event-driven encoding. However, humidity signals change gradually and continuously, lacking intrinsic thresholds required for spike generation. This mismatch poses a fundamental challenge for applying in-sensor spiking computation to humidity. Here, a neuromorphic humidity-sensing platform based on a threshold-switching memristor with an asymmetric Ag/Nafion/ITO structure is reported. The Nafion layer serves both as a humidity transduction medium and as an electrochemical matrix for silver filament formation. Increased ambient humidity reduces ionic migration barriers, enabling volatile conductance changes over six orders of magnitude and switching speeds down to 60 ns. Importantly, the switching threshold decreases from 0.8 V at 50% relative humidity to 0.2 V at 90%, providing an embedded gating mechanism that produces spikes only when humidity exceeds defined levels. To evaluate the system in practical scenarios, real-time classification of spatiotemporal humidity gradients for wind direction inference, as well as noise-resilient speech recognition via exhalation-induced humidity cues is demonstrated. These results demonstrate a hardware-level strategy for event-driven encoding of slow environmental dynamics, offering a pathway toward efficient, low-power sensory systems for edge-intelligent applications.