Biomimetic-voltage nitride-enhanced memristor arrays for femtojoule in-memory biosignal processing.

Edge audio recognition demands extreme energy efficiency and noise resilience. Conventional separate denoising and recognition modules incur high overhead and latency, hindering deployment on resource-constrained edge devices. Inspired by biological audition, here we show a memristor-based neuromorphic system that monolithically integrates on-chip audio denoising and recognition. Our Pt/IGZO/SiNx/Ta memristor achieves a 19 mV biomimetic voltage and high endurance of 106 cycles. Fabricated into a uniform 1-kb crossbar array, the system supports partitioned processing: a volatile-switching denoising region enables real-time noise suppression, activating based on input signal strength without erasure steps, while a non-volatile region performs high-precision classification. This end-to-end solution attains 100% accuracy on 10-class audio signals (500 samples) after 15 training epochs with post-denoising, consuming 1.44 fJ per denoising operation, outperforming non-denoised approaches in convergence speed and robustness. Confusion matrices confirm >90% class-specific accuracy under noise, establishing a pathway for miniaturized, energy-scalable edge hardware.