Polarity-dependent dual-mode AlN-embedded RRAM with improved stochastic switching and synaptic modulation for neuromorphic computing

Abstract

We present a Pt/Al/TaOx/AlN/Al2O3/Pt resistive random-access memory device that enables polarity-dependent, dual-mode switching within a single cell, exhibiting abrupt digital and gradual analog conductance modulation. The incorporation of an AlN layer between the TaOx switching layer and the Al2O3 tunnel barrier (with a thickness of 1.2 nm) functions as a built-in current limiter, promoting controlled filament formation and inherent self-compliance without the need for external circuitry. Under positive bias, localized soft breakdown near the Al/TaOx interface induces abrupt switching with a high ON/OFF ratio and reliable endurance over 100 cycles. Conversely, negative bias facilitates stepwise filament growth near the AlN/Al2O3 interface, enabling smooth analog switching and precise control of multilevel conductance. Using an incremental step pulse with a verify algorithm, the device achieved 6-bit resolution, excellent analog endurance over 500 cycles, and retention >10 000 s. In addition, the device successfully emulates biologically relevant forms of synaptic plasticity, including spike-amplitude-dependent, spike-rate-dependent, and spike-width-dependent-under fixed amplitude stimulation conditions. The device's layered architecture not only ensures stable switching behavior but also enhances device reliability by suppressing current overshoots. These results highlight the device's strong potential for energy-efficient, hardware-level neuromorphic computing, as demonstrated by a multilayer perceptron that achieved 93.5% classification accuracy on the Modified National Institute of Standards and Technology dataset using experimentally extracted conductance values without quantization or preprocessing.