Nonlinear quantized conductance dynamics in vertical SiN RRAM for scalable memory-learning integration

Abstract

We report a vertical resistive random-access memory device based on a Pt/SiN/Ti stack, designed for multi-bit storage and neuromorphic computing. The device exhibits stable bipolar switching and achieves up to 7-bit (128-level) conductance states through precise control of compliance current and reset voltage. Quantized conductance plateaus, corresponding to integer and half-integer multiples of the quantum conductance G<inf>0</inf> = 2e2/h, reveal atomic-scale filament dynamics governed by nonlinear conduction processes. Diverse synaptic plasticity functions, including spike-number-, spike-rate-, spike-duration-, and spike-amplitude-dependent plasticity, were experimentally emulated. Neuromorphic simulations for the Modified National Institute of Standards and Technology dataset achieved classification accuracies exceeding 94 %, confirming the device's suitability for high-precision weight modulation. The vertical architecture ensures scalability toward three-dimensional integration, while robust retention and compatibility with current-based multi-bit modulation highlight its potential for complex-system-inspired edge AI and in-memory computing hardware.