Coexisting Volatile and Nonvolatile Switching in 3D ALD‐IGZO Vertical RRAM for Fully Hardware‐Based Wide Reservoir Computing

ABSTRACT

Although indium‐gallium‐zinc oxide (IGZO) has been widely used in thin‐film transistors (TFTs), its application as an active switching medium in resistive random‐access memory (RRAM) has remained relatively less explored, particularly for vertical RRAM (VRRAM) architectures. Here, we report an ALD‐IGZO VRRAM platform in which the IGZO switching layer is conformally formed along vertical sidewalls, providing excellent step coverage and atomic‐level compositional controllability for uniform and reliable switching. This platform realizes a compact dual‐mode device, where the volatile switching mode before forming serves as the reservoir layer and the nonvolatile switching mode after forming functions as the readout layer. Leveraging these coexisting dynamics, we implement a fully hardware‐based reservoir computing (RC) system with a wide architecture by operating multiple sub‐reservoirs in parallel under distinct volatility conditions, achieving enhanced nonlinear mapping and feature separation. The proposed 2F ALD‐IGZO VRRAM wide RC achieves 91.3% accuracy on the MNIST dataset with an estimated energy consumption of 3.70 nJ per classification, offering a scalable route toward learning‐capable in‐memory neuromorphic computing.