Ferroelectric Charging Tunnel Junctions: Resolving Transport Trade-Offs via Trap-Assisted Barrier Modulation for Physical Reservoir Computing

초록

Dynamic memristors that combine a wide dynamic range, intrinsic self-rectification, and rich temporal dynamics are key building blocks for physical reservoir computing, but state-of-the-art HfO2-based ferroelectric tunnel junctions (FTJs) suffer from thickness-limited transport trade-offs and poorly resolved conduction mechanisms. Here, a ferroelectric charging tunnel junction (FCTJ) based on an IGZO/HZO stack is demonstrated as a scalable two-terminal reservoir node. Introducing a carrier-charging IGZO interlayer into a W/HZO/W stack modulates the tunneling barrier and couples ferroelectric polarization to interface oxygen-vacancy trapping, yielding both an Ion/Ioff and a rectification ratio exceeding 104 and satisfying over 10% read margin in N x N passive crossbar arrays with N = 2000. Temperature- and time-dependent transport, XPS depth profiling, and conductance-method analysis quantitatively link shallow, while reverse leakage is suppressed by depletion and oxygen-reservoir effects in IGZO. These composite ferroelectric-trapping dynamics support multi-timescale synaptic plasticity with modest variability. Leveraging experimentally calibrated device dynamics, FCTJ-based physical reservoirs outperform a multilayer perceptron using only quasi-static conductance states for MNIST classification and accurately process UCR gesture signals and chaotic H & eacute;non/Lorenz trajectories with low prediction error. This work establishes IGZO/HZO FCTJs as CMOS-compatible reservoir nodes and outlines a general design strategy for ferroelectric and oxide-based physical reservoir computing hardware.

키워드