Filamentary-based gradual dual resistive switching in the TiO2/Al2O3 bilayer for remote supervised method

Abstract The development of analog memristive devices with symmetric and controllable switching is critical for the implementation of neuromorphic hardware capable of spike-based supervised learning. Here, we report a TiO 2 /Al 2 O 3 bilayer memristor that exhibits filamentary-driven, gradual dual bipolar resistive switching, enabling both spike-timing-dependent plasticity (STDP) and anti-STDP within a single device. The resistance of the TiO 2 layer is modulated by the growth dynamics of oxygen vacancy filaments in the Al 2 O 3 layer, which in turn alters the electric field distribution and enables progressive filament formation and rupture. This mechanism allows for the quasi-symmetric, polarity-reversible switching without external circuitry or explicit mode-switching procedures. The device demonstrates a high functional yield of 95% and a low switching voltage variation (coefficient of variation < 3%). Crossbar array integration confirms reliable bidirectional synaptic updates in response to identical input spikes. Furthermore, a simulation incorporating experimentally extracted switching parameters achieves 84.5% classification accuracy on the MNIST dataset using the remote supervised method. These results establish the TiO 2 /Al 2 O 3 bilayer memristor as a scalable, CMOS-compatible platform for implementing biologically inspired learning rules in next-generation neuromorphic systems.