TiN/TiOx/BaTiO3/Pt heterostructure memristors for adaptive neuromorphic systems

Abstract

Ferroelectric memristors offer a transformative solution to the von Neumann bottleneck by integrating memory, learning, and perception into a unified platform that is ideal for neuromorphic computing. In this study, we present a TiN/TiOx/BaTiO₃/Pt heterojunction memristor fabricated via radiofrequency magnetron sputtering, demonstrating high-performance analog resistive switching characterized by a switching ratio of ~50, ultralow operating voltage (~0.6 V), low-reset variability (4.86 %), and energy-efficient operation (1.76 pJ). The bilayer design enables precise control over 32 discrete conductance levels, supporting reliable 5-bit data storage. In addition to memory functionality, the device emulates a broad range spectrum of synaptic plasticity behaviors, such as long-term potentiation/depression (LTP/LTD), paired-pulse facilitation (PPF), post-tetanic potentiation (PTP), spike-timing-dependent plasticity (STDP), and spike-voltage-dependent plasticity (SVDP), all of which are enabled by tunable oxygen vacancy filament dynamics. Remarkably, the memristor also exhibits biomimetic nociceptive features such as threshold activation, non-adaptive response, allodynia, and hyperalgesia, establishing an artificial pain perception mechanism in a compact two-terminal structure. When employed in artificial neural network simulations for Modified National Institute of Standards and Technology handwritten digit classification, the device achieves an accuracy of 94.7 %, which closely approaches the software-based ideal performance of 95.1 %. Moreover, accuracy is maintained at values greater than 94 % across multiple LTP/LTD cycles, confirming excellent reliability. These results render TiOx/BaTiO₃ bilayer memristors as powerful candidates for next-generation neuromorphic platforms with embedded hazard awareness and cognitive adaptability. © 2025 Elsevier B.V.