An electroforming-free, analog interface-type memristor based on a SrFeOx epitaxial heterojunction for neuromorphic computing. (May 2021)

• Record Type: Journal Article
• Authors: Rao, J.; Fan, Z.; Hong, L.; Cheng, S.; Huang, Q.; Zhao, J.; Xiang, X.; Guo, E.-J.; Guo, H.; Hou, Z.; Chen, Y.; Lu, X.; Zhou, G.; Gao, X.; Liu, J.-M.
• Is Part Of: Materials today physics. Volume 18(2021)
• Publication Date: 2021-05
• Subjects: Interface-type resistive switching -- SrFeOx -- Topotactic phase transformation -- Memristors -- Neuromorphic computing; Materials science -- Periodicals; Physics -- Periodicals; Electronic journals; 530.41
• DOI: 10.1016/j.mtphys.2021.100392
• Languages: English
• ISSNs: 2542-5293

Abstract

Abstract: Distinct from the conductive filament-type counterparts, the interface-type resistive switching (RS) devices are electroforming-free and exhibit bidirectionally continuous conductance changes, making them promising candidates as analog synapses. While the interface-type RS devices typically operate through the interfacial oxygen migration, materials which can tolerate a wide range of oxygen non-stoichiometry and possess high oxygen mobility are therefore demanded. SrFeO x (SFO), which can easily transform between a conductive, oxygenated perovskite SrFeO3 (PV-SFO) phase and an insulating, oxygen-vacancy-rich brownmillerite SrFeO2.5 (BM-SFO) phase under electric field, emerges as a suitable material. Herein, an interface-type RS device is ingeniously structured by two epitaxial SFO layers: a PV-SFO matrix layer and an ultrathin BM-SFO interfacial layer, aiming to leverage the oxygen migration-induced interfacial BM-PV phase transformation to realize the gradual conductance modulation. Experimentally, the fabricated device exhibits electroforming-free, analog memristive behavior. This device also emulates essential synaptic functions, including excitatory postsynaptic current, paired-pulse facilitation, transition from short-term memory to long-term memory, spike-timing-dependent plasticity, and potentiation/depression. A simulated neural network built from the SFO-based synapses achieves accuracies above 88% for image recognition. This work provides a novel approach to use the SFO family of topotactic materials for developing analog synapses as building blocks for neuromorphic computing circuits. Graphical abstract: Image 1 Highlights: SrFeOx epitaxial heterojunction-based interface-type memristors are developed. Electroforming-free, analog memristive behavior is achieved. Synaptic functions including EPSC, PPF, STDP, and STM/LTM are mimicked. Simulated neural network achieves accuracies above 88% for image recognition.