Artificial synapses with a sponge-like double-layer porous oxide memristor

Gao et al. NPG Asia Materials (2021)13:3 https://doi.org/10.1038/s41427-020-00274-9 ARTICLE NPG Asia Materials Open Access Artificial synapses with a sponge-like double-layer porous oxide memristor Qin Gao1, Anping Huang 1, Jing Zhang2, Yuhang Ji1, Jingjing Zhang1, Xueliang Chen1, Xueli Geng1, Qi Hu3, Mei Wang1, Zhisong Xiao1 and Paul K. Chu 4 1234567890():,; 1234567890():,; 1234567890():,; 1234567890():,; Abstract Closely following the rapid development of artificial intelligence, studies of the human brain and neurobiology are focusing on the biological mechanisms of neurons and synapses. Herein, a memory system employing a nanoporous double-layer structure for simulation of synaptic functions is described. The sponge-like double-layer porous (SLDLP) oxide stack of Pt/porous LiCoO2/porous SiO2/Si is designed as presynaptic and postsynaptic membranes. This bionic structure exhibits high ON–OFF ratios up to 108 during the stability test, and data can be maintained for 105 s despite a small read voltage of 0.5 V. Typical synaptic functions, such as nonlinear transmission characteristics, spike-timingdependent plasticity, and learning-experience behaviors, are achieved simultaneously with this device. Based on the hydrodynamic transport mechanism of water molecules in porous sponges and the principle of water storage, the synaptic behavior of the device is discussed. The SLDLP oxide memristor is very promising due to its excellent synaptic performance and potential in neuromorphic computing. Introduction The development of deep learning is closely associated with the advancement of artificial intelligence, which is inseparable from brain science and neurobiology. Many recent research activities are fog on the biological mechanisms of neurons and synapses1,2, and a good understanding of the biological brain is crucial to the design of intelligent machines3–6. The synaptic signal is transmitted from one neuron to the next, while the signal passing through the neural pathway is remembered through the stimulation of a pulsed signal7,8. Memristors with a small size, low energy consumption, and nonvolatile performance are vital to many applications related to information storage, analog circuits, artificial intelligence, and analog neural networks9,10. Because of the similarity between the memristor and synapse, studies on Correspondence: Anping Huang () 1 Key Laboratory of Micro-nano Measurement-Manipulation and Physics Ministry of Education, Department of Physics, Beihang University, 100191 Beijing, China 2 School of Information Science and Technology, North China University of Technology, 100144 Beijing, China Full list of author information is available at the end of the article the mechanisms and materials simulating the behavior and functions of nerve cells are crucial to the development of biologically inspired devices and prototypes10,11. Materials suitable for memristors mainly include oxides12, sulfides13, perovskites14,15, two-dimensional materials16,17, and related materials18,19, and in particular, oxide-based memristors have been widely studied due to their high switching speed, large density, and compatibility with complementary metal–oxide–semiconductor processing20,21. The conduction mechanism of most of these materials is based on the formation and breakage of conductive metallic filaments via the accumulation of oxygen vacancies or metal atoms, such as Cu or Ag22–25. However, the stochastic and unpredictable formation process results in device variability, fluctuation of resistance states, and excessive “write” noise, consequently undermining the stability of the materials23,25. Therefore, it is necessary to control the switching filament formation and dynamics of memristors. To overcome these hurdles, approaches, including stoichiometry control26, thermal annealing27, and insertion of thin metal layers28, have been suggested, and the control of filaments and their © The Author(s) 2021 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Gao et al. NPG Asia Materials (2021)13:3 formation at desirable locations must be improved and better understood. Introducing a porous structure into the resistive layer is one of the viable methods to provide channels for ion transport. For example, Wang et al. prepared multiple nanoporous (NP) SiOx layers to control filament formation29,30, and Tour et al. proposed a threedimensional NP Ta2O5−x structure with graphene boasting high-density storage and low power consumption31,32. Obviously, a porous structure can regulate the conducting channel and enhance the device performance33. Combining bionic ideas with device design is an effective way to study defects in electronic devices and the underlying mechanisms. Zhang et al. proposed a threedimensional spiral microstructure as the basic unit and constructed bionic soft three-dimensional mesh materials with defect-insensitive characteristics, by varying the spatial topology to reproduce the anisotropic nonlinear mechanical response of biological tissues34. Fan et al. prepared perovskite nanowires for an electrochemical eye with a hemispherical retina that can imitate the photoreceptors in the human retina11. Furthermore, the transport characteristics of water in a natural cave structure have been combined with hydrodynamics and ion transport to design a karst-like hierarchical porous structure with good characteristics35. However, the nonuniform pore diameters in the silicon oxide layer cause large variations in the essential switching elements, and random and nonuniform ion migration in LiCoO2 hampers the optimal switching functionalities. Lithium ion migration in LiCoO2-based memristors is similar to the information exchange process between synapses and neurons in the brain36–40. LiCoO2 is used in Li-ion batteries to improve the capacity and mitigate the volume expansion during deintercalation of lithium ions under repetitive charging and discharging41–43. It is also important to investigate the relationship between the device performance and NP structure, as well as potential device applications. In fact, it is advisable to adopt the bionic route in designing and utilizing the functions (...truncated)