Multilevel Resistance Modulation of Au-Nanodot Embedded Self-Rectifying Memristor for Neuromorphic Hardware

The memristor-based materials have been researched as next-generation computing devices due to their simple metal-insulator-metal (MIM) structure and ease of device fabrication. The memristive crossbar array (CBA) is effective for the implementation of multiply-and-accumulate (MAC) operations by simply collecting current at the single bit line (BL) from the multiple biased memristors through word lines (WLs), which follows Kirchhoff’s current law (KCL). Since the memristive CBA consists of multiple BLs, the multiple MAC operations can be performed in parallel, proving a more efficient computing device for ca. neuromorphic system as the size of the CBA increases.^{[1, 2]}<br/>Despite the promising structural advantages of memristive CBA for MAC operations, the non-ideal characteristics, such as sneak current, line resistance, variations, and electroforming, hinder more practical applications. Such a problem can be mitigated by introducing additional selecting devices, known as selectors or transistors, to the memristors.^[3] However, engineering the electrical characteristic to match the appropriate operational voltage and the current level is challenging, especially when processing variation exists throughout the CBA structure. Another aspect to consider for MAC operations is energy consumption. Introducing other three-terminal devices, however, often induces higher operational current levels in the range from several hundred μA to mA and the operational voltage, resulting in significant additional energy consumption in the larger CBA structure.<br/>To alleviate the difficulties, self-rectifying memristor (SRM)-based CBA is introduced that uses a bi-layered oxide structure instead of introducing the additional selecting devices and has a much lower operation current level with a range of sub-μA to nA. Since the switching mechanism is governed by trap-rich and insulating layers with Schottky contact, the electroforming process is not required in the CBA structure, and the reversed current is significantly suppressed to prevent the sneak current.^{[4, 5]}<br/>However, SRM-based CBA still requires additional control in cycle-to-cycle variations and retention of states, resulting from the variation in the electronic trapping and detrapping in trap levels, which are essential parameters to determine the accuracy of the MAC operations and multilevel states in the implementation of the artificial synapse.<br/>In this work, nanometer-sized gold nanodots (Au NDs) were formed at the interface of switching and rectifying layers to promote the electric field concentration in alignment with trap states. Enhancing the electric field concentration improves the overall cycle-to-cycle variations and retention characteristics. The conduction model to explain the improvement from Au NDs in Pt/Ta₂O₅/HfO_x/TiN structure is proposed based on the experimental result from the modulation of the size and position of Au NDs with transmission electron microscope (TEM) analyses. With conductive atomic force microscopy analysis and the finite element method simulations, the distribution of the internal electric field and the current density could be verified inside Au-NDs embedded SRM. Finally, the demonstration of MNIST classification in neural network application using eight stable states (3-bit) of the SRM-based CBA shows improved accuracy compared to non-inserted Au NDs SRM.<br/><br/><b>References</b><br/>[1] J. Kim, H. C. Woo, S. Lee, B. J. Lee, T. Park, <i>Adv. Intell. Syst.</i> <b>2022</b>, 4 (8), 2100256<br/>[2] T. Park, S. S. Kim, B. J. Lee, T. W. Park, H. J. Kim, C. S. Hwang, <i>Nanoscale</i> <b>2023</b>, 15, 6387-6395<br/>[3] H. C. Woo, J. Kim, S. Lee, H. J. Kim, C. S. Hwang, <i>Adv. Electron. Mater.</i> <b>2022</b>, 8 (12), 2200656<br/>[4] J. H. Yoon. S. S. Song, I.-H. Yoo, J. Y. Seok, K. J. Yoon, D. E. Kwon, T. H Park, C.S. Hwang, <i>Adv. Funct. Mater.</i> <b>2014</b>, 24 (32), 5086-5095<br/>[5] S. S. Kim, S. K. Yong, J. Kim, J. M. Choi, T. W. Park, H. Y. Kim, H. J. Kim, C. S. Hwang, <i>Adv. Electron. Mater.</i> <b>2023</b>, 9 (3), 2200998