Linear Analog Weight Update and Long-Term Plasticity in Pt/p-LiCoO_x/n-IGZO/Pt and Pt/p-LiCoO_x/p-NiO/Pt Memristors by Controlling Li Ion Distribution for Neuromorphic Computing

Emerging data-centric applications such as artificial intelligence, IoT, and autonomous driving systems requires novel high energy-efficient computing architectures overcoming von-Neumann bottleneck problem of delayed data transfer in current computing systems. Neuromorphic computing is one of promising candidates for new computing architectures, mimicking human brain that processes complicated tasks with a low energy consumption of ~ 20 W [1]. To realize neuromorphic computing, a highly linear and symmetric weight update and its long-term plasticity with a low operation voltage for potentiation and depression are essential features of an artificial synapse. In this study, two types of memristors are proposed as artificial synapses controlling movable Li⁺ ion redistribution that is responsible for conductance modulation in these devices. Because Li⁺ ion is highly movable element and act as dopant in metal-oxides, it can change the resistance states in many metal-oxide systems. Therefore, it has a potential to provide a plausible route to realize low-power artificial synapse through controlling its distribution in the memristors.<br/>The first one is the pn-junction memristor with Pt/p-LiCoO_x/n-InGaZnO(IGZO)/Pt structure and it is operated by modulated interfacial energy barrier between p-LiCoO_x and n-IGZO as a result of Li⁺ ion migration. Because Li⁺ ions act as p-type dopants in IGZO channel layer [2], Li⁺ ions in p-LiCoO_x layer migrate to n-IGZO layer upon voltage application and modulate the interface energy barrier between p-LiCoO_x and n-IGZO. This voltage-driven Li⁺ ion migration enables the analog conductance modulation in pn-junction memristor. In particular, because Li⁺ ions are highly movable even under low bias, the conductance could be tuned by two orders of magnitude even at low voltage less than ±3 V. In addition, engineering of the device structure, including insertion of ion blocking layer, local encapsulation of Li⁺ ions, and use of UV/ozone treatment on the layers could enhance long-term plasticity through controlling Li⁺ ions more stably. Thanks to its pn diode characteristics, it can also mitigate sneak current problems in high-density crossbar array architecture as a self-rectifying synaptic memristor.<br/>The other one is Pt/p-LiCoO_x/p-NiO/Pt memristor (pp-junction), which is operated by modulated interfacial energy barrier between p-NiO and Pt bottom electrode with respect to the Li⁺ ion distribution in NiO layer. Because this device utilizes the low energy barrier modulation between p-NiO and high work function Pt electrode, the weight update could be highly linear and symmetric through even minute change in the barrier height. Also, this modulation comes from the redistribution of movable Li⁺ ions, the operation voltage could be reduced to the range of ±2 V. In addition to analog weight update, the pp-junction memristor shows good synaptic properties such as paired pulse facilitation (PPF), short-term and long-term plasticity (STP and LTP) depending on stimulation conditions. The four retained long-term resistance states could be achieved with respect to the applied voltage amplitude. Use of tunable interfacial energy barrier by Li⁺ ion redistribution facilitates analog conductance modulation in a high speed with a low voltage.<br/>These results demonstrate potential applications of Pt/p-LiCoO_x/n-IGZO/Pt and Pt/p-LiCoO_x/p-NiO/Pt memristors as artificial synapses with linear, symmetric, and analog conductance change at low operation voltage with a long-term plasticity by energy barrier modulation with Li⁺ ion distribution.<br/><br/><b>References</b><br/>[1] Jiang, S., Nie, S., He, Y., Liu, R., Chen, C., & Wan, Q. (2019). <i>Mater. Today Nano</i>, <i>8</i>, 100059.<br/>[2] Jeong, B., Han, J., Noh, T., & Yoon, T. S. (2023). <i>ACS Appl. Electron. Mater.</i>, <i>5</i>(6), 3470-3479.