A Study on Liquid-based Memory Devices for Synaptic Applications

Abstract

In the trend of high computing performances of the electronic machines, the demand for memory devices with small size, low voltage, low power consumption, and superior performance has been under extensive consideration. However, silicon-based flash memory, which dominates the market of data storage devices, has been difficult in meeting the needs of future development of data storage devices due to its physical and technological limitations, such as high operation voltage, high power consu mption and low retention capacity. Resistivity switching memory (RSM) devices have emerged as the next-generation memory devices owe to the simple device structure, good scalability, and the possibility of using various materials. RSM devices typically have a two-terminal structure of metal/insulator/metal. The characteristics of the RSM can be designed for high-performance such as low operating energy, fast operating speed, and high stability depending on the material of the insulating layer. In addition, RSM devices are possible to emulate synaptic functions by designing analog resistivity switching characteristics. Emulating the synaptic functions has emerged as an application of memory devices beyond conventional computing technology because it has the advantages for the improving the computing performance. Neuromorphic computing emulates highly efficient computing processes in the brain which has a high-efficiency, high-performance computing functions that can handle complex computation with the extremely low-energy consumption. Neuromorphic computing emulates highly complex computation processes of the brain such as synaptic plasticity, backpropagation learning, and nonlinear synaptic weight update. The synaptic plasticity indicates the variation in synaptic weights as a function of synaptic activity. Diverse functional materials have been used in the fabrication of artificial synapses. The critical parameters for neuromorphic computation are the number of memory states, energy consumption, and the operation endurance characteristics. For the neuromorphic devices, ion-based materials are effective as insulating materials in the RSM devices. Ion-based materials have advantages to induce gradual resistivity changing characteristics. Owe to the movement of ions is slower than that of electrons, it is possible to control them in more precisely. However, in the case of ion-based materials, the mass of charges is heavier than that of electrons, therefore higher energy is required to migrate the ions in the RSM device. To solve this problem, a liquid phase-based resistance change memory devices have been proposed. Diffuse coefficient of ions is generally higher in liquid phase materials than solid phase materials. Based on these features, liquid-based RSM devices can achieve a low power and high efficiency characteristics by minimizing the migration energy of the ions. Furthermore, liquid-based devices have advantages for the emulating the synaptic characteristics because the synaptic functions are realized by the movement of ions and neurotransmitters between cerebrospinal fluid and neurons in the biological synapse. This dissertation focuses on realization of a liquid-based synaptic device. I have demonstrated resistance switching memory devices based on liquid phase materials. Furthermore, synaptic characteristics were emulated using liquid-based devices. It was shown that various types of devices can be made using liquids, and the characteristics of the nervous system were imitated using each device. These works provide a promising route to emulate synaptic functions using the liquids. In chapter 1, I briefly introduce the background for these researches for the development of liquid-based synaptic devices. In chapter 2, I report a RSM device using a liquid for the low voltage operation characteristics. I implemented a resistance change memory device with a liquid phase-based two terminal structure. A device was designed using silver, an aqueous silver nitrate solution, and a gold electrode. Low voltage operating characteristics could be obtained by minimizing the migration energy of silver ions. The device was able to induce resistance change characteristics by forming a filament inside the liquid through an oxidation-reduction reaction of silver. The device could exhibit both analog switching characteristics and abort switching characteristics. By adjusting the concentration of the aqueous solution, the resistance change characteristic could be implemented, and even the flexible characteristics could be implemented using the liquid material of the device. In addition, a liquid-based synaptic array is implemented by designing and manufacturing an array structure that can hold solutions using a 3D printer. Even though this device is made of liquid, the filament is maintained stably, showing stable memory retention characteristics and switching endurance characteristics. In addition, since the position of the filament formed inside the liquid phase is specified based on entropy, uniform switching characteristics could be implemented. In addition, it was confirmed that the thickness of the filament formed inside the liquid is also specified by entropy, so it shows self-compliance characteristics. As a result, it is confirmed that a resistance change device that can be operated at a stable and low power can be implemented by utilizing a liquid phase. In chapter 3, I reported synaptic devices using liquid materials. Using the diffusion behaviors of the ions in the liquid, synaptic devices were demonstrated with liquid-phase materials. Liquid-based synaptic devices exhibited potentiation, depression, EPSC, IPSC, action potential, excitatory-inhibitory balancing, and STDP characteristics. First, it was possible to emulate the synaptic plasticity behaviors using the liquid-based synaptic devices. Using the electrochemical reaction of the ions in the liquid, synaptic devices were demonstrated. The synaptic behaviors were emulated by controlling the conductivity of the liquid materials in the 2-terminal structured memory devices. In addition, the action potential characteristic was emulated by mimicking a neural cell that controls the potential difference between the inside and outside of the cell using ionic channel and ionic pump. In particular, the rectification characteristics of ionic diodes were demonstrated through injection of neurotransmitters, thereby emulating the behaviors of chemical synapses. Finally, using PEDOT:PSS and membrane materials, a device that exhibits an excitatory reaction and an inhibitory reaction according to the type of ions contained in the solution is implemented. By inducing the interaction of excitatory and inhibitory reactions in one device, the enhancing properties of the nervous system were also emulated.