검색 상세

A Study on Multivalent Metal Oxides-Based Neuromorphic Memory Devices

  • IQBAL SHAHID (IQBAL SHAHID, 아주대학교 일반대학원)

초록/요약

In the past few decades, because of Moore's Law and the rapid growth of information technology markets, computing system performance has been dramatically improved over time. As a result, the traditional electromagnetic passive circuit components of capacitor, inductor, and resistor are insufficient to describe the proportion of circuit theory tying all electromagnetic characteristics. Numerous novel electronic aspects have been investigated in this regard so far. The development of resistive random-access memories (ReRAM) is a new type of non-volatile memory that uses electrically driven resistive switching of an active/insulator material. Among the different types of elements, memristors are the emerging two-terminal memory technologies with a number of attractive prospects. Since the discovery of memristors in ReRAM, various types of memristor devices have been reported, including phase-change random-access memory (PRAM), magneto-resistance random-access memory (MRAM), and ferroelectric random-access memory (FeRAM). Moreover, unlike the two-terminal switch, the transistor serves as the processing hardware’s backbone because it permits for improved control of charge migration. Due to the leakage issues in two-terminal neuromorphic device, memtransistors (memristor + transistor) are thought to be promising candidates as three-terminal devices. Various types of metal-oxides as an active/insulator layer have been reported in memristor/memtransistors, Mott metal-to-insulator transition (MIT) as a “Mott memory” in various active/insulator materials under an application of electric field has received a lot of attention, owing to its fast resistive switching and low-power consumption. In this regard, memristor and memtransistor architectures based on multivalent metal oxides have indicated significant potential for replacement or supplementing traditional computing programs based on the von Neumann architecture, which deal with big-data challenges such as the memory wall. However, several key challenges have not been adequately investigated in two- and three-terminal neuromorphic memory devices yet. Here, in this thesis, we addressed such key challenges in multivalent metal oxides-based memristor and memtransistor devices. In chapter 2, the development of VOx based memtransistor device is introduced to achieve the potential of Mott transition of multiphasic vanadium oxides (VOx) for emerging memory applications. The volatile memory behaviors of the VOx memtransistors are observed in both two- and three-terminal measurements using electrical characterizations. The mixed VOx/SiO2 interface is strongly related to their capacitive memory effect and resistive switching mechanisms (called VSiOx). At low bias voltages (0.5 V), VSiOx enhances the Mott transition in VOx, resulting in a low power consumption of the memtransistor. Furthermore, the memtransistors' synaptic functions (synaptic weight) are demonstrated, as well as their fast-switching time of ≈35 ns and tunable memory retention (via drain and gate biases). Altogether, the outcomes pave the way for the development of ultrafast and femto-joule power-consuming neuromorphic devices. In chapter 3, two-terminal memristor devices are being investigated as a possible candidate for mimicking the brain-like functionality in order to facilitate artificial synaptic functions. Because of their small features, extreme compactness, and low power to measure synaptic functionalities, two-terminal nanoscale level devices are considered as a promising candidate for realizing biosynapses. We show how to tune the nanoscale charge transport and resistive switching behavior of VOx thin films by changing the substrate morphology. Surprisingly, conductive atomic force microscopy (c-AFM)-based current maps discovered electric field inhomogeneity caused by morphology changes in the substrate. The corresponding etched devices have shown reliable bipolar resistive switching behavior. An increase in channel area and a decrease in work function were observed as the etching time of the substrate increased. Additionally, localized synaptic characteristics were produced from the subsequent devices, revealing a solid conduction path at the devices' specific bright spots. Furthermore, finite element simulations confirm that applying tip voltages modulates the production of nanoscale current conduction in certain etched devices. These findings represent a novel method for producing nanoscale artificial synaptic functions. In chapter 4, we demonstrated that the memristor devices, which are inspired by biosynapses, have received a lot of consideration as an important step toward high-performance artificial neuromorphic functions. Regardless of recent significant advances in this area for the practical applications of artificial neural networks, there are still technical challenges such as well-stabled devices, low-power switching, and robustness, all of which are critical for achieving ideal neuromorphic functioning. We present a simple method for fabricating a highly reproducible thermal nanostructured-based memristor device. Our memristor device has a remarkable rectifying resistive switching behavior with distinct synaptic functions, such as bulk (104 pulses) and nanoscale synaptic weight, paired-pulsed potentiation/depression, and learning and forgetting abilities similar to those of the human brain, due to the asymmetric interfaces. Furthermore, our nanostructured memristor device successfully mimics Pavlovian associative learning behaviors. Overall, our findings point to a promising path forward for artificial neuromorphic computing. In chapter 5, we demonstrate the Mott transition behavior, which is one of the most dramatic phenomena involving insulator-to-metal transition (IMT) in some metal oxides. Recently, it has become one of the hottest research trends, and various Mott devices have been investigated for memory applications. That has inspired us to create a proximity-oxidation-produced Ti–V–O thin film sputtered on n-Si substrate. The key finding here is the mixed phase of the Ti–V–O film and the interface of Ti–V–O/n-Si enabling a double Mott switching characteristic – in other words, a Ti–V–O device can switch from two HRS regions to two LRS regions. Particularly, the Ti–V–O/n-Si interface plays two roles: supporting the double Mott switching and facilitating a controlled Mott transition behavior up to 100 °C. Due to the remarkably endurant synaptic weight behavior up to 100 °C, the usability of our Ti–V–O devices for synaptic applications is demonstrated and discussed in detail to pave a way for uses in advanced neuromorphic applications.

more

목차

Chapter 1. Introduction 1
1.1. Trends in Memory Technologies 1
1.2. Resistive Switching Devices 3
1.2.1. Materials 5
1.2.2. Modes 8
1.2.3. Operation Principles 9
1.3. Mott Transition 13
1.4. Neuromorphic Memory Devices 14
1.4.1. Two-Terminal Neuromorphic Devices 17
1.4.2. Three-Terminal Neuromorphic Devices 22
1.5. Dissertation Overview 25
1.6. Associated publications 29
Chapter 2. Multiphasic VOx Memtransistor 30
2.1. Introduction 30
2.2. Experimental Section 32
2.2.1. Device Fabrication 32
2.2.2. Device Characterization 33
2.2.3. Device Measurements 33
2.3. Results and Discussions 34
2.3.1. Proposed VOx-Based Memtransistor Structure 34
2.3.2. TEM-EELS Analysis 40
2.3.3. Electrical Measurements of VOx Memtransistor 46
2.3.4. Nanosecond Artificial Synaptic Characteristics 59
2.4. Summary 67
Chapter 3. Surface-engineered VOx memristor 70
3.1. Introduction 70
3.2. Experimental Section 73
3.2.1. Device Fabrication 73
3.2.2. Device characterization/measurements 74
3.2.3. Finite element simulation 74
3.3. Results and Discussion 74
3.3.1. Device Fabrication Method 74
3.3.2. Surface characterization 77
3.3.3. c-AFM and KPFM Analysis 80
3.3.4. Nanoscale Synaptic Characteristics 88
3.4. Summary 93
Chapter 4. Rectifying Resistive Switching of CoOx 95
4.1. Introduction 95
4.2. Experimental Section 97
4.2.1. Device Fabrication 97
4.2.2. Characterization of the Device 98
4.2.3. Electrical measurements 99
4.3. Results and discussion 99
4.3.1. Device Fabrication and Characterizations 99
4.3.2. Rectifying Resistive Switching and Mechanism 104
4.3.3. Synaptic Functions of Thermal Nanostructured Memristor 111
4.3.4. Pavlovian Learning by CoOx Memristor 118
4.4. Summary 121
Chapter 5. Proximity-Oxidation-Induced Mott Transition 123
5.1. Introduction 123
5.2. Experimental Section 126
5.2.1. Device Fabrication 126
5.2.2. Device Characterization 127
5.2.3. Characterization Tools 127
5.3. Results and Discussions 128
5.3.1. Fabrication and Characterizations of the Ti–V–O device 128
5.3.2. Elemental Composition of Ti–V–O Device 132
5.3.3. Mott Switching of the Ti–V–O devices 135
5.3.4. Synaptic Characteristics of Ti–V–O Device 140
5.4. Summary 145
Chapter 6. Conclusion and Future Work 147
6.1. Conclusion 147
6.2. Future Work 149
Bibliography 150

more
반출 Meta View 목록