서브메뉴
검색
Synthesis, Fabrication, and Characterization of Two-Dimensional Neuromorphic Electronic Nanomaterials
Synthesis, Fabrication, and Characterization of Two-Dimensional Neuromorphic Electronic Nanomaterials
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211151031
- ISBN
- 9798382760346
- DDC
- 620.5
- 서명/저자
- Synthesis, Fabrication, and Characterization of Two-Dimensional Neuromorphic Electronic Nanomaterials
- 발행사항
- [Sl] : Northwestern University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 196 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-11, Section: B.
- 주기사항
- Advisor: Hersam, Mark C.
- 학위논문주기
- Thesis (Ph.D.)--Northwestern University, 2024.
- 초록/해제
- 요약Increased reliance on artificial intelligence and machine learning for modern day computing has imposed great strain on the performance of the underlying hardware. As conventional computing operates on the continuous transfer of data between the processor and storage units, executing complex algorithms with high computational overhead will incur energy-inefficiencies that will present an outstanding challenge in eventual scale-up of artificial intelligence. Neuromorphic computing is an emerging bio-realistic computing paradigm primarily motivated by the marvels of the human brain, the world's most efficient processor. Neuromorphic demonstrations such as two-terminal memristors, phase change memory, and three-terminal synaptic transistors have achieved limited success due to their restrictive geometries and rigid learning capabilities. The two-dimensional (2D) memtransistor, in contrast, exploits the strong electrostatics of atomically thin materials to provide co-location of the field-effect volatile transistor and nonvolatile memristor capabilities in a lateral geometry that enables multi-terminal architecture and gate-tunable learning. By integrating atomically thin two-dimensional materials that enable enhanced electrostatic tunability, monolayer polycrystalline MoS2 memtransistors have achieved gate-tunable memristive switching, linearity, and reconfigurability. Similarly, four-terminal dual-gated MoS2 memtransistors have demonstrated reduced crosstalk and sneak currents in scalable crossbar architectures, simplifying integration challenges that have hindered memristive architectures based on bulk materials. Despite the unique attributes of memtransistors, their implementation in neuromorphic architectures has been limited to conventional artificial neural networks, suggesting that their full potential for artificial intelligence and machine learning has not yet been realized. Moreover, the incumbent memtransistor demonstrations, require large operating voltages that are impractical for eventual integration and scale.This thesis addresses three tenets critical to materials science and engineering - synthesis, characterization, and fabrication - with the aim of progressing towards more bio-realistic neuronal computing networks. The enclosed chapters focus on these themes to various degrees, underscoring the interdisciplinary nature and scope of neuromorphic computing engineering research. While the efforts and strategies presented here focus on MoS2, the concepts and process flows can be generalized to other two-dimensional van der Waals materials, especially p-doped semiconducting transition metal dichalcogenides, to engineer emergent synaptic behaviors at low operating voltages for next-generation memtransistor devices. The first work presented will introduce MoS2 memtransistors with a wide range of learning behaviors achieved through a combination of enhanced electrostatic control and tailored gate bias pulsing profiles. Using monolayer MoS2 grown on sapphire, long-term potentiation and depression behaviors are singularly modulated by the gate electrode polarity, an observation that parallels the synaptic weight update and neuroplasticity in biological systems. By enhancing the relative importance of the vertical field effect from the gate voltage compared to the lateral field from the drain voltage, these devices show a greater reconfigurability of synaptic behavior compared to previously reported MoS2 memtransistors grown on SiO2. Different gate bias pulsing strategies further diversify the library of learning curves. The resulting gate-tunable learning behavior is then modeled in a simplified spike-timing-dependent plasticity scheme to perform unsupervised continuous learning in a simulated spiking neural network. We show that continuous learning, a previously underexplored cognitive concept in hardware neuromorphic computing, circumvents traditional trade-offs between image recognition accuracy and resource allocation. Overall, this work demonstrates that reconfigurable MoS2 memtransistors provide unique hardware accelerator opportunities for energy-efficient artificial intelligence.The second work presents the most competitive MoS2 memtransistor to date. Fabrication of back-gated devices with small channel length ( 400 nm) and high-κ Al2O3 as gate dielectric enables low-voltage operation (sub-1 V) with a 104 transistor ON/OFF and 10 switching ON/OFF ratios that outcompete state-of-the-art memtransistor prototypes. The small channel length and large channel width, defined with respect to the MoS2 grain size, augment the points of intersection of grain boundaries and contact edges, where resistive switching is hypothesized to occur. Finite-element simulations reveal enhanced electrostatic effects and band bending near the source contact in the back-gated design compared to incumbent top-gated memtransistors that enable voltage reduction without compromising memristive and transistor performance.
- 일반주제명
- Nanotechnology
- 일반주제명
- Materials science
- 일반주제명
- Engineering
- 키워드
- Machine learning
- 키워드
- Memtransistors
- 키워드
- Neural networks
- 기타저자
- Northwestern University Materials Science and Engineering
- 기본자료저록
- Dissertations Abstracts International. 85-11B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017160506
■00520250211151031
■006m o d
■007cr#unu||||||||
■020 ▼a9798382760346
■035 ▼a(MiAaPQ)AAI30997256
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620.5
■1001 ▼aLiu, Stephanie E.
■24510▼aSynthesis, Fabrication, and Characterization of Two-Dimensional Neuromorphic Electronic Nanomaterials
■260 ▼a[Sl]▼bNorthwestern University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a196 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 85-11, Section: B.
■500 ▼aAdvisor: Hersam, Mark C.
■5021 ▼aThesis (Ph.D.)--Northwestern University, 2024.
■520 ▼aIncreased reliance on artificial intelligence and machine learning for modern day computing has imposed great strain on the performance of the underlying hardware. As conventional computing operates on the continuous transfer of data between the processor and storage units, executing complex algorithms with high computational overhead will incur energy-inefficiencies that will present an outstanding challenge in eventual scale-up of artificial intelligence. Neuromorphic computing is an emerging bio-realistic computing paradigm primarily motivated by the marvels of the human brain, the world's most efficient processor. Neuromorphic demonstrations such as two-terminal memristors, phase change memory, and three-terminal synaptic transistors have achieved limited success due to their restrictive geometries and rigid learning capabilities. The two-dimensional (2D) memtransistor, in contrast, exploits the strong electrostatics of atomically thin materials to provide co-location of the field-effect volatile transistor and nonvolatile memristor capabilities in a lateral geometry that enables multi-terminal architecture and gate-tunable learning. By integrating atomically thin two-dimensional materials that enable enhanced electrostatic tunability, monolayer polycrystalline MoS2 memtransistors have achieved gate-tunable memristive switching, linearity, and reconfigurability. Similarly, four-terminal dual-gated MoS2 memtransistors have demonstrated reduced crosstalk and sneak currents in scalable crossbar architectures, simplifying integration challenges that have hindered memristive architectures based on bulk materials. Despite the unique attributes of memtransistors, their implementation in neuromorphic architectures has been limited to conventional artificial neural networks, suggesting that their full potential for artificial intelligence and machine learning has not yet been realized. Moreover, the incumbent memtransistor demonstrations, require large operating voltages that are impractical for eventual integration and scale.This thesis addresses three tenets critical to materials science and engineering - synthesis, characterization, and fabrication - with the aim of progressing towards more bio-realistic neuronal computing networks. The enclosed chapters focus on these themes to various degrees, underscoring the interdisciplinary nature and scope of neuromorphic computing engineering research. While the efforts and strategies presented here focus on MoS2, the concepts and process flows can be generalized to other two-dimensional van der Waals materials, especially p-doped semiconducting transition metal dichalcogenides, to engineer emergent synaptic behaviors at low operating voltages for next-generation memtransistor devices. The first work presented will introduce MoS2 memtransistors with a wide range of learning behaviors achieved through a combination of enhanced electrostatic control and tailored gate bias pulsing profiles. Using monolayer MoS2 grown on sapphire, long-term potentiation and depression behaviors are singularly modulated by the gate electrode polarity, an observation that parallels the synaptic weight update and neuroplasticity in biological systems. By enhancing the relative importance of the vertical field effect from the gate voltage compared to the lateral field from the drain voltage, these devices show a greater reconfigurability of synaptic behavior compared to previously reported MoS2 memtransistors grown on SiO2. Different gate bias pulsing strategies further diversify the library of learning curves. The resulting gate-tunable learning behavior is then modeled in a simplified spike-timing-dependent plasticity scheme to perform unsupervised continuous learning in a simulated spiking neural network. We show that continuous learning, a previously underexplored cognitive concept in hardware neuromorphic computing, circumvents traditional trade-offs between image recognition accuracy and resource allocation. Overall, this work demonstrates that reconfigurable MoS2 memtransistors provide unique hardware accelerator opportunities for energy-efficient artificial intelligence.The second work presents the most competitive MoS2 memtransistor to date. Fabrication of back-gated devices with small channel length ( 400 nm) and high-κ Al2O3 as gate dielectric enables low-voltage operation (sub-1 V) with a 104 transistor ON/OFF and 10 switching ON/OFF ratios that outcompete state-of-the-art memtransistor prototypes. The small channel length and large channel width, defined with respect to the MoS2 grain size, augment the points of intersection of grain boundaries and contact edges, where resistive switching is hypothesized to occur. Finite-element simulations reveal enhanced electrostatic effects and band bending near the source contact in the back-gated design compared to incumbent top-gated memtransistors that enable voltage reduction without compromising memristive and transistor performance.
■590 ▼aSchool code: 0163.
■650 4▼aNanotechnology
■650 4▼aMaterials science
■650 4▼aEngineering
■653 ▼aMachine learning
■653 ▼aMemtransistors
■653 ▼aNeuromorphic computing
■653 ▼aGate-tunable learning
■653 ▼aNeural networks
■690 ▼a0794
■690 ▼a0652
■690 ▼a0800
■690 ▼a0537
■71020▼aNorthwestern University▼bMaterials Science and Engineering.
■7730 ▼tDissertations Abstracts International▼g85-11B.
■790 ▼a0163
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17160506▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


