187037 research outputs found
Sort by
고유연 플렉서블 태양전지와 광검출기를 위한 콜로이드 양자점 박막의 균열 억제에 관한 연구
No full text학위논문(박사) - 한국과학기술원 : 전기및전자공학부, 2025.2,[vi, 81 p. :]Infra-red sensitive optoelectronic devices, with their wide application areas, have been utilized in biomedical imaging, autonomous vehicles, and alternative power sources. To extend their use in a broader perspective, enhancing the flexibility of these devices has been an extensively researched topic. Colloidal quantum dots (CQD), with their adjustable absorption spectrum extending to infra-red region, have been a promising alternative to be used in optoelectronic devices. However, their mechanical properties were lacking. In this dissertation, we aim to enhance the flexibility of CQD films by utilizing flexible polymers. We investigate a surface anchoring polymer treatment to strengthen CQD films against crack formation. In addition, we offer a simple method to fabricate CQD/polymer mixed films to achieve flexibility by stopping crack propagation. We discuss the fabrication details of CQD films in detail, and study the crack formation, charge transport and device performance as photovoltaic cells and photodetectors.한국과학기술원 :전기및전자공학부
확산 기반 생성 모델을 위한 조건부 정보 추정 향상
No full text학위논문(박사) - 한국과학기술원 : 산업및시스템공학과, 2025.2,[viii, 96 p. :]Diffusion-based generative models have recently received considerable attention due to their remarkable performance across a range of generative tasks. However, diffusion models still have limitations and can produce unsatisfactory results in some cases. In particular, accurate estimation of various conditional information is crucial during the training and inference processes of diffusion models. For example, it is necessary to estimate the perturbed sample at the current diffusion timestep based on the perturbed sample from the previous step, or to accurately infer conditional information for generating samples that align with given conditions. This dissertation explores methods to improve the estimation of conditional information, thereby improving the overall performance of diffusion models. Specifically, the study proposes approaches to improve performance by refining the transition kernel, handling label noise, and optimizing text condition embeddings.
First, we address the problem of denoising estimation in the sampling process of pre-trained diffusion models. We propose Diffusion Rejection Sampling (DiffRS), which uses a rejection sampling scheme that aligns the sampling transition kernels with the true ones at each timestep. The proposed method can be viewed as a mechanism that evaluates the quality of samples at each intermediate timestep and refines them with varying effort depending on the sample. Theoretical analysis shows that DiffRS can achieve a tighter bound on sampling error compared to pre-trained models. Empirical results demonstrate the state-of-the-art performance of DiffRS on the benchmark datasets and the effectiveness of DiffRS for fast diffusion samplers and large-scale text-to-image diffusion models.
Second, we address the training of conditional diffusion models in the presence of noisy conditional inputs, referred to as noisy labels, in the training dataset. We demonstrate that such label noise leads to condition mismatch and quality degradation of generated data. Then, we propose Transition-aware weighted Denoising Score Matching (TDSM) objective for training conditional diffusion models with noisy labels. The TDSM objective contains a weighted sum of score networks, incorporating instance-wise and time-dependent label transition probabilities. We introduce a transition-aware weight estimator, which leverages a time-dependent noisy-label classifier distinctively customized to the diffusion process. Through experiments across various datasets and noisy label settings, TDSM improves the quality of generated samples aligned with given conditions.
Finally, we address the problem of conditional embeddings during the sampling process of pre-trained conditional diffusion models. Conditional embeddings are typically extracted from the pre-trained model of conditions and are used the same for all timesteps of the diffusion model. We propose Dynamic Adaptive Text Embedding (DATE) to improve the conditional information by optimizing the conditional embedding at each diffusion timestep during the sampling process. To optimize the conditional embedding, we propose an objective function that ensures that the final sample predicted at the current timestep aligns well with the given condition. The optimal conditional embedding is obtained by computing the gradient of the objective function using a first-order Taylor approximation, and no additional model training is required. Experimental results with text-to-image diffusion models show improvements in image quality and text-image alignment.한국과학기술원 :산업및시스템공학과
블록체인 시스템 내의 전략적 참여자들에 대한 모델링 및 분석
No full text학위논문(박사) - 한국과학기술원 : 산업및시스템공학과, 2025.2,[vi, 108 p. :]This dissertation examines economic mechanisms in blockchain systems through mathematical modeling and analysis. The examined problems span multiple protocol layers, including consensus mechanisms, transaction processing, and automated market making. The analysis employs techniques from optimal control theory, queueing theory, and stochastic processes to derive quantitative results about system behavior.
In Chapter 2, we analyze intentional block delays in Proof-of-Work blockchain systems through a game-theoretic model of mining behavior. The analysis establishes the necessary and sufficient condition for the existence of mining gaps, where miners strategically delay operations based on reward and cost parameters. A two-player game formulation demonstrates how mining gaps interact with difficulty adjustment algorithms, potentially leading to system instability. The theoretical results are tested against empirical data from the Bitcoin network to determine stability bounds under various reward and fee structures.
In Chapter 3, we model blockchain transaction processing as an M/GK/1 priority queue where users compete through fee bidding. The analysis derives semi-closed form expressions for steady-state quantities and establishes the relationship between user delay costs and optimal bidding strategies under general block generation time distributions. Numerical results from Bitcoin network data demonstrate how fee structures and user behavior vary across different blockchain implementations with distinct consensus parameters.
Chapter 4 presents a mathematical model of concentrated liquidity provision in constant product automated market makers. The model assumes geometric Brownian motion price dynamics and derives optimal liquidity provision boundaries through stochastic control theory. The analysis determines optimal position ranges and expected returns under various market conditions, accounting for both changes in position value and accumulated fees from trading activity.
The mathematical frameworks developed in this dissertation enable quantitative analysis of blockchain system behavior across multiple protocol layers. The results establish precise conditions for various economic phenomena in blockchain systems and determine bounds on system stability and efficiency under different parameter regimes. This research contributes to the emerging field of cryptoeconomics by developing theoretical foundations that account for the unique characteristics of blockchain-based financial infrastructure.한국과학기술원 :산업및시스템공학과
버클링 거동을 고려할 수 있는 딥러닝 트러스 유한요소의 개발
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[iv, 88 p. :]In this study, we propose a truss finite element that can represent buckling behavior using deep learning. Truss structures may experience buckling at specific members under extreme loads, impacting the overall structural failure. When modeling a truss using truss finite elements, it is challenging to consider the buckling of each member. To account for buckling, beam finite elements need to be employed, but this requires significantly more degrees of freedom compared to using truss elements. Additionally, nonlinear analysis of complex truss structures can pose convergence and computational efficiency issues. In this research, we model a single truss member using beam finite elements for nonlinear buckling analysis and calculate the relationship between axial force and axial deformation. Utilizing artificial neural network training, we construct a machine learning element that can predict the behavior of truss structures under various loading conditions. When conducting finite element analysis of truss structures, if truss elements requiring nonlinear analysis are present, we replace them with pre-built machine learning elements that consider buckling, thus performing analysis with reduced degrees of freedom. Nonlinear analysis in finite element methods typically involves iterative algorithms in each incremental step to achieve equilibrium. However, the truss element proposed in this paper utilizes the relationship between converged axial force and axial deformation to generate element stiffness and internal forces, eliminating the need for iterative processes for convergence. This approach significantly reduces the required degrees of freedom in finite element analysis and improves the convergence and time efficiency of nonlinear analysis.한국과학기술원 :기계공학과
Atomic-referenced, low noise, high-repetition-rate optical frequency comb
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[vii, 89 p. :]광 주파수 빗은 우수한 타이밍 지터와 높은 주파수 안정도에 힘입어 아날로그-디지털 컨버터, 정밀 거리 측정, 마이크로파 생성 등 다양한 분야에서 활용되고 있다. 특히, 원자 시계에 안정화된 광 주파수 빗은 높은 주파수 안정도와 정확도를 제공하기 때문에 시각 동기, 분자 분광학, 초장기선 전파 간섭계와 같은 응용에서 중요한 신호원으로 사용된다. 그러나 이를 구현하기 위해서는 광 주파수 빗과 원자 시계 기반 레퍼런스 사이의 타이밍 오차를 정밀하게 검출하여 원자 시계의 성능 손실을 최소화 해야한다. 본 연구에서는 전-광 샘플링 기반 타이밍 검출기를 사용하여 원자 시계에 안정화된 광 주파수 빗을 생성하였다. 먼저 원자 시계의 성능 손실 없이 그 안정도를 광 주파수 빗에 전달하기 위해 전-광 샘플링 기반 타이밍 검출기의 이론적인 한계에 도달하기 위한 방법을 제시하고 아토초 수준 광-마이크로파 동기화를 구현하였다. 이를 기반으로 초장기선 전파 간섭계에서의 응용을 위해 수소 메이져에 동기화된 모드-잠금 광섬유 레이져로부터 고안정도 마이크로파 및 광대역 RF 빗을 생성하였다. 또한, 고-반복률 광 주파수 빗인 마이크로콤으로부터 생성한 포토닉 마이크로파를 광 레퍼런스에 안정화된 저잡음 모드-잠금 광섬유 레이져에 동기화하여 저잡음, 고안정도 마이크로콤을 생성하였으며, 광 원자 시계에 안정화된 마이크로콤의 생성 가능성에 대해 다루었다.한국과학기술원 :기계공학과
노면 정보 추정을 이용한 노면소음 능동소음제어에서의 응용
No full text학위논문(박사) - 한국과학기술원 : 기계공학과, 2025.2,[vii, 93 p. :]Road profile information represents the vertical displacement relative to the road direction, capturing the physical characteristics of the surface. This study aims to estimate road profile information by utilizing acceleration signals measured by accelerometers mounted on a vehicle’s chassis. To achieve this, we derive the transfer function between vehicle acceleration and road profile information, using it to extract road profile data from the measured acceleration signals. The estimation performance was validated under various speed conditions and vehicle characteristics. The estimated profile information is stored in a server database, enabling information sharing across different types of vehicles. This shared road profile information can be applied to active noise control (ANC) of road noise. In-vehicle road noise has complex propagation characteristics, requiring multiple sensors and speakers and high-performance signal processing units. However, constraints due to in-vehicle installation and hardware limitations restrict computational performance. Moreover, as road conditions change during driving, the ANC control filter must be updated, which may reduce noise control performance. To address these issues, this study proposes an ANC method that utilizes road profile information. By using the stored road profile information, the control filter can be adjusted preemptively before the vehicle encounters a change in the road profile, thereby enhancing control performance. The optimal filter for each predicted profile is obtained by generating a reference signal spectrum from the profile information of that surface, which is then used to design the control filter. This study proposes a method for obtaining a control filter based on the reference signal spectrum, which was validated through simulation. Additionally, a sub-optimal control filter was developed for ANC of road noise, and its effectiveness was evaluated. Lastly, a method for determining the optimal filter under conditions of limited measurement signals was proposed and verified using actual road profile data. This study confirms that the proposed road estimation process can effectively enhance ANC performance.한국과학기술원 :기계공학과
전도성 유무기 소재를 활용한 에너지 생산 및 저장 시스템의 메커니즘에 관한 연구
No full text학위논문(박사) - 한국과학기술원 : 신소재공학과, 2025.2,[ix, 127 p. :]Over the past century, the utilization of electrical energy has surged, making the sustainability of energy production and consumption a significant issue. As of 2021, global electricity consumption reached 2.8 × 10^4 TWh, raising concerns regarding the environmental burden of fossil fuels and the risk of resource depletion. Particularly, South Korea's renewable energy share remains low at 9.22%, although 2023 presents an opportunity for approximately 30% of total electricity generation to be supplied by renewable sources. This dissertation analyzes the energy production and storage mechanisms of water-based energy harvesters and lithium-sulfur batteries utilizing conductive organic and inorganic materials. Water-based energy harvesters enable efficient energy production in low-power environments, while lithium-sulfur batteries demonstrate potential as nextgeneration energy storage devices based on their high theoretical capacity. The present research aims to address the technical limitations identified within these systems by exploring new solutions through model experiments and electrochemical analysis.한국과학기술원 :신소재공학과
다기능 뉴로모픽 소자용 다원소 필라멘트 멤리스터에 대한 연구
No full text학위논문(박사) - 한국과학기술원 : 신소재공학과, 2025.2,[xi, 80 p. :]Memristor exhibits changes in internal resistance in response to external voltage, leading to diverse research in material science. In recent years, various applications of memristor have emerged, including not only in non-volatile memory but also in neuromorphic devices, sensors, computing devices, and more, making memristor highly valuable in electronics. These applications leverage various mechanisms underlying memristors. Therefore, understanding the mechanisms behind memristor and exploring their applications is an exciting endeavor. In this dissertation, we introduce a new type of multielement filament memristor characterized by hybrid mechanisms and its potential for multifunctional applications. The multielement filament memristor we propose combines the alloyed metallic filament with the oxygen vacancy filament, achieving both the high on/off ratio and the analog properties with high reliability. Also, by controlling the compliance current, both resistive and threshold switching characteristics are achieved, making them multifunctionally available for synaptic and neuron devices. In particular, when used as a neuron device, it exhibits novel multimodal nociceptive behaviors, responding not only to electrical signals but also to thermal and mechanical stimuli.한국과학기술원 :신소재공학과
투과전자현미경을 이용한 원자단위의 비스무트 옥시칼코겐나이트 기반 계면 관찰
No full text학위논문(박사) - 한국과학기술원 : 신소재공학과, 2025.2,[xi, 146 p. :]Bismuth oxychalcogenides (Bi2O2Se, Bi2O2S), as oxide-based materials, exhibit structural stability when exposed to air, and when grown on SrTiO3 or mica substrates, they demonstrate high charge mobility due to the absence of surface and interface defects. This makes them promising candidates for future field-effect transistors (FETs), potentially overcoming the limitations of silicon-based semiconductors. While most previous research has focused on the discovery and synthesis of new materials, recent studies emphasize that the growth of next-generation semiconductor materials, often just a few nanometers thick, can be significantly influenced by the interface. It has been reported that the interface of two-dimensional semiconductors can alter the intrinsic properties of the material. Typically, interfaces consist of only one or two atomic layers, and the most effective method for their structural and chemical analysis is cross-sectional transmission electron microscopy (TEM). By analyzing interfaces at the atomic scale and imaging the positions of oxygen and transition metal atoms, it is possible to visualize defects, vacancies, and potential variations within the interface through crystallographic analysis. Furthermore, the atomic structure of the semiconductor-interface may differ depending on the type of substrate, potentially affecting the semiconductor's electrical properties. Moreover, phenomena that occur exclusively at the atomic-scale interface can also be observed. Investigating these phenomena at semiconductor interfaces holds significant academic impact and, through the development of advanced analytical systems, can aid in the selection and development of next-generation semiconductor materials, providing a competitive edge in materials technology.한국과학기술원 :신소재공학과
원자간력 현미경을 활용한 데이터 기반 통찰과 시각화: 압전 생체재료의 이해와 응용
No full text학위논문(박사) - 한국과학기술원 : 신소재공학과, 2025.2,[vii, 95 p. :]Piezoelectricity, defined by the linear coupling between mechanical stress and electric polarization, is a fundamental property of many biological systems. This phenomenon arises from the non-centrosymmetric crystal structures found in biomaterials, including amino acid proteins, collagen, deoxyribonucleic acid (DNA), and bacteriophage. Elucidating the intrinsic electromechanical properties of these biomaterials and their functional implications within biological systems has become a key motivation driving research into piezoelectricity in both natural and bioinspired materials at the nanoscale. This research explores the intersection of materials science and biology, emphasizing the application of atomic force microscopy (AFM) and data-driven analysis to study piezoelectric biomaterials. By integrating advanced AFM techniques with computer vision-based algorithms, this study aims to achieve a comprehensive understanding of the electromechanical properties of biomaterials, which are crucial for various medical and technological applications.
In the first study, we investigate the enhanced piezoresponse of the ferroelectric polymer, poly(vinylidene fluoride-co-trifluoro ethylene) [P(VDF-TrFE)]. The negative piezoelectricity in ferroelectric polymers cannot be fully explained by their crystal structure alone. Instead, the semi-crystalline nature of these materials suggests a complex coupling between crystalline and amorphous phases. This study focuses on elucidating the role of the phase boundary gradient – a zone of diminishing crystallinity – in enhancing piezoresponse. Using X-ray diffraction (XRD) and advanced piezoresponse force microscopy (PFM) techniques, we examine how variations in thermal treatments and ceramic filler incorporation impact the nanostructure and electromechanical behavior of P(VDF-TrFE). Results show that controlled thermal processing within the ferroelectric temperature range modifies the nanostructure, expanding phase boundary gradients and enhancing the piezoresponse, despite reduced overall crystallinity. In barium titanate composites, radially varying piezoresponse patterns further highlight the potential of phase boundary engineering in improving flexible ferroelectric performance for applications such as sensors, actuators, and energy harvesters.
Next, we examine the piezoelectric and topographic properties of P(VDF-TrFE) scaffolds designed for bone regeneration. Bone regeneration necessitates a combinatorial approach that integrates mechanical, electrical, and biological stimuli to mimic the native cellular microenvironment. While this has led to multiple studies on biocompatible piezoelectric scaffolds, their application in bone tissue engineering has been constrained by the absence of a material that can simultaneously provide the complex electromechanical environment of native bone tissue. In this study, a novel biomimetic scaffold incorporating hydroxyapatite (HAp) into P(VDF-TrFE) in a freestanding form is introduced, leveraging HAp’s natural osteogenic potential within a piezoelectric framework. In vitro and in vivo experiments demonstrate the remarkable potential of these scaffolds to promote bone regeneration, facilitated by electrical, topographical, and paracrine mechanisms.
Finally, we explore the distinct piezoelectric properties of beta-amyloid (Aβ) fibrils, implicated in Alzheimer’s disease pathology. Vector piezoresponse force microscopy analysis reveals that Aβ fibrils possess spiraling piezoelectric domains along their length and exhibit a lateral piezoelectric constant of 44.1 pC N-1. Additionally, continuous sideband Kelvin probe force microscopy (KPFM) imaging indicates that the charge-induced surface potential of a single Aβ fibril can exceed +1,700 mV in response to applied forces. These findings underscore the exceptional mechano-electrical surface traits of pathological Aβ fibrils, which surpass those of typical biological components and may correlate with the neurodegenerative symptoms associated with Alzheimer’s disease.
The studies presented collectively emphasize the concept of emergent properties, where interactions within a system result in behaviors not found in individual components. For instance, HAp, when combined with P(VDF-TrFE) and applied to bone, results in the emergence of medical functionality. Similarly, Aβ peptides in their individual forms are trivial, but when aggregated into fibrillar form, serve as major indicators of Alzheimer’s. From engineered biomaterials designed for bone regeneration to the pathological attributes of Aβ fibrils, these studies illustrate how piezoelectricity and electromechanical properties can arise from complex material interactions. Together, they underscore the potential of nanoscale studies to unravel the collective behavior of materials, with significant implications for advancing medical and technological advancements.한국과학기술원 :신소재공학과