Search results
Publication Iron-Based Microstructures For Biodegradable And Non-Biodegradable Applications(2022-01-01) Zhang, TaoTraditional microfabrication techniques for microelectromechanical systems (MEMS) have emphasized silicon and silicon-related materials. As more MEMS applications emerge, including the Internet of Things (IoT) and biodegradable applications, widely expanded materials sets are being considered. Since iron (Fe) has good electrical and magnetic properties, it undergoes degradation in moist or oxygenated environments, and since it is prevalent in both biological and environmental systems, it is being seriously considered as a candidate material to satisfy these new needs. In this dissertation, we developed iron-based microstructures for use in biodegradable and non-biodegradable applications. Specifically, electrical interconnects were developed with biodegradable iron/polymer composites. From the materials perspective, electrical, mechanical, and electrochemical properties of the composite material under physiological degradation were investigated. Stable electrical resistivity of the composite was shown over 20 days in degradation, even if under mechanical straining. Good adhesion to similarly biodegradable substrates of the composite was also shown in degradation. Further, the biodegradability of the composite was demonstrated by characterizing its degradation behavior. From the application perspective, a functional lifetime of over 4 days of the electrical interconnects with the packaging was achieved under physiological degradation to prove the capability of iron-based materials in biodegradable applications. In addition, miniaturized step-up transformers with laminated electrodeposited iron-alloy cores were developed. From the materials perspective, the magnetic permeability of the iron-alloy thin films at micro-scale thicknesses was characterized to demonstrate the retained magnetic properties of these materials at smaller scales. From the application perspective, the performance of the transformers was characterized at the device and circuit level to demonstrate the utility of iron-based bulk material in non-biodegradable magnetic applications. Finally, electrostatically tunable inductors fabricated from iron/ceramic composites were developed. From the materials perspective, the magnetic permeability tunability of the iron/ceramic composites was demonstrated. From the application perspective, the reasonable tuning ability of the device with minimum power consumption was shown to prove the utility of iron-based composite materials in non-biodegradable magnetic applications.Publication Extracting Generalizable Hierarchical Patterns Of Functional Connectivity In The Brain(2022-01-01) Sahoo, DushyantThe study of the functional organization of the human brain using resting-state functional MRI (rsfMRI) has been of significant interest in cognitive neuroscience for over two decades. The functional organization is characterized by patterns that are believed to be hierarchical in nature. From a clinical context, studying these patterns has become important for understanding various disorders such as Major Depressive Disorder, Autism, Schizophrenia, etc. However, extraction of these interpretable patterns might face challenges in multi-site rsfMRI studies due to variability introduced due to confounding variability introduced by different sites and scanners. This can reduce the predictive power and reproducibility of the patterns, affecting the confidence in using these patterns as biomarkers for assessing and predicting disease. In this thesis, we focus on the problem of robustly extracting hierarchical patterns that can be used as biomarkers for diseases. We propose a matrix factorization based method to extract interpretable hierarchical decomposition of the rsfRMI data. We couple the method with adversarial learning to improve inter-site robustness in multi-site studies, removing non-biological variability that can result in less interpretable and discriminative biomarkers. Finally, a generative-discriminative model is built on top of the proposed framework to extract robust patterns/biomarkers characterizing Major Depressive Disorder. Results on large multi-site rsfMRI studies show the effectiveness of our method in uncovering reproducible connectivity patterns across individuals with high predictive power while maintaining clinical interpretability. Our framework robustly identifies brain patterns characterizing MDD and provides an understanding of the manifestation of the disorder from a functional networks perspective which can be crucial for effective diagnosis, treatment and prevention. The results demonstrate the method's utility and facilitate a broader understanding of the human brain from a functional perspective.Publication Additively Fabricated Laminated Inductors For Miniaturized Switching Power Converters(2022-01-01) Pyo, Jun BeomThis thesis work proposed an innovative idea to solve the problem of building a small integrated inductor, a bottleneck of realizing power supply-on chip (PwrSoC): utilization of material that has intermediate electrical conductivity as an interlamination material for a magnetic core We theoretically analyzed that the electrical conductivity of 0.1 S/m - 1 S/m is a desirable range for an interlamination material that enables sequential electrodeposition (CMOS-compatible process) yet suppression of eddy-current loss. Polypyrrole (PPy) was selected as such interlamination material to validate our idea. We empirically demonstrated sequential stacking of PPy and metallic magnetic alloy, NiFe via continuous and additive electrodeposition process. We also experimentally showed that PPy interlamination effectively suppresses eddy-current loss by building and testing inductors. A laminated 10-layer NiFe inductor showed higher inductance retention (88% vs. 21%), and lower AC resistance (1.68 ohm vs. 12.7 ohm) at 8 MHz compared to a single layer NiFe inductor having comparable total NiFe thickness; both higher inductance retention and lower AC resistance are signs of suppressed eddy current loss. We then investigated the practicality of our findings. A thick core having 45 NiFe layers was fabricated to show the developed technology can achieve tens of microns in thickness which is necessary for watt scale power converters. The air gap was introduced to increase saturation current up to 500 mA which was then implemented in a buck converter. Our toroid inductor and commercial inductor showed comparable power efficiency of ~75% in a buck converter that steps down 2.7V to 1.2 V at 20 MHz switching frequency. Although our inductor was larger than the commercial inductors, utilization of higher saturation flux density material like CoNiFe and optimization of inductor design expects to reduce inductor size.Publication Virtualizing Reconfigurable Architectures: From Fpgas To Beyond(2022-01-01) Zha, YueWith field-programmable gate arrays (FPGAs) being widely deployed in data centers to enhance the computing performance, an efficient virtualization support is required to fully unleash the potential of cloud FPGAs. However, the system support for FPGAs in the context of the cloud environment is still in its infancy, which leads to a low resource utilization due to the tight coupling between compilation and resource allocation. Moreover, the system support proposed in existing works is limited to a homogeneous FPGA cluster comprising identical FPGA devices, which is hard to be extended to a heterogeneous FPGA cluster that comprises multiple types of FPGAs. As the FPGA cloud is expected to become increasingly heterogeneous due to the hardware rolling upgrade strategy, it is necessary to provide efficient virtualization support for the heterogeneous FPGA cluster. In this dissertation, we first identify three pairs of conflicting requirements from runtime management and offline compilation, which are related to the tradeoff between flexibility and efficiency. These conflicting requirements are the fundamental reason why the single-level abstraction proposed in prior works for the homogeneous FPGA cluster cannot be trivially extended to the heterogeneous cluster. To decouple these conflicting requirements, we provide a two-level system abstraction. Specifically, the high-level abstraction is FPGA-agnostic and provides a simple and homogeneous view of the FPGA resources to simplify the runtime management and maximize the flexibility. On the contrary, the low-level abstraction is FPGA-specific and exposes sufficient low-level hardware details to the compilation framework to ensure the mapping quality and maximize the efficiency. This generic two-level system abstraction can also be specialized to the homogeneous FPGA cluster and/or be extended to leverage application-specific information to further improve the efficiency. We also develop a compilation framework and a modular runtime system with a heuristic-based runtime management policy to support this two-level system abstraction. By enabling a dynamic FPGA sharing at the sub-FPGA granularity, the proposed virtualization solution can deploy 1.62x more applications using the same amount of FPGA resources and reduce the compilation time by 22.6% (perform as many compilation tasks in parallel as possible) with an acceptable virtualization overhead, i.e., Finally, we use Liquid Silicon as a case study to show that the proposed virtualization solution can be extended to other spatial reconfigurable architectures. Liquid Silicon is a homogeneous reconfigurable architecture enabled by the non-volatile memory technology (i.e., RRAM). It extends the configuration capability of existing FPGAs from computation to the whole spectrum ranging from computation to data storage. It allows users to better customize hardware by flexibly partitioning hardware resources between computation and memory based on the actual usage. Instead of naively applying the proposed virtualization solution onto Liquid Silicon, we co-optimize the system abstraction and Liquid Silicon architecture to improve the performance.Publication Statistical Learning For System Identification, Estimation, And Control(2022-01-01) Tsiamis, AnastasiosDespite the recent widespread success of machine learning, we still do not fully understand its fundamental limitations. Going forward, it is crucial to better understand learning complexity, especially in critical decision making applications, where a wrong decision can lead to catastrophic consequences. In this thesis, we focus on the statistical complexity of learning unknown linear dynamical systems, with focus on the tasks of system identification, prediction, and control. We are interested in sample complexity, i.e. the minimum number of samples required to achieve satisfactory learning performance. Our goal is to provide finite-sample learning guarantees, explicitly highlighting how the learning objective depends on the number of samples. A fundamental question we are trying to answer is how system theoretic properties of the underlying process can affect sample complexity. Using recent advances in statistical learning, high-dimensional statistics, and control theoretic tools, we provide finite-sample guarantees in the following settings. i) System Identification. We provide the first finite-sample guarantees for identifying a stochastic partially-observed system; this problem is also known as the stochastic system identification problem. ii) Prediction. We provide the first end-to-end guarantees for learning the Kalman Filter, i.e. for learning to predict, in an offline learning architecture. We also provide the first logarithmic regret guarantees for the problem of learning the Kalman Filter in an online learning architecture, where the data are revealed sequentially. iii) Difficulty of System Identification and Control. Focusing on fully-observed systems, we investigate when learning linear systems is statistically easy or hard, in the finite sample regime. Statistically easy to learn linear system classes have sample complexity that is polynomial with the system dimension. Statistically hard to learn linear system classes have worst-case sample complexity that is at least exponential with the system dimension. We show that there actually exist classes of linear systems, which are hard to learn. Such classes include indirectly excited systems with large degree of indirect excitation. Similar conclusions hold for both the problem of system identification and the problem of learning to control.Publication Learning Environmental Models With Multi-Robot Teams Using A Dynamical Systems Approach(2022-01-01) Salam, TahiyaRobots monitoring complex, spatiotemporal phenomena require rich, meaningful representations of the environment. This thesis presents methods for representing the environment as a dynamical system with machine learning techniques. Specifically, we formulate machine learning methods that lend to data-driven modeling of the phenomena. The data-driven modeling explicitly leverages theoretical foundations of dynamical systems theory. Dynamical systems theory offers mathematical and physically interpretable intuitions about the environmental representation. The contributions presented include distributed algorithms, online adaptation, uncertainty quantification, and feature extraction to allow for the actualization of these techniques on-board robots. The environmental representations guide robot behavior in developing strategies such as optimal sensing and energy-efficient navigation. The methods and procedures provided in this thesis were verified across complex, spatiotemporal environments and on experimental robots.Publication Graph Convolutions For Teams Of Robots(2021-01-01) Khan, ArbaazIn many applications in robotics, there exist teams of robots operating in dynamic environments requiring the design of complex communication and control schemes. The problem is made easier if one assumes the presence of an oracle that has instantaneous access to states of all entities in the environment and can communicate simultaneously without any loss. However, such an assumption is unrealistic especially when there exist a large number of robots. More specifically, we are interested in decentralized control policies for teams of robots using only local communication and sensory information to achieve high level team objectives. We first make the case for using distributed reinforcement learning to learn local behaviours by optimizing for a sparse team wide reward as opposed to existing model based methods. A central caveat of learning policies using model free reinforcement learning is the lack of scalability. To achieve large scale scalable results, we introduce a novel paradigm where the policies are parametrized by graph convolutions. Additionally, we also develop new methodologies to train these policies and derive technical insights into their behaviors. Building upon these, we design perception action loops for teams of robots that rely only on noisy visual sensors, a learned history state and local information from nearby robots to achieve complex team wide-objectives. We demonstrate the effectiveness of our methods on several large scale multi-robot tasks.Publication Machine Learning On Large-Scale Graphs(2022-01-01) Ruiz, LuanaGraph neural networks (GNNs) are successful at learning representations from most types of network data but suffer from limitations in the case of large graphs. Challenges arise in the very design of the learning architecture, as most GNNs are parametrized by some matrix representation of the graph (e.g., the adjacency matrix) which can be hard to acquire when the network is large. Moreover, in many GNN architectures graph operations are defined through convolutional operations in the spectral domain. In this case, another obstacle is the obtention of the graph spectrum, which requires a costly matrix eigendecomposition. Yet, large graphs can often be identified as being similar to each other in the sense that they share structural properties. We can thus expect that processing data supported on such graphs should yield similar results, which would mitigate the challenge of large size since we could then design GNNs for small graphs and transfer them to larger ones. In this thesis, I formalize this intuition and show that this graph transferability is possible when the graphs belong to the same "family", where each family is identified by a different graphon. A graphon is a function W(x,y) that describes a class of stochastic graphs with similar shape. One can think of the arguments (x,y) as the labels of a pair of nodes and of the graphon value W(x,y) as the probability of an edge between x and y. This yields a notion of a graph sampled from a graphon or, equivalently, a notion of a limit as the number of nodes in the sampled graph grows. Graphs sampled from a graphon almost surely share properties in the limit such as homomorphism densities which, in practice, implies that graphons identify families of networks that are similar in the sense that the density of certain "motifs" is preserved. This motivates the study of information processing on graphons as a way to enable information processing on large graphs. The central component of a signal processing theory is a notion of shift that induces a class of linear filters with a spectral representation characterized by a Fourier transform (FT). In this thesis, we show that graphons induce a linear operator which can be used to define a shift and therefore graphon filters and the graphon FT. Building on the convergence properties of sequences of graphs and associated graph signals, it is then possible to show that for these sequences the graph FT converges to the graphon FT and that graph filter outputs converge to the outputs of the graphon filter with same coefficients. These theorems imply that for graphs that belong to certain families, graph Fourier analysis and graph filter design have well defined limits. In turn, these facts enable graph information processing on graphs with large number of nodes, since information processing pipelines designed for limit graphons can be applied to finite graphs. We further define graphon neural networks (WNNs) by composing graphon filters banks with pointwise nonlinearities. WNNs are idealized limits which do not exist in practice, but they are a useful tool to understand the fundamental properties of GNNs. In particular, the sampling and convergence results derived for graphon filters canbe readily extended to WNNs, allowing to show that GNNs converge to WNNs as graphs converge to graphons. If two GNNs can be made arbitrarily close to the same WNN, then by a simple triangle inequality argument they can also be made arbitrarily close to one other. This result formalizes our intuition that GNNs are transferable between similar graphs. A GNN can be trained on a moderate-scale graph and executed on a large-scale graph with a transferability error dominated by the inverse of the size of the smallest graph. Interestingly, this error increases with the variability of the spectral response of the convolutional filters, revealing a trade-off between transferability and spectral discriminability that is inherited from graph filters. In practice, this trade-off is less present in GNNs due to nonlinearities, which are able to scatter spectral components of the data to different parts of the eigenvalue spectrum where they can be discriminated. This explains why GNNs are more transferable than graph filters.Publication Integrated Diamond-Based Devices For Quantum Sensing And Communication(2022-01-01) Huang, Tzu-YungThe nitrogen-vacancy (NV) center in diamond has been integral to the advancement of quantum technologies. It has enabled key demonstrations in quantum networks and distributed quantum computing, as well as impressive proof-of-concept devices in nanoscale, sub-cellular imaging and sensing. The NV center | a point defect in diamond with an optically addressable electron spin | boasts several advantages such as long coherence times and quantum control in ambient conditions. At the same time, the NV center still faces significant engineering challenges in the realization of scalable quantum devices. The high-refractive index of diamond in the visible spectrum results in inefficient photon collection, and the NV center's sensitivity to charge and spin fluctuations at surfaces and interfaces necessitates deeply embedded NV centers for coherence-limited applications, which further worsens readout efficiency. The challenging optical interface has necessitated the use of high numerical-aperture objectives and free space optics, which are non-ideal components for scalable devices. On the other hand, benchtop frequency synthesizers and sequence generators typically used in laboratory experiments inhibit the packaging of NV centers for deployment outside of laboratory environments. This thesis focuses on the realization of integrated quantum devices using the NV center and targets miniaturization of the host crystal, the optical interface, and quantum control sequence generators. First, it discusses efforts and progress in realizing chemically active quantum sensors based on NV centers embedded in nanodiamonds. Following, it presents the miniaturization of collection optics through the coupling of an immersion metalens fabricated on bulk diamond to a single NV center in the crystal. Finally, it lays out performance considerations for integrated sequence generators to preserve the NV center's coherence properties, and demonstrates quantum control at room temperature of a NV center using a CMOS control signal generator. To conclude, this thesis discusses future potential for further integrating and miniaturizing diamond-based quantum devices.Publication Engineering Nanocrystal Devices Through Cation Exchange And Surface Modifications(2020-01-01) Wang, HanColloidal nanocrystals (NCs) are promising materials for electronics and optoelectronics, due to their (i) solution processability; (ii) size-dependent electronic band structures; and (iii) large surface-to-volume ratio. Exploring these features has led to great advances in our fundamental understanding of carrier mobility and lifetime in NC assemblies, which governs the performance of their devices. To continue to push forward NC technologies requires further development and optimization of materials and device design and fabrication process, and efforts are needed to replace the toxic elements found in the most studied NC compositions for many device applications. In this thesis, we develop solution-processable, all NC field-effect transistors (FETs) with metallic (Ag, In), semiconducting (CdSe), and insulating (Al2O3) NCs. We modify NC surfaces and the materials, yielding a chemically compatible, structurally stable, and physically cooperative device integration on flexible polymer substrates. All NC FETs show similar device performances with FETs built with NC semiconducting channels and conventionally vacuum-grown materials in other device layers. To address toxicity concerns, we develop a post-deposition cation exchange process to achieve high-performance CuInSe2 NC FETs and circuits. We start with high-quality CdSe NCs, and exchange Cd2+ with Cu+ and In3+. We integrate the high-quality CuInSe2 NC thin film into FETs and proof-of-concept circuits, by applying surface modifications to introduce fusion among NCs and to add In-doping, which show comparable device and circuit performances with the starting CdSe NC thin-film devices. We continue to optimize the NC device design to achieve shortwave infrared (SWIR) photodetection. We build photo FETs with CdSe NCs by surface modification to introduce In doping. We achieve strong quantum confinement in high mobility CdSe NC thin films, which we hypothesize makes intra sub-band transitions possible and allows us to achieve high photoresponsivity broad spectrum photodetection, including at SWIR wavelengths. To improve the compatibility with complementary metal oxide semiconductor (CMOS) technology, we build Cd- and Pb-free photodiodes by integrating InAs NCs with crystalline Si. We introduce a two-step surface modification process to enhance interparticle coupling, passivate trap states, and introduce doping. We show promising external quantum efficiencies (EQEs) at SWIR wavelengths.
- «
- 1 (current)
- 2
- 3
- »