The subject of this dissertation focuses on designing biohybrid systems for efficient solar-fuel conversion. As the unlimited energy source for the ecosphere, solar energy could be captured via various devices to meet the rising demand on Earth’s energetic hungry. Among all efforts, solar-driven conversion of carbon dioxide to value-added multi-carbon products is an ambitious objective, which recycles the waste CO2 back into useful chemicals as the energy fuel, polymer, and pharmaceutical precursor. However, pure inorganic CO2 (photo)electrocatalysts suffer from low selectivity, high overpotential, and poor mass transfer. Therefore, we proposed integrating inorganic semiconducting nanomaterials that harvesting solar energy with biological organisms to microbial fix CO2 into target product. By taking advantage of solar energy harvesters, the metabolic pathway in the living organism is essentially powered by solar energy. This work herein explores the novel nanomaterial-biological photosynhtetic biohybrid system for efficient artificial photosynthesis, as well as cytoprotective strategy for further developments. Additionally, elucidation of charge transfer mechanisms at abiotic-biotic interface remains challenge, and this work delves into model system to probe the possible charge transfer process.

Chapter 1 introduces a brief history of new emerging biohybrid system. The diversity of possible biohybrid architectures has been thoroughly discussed, depicting the microorganisms coupled with photovoltaic devices, interfaced with photoelectrodes, and membrane-bound photocatalysts. The synergistic biohybrid design is capable to address the grand challenges for the efficient generation of solar fuels and chemicals. In parallel, the living anaerobe with less oxidative stress tolerance has been carefully treated with various cytoprotective strategies. To get insight of modern nano-biotechnology, the current progress on energy transfer mechanism and the challenges associated with this technology has been examined.

Chapter 2 presents one type of biohybrid system with gold nanoclusters as the photosensitizers. Gold nanocluster with discrete energy levels is regard as the light absorber, and translocation of these gold nanoclusters into nonphotosynthetic bacteria, Moorella thermoacetica, enabled photosynthesis of acetic acid from CO2. Due to the small-size, gold nanocluster enables strong direct interface between organism, which could circumvent the sluggish kinetics of electron transfer for existing biohybrid systems. Besides, biocompatible gold nanoclusters also serve as reactive oxygen species inhibitors to maintain high bacterium viability, realizing CO2 fixation continuously over several days.

Chapter 3 introduces a uniformly cyprotective strategy for anaerobes using metal-organic framework (MOF) monolayer. Biohybrid system has recently being taken as a promising approach to sustainable energy. Among these chemical-producing microbes are anaerobic bacteria, inherently susceptible to O2 and reactive oxygen species that are inevitably generated on anodes. Moorella thermoacetica as the strict anaerobic bacteria was dynamically wrapped by single MOF layer. The high definition of the MOF-bacteria interface involves direct bonding between phosphate units on the cell surface and zirconium clusters on MOF monolayer, enabling cell elongation and separation. The catalytic activitly of MOF enclosure toward decomposition of reactive oxygen species reduces the death of bacteria by fivefold.

Chapter 4 discusses the possible electron transfer process based on the work in chapter 2. The excited states behavior of gold nanoclusters attracts more and more attention due to their vast applications in artificial photosynthesis. The influence of charge-state on the cluster was studied by transient absorption spectroscopy, and interesting correlation between the core-gold and shell-gold relaxation was observed. The long-lived excited states components have been identified as ligand-to-metal transfer of gold nanoclusters with two different sizes. Their photocatalytic activity was evidenced from the electron transfer to methyl viologen. The elucidation of excited states and transfer process in model system could facilitate the understanding of complex biohybrid system.

Innovative micromechanical-based isotropic formulations have been proposed to predict the mechanical behavior of the asphalt concrete materials featuring high toughness, low viscosity nano-molecular resins and employed for comparisons between model predications and experimental measurements. A class of isotropic elastic-damage model based on a continuum thermodynamic framework is developed within an initial elastic strain energy based formulation. A 3-D analytical modeling methodology is established by treating the revolutionary asphalt concrete material as an asphalt mastic composite matrix containing fine aggregates, asphalt binder, polymerized DCPD and air voids with coarse aggregates inhomogeneities distributed in it . The effective elastic moduli of the asphalt mastic composites are homogenized by a newly proposed multilevel homogenization approach based on an ensemble-volume averaged pairwise interacting theory. The coarse aggregates are represented by spherical multilayer-coated particles in certain sizes. A governing damage evolution criterion is characterized through the net stress concept in conjunction with the hypothesis of strain equivalence.

An analytical formulation to predict the isotropic viscoelastic properties of the multiphase asphalt mastic composites is proposed within the micromechanical framework based on the concept of the correspondence principle along with the Laplace transform. The viscous behavior induced by the asphalt binder phase is characterized by a 4-parameter Burgers model, from where a multilevel homogenization approach similar to the elastic framework is employed to evaluate the effective viscoelastic mechanical properties of the heterogeneous asphalt mastic composites.

A class of isotropic elastoplastic-damage framework and isotropic thermo-elasto- viscoplastic-damage framework are developed following a similar methodology of the isotropic elastic-damage framework. The plastic and viscoplastic flows are introduced by means of an additive split of the stress tensor and the energy norm of strain tensor is redefined and employed as the equivalent strain. In particular, the Drucker-Prager yield condition is employed to characterize the plasticity behavior and a linear Perzyna type associative flow rule and viscoplastic hardening law are characterized in terms of the effective quantities in the effective stress space. A rate-dependent damage evolution criterion is introduced within the initial elastic strain energy based micromechanical framework to implement the damage behavior. An Arrhenius-type temperature term, uncoupled with Helmholtz free energy potential, is introduced to account for the effect of temperature. The computation of coupled elastoplastic/ elasto-viscoplastic damage behaviors are realized by a two-step operator splitting methodology.

Numerical simulations are conducted based on the proposed formulations. Particularly, the 3-D modeling is achieved by Python scripting in ABAQUS to make sure there is no overlapping among spherical particle inclusions; while the two-step operator splitting computational algorithms are accomplished through Fortran UMAT. Prediction results are compared with suitably designed experimental data, showing reasonably good agreement.

Internet on-demand video traffic has seen explosive growth in recent years. This brings a number of challenges in deploying video-on-demand services at large scale. The first challenge has to do with the enormity of the video catalog size. Traditionally, content providers replicate the entire video library in different locations, but this is wasteful and non-scalable as the video catalog size expands. The second challenge comes with the increase in video quality, which calls for efficient utilization of the scarce network bandwidth resources that continue to be economically expensive to expand. The emergence of different device modalities, including smart-phones, high-definition and 3D TV, tablets and etc poses another challenge in designing efficient systems and algorithms that cater to all device characteristics and user needs.

In this dissertation, we aim to present a general approach to designing, optimizing and architecting a video-on-demand system. Our approach considers the practical constraints of disk space, network link bandwidth, and node connection degree bound. In general, the joint optimization problem is combinatorially difficult. To tackle this, we first design a simple fractional storage architecture, which uses a class of regeneration codes that fluidifies the content, thereby enabling a distributed content placement and link rate allocation algorithm. We show that by storing only a fractional of the entire catalog everywhere, the system is able to fully support user demand at large scale. Second, we develop a Markov approximation technique to solve the problem of topology selection under node degree bound using a simple distributed algorithm. We prove that our algorithm achieves close-to-optimal solution, which we verify using extensive realworld trace simulations.

On the system side, we show extensive results to test the algorithm's scalability and robustness to changes in user dynamics and demand patterns. We show that our solution achieves high utilization of cache nodes storage and bandwidth resources, and automatically learns and caches the video according to the demand patterns. We observe that there exists a complex interplay between disk space, network bandwidth and node degree bound. We also present guidelines to important practical design choices including caching update intervals, demand prediction and provisioning. We also demonstrate the feasibility and efficiency of our design choice by building and experimenting a prototype system at Berkeley.

Deep learning (DL)’s dramatic rise in popularity across the domain sciences and industry has been accompanied by a correspondingly aggressive increase in the scale and computational complexity of DL workloads. In order to adopt state-of-the-art techniques, practitioners must wrestle with systems challenges of performance, cost, and scalability. In this dissertation, we identify the need for orchestration systems, which ease scaling burdens across the DL lifecycle through holistic, workload-aware optimizations. Drawing on both established techniques from data management research and new bespoke algorithms, we build practical orchestration engines to optimize three common DL workloads in the large-scale setting: model selection, data processing, and high-throughput serving. Our systems — which exploit workload- and context- specific opportunities — address a new layer of the large-scale DL optimization stack, more granular than current cluster managers and data systems, but still abstracted away low-level kernel & compiler optimizations. Empirical evaluations show that our orchestration techniques and systems can accelerate large-scale DL workloads by a large margin, even in complex, real-world settings. Our approach introduces a new technical lens, unifying systems, databases, and DL research, ultimately focused on democratizing and amplifying state-of-the-art DL innovations. Some of the systems proposed in this dissertation have already been adopted in production-scale industry pipelines, demonstrating the value of such orchestration optimizers for real-world DL.