UC Santa Barbara

This thesis focuses on the design, fabrication, and characterization of gold microelectrode array (MEA) devices for applications in electrochemical aptamer-based (EAB) sensing. EAB sensors leverage target-induced conformational changes in surface-bound aptamers to transduce molecular recognition phenomena into measurable electrochemical signals, enabling real-time, reversible, and reagentless detection of target molecules in complex biological environment. Such sensors are highly necessary for pharmacokinetic measurements, where continuous monitoring of target concentrations is required.Motivated by the need to extend EAB sensing beyond single-point measurements toward spatially resolved target monitoring, this work develops planar, multi-site gold microelectrode arrays fabricated using standard microfabrication techniques. The MEAs were designed to support dense packing of sensing sites while maintaining well-defined electrode geometry and compatibility with aptamer functionalization. A custom PCB interface was developed to enable reliable electrical connections to external instrumentation.Electrochemical characterization of the fabricated devices was performed using cyclic voltammetry and square-wave voltammetry to assess electrode quality, surface area, and stability before and after aptamer immobilization. Sensor performance was evaluated using kinetic differential measurements (KDM) to enhance signal robustness and suppress common-mode drift. The effects of electrode geometry, square wave frequency, and biological media on signal stability were analysed.This thesis demonstrates that MEA based EAB sensors exhibit comparable electrochemical behavior and sensing performance to conventional gold wire electrode, while additionally enabling multi-site, spatially resolved measurements.Importantly, this work also shows that planar gold microelectrode arrays can be functionalized with aptamers without electrochemical surface roughening while still achieving kinetic differential measurement (KDM) responses comparable to electrochemically roughened gold wire control electrodes. This simplifies sensor fabrication while preserving robust EAB signaling, supporting the use of planar, microfabricated platforms for scalable EAB sensor implementations. Overall, this thesis lays the foundation for reproducible and scalable MEA-based platform for EAB sensing and for the future implementation of densely packed, minimally invasive sensor arrays capable of spatially resolved pharmacokinetic measurements real-time continuous in vivo monitoring.

The exponential increase of global energy demand has positioned proton exchange membrane fuel cells (PEMFC) as an attractive alternative to conventional fossil fuel powered electricity generators. Their compact designs, high efficiency, mitigated emissions and lower harmful impact on the environment are key factors that justify this enthusiasm. PEMFC are part of a growing family of electrochemical devices that can efficiently convert chemical energy to electrical energy that will lead to the advancement of integrating hydrogen in power systems. The key issues preventing rapid commercialization and integration of fuel cells into the global energy infrastructure is their high cost and low durability. In this thesis, a pulsed electrochemical deposition method has been employed to develop improved cathode electrocatalyst in PEM fuel cells. The electrochemical deposition of Ni and Co has been studied and further incorporated with Pt to develop effective fuel cell cathodes. We found that using an optimized pulse potential deposition can produce multiple shapes and sizes of Ni and Co nanoparticles depending on the electrolyte material. We determined critical parameters to employ to our electrodeposition pulse that are crucial to produce controlled size distribution for both Pt and Co and Ni in large densities. The cathode electrode was redesigned to improve catalytic activity by layering Pt nanoparticles on top of a earth abundant metal directly onto a carbon gas diffusion layer, so that they may be directly used in a fuel cell device.

The III-Nitrides remain a rich material family for study and new applications. Emerging ultra-wide bandgap (UWBG) semiconductors, such as high Al-content AlGaN, are attractive due to their large direct bandgaps, resulting high critical electric field, and large saturation velocity for electrons. While this materials system has been the subject of intense research and commercial success for several decades, there are many challenges still associated with making practical devices from UWBG III-nitrides, including lattice defects, low mobilities and conductivities, and difficulty forming contacts. In this work, I seek to improve the understanding of these materials when grown by ammonia molecular beam epitaxy (NH3-MBE), first by establishing suitable homoepitaxial AlN buffer layers, followed by characterizing and later avoiding plastic relaxation in graded alloys, and then by integrating these materials into polarization-doped field effect transistors for next-generation RF power amplifiers. Finally, I will present on the growth of an emerging ultra-wide bandgap nitride – aluminum boron nitride (AlBN).AlN films were grown by NH3-MBE with varying V/III ratios and substrate temperatures. The surface morphology was examined by atomic force microscopy and the growth rate was calculated from the Pendellösung fringe spacing obtained from high resolution x-ray diffraction ω-2θ scans. Three growth regimes based on different surface morphologies were identified, most importantly the N-rich step flow growth regime at high temperatures. The differences in surface morphology illustrate the effects of the growth parameters on adatom surface mobility. On and off-axis rocking curve widths of films grown at different V/III ratios were compared to those of the bare substrates, and no additional broadening was observed which indicates that no significant defect formation takes place in the MBE-grown films, which is confirmed by planview transmission electron microscopy. This work shows that NH3-MBE AlN can serve as an insulating buffer layer for high quality electronic devices enabled by precise control over growth regimes.There is a significant lattice mismatch between AlN and lower-Al alloy layers needed for transistor barriers, channels, and contacts. This work demonstrates fully coherent graded AlGaN heterostructures based on AlN templates which could be implemented into future devices. We show the importance of well-controlled growth conditions, AlGaN layer thicknesses, and composition steps through transmission electron microscopy, atomic force microscopy, and high-resolution x-ray diffraction measurements to guide the epitaxial design of future generations of AlGaN-based transistors. We also emphasize the importance of careful analysis of high-resolution x-ray diffraction data in the context of a heterogeneous microstructure, such as one exhibiting crosshatch, in which there are simultaneously heavily dislocated and coherent crystalline regions in a thin layer.To realize high performance, ultra-wide bandgap AlGaN transistors for high-frequency, high-power applications, it is imperative to minimize the access resistance to the conductive, high mobility channel to realize high frequency operation. The relatively low electron affinity of AlGaN presents serious challenges for forming ohmic contacts. In this work, we report on low resistance compositionally graded AlGaN contacts to Al0.75Ga0.25N in a PolFET. The resistances of Si-doped graded contacts are compared to standard alloyed vanadium-based contacts to identical graded PolFET channels. By optimizing the donor density in the graded contacts to counteract the negative volume polarization charge induced by the compositional grade, we achieve one of the lowest-reported specific contact resistances to high-Al content AlGaN – ρc = 7.2×10-7 Ω-cm2.Finally, we demonstrate growth of AlBN by NH3-MBE using a high temperature boron effusion cell. Despite challenging growth conditions, we are able to controllably incorporate boron up to alloy fractions of 6% and maintain the epitaxial wurtzite crystal structure. This represents an important step towards integrating emerging nitride materials with the existing III-nitride ecosystem.

The diffusion of ideas, innovations and behaviors is a complex sociological phenomena for which various computational models have been proposed by economists, algorithmists and ML researchers. However, it remains insufficiently understood due to lack of propagation data on networks and the complex interplay between dynamic network structure and node-level features. Moreover, real-world networks exhibit skewed degree distributions and propagation size distributions, thereby necessitating learning models that account for imbalanced distributions and data scarcity. The goal of this thesis is two-pronged: (1) How do coupled representation and classifier learning models cope with imbalanced data distributions, particularly with minority classes, and how can their weaknesses in limited data settings be mitigated?, (2) How can dynamics of propagation on real-world networks, themselves admitting a natural power law degree distribution, be understood using the machinery of gradient-based optimization and high-dimensional feature representations? The two goals are naturally coupled; distributional analysis is a powerful tool for network (propagation) analysis, and network propagation modeling suffers from a lack of ground truth data, which necessitates a combination of data-driven and synthetic models rooted in sociological insights. Therefore, the major theme of this thesis is models of propagation in networks, and the minor is the limited data learning setting.

Wildfire fuel heterogeneity is a fundamental yet persistently underrepresented source of uncertainty in fire consumption and emissions estimates. The physical and chemical characteristics of vegetation that burns in a fire vary at fine spatial scales that standard fuel representations cannot capture. Operational fire consumption and emissions modeling relies on categorical approaches that assign static, uniform values to fuel properties across what are often highly variable landscapes with no mechanism for quantifying how uncertainty in those inputs propagates to emissions. Remote sensing data offers an avenue toward continuous, spatially explicit fuel characterization that can improve estimates of fire emissions. Specifically, imaging spectroscopy and lidar data provide complimentary observations such as foliar chemical and biophysical traits, fractional cover, and three-dimensional structure that can be used to effectively characterize and quantify spatial variability in vegetation. In this dissertation, I develop and evaluate a remote sensing methodology for continuously mapping pre-fire fuel characteristics with quantified uncertainty using imaging spectroscopy and lidar and demonstrate how individual fuel characteristics and their uncertainty propagate through fire emission models to consumption and emissions estimates.In Chapter 1, I use airborne imaging spectroscopy data from the Advanced Visual InfraRed Imaging Spectrometer (AVIRIS) and lidar alongside field measurements to develop models to map spatially variable fuel characteristics across a wildfire and prescribed burns in Washington and Utah. Where imaging spectroscopy coverage was unavailable, I used multispectral satellite imagery and topographic data to fill spatial gaps, enabling complete coverage across burned areas. Continuous pre-fire fuel characteristic models and maps were produced with accompanying uncertainty maps and high confidence spatial masks. The models for most fuel characteristics achieved normalized root mean square errors at or below 20%. In Chapter 2, I evaluate the portability of this remote sensing methodology across a broader suite of wildland and prescribed fires in the western U.S. and application to data from additional sensors like the National Ecological Observatory Network Airborne Observation Platform (AOP). The methodology resulted in consistent model performance across the study fires when applied with site-specific model development, supporting broad applicability of this methodology across the western U.S. Analysis of the structure of model predictors provides insight into environmental drivers of fuel variability. In Chapter 3, I develop a spatially explicit approach to propagate fuel characteristic uncertainty to estimates of consumption and emissions using an existing fire emissions model. The spatially variable fuel maps from Chapters 1 and 2 were used as inputs and consistently produced lower total consumption and emissions estimates than standard categorical inputs, ranging from 42% to 80% and 36.6% to 78.12% of standard estimates across the study fires. Sensitivity analysis identified canopy structural characteristics as the primary drivers of differences in emissions magnitude while ground fuels, particularly duff, contributed disproportionately to emissions uncertainty. This research advances both the remote sensing of vegetation and the scientific understanding of how fire fuel heterogeneity and its uncertainty contribute to fire emissions, establishing for the first time that uncertainty from fuels can be statistically propagated to emissions estimates.

Functional verification through simulation is one of the most time-consuming phases of the digital design cycle. Graphics processing units offer a natural avenue for acceleration, since nodes within a single topological level of a synthesized netlist can be evaluated independently and in parallel. Prior GPU-based approaches operate at the granularity of individual logic gates, inheriting the full complexity of the original netlist. This work presents an alternative: we first subject the input design to logical synthesis using Yosys, which restructures the logic into a smaller number of multi-input lookup tables (LUTs) of up to eight inputs, and then simulate the simplified netlist on the GPU. Across a suite of 13 benchmark designs ranging from a 2-gate half adder to a 14,000-gate FIR filter, synthesis reduces the number of logical components by 66–98% for non-trivial designs. When simulating for 10,000 cycles on an NVIDIA RTX A4000, LUT-level simulation achieves GPU kernel speedups of 3.9x–19.6x over gate-level simulation for combinational- heavy designs, with the largest benefits appearing at scale. We present in this thesis a fully automated end-to-end pipeline, from Verilog source through synthesis, parsing, and GPU simulation.

Vision Transformers (ViTs) achieve state-of-the-art performance but suffer from the O(N2 ) complexity of self-attention, making inference costly for high-resolution inputs. To address this bottleneck, token pruning has emerged as a critical technique to accelerate inference. Most existing methods rely on the [CLS] token to estimate patch importance. However, we argue that [CLS] token can be unreliable in early layers where semantic representations are still immature. As a result, pruning in the early layer often leads to inaccurate importance estimation and unnecessary information loss. In this work, we propose a training-free token importance metric, namely Col-Ln, which is derived from Rényi entropy that enables the identification of informative tokens from the first layer of the network, thereby enabling more reliable pruning in token reduction. Extensive experiments on ViTs and Large Vision-Language Models (LVLMs) demonstrate that our approach consistently outperforms state-of-the-art pruning methods across diverse benchmarks.

Decarbonizing the global energy and chemical infrastructure is a long-term challenge, owing in large part to the pervasive role of fossil fuels in the modern economy. Their vast availability and low cost have placed them in a central position to the operation of energy and chemical industries, making rapid and complete replacement difficult despite the urgency of reducing greenhouse gas emissions. As a result, near-term decarbonization will depend not only on the gradual replacement of fossil fuels, but also on the development of low-carbon methods to utilize them to help bridge the transition to a zero-carbon economy. This has motivated growing interest in finding efficient pathways for converting hydrocarbons into hydrogen and other value-added products without direct CO2 emission. Plasma-driven processes are particularly attractive because they can be driven electrically, making them compatible with renewable energy sources, and can generate reactive species, such as radicals and ions, that are not readily available under thermal conditions, thereby enabling additional reaction pathways.Despite this promise, the understanding of mechanisms and process design principles governing plasma-driven hydrocarbon processes remains insufficient. In particular, the relative importance of non-equilibrium plasma chemistry, thermal effects, and transport can vary widely with discharge conditions, pressure, and reaction environment, making product selectivity and energy efficiency difficult to predict. This lack of clarity continues to limit the design and optimization of plasma processes for practical applications. This thesis investigates the mechanisms and efficacy of non-thermal plasma for methane and light hydrocarbon conversion into hydrogen and value-added hydrocarbon products. Methane conversion was examined using a custom pulsed plasma source to systematically determine the influence of plasma pulse energy and excitation timescale (nanoseconds to millisecond) on discharge characteristics, conversion and product selectivity. Quantitative mass spectrometry and high-resolution optical emission spectroscopy were used to characterize product distributions, gas temperature and electron density. This work demonstrates that methane dissociation proceeds through both electron-driven and thermal pathways in short discharge pulses; however, the overall product distribution is largely governed by thermal dissociation. These studies are complemented by high-pressure batch reactor experiments that measure methane dissociation rates under the influence of pressure and hydrogen dilution. Results indicate that increasing reactant density improves process energy efficiency and methane dissociation can be facilitated by H-abstraction reactions. Building upon these gas-phase studies, the work is extended to direct liquid hydrocarbon conversion, where the electrical discharge is heavily influenced by the local hydrodynamic environment, introducing an additional layer of complexity. The product distribution, conversion and energy efficiency were strongly affected by plasma excitation frequency and fluid flow velocity. To better understand the physical basis of these effects, the breakdown mechanism of liquid hydrocarbon was further investigated using time-resolved shadow graph imaging, and the influence of flow profile on bubble dynamics and discharge stability was examined in a liquid venturi jet reactor. Overall, this work shows that non-thermal plasma offers a promising pathway for hydrocarbon reforming and identifies important operating parameters for future process optimization studies. Moreover, the demonstrated viability of direct liquid hydrocarbon conversion points towards the potential for direct reforming of liquefied natural gas (LNG). The development of such a process could make use of existing LNG infrastructure and facilities, making the transition towards sustainable energy sources more economically feasible.

Many problems in combinatorial optimization, probabilistic inference, and machine learning require exploring extremely large solution spaces and are difficult to solve efficiently using conventional deterministic computing architectures. Such problems often involve rugged energy landscapes and require stochastic exploration to avoid local minima. These challenges have motivated the development of probabilistic computing systems that exploit randomness and parallelism directly in hardware.This dissertation presents the design and implementation of hardware architectures for probabilistic computing based on interacting stochastic units modeled after statistical physics systems. First, a fully connected 200-spin Ising machine implemented in 65 nm CMOS technology is presented. The architecture employs a dense SRAM-based weight storage array and mixed-signal bitline accumulation circuits to perform the weighted summation required for Ising computation. A 256 × 200 SRAM array stores the interaction matrix with discrete weights {−1, 0, +1}, enabling efficient realization of fully connected optimization networks.Second, a probabilistic computing chip integrating 440 interacting probabilistic bits arranged in a Chimera topology is demonstrated. The system utilizes mixed-signal stochastic neuron circuits and asynchronous Gibbs sampling to realize compact and energy-efficient probabilistic computation. Hardware-aware learning techniques are used to compensate for circuit mismatch and variability, enabling robust stochastic behavior across the chip.Finally, a scalable hardware accelerator implementing a Potts-model solver is developed on FPGA platforms to explore large-scale combinatorial optimization problems. The architecture employs chromatic updates based on graph coloring to ensure conflict-free state transitions while maintaining hardware efficiency. The system supports programmable annealing schedules and implements invertible logic circuits that allow arithmetic operations to be solved in both forward and reverse directions. Using this framework, the solver is applied to integer factorization by mapping an array multiplier to a probabilistic network and clamping the product output. Simulation results demonstrate successful recovery of the factors of an 18-bit semiprime through stochastic annealing dynamics.

Li-ion batteries have established themselves as a key form of energy storage for portable electronics, electric vehicles (EVs), and grid-level storage. As Li-ion batteries become more broadly adopted, there remains a need to understand and address sources of battery degradation to ensure their optimal performance and safe operation, particularly in extreme environments. One source of degradation known to negatively impact Li-ion battery performance is temperature with both high and low temperatures inducing processes which reduce battery capacity, cycling rate performance and can even lead to thermal runaway. Importantly, however, understanding of the impact of temperature on Li-ion batteries has been largely limited to uniform high and low temperature fields. While a few preliminary studies have identified non-uniform aging mechanisms in response to temperature hot spots and gradients, such studies have been performed operando and thus the degradation mechanisms are electrochemical. Meanwhile, the effects of passive exposure to complex thermal fields remain largely unknown.In the following chapters, I will elaborate on the work I have performed to characterize the effects of passive exposure to spatially non-uniform and temporally evolving temperatures. Initial discussion will focus on characterization of the effect that a microscale temperature hotspot has on a Li-ion battery with in situ micro-Raman spectroscopy, in situ optical microscopy and thermal simulations. In this study, mild temperature heterogeneity was observed to cause Li to leach out from lithiated phases of graphite in the absence of an applied current. It was proposed that temperature heterogeneity causes Li+ to redistribute within the graphite electrode to maintain a uniform potential, which can lead to Li leaching. Subsequent discussion will shift to work assessing the viability of Li-ion batteries for applications in space by determining how controlled-rate thermal cycling to extreme low temperatures affects different battery components. In one study, the solidification and melting dynamics of carbonate-based Li-ion battery electrolytes were characterized during controlled-rate thermal cycling with micro-Raman spectroscopy and differential scanning calorimetry. In the bulk, supercooled solidification was observed during cooling even in the presence of nucleating agents whereas during melting defined solidus and liquidus transitions occur. Locally, regions with elevated [Li+] were observed below the bulk solidus point that were proposed to have formed from partial solidification of the electrolyte solvents expelling Li+ into the liquid regions. In another study, in situ transient grating spectroscopy was utilized to characterize changes in time-of-flight (ToF) within a porous composite graphite electrode during controlled-rate thermal cycling to cryogenic temperatures. From these experiments, reversible changes in ToF were identified which were attributed to reversible changes in the effective Young’s modulus of the graphite electrode. Collectively, these studies highlight the unique ways in which spatially non-uniform and temporally evolving temperatures impact Li-ion batteries and expand understanding beyond the effects of uniform temperature fields.