Electrical and Computer Engineering, 1932-2025

High-density microelectrode arrays make it possible to obtain detailed recordings of neural population activity. However, they are fundamentally limited by the number of available readout channels in a given system. Switch-matrices attempt to address this issue by flexibly routing many electrodes to a smaller set of amplifiers, but existing implementations are limited to static configurations during each recording. Consequently, continuous temporal coverage is restricted to a subset of electrodes. In this project, I am investigating whether dynamic time-multiplexed switching can be adapted to a PCB-based recording system. I have designed a dynamically reconfigurable switch matrix that routes 480 electrodes, arranged in a 24 x 20 array, to 120 headstage inputs by rapidly cycling though 4 electrode groups of 120 each. The feasibility of routing is validated using a min-cost/max-flow algorithm to generate the best switch configurations. A microcontroller-based control system sends configuration bitstreams to MAX14803 chips distributed throughout the system and applies atomic switching updates while maintaining settling delays to mitigate switching artifacts. I will analyze the constraints surrounding the switching speed, signal integrity, and noise as well as the trade-offs of working with discrete analog switches on a PCB. This paper aims to show that switch matrices based on PCBs can provide sufficient flexibility and timing performance for electrophysiological exploration, creating a cost-effective platform for testing time-multiplexed neural recording architectures.

Air handwriting enables text input without physical contact, but existing approaches rely mainly on computer vision and IMUs, which perform position-aware trajectory reconstruction. My senior thesis design project compares alternative sensing paradigms for air handwriting recognition using non-positional, wearable sensing modalities, specifically EMGs and piezoresistive bend and stretch sensors to infer handwriting patterns. This project presents a three-stage machine learning pipeline developed to interpret noisy time-series signals, consisting of spatial modeling for classifying directional primitives, sequential modeling for individual letter recognition, and continuous sequence decoding for real-time multi-letter inference. The intention is for the device to be as minimally invasive as possible and avoid restriction of finger movement, while proving the feasibility of bend and stretch sensors or EMGs in creating successful dynamic handwriting detection at finger scale. The project evaluates the efficacy of various model architectures used in conjunction with one another to track the user’s air handwriting. Experimental results demonstrated that despite the limitations of these low-cost nonpositional sensors, appropriate modeling and design can achieve robust, meaningful results. These findings highlight the capacity of different algorithmic approaches in compensating for limited sensing fidelity, and the separability of patterns in small-scale finger and wrist scale movements. Lastly, the project discusses future directions in which the system can be designed to be more robust to more sources of variability.

The purpose of the human intellect is to identify and name the phenomena around us and manipulate them constructively. Many such phenomena have been completely classified, some with simple and methodical methods and others with algorithmic approaches. Other phenomena remain out of grasp of our ability to internalize and map their workings. Entanglement among arbitrary quantum states, particularly bound entanglement, fits into this latter type of phenomena with the possibility or impossibility of deterministically detecting this entanglement remaining an open question. Machine learning approaches have offered an unparalleled opportunity to explore previously unfeasible problems as well as expand human pattern recognition to the perfection of machine algorithms. In this work, I analyze the efficacy of 28 different machine learning methods in detecting entanglement in quantum states and compare this efficacy to that of the most advanced algorithmic approaches for this problem. In pursuit of this goal, I have produced a semi-comprehensive dataset generation and feature extraction package to benchmark entanglement detection methods. Additionally, I have implemented limited versions of two proposed algorithmic approaches in a manner feasible for modern computing capabilities. Incredibly, my results show that near-deterministic was possible for all tested splits of the data with over 90% accuracy being attained being attained for training sets as small as 50 quantum states. This high accuracy was seen even for runs with training and testing splits between bound entanglement families, implying generalizability to bound entanglement learning. Moreover, simple baseline ML methods proved to be more effective than modern high-capacity methods when feature engineering was implemented, implying that feature extraction contains most of the information relevant for classification of the dataset.

Semiconductor lasers can be used as physical reservoir computers (RCs) that can perform ML tasks faster and more efficiently than traditional neural networks. However, most setups rely on fiber-optic delay lines or external cavities which hinder scalability. This thesis uses numerical simulation to investigate laser-based RC setups which do not rely on time delays to remember past inputs, instead exploiting lasers’ relaxation oscillations. An analysis of a a reservoir computer composed of uncoupled lasers demonstrated strong linear memory and state-of-the-art performance on the NARMA-10 benchmark. To increase nonlinear processing capacity, unidirectionally coupled laser chain reservoirs were evaluated, showing improved accuracy but requiring an impractically large number of lasers. A novel binary tree laser reservoir topology was developed, where pairs of dissimilar lasers couple into subsequent layers to exponentially increase nonlinear interactions. This binary tree architecture achieved state-of-the-art performance on the NARMA-10 benchmark (NMSE = 0.100) utilizing a total of only 63 lasers, requiring just 32 lasers in the first layer to be directly driven by the modulation current.

Safety certifications are crucial for complex robotic systems operating in diverse environments alongside humans. Current approaches often tightly couple safety mechanisms with specific tasks or environments, creating challenges for certifying systems deployed in real-world settings like warehouses or homes. This paper proposes a pipeline to design an Neural Control Barrier Function (NCBF) that serves as a rigid safety filter, decoupling safety guarantees from task execution to ensure liveness, scalability, and robustness. We simultaneously train an optimized data-driven, actor-critic safety policy to steer the robot away from the unsafe set of states. This thesis relies on existing semantic counterfactual reasoning approaches involving vision-language models (VLMs) to "imagine" unsafe scenarios from real safe trajectories, addressing the data sparsity problem in training learned CBFs. It expands upon these by incorporating world models that can rollout trajectories in latent space and attempts to test the pipeline on two datasets taken on different robots, a manipulator and a navigator. The navigation pipeline was implemented fully, whereas the manipulation pipeline was shown to be possible but not fully implemented due to compute constraints. In testing the developed Control Barrier Function on the navigation dataset, it was found that although the safe and unsafe boundary could be learned to a high accuracy, the optimizer function trained as a safety policy often reached a trade-off and frequently misfired on safe episodes. The results suggested this pipeline may be useful but requires further development on the preprocessing and training for the final Control Barrier Function and optimizer.

The Deterministic Quantum Computation with One Clean Qubit (DQC1) model provides a framework for evaluating the computational power of highly mixed quantum states. Motivated by the constraints of physical quantum system implementations, DQC1 relies on initializing only a single control qubit in a pure state while the remaining register is maximally mixed. This work provides a comprehensive characterization of the DQC1 complexity class and surveys its algorithmic capabilities.

We first explore the model’s foundational mechanism of unitary trace estimation an rigorously prove that supplementing the system with a logarithmic number of pure ancilla qubits (k-clean-qubits) does not alter its fundamental complexity class. Furthermore, we determine the limits of this model by demonstrating its inability to obliviously simulate universal Bounded Error Quantum Polynomial-Time (BQP) circuits, establishing a definitive difference between DQC1 and BQP.

Finally, we examine applications that leverage this architecture: the #P-hard approximation of Jones polynomials for trace closures of braids, the demonstration of quantum advantage driven by quantum discord rather than entanglement, the noise-resilient estimation of kernels for supervised quantum machine learning, and the evaluation of constrained Quadratically Signed Weight Enumerators (QSWEs). Together, these analyses highlight DQC1 as a bridge between classical and quantum computing, capable of efficiently solving specific, classically intractable problems despite limitations on quantum purity.

Reservoir computing is an emerging paradigm in neuromorphic computing, particularly well-suited for processing temporal data such as speech recognition, time-series forecasting, and real-time signal analysis. The use of photonic devices offers significant advantages over traditional electronic implementations, such as higher speed, broader bandwidth, and improved energy efficiency.

The core focus of this Senior Thesis is the design and fabrication of strongly coupled diode lasers integrated on a single semiconductor chip. An emphasis is placed on the fabrication procedures of such lasers intersecting at shallow angles, forming a photonic reservoir. By embedding these components within a monolithic chip, the reliance on external optics and complex alignment procedures is eliminated. This streamlined integration not only simplifies the overall architecture, but also reduces the size, cost, and power consumption of the system, paving the way for scalable and practical photonic reservoir computing platforms.

Textiles are increasingly recognized in architecture and robotics for their adaptability to diverse shapes, low environmental impact, and lightweight character. While automated knitting has enabled the production of three dimensional fabric shapes, the low internal strength and geometric limits of knitted fabrics restrict their structural potential. Crochet, by contrast, offers higher tensile strength, greater three dimensional extensibility, and the ability to generate arbitrary topological surfaces through its flexible stitch geometry. This project introduces a new framework for robotic crochet which translates the handcraft from a hook to a loom. Through a manual study, we established the ability to create fabric that is topologically equivalent to crochet on a pegboard loom. We demonstrated that on a loom it is possible to create any crochet stitch that does not require front or back loop insertion points, and that loom crochet permits the realization of crochet patterns with stitches that cross several rows. To characterize the created fabric relative to crochet, we then performed a tensile test on swatches created on different loom sizes. For looms up to twice the natural gauge of crocheted fabric, tensile strength was equivalent to controls (p = .048, TOST; 90% CI [→0.108, 0.100]; equivalence bounds = ±15%), indicating the utility of loom created crochet pieces for high strength applications. We then created a robotic prototype to automate loom crochet production using a 3D printed end effector for a UR3 robotic arm. The robot was able to successfully create chains, slip stitches, and decreases, with a 74% stitch level success rate for slip stitches. This work establishes the viability of loom crochet for automated production of complex, high strength textiles.

MC (in-memory computing) is a ground-breaking paradigm for machine learning acceleration. While several architectures have been proposed, most IMC architectures revolve around weight-stationary SRAM crossbar arrays. While these arrays can provide throughput and efficiency gains, they are often bottlenecked by activation- and weight- loading costs. By analyzing how activations travel through a hierarchical IMC on-chip network, we hope to develop a model for power and energy usage attributed to activation loading. This paper establishes several parameters that are important to determining how activations move through IMC networks. Based on these parameters, it derives a high-level model for power and energy usage for different mappings of an ML (transformer) model to CIMUs (compute-in-memory units). The different parallelism parameters (Mo, Mi, and D), along with model size, strongly affect the activation-loading overhead. Furthermore, this paper looks to low-swing signaling as a potential avenue to reduce power, but also uses multiple drivers and receivers to reduce delay. The two receivers are the StrongARM latch (clocked) and an unclocked low-swing receiver.

Although de novo long-read genome assembly enables highly contiguous reconstruction of complex genomes, it remains computationally expensive due to the scale of modern sequencing datasets and the computational demands of key assembly stages. Flye is a widely used de novo long-read assembler that resolves repeats using a repeat graph representation followed by iterative polishing. Profiling across four genome datasets shows that Flye’s polishing stage accounts for an average of 37% of total assembly time, driven by repeated dynamic programming-based alignment and single-base edit evaluation across large numbers of bubbles and supporting branches. This work analyzes Flye’s polishing algorithm to identify performance bottlenecks and proposes GPUFlye, a CUDA-accelerated polisher that preserves Flye’s scoring and consensus mechanisms while exploiting fine-grained parallelism and high memory bandwidth on modern GPUs. Evaluated across four genome datasets, GPUFlye achieves polishing speedups between 1.22–5.95× and reduces total assembly time by up to 1.39×, with the NVIDIA L4 reducing full cloud assembly cost by up to 20% compared to a 16-thread CPU baseline.

Holographic displays are limited by experimental imperfections found in current spatial light modulators (SLMs). Some examples include phase quantization, optical aberrations, non-uniform illumination, and finite spectral bandwidth, all of which degrade reconstructed holographic image quality compared to ideal simulations. Compensating for these imperfections requires an accurate, differentiable forward model of the full optical pipeline from the SLM phase pattern to the captured holographic image. The first portion of this project develops a camera-in-the-loop (CITL) calibration and optimization framework, using real camera feedback to iteratively correct SLM phase patterns for hardware nonidealities such as optical distortions and misalignment. The second portion of the project extends this approach to a simulation-only setting by training a learned differentiable proxy model (PLMPropagate) against a physically grounded simulator (PLMSimulator), incorporating Zernike aberrations, non-uniform illumination, detector noise, and Gaussian spectral bandwidth. Phase patterns are then re-optimized through the learned proxy to account for these imperfections without requiring physical hardware access. Evaluated on ten test images, the proxy-optimized phase achieves an average improvement of +2.58 dB PSNR and +0.109 SSIM over the naive baseline, demonstrating that learned forward models can substantially improve holographic image quality even in the absence of a physical camera-in-the-loop.

The decarbonization of India’s chemical sector is crucial for India to achieve its net-zero targets. The production of ethylene, a key building block of the chemical industry, currently relies on fossil fuel combustion to supply high-temperature heat, contributing approximately half of India’s chemical sector emissions and prompting significant interest in exploring decarbonization strategies. This thesis builds a bottom-up techno-economic model evaluating four decarbonized approaches to supply industrial heat for ethylene production: a) electrification through grid connection, b) electrification through captive renewable energy, c) green hydrogen combustion, and d) blue hydrogen combustion. These pathways are evaluated based on the calculated levelized cost of carbon abatement (LCCA), feasibility considerations, and technoeconomic assumptions specific to the Indian context, with a sensitivity analysis conducted to identify the key drivers affecting pathway viability. The two electrification pathways with captive renewable energy and grid connection exhibited comparable base-case LCCAs of $136/tCO2 and $146/tCO2, respectively. The blue hydrogen pathway demonstrated a slightly higher LCCA of $213/tCO2, while the green hydrogen pathway exhibited the highest LCCA of $534/tCO2. The electrification through captive renewable energy pathway appears to be best suited for India given its lowest cost of carbon abatement and lower dependence on highly uncertain risk factors. The robust, immediate value proposition of the electrification pathway found in this study differs from the academic consensus, offering surprising implications for the decarbonization of India’s ethylene sector.

The concept of using CMOS at cryogenic temperatures is fairly novel, so its behavior in comparison to CMOS at room temperature is not fully understood; this is mainly due to how difficult it is to perform high volumes of tests on these devices at cryogenic temperatures like is done for higher temperatures. That being said, the newly available and currently undergoing research on circuitry for read-out and control of quantum devices using closely integrated cryo-CMOS technology demonstrates massive improvements to the pitfalls associated with using environmentally separate instrumentation. This thesis aims to contribute to the growing use of cryo-CMOS in such applications by designing circuits for effective data collection on cryo-CMOS devices. Specifically, this thesis performs schematic capture and layout of circuits for on-chip thermometry and DC characterization of CMOS devices at cryogenic temperatures.

As the global temperature rises, the ocean continues to absorb most of the heat, which in turn causes ocean patterns to change [1]. The island of Puerto Rico has long felt the impacts of climate change, while a large part of the population has been affected by an increase in tropical cyclone activity and rising sea levels [2]. As a result of the rising temperatures, ocean acidification has increased by around 25 percent in the past three centuries and might increase to 40-50 percent by 2100 [3]. Recent research has employed marine robotic vehicles as tools for monitoring ocean resources and ecosystems. However, most existing vehicles are costly to both develop and manufacture, which restricts their use and can further decelerate efforts of research. This makes it challenging to use them to navigate and collect important samples in areas with obstacles, such as the coral reef. To tackle this, we propose to develop a pressure-sensing system that can enable the robotic fish, BlueGuppy swarm, to collect hydrodynamic flow data in real-time for optimized swimming efficiency, and additionally to improve environmental monitoring technologies.

Superconducting qubits are among the leading platforms for quantum computing, where material choice critically impacts qubit coherence and scalability. Tantalum (Ta) has emerged as a promising candidate due to its stable, self-limiting oxide Ta_2O_5, which hosts fewer dielectric two-level systems (TLS). Ta transmons on high-resist silicon exhibit lifetimes T_1 up to 1.68 ms and Q factors up to 2.5*10^7 \cite{bland2025}. Rhenium (Re) also forms an exceptionally thin and inert oxide layer, and has demonstrated resonator quality factors exceeding 2 * 10^6 \cite{crisa2025}. By alloying Ta with Re, we aim to combine Ta's low-loss coherence performance with Re's enhanced surface and structural properties. Ta--Re alloys may offer tailored superconducting behavior, improved oxidation resistance, and reduced TLS density---potentially surpassing pure Ta in coherence and fabrication resilience. This research explores the stability and oxidation behavior of thin-film Ta-Re alloys deposited on sapphire substrate.

The Gambia’s electricity sector remains heavily dependent on imported energy, exposing the country to price volatility, foreign exchange pressures, and persistent supply insecurity. This thesis investigates the least-cost pathway to reduce this dependence using a capacity expansion model applied across three planning horizons, 2025, 2030, and 2035, drawing on a real demand profile constructed from 329 daily operation reports from The Gambia’s National Water and Electricity Company (NAWEC) covering the 2025 calendar year. Modelling results show that full energy independence requires between 584 MW and 1,156 MW of solar PV and between 130 MW and 222 MW of battery storage, at an additional annualised cost ranging from $7.09 million per year at current demand to $15.14 million per year by 2035. The least-cost pathway is unambiguously a clean energy pathway, new diesel generation was found to be economically uncompetitive across all scenarios. Sensitivity analysis identifies an import tariff crossover of approximately $131/MWh, above which full energy independence becomes economically optimal without policy support. The thesis also examines the institutional, financial, and technical barriers to implementation, and provides evidence-based recommendations to support a realistic and affordable energy transition for national planners and policymakers.

Robotic drone platforms require robust, precise control solutions that are able to preserve safety while achieving high task performance in order to operate in high uncertainty, constrained environments. Safety filters provide a scalable method to ensure the safety of the system, but due to the dynamical complexity of drone platforms, synthesizing safety monitors and intervention systems can be challenging. This project presents Safe-Aviary, a modular, gym-like simulation environment that extends the gym-pybullet-drones simulator. Safe-Aviary is optimized for training safety fallback policies using the Iterative Soft Adversarial Actor Critic for Safety (ISAACS) deep reinforcement learning algorithm. We demonstrate that the resulting safety filters are effective in reducing safety violations, and assess the impact that filtering during conventional policy training can have in Safe Reinforcement Learning contexts.

Millimeter-Wave Communications (mmWave) enable highly directional signal transmissions, but remain inherently vulnerable to passive interception due to the broadcast nature of wireless propagation. This motivates the need for physical-layer security which is not addressed by conventional encryption operating at higher layers of the communication stack. Antenna Subset Modulation has been proposed as a transmission obfuscation technique that generates randomness in sidelobe emissions while preserving coherent modulation toward an intended receiver.

This work evaluates ASM under both single and coordinated passive adversarial models, with a focus on realistic deployment conditions. In particular, we investigate the impact of multipath propagation on both system performance and adversarial decoding capability. We show that previously proposed coordinated attacks relying on idealized channel assumptions degrade significantly in the presence of multipath, showing ASM's security dependence on complete channel knowledge. However, when the attack is extended to a complete sensing matrix capturing both line-of-sight and multipath components of the environment, and Moore-Penrose pseudoinverse reconstruction is applied, coordinated adversaries can recover transmitted signals. These results demonstrate that ASM does not provide unconditional security, but offers strong practical resilience in realistic environments where precise channel knowledge is difficult to obtain.

With the upcoming launch of a new generation of space stations designed to operate unmanned over long durations, the need for continuous monitoring of this infrastructure is becoming increasingly important. Robotic swarms offer a promising solution for maintaining the health and integrity of these structures. We begin by introducing the Tumblenauts, a bacteria-inspired robot swarm designed to perform intra-vehicular inspection in microgravity using stochastic “run-and-tumble” locomotion. However, evaluating the performance of these microgravity robots on Earth remains a key challenge. Existing experimental methods such as air-bearing tables and parabolic flights are costly and spatially confined which contributes to the difficulty of developing and validating swarm algorithms with these methods. Therefore, we introduce the Hovernauts, a low-cost robotic swarm testbed that replicates the stochastic “run and tumble” motion of the Tumblenauts in two-dimensions. The Hovernauts can glide on various flat surfaces, such as a table or a floor, without requiring any external infrastructure. They are extremely simple to manufacture and are specifically designed for swarm-level experimentation. Using this platform, we explore several active random walks inspired by bacterial run and tumble motion. Our analysis reveals that by controlling the frequency of thruster activation, we can modulate the robot’s exploration dynamics. We further investigate how dynamically adjusting the run probability enables the robots to climb gradients. Overall, the Hovernauts provide a low-cost and accessible platform that enables the development and validation of novel stochastic swarm algorithms for future space inspection applications.

Rooftop solar photovoltaic (PV) systems present a valuable opportunity for households to meet their own electricity needs while contributing to broader clean energy goals. Although some countries have promoted rooftop solar through subsidies, net metering, and other policy incentives, its global potential remains underutilized, particularly in developing countries. Understanding the factors that drive residential rooftop solar adoption is important for designing effective policy and accelerating deployment. This study develops a cross-country framework to evaluate the drivers of rooftop solar adoption across different national contexts. A state-level analysis of India is conducted and compared with existing literature on the United States and Australia to examine how adoption dynamics differ between developing and developed countries. The results show that economic conditions and energy-use patterns have the strongest influence on rooftop solar adoption in India, but they do not fully explain adoption on their own. Education, awareness, policy accessibility, financing mechanisms, and subsidy implementation also play an important role in enabling adoption. These findings suggest that, as a developing country and emerging rooftop solar market, India must place greater emphasis on policy delivery, financial accessibility, and public awareness than countries in later stages of solar market development.