<![CDATA[Ph.D. Proposal Oral Exam - Chun-wei Ho]]>28475Title: A Knowledge-driven approach to Audio Segmentation, Target Music Extraction and Cinematic Audio Source Separation
Committee:
Dr. Chin-Hui Lee, Advisor
Dr. Moore, Chair
Dr. Anderson
]]> Daniela Staiculescu117763659022026-04-16 18:58:2217763659732026-04-16 18:59:3300eventThe objective of the proposed research is to develop a unified knowledge-guided framework for audio processing that integrates audio segmentation, target music extraction (TME), and cinematic audio source separation (CASS). Real-world audio recordings exhibit highly diverse and overlapping acoustic conditions, where precise temporal boundaries are ambiguous, target sources are not clearly separated, and many sound events are rare or poorly represented in training data. Conventional data-driven approaches struggle under such conditions due to limited supervision and the lack of structured guidance. This research investigates how auxiliary knowledge, such as speech transcripts, musical scores, instrument and artist information, and language-universal speech attributes, can be systematically incorporated to constrain and guide model learning, particularly in resource-limited and real-world scenarios.]]>2026-04-30T09:00:00-04:002026-04-30T11:00:00-04:002026-04-30T11:00:00-04:002026-04-30 13:00:002026-04-30 15:00:002026-04-30 15:00:002026-04-30T09:00:00-04:002026-04-30T11:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-30 09:00:002026-04-30 11:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Ph.D. Proposal Oral Exam - Jungjin Park]]>28475Title: Adaptation technique for energy-efficient digital baseband processor
Committee:
Dr. Sathe, Advisor
Dr. Romberg, Chair
Dr. Raychowdhury
]]> Daniela Staiculescu117763650692026-04-16 18:44:2917763651712026-04-16 18:46:1100eventThe objective of the proposed research is to develop adaptation techniques for energy-efficient digital baseband processors in GPS acquisition and adaptive beamforming applications. Three techniques are presented: (1) an adaptive compressive sampling GPS correlator that reduces acquisition energy by 4.7x–51x in 65 nm CMOS, (2) a beamspace MVDR algorithm that replaces O(N³) matrix inversion with O(N log N) FFT-based operations, and (3) a gradient-descent-based beam tracker with adaptive step size and update-period control, validated in 16 nm and 28 nm CMOS implementations. Together, runtime adaptation to signal conditions delivers significant energy savings without sacrificing the digital baseband processing performance.]]>2026-05-14T13:00:00-04:002026-05-14T15:00:00-04:002026-05-14T15:00:00-04:002026-05-14 17:00:002026-05-14 19:00:002026-05-14 19:00:002026-05-14T13:00:00-04:002026-05-14T15:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-05-14 01:00:002026-05-14 03:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Zoom link]]><![CDATA[Ph.D. Proposal Oral Exam - Prithwijit Chowdhury]]>28475Title: Necessity and Sufficiency: A Framework for Evaluating Information Flow in Human-AI Interaction
Committee:
Dr. AlRegib, Advisor
Dr. Muthukumar, Chair
Dr. Heck
]]> Daniela Staiculescu117763646602026-04-16 18:37:4017763647342026-04-16 18:38:5400eventThe objective of the proposed research is to develop a unified framework for evaluating and improving the bidirectional flow of information between humans and AI systems by treating both explanations and prompts through the lens of the Information Bottleneck principle. The core idea is that human AI interaction is a communication problem in two directions. In the model to human direction, explanations must compress the model’s internal reasoning into a form that preserves what a person needs to judge, trust, or correct the output. In the human to model direction, prompts must compress the user’s intent into a form that is sufficient to guide model behavior without adding redundant or conflicting signal. This proposal argues that necessity and sufficiency are the right principles for analyzing both directions under one theory. A representation is necessary if removing it changes the outcome, and sufficient if providing it can produce the desired outcome. Building on this view, the research first develops causal, interventional measures of necessity and sufficiency and uses them to study existing explanation methods, showing in preliminary tabular experiments that highly ranked features are not always both necessary and sufficient. It then extends the framework to image models, where it investigates whether standard evaluation metrics truly measure information that survives the model’s internal compression. Finally, it applies the same framework to interactive segmentation, where prompts are analyzed as units of information transfer, classified by their contribution, and selected actively through predictive disagreement using a Bayesian extension of SAM. Together, the proposed work aims to establish a single information theoretic account of how AI systems should communicate with humans and how humans should communicate with AI systems for effective collaborative decision making.]]>2026-04-30T13:30:00-04:002026-04-30T15:30:00-04:002026-04-30T15:30:00-04:002026-04-30 17:30:002026-04-30 19:30:002026-04-30 19:30:002026-04-30T13:30:00-04:002026-04-30T15:30:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-30 01:30:002026-04-30 03:30:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Microsoft Teams Link ]]><![CDATA[Ph.D. Proposal Oral Exam - Faaiq Waqar]]>28475Title: Bottom-Up System-Technology Co-design of Emerging Amorphous Oxide Semiconductor Memories for High-Performance and Reconfigurable Computing Machines
Committee:
Dr. Yu, Advisor
Dr. Hao, Chair
Dr. Mukhopadhyay
]]> Daniela Staiculescu117763631592026-04-16 18:12:3917763641792026-04-16 18:29:3900eventThe objective of the proposed research is to address the scaling bottlenecks and static power inefficiencies of SRAM by investigating monolithic 3D (M3D) integration of back-end-of-line (BEOL)-compatible amorphous oxide semiconductor (AOS) memories, including capacitive topologies such as eDRAMs and gain cells and AOS-based SRAM, to address conventional SRAM’s scaling and power limitations while preserving its most desirable attributes. Specifically, this work seeks to determine how key AOS device characteristics, including operating voltage, current drive, parasitic capacitance, retention, and variation tolerance, propagate through bitcell and array design to shape the energy, latency, density, and scalability of monolithically integrated 3D memories. To this end, the research develops compact-model and logical-effort-based evaluation frameworks that connect device calibration and SPICE-level circuit analysis to cycle-accurate architectural benchmarking, enabling rigorous study of AOS memories in both conventional memory hierarchies and reconfigurable systems.]]>2026-04-27T13:00:00-04:002026-04-27T15:00:00-04:002026-04-27T15:00:00-04:002026-04-27 17:00:002026-04-27 19:00:002026-04-27 19:00:002026-04-27T13:00:00-04:002026-04-27T15:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-27 01:00:002026-04-27 03:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Microsoft Teams Link ]]><![CDATA[Ph.D. Proposal Oral Exam - Yavuz Yarici]]>28475Title: Understanding and Utilizing Domain-Dependent Information Structure in Multimodal Systems
Committee:
Dr. AlRegib, Advisor
Dr. Vela, Chair
Dr. Calhoun
]]> Daniela Staiculescu117763639622026-04-16 18:26:0217763640312026-04-16 18:27:1100eventThe objective of the proposed research is to characterize the domain-dependent structure of task-relevant information in multimodal systems and to develop methods that maintain robust performance under distribution shift. One of the main promises of multimodal learning is to combine data from multiple sensing modalities via the use of their complementary information for a targeted downstream task. However, the benefits of multimodal fusion are typically demonstrated when training and evaluation data share the same distribution. Domain generalization methods address distribution shift for single-modality systems, but do not account for the interactions that arise when multiple encoders are trained jointly through a shared objective. While information-theoretic frameworks provide a principled decomposition of task-relevant information into redundant, unique, and synergistic components, they assume stationary distributions. How this information structure changes under domain shift, and how such changes interact with joint optimization, remain unexplored. This proposal addresses this gap by analyzing how each component of multimodal information varies across domains and by developing training procedures that preserve robust representations under domain shift. Preliminary research establishes that multimodal fusion improves accuracy under matched conditions, but this advantage diminishes under distribution shift. The proposed research characterizes how domain shift affects multimodal information structure and develops methods to maintain robust performance.]]>2026-04-22T15:00:00-04:002026-04-22T17:00:00-04:002026-04-22T17:00:00-04:002026-04-22 19:00:002026-04-22 21:00:002026-04-22 21:00:002026-04-22T15:00:00-04:002026-04-22T17:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-22 03:00:002026-04-22 05:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Microsoft Teams Link ]]><![CDATA[Ph.D. Proposal Oral Exam - Shubham Sahasrabudhe]]>28475Title: Dual-mode Low Phase Noise Oscillator Design for Precision Self-temperature Sensing in a Distributed Lam?? Resonator
Committee:
Dr. Ayazi, Advisor
Dr. Shaolan Li, Chair
Dr. Ansari
]]> Daniela Staiculescu117761710762026-04-14 12:51:1617761710892026-04-14 12:51:2900eventThe objective of my proposed research is to develop a versatile, low phase noise dual-mode oscillator architecture that is applied to a distributed Lamé resonator (DLR) for high-performance timing and temperature sensing. Dual-mode MEMS resonators enable a unique method for precision self-temperature sensing for temperature-compensated and oven-controlled MEMS oscillators. A high-performance, low phase noise dual-mode oscillator design is crucial to achieving high-resolution and high-bandwidth resonator self-temperature measurement. I further demonstrate the applications of the self-temperature sensing DLR oscillator in temperature-compensated and oven-controlled oscillator applications.]]>2026-04-21T15:00:00-04:002026-04-21T17:00:00-04:002026-04-21T17:00:00-04:002026-04-21 19:00:002026-04-21 21:00:002026-04-21 21:00:002026-04-21T15:00:00-04:002026-04-21T17:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-21 03:00:002026-04-21 05:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Ph.D. Proposal Oral Exam - Dong Suk Kang]]>28475Title: Time-Oriented Data Conversion Techniques for High-Speed and Energy-Efficient Analog-to-Digital Converters
Committee:
Dr. Shaolan Li, Advisor
Dr. Gu, Chair
Dr. Sathe
]]> Daniela Staiculescu117761684802026-04-14 12:08:0017761685732026-04-14 12:09:3300eventThe object of the proposed research is to develop a high-speed, energy-efficient analog-to-digital converter architecture based on hybrid voltage-time-domain signal processing. This work explores converting voltage signals into time-domain representations to achieve energy efficiency and scalability in advanced CMOS process.]]>2026-04-27T14:00:00-04:002026-04-27T16:00:00-04:002026-04-27T16:00:00-04:002026-04-27 18:00:002026-04-27 20:00:002026-04-27 20:00:002026-04-27T14:00:00-04:002026-04-27T16:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-27 02:00:002026-04-27 04:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Microsoft Teams Meeting link]]><![CDATA[Ph.D. Proposal Oral Exam - Jiacheng Wang]]>28475Title: A Hybrid Atomic Mems Platform for Miniature Atomic Beam Devices and Applications
Committee:
Dr. Ayazi, Advisor
Dr. Ghalichechian, Chair
Dr. Raman
]]> Daniela Staiculescu117761683762026-04-14 12:06:1617761683912026-04-14 12:06:3100eventThe objective of the proposed research is the design, fabrication, characterization, and integration of a hybrid MEMS–atomic beam platform for fully integrated chip-scale atomic beam devices. The work addresses a key limitation in current atomic clock and quantum sensor technologies, namely the reliance on manual assembly and bulk optical components, which constrain scalability, increase cost, and limit manufacturing yield. The proposed approach introduces a wafer-level fabricated, piezoelectrically actuated micromirror capable of achieving a precisely controlled 45° tilt with high optical reflectivity. This micromirror is integrated with a microfabricated silicon collimator to realize a planar atomic beam architecture. By eliminating the need for buffer gas, the system enables nearly transit-time-limited linewidths, improving long-term stability while mitigating drift mechanisms associated with buffer gas interactions.]]>2026-04-21T10:00:00-04:002026-04-21T12:00:00-04:002026-04-21T12:00:00-04:002026-04-21 14:00:002026-04-21 16:00:002026-04-21 16:00:002026-04-21T10:00:00-04:002026-04-21T12:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-21 10:00:002026-04-21 12:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Ph.D. Dissertation Defense - Venkatesh Avula]]>36804Title: Multiphysics Design and Modeling for Advanced Packaging and Heterogeneous Integration of Power Delivery
Committee:
Dr. Peterson, ECE, Advisor
Dr. Naeemi, ECE
Dr. Losego, MSE
Dr. Cohen, ECE
Dr. Sitaraman, ME
]]> jjones779117757613622026-04-09 19:02:4217761643882026-04-14 10:59:4800eventThe objective of this research is to solve the multiphysics design challenges in integrated power delivery for advanced electronic packaging and heterogeneous integration architectures. Embedded inductors, essential for efficient power delivery, must maintain high performance at elevated temperatures, minimizing Joule heating and inductance loss. Designing such components requires robust multiphysics modeling, yet conventional methods are computationally intensive. This thesis proposes solutions for both efficient embedded inductor design and computationally efficient multi-physics modeling. A dual-core inductor design is introduced, featuring two coils separated by magnetic cores and an air gap. This configuration enhances saturation current capacity and overall performance, mitigating the common low-saturation current limitation in embedded inductors. For multi-physics modeling, an augmented computational method is developed, using a frequency-domain approach with Fourier series–based Toeplitz matrices. This method provides a unified framework for transient electrical simulations and steady-state thermo-mechanical analysis, representing time-varying system elements in the frequency domain to improve stability and efficiency. To further accelerate computations, a parallelized version of the augmented method is implemented, leveraging modern CPU and GPU parallel architectures. ]]>2026-04-21T15:00:00-04:002026-04-21T17:00:00-04:002026-04-21T17:00:00-04:002026-04-21 19:00:002026-04-21 21:00:002026-04-21 21:00:002026-04-21T15:00:00-04:002026-04-21T17:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-21 03:00:002026-04-21 05:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Zoom Link ]]><![CDATA[Ph.D. Dissertation Defense - Janak Sharda]]>28475Title: Electrical-thermal Co-design and System-level Analysis of Advanced Packaging Solutions for Hardware Accelerators and Sensor Technologies
Committee:
Dr. Shimeng Yu, ECE, Chair, Advisor
Dr. Muhannad Bakir, ECE
Dr. Saibal Mukhopadhyay, ECE
Dr. Tushar Krishna, ECE
Dr. Yingyang (Celine) Lin, CoC
]]> Daniela Staiculescu117752550632026-04-03 22:24:2317752551412026-04-03 22:25:4100eventDeep learning solutions are becoming increasingly important for a lot of tasks. Tasks such as object detection from image rely on additional sensors and tasks such as natural language processing performed using large language models (LLMs) require efficient access to large memory capacity. As a result, they require close interconnection between different components for efficient data and signal movement. Heterogeneous integration has been proposed recently to address such concerns. This dissertation proposes designs for various problems ranging from sensing-based edge technologies to large server-based hardware accelerators for LLMs. The work proposes designs and analysis different packaging schemes depending upon the requirements. It proposes heterogeneous integration of different components including emerging memories to be able to design efficient hardware for such applications. Further, this study addresses thermal challenges associated with such heterogeneous integrated designs. Evaluations are performed to compare various designs and compared against the state-of-the-art. Further, the study proposes path for designing and evaluating future memories and interconnect technologies, providing design insights into various bottlenecks and an evaluation methodology for performing an electrical-thermal system-level co-design. Building upon the CIS and LLM accelerator designs, this dissertation develops a systematic system-technology co-optimization (STCO) methodology for evaluating advanced packaging configurations. The methodology is applied to compare 2.5D interposer, 3D HBM-on-logic, and selective layer transfer (SLT) integration for LLM accelerators across inference and training workloads. Detailed thermal and power delivery analysis is performed for 3D stacked HBM and compute accelerators, examining the trade-offs between memory-on-logic and logic-on-memory configurations. A hybrid design combining elements of both approaches is proposed and shown to achieve favorable thermal profiles while maintaining high memory bandwidth. Finally, a generalized software-hardware co-design framework, NS-AP (NeuroSim for Advanced Packaging), is developed that extends the analysis to arbitrary packaging configurations and workloads. The framework operates at three levels: component, chip, and package, and incorporates physical design, thermal simulation, and power delivery analysis.]]>2026-04-23T13:00:00-04:002026-04-23T15:00:00-04:002026-04-23T15:00:00-04:002026-04-23 17:00:002026-04-23 19:00:002026-04-23 19:00:002026-04-23T13:00:00-04:002026-04-23T15:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-23 01:00:002026-04-23 03:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Microsoft Teams Meeting link]]><![CDATA[Ph.D. Dissertation Defense - Shambavi Ganesh]]>28475Title: Multimodal Predictive Frameworks for Cardiac Computed Tomography Gating
Committee:
Dr. Pamela Bhatti, ECE, Chair, Advisor
Dr. Srini Tridandapani, UAB, Co-Advisor
Dr. Brooks Lindsey, BME
Dr. Elliot Moore, ECE
Dr. Shamali Dharmadhikari, Emory
Dr. Xiangyang Tang, Emory
]]> Daniela Staiculescu117752543722026-04-03 22:12:5217752544482026-04-03 22:14:0800eventCardiovascular disease (CVD) is the leading cause of mortality worldwide, underscoring the need for accurate and accessible cardiac imaging. Although coronary catheter angiography is the gold standard for CVD evaluation, it is invasive and relatively expensive. In contrast, cardiac computed tomography angiography (CTA) is a more practical alternative because it is less invasive, more affordable, and more accessible. Despite these advantages, its image quality is often degraded by motion artifacts from the continuously beating heart. To reduce these artifacts, CTA commonly relies on electrocardiogram (ECG) gating to estimate cardiac quiescence, defined as the period of minimal cardiac motion. However, ECG reflects electrical activity rather than mechanical motion, making it an indirect surrogate for identifying true quiescent periods. Mechanical signals such as seismocardiography (SCG) and ultrasound (US) therefore offer promising alternatives for improving quiescence prediction. Although ultrasound has previously shown promise for identifying cardiac quiescence, its use during CT acquisition was limited by streak artifacts from conventional transducers. This dissertation develops multimodal gating frameworks to improve cardiac quiescence prediction for CTA. First, an SCG-ECG prediction framework was extended with a regression-based artificial neural network and evaluated using patient data, with diagnostic image quality assessed through a blinded reader study. Second, a deep learning-based US-ECG gating framework incorporating convolutional neural networks was developed and evaluated using multimodal datasets acquired from a dynamic heart motion phantom, enabling multimodal data acquisition under varied motion conditions. Third, the effect of US-ECG-based quiescence prediction on diagnostic image quality was assessed through a phantom-based CT reader study comparing reconstructed CTA images with conventional gating strategies. These contributions highlight the potential of multimodal gating to improve cardiac CTA and advance translation toward more reliable clinical imaging workflows.]]>2026-04-21T12:00:00-04:002026-04-21T14:00:00-04:002026-04-21T14:00:00-04:002026-04-21 16:00:002026-04-21 18:00:002026-04-21 18:00:002026-04-21T12:00:00-04:002026-04-21T14:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-21 12:00:002026-04-21 02:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Zoom link]]><![CDATA[Ph.D. Proposal Oral Exam - Hua Chen]]>28475Title: Efficient Power Delivery Using High-voltage DC-DC Converters for Data-Centers
Committee:
Dr. Raychowdhury, Advisor
Dr. Li, Chair
Dr. Sathe
]]> Daniela Staiculescu117752510482026-04-03 21:17:2817752511362026-04-03 21:18:5600eventThe objective of this research is to develop high-voltage hybrid DC-DC converters with compact designs that achieve high power density for data center applications. The growing demand for non-isolated, high-voltage step-down converters that are energy-efficient and capable of supplying heavy load currents motivates this work. We focus on converter shrinking techniques to improve power density. First, we introduce a merged cross-connected double step-down (DSD) converter that enables high-frequency power delivery. Next, we present a dual-path hybrid Dickson (DPHD) converter, which reduces inductor RMS current and minimizes the need for external bootstrap capacitors, paving the way for higher current density. Finally, we compare these designs with existing solutions and propose two promising directions for future investigation: a Fibonacci hybrid converter and a 6S quadruple step-down (QSD) hybrid converter.]]>2026-04-24T10:00:00-04:002026-04-24T12:00:00-04:002026-04-24T12:00:00-04:002026-04-24 14:00:002026-04-24 16:00:002026-04-24 16:00:002026-04-24T10:00:00-04:002026-04-24T12:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-24 10:00:002026-04-24 12:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]><![CDATA[Microsoft Teams Meeting link]]><![CDATA[Ph.D. Proposal Oral Exam - Zahra Mohseni]]>28475Title: Design and Noise Analysis of a D-band Passive Imaging Receiver
Committee:
Dr. Gu, Advisor
Dr. Ghalichechian, Co-Advisor
Dr. Cressler, Chair
Dr. Shaolan Li
]]> Daniela Staiculescu117743910452026-03-24 22:24:0517743911332026-03-24 22:25:3300eventThe objective of the proposed research is to develop a passive imaging receiver with flicker noise and multiplicative noise suppression capability to improve the resolution of the system. In the first part of the project, a Dicke switch-based system which has 50% time efficiency is designed and implemented. A comprehensive noise analysis is done to investigate the parameters that degrade low frequency noise suppression capability of the system. This investigation reveals that clock duty cycle of the chopper degrades the noise suppression performance. Therefore, a structure is proposed to overcome the sensitivity of the multiplicative noise suppression to duty cycle offset and also it has 100% time efficiency which improves the resolution of the system.]]>2026-04-20T15:00:00-04:002026-04-20T17:00:00-04:002026-04-20T17:00:00-04:002026-04-20 19:00:002026-04-20 21:00:002026-04-20 21:00:002026-04-20T15:00:00-04:002026-04-20T17:00:00-04:00America/New_YorkAmerica/New_Yorkdatetime2026-04-20 03:00:002026-04-20 05:00:00America/New_YorkAmerica/New_Yorkdatetime<![CDATA[]]>