Keywords: Photonic Sensors, Integrated Photonics
Abstract: Monolithic Active Pixel Sensors (MAPS) have transformed radiation detection and imaging by enabling the integration of sensing and increasingly sophisticated signal processing within a single silicon substrate. Over more than three decades, MAPS have evolved through successive generations, driven by advances in semiconductor technologies, detector architectures, and mixed-signal integrated circuits. The aim has been development of featherlight devices suitable for accurate position sensitive recording of charged relativistic particles. This keynote presents a generational review of MAPS developed for scientific applications beyond visible-light imaging. Beginning with the first diffusion-based monolithic sensors, the talk follows the evolution toward large-scale detectors deployed in high-energy physics experiments, modern depleted MAPS, and silicon-on-insulator (SOI) technologies. Attention is given to the interplay between semiconductor process technology, sensor electrostatics, low-noise analog front-ends, in-pixel processing, sparse readout architectures, timing performance, radiation tolerance, and spectroscopic capability. The lecture concludes with emerging directions for the next generation of MAPS, including electrostatically engineered sensors, three-dimensional integration, event-driven architectures, near-sensor intelligence, and highly integrated detector systems. Rather than viewing MAPS simply as imaging devices, the presentation highlights their evolution.

Keywords: bio-inspired sensors, Low SWaP/Form Factor, Bio-inspired Systems
Abstract: Electrochemical aptamer sensors, introduced in 2005, have yet to match the commercial success of continuous glucose monitors. We have overcome key barriers: achieving multiweek stability, integrating with proven needle-strip formats, and targeting clinically relevant analytes. We present advances enabling multi-month operation and introduce a multiplex approach for human performance monitoring with high sensitivity and specificity across multiple physiological states.

Keywords: Brain machine interfaces BMI. and Autonomy in Medicine, Integrated Human Effects Models
Abstract: Endogenous retroviruses are inherited viral elements embedded within the human genome and may contribute to neurodegenerative disease when they become transcriptionally active. This presentation explores how innate immune surveillance detects and responds to the activation of these latent viral elements and how similar principles may inspire the detection and mitigation of unknown threats embedded within software supply chains and later triggered on cyber-electronic warfare systems.

Keywords: Integrated Human Effects Models, Enabling Technology for Artificial Intelligence, AI for Cognitive Sensing and Data Fusion
Abstract: Abstract—G-induced Loss of Consciousness (GLOC) is a persistent hazard in high-performance aviation that can result in catastrophic loss of control. This study develops a machine learning model using behavioral and physiological signals to predict imminent GLOC from centrifuge trial data provided by the Air Force Research Laboratory (AFRL). After systematically eliminating sources of data leakage, including centrifuge telemetry, device accelerometers, motion-contaminated EEG signals, and an experimental condition confound, a leakage-free Random Forest classifier trained exclusively on Non-AFE rapid-onset runs achieves a Recall of 0.86, Precision of 0.72, and F1 score of 0.78 in leave-one-subject-out cross-validation. Key predictive signals include cerebral oxygenation (fNIRS), joystick tracking degradation, and pupillometry. Results demonstrate that cockpit-deployable sensors alone are sufficient to predict GLOC onset with clinically meaningful accuracy.

Keywords: Enabling Technology for Artificial Intelligence, Collaborative/Multi-Agent Systems, AI for Cognitive Sensing and Data Fusion
Abstract: Contemporary AI/ML tasks such as foundation modeling pose significant challenges for networks comprised of heterogeneous collections of assets. This talk will overview recent advances to the so-called intelligence orchestration problem, aiming to jointly optimize task performance, resource efficiency, and service latency in deciding where, when, and how learning is conducted.

Keywords: Autonomous Guidance, Navigation and Control, Simulation-Based Validation and Verification, Software and Networking
Abstract: Existing trajectory tracking approaches for autonomous vehicle control use either kinematic or dynamic models, where both models exhibit varying success when deployed independently. In this paper, a hybrid vehicle model is proposed that combines kinematic and dynamic models to accurately describe vehicle motion at both low and high speeds. A hybrid model predictive controller (hMPC) is proposed, consisting of two model predictive controller (MPC) architectures governed by hysteresis switching logic. A combined state and error, caused by packet drops, estimator-MPC (SEE-MPC) compensator is proposed. The SEE comprises an unknown input observer (UIO) and packet drops estimator. The SEE-MPC compensator performance is analyzed in the presence of malicious packet drops on the vehicle's communication channels. Simulation experiments are performed using MATLAB's RoadRunner software.

Keywords: Autonomous Guidance, Navigation and Control, Architecture, UAVs
Abstract: This paper presents a physics-structured multi-scale feature layer for compact helicopter in-ground-effect modeling. The method combines a blade-element thrust prior with filtered collective-history features to capture transient inflow effects. Closed-loop evaluation on unseen sine and non-sinusoidal inputs shows improved thrust prediction over raw neural and analytical baselines.

Keywords: Microelectronics, Power Electronics, Wider Bandgap Semiconductors
Abstract: New material, new switching devices, and new applications are three major driving forces behind power electronics innovations. This talk will start with the new applications and related opportunities and challenges for power electronics and then use examples to show how new material and new devices can enable practical yet innovative power electronics solutions.

Keywords: Extreme Environment Electronics, Wider Bandgap Semiconductors, Power Electronics
Abstract: We have developed a machine learning model which can predict the failure of devices and provide a predicted neutron fluence at which failure occurs. This model utilizes the static characteristics of a device such as its gate leakage, body diode characteristics, and threshold voltage as well as data collected during in-situ failure-in-time testing of the device such as temperature, irradiation angle, bias ratio to rated voltage, and neutron fluence. This combination of parameters allows us to describe many kinds of devices without knowledge of their exact specifications. The model uses gradient-boosted trees to build a two-stage model with first classification of samples into failed and not failed and then regression analysis to predict a failure fluence. This model has been validated using data collected during in-situ failure in time testing at the Los Alamos Neutron Science Center (LANSCE) ICE House-1 facility The neutron beamline at this facility produces atmospheric-spectrum neutrons at a fluence 1 million times higher than sea level. SiC Power MOSFETs were statically characterized and exposed to the beam until failure. The model has an accuracy of 90% when classifying failures. The models show strong accuracy for all device sets, with mispredictions being mostly false negatives; predicting that a device did not fail when it did fail. The regression model has prediction error no worse than 0.2 orders of magnitude.

Keywords: Semiconductor Devices, Microelectronics, Microfabrication
Abstract: NAND Flash is foundational to modern storage but remains vulnerable to radiation-induced reliability degradation in space environments. This talk will discuss how ferroelectric NAND Flash can provide semiconductor storage with improved radiation tolerance for aerospace and defense systems, orbital AI infrastructure, and deep-space missions, while enabling significant performance improvements for terrestrial applications.

Keywords: bio-inspired sensors
Abstract: Medical inhaler use was identified via analysis of exhaled breath with a novel terahertz / millimeter-submillimeter spectrometer. Breath samples of 10 volunteers were collected before and after use of a pulse-metered-dose inhaler. Identification of the exogenous inhaler propellant was unambiguous immediately post use and possible up to 30 minutes later.

Keywords: Brain machine interfaces BMI. and Autonomy in Medicine, Enabling Technology for Artificial Intelligence, Integrated Human Effects Models
Abstract: Nonalcoholic fatty liver disease (NAFLD) and its advanced form, nonalcoholic steatohepatitis (NASH), represent a growing global health burden with limited diagnostic tools. Accurate quantification of hepatic steatosis, one of the four cardinal histological features of NASH, is clinically critical yet labor-intensive when performed manually by pathologists. In this paper, we propose a weakly supervised deep learning pipeline for automatic segmentation of steatosis in whole slide images (WSIs). A weakly supervised segmentation mask generation approach based on classical image processing is introduced to reduce the need for pixel-level manual annotation substantially. The generated masks are validated by expert pathologists and used to train a Recurrent Residual U-Net (R2U-Net) model. A pixel-ratio thresholding scheme then maps patch-level segmentation results to four steatosis classes (Class 0–III). Experimental results on a publicly available NAFLD dataset demonstrate that the R2U-Net model achieves an Intersection-over-Union (IoU) of 0.763 and a Dice coefficient of 0.865 on the steatosis segmentation task. We further present an end-to-end pipeline that extends patch-level predictions to whole slide image (WSI) or region-of-interest (ROI) level steatosis grading. The proposed system shows strong agreement with expert pathologists and has the potential to reduce pathologists' workload in clinical practice substantially.

Keywords: Neuromorphic Computing, Trusted and Assured Hardware, Enabling Technology for Artificial Intelligence
Abstract: Deep neural networks (DNNs) deployed on resource-constrained neuromorphic hardware face three concurrent challenges: the need for model compression through pruning, vulnerability to adversarial input perturbations, and susceptibility to hardware-induced weight faults such as stuck-at-zero errors. While each of these factors has been studied in isolation, their combined effects on model reliability have received little attention. This paper presents an empirical investigation of how pruning, adversarial training, and hardware fault injection interact to affect the robustness of convolutional neural networks. Using a compact three-layer CNN trained on MNIST, we conduct three experiments: (1) comparing the fault tolerance of naturally and adversarially trained models under simultaneous hardware faults and adversarial attacks, (2) evaluating how pruning affects adversarial robustness, and (3) characterizing the joint accuracy surface across fault rates, adversarial perturbation magnitudes, and pruning levels. Our results show that adversarial training improves robustness against input perturbations but increases sensitivity to stuck-at-zero weight faults. Contrary to intuition, pruning did not significantly increase fault sensitivity, and varying the pruning level had little effect across fault rates and attack strengths. These results highlight the need to jointly consider adversarial robustness and hardware reliability.

Keywords: Communication, Digital Engineering, AI for Cognitive Sensing and Data Fusion
Abstract: Efficient and reliable transmission of channel state information (CSI) is crucial for maintaining optimal performance in massive multiple-input multiple-output (MIMO) wireless communication systems. However, the feedback overhead remains a significant challenge, particularly in frequency division duplex (FDD) systems with large antenna arrays. In this paper, we propose a novel CSI feedback compression scheme based on sparse residual component selection. The proposed method builds a training-set statistical mean baseline from the training dataset and identifies the K positions whose expected residual magnitude is largest, yielding a fixed importance mask shared between the user equipment (UE) and the base station (BS). Only the residual values at selected positions are encoded and transmitted using a CsiNet-style convolutional autoencoder, while the remaining positions are approximated by the statistical baseline at the BS without any additional feedback. As compression becomes more aggressive, the encoder and decoder operate on a smaller spatial grid at high compression, directly reducing computational cost. Experiments on the COST~2100 dataset show that the proposed method achieves competitive reconstruction quality compared with CsiNet, CsiNet+, and Sobel selective processing, while significantly reducing computational overhead at intermediate and high compression levels.

Keywords: Wider Bandgap Semiconductors, Microfabrication, Microelectronics
Abstract: AFRL's "Lab-to-Fab" strategy accelerates critical microelectronics for the warfighter. This talk details our dual-pronged approach: pioneering advanced RF front-ends like sub-100nm GaN, while simultaneously driving next-generation digital back-ends on advanced silicon nodes. We will discuss how this integrated model delivers revolutionary, de-risked sensor sub-systems.

Keywords: Extreme Environment Electronics, Microfabrication, Wider Bandgap Semiconductors
Abstract: Traditional Ni/Au Schottky contacts used in AlGaN/GaN High Electron Mobility Transistors (HEMTs) are prone to diffusion related degradation within the gate stack at high temperatures (HT). In this manuscript, we consider a Pt interlayer between Ni and Au as a candidate diffusion barrier to improve the high temperature stability of 750 nm planar-gated devices. These devices were characterized before and after HT exposure to evaluate changes in their DC performance compared to standard Ni/Au gated control samples. The Ni/Pt/Au gated devices maintained superior channel control, lower gate leakage, lower off state drain current, and exhibited different types of shifts in threshold voltage.

Keywords: Microelectronics, Packaging & Integration
Abstract: From advanced propulsion to autonomous systems, aerospace system designers have less room for thermal management to meet aggressive form factors and weight goals. Existing and emerging technologies each have their place to provide a range of performances - from computing to high power to high frequency. Let’s explore SWAP optimization and tradeoffs through the lens of high temperature electronics and how it ties in to, and is affected by, supply chain.

Keywords: Extreme Environment Electronics, Microelectronics, Semiconductor Devices
Abstract: Space is a constantly evolving theater especially in terms of radiation effects on electronics and materials. Technologies of the past, present, and future each have unique concerns. The landscape of space missions is fundamentally changing with technology. Industry, policies, and international standards will need to adapt to meet these evolving needs.

Keywords: Enabling Technology for Artificial Intelligence
Abstract: With the emergence of the Internet of Things (IoT), security of edge devices has moved to the forefront of system design. Recent attacks on microchips and circuits challenge the concept of hardware trust anchors, while the microelectronics business model and supply-chain ecosystem create unprecedented security concerns. This talk covers challenges in hardware security and emerging solutions for creating secure, trustworthy hardware for IoT applications. It addresses design-for-security, test and validation, the growing interplay between security and verification, learning-guided security design and verification, and the use of generative AI to create a new paradigm of hardware-centric IoT security.

Keywords: Digital Twins, Terrestrial and Space Applications.
Abstract: Digital Twin technology is widely applied across industries but under-explored for spacecraft launch testing. This paper reviews existing research and proposes a DT‑based launch‑testing framework, highlighting opportunities for anomaly detection and virtual fault analysis to improve safety, reduce costs, and enable more realistic pre‑launch evaluation.

Keywords: Simulation-Based Validation and Verification, Architecture
Abstract: Tensegrity form-finding and physical property prediction are fundamental inverse problems in structural mechanics, which aim to determine equilibrium configurations and internal force distributions. These problems are challenging due to strong nonlinearity arising from the coupling between geometry and forces, the need to ensure structural stability, and the enforcement of constraints such as boundary conditions and symmetry. Moreover, traditional methods often lack robustness to noise and outliers. This paper proposes an energy-based learning framework for clustered tensegrity form finding and physical property prediction. The proposed approach incorporates total potential energy minimization and constitutive relations into the training objective, enabling the simultaneous prediction of equilibrium nodal configurations and associated physical quantities, including member forces and force densities. By incorporating energy-based physical losses directly into the learning process, the framework improves physical consistency, robustness, and data efficiency. Numerical experiments on tensegrity structures, including prism and lander systems, show the great potential of the proposed approach and demonstrate its capability for scalable form finding and accurate prediction of structural properties.

Keywords: Networking, Digital Engineering, AI for Cognitive Sensing and Data Fusion
Abstract: Deep learning based intrusion detection systems (IDSs) achieve strong closed-set accuracy but fail silently when novel attacks appear after deployment. Existing out-of-distribution (OOD) detectors relying on softmax confidence, predictive uncertainty, or logit energy operate on the classification head and miss novel attacks that align with known-class logit directions. We propose Local Manifold Coverage Deficit (LMCD), an OOD criterion on L2-normalised deep features requiring no additional trainable parameters. LMCD computes class-conditional local support from the nearest known-class manifolds and amplifies uncertain boundary cases through an interpretable nearest-to-second-nearest distance-ratio margin. The threshold is calibrated on held-out known validation traffic to target a specified false-alarm rate. Under the single-novel-class assumption, detected candidates are provisionally assigned to one new class without using held-out labels during candidate selection. On USTC-TFC2016 and CIC-IDS-2017, LMCD raises novel-attack detection recall from 25.96% to 65.88% and from 57.84% to 71.27%, achieving mean updated accuracies of 97.23% and 96.55% across all leave-one-label-out scenarios. In selected hard scenarios, LMCD further reaches near-perfect recall while maintaining precision above 96%.

Keywords: Enabling Technology for Artificial Intelligence, Networking, Software and Networking
Abstract: Cyber ranges have evolved significantly over the past 18 years since they were first introduced by the Defense Advanced Research Projects Agency. This paper addresses how Large Language Models (LLM) equipped with a Retrieval Augmented Generation (RAG) database can be used to automate the manual process of range creation for the benefit of cybersecurity training.

Keywords: Enabling Technology for Artificial Intelligence, Simulation-Based Validation and Verification, AI for Cognitive Sensing and Data Fusion
Abstract: Infrared Search and Track (IRST) systems utilized for remote-sensing applications, such as missile defense and surveillance, are often hindered by their inability to overcome noise and clutter within complex, dynamic scenes of infrared imagery captures. While performance in IRST systems is greatly augmented by the integration of convolutional- and deep-neural networks, vulnerability to adversarial attacks exist within neural networks underlying operationally advantageous infrared small and dim target detection models. This study aims to test an IRST-applicable convolutional neural network against gradient-based adversarial attacks. This 3D U-Net is a novel infrared target segmentation model developed by Eismann et al. in 2025. The dataset utilized was generated by the Air Force Institute of Technology Sensor and Scene Emulation Tool and contains remote-sensing infrared images of aerial targets. The implemented attacks include both untargeted (offline) and targeted (online) single-step Fast Gradient Sign Method attacks. This paper primarily uses mean dice scores to analyze attacks performed against the model, with some emphasis on precision and recall for analysis of attack effects. Despite achieving a clean mean dice score of 0.8744, the 3D U-Net model proves vulnerable to visually imperceptible adversarial perturbations: at ϵ = 0.003, the untargeted and targeted attacks reduce performance to dice scores of 0.4439 and 0.2706, respectively.

Keywords: Electronic Warfare and Countermeasures, UAVs
Abstract: We present a passive UAV detection system exploiting electromagnetic motor signatures via circular coil arrays. Through FFT analysis of five spectral bands, the system identifies UAV-specific harmonic fingerprints and estimates azimuth within 0.5% error. Proof-of-concept experiments demonstrate 3-meter detection, with theoretical analysis indicating kilometer-scale range through ferromagnetic core optimization.

Keywords: Electronic Warfare and Countermeasures, AI for Cognitive Sensing and Data Fusion
Abstract: This paper presents a Resource-aware Decision- Making adaptive radar anti-jamming (RADAMAJ) framework in dynamic and adversarial environments. Rather than optimizing a single mitigation technique, the proposed approach formulates anti-jamming as a sequential decision problem over a heterogeneous set of countermeasures, including deep learning based denoising, spectral filtering, spatial beamforming, and frequency agility. A Deep Q-Network (DQN) is employed as a meta-controller to select actions based on real-time observations of jammed radar signals. To bridge the gap between algorithmic performance and practical deployment, a novel reward function is designed to explicitly balance target detection accuracy, signal fidelity, and computational resource consumption. The RADAMAJ framework is evaluated in a realistic simulation environment that models a Frequency-Modulated Continuous-Wave (FMCW) radar with adaptive and multi-modal interference. Experimental results demonstrate that the learned policy achieves superior overall performance compared with static strategies by dynamically trading off effectiveness and cost, selectively deploying high-complexity methods only when necessary. These results highlight the importance of resource-aware control in next-generation cognitive radar systems operating under adversarial conditions.

Keywords: Enabling Technology for Artificial Intelligence, SARs
Abstract: Classifier performance decomposition methods like Distribution Shift DEcomposition (DISDE) can provide valuable insight into the failures of automatic target recognition (ATR) algorithms under domain mismatch by attributing the errors to covariate or conditional shifts. This information enables developers to design specific interventions to recover algorithm performance. However, when the domain shift can be expressed in terms of changes in operating conditions (OCs)—target, sensor, or environment variables that affect the data—this information is frequently ignored, resulting in a decomposition that leads to an ineffective intervention. To address this, we propose OC-DISDE, a simple modification to DISDE that performs the decomposition in a joint space shared by the classifier features and OCs. This enables the decomposition to leverage the direct causal link between ATR performance and the underlying physics of the problem and correctly separate covariate shift from conditional shift. Through experiments with synthetic data and real-world synthetic aperture radar (SAR) images, we show that our proposed approach highlights OC-induced covariate shift while remaining robust against uninformative OCs. This work motivates future efforts to design interventions that leverage OC-driven performance decompositions.

Keywords: Wider Bandgap Semiconductors, Extreme Environment Electronics, Power Electronics
Abstract: Eight 200µm n-type Gallium Oxide (GaO) Schottky diodes were exposed to 100krad of 20MeV proton radiation at the ATLAS beamline at Argonne National Laboratory. A custom fixture was developed to allow irradiation of multiple devices in a series in a vacuum chamber with applied biasing. Devices were biased at either 0V, 2V, or -100V. Irradiation was divided into eight stages with voltage sweeps done between each stage. One device at -100V was observed to fail during radiation. Another device was damaged due to being on the same package as the first. Three other devices showed increased leakage current between one to two orders of magnitude above their specifications in post-radiation observation. There was no observed difference between the devices biased at 2V and those at 0V.

Keywords: Microelectronics, bio-inspired sensors
Abstract: This paper presents a reconfigurable capacitively coupled chopper amplifier (CCOPA) that proposes a cascaded-beta architecture to achieve a highly power-efficient closed-loop gain of 144V/V. By dividing the high target closed loop gain across two identical stages each of gain 12V/V, the exponential power penalty of high-gain single-beta topologies is successfully circumvented. This cascaded approach eliminates the need to massively over-provision the gain-bandwidth product (GBW) to maintain closed-loop bandwidth at extreme gains, achieving significantly lower power consumption than an equivalent single-beta-stage implementation. The core of each stage is a single-stage regenerative feedback amplifier capable of a high open-loop DC gain of 145dB by utilizing controlled positive feedback to minimize the effective output conductance. Designed in a standard 180nm CMOS process, the system features a chopping and auto-zeroing scheme to suppress 1/f noise and offset, alongside ripple reduction. The architecture is dynamically reconfigurable, operating as a highly efficient single-beta-stage for lower gains (1 to 12) and switching to the cascaded-beta configuration for high-gain (12-144) sensing applications.

Keywords: Terrestrial and Space Applications.
Abstract: Autonomous satellite constellation missions demand real-time telemetry architectures that can fuse heterogeneousdata streams into state representations suitable for machine learning-based coordination. Existing systems, designed primarily for human operators, lack the integration, latency, and data structure requirements of reinforcement learning algorithms. We present a telemetry bridge architecture for Basilisk-based satellite constellation simulation that addresses this gap. The architecture unifies controller kinematics, resource telemetry, and gossip-based communication data into structured state snapshots across Low Earth Orbit (LEO), Medium Earth Orbit (MEO), and Geostationary Earth Orbit (GEO) configurations. A dual-mode streaming layer built on Apache Kafka and Server-Sent Events (SSE) delivers these snapshots at sub-second latency to both realtime dashboards and batch learning pipelines. Integrated realtime fault detection and a web-based interface for task creation, assignment, and reassignment support both autonomous coordination and human-in-the-loop oversight. Experimental validation on an 18-satellite constellation demonstrates fusion latency of 12.3 ms, state consistency above 99.89%, and fault detection within 3.2 seconds, establishing a real-time telemetry substrate for future autonomous coordination and task-management studies. Index Terms—Satellite Constellation Simulation, Real-Time Spacecraft Interfacing , Telemetry Integration in Basilisk.

Keywords: Adaptive Algorithms, Enabling Technology for Artificial Intelligence, Robotics
Abstract: While the importance of data quantity is well understood within robotic imitation learning, little is known about quantifying data quality. This paper studies how data collection with two popular robot teleoperation modalities affects downstream pick-and-place task performance, given similar amounts of training data.

Keywords: Robotics, Terrestrial and Space Applications., Simulation-Based Validation and Verification
Abstract: Servicing space satellites presents significant challenges, particularly with increasing debris and satellite obsolescence. A promising solution involves utilizing a high-degree-offreedom space manipulator system to deploy a deputy satellite for maintenance, inspection, and debris removal. This work proposes an approach that leverages the manipulator’s throwing motion to launch the deputy toward a client satellite, reducing fuel consumption and improving accessibility for orbital maintenance. In simulations, the deputy is released from a chief satellite in geostationary orbit toward a client satellite, targeting a 3-hour rendezvous over a 1 km distance. The mission achieves this with imparted velocities of approximately 10 cm/s, assuming no external thrusters are used for deputy attitude or trajectory control.

Keywords: Integrated Photonics, Communications through Interconnects and/or Free Space
Abstract: A proteretic device design model using five Semiconductor Ring Lasers (SRLs) has been presented. The proposed device is the reverse of a hysteretic system, exhibiting a predictive rather than delayed response. Unlike traditional hysteretic systems characterized by delayed state transitions, proteresis describes a predictive inverse-hysteretic phenomenon where transitions occur in advance of the triggering signal thresholds. The architecture is implemented using cascaded Schmitt trigger structures realized through SRLs. By exploiting injection locking and gain competition mechanisms, the system demonstrates a stable switching speed of 500 MHz in simulation. Technical analysis shows that proteretic behavior is achieved when specific hierarchy constraints are satisfied between the cascaded Schmitt triggers. The proposed approach provides a pathway for mitigating propagation-delay limitations in high-speed signal processing, enabling potential applications in all-optical analog-to-digital converters (ADCs) with improved noise immunity.

Keywords: Architecture, AI for Cognitive Sensing and Data Fusion, UAVs
Abstract: Drones increasingly depend on real-time object detection for surveillance, search-and-rescue, and infrastructure inspection, yet current onboard pipelines rely on supervised models such as You Only Look Once (YOLO) that must be retrained whenever the target class set changes. Open-vocabulary foundation models offer a possible escape from this retraining loop, but their effectiveness on the small objects, top-down viewpoints, and cluttered scenes typical of aerial imagery is unknown. I present, to my knowledge, the first quantitative evaluation of Segment Anything Model 3 (SAM 3) on drone footage, applying the model zero-shot to the VisDrone benchmark and the Semantic Drone Dataset using only plain-text class prompts. SAM 3 reaches a per-class mean Intersection-over-Union (IoU) of 0.241 on VisDrone (10 classes) and 0.282 on the Semantic Drone Dataset (22 classes), with strong results on large vehicles (car: 0.872 on Semantic Drone, 0.433 on VisDrone) and near-zero results on small or ambiguous classes (bicycle: 0.096; obstacle: 0.009). Recall consistently exceeds precision by roughly a factor of two, indicating systematic over-segmentation. The findings suggest SAM 3 is a useful starting point for flexible aerial perception, but is not yet a drop-in replacement for trained drone detectors, and they align with concurrent observations on other open-vocabulary detectors evaluated on aerial data.

Keywords: UAVs, Networking, Communication
Abstract: Unmanned Aerial Vehicle (UAV) communication systems are increasingly deployed to provide connectivity in dynamic and infrastructure-limited environments, where maintaining continuous service during topology changes remains a key challenge. This paper presents a hybrid experimental framework that combines Network Simulator 3 (ns-3) with real-time hardware-in-the-loop (HIL) emulation through Linux TAP (virtual Layer-2 network tap) interfaces. The proposed system enables controlled evaluation of service continuity across access, backhaul, and inter-UAV links under configurable impairments such as delay and packet loss. The framework is designed to capture network behavior during link disruptions and transition events, including UAV handoff and associated failover mechanisms. Preliminary results demonstrate that the proposed setup provides a practical and flexible platform for analyzing service interruption and recovery behavior, offering insights into improving service continuity in UAV communication systems with multiple communication links.

Keywords: Software and Networking, Trusted and Assured Hardware, Collaborative/Multi-Agent Systems
Abstract: Traditional firmware security design suffers from a sluggish offense–defense iteration cycle, making it difficult to anticipate complex logical vulnerabilities before deployment. This paper presents a methodology that uses large language models (LLMs) to accelerate that cycle and report its application to 40 independently developed firmware implementations from the MITRE eCTF competition.

Keywords: Trusted and Assured Hardware
Abstract: Cryptographic accelerators on Field-Programmable Gate Arrays (FPGAs) are vulnerable to bitstream-level fault injection attacks, but spatially precise attacks on proprietary bitstreams remain difficult. This paper presents an automated framework that uses Large Language Models (LLMs) to guide FPGA bitstream manipulation without requiring original HDL source code or post-synthesis netlists. Using the Project X-Ray toolchain, Xilinx Artix-7 bitstreams are translated into semantic FPGA Assembly (FASM), allowing configuration data to be examined at a readable level. An entropy heuristic identifies high-density Look-Up Table (LUT) regions associated with Advanced Encryption Standard (AES) S-Boxes, while an LLM assists in synthesizing stuck-at fault mutations.

Evaluation on a ChipWhisperer CW305 target showed that targeted mutation of the X40 slice column did not cause localized data-path corruption, but instead produced catastrophic state-machine failure. The modified hardware consistently generated a null-state ciphertext array of all zeros, regardless of plaintext input. This suggests that standard synthesis tools may co-locate sequential control routing and combinational cryptographic logic, creating a Denial-of-Service (DoS) vulnerability. These findings show the feasibility of neutralizing secure cryptographic wrappers through targeted bitstream mutation and highlight the need for strict physical isolation boundaries, such as p-block constraints, in secure FPGA floorplanning

Keywords: Software and Networking, Enabling Technology for Artificial Intelligence, Collaborative/Multi-Agent Systems
Abstract: Physical Unclonable Functions (PUFs) are widely used in hardware security due to their lightweight authentication capabilities. However, strong PUFs remain vulnerable to machine learning–based modeling attacks under limited challenge-response pair (CRP) access conditions. This paper presents a reinforcement learning–based adaptive attacker for efficient CRP selection in PUF modeling attacks. Using Proximal Policy Optimization (PPO) under a constrained 5% CRP access scenario, the proposed method achieves approximately 69% prediction accuracy without data augmentation. The results demonstrate that adaptive CRP selection improves attack efficiency and highlights stronger vulnerabilities in PUF-based security systems.

Keywords: Trusted and Assured Hardware
Abstract: This paper presents a polymorphic logic based approach for mitigating power side-channel leakage in RTL hardware designs. Selective polymorphic gate replacement introduces implementation diversity and varying switching characteristics, reducing stable leakage signatures across operating conditions. Illustrative RTL-level analytical evaluation indicates improved leakage resilience with moderate area and power overheads.

Keywords: Trusted and Assured Hardware
Abstract: This paper proposes configurable inter-tier connectivity for secure 3D integrated circuit design. Multiple valid vertical communication configurations are introduced to increase structural uncertainty and complicate reverse-engineering attacks targeting inter-tier connectivity. Illustrative analytical evaluation examines connectivity ambiguity, security implications, and associated implementation overheads in representative 3D IC architectures.

Keywords: Communication, Brain machine interfaces BMI. and Autonomy in Medicine, Enabling Technology for Artificial Intelligence
Abstract: Effective disclosure of medical error requires factual clarity, accountability, empathy, and a prevention-oriented plan. Automated assessment is challenging as clinically sensitive human-rated disclosure interactions are scarce and consented. This study presents HEART-Synth, a 400-transcript resident-patient disclosure synthetic dataset for scoring development-stage rubrics. Each conversation is presented as transcript with multi-turn dialogue and ordinal scores ranging from 0 to 3 for the HEART domains used in our medical error disclosure curriculum. The dataset was produced with independent per-domain performance sampling, separated dialogue generation and blind scoring, conservative second-pass review, duplicate rejection, and audit-only metadata excluded from model inputs. It is evaluated with audits for scenario-score coupling, lexical shortcuts, adversarial masking, linguistic diversity, and score coverage. HEART-Synth lowers the mean normalized mutual information between scenario identification and score from 0.825 to 0.068 when compared to an initial simple-generator baseline. Linguistic diversity (distinct-2 = 0.265; self-BLEU = 0.146) is preserved, all score levels in every domain are covered, and the best TF-IDF lexical baseline macro-F1 is reduced from 0.886 to 0.560. These findings support the technical quality and development-stage use of HEART-Synth, though future applications should integrate human score validation to ensure clinical validity.

Keywords: Robotics, Autonomous Guidance, Navigation and Control, Terrestrial and Space Applications.
Abstract: The Universal Manipulation Interface (UMI), from Stanford's Robotics and Embodied AI Lab, trains manipulation policies for terrestrial robots using imitation and diffusion based learning. Our long term goal is to extend this toward autonomous in-space servicing, a domain constrained by limited power, computation, latency, and scarce physical validation. As a first step, we reconstruct a Franka Emika Research 3 manipulator with a custom gripper inside NVIDIA Isaac Sim and deploy a UMI trained diffusion policy without fine tuning. The policy uses a ViT-B/16 backbone, and the Lula inverse kinematics solver performs control from its end effector commands. We formalize the coordinate and action frame transformations needed to transfer a policy trained on handheld human demonstrations onto a simulated embodiment, and show subcentimeter tracking. Across 100 autonomous pick and place episodes, geometry and timing transfer faithfully, while the visual domain gap between rendered and real observations emerges as the dominant barrier, bottlenecking placement despite reliable grasping. By isolating the reality gap to a single layer, this Real2Sim verification establishes a validated testbed toward future free floating microgravity manipulation and dual arm object handover.