The rapid growth of artificial intelligence, electric vehicles, and high-performance computing has accelerated the demand for miniaturized and thermally reliable electronic packages. As front-end scaling approaches its physical limits, advanced packaging has become central to improving system performance. However, heterogeneous integration, 2.5D/3D packaging, and chiplet architectures intensify thermomechanical reliability concerns because materials with different coefficients of thermal expansion are integrated within confined structures. Conventional Sn–Pb solders offered low processing temperatures and excellent ductility, but their toxicity drove the transition to lead-free systems. Sn–Ag–Cu solders, especially SAC305, have been widely adopted for their mechanical strength and fatigue resistance; nevertheless, their high melting temperature aggravates warpage, interfacial delamination, and thermal stress. In this context, Sn–Bi solders have attracted attention as low-temperature candidates, since eutectic Sn–58Bi melts near 138–139 oC. Despite this advantage, they suffer from brittleness, Bi-rich phase segregation, microstructural coarsening, electromigration-related degradation, and excessive interfacial intermetallic growth. This manuscript reviews the evolution of conventional solder systems, the metallurgical characteristics of Sn–Bi alloys, and recent strategies including micro-alloying and nanocomposite reinforcement for improving the reliability of low-temperature solder joints in advanced packaging applications.
In automated AI systems for urban change detection, deforestation tracking requires effective, scalable approaches that generalize across different ecosystems. Current Normalized Difference Vegetation Index (NDVI)-based approaches and single-biome deep learning networks lack good generalizability across regions and are energy-intensive. NDVI is prone to saturation over closed vegetation and fails over dry canopy; biome-specific networks overfit and require extensive labeled datasets. These limitations hinder accurate monitoring in diverse environmental settings. To this end, a spectral-attention U-Net-based model is introduced using Short-Wave Infrared (SWIR) bands and multi-temporal Landsat 8/9/Sentinel-2 imagery for biome-flexible deforestation monitoring with transfer learned ResNet-50 and carbon-friendly training regimes. SWIR improves vegetation distress surveillance and is less susceptible to atmospheric perturbations; transfer learning reduces data demands and training energy with sustainability improvements. Also, Spectral Attention Gates (SAGs) are proposed that dynamically weight spectral inputs for improved feature extraction, and integrate a lightweight decoder design using transposed convolution for sharper boundary detection. The proposed model achieves 92.3% IoU (Intersection over Union) in the Amazon and 88.7% in Jordan’s temperate forests with up to 22% higher precision in arid regions compared to the NDVI approach, while maintaining low energy costs (1.43 kgCO2e/10 km2 on NVIDIA A100). Hotspot analysis reveals 42.3 km2/year loss in the Amazon (Z=4.71) and 12.5 km2/year in Jordan (Z=3.28), validated against 5000 ground truth patches (κ = 0.82).
Dynamic task assignment in multi-agent systems aims to distribute limited agent resources among time-varying tasks to improve operational efficiency and reduce completion time. This paper develops a game-theoretic distributed learning framework for dynamic task assignment by modeling the problem as an exact potential game that aligns individual agent utilities with a global performance objective. The task dynamics capture time-varying task loads under fixed agent execution capacities, and an analysis of optimal static assignment characterizes completion-time minimization. Based on this potential game formulation, a synchronous distributed learning algorithm is proposed that utilizes the potential function as a Lyapunov measure, ensuring monotonic improvement and convergence to a pure-strategy Nash equilibrium without centralized coordination. The algorithm relies only on local information and supports parallel updates, enabling scalability and adaptability in dynamic environments. Numerical simulations demonstrate that the proposed method achieves improved convergence behavior and higher task completion efficiency compared with representative baseline algorithms.
Fibonacci words represent a well-known category of infinite sequences of words that have garnered significant attention in the fields of word combinatorics and theoretical computer science. Fibonacci words possess exceptional structural properties that make them stand out significantly. This study analyzes the languages produced by involutive Fibonacci words, a distinct category of Fibonacci words. Fuzzy involutive Fibonacci words are defined by incorporating membership degree into the rules of production. Their properties are examined, and the formal languages linked to these words are defined.
Embedded voice-command systems are increasingly adopted in cyber-physical and industrial environments, yet their machine-learning pipelines remain vulnerable to data poisoning attacks that can silently degrade recognition performance or induce targeted misclassifications. This paper proposes a multi-stage poisoning detection pipeline tailored for embedded voice-command applications, combining fast integrity screening with progressively stronger detection models. First, an audio integrity layer analyzes waveform and spectrogram consistency to flag abnormal patterns indicative of injected perturbations or corrupted samples. Second, a lightweight machine-learning detector based on Random Forests operates on compact statistical descriptors extracted from spectrogram representations, enabling efficient on-device or edge-level screening. Third, a spectrogram-based Convolutional Neural Network provides high-sensitivity discrimination between clean and poisoned samples, acting as a robust validation stage in higher-assurance deployments. To increase resilience, the training procedure integrates regularization and controlled augmentation to mitigate overfitting to spurious artifacts and reduce poisoning sensitivity. The proposed approach is designed to be deployable under embedded constraints by separating low-cost screening from deeper analysis, and it offers a structured methodology for securing voice-command pipelines against poisoning threats.
Massive Multiple-Input Multiple-Output (mMIMO) is a key enabling technology for beyond fifth-generation (B5G) and sixth-generation (6G) wireless systems; however, its performance critically depends on accurate channel state information (CSI). Conventional minimum mean square error (MMSE) channel estimators are typically derived under the assumption of spatially uncorrelated fading, which rarely holds in practical deployments due to antenna coupling, limited spacing, and structured propagation environments. As a result, spatial correlation can significantly degrade estimation accuracy. This paper investigates uplink channel estimation for mMIMO systems under spatially correlated channels. First, an analytical characterization of the effect of correlation on the traditional MMSE estimator is presented. Then, two correlation-aware estimation techniques are developed. The first approach introduces a regularized MMSE formulation to improve robustness in correlated scenarios. The second approach proposes a pre-whitening-based MMSE estimator that exploits channel covariance information to transform the correlated channel into an equivalent white channel prior to estimation. Closed-form solutions for the optimal filters are derived, and their complexity is calculated. The results of the simulation verify that the pre-whitening based estimator outperforms the conventional and regularized MMSE estimators over a wide interval for the value of the signal to noise ratio (SNR) and the correlation coefficient.
Consider an active target search and coverage problem in which a mobile 3D sensor aims both to quickly find relevant targets in its environment and to quickly scan these targets so as to cover them. We propose a deep reinforcement learning method to find a sequence of sensor poses that solves this problem. The method uses a deep hierarchical encoder-decoder architecture that heavily modifies the point cloud transformer network. To reduce computation, a new way is proposed to select representative points from the cloud, based on cosine similarities between feature vectors. In simulated experiments, the novel encoder-decoder architecture significantly improves performance compared both to a greedy baseline, and to a network structure closer to the point cloud transformer. The architecture also improves classification and segmentation performance in supervised learning on point clouds. We successfully train much deeper networks than those usually employed in reinforcement learning, and we believe the ideas used to achieve this can be adapted to deep reinforcement learning in general.
Root cause analysis (RCA) has the capability to uncover the underlying reasons for anomalies and offers robust support for network systems recovery. However, one of the most significant challenges in RCA is the difficulty in capturing and analyzing the finegrained human factors that trigger anomalies. To address this issue, this paper presents an approach for correlating fine-grained events with key performance indicators (KPIs). Specifically, in this approach, captured function events, treated as fine-grained human factors, are analyzed with modified Petri Nets to determine the root cause of anomalies. The experimental results demonstrate that the proposed approach is highly effective in tracing fine-grained human factors during RCA. Moreover, the F1-score in anomaly recognition can reach approximately 0.79.
The increasing use of large language models (LLMs) in conversational systems raises concerns about the credibility and verifiability of the information they generate. These models often produce fluent and convincing responses that may lack factual basis or supporting evidence. To address this challenge, a reference recommendation approach is proposed to retrieve relevant citations from LLM-generated text. The proposed approach treats the LLM output as a query and employs Sentence-BERT to create deep contextual embeddings of documents. Retrieval performance is further enhanced by integrating Siamese and Triplet neural network architectures to model semantic similarity and applying a submodular scoring function to ensure relevance and diversity in recommended references. Performance tests on a domain-specific dataset demonstrate that the proposed approach outperforms traditional retrieval approaches and recent baselines in standard evaluation metrics, including F1 @ k and mean recurrence rank (MRR). This work offers a scalable and effective solution for improving the reliability of AI-generated content through evidence-based support.
The weighted independent set problem is a classic combinatorial optimization problem with broad applications. However, since practical applications require high solution quality, effectively balancing solution efficiency and quality remains a significant challenge. with an exact search, using heuristics to reduce the search space effectively. Based on the predicted probability, the solving process is divided into two stages. First, a graph neural network is used to predict the marginal probability of vertices belonging to the solution, and deep reinforcement learning trains a policy to construct a high-quality initial solution. Second, an adaptive search strategy is proposed, which dynamically defines a search range based on the credibility of the predicted probability. By reducing the size of the space that the exact solver needs to search, the solution time can be significantly decreased. The experimental results show that the proposed hybrid method improves the average solving speed by 86% while maintaining the same solution quality as the exact solver. Meanwhile, it demonstrates good scalability for large-scale graphs.
The paper presents an approach of identifying linear systems without finite zeros. The identification of parameters for systems with one input and one output is observed. Using this approach, it is possible to identify the parameters of the system. The approach is practically verified on a series of identification examples, which confirm the justification of its application.
This paper introduces a unified approach to multilingual visual speech recognition (VSR) that combines cross-modal phonetic modeling with large-scale language decoding to enable robust generalization across low-resource and previously unseen languages. The architecture within the approach includes a Cross-Modal Transcriber that encodes synchronized audio-visual speech inputs into a language-agnostic phoneme space via a fine-grained cross-attention mechanism. To bridge perception and language understanding, two decoding pathways are explored: (1) a modular configuration that maps phonetic sequences to text using a pretrained large language model (LLM), and (2) an end-to-end formulation in which fused visual features are projected into the LLM's embedding space via a lightweight adapter for direct transcription. Experimental evaluations on the mTEDx multilingual corpus show that the architecture surpasses state-of-the-art VSR models, achieving up to a 6% absolute improvement in WER across Latin-derived languages.
In order to cope with the complex features of ambiguity, randomness and uncertainty in multi-attribute decision-making problems, this paper introduces the Dempster-Shafer evidence theory in the framework of cloud modeling. First, a cloud model is used to calculate the affiliation of each evaluation metric, which was subsequently converted to a basic confidence assignment function. Second, the game theory idea is borrowed to combine the dynamic and static weights of the evidence in the game, to improve the traditional evidence theory and realize the effective integration of information. The idea of average fit is identified again, and a comprehensive evaluation conclusion is drawn by comparing the closeness of the evaluation object to the optimal and worst solutions. The new electric power system investment project is illustrated, and the applicability of the algorithm is verified.
This paper introduces a timed process calculus designed to model reactive systems that can dynamically adjust their spatial organization and communication structure. Authors' calculus, named rTIMO, facilitates timed agent migration between distributed locations, with explicit timers governing both communication and movement actions. New behavioural equivalence relations are proposed aiming at identifying behaviours that closely align with desired specifications. To illustrate this, the paper includes an example of resource-sensitive routing, in which a driver aims to reach a specific location. The calculus proposed in this paper provides a formal foundation for compositional reasoning about timed, mobile multi-agent systems.
This paper presents a novel current-mode capacitor multiplier that offers a superior alternative to existing solutions. By simply adding a resistor, the proposed circuit achieves multiplied equivalent capacitance, outperforming state-of-the-art implementations. Simulation results demonstrate the circuit's effectiveness in a noise filtering application for a voltage reference, requiring only a 100pF integrated capacitor and 2 & micro;A quiescent current to achieve a 16Hz low cut-off frequency. Compared to traditional solutions, the proposed multiplier reduces the required capacitor size from 10nF to 100pF. The new design excels in both power supply rejection ratio and noise performance, making it an attractive solution for low-power, high-precision applications.
The detection of nitrophenolic compounds, including 4-nitrophenol (4-NP) and 2,4-dinitrophenol (2,4-DNP), is essential because of their significant environmental persistence and toxic effects on living organisms. In this work, we present a sensitive and selective approach based on a polyaniline (PANI)-coated fiber optic-surface plasmon resonance (FOSPR) sensor for real-time monitoring of these pollutants in aqueous environments. The PANI layer was fabricated directly onto the gold-coated sensing region of an unclad optical fiber via chemical oxidative polymerization, ensuring uniform coverage and strong adhesion. The PANI layer was characterized using scanning electron microscopy (SEM) and X-ray photoelectron spectroscopy (XPS), which provided detailed information on its morphological features and chemical composition. Moreover, the sensor performance was evaluated by tracking the shifts in the SPR wavelength upon analyte interaction. The developed sensor exhibited a clear linear response to both target compounds, achieving detection limits of 33 nM for 4-NP and 14 nM for 2,4-DNP. Moreover, the proposed FO-SPR sensor offers key advantages, including low cost, label-free operation, and rapid response, making it highly suitable for continuous environmental monitoring. Overall, this FO-SPR sensing technology represents a promising and practical solution for detecting hazardous nitrophenolic contaminants in water systems.
Bisphenol A (BPA) is widely used in plastic manufacturing and is known to have adverse effects on both human health and the environment. Among the available analytical approaches for BPA detection, electrochemical sensors represent a promising alternative due to their sensitivity, simplicity, and low cost. In this paper, the authors report the electrochemical synthesis of silver nanostructures-carbon-based composites involving green electrolytes, namely deep eutectic solvents (DESs). Single-walled carbon nanotubes (SWC-NTs) and graphene oxide (GO) were employed as carbon nanomaterials. The influence of the carbon nanostructure type and electrochemical synthesis parameters (pulse duration, frequency, duty cycle) on the morphology of the samples was discussed. A wide range of silver morphologies was obtained, from spherical nanoparticles to cubic structures. The composites were characterized by scanning electron microscopy coupled with energy dispersive X-ray spectroscopy (SEM-EDX), Raman spectroscopy, atomic force microscopy (AFM), and X-ray diffraction (XRD). The nanocomposite material was drop-casted on a screen-printed carbon electrode (SPCE) for the development of an environmental sensing device for the endocrine disruptor BPA. The AgNPs-GO/SPCE sensor exhibited a calculated limit of detection (LOD) of 0.46 & micro;M and a sensitivity of 0.01 A mol-1 L-1, with a wide linear concentration range of 1 & times; 10-7to 8 & times; 10-4 M for BPA detection.
This paper presents the design and experimental validation of a wideband, real-time spectrum monitoring system implemented on a multi-channel software-defined radio (SDR) architecture. The platform utilizes a USRP X310 equipped with dual TwinRX daughterboards to provide four independent receive chains. It has been demonstrated that a RFNoC-based acquisition chain performs FPGA rate reduction, while host-side processing in GNU Radio applies high-resolution FFT based spectrum estimation and real-time visualization. A spectrogram graphical user interface (GUI) is presented, incorporating percentile-driven auto-leveling with smoothing and display-domain frequency pooling to enhance wideband readability. A concatenation and trim method facilitate stitched wideband observation in the 5.8 GHz band. Controlled laboratory measurements in the 2.4 and 5.8 GHz ISM bands, including sweep injection and multi-UAV scenarios under Wi-Fi congestion, confirm stable real-time operation, continuity across the stitching boundary, and clear visibility of dominant drone downlink activity.
This paper presents a comparative study of surface acoustic wave (SAW) humidity sensors fabricated on Sc0.3Al0.7N/Si, functionalized with HfO2 and SiO2 thin films, operating in both Rayleigh and Sezawa propagation modes. The paper describes the manufacturing of SAW humidity sensors by advanced nanolithographic processing tools and deposition methods of two different functionalization layers. The effect of HfO2 and SiO2 films on the resonance frequencies of the devices is analyzed for both propagation modes. The devices were characterized under controlled relative humidity conditions (20. . . 90 %) by monitoring the resonance frequency shifts of both acoustic modes. Depending on the type of functionalization layer, the analyzed sensors exhibit distinct responses to humidity: the SAW sensors covered with HfO2 show an increase of the resonance frequency with the relative humidity, while the SAW sensor having SiO2 layer has an opposite response, the resonance frequency decreases with the increase of the humidity. This contrasting behavior is explained by the sensing mechanism that becomes dominant in each case: the stiffening effect for HfO2 and the mass loading effect for SiO2. The high values obtained for the humidity sensitivities confirm the advantage of SAW sensors manufacturing on III-Nitride/Si layered structures, allowing an increase of the resonance frequencies above 5 GHz.
This paper presents a solution for implementing frequency-compensation of low-power automotive output-capacitor-less linear regulators able to operate over a wide supply voltage range. First, a novel Active-Capacitive-Multiplier (ACM) is introduced that emulates the passive, R-series-C, frequency compensation network of a conventional NMOS regulator. Two design examples are shown that follow the same approach: the conventional NMOS regulator with RC compensation is sized to meet a set of real-life requirements; next, the RC network is replaced by the proposed ACM, sized so the resulting regulator has loop gain frequency characteristics similar to the conventional one; then the performances of regulators with conventional-RC and ACM-based frequency-compensation are compared. The ACM-based regulator exhibits significantly better response to fast transients: the output voltage undershoot/overshoot is reduced by about 3x for the NMOS regulator. The ACM has a negative impact on the power supply rejection (PSR) of the NMOS regulator. Third, an effective way of scaling up the load current of such regulators is presented: starting from the NMOS ACM-based regulator described above, with a maximum load current of 100 mA, an AI-assisted optimized sizing procedure was developed to re-size the pass-device and key ACM elements, so that the resulting regulator can handle loads up to 200 mA.
