Abstract Reliable air-quality monitoring is fundamental to environmental sustainability, public-health surveillance, climate resilience, and smart city governance. Among atmospheric pollutants, fine particulate matter with an aerodynamic diameter of <2.5 μm (PM2.5) is particularly hazardous because prolonged exposure is strongly associated with respiratory illness, cardiovascular disease, and premature mortality. Although distributed low-cost environmental sensing systems have emerged as scalable alternatives to traditional regulatory-grade monitoring stations, their operational... Reliable air-quality monitoring is fundamental to environmental sustainability, public-health surveillance, climate resilience, and smart city governance. Among atmospheric pollutants, fine particulate matter with an aerodynamic diameter of <2.5 μm (PM2.5) is particularly hazardous because prolonged exposure is strongly associated with respiratory illness, cardiovascular disease, and premature mortality. Although distributed low-cost environmental sensing systems have emerged as scalable alternatives to traditional regulatory-grade monitoring stations, their operational reliability is frequently compromised by missing observations, communication instability, calibration degradation, and long-term sensor drift. These issues significantly affect data quality and may reduce the reliability of downstream predictive environmental analytics. This study presents a comprehensive analytical framework for evaluating missing-value behavior and sensor drift within a large-scale PM2.5 microclimate sensing network. The analysis utilizes 897,085 environmental observations collected from 12 distributed sensing devices between May 2022 and May 2026. The proposed framework integrates device level and feature level missingness analysis, temporal missingness evaluation, rolling statistical drift analysis, baseline deviation monitoring, Kolmogorov Smirnov distributional drift testing, and z-score anomaly detection. The findings reveal substantial heterogeneity in sensing reliability across devices and environmental variables. Device-level analysis demonstrates severe acquisition instability in selected sensing nodes, while feature level evaluation indicates that wind related variables exhibit the highest missingness behavior. Sensor drift analysis further identifies significant long term deviations in PM2.5 sensing behavior, suggestingprogressive calibration instability and operational degradation in multiple sensing devices. Statistical drift testing confirms substantial distributional evolution across sensing periods. The proposed framework contributes toward trustworthy environmental intelligence by establishing a scalable methodology for evaluating sensing reliability in distributed smart city infrastructures. The findings additionally emphasize the importance of continuous quality assurance, adaptive recalibration, and automated drift monitoring for maintaining robust environmental sensing ecosystems Read More Read Less

Abstract The increasing complexity of System on Chip architectures has significantly intensified semiconductor verification challenges. Traditional workflows require extensive manual effort for testcase generation, constraint extraction, interface interpretation, and validation of complex hardware interactions. Recent advances in Large Language Models enable automation via long context reasoning, prompt engineering, and AI assisted test synthesis. This study presents a comprehensive analytical framework for LLM-driven semiconductor verification using structured... The increasing complexity of System on Chip architectures has significantly intensified semiconductor verification challenges. Traditional workflows require extensive manual effort for testcase generation, constraint extraction, interface interpretation, and validation of complex hardware interactions. Recent advances in Large Language Models enable automation via long context reasoning, prompt engineering, and AI assisted test synthesis. This study presents a comprehensive analytical framework for LLM-driven semiconductor verification using structured prompt orchestration, prompt chaining, semantic context engineering, and token aware optimisation. The analysis evaluates single shot and multi shot prompting, hierarchical prompt templates, staged reasoning pipelines, context management strategies, and token utilization efficiency for complex RTL verification tasks. The framework also investigates long context processing using Gemini 1.5 based architectures capable of handling large scale SoC repositories without conventional chunk fragmentation limitations. Comparative evaluation shows that multi shot prompt orchestration significantly improves reasoning continuity, verification completeness, and hallucination resistance while enabling scalable handling of interdependent RTL modules. The proposed framework contributes to scalable AI assisted verification systems for next generation Electronic Design Automation workflows. Read More Read Less

Abstract The increasing complexity of modern electronic systems has created a growing demand for intelligent Electronic Design Automation (EDA) frameworks capable of understanding circuit topology, signal propagation, and component interaction behavior. Traditional machine learning approaches are limited in capturing the non-Euclidean structural relationships inherent in electronic circuits because they operate primarily on independent feature vectors rather than connectivity aware representations. This study presents a graph-based analytical... The increasing complexity of modern electronic systems has created a growing demand for intelligent Electronic Design Automation (EDA) frameworks capable of understanding circuit topology, signal propagation, and component interaction behavior. Traditional machine learning approaches are limited in capturing the non-Euclidean structural relationships inherent in electronic circuits because they operate primarily on independent feature vectors rather than connectivity aware representations. This study presents a graph-based analytical framework for intelligent circuit understanding using a Heterogeneous Graph Attention Message Passing Network (HGAMPN). The proposed framework combines heterogeneous graph representation, Message Passing Neural Networks (MPNNs), Graph Attention Networks (GATs), hierarchical graph pooling, and explainable artificial intelligence mechanisms to model connectivity behavior, signal propagation, component dependencies, and hierarchical design structures within electronic circuits. The dataset employed in this study consists of structured circuit topology data obtained from the Kaggle Electronic Circuit for Machine Learning dataset. The dataset contains component level and connectivitylevel information describing electronic circuits composed of resistors, capacitors, transistors, diodes, LEDs, voltage sources, switches, and sensor elements. Each circuit is represented through structured metadata including component identifiers, component categories, pin level relationships, electrical values, and netbased interconnections. This representation enables the reconstruction of circuit structures as heterogeneous graphs where components and nets are modeled as nodes and electrical interconnections are represented asedges. The proposed framework demonstrates how graph-based learning architectures can capture circuit connectivity patterns, identify dominant signal propagation pathways, and discover hierarchical functional modules within electronic systems. The study further illustrates how explainable graph learning mechanisms can identify salient subgraphs, influential electrical paths, and critical component interactions. The resulting framework provides a strong foundation for intelligent EDA, autonomous circuit understanding, graphbased hardware learning, and explainable AI-driven electronic system analysis. Read More Read Less