Search Results (1,446)

Embedded systems increasingly require on-device deep learning, yet their processors must simultaneously handle real-time sensing, networking administration, and data control. Existing Field-Programmable Gate Array (FPGA) accelerators typically target peak per-operator throughput without addressing concurrent execution demands of real-time embedded platforms. This paper presents a systolic array-based accelerator prototype implemented on the Zynq-7000 SoC integrated directly into PyTorch, enabling dense linear algebra to be delegated to the FPGA chip while Cortex-A9 continues executing the software stack uninterrupted. Unlike traditional accelerators optimized for peak per-operator speed, this design prioritizes asynchronous co-executionbetween the processing system (PS, the dual-core Cortex-A9) and the programmable logic (PL): The PL performs tiled matrix multiplication, while the PS executes preprocessing, orchestration, and I/O data concurrently, increasing effective end-to-end throughput regardless of the relative isolated performance of CPU and FPGA. The proposed module includes high-level-synthesis (HLS)-based matrix multiplication, activation functions, and Advanced eXtensible Interface (AXI)-Stream Direct Memory Access (DMA) interfaces, wrapped as custom PyTorch kernels under the PetaLinux operating system. The results obtained on the PYNQ-Z2 board show that, once the DMA transfer time is included in the measurement, the FPGA path does not surpass Cortex-A9 in isolated per-operator latencies across the evaluated range; the benefit lies instead in delegating the matrix compute to the fabric at low incremental power while the host CPU cores stay available for concurrent tasks. A concurrent workload sweep across matrix sizes from $8\times 8$ to $256\times 256$ confirms that the co-execution mode sustains 98–99% of available PS compute throughput compared with a constant ≈50% in single-core blocking mode; the difference is statistically significant for all evaluated sizes (see Mann–Whitney U: $U=25$, $p=3.97\times {10}^{-3}$, perfect discrimination, $n=5$). A fair dual-core CPU-only baseline attains comparable PS availability, so this figure reflects the dual-core scheduling that co-execution enables rather than a per-operator advantage of the fabric; the accelerator’s distinct role is to perform the matrix arithmetic off the general-purpose cores at low incremental power. The design occupies only 8% of available look-up tables (LUTs) and 5% of digital signal processing (DSP) blocks, maintains 1.69 W power with a junction temperature of 44.5 °C, and achieves 96.10% MNIST accuracy under fixed-point arithmetic. Full article

A cascade impactor equipped with microparticle surface acoustic wave sensors along with a surface cleaning system is an innovative system developed by our team to measure particles and monitor air quality. The system has been proven to function properly under various particle concentrations. Nevertheless, long exposure times and heavily polluted media impose a limitation on cascade impactors, known as surface saturation. This problem affects the sensitivity of our sensors, which tends to degrade with particles fouling the surface. To overcome this issue, a surface cleaning system that uses a Rayleigh wave-actuated water droplet has been implemented and tested. Rayleigh waves were generated on an innovative SAW chip, thus exciting the droplet using Radio-Frequency power. NaCl solutions and SiC particles were considered. The optimal droplet size was determined along with the required Radio-Frequency power. Experiments showed that the SAW-driven droplet successfully displaced the collected particles outside of the sensing zone, irrespective of their nature, without affecting the sensor’s performance. The results found in this study provide further improvements to our particle measuring system, advancing it towards an autonomous self-regenerating prototype. Full article

Lead (Pb)-based paints and other metal-bearing phases sourced from home exteriors can be added to adjacent soils, which break down over time, altering potential bioaccessibility. However, data on how particle size and morphology influence metal bioaccessibility remains limited, which was investigated in this study by using a simulated gastric acid (GA) extraction and characterization at the sub-grain scale. Soil samples were collected along horizontal transects at ~1 m intervals from three homes within the Akron metropolis, OH (USA). Each home had Pb-based paint on its exterior before recent renovations (2, 11, and 25 years ago, respectively). Soils were fractionated into sand, silt, and clay–fine silt using sieving, sedimentation, and laser scattering techniques, and analyzed by ICP-OES, XRD, and SEM-EDS. Lead content peaked within 0–2 m from the homes and declined with distance (maxima: 14,583 mg/kg, 8503 mg/kg, 2393 mg/kg at Sites 1–3, respectively), and primarily in the clay-fine silt fraction. Physical speciation of Pb at each site was invariant across the transect, but the percent Pb in the clay–fine silt Pb increased from Site 1 to Site 2, then declined at Site 3, suggesting the loss of fine particles at Site 3. Across sites 1 to 3, the paint chip abundance and size decreased, and secondary Pb-bearing phases appeared as discrete grains and secondary coatings. A similar trend is exhibited by Cu, Cd, and Zn, over time and space. These results highlight the ongoing breakdown of paint and re-sequestration of metals, which may elevate exposure risk from fine particles after housing renovations. Full article

Enzyme immobilisation within Metal–organic Frameworks (MOFs) provides a promising strategy for improving enzyme dispersion and local environment control, although the resulting performance depends strongly on the host materials, enzyme type and immobilisation conditions. Conventional in situ biomimetic mineralisation typically produces enzyme–MOF composites (enzyme@MOFs) with irregular morphologies, broad particle size distributions and aggregation, which can compromise catalytic performance and reproducibility. This study presents a flow-based microfluidic biomimetic mineralisation strategy for preparing horseradish peroxidase-encapsulated ZnBDC-NH₂ MOF composites. A flow-focusing microfluidic chip containing multiple rectangular baffle structures was designed to enhance transverse mixing, extend the effective residence time, and mitigate clogging during particle formation. Under the selected conditions, homogeneous HRP@ZnBDC-NH₂ particles with an average hydrodynamic diameter of 868.5 nm and a polydispersity index of 0.266 were obtained. The homogeneous HRP@ZnBDC-NH₂ showed an encapsulation efficiency of 56.17% and a loading content of 1.49%. Michaelis–Menten analysis gave a Km value of 52.49 μM for HRP@ZnBDC-NH₂, suggesting improved apparent substrate affinity compared with the corresponding bulk-synthesised sample. The results support the use of baffle-structured microfluidics as a controllable platform for enzyme@MOF synthesis, while further studies on enzyme leaching, reusability, long-term stability and extended chip operation are required to evaluate its operational robustness. Full article

Close visual inspection (CVI) forms a cornerstone of asset integrity. Advances in access technologies, including drones, have led to their increased use for remote visual inspection (RVI). However, comparative studies of RVI and CVI, in terms of defect detection, are currently limited. In this study, controlled trials were conducted with multiple industrial participants operating drones to inspect a concrete block wall containing representative defects. RVI performance was assessed in terms of defect detection, identification and sizing. RVI demonstrated moderate performance, with an overall defect detection rate of approximately 50% and no participant exceeding 0.6. Detection was strongly dependent on defect type, with larger defects such as spalling and chipping consistently identified, while finer defects such as cracking were frequently missed. Identification of defect type was less reliable and influenced by inspector experience, while sizing capability was limited, with only one participant providing approximate measurements for larger defects. An automated visual inspection device, termed ALICS (Adaptive Lighting for the Inspection of Concrete Structures), was deployed on two samples. Images were captured of the concrete surface under varying lighting conditions to enhance the visibility of any present defects. Images were then analysed using artificial intelligence (AI), with the device identifying all defects in the tested areas. These results highlight both the current limitations of RVI and the potential of illumination-enhanced automated approaches to improve inspection reliability. Full article

Introduction: Bronchopulmonary dysplasia (BPD) remains the most significant complication of extreme prematurity, affecting long-term respiratory outcomes. Because current diagnostic criteria identify only established lesions at 36 weeks postmenstrual age, early non-invasive biomarkers are needed. This pilot study aimed to identify a candidate salivary miRNA panel associated with BPD risk and to explore its pathogenetic relevance through in silico analysis. Methods: Saliva was collected from 20 preterm infants (10 with BPD and 10 controls), and miRNA expression was profiled using the GeneChip™ miRNA 4.1 Array Plate. Discriminatory performance was explored by ROC analysis within this discovery cohort, together with power and Spearman correlation analyses. Results: Expression of hsa-let-7b-5p, hsa-let-7c-5p, and hsa-miR-4454 was significantly elevated in the BPD group (p < 0.05, log₂FC ≥ 1.0), with no significant correlation with gestational age or birth weight. Bootstrap-corrected AUC values ranged from 0.905 to 0.937 and were supported by leave-one-out cross-validation. All three miRNAs showed very large effect sizes exceeding the minimum detectable effect at 80% power. Conclusions: In this pilot study, salivary miRNAs represent a hypothesis-generating candidate biomarker signal for BPD that requires external validation in larger, independent cohorts before any diagnostic or prognostic application can be considered. Full article

Total internal reflection (TIR) loss is a critical bottleneck limiting light extraction in AlGaN-based deep-ultraviolet (DUV) light-emitting diodes (LEDs), primarily due to the large refractive-index contrast at the light-emitting interface. Here, pyramid-shaped dielectric metasurfaces are designed and fabricated at the sapphire/air interface of flip-chip AlGaN-based DUV LEDs using a scalable nanoimprinting process. The metasurface functions as a light outcoupling layer that modifies the interfacial momentum-matching condition and redistributes photon propagation directions. Experimental results and theoretical simulations show that metasurfaces with different feature sizes enhance light extraction through distinct mechanisms. The subwavelength pyramid nanoarray perturbs the local optical field and provides additional in-plane momentum components, facilitating the coupling of high-angle photons into radiative channels, whereas the larger pyramid void structure mainly promotes photon extraction through geometrical redirection, tilted output interfaces, and dry-etching-induced rough surface scattering. As a result, an average light output power (LOP) enhancement of over 8% is achieved for AlGaN-based DUV LEDs emitting at approximately 275 nm. This work demonstrates a low-cost, scalable, and effective strategy for enhancing the LEE of DUV LEDs, with promising potential for high-efficiency ultraviolet optoelectronic application. Full article

The Dingtao M2 Tomb, the largest, highest-specification, and best-preserved “Huangchangticou” tomb currently discovered in China, is of great significance for cultural relic conservation. During its dismantling and protection, extensive white filamentous fungal contamination was observed on the surface of sand-buried wood components. To clarify the dominant fungal species and its impact on wood cultural relics, the dominant fungus was isolated and purified from contaminated wood samples, and identified as Serpula himantioides (designated as DTW) via molecular and morphological methods. Systematic studies were conducted on DTW’s wood degradation capacity, cellulase activities, regulatory effects of Fe³⁺ and Ca²⁺ on cellulase activities, whole-genome characteristics, and sensitivity to fungistatic agents. The results indicated that DTW exhibited strong wood degradation ability: after 60 days of inoculation, the maximum force, elongation at break, and tensile strength of wood chips decreased significantly, by 88.0%, 54.6%, and 88.6%, respectively, compared with the control group. The cultural relic microenvironment colonized by strain DTW is rich in Fe³⁺ and Ca²⁺. Both ions can significantly suppress the activity and specific activity of cellulase from strain DTW at elevated concentrations. Whole-genome sequencing revealed that DTW had a genome size of 68,519,434 bp with a Guanine–Cytosine (GC) content of 44.01% and 18,373 genes; Carbohydrate-Active Enzyme (CAZy) and Cell Wall-Degrading Enzyme (CWDE) database annotations confirmed its strong plant cell wall degradation ability. Additionally, DTW was sensitive to screened green fungistatic agents, which effectively inhibited its growth. This study clarifies the species and wood degradation mechanism of the dominant fungus on Dingtao M2 Tomb’s sand-buried wood, providing theoretical and technical support for the protection of the tomb’s wood cultural relics. Full article

Wireless edge systems increasingly rely on on-device learning for tasks such as user localization and misuser detection, but centralized training is often infeasible because of privacy and bandwidth constraints. In this work, we study communication-efficient federated learning schemes that jointly train lightweight neural models across clients. We compare a multi-task shared-backbone strategy with mid-training backbone freezing against task-separate baselines. Using a wireless graph simulator, we prototype both multilayer perceptrons (MLPs) and an edge-aware graph neural network (GNN) encoder trained with FedAvg, and we evaluate localization error, misuser F1-score, and byte-level communication over training rounds. The results show that the GNN-Shared model achieves a 28.9% lower localization error than the MLP-Shared baseline and a 17.7% lower localization error than the GNN-Loc_only baseline, while improving the best misuser F1-score by 2.7% compared with GNN-Mis_only. In terms of communication, the GNN-Shared model reduces the total communication cost by 40.1% compared with MLP-Shared over 40 federated rounds. Additional experiments with FedProx, GraphSAGE, and GAT baselines show that the proposed federated graph-learning framework is flexible across different graph backbones. Multi-seed experiments over five random seeds further confirm the robustness of graph-based federated learning, with GraphSAGE-FedAvg achieving strong average localization and misuser detection performance. Scalability experiments with up to 20 clients, 150 nodes, and different Dirichlet non-IID parameters show that the framework remains stable under larger graph sizes, while communication cost grows mainly with the number of participating clients. To move the framework closer to edge deployment, this paper also introduces a lightweight hardware-oriented GNN-Lite inference prototype with multi-neighbor accumulation, finite-state-machine-based sequential computation, and read-only-memory-based coefficient storage. The prototype achieved timing closure on an Artix-7 field-programmable gate array with 213 LUTs, 111 FFs, 18 DSPs, 0 BRAMs, and 0.106 W estimated total on-chip power. The dynamic power was only 0.001 W, and the estimated total energy consumption was 10.6 nJ per inference. Overall, the results show that federated graph learning is a promising direction for wireless edge intelligence, and that different graph backbones can be selected depending on the target trade-off among localization accuracy, misuser detection performance, communication cost, robustness, and hardware deployment. Full article

Introduction: Circulating cell-free DNA (cfDNA) has emerged as a promising minimally invasive biomarker in hepatocellular carcinoma (HCC), with potential applications in disease detection and prognostic stratification. This exploratory study aimed to evaluate the relationship between circulating cell-free DNA (cfDNA) concentration, liver dysfunction parameters, and hepatocellular carcinoma stage in an exploratory cohort: we sought to explore the extent to which cfDNA variability may be explained by the underlying liver disease environment and whether cfDNA concentration provides incremental information beyond routinely available markers of liver reserve for the discrimination between early- and late-stage hepatocellular carcinoma. Methods: Sixty-four newly diagnosed HCC patients were included. Clinical, laboratory, and staging data were collected. cfDNA was isolated from plasma, confirmed by on-chip electrophoresis, and quantified by fluorimetry. Logistic regression and ROC curve analyses were performed to assess the ability of several biomarker panels to discriminate early-stage HCC (BCLC 0–A) from intermediate/advanced-stage disease (BCLC B–D). Bootstrap resampling (2000 iterations) evaluated model robustness and coefficient stability. Additional linear regression analyses explored associations between cfDNA concentration and liver dysfunction parameters. Results: Linear regression demonstrated that liver dysfunction parameters explained approximately 53% of cfDNA variability, while HCC stage contributed minimally after adjustment. Models incorporating albumin, bilirubin, and platelet count as individual parameters demonstrated the best discriminatory performance after adjustment for liver disease etiology, achieving AUROCs up to 0.857. The only incremental value that cfDNA concentration added to the panel was an increase to its specificity (from 79.4% to 94.1%), while reducing sensitivity by 13.4%. Bilirubin and platelet count remained the most stable predictors after bootstrap, whereas cfDNA concentration was unstable. Limitations: The study was limited by its small sample size, cross-sectional design, and lack of longitudinal outcome assessment. External validation in larger prospective cohorts is necessary. Conclusion: A substantial proportion of cfDNA concentration variability (approximately 53%) was explained by routinely available liver dysfunction parameters, whereas HCC stage had minimal contribution. These findings suggest that circulating cfDNA concentration in patients with HCC may be influenced to a greater extent by the underlying cirrhotic liver environment and hepatocyte injury than by tumor burden alone. Integrated multimarker panels combining liver reserve parameters and liver disease etiology may be of interest for minimally invasive stratification of HCC. Although cfDNA concentration increased specificity for early- versus advanced-stage disease discrimination, its incremental value was low. Further validation in larger prospective cohorts is required. Full article

Orbital remote sensing platforms increasingly rely on CNN-based object detection for real-time situational awareness. However, deploying these models on spaceborne edge devices is challenging because of stringent Size, Weight, and Power (SWaP) constraints. In addition, the branch-and-merge topology of conventional single-stage detectors increases on-chip memory usage and introduces pipeline stalls, limiting efficient FPGA implementation. To address these challenges, we proposed RS-YOLO, an object detection algorithm developed through a hardware–software co-design approach. Structural re-parameterization converts heterogeneous branches into a sequential stream of padding-free convolutions, producing a deterministic dataflow and reducing per-state combinational control complexity and data-path multiplexing overhead. To mitigate the high-entropy concentration at the center of the re-parameterized kernels, we further introduce a spatial heterogeneous quantization (SHQ) engine. The SHQ engine assigns 16-bit precision to the central coefficients while preserving vectorized 8-bit computation for peripheral elements, reducing quantization errors for small targets with minimal hardware overhead. Experimental results on the Xilinx Zynq-7020 platform show that the proposed system consumes only 2.24 W while achieving a mean Average Precision (mAP) of 0.887 on the NWPU VHR-10 dataset, representing a 1.4% decrease compared with the FP32 baseline. The system also achieves an energy efficiency of 15.19 GOPS/W, demonstrating an effective balance between hardware efficiency and detection performance for resource-constrained edge platforms such as micro-satellite payloads. Full article

Spaceborne Synthetic Aperture Radar (SAR) provides all-weather, day and night and wide-area imaging capability, and plays a critical role in maritime surveillance. While substantial progress has been achieved in SAR ship detection, SAR ship classification remains relatively underexplored, mainly due to the scarcity of reliable category labels. Automatic Identification System (AIS) provides vessel identity, type, and dynamic trajectory information, and thus offers vessel type information that is difficult to obtain directly from SAR imagery. This paper proposes a fine-grained nine-category SAR ship classification method based on AIS association, which reorganizes the original AIS vessel types into nine fine-grained categories of SAR ship, transfers AIS vessel type information to SAR detection through a global optimal matching strategy, and supports SAR-only vessel category recognition. By retaining only high-confidence SAR and AIS matched pairs and cropping the corresponding SAR ship chips, an SAR ship classification dataset containing 4472 ship chips across the nine categories is constructed. In Monte Carlo experiments based on real AIS records, the proposed association strategy achieves more reliable high-confidence label generation than the compared association methods under close ship ambiguity, spatial perturbation, distractor AIS candidates, and AIS static size errors. In the benchmark experiment on the constructed classification dataset, ConvNeXt-Tiny achieves the best performance among the compared mainstream classifiers. These results demonstrate that AIS association can provide reliable category supervision for SAR ship classification, and the trained classifier can perform ship classification using SAR imagery alone. Full article

Background: We present a miniature (30 × 34 mm) 64-channel data acquisition headstage optimized for high-density surface electromyography. Methods: The headstage is made up of a multi-channel ASIC analogue front-end utilizing only MOS transistors, fabricated in 350 nm CMOS technology (IC die dimensions 6.9 × 1.8 mm), combined with an off-the-shelf multi-channel current-input ADC (DDC264, Texas Instruments). The ASIC analogue front-end employs MOS-based capacitors for both processing and AC-coupling. Results: The combination of these two sub-circuits enables the simultaneous recording of 64 channels at a typical sampling rate of 4 KHz with a maximum analogue bandwidth of 0.5–1500 Hz and a resolution of 20-bits. Typical input-referred-noise, determined by the analogue front-end, is 3.5 μV_RMS for a surface EMG bandwidth of interest of 20–500 Hz. This two-chip solution results in a power consumption of 5 mW per channel. Analogue performance variability of the custom ASIC was characterized across a dataset of 960-channels (15 dies) from two fabrication runs. Conclusions: This work practically demonstrates the viability of using both a MOS-only analogue front-end and commercially available off-shelf high-performance back-end hardware already developed for medical imaging applications to record high-density surface biosignals. The aforementioned techniques can be employed to reduce the size and cost for systems or wearable devices; facilitating the translation of high-density bio-acquisition setups from the research environment to more affordable commercial products. Full article

Compact malaria detectors for microcontrollers are almost always benchmarked on the NIH Malaria dataset with a per-cell random split. This leaks slide identity because the cells come from only about 200 slides and a random split mixes same-slide cells across training and testing. The leakage also distorts architectural conclusions: under a leakage-free slide-disjoint protocol, per-module ablation gains collapse to seed noise and an apparent cross-site robustness variant loses most of its advantage. Headline accuracy falls from 97.1% to 95.6%, a gap that sits within the cross-seed noise, and all eight tested architectures move the same way. The evidence is this unanimous direction, not the size of any single gap. This benchmarking finding is our main contribution. Two results survive. First, MalariaNet, our 21 K-parameter detector, reaches about 95.6% accuracy at 23.5 KB of INT8 weights, with a numerically faithful on-chip forward on an STM32H743 at a 1.2 FPS triage rate. Second, it is among the most interference-robust of the eight networks and the most robust microcontroller-deployable model. Scope is limited to single P. falciparum thin-smear cells. Slide-disjoint evaluation should become standard, and we provide MalariaNet as the first leakage-free, on-device-validated point-of-care malaria reference. Full article

The exponential growth of AI-driven data traffic necessitates the evolution of Data Center Networks toward high bandwidths and sub-microsecond latency. While co-packaged optics (CPO) offer a pathway to reduced energy consumption and increased capacity, they introduce significant challenges in optical chip coupling and packaging complexity. This study explores monolithically integrated metasurfaces as an alternative for optical interfaces, potentially reducing the need for bulky external microlens arrays or extremely precise mechanical alignment. We design an amorphous silicon (a-Si) metasurface on a Silicon-On-Insulator (SOI) platform operating at 1310 nm. By spatially mapping nanopillar radii to satisfy a spherical phase profile, we achieved near-vertical beam emission with an emission angle of 0.88° focused at a focal length of 98.99 μm. Broadband characterization across a 20 nm band confirms stable focusing and a confined spot size with moderate roll-off toward the band edges. The sensitivity of the emission profile of the device to fabrication imperfections in pillar radius, height, and sidewall taper is quantified. The coupling to a polymer-based optical redistribution layer (ORDL) is also studied, and the corresponding modal analysis demonstrates a maximum coupling efficiency of 68.2% into an SU-8 polymer waveguide. Tolerance analysis results reveal deterioration of 0.9 dB and 0.4 dB for $\pm 0.6 \mathsf{\mu }\mathrm{m}$ horizontal and $\pm 1.5 \mathsf{\mu }\mathrm{m}$ vertical misalignment respectively, making the interface compatible with relaxed alignment assembly assumptions, although experimental packaging validation remains required. The methodology is further validated at 1550 nm, demonstrating its applicability across telecom bands. These results suggest that integrated metasurfaces may simplify the packaging stack and enhance density for next-generation CPO links by providing precise, on-chip wavefront manipulation. Full article