Xiangning Lu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Xiangning Lu has a total of 71 co-authors. Here are the top co-authors by number of shared publications:
- Zhenzhi He
School of Mechatronic Engineering, Jiangsu Normal University, Xuzhou, China
14 shared publications - Lianchao Sheng
Jiangsu Normal University
7 shared publications - Tianchi Chen
Jiangsu Normal University
7 shared publications Zhenzhi He
6 shared publications- Lei Su
Jiangnan University
5 shared publications Tianchi Chen
5 shared publications
Latest Publications
- 30 July 2026
- Published by Springer Nature in The European Physical Journal Plus
- Vol. 141 (7) , 876
Design and simulation of a manifold–pin–fin hybrid heat sink for non-uniform heat sources
- 30 July 2026
- Published by Springer Nature in The European Physical Journal Plus
- Vol. 141 (7) , 876
Design and simulation of a manifold–pin–fin hybrid heat sink for non-uniform heat sources
- 2 July 2026
- Published by Elsevier in Microelectronics Journal
- Vol. 176
Wafer-YOLO: A lightweight object detection network for wafer inspection
- 2 July 2026
- Published by Elsevier in Microelectronics Journal
- Vol. 176
Wafer-YOLO: A lightweight object detection network for wafer inspection
- 18 June 2026
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 37 (27) , 276305
A lightweight high precision real-time system for bonding ball-pad localization in industrial packaging inspection
As a critical interconnection technology in chip packaging, Wire bonding imposes stringent requirements on the localization accuracy of bonding balls and pads. Conventional vision-based inspection methods often struggle to satisfy real-time constraints when targets are extremely small, densely distributed, and embedded in cluttered backgrounds. To meet these demands, a real-time high-precision localization approach was developed on an improved RT-DETR framework, termed Dynamic-Fast-Aware DETR (DFA-DETR). DFA-DETR reflects three key properties of the design, namely dynamic feature fusion, fast inference, and awareness enhancement for critical regions. Specifically, the backbone was redesigned as a lightweight ResNet-FasterBlock network, where partial-channel spatial convolution was employed to reduce redundant computation while maintaining feature expressiveness. Moreover, cross-scale feature fusion was refined using a DySample-Slimneck module, in which content-adaptive dynamic sampling replaces fixed upsampling to mitigate spatial misalignment across feature levels. In addition, a magnitude-aware attention module ( MALA-AIFI ) was incorporated to enhance region selectivity and suppress background-dominated responses under challenging industrial imaging conditions. Experiments on the bonding ball-pad positioning task demonstrate 77.92% mAP50-95 and 90.47% mAP@0.75, together with an overall F1-score of 96.62%. The resulting model runs at 86.81 FPS with only 16.30M parameters, indicating strong potential for real-time industrial deployment. Ablation studies and qualitative analyses further confirm the effectiveness of each component and their complementary contributions. Overall, an accurate and efficient solution is provided for online visual inspection in packaging wire-bonding processes.
- 18 June 2026
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 37 (27) , 276305
A lightweight high precision real-time system for bonding ball-pad localization in industrial packaging inspection
As a critical interconnection technology in chip packaging, Wire bonding imposes stringent requirements on the localization accuracy of bonding balls and pads. Conventional vision-based inspection methods often struggle to satisfy real-time constraints when targets are extremely small, densely distributed, and embedded in cluttered backgrounds. To meet these demands, a real-time high-precision localization approach was developed on an improved RT-DETR framework, termed Dynamic-Fast-Aware DETR (DFA-DETR). DFA-DETR reflects three key properties of the design, namely dynamic feature fusion, fast inference, and awareness enhancement for critical regions. Specifically, the backbone was redesigned as a lightweight ResNet-FasterBlock network, where partial-channel spatial convolution was employed to reduce redundant computation while maintaining feature expressiveness. Moreover, cross-scale feature fusion was refined using a DySample-Slimneck module, in which content-adaptive dynamic sampling replaces fixed upsampling to mitigate spatial misalignment across feature levels. In addition, a magnitude-aware attention module ( MALA-AIFI ) was incorporated to enhance region selectivity and suppress background-dominated responses under challenging industrial imaging conditions. Experiments on the bonding ball-pad positioning task demonstrate 77.92% mAP50-95 and 90.47% mAP@0.75, together with an overall F1-score of 96.62%. The resulting model runs at 86.81 FPS with only 16.30M parameters, indicating strong potential for real-time industrial deployment. Ablation studies and qualitative analyses further confirm the effectiveness of each component and their complementary contributions. Overall, an accurate and efficient solution is provided for online visual inspection in packaging wire-bonding processes.
- Open Access
- 26 February 2026
- Published by Elsevier in SSRN Electronic Journal
Wafer-YOLO: A Lightweight Object Detection Network for Wafer Inspection
Wafer defect detection, as a core process in the back-end of chip manufacturing, is critical for chip quality. Currently, machine vision is predominantly used to identify surface defects of the wafer. However, the inference speed of detection models often fails to meet the requirements of the production lines in real-time detection, potentially causing operational delays or defect accumulation. We proposed an algorithm of Wafer-YOLO based on the YOLOvl0n by integrating the models of Liteshnfthead, PCony, BoT, and SGE. The proposed Wafer-YOLO can enhance the ability of feature extraction and reduce convolutional redundancy, also optimizes the computational resources. These improvements simplify computation, expand the receptive field, and accelerate detection. Wafer-YOLO reduces GFLOPs, inference time, and parameters by 40.2%, 34.4%, and 11.5%, respectively, achieving 2.6M parameters and improving FPS from 344.2 to 526.3 (a 52.9% increase) with an inference time under 2ms. Overall, Wafer-YOLO surpasses other lightweight models in wafer detection speed and efficiency, proving its effectiveness and superiority.
- 26 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessWafer-YOLO: A Lightweight Object Detection Network for Wafer Inspection
Wafer defect detection, as a core process in the back-end of chip manufacturing, is critical for chip quality. Currently, machine vision is predominantly used to identify surface defects of the wafer. However, the inference speed of detection models often fails to meet the requirements of the production lines in real-time detection, potentially causing operational delays or defect accumulation. We proposed an algorithm of Wafer-YOLO based on the YOLOvl0n by integrating the models of Liteshnfthead, PCony, BoT, and SGE. The proposed Wafer-YOLO can enhance the ability of feature extraction and reduce convolutional redundancy, also optimizes the computational resources. These improvements simplify computation, expand the receptive field, and accelerate detection. Wafer-YOLO reduces GFLOPs, inference time, and parameters by 40.2%, 34.4%, and 11.5%, respectively, achieving 2.6M parameters and improving FPS from 344.2 to 526.3 (a 52.9% increase) with an inference time under 2ms. Overall, Wafer-YOLO surpasses other lightweight models in wafer detection speed and efficiency, proving its effectiveness and superiority.
- 5 January 2026
- Published by Springer Nature in International Journal of Dynamics and Control
- Vol. 14 (2) , 49
Dynamic characteristics of shearer gear transmission system with spalling fault
- 5 January 2026
- Published by Springer Nature in International Journal of Dynamics and Control
- Vol. 14 (2) , 49
Dynamic characteristics of shearer gear transmission system with spalling fault
- 1 January 2026
- Published by Canadian Science Publishing in Transactions of the Canadian Society for Mechanical Engineering
- Vol. 50, 1-14
Dynamic characteristics of electromechanical coupling in the cutting drive system of coal shearer with spalling fault
Due to the electromagnetic vibration of the driving motor and the dynamic meshing excitation of the transmission gears, it is very prone to electromechanical coupling vibration. The superimposed vibration frequency is a significant challenge to vibration source location and vibration suppression. An investigation was performed to gain insights into the operation of a coal mining machine’s transmission system during problematic conditions. First, a math model for controlling the permanent magnet motor is built to find the control settings. The decoupling mechanism of motor current was revealed. Subsequently, time-varying mesh stiffness models were developed for both normal operating conditions and gear spalling fault scenarios. The changes in how stiffness varies over time when there are spalling faults were explained. Finally, considering factors such as gear error, tooth side clearance, and bearing stiffness, the dynamic model of the cutting part of the coal shearer was established. The influence laws of internal excitation, load impact, and spalling failure on the dynamic characteristics of the system’s electromechanical coupling were studied, and the typical features of gear spalling failure were extracted. The study’s results are crucial for comprehending how to assess the state and minimize vibrations in the transmission systems of coal mining equipment.
- 1 January 2026
- Published by Canadian Science Publishing in Transactions of the Canadian Society for Mechanical Engineering
- Vol. 50, 1-14
Dynamic characteristics of electromechanical coupling in the cutting drive system of coal shearer with spalling fault
Due to the electromagnetic vibration of the driving motor and the dynamic meshing excitation of the transmission gears, it is very prone to electromechanical coupling vibration. The superimposed vibration frequency is a significant challenge to vibration source location and vibration suppression. An investigation was performed to gain insights into the operation of a coal mining machine’s transmission system during problematic conditions. First, a math model for controlling the permanent magnet motor is built to find the control settings. The decoupling mechanism of motor current was revealed. Subsequently, time-varying mesh stiffness models were developed for both normal operating conditions and gear spalling fault scenarios. The changes in how stiffness varies over time when there are spalling faults were explained. Finally, considering factors such as gear error, tooth side clearance, and bearing stiffness, the dynamic model of the cutting part of the coal shearer was established. The influence laws of internal excitation, load impact, and spalling failure on the dynamic characteristics of the system’s electromechanical coupling were studied, and the typical features of gear spalling failure were extracted. The study’s results are crucial for comprehending how to assess the state and minimize vibrations in the transmission systems of coal mining equipment.
- 24 December 2025
- Published by IOP Publishing in Engineering Research Express
- Vol. 7 (4) , 0455d9
A lightweight ghost-driven attention network for wire bond positioning in measurement
Wire bonding is one of the key interconnection processes in semiconductor packaging. As the shrinking of feature size, it becomes challenging to measure the critical dimension precisely online in wire bonding. To achieve this goal, a lightweight ghost-driven attention network (GDA-UNet) was proposed for positioning the regions of interest (ROIs) in measurement of wire bonding. The Ghost module V1 was adopted to reduce unnecessary computational load. And the gated channel spatial attention (GCSA) module can highlight the key features with extremely low overhead. The experimental results show that GDA-UNet achieves high-precision with a mIoU of 96.04% and a Dice coefficient of 97.95%. Moreover, it reduces the number of parameters from 31.03M to 4.44M and FLOPs from 218.63G to 46.04G. The proposed approach achieves the best balance between accuracy and computational parsimony, which provides a new solution for automatic optical inspection of electronic packaging processes.
- 24 December 2025
- Published by IOP Publishing in Engineering Research Express
- Vol. 7 (4) , 0455d9
A lightweight ghost-driven attention network for wire bond positioning in measurement
Wire bonding is one of the key interconnection processes in semiconductor packaging. As the shrinking of feature size, it becomes challenging to measure the critical dimension precisely online in wire bonding. To achieve this goal, a lightweight ghost-driven attention network (GDA-UNet) was proposed for positioning the regions of interest (ROIs) in measurement of wire bonding. The Ghost module V1 was adopted to reduce unnecessary computational load. And the gated channel spatial attention (GCSA) module can highlight the key features with extremely low overhead. The experimental results show that GDA-UNet achieves high-precision with a mIoU of 96.04% and a Dice coefficient of 97.95%. Moreover, it reduces the number of parameters from 31.03M to 4.44M and FLOPs from 218.63G to 46.04G. The proposed approach achieves the best balance between accuracy and computational parsimony, which provides a new solution for automatic optical inspection of electronic packaging processes.
- 21 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Components, Packaging and Manufacturing Technology
- Vol. 16 (6) , 1321-1331
A Novel Annular-Dielectric-Layer-Structured TSV in Silicon Interposer for 2.5-D/3-D Integration
Silicon interposer is extensively used in 2.5D/3D electronic packaging, which has ultra-high densities of Redistribution Layers (RDLs) and Through Silicon Vias (TSVs), but the signal integrity degrades significantly due to the combined effects of high transmission loss and enhanced mutual coupling in ultra-dense interconnects. A novel annular dielectric layer (ADL) TSV structure was proposed to address the signal integrity challenges encountered in conventional silicon interposer. In ADL-TSV structure, low-k dielectric fillers are added around the signal TSVs, which may reduce the effective dielectric constant between TSVs, thereby diminishes capacitive coupling among TSVs. It also increases their characteristic impedance and signal propagation speed, while the associated risk of signal delay remains well within controllable limits. To thoroughly investigate the impact of the ADL on signal integrity, an equivalent circuit model based on RLGC parameters was established. Compared to the conventional TSV, ADL-TSV reduced the return loss (RL) by 0.75 dB, insertion loss (IL) by 0.031 dB, and far-end crosstalk (FEXT) by 4.78 dB at the signal frequency of 45 GHz. In the eye diagram, ADL-TSV is larger than conventional TSV by 0.003 V and 0.12 ps in eye height and eye width respectively. Furthermore, the performance improvement achieved by the ADL is more pronounced under small-pitch TSV configurations. The results show that the proposed ADL-TSV has obvious advantages in signal transmission quality at high-frequency, and thus can be applied to high-performance, high-reliability heterogeneous integration systems.
- 21 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Components, Packaging and Manufacturing Technology
- Vol. 16 (6) , 1321-1331
A Novel Annular-Dielectric-Layer-Structured TSV in Silicon Interposer for 2.5-D/3-D Integration
Silicon interposer is extensively used in 2.5D/3D electronic packaging, which has ultra-high densities of Redistribution Layers (RDLs) and Through Silicon Vias (TSVs), but the signal integrity degrades significantly due to the combined effects of high transmission loss and enhanced mutual coupling in ultra-dense interconnects. A novel annular dielectric layer (ADL) TSV structure was proposed to address the signal integrity challenges encountered in conventional silicon interposer. In ADL-TSV structure, low-k dielectric fillers are added around the signal TSVs, which may reduce the effective dielectric constant between TSVs, thereby diminishes capacitive coupling among TSVs. It also increases their characteristic impedance and signal propagation speed, while the associated risk of signal delay remains well within controllable limits. To thoroughly investigate the impact of the ADL on signal integrity, an equivalent circuit model based on RLGC parameters was established. Compared to the conventional TSV, ADL-TSV reduced the return loss (RL) by 0.75 dB, insertion loss (IL) by 0.031 dB, and far-end crosstalk (FEXT) by 4.78 dB at the signal frequency of 45 GHz. In the eye diagram, ADL-TSV is larger than conventional TSV by 0.003 V and 0.12 ps in eye height and eye width respectively. Furthermore, the performance improvement achieved by the ADL is more pronounced under small-pitch TSV configurations. The results show that the proposed ADL-TSV has obvious advantages in signal transmission quality at high-frequency, and thus can be applied to high-performance, high-reliability heterogeneous integration systems.
- Open Access
- 12 November 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (11) , 408
Wafer Defect Detection Technology Based on CTM-IYOLOv10 Network
The continuous scaling of semiconductor devices has increased the density and complexity of wafer dies, making precise and efficient defect detection a critical task for intelligent manufacturing. Traditional manual or semi-automated inspection approaches are often inefficient, error-prone, and susceptible to missed or false detections, particularly for small or irregular defects. This study presents a wafer defect detection framework that integrates clustering–template matching (CTM) with an improved YOLOv10 network (CTM-IYOLOv10). The CTM strategy enhances die segmentation efficiency and mitigates redundant matching in multi-die fields of view, while the introduction of a modified GhostConv module and an enhanced BiFPN structure strengthens feature representation, reduces computational redundancy, and improves small-object detection. Furthermore, data augmentation strategies are employed to improve robustness and generalization. Experimental evaluations demonstrate that CTM-IYOLOv10 achieves a detection accuracy of 98.1%, reduces inference time by 23.2%, and compresses model size by 52.3% compared with baseline YOLOv10, and consistently outperforms representative detectors such as YOLOv5 and YOLOv8. These results highlight both the methodological contributions of the proposed architecture and its practical significance for real-time wafer defect inspection in semiconductor manufacturing.
- 12 November 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (11) , 408
Open AccessWafer Defect Detection Technology Based on CTM-IYOLOv10 Network
The continuous scaling of semiconductor devices has increased the density and complexity of wafer dies, making precise and efficient defect detection a critical task for intelligent manufacturing. Traditional manual or semi-automated inspection approaches are often inefficient, error-prone, and susceptible to missed or false detections, particularly for small or irregular defects. This study presents a wafer defect detection framework that integrates clustering–template matching (CTM) with an improved YOLOv10 network (CTM-IYOLOv10). The CTM strategy enhances die segmentation efficiency and mitigates redundant matching in multi-die fields of view, while the introduction of a modified GhostConv module and an enhanced BiFPN structure strengthens feature representation, reduces computational redundancy, and improves small-object detection. Furthermore, data augmentation strategies are employed to improve robustness and generalization. Experimental evaluations demonstrate that CTM-IYOLOv10 achieves a detection accuracy of 98.1%, reduces inference time by 23.2%, and compresses model size by 52.3% compared with baseline YOLOv10, and consistently outperforms representative detectors such as YOLOv5 and YOLOv8. These results highlight both the methodological contributions of the proposed architecture and its practical significance for real-time wafer defect inspection in semiconductor manufacturing.