2025.12. ~ present
Shenzhen Baolian Artificial Intelligence Technology Co., Ltd
Software development for silicon photonic chip four-sided inspection equipment (ongoing)
Participated in the development of software system for silicon photonic chip inspection equipment; the project is still ongoing. Responsible for building software architecture and engineering delivery chains for the four inspection surfaces: Top, AR, HR, and Bottom, covering modules such as central release, algorithm package management, recipe version management, online detection deployment package, offline re-inspection deployment package, and algorithm debugging tools. The central end is based on WPF/MVVM, ASP.NET Core Web API, EF Core, and SQLite to implement management of models, defect dictionaries, inspection items, algorithm packages, recipe versions, and release packages; On the algorithm side, comprehensive algorithm packages are designed at the surface level, supporting the packaging of C#/.NET plugin DLLs, HALCON runtime environments, package-manifest.json metadata, models/templates/configurations, and other artifacts, and generating online/offline workstation deployment packages that are importable, traceable, and verifiable through the release process. Currently, the Top Plane integrated inspection plugin and debugging workbench have been promoted, focusing on standard coordinate templates, chip body positioning, ROI projection, defect distribution, and unified inspection result output, supporting integrated verification of detection items such as twining, chipping, metal ID defects, cut offsets, epitaxial defects, foreign object appearances, groove anomalies, substrate leakage, residual gold, scratches, and overflowing plating.
Gerber file parsing and Mark/Bump data export plugin
Supporting other project teams in developing Gerber file reading and data export tools to meet the extraction needs of Mark points, bump arrays, and rectangular areas in PCB/semiconductor inspection projects. Gerber RS-274X file parsing based on C++, supporting %FS coordinate format, mm/inch unit recognition and conversion, D-size Aperture parameter configuration, D03 flash circular exposure point extraction, rectangular aperture region extraction, and duplicate point deduplication; Ultimately, MarkList, BumpList, and RectangleList are output according to the agreed JSON structure for subsequent HALCON/visual inspection workflows to generate ROI, positioning benchmarks, and inspection object data, reducing the cost of manual Gerber data organization by project teams and improving the efficiency of recipe generation.
UVI Dispensing Online Inspection System Validation Tool
Based on C# WPF, . NET 8 and HALCON 20.11 have developed UVI online dispensing and verification tools for online visual inspection scenarios after flexible board soldering and before glue curing. The system builds a workflow for line scan image simulation acquisition, integrity state machine, frame buffer splicing, ROI teaching recipes, Mark positioning, HALCON detection algorithms, and result visualization processes centered on requirements such as detection of phosphor colloid, glue coverage area assessment, and whether glue exceeds board edges; It also encapsulates the MVSDK camera access capability, supporting both simulated image and real line scan camera verification paths, providing a software validation foundation for online detection solution evaluation, algorithm debugging, and equipment selection within 25 seconds of CT.
AOI OpenCV checkerboard automatic calibration extension plugin
Supports other project teams in developing AOI camera automatic calibration extensions, engineering and packaging the original Python pre-research process into C++ OpenCV DLLs, and integrating the upper computer plugin system via C#/Prism modules. Functions cover checkerboard camera distortion calibration, de-distortion verification, affine calibration from pixel coordinates to platform coordinates, exposure/gain parameter tuning evaluation, multi-field PCB image splicing, and measurement conversion from stitched images to platform coordinates; It also provides WinForms verification demos and operation documentation, making it easy for project teams to calibrate on-site according to the workflow of "capture checkerboard image -> generate camera_calib.json -> fitting pixel_stage_map.json -> multi-field stitching/measurement." This extension reduces the integration costs of camera calibration, platform coordinate conversion, and multi-field stitching in AOI projects, providing a reusable calibration toolchain for subsequent PCB inspection, ROI positioning, and workstation vision tuning.