Our Automated Microscope for Semiconductor Inspection
Technology · Photonics · Semiconductor Inspection
Our laboratory is equipped with an automated optical microscope capable of simultaneous operation in the visible (VIS) and shortwave infrared (SWIR) spectral ranges. Developed in collaboration with industry partners, the system is specifically designed for automated inline quality control in semiconductor chip manufacturing, combining multi-objective optics, motorised scanning stages, and Machine Learning-based defect classification in a single integrated platform.
Two spectral windows, one platform
What makes this system technically distinctive is its dual-band imaging capability. Conventional optical microscopes operate exclusively in the visible spectrum (roughly 400–700 nm). This instrument extends imaging into the SWIR range (approximately 900–1700 nm), a spectral region where silicon and other semiconductor materials become partially or fully transparent. As a result, defects and structural features that are completely invisible under standard white-light illumination — such as subsurface cracks, buried layers, or internal bonding interfaces — can be detected without any physical sectioning or destructive preparation of the sample. Furthermore, specifically for the SWIR band, the microscope has been tested with high-speed hyperspectral line-scan cameras with high spectral resolution, enabling increased contrast for defects with relevant spectral signatures within that band.
This dual-band approach means that a single inspection pass can simultaneously capture surface morphology in VIS and subsurface structural information in SWIR, significantly increasing the information density per measurement and reducing total inspection time.
Full automation and machine learning integration
The microscope is equipped with high-precision motorised XYZ stages and automatic focus control, enabling fully automated raster scanning of large sample areas without operator intervention. The system is capable of stitching high-resolution images from individual fields of view into a complete map of the inspected surface, even when the area to be scanned is orders of magnitude larger than the defects being sought — a common challenge in semiconductor quality control.
Beyond image acquisition, the platform integrates Machine Learning algorithms for defect classification. These models can be trained on labelled image datasets to automatically identify and categorise anomalies such as contamination particles, scratches, lithographic defects, or structural irregularities. Once trained, the classifier operates in real time during the scan, flagging regions of interest and assigning defect categories without human review of every image frame.
Industrial context and availability
Manual chip inspection using conventional microscopes — whether optical or electron-based — is inherently limited by human attention and throughput. Scanning a full wafer or production batch at the resolution required to detect micron-scale defects demands hours of operator time per lot, and the repetitive nature of the task introduces fatigue-related error rates that are difficult to control. Automated systems address both constraints simultaneously.
Despite the clear technical need, the market for fully automated, customisable inspection systems tailored to the specific requirements of small and mid-sized chip manufacturers remains limited. This system was developed in collaboration with industry partners precisely to address that gap, combining optics, automation, and AI in a single integrated platform.
https://www.linkedin.com/feed/update/urn:li:activity:7472230913439895552