Optical Character Recognition on Integrated Circuits
Read curved strings, low contrast characters, and deformed, skewed, and poorly etched codes
After going through the strenuous process of the package test, the semiconductor chip finally gets its identification number which contains both manufacturer information and the integrated circuit’s technical specifications. This alphanumeric code is stamped on the top surface of the IC. The readability of this information is critical to the semiconductor manufacturer for internal and external traceability. Downstream parties such as system design houses and board assembly companies read these codes to verify both that they received the correct chips and the right chips were attached to the PCB. Identifying and confirming this information is traditionally done by rule-based machine vision. However, traditional algorithms struggle reading the extremely small and variable text strings that are laser marked or chemically etched onto ICs. Other common problems that affect readability are highly textured surfaces and ambient lamination, which deform the characters in the image.
Cognex Deep Learning technology addresses the challenges that cannot be solved with rule-based image processing technology. The Cognex Deep Learning OCR tool reads curved strings, low contrast characters, and deformed, skewed, and poorly etched codes using a built-in library that is pre-trained with a thousand characters. The OCR tool also provides re-train capability, so users can solve new or specific characters that were not automatically identified the first time around. Reading a chip’s identification number quickly and accurately improves traceability and ensures the correct information is captured and makes it available if needed in the future.