Four Defect Categories Assessed Quickly and Accurately by AI OKI has developed an AI-based visual inspection technology (hereinafter "this technology") that reduces false detection of solder defects identified during AOI (Automated Optical Inspection) (Note 1) after component mounting on PCBs (printed circuit boards), cutting post-AOI visual inspection time by approximately 80%. Deployment will begin on July 1, 2026, on production lines serving customers of the "Marugoto EMS" service. By training the AI exclusively on good product data (Note 2) and combining proprietary algorithms, this technology shortens visual inspection time while improving inspection accuracy, thereby enhancing customer service quality. In recent years, driven by the rapid advancement and proliferation of AI, semiconductors mounted on PCBs have become increasingly larger, finer, and multi-layered. In AI servers in particular, numerous large components and ultra-fine pitch components with terminal spacing of just a few microns are mounted, making solder defect detection significantly more challenging. The newly developed technology integrates OKI's manufacturing expertise for large, high-density PCBs into the AOI system's inspection algorithms, implementing a dedicated AI model for mounted boards. It uses AI to inspect four key defect categories—"solder wetting" (Note 3), "misalignment," "missing components," and "lifting"—enabling fast and highly accurate judgment. Additionally, the system includes a proprietary program that recognizes component-specific characteristics, treating variations such as manufacturing or lot numbers and minor printing smudges—factors that do not affect functionality—as acceptable good products. Furthermore, by adopting a "good product-only learning" approach, which trains the AI using only a small volume of good product data, OKI has optimized the AI technology for high-mix, low-volume production involving tens of thousands of component types and thousands of equipme