一家知名半導體封裝廠於其IC載板製程中，導入先進自動光學檢測（AI AOI）系統，應用於二階段檢測，以優化自動視覺檢測（AVI）後的瑕疵篩選作業。這套AI AOI解決方案結合由和碩開發的全流程AI平台PEGAAi，以及東擎科技（ASRock Industrial）工業級強固邊緣AIoT系統平台iEPF‑9030S，並運用邊緣AI進行異常判斷，大幅減少人工複檢的負擔，提升檢測效率並穩定製程良率。此系統展現AI軟體與高性能硬體的強大協同效益，加速半導體智慧製造發展進程。目前全球逾13座和碩廠區已部署PEGAAi平台，累計應用規模達400至500個生產站點，充分展現擴充彈性與實際應用價值。
To address these challenges, the leading global semiconductor packaging provider implemented a comprehensive AI-driven inspection solution that integrates PEGATRON’s PEGAAi platform with ASRock Industrial’s iEPF-9030S Robust Edge AIoT Platform to enhance IC substrate quality control. The AI station, powered by PEGAAi platform and iEPF-9030S, performs second-stage screening between the Auto Vision Inspection (AVI) and Vision Repair System (VRS) stages, automatically classifying and filtering out false positives to minimize unnecessary human intervention.
ASRock Industrial’s iEPF-9030S, equipped with Intel® 14th Gen Core™ processors and NVIDIA RTX 5000 ADA GPUs, provides the high computing power required to execute AI model training on-site and perform real-time AI inference of defects inspection. The solution further incorporates Auto MLOps workflows for model optimization and AI performance dashboards, enabling continuous feedback from operator re-inspection to refine the AI model. This end-to-end integration not only streamlines the inspection process but also ensures adaptability, scalability, and sustained accuracy in demanding semiconductor production environments.
The implementation of PEGATRON’s PEGAAi platform in combination with ASRock Industrial’s iEPF-9030S Robust Edge AIoT Platform for IC substrate inspection has delivered measurable improvements in inspection accuracy, labor efficiency, and system scalability within high-volume semiconductor manufacturing environments.
By introducing AI-powered second screening before operator involvement, the system reduced excessive false positives generated by AVI machines. This directly lowered the need for labor-intensive visual re-inspection, cutting manual workload by 50-70% and improving workforce allocation.
Through accurate anomaly classification and continuous AI model refinement, the solution minimized overkill incidents and false defect identifications. This led to improved defect detection precision, enhanced yield stability, and more consistent product quality.
The integration of high-performance edge computing enabled real-time inference and on-site retraining, solving delays caused by disconnected or centralized inspection systems. This improved responsiveness and supported faster, more intelligent decision-making on the production line.
This improved responsiveness and supported faster, more intelligent decision-making on the production line, leading to reduced production delays and optimized process control.
