Agentic Vision AI for Smart Inspection
Vision technology plays a critical role in semiconductor manufacturing, where high-speed, high-precision inspection is essential for detecting subtle defects, maintaining process quality, and improving yield. AI-powered vision has significantly advanced these capabilities. However, in our work with many partners, we have found that adapting vision systems to new products, processes, and defect types can still require substantial effort, particularly in data preparation, augmentation, and curation. Recent advances in agentic AI are creating new possibilities for vision systems. Beyond perception, these systems can reason, use tools, adapt to evolving requirements, and learn from real-world feedback.
This presentation explores how NVIDIA is bringing agentic capabilities to vision model development and evolving its tools for semiconductor manufacturing. Through practical inspection use cases, we will demonstrate how agentic workflows and data-enhanced fine-tuning can improve performance on downstream inspection tasks. We will also share how these capabilities can accelerate model customization and deployment while continuously improving inspection workflows through closed-loop feedback from production environments.
Attendees will gain practical insights into how Agentic Vision AI can help semiconductor manufacturers build smarter, more accurate, and more scalable inspection and quality-control operations.
Key Technologies Covered
- Vision model training, fine-tuning, optimization, and deployment
- Data Enhanced Fine-Tuning, data mining, and gap analysis
- Digital Twin to Synthetic Data Generation
- Data Augmentation
- Semiconductor AOI, defect classification, and vision-model data flywheels
