WaferSAGE uses LLMs to analyze semiconductor defects with synthetic data

Researchers have developed WaferSAGE, a framework utilizing a 4B-parameter Qwen3-VL model for visual question answering on wafer defects in semiconductor manufacturing. The system addresses data scarcity by employing a three-stage synthetic data generation pipeline guided by structured rubrics. This approach allows for precise evaluation and covers defect identification, spatial distribution, morphology, and root cause analysis, enabling on-premise deployment and cost-effective solutions. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Demonstrates how smaller, domain-specific models can achieve high performance in specialized industrial tasks, enabling privacy-preserving on-premise deployments.

RANK_REASON This is a research paper detailing a new framework and model for a specific industrial application.

Read on arXiv cs.AI →

• WaferSAGE
• Qwen3-VL
• Gemini-3-Flash
• arXiv

• paper
• model release