SemiFA: An Agentic Multi-Modal Framework for Autonomous Semiconductor Failure Analysis Report Generation

Shivam Chand Kaushik's SemiFA framework represents a breakthrough in semiconductor manufacturing automation, transforming a process that typically consumes hours of expert engineering time into a sub-minute automated workflow. The system employs an agentic architecture built on LangGraph, where specialized AI agents handle different aspects of failure analysis: a DefectDescriber classifies and narrates defect morphology using DINOv2 and LLaVA-1.6 models, a RootCauseAnalyzer fuses SECS/GEM equipment telemetry with historical defect data from a Qdrant vector database, a SeverityClassifier assesses impact, and a RecipeAdvisor proposes corrective actions.

Trained on the SemiFA-930 dataset containing 930 annotated semiconductor defect images across nine classes, the framework achieves 92.1% accuracy on validation images with a macro F1 score of 0.917. Crucially, the multi-modal approach—combining visual inspection with equipment telemetry—improves root cause reasoning by +0.86 composite points on a 1-5 scale compared to image-only analysis. The complete pipeline generates structured PDF reports in just 48 seconds when running on an NVIDIA A100-SXM4-40GB GPU, marking what the researchers claim is the first integration of SECS/GEM equipment telemetry into a vision-language model pipeline for autonomous failure analysis.