An Integrated AI Platform for Accelerating Semiconductor Failure Analysis: Combining Similarity Search, Deep-Learning Image Retrieval, Automated Labeling, and Ontology-Guided Workflow Assistance

Semiconductor failure analysis (FA) requires rapid access to relevant historical cases, consistent failure labeling, image-based pattern recognition, and methodologically correct investigation planning. In practice, these tasks are often supported by disconnected tools, leading to workflow friction and delayed root-cause identification. This paper presents AiFAct, an integrated, on-premise AI platform that supports the FA workflow end-to-end through six coordinated capabilities: multi-criteria similarity search with pair-specific reranking feedback, ontology-based automatic label prediction, quiz-based human verification for continuous model refinement, on-premise large language model (LLM) support for structured summaries and next-step proposals, deep-embedding image similarity for microscopy retrieval, and process-ontology-constrained workflow guidance. Unlike point solutions, the platform is designed as a closed feedback system where analyst interactions continuously improve ranking, labeling, and recommendation quality during routine work. In production-scale deployment across multi-site FA data, core candidate retrieval and reranking remains sub-second for candidate pools above 50,000 jobs, and analyst-facing query responses are delivered in under five seconds. The key contribution is workflow-level orchestration and feedback coupling rather than a single new standalone model. This architecture improves analyst trust through transparent scoring, supports knowledge transfer across sites, and preserves data sovereignty via fully on-premise operation.