| Abstract Scope |
Characterizing two-dimensional materials demands months of specialist training and remains error-prone even for experts—bottlenecks that impede materials discovery. We present two foundation model-based frameworks that eliminate these barriers. ATOMIC integrates the Segment Anything Model with GPT-based reasoning, topological analysis, and automated microscope control for zero-shot optical characterization across graphene, MoS₂, WSe₂, and SnSe—achieving 99.7% accuracy in monolayer identification without task-specific training, while resolving grain boundaries invisible to human observers. RamanNet applies LLM peak analysis and GPT-4o classification with ELO scoring to enable autonomous Raman spectra interpretation from minimal data. Together, these frameworks demonstrate that foundation models can serve as generalizable, modality-agnostic characterization engines, collapsing specialist onboarding into deployable pipelines. These results establish a scalable paradigm accelerating autonomous materials characterization and discovery. |