Weimin Fu

LLMs generating hardware code often fail *after* synthesis, and the type of failure (elaboration errors vs. missing wrappers) systematically depends on whether the model is proprietary or open-weight.