SpecLens: LLM-Based Verilog Generation with Specification-Derived Constraints via Behavioral Divergence

Large language models (LLMs) have recently shown promise in Verilog generation, but producing functionally correct RTL directly from natural-language specifications remains a highly challenging task. Existing approaches improve LLM-based Verilog generation mainly with retrieval-augmented generation