QiMeng-CRUX: Narrowing the Gap between Natural Language and Verilog via Core Refined Understanding eXpression

2025-11-25

AI Summary

Existing natural language (NL)–to–Verilog synthesis methods suffer from ambiguous and unstructured NL inputs, resulting in a large semantic gap and low generation accuracy. Method: We propose CRUX (Core Refined Understanding eXpression), a structured intermediate representation that explicitly models user intent as a hardware semantic graph to bridge the semantic gap between NL and Verilog. We further design a two-stage joint training framework: Stage I learns the NL→CRUX mapping, while Stage II optimizes CRUX→Verilog generation and supports cross-model transfer. Contribution/Results: Our approach achieves state-of-the-art performance among general-purpose large language models on multiple Verilog synthesis benchmarks, with particularly significant improvements on complex circuit design tasks. Moreover, CRUX serves as a plug-and-play enhancement module for other code generation models, demonstrating strong generalizability and practical utility.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Large language models (LLMs) have shown promising capabilities in hardware description language (HDL) generation. However, existing approaches often rely on free-form natural language descriptions that are often ambiguous, redundant, and unstructured, which poses significant challenges for downstream Verilog code generation. We treat hardware code generation as a complex transformation from an open-ended natural language space to a domain-specific, highly constrained target space. To bridge this gap, we introduce Core Refined Understanding eXpression (CRUX), a structured intermediate space that captures the essential semantics of user intent while organizing the expression for precise Verilog code generation. We further design a two-stage training framework, comprising Joint Expression Modeling and Dual-Space Optimization, to enhance the quality of both CRUX and Verilog code. Experiments across multiple Verilog generation benchmarks demonstrate that our model, CRUX-V, achieves state-of-the-art performance among general models, particularly under challenging design tasks. Furthermore, the CRUX space proves transferable and beneficial when used as input prompts for other code models, highlighting its effectiveness in narrowing the gap between free-form natural language descriptions and precise Verilog generation.

Problem

• Bridging ambiguous natural language and precise Verilog generation
• Creating structured intermediate representation for hardware semantics
• Improving code quality via dual-space optimization framework

Innovation

• Introducing structured CRUX intermediate representation space
• Implementing two-stage training with joint modeling
• Optimizing dual-space for Verilog code quality