IncreRTL: Traceability-Guided Incremental RTL Generation under Requirement Evolution
Summary: arXiv:2603.25769v1 Announce Type: cross
Abstract: Large language models (LLMs) have shown promise in generating RTL code from natural-language descriptions, but existing methods remain static and struggle to adapt to evolving design requirements, potentially causing structural drift and costly full regeneration. We propose IncreRTL, a LLM-driven framework for incremental RTL generation under requirement evolution. By constructing requirement-code traceability links to locate and regenerate affected code segments, IncreRTL achieves accurate and consistent updates. Evaluated on our newly constructed EvoRTL-Bench, IncreRTL demonstrates notable improvements in regeneration consistency and efficiency, advancing LLM-based RTL generation toward practical engineering deployment.
Introduction
The field of hardware design increasingly relies on the capabilities of large language models to convert natural-language specifications into Register Transfer Level (RTL) code. However, as design requirements evolve, existing methods often fall short, leading to inefficiencies and potential errors in the generated code. In response to these challenges, the IncreRTL framework introduces a novel approach that maintains traceability between requirements and the corresponding code segments.
The Challenge of Requirement Evolution
As projects advance, design requirements are frequently updated, necessitating modifications to the RTL codebase. Traditional methods of RTL code generation are largely static, meaning they regenerate the entire codebase in response to even minor changes. This full regeneration can lead to:
- Structural drift, where the updated code diverges from the intended design.
- Increased development time and resource consumption.
- Heightened risk of introducing errors during the regeneration process.
Introducing IncreRTL Framework
IncreRTL offers a solution to these challenges by utilizing a traceability-guided approach. The framework constructs links between requirements and the code, enabling it to identify which segments of the RTL code need to be regenerated when requirements change. This targeted regeneration process enhances:
- Consistency: Ensures that the updated code remains aligned with the specified requirements.
- Efficiency: Minimizes the amount of code that needs to be processed, reducing time and resource expenditure.
- Accuracy: Reduces the likelihood of errors by focusing only on the impacted code segments.
Evaluation and Results
To evaluate the effectiveness of IncreRTL, the authors developed the EvoRTL-Bench, a benchmark suite designed to assess the performance of incremental RTL generation methods. The results demonstrated that IncreRTL not only improved regeneration consistency but also significantly enhanced efficiency compared to traditional methods. Key findings include:
- A marked reduction in operational time for code regeneration.
- A higher degree of accuracy in the generated RTL code.
- Increased developer confidence in the reliability of the updates.
Conclusion
IncreRTL represents a significant advancement in the field of RTL generation, addressing the shortcomings of static methods and paving the way for more dynamic approaches. By facilitating traceability and targeted regeneration, IncreRTL stands to revolutionize how engineers adapt to evolving design requirements, ultimately enhancing the practicality of LLM-based RTL generation in real-world engineering applications.