Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

REvolution: An Evolutionary Framework for RTL Generation driven by Large Language Models

Kyungjun Min, Kyumin Cho, Junhwan Jang, Seokhyeong Kang

TL;DR

REvolution tackles the dual challenge of producing RTL code that is both functionally correct and optimized for power, performance, and area. It fuses Large Language Models with Evolutionary Computation to explore a broad design space in parallel, using a dual-population scheme and adaptive prompt strategies. Across VerilogEval and RTLLM benchmarks, REvolution raises initial pass rates and achieves final passes up to 95.5% with notable PPA improvements, all without fine-tuning or bespoke domain tools. This work demonstrates that combining generative AI with evolutionary search provides a practical, scalable path to high-quality RTL design.

Abstract

Large Language Models (LLMs) are used for Register-Transfer Level (RTL) code generation, but they face two main challenges: functional correctness and Power, Performance, and Area (PPA) optimization. Iterative, feedback-based methods partially address these, but they are limited to local search, hindering the discovery of a global optimum. This paper introduces REvolution, a framework that combines Evolutionary Computation (EC) with LLMs for automatic RTL generation and optimization. REvolution evolves a population of candidates in parallel, each defined by a design strategy, RTL implementation, and evaluation feedback. The framework includes a dual-population algorithm that divides candidates into Fail and Success groups for bug fixing and PPA optimization, respectively. An adaptive mechanism further improves search efficiency by dynamically adjusting the selection probability of each prompt strategy according to its success rate. Experiments on the VerilogEval and RTLLM benchmarks show that REvolution increased the initial pass rate of various LLMs by up to 24.0 percentage points. The DeepSeek-V3 model achieved a final pass rate of 95.5\%, comparable to state-of-the-art results, without the need for separate training or domain-specific tools. Additionally, the generated RTL designs showed significant PPA improvements over reference designs. This work introduces a new RTL design approach by combining LLMs' generative capabilities with EC's broad search power, overcoming the local-search limitations of previous methods.

REvolution: An Evolutionary Framework for RTL Generation driven by Large Language Models

TL;DR

REvolution tackles the dual challenge of producing RTL code that is both functionally correct and optimized for power, performance, and area. It fuses Large Language Models with Evolutionary Computation to explore a broad design space in parallel, using a dual-population scheme and adaptive prompt strategies. Across VerilogEval and RTLLM benchmarks, REvolution raises initial pass rates and achieves final passes up to 95.5% with notable PPA improvements, all without fine-tuning or bespoke domain tools. This work demonstrates that combining generative AI with evolutionary search provides a practical, scalable path to high-quality RTL design.

Abstract

Large Language Models (LLMs) are used for Register-Transfer Level (RTL) code generation, but they face two main challenges: functional correctness and Power, Performance, and Area (PPA) optimization. Iterative, feedback-based methods partially address these, but they are limited to local search, hindering the discovery of a global optimum. This paper introduces REvolution, a framework that combines Evolutionary Computation (EC) with LLMs for automatic RTL generation and optimization. REvolution evolves a population of candidates in parallel, each defined by a design strategy, RTL implementation, and evaluation feedback. The framework includes a dual-population algorithm that divides candidates into Fail and Success groups for bug fixing and PPA optimization, respectively. An adaptive mechanism further improves search efficiency by dynamically adjusting the selection probability of each prompt strategy according to its success rate. Experiments on the VerilogEval and RTLLM benchmarks show that REvolution increased the initial pass rate of various LLMs by up to 24.0 percentage points. The DeepSeek-V3 model achieved a final pass rate of 95.5\%, comparable to state-of-the-art results, without the need for separate training or domain-specific tools. Additionally, the generated RTL designs showed significant PPA improvements over reference designs. This work introduces a new RTL design approach by combining LLMs' generative capabilities with EC's broad search power, overcoming the local-search limitations of previous methods.
Paper Structure (17 sections, 4 equations, 5 figures, 2 tables)

This paper contains 17 sections, 4 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. LLMs for RTL Generation
  4. Integrating LLMs and Evolutionary Computation
  5. Methodology
  6. Overview
  7. Evolutionary Representation
  8. Dual-Population Algorithm
  9. Prompt Strategies
  10. Adaptive Prompt Strategy Selection
  11. The REvolution Algorithm
  12. Experimental Results
  13. Experimental Setup
  14. Performance Analysis and Comparison
  15. Case Study: 8-bit Signed Adder
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Overall framework of the REvolution. The framework initializes a population from a specification (functional description) and refines it through an evolutionary loop of offspring generation, evaluation, and survivor selection until termination.
  • Figure 2: Structure of Individual. Individual is a (Thought, Code, Feedback) tuple, with Feedback generated based on the Code's functional correctness.
  • Figure 3: Example of an Evolutionary Prompt. The available prompt strategies vary depending on the population, and each prompt is constructed from a general template.
  • Figure 4: Evolutionary Trajectory of DeepSeek-V3 on VerilogEval Prob033 problem.
  • Figure 5: Power-area distribution of design candidates in the evolutionary process of DeepSeek-V3 on VerilogEval Prob033.