DeepCircuitX: Repository-Level RTL Dataset for Code Understanding, Generation, and Multimodal Analysis

Zeju Li, Changran XU, Zhengyuan Shi, Zedong Peng, Yi Liu, Yunhao Zhou, Lingfeng Zhou, Chengyu Ma, Jianyuan Zhong, Xi Wang, Jieru Zhao, Zhufei Chu, Xiaoyan Yang, Qiang Xu

Keywords: Multi-modal dataset of EDA, AI for EDA, LLM for RTL, Circuit Learning

Abstract:

This paper introduces DeepCircuitX, a comprehensive multimodal dataset designed to advance RTL code understanding, generation, and completion tasks in hardware design automation. Unlike existing datasets, which focus either on file-level RTL code or downstream netlist and layout data, DeepCircuitX spans repository, file, module, and block-level RTL code, providing a more holistic resource for training and evaluating large language models (LLMs). The dataset is enriched with Chain of Thought (CoT) annotations that offer detailed functionality and structure descriptions at multiple levels, enhancing its utility for RTL code understanding, generation, and completion.

In addition to RTL data, DeepCircuitX includes synthesized netlists and power-performance-area (PPA) metrics, allowing for early-stage design exploration and PPA prediction directly from RTL code. We establish comprehensive benchmarks for RTL code understanding, generation, and completion using open-source models such as CodeLlama, CodeT5+, and CodeGen, demonstrating substantial improvements in task performance. Furthermore, we introduce and evaluate models for PPA prediction, setting new benchmarks for RTL-to-PPA analysis. We conduct human evaluations and reviews to confirm the high quality and functionality of the generated RTL code and annotations. Our experimental results show that DeepCircuitX significantly improves model performance across multiple benchmarks, underscoring its value as a critical resource for advancing RTL code tasks in hardware design automation.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7380