Elamur Research⌕
☰
home / papers / 2026 / SYMDIREC: A Neuro-Symbolic Divide-Retrieve-Conquer Framework for Enhanced RTL Synthesis and Summarization

SYMDIREC: A Neuro-Symbolic Divide-Retrieve-Conquer Framework for Enhanced RTL Synthesis and Summarization

Paper · 2026 · Prashanth Vijayaraghavan, Apoorva Nitsure, Luyao Shi et al.

Summary

Writing and explaining hardware code at the Register-Transfer Level (RTL) is hard for LLMs. HDL syntax is strict, annotated data is scarce, and Verilog/VHDL map poorly onto natural language. Prompting and RAG approaches also lack an explicit symbolic plan. The IBM Research authors propose SYMDIREC, a three-stage neuro-symbolic pipeline that handles both RTL synthesis (NL spec to code) and RTL summarization (code to NL):

  • Divide: an LLM splits the input into sub-components, each paired with a symbolic logic form (Boolean, temporal or dataflow). For summarization, the split follows the code's AST.
  • Retrieve: a retriever fine-tuned on the curated RTL-IR dataset uses those symbolic forms to fetch matching code fragments.
  • Conquer: an LLM checks the candidates against the symbolic intent and assembles the final module or summary.

Only the retriever is trained; the LLM itself is not fine-tuned. Across Verilog and VHDL benchmarks, the method improves Pass@1 by about 20% and ROUGE-L by 15–20% over prompting and RAG baselines. This suggests symbolic scaffolding is a lightweight way to make hardware code generation more precise.

Key points

  • The authors describe SYMDIREC as the first system to tackle synthesis and summarization jointly across both Verilog and VHDL.
  • The retriever is trained on RTL-IR, a dataset drawn from GitHub projects released under reuse-friendly licenses (MIT, BSD, Apache-2.0). Public benchmarks such as Verilog-Eval are also used.
  • Hyperparameter study:
    • Performance rises as the number of sub-components N grows to 6, then falls.
    • Retrieval depth helps up to k = 5, after which it plateaus or drops.
    • The recommended default is N = 4, k = 5.
  • Error analysis:
    • About 8–10% of sub-components get flawed symbolic expressions.
    • About 12–15% of retrieved candidates match only partially.
    • About 6–8% of sub-components suffer assembly failures.
    • Together these leave a ~10–15% Pass@1 gap to the oracle variant SYMDIREC-GT.
  • Main limitation: only single-file RTL designs are supported, not hierarchical or multi-file projects. Weaker LLMs also produce less useful symbolic decompositions.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢IBMArmonk · United States154 works · 72,606 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksHardware and Architecture · 167 worksEngineering · 188 worksElectrical and Electronic Engineering · 33 works

🧭 Topics 1Research subjects and languages the work is about.Summarization · 243 works

Cite

Cite as Prashanth Vijayaraghavan, Apoorva Nitsure, Luyao Shi, Charles Mackin, Ashutosh Jadhav, David Beymer, Ehsan Degan and Vandana Mukherjee. 2026. SYMDIREC: A Neuro-Symbolic Divide-Retrieve-Conquer Framework for Enhanced RTL Synthesis and Summarization. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-industry.66

@inproceedings{vijayaraghavan2026symdirec,
  title = {{SYMDIREC: A Neuro-Symbolic Divide-Retrieve-Conquer Framework for Enhanced RTL Synthesis and Summarization}},
  author = {Prashanth Vijayaraghavan, Apoorva Nitsure, Luyao Shi, Charles Mackin, Ashutosh Jadhav, David Beymer, Ehsan Degan and Vandana Mukherjee},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 5: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-industry.66},
  url = {https://doi.org/10.18653/v1/2026.eacl-industry.66}
}

References

35 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2023. Self-rag: Learning to retrieve, generate, and critique through self-reflection. In The Twelfth International Conference on Learning Representations.
  2. Yongchao Chen, Yilun Hao, Yang Zhang, and Chuchu Fan. 2025. Code-as-symbolic-planner: Foundation model-based robot planning via symbolic code generation. arXiv preprint arXiv:2503.01700. — [arXiv:2503.01700](https://arxiv.org/abs/2503.01700) find
  3. Elastic. 2024. Machine learning: Natural language processing with elastic. https://www.elastic.co/guide/en/machine-learning/current/ ml-nlp-elser.html. Accessed: March 21, 2025. ↗
  4. Thibault Formal, Carlos Lassance, Benjamin Piwowarski, and Stéphane Clinchant. 2021. Splade v2: Sparse lexical and expansion model for information retrieval. arXiv preprint arXiv:2109.10086.
  5. Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang. 2020. Retrieval augmented language model pre-training. In International conference on machine learning, pages 3929–3938. PMLR.
  6. Chia-Tung Ho, Haoxing Ren, and Brucek Khailany. 2025. Verilogcoder: Autonomous verilog coding agents with graph-based planning and abstract syntax tree (ast)-based waveform tracing tool. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 300–307. find
  7. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing Systems, 33:9459–…
  8. Raymond Li, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia LI, Jenny Chim, Qian Liu, and 1 others. 2023a. Starcoder: may the source be with you! Transactions on Machine Learning Research. find
  9. Zehan Li, Xin Zhang, Yanzhao Zhang, Dingkun Long, Pengjun Xie, and Meishan Zhang. 2023b. Towards general text embeddings with multi-stage contrastive learning. arXiv preprint arXiv:2308.03281.
  10. Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text Summarization Branches Out.
  11. Mingjie Liu, Nathaniel Pinckney, Brucek Khailany, and Haoxing Ren. 2023. Verilogeval: Evaluating large language models for verilog code generation. In 2023 IEEE/ACM International Conference on Computer Aided Design (ICCAD), pages 1–8. IEEE. find
  12. Shang Liu, Wenji Fang, Yao Lu, Jing Wang, Qijun Zhang, Hongce Zhang, and Zhiyao Xie. 2024. Rtlcoder: Fully open-source and efficient llm-assisted rtl code generation technique. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems. find
  13. Mayank Mishra, Matt Stallone, Gaoyuan Zhang, Yikang Shen, Aditya Prasad, Adriana Meza Soria, Michele Merler, Parameswaran Selvam, Saptha Surendran, Shivdeep Singh, and 1 others. 2024. Granite code models: A family of open foundation models for code intelligence. CoRR. find
  14. Niklas Muennighoff, Nouamane Tazi, Loic Magne, and Nils Reimers. 2023. Mteb: Massive text embedding benchmark. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics. Association for Computational Linguistics.
  15. Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, and Caiming Xiong. 2022. Codegen: An open large language model for code with multi-turn program synthesis. In The Eleventh International Conference on Learning Representations. find
  16. Liangming Pan, Alon Albalak, Xinyi Wang, and William Yang Wang. Logic-lm: Empowering large language models with symbolic solvers for faithful logical reasoning. In The 2023 Conference on Empirical Methods in Natural Language Processing.
  17. Md Rizwan Parvez, Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, and Kai-Wei Chang. 2021. Retrieval augmented code generation and summarization. In EMNLP-Findings.
  18. F Petroni, A Piktus, A Fan, PSH Lewis, M Yazdani, ND Cao, J Thorne, Y Jernite, V Karpukhin, J Maillard, and 1 others. 2021. Kilt: a benchmark for knowledge intensive language tasks. In NAACL-HLT, pages 2523–2544. Association for Computational Linguistics.
  19. Heng Ping, Shixuan Li, Peiyu Zhang, Anzhe Cheng, Shukai Duan, Nikos Kanakaris, Xiongye Xiao, Wei Yang, Shahin Nazarian, Andrei Irimia, and 1 others. 2025. Hdlcore: A training-free framework for mitigating hallucinations in llm-generated hdl. arXiv preprint arXiv:2503.16528. — [arXiv:2503.16528](https://arxiv.org/abs/2503.16528) find
  20. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992.
  21. Stephen Robertson, Hugo Zaragoza, and 1 others. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends® in Information Retrieval, 3(4):333–389. find
  22. Baptiste Roziere, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Jérémy Rapin, and 1 others. 2023. Code llama: Open foundation models for code. arXiv preprint arXiv:2308.12950.
  23. Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper, and Bryan Catanzaro. 2019. Megatron-lm: Training multi-billion parameter language models using model parallelism. arXiv preprint arXiv:1909.08053. find
  24. Aivin V. Solatorio. 2024. Gistembed: Guided in-sample selection of training negatives for text embedding fine-tuning. arXiv preprint arXiv:2402.16829. — [arXiv:2402.16829](https://arxiv.org/abs/2402.16829) find
  25. Prashanth Vijayaraghavan, Apoorva Nitsure, Charles Mackin, Luyao Shi, Stefano Ambrogio, Arvind Haran, Viresh Paruthi, Ali Elzein, Dan Coops, David Beymer, and 1 others. 2024a. Chain-of-descriptions: Improving code llms for vhdl code generation and summarization. In Proceedings of the 2024 ACM/IEEE I… find
  26. Prashanth Vijayaraghavan, Luyao Shi, Stefano Ambrogio, Charles Mackin, Apoorva Nitsure, David Beymer, and Ehsan Degan. 2024b. Vhdl-eval: A framework for evaluating large language models in vhdl code generation. In 2024 IEEE LLM Aided Design Workshop (LAD), pages 1–6. IEEE. find
  27. Yue Wang, Weishi Wang, Shafiq Joty, and Steven CH Hoi. 2021. Codet5: Identifier-aware unified pre-trained encoder-decoder models for code understanding and generation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8696–8708. find
  28. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  29. Shi-Qi Yan, Jia-Chen Gu, Yun Zhu, and Zhen-Hua Ling. 2024. Corrective retrieval augmented generation. arXiv preprint arXiv:2401.15884.
  30. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR).
  31. Xufeng Yao, Haoyang Li, Tsz Ho Chan, Wenyi Xiao, Mingxuan Yuan, Yu Huang, Lei Chen, and Bei Yu. 2024. Hdldebugger: Streamlining hdl debugging with large language models. ACM Transactions on Design Automation of Electronic Systems. find
  32. Dun Zhang. 2023. stella_en_400m_v5. https://huggingface.co/dunzhang/stella_en_400M_v5. Hugging Face model card, accessed on March 21, 2025. ↗
  33. Yang Zhao, Di Huang, Chongxiao Li, Pengwei Jin, Muxin Song, Yinan Xu, Ziyuan Nan, Mingju Gao, Tianyun Ma, Lei Qi, and 1 others. 2025. Codev: Empowering llms with hdl generation through multi-level summarization. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems. find
  34. Shuyan Zhou, Uri Alon, Frank F Xu, Zhengbao Jiang, and Graham Neubig. 2022. Docprompting: Generating code by retrieving the docs. In The Eleventh International Conference on Learning Representations. find
  35. Jian Zuo, Junzhe Liu, Xianyong Wang, Yicheng Liu, Navya Goli, Tong Xu, Hao Zhang, Umamah-eswara Rao Tida, Zhenge Jia, and Mengying Zhao. 2025. Complexvcoder: An llm-driven framework for systematic generation of complex verilog code. arXiv preprint arXiv:2504.20653. — [arXiv:2504.20653](https://arxiv.org/abs/2504.20653) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vijayaraghavan-2026-symdirec-neuro-symbolic-divide-retrieve/