Pith. sign in

REVIEW 4 major objections 5 minor 28 references

Front-end chip design can move from LLM task helpers to closed-loop agents that keep design intent consistent across stages.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 01:42 UTC pith:A5HD77RT

load-bearing objection Useful DAC-style roadmap that packages the authors' own front-end LLM results under an agentic framing; the multi-stage consistency claim stays aspirational. the 4 major comments →

arxiv 2607.09616 v1 pith:A5HD77RT submitted 2026-07-10 cs.ET cs.ARcs.LGcs.SYeess.SY

LLM for EDA in Front-End Design: Challenges and Opportunities

Kangwei Xu , Bing Li , Ulf Schlichtmann This is my paper
classification cs.ET cs.ARcs.LGcs.SYeess.SY
keywords large language modelselectronic design automationfront-end designHDL generationtestbench generationhigh-level synthesisagentic AIsemantic consistency
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper argues that large language models are already useful for front-end electronic design automation—interpreting specifications, drafting hardware description language, building testbenches, and repairing high-level synthesis code—but the real prize is turning them into agents that run full design-verify-repair loops. The bottleneck is not isolated code generation; it is preserving the original design meaning as work moves from natural-language specs through system models, HDL, and verification. By reviewing consistency-ranking methods for HDL candidates, hybrid driver-checker testbench generators with self-correction, and agentic HLS repair-test-optimize pipelines, the authors claim measurable gains on academic benchmarks while mapping a path from localized assistance to tool-orchestrating autonomy. A sympathetic reader cares because front-end design is now a major schedule bottleneck; if agents can coordinate tools, interpret feedback, and keep stages consistent, engineers can spend more time on architecture and less on mechanical translation and debug.

Core claim

LLMs can evolve from stage-local assistants into autonomous front-end EDA agents that generate, verify, and refine designs in closed loops while preserving semantic consistency across heterogeneous representations—specifications, system models, HDL, and testbenches—provided generation is tightly coupled to simulation feedback, tool orchestration, and iterative repair.

What carries the argument

The closed-loop agentic front-end flow: multi-candidate HDL generation filtered and ranked by reasoning density and simulation self-consistency; hybrid testbench generation (driver plus Python checker) with functional self-validation and repair; and staged HLS agents that repair synthesizability, check equivalence, and rewrite for PPA—all coordinated so early intent survives later stages.

Load-bearing premise

That gains from ranking, hybrid checkers, and self-correction on current academic circuit benchmarks will still hold for industrial designs that need full coverage, hard corner cases, and multi-stage semantic consistency.

What would settle it

Run the same ranking-plus-self-correction HDL and testbench pipelines on a multi-module industrial-scale design with timing boundaries and concurrent interactions; if majority-consistent clusters still pass local tests while missing real semantic bugs, or if self-validators fail to flag incomplete coverage, the agentic roadmap does not yet transfer.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This DAC perspective paper argues that LLMs can move front-end EDA from localized assistance (Q&A, drafting) toward autonomous agentic execution that coordinates tools, interprets feedback, and preserves semantic consistency across specification, system models, HDL, and testbenches. It sketches an evolution path (Fig. 1) and an agentic flow (Fig. 2), then reviews three clusters of results: simulation self-consistency ranking for HDL generation (VRank/VFocus), hybrid/self-correcting testbench generation (AutoBench/CorrectBench/ConfiBench), and LLM-aided HLS repair, testing, and rewrite (HLSRepair/HLSTester/HLSRewriter). Section 4 lists open challenges (datasets, specification refinement, multi-agent systems, efficiency) and positions OpenClaw-style agents as a roadmap rather than a finished system.

Significance. If the roadmap is right, the paper would help organize a fast-moving area around a clear failure mode—early semantic drift that only appears as low-level symptoms—and around closed-loop agent design rather than one-shot code generation. The manuscript is useful as a compact map of front-end tasks and as an explicit call for aligned corpora, structured tool interfaces, and verifiable loops. Strengths include a coherent stage model (Fig. 2), concrete quantitative anchors from the cited generation/testbench/HLS studies (e.g., Pass@1 gains in Figs. 4 and 6–8), and an honest challenges section that names semantic consistency, token cost, and missing industrial proxies. The contribution is primarily organizational and prospective, not a new multi-stage experiment.

major comments (4)
  1. [§1, §3, §5] The load-bearing claim in the Introduction and Conclusion—that LLMs can evolve into agents that preserve semantic consistency across heterogeneous representations—is not supported by multi-stage evidence in this manuscript. Section 3 reports single-stage gains (VerilogEval-style Pass@1 via ranking/filtering; testbench pass rates via hybrid checkers and imperfect-DUT validation; HLS synthesizability/PPA). Those results are consistent with the cited figures but do not demonstrate that high-level intent survives re-partitioning, re-synthesis, and re-verification—the exact failure mode §1 flags. Either add a multi-stage case study/consistency metric, or systematically demote the agentic claim to an explicitly aspirational roadmap and state what would falsify it.
  2. [§3.1–§3.3, References [9–13,19–23,28]] The quantitative backbone of §3 is almost entirely prior work from the same author group (VRank, VFocus, AutoBench, CorrectBench, ConfiBench, HLSRepair, HLSTester, HLSRewriter, plus related MEIC/CorrectHDL citations). For a paper framed as a “systematic perspective,” this creates a circularity risk: the same line of work supplies the evidence and the synthesis. Broaden §3 with independent baselines and external systems (e.g., ChipNeMo, ChatEDA, VerilogEval-era generators, other assertion/stimulus agents) under comparable metrics, or reframe §3 explicitly as “progress in our recent line of work” rather than a field-wide breakthrough narrative.
  3. [§3.1 (VFocus), §3.2 (CorrectBench/Fig. 6)] Several free parameters that drive the reported gains are not stress-tested for transfer: VFocus Lmin/Lmax “reasoning sweet spot,” CorrectBench’s 70%-wrong validator threshold, candidate counts, and syntax-retry limits. Fig. 6 already shows criterion sensitivity for validators, but the manuscript still treats the chosen settings as generally reliable proxies for industrial coverage and corner-case correctness. Add sensitivity analysis or clearly bound claims to the academic suites used, and state that simulation agreement under shared imperfect testbenches is not equivalent to coverage closure or cross-stage semantic fidelity.
  4. [Abstract, Fig. 1, §4] Section 4’s OpenClaw-style roadmap (tool orchestration, compressed skill interfaces, persistent memory) is plausible but currently axiomatic: no EDA-specific agent experiment, token/latency budget, or failure analysis is reported here. Given that §4 itself lists semantic consistency and efficiency as open problems, the abstract and Fig. 1 “S5: Autonomy” framing should be aligned with that caution so readers do not infer demonstrated autonomy from localized Pass@1 improvements.
minor comments (5)
  1. [Fig. 3–6] Figure numbering/captioning is inconsistent in places (e.g., embedded “Fig. 2: Overall framework of VFocus” inside the manuscript text vs. paper Fig. 3). Clean numbering so every panel maps to a single caption.
  2. [References] Duplicate bibliography entries appear for the same “Revolution or Hype?” paper ([25] and [26]). Deduplicate and normalize arXiv/venue metadata.
  3. [§1, [8]] OpenClaw is cited as a 2026 GitHub repo with an access date in 2026; for archival stability, add a commit hash or archived snapshot and briefly define which agent properties are assumed transferable to EDA.
  4. [§3] Pass-rate comparisons mix Pass@1, Eval0/1/2, and “overall pass rate” across figures; a short metric glossary would help non-specialist readers compare HDL vs. testbench vs. HLS results.
  5. [§1, §5] Minor prose issues: “propelling … frontier,” repeated “great potential,” and occasional tense shifts between review and roadmap. Tighten for a 5-page DAC format.

Circularity Check

0 steps flagged

No definitional or predictive circularity; heavy author self-citation of prior empirical systems is normal for a perspective review and does not force the agentic claim by construction.

full rationale

This is a short perspective/review paper, not a first-principles derivation. It contains no equations that define a quantity in terms of the quantity being predicted, no parameters fitted to data and then re-labeled as predictions, no uniqueness theorems imported from the authors, and no ansatz smuggled in via citation. The narrative of localized progress (VRank/VFocus Pass@1 gains, AutoBench/CorrectBench/ConfiBench testbench pass rates, HLSRepair/HLSTester/HLSRewriter synthesizability and PPA improvements) is supported by citations to published empirical systems, many of which share authors with the present manuscript. Those prior results are externally falsifiable on public or released benchmarks (VerilogEval-style suites, sequential-circuit suites, C/C++ HLS suites) and are not re-derived inside this paper; the present text simply summarizes them with figures and then lists open challenges (semantic consistency across stages, token cost, missing aligned corpora). The agentic closed-loop roadmap (OpenClaw-style orchestration, multi-agent systems, design memory) is explicitly framed as aspirational future work rather than a result forced by the cited numbers. Self-citation of a research group’s own line of work is therefore present but not load-bearing circularity under the defined patterns. Score 1 reflects only the minor, non-forcing self-citation density typical of a group perspective; the derivation chain itself is empty of circular reductions.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 1 invented entities

As a perspective paper the central claim rests on domain assumptions about LLM tool-use and EDA semantics rather than fitted constants or new physical entities. Free parameters appear only inside the summarized prior systems (reasoning-length bounds, majority-cluster ranking, 70%-wrong validation threshold). No new particles or forces are postulated; the 'agentic EDA agent' is an architectural aspiration, not an invented physical entity with independent evidence.

free parameters (3)
  • VFocus reasoning-length sweet-spot bounds (Lmin, Lmax)
    Empirically chosen per model to filter candidates before ranking; directly affects which HDL samples enter the self-consistency stage reported as a 30.9% gain.
  • CorrectBench 70%-wrong validation threshold
    Hand-selected percentage that decides whether a testbench is judged wrong and triggers self-correction; authors note it may not be optimal yet use it for the headline 70.13% pass rate.
  • Number of sampled candidates / retry limit (e.g., five syntax retries)
    Sampling budget and retry caps in VRank/VFocus-style pipelines are free engineering choices that change reported Pass@1.
axioms (4)
  • domain assumption Majority behavioral agreement under a shared (possibly imperfect) testbench correlates with functional correctness of HDL candidates.
    Core of VRank/VFocus ranking (Eqs. 2–3 in the embedded VFocus description); assumed throughout §3.1.
  • domain assumption Semantic mismatches introduced early become harder to diagnose after synthesis/netlist stages, so closed-loop agents are required.
    Stated in Introduction and Fig. 2 discussion; underpins the entire agentic roadmap.
  • ad hoc to paper OpenClaw-style tool orchestration, persistent memory, and skill retrieval transfer usefully to EDA once skill documents are compressed into structured interfaces.
    §4 roadmap; OpenClaw is cited as a pioneering system but no EDA-specific transfer proof is given in this paper.
  • domain assumption Python checkers are more reliably generated by LLMs than pure Verilog checkers and still validate hardware behavior.
    Design choice in AutoBench/CorrectBench (§3.2).
invented entities (1)
  • LLM-based agentic front-end EDA agent (OpenClaw-style for chip design) no independent evidence
    purpose: Orchestrate multi-stage design, verification, and optimization while maintaining long-horizon semantic consistency.
    Presented as the strategic target of the paper; no deployed industrial instance or independent falsifiable measurement is provided beyond cited partial systems.

pith-pipeline@v1.1.0-grok45 · 18350 in / 3221 out tokens · 37487 ms · 2026-07-13T01:42:08.426336+00:00 · methodology

0 comments
read the original abstract

As chip complexity increases and time-to-market pressures grow, front-end design has become a critical bottleneck in chip development. Recently, Large Language Models (LLMs) have shown great potential in Electronic Design Automation (EDA). Beyond specification understanding, LLMs show the potential to serve as a unified intelligent interface for hardware description language (HDL) generation, testbench construction, and design space exploration. The rise of agentic AI, represented by pioneering systems such as OpenClaw, offers a strategic roadmap for the next generation EDA. From this perspective, this paper discusses the evolution of EDA from localized assistance to autonomous agentic execution. Then, we review representative advances of LLMs in front-end design, focusing on key tasks such as circuit and testbench generation from a shared specification, as well as design quality improvement in established workflows such as high-level synthesis. Finally, we discuss the key challenges and limitations of integrating LLMs into EDA, and outline future opportunities for advancing LLM-enabled front-end design, offering a systematic perspective for researchers interested in leveraging agentic AI technologies for EDA.

Figures

Figures reproduced from arXiv: 2607.09616 by Bing Li, Kangwei Xu, Ulf Schlichtmann.

Figure 1
Figure 1. Figure 1: Evolution Path of LLM Capabilities in Chip Design. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: LLM-Based Agentic Chip Design Flow. automation of individual tasks, but preserving semantic consistency across different design stages. In hardware design, an implementa￾tion may be syntactically correct and pass basic test vectors, while still containing errors in timing boundaries, concurrent interactions, or corner cases. More importantly, even minor semantic mismatches introduced in early stages can pr… view at source ↗
Figure 2
Figure 2. Figure 2: Overall framework of VFo Figure 3: Overview of Automated HDL Design with Density [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Outline of AutoBench workflow and AutoEval eval Figure 5: Overview of Automated Testbench Generation [11–13]. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 8
Figure 8. Figure 8: Results of HLS Program Repair in Pass Rate [21]. [PITH_FULL_IMAGE:figures/full_fig_p003_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Future Opportunities for EDA in Front-End Design. [PITH_FULL_IMAGE:figures/full_fig_p004_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 2 linked inside Pith

  1. [1]

    SpecLLM: Exploring Generation and Review of VLSI Design Specification with Large Language Model,

    M. Li, W. Fang, Q. Zhang, and Z. Xie, “SpecLLM: Exploring Generation and Review of VLSI Design Specification with Large Language Model, ” inProceedings of the International Symposium of Electronics Design Automation (ISEDA), 2025

    2025

  2. [2]

    ChipNeMo: Domain-Adapted LLMs for Chip Design,

    M. Liuet al., “ChipNeMo: Domain-Adapted LLMs for Chip Design, ”arXiv preprint arXiv:2311.00176, 2023

    Pith/arXiv arXiv 2023

  3. [3]

    Invited Paper: VerilogEval: Evaluating Large Language Models for Verilog Code Generation,

    M. Liu, N. R. Pinckney, B. Khailany, and H. Ren, “Invited Paper: VerilogEval: Evaluating Large Language Models for Verilog Code Generation, ” inProceedings of the IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2023

    2023

  4. [4]

    LLM4DV: Using Large Language Models for Hardware Test Stimuli Generation,

    Z. Zhang, B. Szekely, P. Gimenes, G. Chadwick, H. McNally, J. Cheng, R. D. Mullins, and Y. Zhao, “LLM4DV: Using Large Language Models for Hardware Test Stimuli Generation, ” inProceedings of the IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM), 2025

    2025

  5. [5]

    ChatEDA: A Large Language Model Powered Autonomous Agent for EDA,

    H. Wu, Z. He, X. Zhang, X. Yao, S. Zheng, H. Zheng, and B. Yu, “ChatEDA: A Large Language Model Powered Autonomous Agent for EDA, ”IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 43, no. 10, 2024

    2024

  6. [6]

    Towards LLM-based Root Cause Analysis of Hardware Design Failures,

    S. Qiu, M. Wang, R. Afsharmazayejani, M. M. Shahmiri, B. Tan, H. Pearce, “Towards LLM-based Root Cause Analysis of Hardware Design Failures, ” inProceedings of the IEEE International Conference on Omni-layer Intelligent Systems, 2025

    2025

  7. [7]

    EDA-Debugger: An LLM- Based Framework for Automated EDA Runtime Issue Resolution,

    J. Li, S. -Z. Wong, G. -W. Wan, X. Wang and J. Yang, “EDA-Debugger: An LLM- Based Framework for Automated EDA Runtime Issue Resolution, ” inProceedings of the IEEE International Symposium on Quality Electronic Design (ISQED), 2025

    2025

  8. [8]

    OpenClaw,

    OpenClaw, “OpenClaw, ” GitHub repository, 2026. [Online]. Available: https://github.com/openclaw/openclaw . Accessed: Apr. 15, 2026

    2026

  9. [9]

    MEIC: Re-thinking RTL Debug Automation using LLMs,

    K. Xu, J. Sun, Y. Hu, X. Fang, W. Shan, X. Wang, and Z. Jiang, “MEIC: Re-thinking RTL Debug Automation using LLMs, ” inProceedings of the IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2024

    2024

  10. [10]

    LLM-Aided Efficient Hardware Design Automation,

    K. Xu, R. Qiu, Z. Zhao, G. L. Zhang, U. Schlichtmann, B. Li, “LLM-Aided Efficient Hardware Design Automation, ”arXiv: 2410.18582, 2024

    Pith/arXiv arXiv 2024

  11. [11]

    AutoBench: Au- tomatic Testbench Generation and Evaluation Using LLMs for HDL Design,

    R. Qiu, G. L. Zhang, R. Drechsler, U. Schlichtmann, and B. Li, “AutoBench: Au- tomatic Testbench Generation and Evaluation Using LLMs for HDL Design, ” in Proceedings of the ACM/IEEE International Symposium on Machine Learning for CAD (MLCAD), 2024

    2024

  12. [12]

    CorrectBench: Automatic Testbench Generation with Functional Self-Correction using LLMs for HDL Design,

    R. Qiu, G. L. Zhang, R. Drechsler, U. Schlichtmann, and B. Li, “CorrectBench: Automatic Testbench Generation with Functional Self-Correction using LLMs for HDL Design, ” inProceedings of the IEEE/ACM Design, Automation & Test in Europe Conference & Exhibition (DATE), 2025

    2025

  13. [13]

    Con- fiBench: Automatic Testbench Generation with Confidence-Based Scenario Mask and Testbench Ensemble using LLMs for HDL Design,

    R. Qiu, G. L. Zhang, R. Drechsler, T.-Y. Ho, U. Schlichtmann, and B. Li, “Con- fiBench: Automatic Testbench Generation with Confidence-Based Scenario Mask and Testbench Ensemble using LLMs for HDL Design, ”ACM Transactions on Design Automation of Electronic Systems (TODAES), 2026

    2026

  14. [14]

    Exploring Code Language Models for Automated HLS-based Hardware Generation: Bench- mark, Infrastructure and Analysis,

    J. Gai, H. M. Chen, Z. Wang, H. Zhou, W. Zhao, N. Lane, and H. Fan, “Exploring Code Language Models for Automated HLS-based Hardware Generation: Bench- mark, Infrastructure and Analysis, ” inProceedings of the IEEE/ACM Asia and South Pacific Design Automation Conference (ASP-DAC), 2025

    2025

  15. [15]

    FPGA HLS Today: Successes, Challenges, and Opportunities,

    J. Cong, J. Lau, G. Liu, S. Neuendorffer, P. Pan, K. Vissers, and Z. Zhang, “FPGA HLS Today: Successes, Challenges, and Opportunities, ”ACM Transactions on Reconfigurable Technology and Systems (TRETS), vol. 15, no. 4, 2022

    2022

  16. [16]

    Are We There Yet? A Study on the State of High-Level Synthesis,

    S. Lahti, P. Sjövall, J. Vanne, T. Hämäläinen, “Are We There Yet? A Study on the State of High-Level Synthesis, ”IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (TCAD), 2019

    2019

  17. [17]

    AssertLLM: Generating Hardware Verification Assertions from Design Specifications via Multi-LLMs,

    Z. Yan, W. Fang, M. Li, M. Li, S. Liu, Z. Xie, H. Zhang, “AssertLLM: Generating Hardware Verification Assertions from Design Specifications via Multi-LLMs, ” in Proceedings of the IEEE/ACM Asia and South Pacific Design Automation Conference (ASP-DAC), 2025

    2025

  18. [18]

    Invited: Agentic AI for Physi- cal Design R&D: Status and Prospects,

    A. Ghose, A. B. Kahng, S. Kundu, B. Pramanik, “Invited: Agentic AI for Physi- cal Design R&D: Status and Prospects, ” inProceedings of the IEEE International Symposium on Physical Design (ISPD), 2026

    2026

  19. [19]

    VRank: Enhancing Verilog Code Generation from LLMs via Self-Consistency,

    Z. Zhao, R. Qiu, I.-C. Lin, G. L. Zhang, B. Li, U. Schlichtmann, “VRank: Enhancing Verilog Code Generation from LLMs via Self-Consistency, ” inProceedings of the IEEE International Symposium on Quality Electronic Design (ISQED), 2025

    2025

  20. [20]

    VFocus: Better Verilog Gener- ation from Large Language Model via Focused Reasoning,

    Z. Zhao, B. Li, G. L. Zhang, and U. Schlichtmann, “VFocus: Better Verilog Gener- ation from Large Language Model via Focused Reasoning, ” inProceedings of the IEEE International System-on-Chip Conference (SOCC), 2025

    2025

  21. [21]

    Automated C/C++ Program Repair for High-Level Synthesis via Large Language Models,

    K. Xu, G. L. Zhang, X. Yin, C. Zhuo, U. Schlichtmann, and B. Li, “Automated C/C++ Program Repair for High-Level Synthesis via Large Language Models, ” in Proceedings of the ACM/IEEE International Symposium on Machine Learning for CAD (MLCAD), 2024

    2024

  22. [22]

    HLSTester: Efficient Testing of Behavioral Discrepancies with LLMs for High-Level Synthesis,

    K. Xu, B. Li, G. L. Zhang, and U. Schlichtmann, “HLSTester: Efficient Testing of Behavioral Discrepancies with LLMs for High-Level Synthesis, ” inProceedings of the IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2025

    2025

  23. [23]

    HLSRewriter: Efficient Refactoring and Optimization of C/C++ Code with LLMs for High-Level Synthesis,

    K. Xu, G. L. Zhang, X. Yin, C. Zhuo, U. Schlichtmann, and B. Li, “HLSRewriter: Efficient Refactoring and Optimization of C/C++ Code with LLMs for High-Level Synthesis, ”ACM Transactions on Design Automation of Electronic Systems, 2026

    2026

  24. [24]

    Logic Design of Neural Networks for High-Throughput and Low- Power Applications,

    K. Xu, et al., “Logic Design of Neural Networks for High-Throughput and Low- Power Applications, ” inProceedings of the IEEE/ACM Asia and South Pacific Design Automation Conference (ASP-DAC), 2024

    2024

  25. [26]

    Revolution or Hype? Seeking the Limits of Large Models in Hardware Design,

    Q. Xu, L. Stok, R. Drechsler, X. Wang, G. L. Zhang, I. L. Markov, “Revolution or Hype? Seeking the Limits of Large Models in Hardware Design, ” inProceedings of the IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2025

    2025

  26. [27]

    HumanEval Pro and MBPP Pro: Evaluating Large Language Models on Self-invoking Code Generation Task,

    Z. Yu, Y. Zhao, A. Cohan, and X.-P. Zhang, “HumanEval Pro and MBPP Pro: Evaluating Large Language Models on Self-invoking Code Generation Task, ” in Findings of the Association for Computational Linguistics (ACL), 2025

    2025

  27. [28]

    CorrectHDL: Agentic HDL Design with LLMs Leveraging High-Level Synthesis as Reference,

    K. Xu, G. L. Zhang, U. Schlichtmann, and B. Li, “CorrectHDL: Agentic HDL Design with LLMs Leveraging High-Level Synthesis as Reference, ”arXiv preprint arXiv:2511.16395, 2025

    arXiv 2025

  28. [29]

    Large Language Models (LLMs) for Electronic Design Automation (EDA) : Special Session Paper,

    K. Xu, D. Schwachhofer, J. Blocklove, I. Polian, P. Domanski, D. Pfluger, S. Garg, R. Karri, O. Sinanoglu, J. Knechtel, Z. Zhao, U. Schlichtmann and B. Li, “Large Language Models (LLMs) for Electronic Design Automation (EDA) : Special Session Paper, ” inProceedings of the IEEE International System-on-Chip Conference, 2025

    2025