Pith. sign in

REVIEW 3 major objections 4 minor 36 references

SCALE: Self-Supervised Constraint-Aware Layout GEneration for Local P&R DRV Fixing at Advanced Nodes

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Pith's one-line read Self-supervised layout generation trains a VLM that lifts DRV repair rates to 97%

desk verdict A credible, honestly-scoped pipeline for DRC-aware VLM guidance; the missing generic-VLM control is the main gap between what's shown and what's claimed. read the letter →

arxiv 2607.21850 v1 pith:EFWGNXRI submitted 2026-07-23 cs.CV cs.AI

classification cs.CVcs.AI
keywords designrulecheckingDRVfixinglayoutgenerationvision-languagemodelself-supervisedlearningsub-2nmtechnologyplaceandroutesyntheticdata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that local design-rule-violation (DRV) repair at sub-2nm nodes can be automated by a two-stage pipeline. First, real back-end-of-line layouts are serialized into text, an autoregressive model learns to reconstruct masked polygons from surrounding context alone, and inference-time natural-language constraints plus high-temperature sampling generate diverse, violation-prone layouts that an industrial design-rule-check (DRC) signoff checker labels. Second, a vision-language model (VLM) fine-tuned on those labeled pairs supplies rule-aware root-cause analysis and ranked fixes to state-of-the-art coding agents, raising their solve rates from 63.0% to 88.0% and from 85.0% to 97.0% on 100 real sub-2nm cases and reducing token cost. If correct, this shows that self-supervised synthetic data can close the data-scarcity and domain-gap barriers that keep general-purpose VLMs from precise geometric reasoning on proprietary design rules.

What carries the argument

The core mechanism is the text-formulated layout representation: each rectangle is encoded as a directional edge walk from its top-left vertex via cardinal deltas (→ width, ↓ height, ← width, ↑ height), grouped by layer headers, with [Poly]/[\Poly] tokens for complete polygons and [BPoly] tokens for boundary-clipped context shapes that can never be generation targets. This serialization lets a language model be trained by masked-polygon reconstruction, and at inference, natural-language constraints (e.g., 'generate more tip-to-tip shapes') steer sampling toward violation-prone configurations that are then validated by an industrial DRC checker. The second piece is the DRC-VLM, a fine-tuned o

What would settle it

A direct falsifier is to strip the DRC-VLM's analysis out of the agent loop while keeping the same coding agents, DRC/LVS checker tools, and checker feedback on the same 100 cases; if the solve rate does not drop from the reported 97%, the central claim fails. A sharper test targets the representation: take a real layout containing a non-rectangular polygon and run the tokenizer; if it cannot serialize that polygon, the training corpus is provably incomplete for that geometry.

Watch

Extended reading notes

Core claim

The central claim is that a domain-adapted DRC-VLM, trained entirely on layouts produced by reconstructing masked polygons from surrounding metal/via context and labeled by a signoff DRC checker, gives a coding agent enough rule-aware geometric guidance to fix local DRC violations without introducing new violations or breaking the connectivity check (LVS). The authors validate this on 100 real sub-2nm cases: the weaker baseline agent solves 63.0% of cases without guidance and 88.0% with it; the stronger baseline agent solves 85.0% and 97.0% respectively. They also show the gain is concentrated in localization rather than rule recall: general-purpose VLMs reach at most 24.6% localization F1 a

Load-bearing premise

The generation stage assumes every polygon can be serialized as an axis-aligned rectangle via a four-step directional edge walk; the paper does not test this restriction, yet if real sub-2nm routing contains non-rectangular multi-vertex shapes, the tokenizer cannot represent them and the corpus silently misses those geometries.

Editorial extensions

If this is right

  • If correct, VLM-based guidance can replace hand-tuned geometric templates for conditional DRC rules, so a new rule family can be covered by generating and labeling more layouts rather than writing new rule-specific repair code.
  • Signoff-validated self-supervised generation offers a practical way to create proprietary-rule training data at scale from real designs, without manual labeling or leaking confidential rule decks.
  • The reported token-cost reductions (24.5% and 2.2%) mean rule-aware analysis shortens the agent's exploratory loop, which matters because each fix is capped at 1800 seconds of compute.
  • The method covers enclosure, width, spacing, color-spacing, and area violations, including conditional and group-dependent rules that template-based engines miss.
  • The remaining unsolved cases cluster in color-spacing, where local edits risk cascading violations; net-level rerouting, named as future work, is the natural next test of the approach's ceiling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the rectangularity restriction of the tokenizer is real, the generated corpus silently excludes non-rectangular multi-vertex routing shapes; a test on such shapes would show whether the diversity numbers overstate coverage.
  • The 100-case evaluation has only 3 area cases, so the Area category's apparent improvement is a single-case swing; the headline gains are carried by enclosure and spacing categories.
  • The reported solve-rate gap does not isolate the causal role of the VLM analysis from the checker-in-the-loop feedback already present in the agents; an ablation that removes only the VLM text would disentangle them.
  • The same generate-annotate-finetune recipe should transfer to other geometry-heavy EDA tasks, such as hotspot detection or lithography repair, where labeled data is scarce and rules are proprietary.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper presents SCALE, a two-phase framework for local P&R DRV fixing at advanced nodes. Phase I serializes layout crops into structured text using cardinal-delta rectangle walks, trains an autoregressive language model to reconstruct masked polygons from BEOL context, and samples at high temperature with rule-conditioned prompts; the resulting layouts are annotated by an industrial DRC checker. A teacher VLM then generates reasoning traces that are filtered and used to fine-tune Qwen3-VL-8B into a domain-adapted DRC-VLM. Phase II supplies the DRC-VLM's structured analysis to coding agents (Claude Code, Codex) that edit local GDS with DRC/LVS checker-in-the-loop feedback. On 100 real sub-2nm cases, the paper reports solve rates rising from 63.0% to 88.0% for Claude Code and from 85.0% to 97.0% for Codex.

Significance. If the reported +12–25% solve-rate improvement is real and attributable to the DRC-VLM, this is a significant contribution: it offers a scalable self-supervised route to DRC-annotated layout–violation training data and demonstrates that a domain-adapted VLM can ground conditional design rules for agentic layout repair. The evaluation's use of an external signoff DRC checker and LVS, the held-out test set from a different IP, and the explicit disclosure of the small/curated dataset and the three-case Area category are strengths. However, the central attribution claim is currently under-supported because the Phase II comparison lacks a non-domain-adapted analysis control, and the layout representation's rectangularity assumption is unstated and untested.

major comments (3)
  1. [§4.3, Table 6] The central claim—that the DRC-VLM's domain-adapted reasoning boosts solve rates by +12–25%—rests on comparing agents with DRC-VLM analysis against agents with no analysis arm. This entangles two variables: the presence of any structured expert-analysis text and the specific domain adaptation of the VLM. Table 5 shows the DRC-VLM's Loc F1@1 is only 23.5%, so a non-domain-adapted VLM (e.g., Gemini-3-Pro or Qwen3-VL-8B-Base) or a template that restates the DRC report might produce a similar boost. Please add a control condition in which the same agents receive an equally structured analysis from a generic VLM or a rule-template parser. Without such a control, the abstract's claim 'This VLM provides rule-aware geometric guidance ... boosting' is not supported.
  2. [§3.1.1, Fig. 2] The text-formulated representation encodes each rectangle as a directional edge walk from its top-left vertex via four cardinal deltas. The paper never states that all target polygons are assumed to be axis-aligned rectangles. Real sub-2nm layouts commonly contain non-rectangular multi-vertex polygons (jogs, via arrays, tapered geometries). If such shapes are either decomposed into multiple rectangles or treated only as boundary-only [BPoly] context, the generated corpus and the diversity/rule-coverage numbers in Table 3 cover only a subset of the layout space, and the t-SNE 'in-distribution' claim in Fig. 3 would overstate coverage. Please clarify whether the representation is lossless for all polygons in the target layer/rule subset, or state the rectangularity restriction and quantify its impact (e.g., fraction of real polygons in the 100 test cases that are exactly rectangular).
  3. [§4.3, Table 6] The solve-rate comparison is based on a single agent run per problem ('We run the agent fix once per problem'). No confidence intervals, standard errors, or repeated trials are reported, and several per-category counts are small (EN: 39, Area: 3). Given that the headline contribution is the +12–25% numerical range, the paper should report binomial confidence intervals or a small-seed sensitivity analysis to show that the differences are not within run-to-run variability of the agents.
minor comments (4)
  1. [§4.1, Table 3] The D3PM baseline is cited to [25] (PatternPaint). Please clarify whether this is the same discrete-diffusion formulation as in [23]/[27], and state any differences in training data or sampling.
  2. [§4.2, Table 5] The ablation names 'Mixed-NoR-Script' and 'Mixed-NoR-AutoReg' are not expanded. Define 'NoR' and specify which data components (reasoning traces? description tasks?) are removed in each ablation.
  3. [§4.3, Table 6] The token-cost parentheses use an unclear sign convention (e.g., Codex Width shows '(-1.4%)' for an increase; Codex Color Spacing shows '(-4.9%)' for an increase). Define the sign uniformly and use a consistent label such as '% change from baseline'.
  4. [§3.1.1, §3.1.2] Reproducibility would benefit from reporting the generation temperature τ, the trace-filtering thresholds used in §3.1.2, and the final corpus sizes after all filtering stages. These are currently described only qualitatively.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: central results are measured against held-out real layouts and an external DRC checker; remaining concerns are experimental confounds, not definitional circularity.

full rationale

The paper's derivation chain is empirically anchored: Phase I trains an autoregressive generator by masked polygon reconstruction from real BEOL context (no violation labels), then an industrial signoff DRC checker (Calibre) provides ground-truth violation labels for the synthetic corpus. The DRC-VLM is fine-tuned on this corpus plus filtered teacher traces, but its detection evaluation (Table 5) uses a 200-sample held-out set from a different IP with disjoint routing patterns, and the fixing evaluation (Table 6) measures solve rates with the criterion 'A case is solved if the target DRV is fixed without introducing new DRVs in the region and the design passes LVS.' Thus the central claim—that DRC-VLM guidance raises solve rates by +12–25%—is not a fitted value or a prediction that reduces to its training inputs; it is an externally measured outcome on real cases. The Phase II comparison lacks a generic-VLM/structured-analysis control, so the improvement could partly reflect the presence of any additional expert hint rather than the domain-adapted DRC-VLM specifically. This is a validity threat but not a circularity: no equation defines the claimed result in terms of its own inputs, and no load-bearing argument rests on a self-citation. Self-citations in the related work (DRC-Coder, Polymath, PatternPaint) are contextual or baseline, not used to prove the main result. The paper's own limitation note about the Area category (3 cases being too small) is acknowledged but does not affect circularity. Overall, the derivation is self-contained against external benchmarks; score 1 reflects only minor, non-load-bearing self-citations and methodological concerns that belong to correctness risk rather than circularity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The central claim rests on four domain-specific assumptions (checker-as-ground-truth, rectangular serialization, teacher-trace quality, learnability of the layout distribution) and one standard statistical estimation method. The main ad-hoc-to-paper element is the rectangle-only serialization, which is neither stated as a limitation nor tested. The free-parameter count is dominated by hand-set generation and curation choices (temperature, filter thresholds, corpus size, rule subset, top-k), none of which are given sensitivity analysis.

free parameters (5)
  • Generation temperature τ = not reported (described only as high temperature)
    Inference-time sampling temperature in Phase I (y_t ~ softmax(z_t/τ), §3.1.1) controls the diversity of generated violation-prone layouts; the paper never gives the value, yet the diversity and rule-coverage results in Table 3 depend on it.
  • Trace-filtering thresholds = 13.40% rejection (image); 30.66% (text)
    Multi-stage filtering of teacher reasoning traces (§3.1.2, Table 2) is a hand-set quality gate that shapes the DRC-VLM's training supervision; the per-stage thresholds are not specified.
  • Synthetic corpus size = ~15K samples per method (16,861 VIA / 17,254 metal for proposed)
    The number of generated samples and the filtering to remove near-duplicate DRVs determine the VLM SFT data; whether ~15K is sufficient is not analyzed, e.g., with a scaling curve.
  • Rule subset and fixing-case distribution = 100 cases: EN 39, Width 27, Spacing 21, Color Spacing 10, Area 3
    The DRC-VLM and the evaluation are scoped to a hand-selected subset of local P&R rules; the 100-case curation (§4.3) sets the denominator for the headline +12–25% numbers, and Area is too small (3 cases) for conclusions.
  • Top-k fix actions = k = 3
    Phase II uses the VLM's top-3 potential fixes as agent guidance (§3.2); no sensitivity analysis on k is reported.
assumptions (5)
  • domain assumption Calibre (or equivalent signoff DRC checker [1]) output is ground truth for violation labels in both synthetic training data and test-set evaluation.
    The entire pipeline uses the industrial checker to annotate generated layouts (§3.1.1) and to judge fixes (§4.3); if the checker's rule interpretation is imperfect, labels and evaluation inherit the error. The paper does not independently verify checker judgments.
  • ad hoc to paper Target-layer geometry can be losslessly serialized as axis-aligned rectangles via four cardinal deltas (→w, ↓h, ←w, ↑h).
    §3.1.1 defines the text formulation with per-rectangle edge walks and [Poly]/[BPoly] tokens. Non-rectangular multi-vertex polygons, common in advanced-node BEOL, are not handled in the generation target representation; the paper does not state this restriction explicitly.
  • domain assumption The teacher model (Gemini-3-Pro) reasoning traces, after filtering, are accurate enough to supervise the DRC-VLM.
    §3.1.2 uses Gemini-3-Pro to generate DRC reasoning traces and retains 86.6% of image-input traces; if the teacher frequently misidentifies rules or locations on retained samples, the VLM inherits the bias. The paper's own Table 5 shows Gemini-3-Pro has Loc F1 @1 of only 14.4%.
  • domain assumption Masked-polygon reconstruction from BEOL context alone teaches the model the true distribution of routed layouts.
    The self-supervised objective in §3.1.1 (teacher-forced NLL over masked polygons) assumes that context is sufficient to predict plausible geometry; Table 3 and t-SNE give partial empirical support, but there is no quantitative coverage test against the real layout distribution.
  • standard math Pass@k estimation from n=10 trials per sample yields unbiased performance estimates for the detection task.
    §4.2 aggregates Rule/Loc F1 using the Chen et al. pass@k estimator [33] over 10 trials; this assumes trial independence and is applied consistently, which is standard practice.
invented entities (2)
  • [BPoly] boundary-polygon token
    purpose: Distinguishes context-only shapes clipped by the crop boundary from complete generation targets, preventing the model from treating clipped shapes as valid outputs or hallucinating edge geometry.
    A representational device introduced in §3.1.1 (Fig. 2). It is not a physical entity and makes no falsifiable prediction outside the pipeline; its value is bound to the paper's own generation quality metrics.
  • Text-formulated layout serialization (layer headers + cardinal-delta rectangle walks)
    purpose: Casts multi-layer GDS geometry into a token stream so an autoregressive LLM can reconstruct masked polygons and decode them back into DRC-annotatable shapes.
    New technical representation (§3.1.1); no independent handle beyond the paper's DRC-checked decoding. Any 'correctness' claim for the representation rests on the paper's own entropy and rule-coverage experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SCALE: Self-Supervised Constraint-Aware Layout GEneration for Local P&R DRV Fixing at Advanced Nodes." pith.science (2026). https://pith.science/paper/EFWGNXRI

@misc{pith2026260721850,
  author       = {Pith},
  title        = {Pith review of: SCALE: Self-Supervised Constraint-Aware Layout GEneration for Local P&R DRV Fixing at Advanced Nodes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EFWGNXRI}},
  note         = {Machine review of arXiv:2607.21850}
}
read the original abstract

As semiconductor manufacturing advances toward sub-2nm nodes, local place-and-route (P&R) design-rule violation (DRV) fixing is increasingly limited by complex rule interactions, dense multi-layer routing geometries, and foundry-specific constraints. While Large Language Models (LLMs) have recently demonstrated strong capabilities in EDA scripting and documentation, their application to visual layout understanding remains largely unexplored: diagnosing DRC violations from layout imagery demands precise geometric reasoning and foundry-specific rule knowledge absent from general-purpose VLM training. We propose SCALE, a framework with a self-supervised layout-generation stage for local DRV fixing at advanced nodes. Multi-layer layout geometry is serialized into structured text, and a fine-tuned language model learns to reconstruct randomly masked polygons from surrounding BEOL context alone without violation labels. At inference, natural-language rule constraints and high-temperature sampling steer generation toward diverse, violation-prone layout variants validated by an industrial signoff DRC checker, producing DRC-annotated layout--violation pairs used to fine-tune a domain-adapted DRC-VLM. This VLM provides rule-aware geometric guidance for local DRV repair, boosting state-of-the-art agents' solve rates by +12--25% (up to 97%) on 100 real sub-2nm cases spanning enclosure, spacing, width, and color-spacing violations.

Figures

Figures reproduced from arXiv: 2607.21850 by the authors.

Figure 1
Figure 1. Off-the-shelf closed-source VLMs hallucinate geometric fea [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Proposed Framework. Phase I: Layouts are serialized into text to train an autoregressive model; text-conditioned sampling generates [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. t-SNE Diversity Analysis. t-SNE projection comparing syn [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: An example of the VLM detection task used for SFT. Each SFT [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Examples of DRC fix approaches for VIA spacing (top) and metal tip-to-tip (bottom) rules. (a) Target layouts with the original [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 6 linked inside Pith

  1. [1]

    Calibre nmDRC hierarchical design rule checking

    Siemens EDA, “Calibre nmDRC hierarchical design rule checking. ” https://eda.sw. siemens.com/en-US/ic/calibre-design/drc/, 2024

  2. [2]

    Innovus implementation system

    Cadence Design Systems, “Innovus implementation system. ” https: //www.cadence.com/en_US/home/tools/digital-design-and-signoff/soc- implementation-and-floorplanning/innovus-implementation-system.html, 2024

  3. [3]

    IC Compiler II place and route solution

    Synopsys, Inc., “IC Compiler II place and route solution. ” https://www.synopsys. com/implementation-and-signoff/physical-implementation/ic-compiler.html, 2024

  4. [4]

    DRCPlus in a router: automatic elimination of lithography hotspots using 2D pattern detection and correction,

    J. Yang, N. Rodriguez, O. Omedes, F. Gennari, Y.-C. Lai, and V. Mankad, “DRCPlus in a router: automatic elimination of lithography hotspots using 2D pattern detection and correction, ” inDesign for Manufacturability through Design-Process Integration IV(M. L. Rieger and J. Thiele, eds.), vol. 7641, p. 76410Q, International Society for Optics and Photonics...

  5. [5]

    Orassistant: A custom rag-based conversational assistant for openroad,

    A. Kaintura, P. R, S. S. Luar, and I. I. Almeida, “Orassistant: A custom rag-based conversational assistant for openroad, ” 2024

  6. [6]

    Drc-coder: Automated drc checker code generation using llm autonomous agent,

    C.-C. Chang, C.-T. Ho, Y. Li, Y. Chen, and H. Ren, “Drc-coder: Automated drc checker code generation using llm autonomous agent, ” inProceedings of the 2025 Interna- tional Symposium on Physical Design, ISPD ’25, (New York, NY, USA), p. 143–151, Association for Computing Machinery, 2025

  7. [7]

    Rule2drc: Benchmarking llm agents for drc script synthesis with execution-guided test generation,

    J. Kim, J. Byun, D. Hwang, S.-J. Park, and H. O. Song, “Rule2drc: Benchmarking llm agents for drc script synthesis with execution-guided test generation, ”arXiv preprint arXiv:2605.15669, 2026

  8. [8]

    Customized retrieval augmented genera- tion and benchmarking for eda tool documentation qa,

    Y. Pu, Z. He, T. Qiu, H. Wu, and B. Yu, “Customized retrieval augmented genera- tion and benchmarking for eda tool documentation qa, ” inProceedings of the 43rd IEEE/ACM International Conference on Computer-Aided Design, ICCAD ’24, (New York, NY, USA), Association for Computing Machinery, 2024

Show all 36 references
  1. [9]

    Qwen3-VL: Enhanced vision-language model

    J. Baiet al., “Qwen3-VL: Enhanced vision-language model. ” https://github.com/ QwenLM/Qwen3-VL, 2025

  2. [10]

    Schemacoder: Auto- matic log schema extraction coder with residual q-tree boosting,

    L. J. Wan, C.-T. Ho, R. Liang, C. Yu, D. Chen, and H. Ren, “Schemacoder: Auto- matic log schema extraction coder with residual q-tree boosting, ”arXiv preprint arXiv:2508.18554, 2025

  3. [11]

    Chipnemo: Domain-adapted llms for chip design,

    M. Liu, T.-D. Ene, R. Kirby, C. Cheng, N. Pinckney, R. Liang, J. Alben, H. Anand, S. Banerjee, I. Bayraktaroglu,et al., “Chipnemo: Domain-adapted llms for chip design, ”arXiv preprint arXiv:2311.00176, 2023

  4. [12]

    Verilogcoder: Autonomous verilog coding agents with graph-based planning and abstract syntax tree (ast)-based waveform tracing tool,

    C.-T. Ho, H. Ren, and B. Khailany, “Verilogcoder: Autonomous verilog coding agents with graph-based planning and abstract syntax tree (ast)-based waveform tracing tool, ” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, pp. 300–307, 2025

  5. [13]

    ScaleRTL: Scaling LLMs with reasoning data and test-time compute for accurate RTL code generation,

    Z. Donget al., “ScaleRTL: Scaling LLMs with reasoning data and test-time compute for accurate RTL code generation, ” inProceedings of the ACM/IEEE Workshop on Machine Learning for CAD (MLCAD), 2025. arXiv:2506.05566

  6. [14]

    CraftRTL: High-quality synthetic data generation for Verilog code models with correct-by-construction non-textual representations and targeted code repair,

    M. Liu, W. Tsai, and H. Ren, “CraftRTL: High-quality synthetic data generation for Verilog code models with correct-by-construction non-textual representations and targeted code repair, ” inProceedings of the International Conference on Learning Representations (ICLR), 2025

  7. [15]

    Fvrulelearner: Operator-level reasoning tree (op-tree)-based rules learning for formal verification,

    L. J. Wan, C.-T. Ho, Y. Bai, C. Yu, D. Chen, and H. Ren, “Fvrulelearner: Operator-level reasoning tree (op-tree)-based rules learning for formal verification, ”arXiv preprint arXiv:2604.03245, 2026

  8. [16]

    Polymath: Self-improving hierarchical workflow for multi-domain problem solving,

    C.-T. Ho, J. Gong, H. Yang, A. Akkur, and H. Ren, “Polymath: Self-improving hierarchical workflow for multi-domain problem solving, ” inProceedings of the 2026 International Symposium on Physical Design, pp. 121–130, 2026

  9. [17]

    Spatialvlm: Endowing vision-language models with spatial reasoning capabilities,

    B. Chen, Z. Xu, S. Kirmani, B. Ichter, D. Driess, P. Florence, D. Sadigh, L. Guibas, and F. Xia, “Spatialvlm: Endowing vision-language models with spatial reasoning capabilities, ” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. ...

  10. [18]

    iVISPAR – an interactive visual-spatial reasoning benchmark for VLMs,

    J. Mayer, M. Ballout, S. Jassim, F. Nosrat Nezami, and E. Bruni, “iVISPAR – an interactive visual-spatial reasoning benchmark for VLMs, ” inProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP), (Suzhou, China), pp. 26757–26781, 2025

  11. [19]

    Deepattern: Layout pattern generation with transforming convolutional auto-encoder,

    H. Yang, S. Li, W. Chen, P. Pathak, F. Gennari, Y.-C. Lai, and B. Yu, “Deepattern: Layout pattern generation with transforming convolutional auto-encoder, ”IEEE Transactions on Semiconductor Manufacturing, vol. 35, no. 1, pp. 67–77, 2022

  12. [20]

    Layout pattern generation and legalization with generative learning models,

    X. Zhang, J. Shiely, and E. F. Young, “Layout pattern generation and legalization with generative learning models, ” in2020 IEEE/ACM International Conference On Computer Aided Design (ICCAD), pp. 1–9, 2020

  13. [21]

    Attentional transfer is all you need: Technology- aware layout pattern generation,

    X. Zhang, H. Yang, and E. F. Young, “Attentional transfer is all you need: Technology- aware layout pattern generation, ” in2021 58th ACM/IEEE Design Automation Con- ference (DAC), pp. 169–174, 2021

  14. [22]

    Layoutransformer: Gener- ating layout patterns with transformer via sequential pattern modeling,

    L. Wen, Y. Zhu, L. Ye, G. Chen, B. Yu, J. Liu, and C. Xu, “Layoutransformer: Gener- ating layout patterns with transformer via sequential pattern modeling, ” inPro- ceedings of the 41st IEEE/ACM International Conference on Computer-Aided Design, ICCAD ’22, (New York, NY, USA),...

  15. [23]

    Diffpattern: Layout pattern generation via discrete diffusion,

    Z. Wang, Y. Shen, W. Zhao, Y. Bai, G. Chen, F. Farnia, and B. Yu, “Diffpattern: Layout pattern generation via discrete diffusion, ” in2023 60th ACM/IEEE Design Automation Conference (DAC), pp. 1–6, 2023. Chia-Tung Ho1,∗, Haoyu Yang1,∗, Guanglei Zhou 1,2,∗, Yoshi Nishi1, Yaguan...

  16. [24]

    Controlayout: Conditional diffusion for style-controllable and violation-fixable layout pattern generation,

    Q. Wang, X. Zhang, M. D. Wong, and E. F. Young, “Controlayout: Conditional diffusion for style-controllable and violation-fixable layout pattern generation, ” in Proceedings of the Great Lakes Symposium on VLSI 2024, GLSVLSI ’24, (New York, NY, USA), p. 511–515, Association fo...

  17. [25]

    Patternpaint: Practical layout pattern generation using diffusion-based inpainting,

    G. Zhou, B. Korrapati, G. R. Reddy, C.-C. Chang, J. Pan, J. Hu, Y. Chen, and D. G. Thakurta, “Patternpaint: Practical layout pattern generation using diffusion-based inpainting, ” inProceedings of the 62nd Annual ACM/IEEE Design Automation Con- ference, IEEE Press, 2025

  18. [26]

    Vario: Enhance pattern diversity with diffusion model,

    G. Zhou, B. Korrapati, G. R. Reddy, J. Zhang, Y. Chen, and D. G. Thakurta, “Vario: Enhance pattern diversity with diffusion model, ” inProceedings of SPIE Advanced Lithography + Patterning, vol. 13425, 2025

  19. [27]

    Diffusion-model-enhanced layout pattern generation for sub-3nm DFM,

    G. Zhou, B. Korrapati, G. R. Reddy, C.-C. Chang, J. Hu, Y. Chen, and D. G. Thakurta, “Diffusion-model-enhanced layout pattern generation for sub-3nm DFM, ” inPro- ceedings of the 44th IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2025. Invited paper

  20. [28]

    Enhanced hotspot detection through synthetic pattern generation and design of experiments,

    G. R. Reddy, C. Xanthopoulos, and Y. Makris, “Enhanced hotspot detection through synthetic pattern generation and design of experiments, ” in2018 IEEE 36th VLSI Test Symposium (VTS), pp. 1–6, 2018

  21. [29]

    Viper: A versatile and intuitive pattern generator for early design space exploration,

    G. R. Reddy, M.-M. Bidmeshki, and Y. Makris, “Viper: A versatile and intuitive pattern generator for early design space exploration, ” in2019 IEEE International Test Conference (ITC), pp. 1–7, 2019

  22. [30]

    Design rule manual and drc code qualification flows empowered by high coverage synthetic layouts generation,

    A. Kabeel, S. Kim, Y. G. Park, D. Kim, J. Kwan, S. Rizk, K. Madkour, M. Shafee, and J. Kim, “Design rule manual and drc code qualification flows empowered by high coverage synthetic layouts generation, ” inDTCO and Computational Patterning II, vol. 12495, pp. 415–428, SPIE, 2023

  23. [31]

    Enhanced OPC recipe coverage and early hotspot detection through automated layout generation and analysis,

    A. Hamouda, M. Bahnas, D. Schumacher, I. Graur, A. Chen, K. Madkour, H. Ali, J. Meiring, N. Lafferty, and C. McGinty, “Enhanced OPC recipe coverage and early hotspot detection through automated layout generation and analysis, ” inOpti- cal Microlithography XXX(A. Erdmann and J...

  24. [32]

    Claude (computer software)

    Anthropic, “Claude (computer software). ” https://anthropic.com, 2026. Version Claude 3.5 Sonnet. Code generated on July 5, 2026

  25. [33]

    Evaluating large language models trained on code,

    M. Chen, J. Tworek, H. Jun, Q. Yuan, H. P. d. O. Pinto, J. Kaplan, H. Edwards, Y. Burda, N. Joseph, G. Brockman,et al., “Evaluating large language models trained on code, ” arXiv preprint arXiv:2107.03374, 2021

  26. [34]

    ” Computer software, 2026

    OpenAI, “Codex. ” Computer software, 2026

  27. [35]

    Claude sonnet 4.6 interaction

    Anthropic, “Claude sonnet 4.6 interaction. ” Generative AI chat, 2026. Chat retrieved on July 9, 2026

  28. [36]

    ChatGPT

    OpenAI, “ChatGPT. ” https://chat.openai.com, 2026. [Large language model]

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.