Pith. sign in

REVIEW 3 major objections 5 minor 35 references

The paper claims a diffusion-based pipeline can generate new, valid RTL circuits and that these improve ML-based PPA prediction.

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 ↗

SynCircuit generates new, structurally valid RTL circuits with a directed-cyclic-graph diffusion model plus post-processing and MCTS, and shows they improve ML-based PPA prediction when added to training data.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A genuinely new DCG-diffusion pipeline for synthetic RTL that overclaims 'valid functionality' — structurally valid and useful for data augmentation, but needs functional verification and claim-tightening. the 3 major comments →

arxiv 2509.00071 v1 pith:TKGSAR55 submitted 2025-08-26 cs.LG cs.AI

SynCircuit: Automated Generation of New Synthetic RTL Circuits Can Enable Big Data in Circuits

Shang Liu , Jing Wang , Wenji Fang , Zhiyao Xie This is my paper
classification cs.LG cs.AI
keywords synthetic RTL generationdirected cyclic graphsdiffusion modelsMonte Carlo tree searchlogic redundancyRTL PPA predictiondata augmentationcircuit graph generation
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.

The reading

SynCircuit tries to solve a concrete bottleneck: AI-assisted chip design needs large amounts of circuit data, but open-source RTL designs are scarce. The paper's claim is that this scarcity can be removed by generating new, valid synthetic circuits automatically—the first such framework, to the authors' knowledge—using a diffusion model over directed cyclic graphs, a post-processing pass that enforces circuit rules, and a Monte Carlo tree search that trims redundant logic. If the claim holds, ML models for early-stage power, performance, and area prediction can be trained on synthetic designs instead of waiting for more real chips to be open-sourced. The paper's experiments show structural similarity to real circuits and improved PPA prediction when synthetic designs are added to small training sets, with the largest gains when only five real designs are available. The central bet, stated plainly by the authors, is that two structural constraints are enough to call a generated circuit 'valid'; their own redundancy metric makes that bet the part to watch.

Core claim

The paper's central claim is that circuit data scarcity can be resolved by generation, not collection. SynCircuit represents RTL designs as directed cyclic graphs and trains a diffusion model to produce new such graphs; it then enforces circuit validity by fixing fan-in counts and eliminating combinational loops, and uses Monte Carlo tree search to cut logic redundancy. The authors report that the resulting graphs match real circuits on degree, clustering, orbit, and correlation statistics better than GraphRNN, D-VAE, GraphMaker, and SparseDigress, and that augmenting small real training sets with 25 synthetic designs improves RTL-stage PPA prediction across register slack, WNS, TNS, and are

What carries the argument

The central object is the directed cyclic graph (DCG) representation of RTL, in which nodes carry operator/register/port types and widths and edges carry signal flow. Generated graphs are treated as adjacency matrices and learned with a discrete diffusion model; directionality is handled by a translated-embedding decoder that scores an edge i→j via P = MLP([(H_i + r) ⊙ H_j] ⊕ d), breaking the symmetry that makes dot-product or distance decoders blind to edge direction. Two refinement stages then carry the argument: a probability-guided sequential post-processor that repairs the graph by adding parent edges in order of model confidence while blocking register-free cycles, and an MCTS phase wh

Load-bearing premise

The paper's load-bearing premise is that a generated circuit is valid once it satisfies two structural rules—each node has the correct number of parent nodes and no register-free cycle exists—without separately proving that the circuit's logic is non-degenerate.

What would settle it

Take a random batch of SynCircuit-generated HDL designs, pass each through a commercial synthesis tool, and count the fraction whose sequential-cell preservation ratio (SCPR) falls in the 70–100% range typical of real designs. If most generated circuits still lose more than half their registers to synthesis—as the paper's own Figure 4(a) shows for the worst Gval examples—then the two structural constraints are insufficient to establish 'valid functionality', and the central claim fails.

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

If this is right

  • With a trained SynCircuit model, a user specifies node count and attributes, and the pipeline returns an unlimited number of distinct HDL designs that can be synthesized into netlists and layouts with commercial tools.
  • Adding 25 SynCircuit-generated designs to a 15-design real training set improves RTL-stage register-slack prediction (MAPE 27% to 16%) and area prediction (RRSE 0.62 to 0.34); gains are larger when the real set is only 5 designs.
  • Generated graphs reproduce the scale-free degree distribution of real circuits, which is what makes them more useful for training than graphs from undirected or DAG-only baselines.
  • Unoptimized synthetic circuits can hurt downstream predictors because synthesis removes too much logic; the MCTS optimization phase is what closes the gap, implying redundancy control is essential for any synthetic-data pipeline.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Read strictly, SynCircuit's validity check guarantees only fan-in counts and loop-freedom; it does not prove a generated design computes a useful function. The paper's own Figure 4(a) shows synthesis can still discard most registers in unoptimized generations, so a natural hardening is to filter or score designs by SCPR before release.
  • The asymmetric translated-embedding decoder is a general trick for directed cyclic graphs. It should transfer to other domains with feedback structure—recurrent architectures, cyclic dataflow, or control-flow graphs—where topological-order autoregressive generation is impossible.
  • The MCTS loop with a learned proxy for synthesis cost points to a general recipe: generate candidates cheaply, then search against a learned surrogate of the expensive tool. Applying it to timing or power instead of area would test whether redundancy is really the only dimension that matters.
  • A large automated sweep—generate hundreds of designs, synthesize all of them, and compare the full SCPR and timing distributions to real designs—would turn the paper's selected-case evidence into a quantitative guarantee.
Share X Bluesky LinkedIn Reddit HN

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

3 major / 5 minor

Summary. SynCircuit proposes a three-phase pipeline for generating synthetic RTL circuits in HDL form: a diffusion-based generative model for directed cyclic graphs (DCG), a probability-guided post-processing step that enforces fanin-count and combinational-loop constraints, and an MCTS-based refinement step that optimizes a post-synthesis size proxy to reduce logic redundancy. The paper evaluates the generated graphs on structural similarity metrics and on their usefulness as extra training data for ML-based RTL PPA prediction, reporting improvements over baselines such as GraphRNN and DVAE. The central advertised claim is that SynCircuit is the first technique to automatically generate brand-new synthetic circuits with 'valid functionalities.'

Significance. If the central claim were fully established, SynCircuit could offer a scalable answer to the circuit-data scarcity problem in AI-assisted IC design, and its DCG diffusion framework with an asymmetric edge decoder would be a useful contribution in its own right. The authors also provide a concrete downstream application (PPA prediction augmentation) and an MCTS optimization method with a novel reward proxy. These are genuine strengths. However, the paper's own evidence does not yet support the 'valid functionality' claim, because the enforced constraints are structural only and no functional verification is provided. The downstream results are promising but lack statistical grounding. The framework is worth considering for publication after the load-bearing validity and evaluation issues are addressed.

major comments (3)
  1. [Section II (Circuit constraints C), Section VI, Figure 4(a)] The definition of 'valid' is purely structural: correct fanin count per node type and absence of combinational loops. These conditions guarantee only that the graph can be parsed into HDL, not that the HDL implements meaningful nontrivial logic. The paper's own SCPR results show that some Gval circuits have SCPR below 10%, meaning synthesis discards over 90% of sequential cells, which is strong evidence that much of the generated logic is constant, unreachable, or otherwise redundant. The constraints also do not check signal widths or operand types, so a graph satisfying C could yield type-incorrect HDL. The Abstract and Section I claims of 'valid functionalities' are therefore unsupported. I request either functional verification (simulation with testbenches, equivalence checking against a reference, or formal property checks; at minimum, report the fraction of generated circuits that s
  2. [Section VII.A (SynCircuit setup), Section VI.B, Figure 3(c)] The MCTS reward uses 'a trained discriminator to approximate the PCS,' but the paper provides no architecture, training data, or accuracy numbers for this discriminator. The optimization results in Figure 4 and the downstream improvements in Table III depend critically on this proxy. If the discriminator is inaccurate, the claimed reduction in logic redundancy may be spurious. Please report the discriminator's correlation or error against actual synthesis PCS on a held-out set, and ideally run an ablation with true PCS on a subset of register cones.
  3. [Table III (downstream PPA prediction)] The reported improvements are point estimates without error bars, multiple seeds, or significance tests. For example, Register Slack MAPE improves from 27% to 16% in the 15-design setting and from 34% to 25% in the 5-design setting, but with only 7 test designs and 25 generated circuits per augmentation, these gains could be within noise. Please report mean and standard deviation over at least 3 random seeds or provide per-test-design results. The paper should also clarify that the graph generative model was trained on the same 15 designs that form the 'basic training data' in Table III(a); the 5-design setting is less affected but still includes overlap, so the augmentation benefit under a true data-scarcity scenario remains to be demonstrated.
minor comments (5)
  1. [Title and Section VI.A] There is a spacing error in the title ('Syn thetic') and a typo in the Section VI.A heading ('Challengs'). Please correct.
  2. [Section IV.D equation] The decoder equation for P(t-1)_E has unclear notation (e.g., the 'n' and the use of ⊙ with concatenation). Please reformat and define all symbols.
  3. [Section VII.B.1, Table II] The W1 distances and expectation ratios are reported as point estimates without the number of generated graphs or variance. Please state how many graphs were generated per model and add confidence intervals or multiple-run statistics so the 'more realistic' claim can be assessed.
  4. [Table III caption and body] The notation 'R→1' is undefined. Clarify that R is the Pearson correlation coefficient and that values closer to 1 are better, and define MAPE/RRSE consistently in one place.
  5. [Overall] The paper uses 'SCPR' and 'register ratio' interchangeably; choose one term. Also, the statement that 'digital circuits are indeed scale-free networks' is a strong empirical claim; cite a circuit-specific study or soften the wording.

Circularity Check

0 steps flagged

No significant circularity: SynCircuit's claims are empirical and externally evaluated; the 'valid functionality' overclaim is an unestablished assumption, not a derivation that reduces to its inputs.

full rationale

This is an empirical systems paper, not a derivation. Phase 1 trains a diffusion model on 15 open-source designs; Phase 2 enforces the two structural constraints C by construction; Phase 3 optimizes an MCTS proxy (PCS) and is evaluated with actual Synopsys synthesis. The downstream PPA experiment is a held-out evaluation: the 7 test designs are excluded from generative-model training, and the augmented training sets are compared against GraphRNN, DVAE, SparseDigress, and GraphMaker using a standard (if same-group) RTL PPA predictor. None of the reported numbers is a fitted parameter renamed as a prediction: the MCTS reward is a surrogate for PCS, but the reported SCPR, WNS, and TNS values come from the real synthesis toolchain, not from that surrogate. The same-group citations (MasterRTL, RTL-Timer) are used as evaluation harnesses, not as evidence for the central claim, so they are not load-bearing. The paper's genuine weakness is a support gap, not circularity: 'valid' is stipulated in Section II as fanin count plus acyclic combinational logic, while the abstract claims 'valid functionalities'; Section VI/Figure 4(a) admits that 'some synthetic designs Gval can not even reach a 10% SCPR,' so the semantic meaningfulness of the generated circuits is not established. That is an assumption about what makes circuits useful, not a step where an output equals an input by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities. Its assumptions are domain-specific: representation equivalence, constraint sufficiency, dataset representativeness, and proxy model accuracy. The most fragile is the constraint sufficiency assumption, which the paper's own redundancy data call into question.

axioms (4)
  • domain assumption The parser mapping between HDL and circuit graph G is a bijection.
    Section II states 'Given a circuit design HDL code D, it can be mapped to a circuit graph representation G through a bijection function f' without proof or discussion of edge cases.
  • ad hoc to paper The constraint set C (correct fanin count and no combinational loops) is sufficient for circuit validity.
    Section II defines validity solely by these two constraints, but the paper's own SCPR results (Section VI) show that many such circuits are mostly optimized away by synthesis, indicating they are not functionally meaningful.
  • domain assumption The 15 real designs used for training the generator are representative of the circuit distribution.
    The generator is trained on only 15 open-source designs (Section VII.A). Generalization to the broader space of RTL designs is assumed without evidence.
  • domain assumption The trained discriminator accurately approximates post-synthesis area (PCS) during MCTS.
    Section VII.A states the synthesis tool is replaced with a trained discriminator, but its architecture, training data, and accuracy are not reported, so the MCTS guidance may be unreliable.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of SynCircuit: Automated Generation of New Synthetic RTL Circuits Can Enable Big Data in Circuits." pith.science (2026). https://pith.science/paper/TKGSAR55

@misc{pith2026250900071,
  author       = {Pith},
  title        = {Pith review of: SynCircuit: Automated Generation of New Synthetic RTL Circuits Can Enable Big Data in Circuits},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TKGSAR55}},
  note         = {Machine review of arXiv:2509.00071}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

In recent years, AI-assisted IC design methods have demonstrated great potential, but the availability of circuit design data is extremely limited, especially in the public domain. The lack of circuit data has become the primary bottleneck in developing AI-assisted IC design methods. In this work, we make the first attempt, SynCircuit, to generate new synthetic circuits with valid functionalities in the HDL format. SynCircuit automatically generates synthetic data using a framework with three innovative steps: 1) We propose a customized diffusion-based generative model to resolve the Directed Cyclic Graph (DCG) generation task, which has not been well explored in the AI community. 2) To ensure our circuit is valid, we enforce the circuit constraints by refining the initial graph generation outputs. 3) The Monte Carlo tree search (MCTS) method further optimizes the logic redundancy in the generated graph. Experimental results demonstrate that our proposed SynCircuit can generate more realistic synthetic circuits and enhance ML model performance in downstream circuit design tasks.

Figures

Figures reproduced from arXiv: 2509.00071 by Jing Wang, Shang Liu, Wenji Fang, Zhiyao Xie.

Figure 1
Figure 1. Figure 1: The overview of SynCircuit. It addresses the serious design [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SynCircuit framework. The circuit generation process starts with the specified node number [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: MCTS-based circuit optimization flow overview. (a) The search tree is on the left and each state node represents an adjacency matrix. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: (a) Logic redundancy metric SCPR is calculated by dividing [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Netlist statistics for the three synthetic datasets and real [PITH_FULL_IMAGE:figures/full_fig_p006_5.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

35 extracted references · 26 canonical work pages · 1 internal anchor

  1. [1]

    Betterv: Controlled ver- ilog generation with discriminative guidance,

    Z. Pei, H. Zhen, M. Yuan, Y . Huang, and B. Yu, “Betterv: Controlled ver- ilog generation with discriminative guidance,” in Forty-first International Conference on Machine Learning (ICML) , 2024

    work page 2024

  2. [2]

    Lamagic: Language-model-based topology generation for analog integrated circuits,

    C.-C. Chang, Y . Shan, S. Fan, J. Li, S. Zhang, N. Cao, Y . Chen, and X. Zhang, “Lamagic: Language-model-based topology generation for analog integrated circuits,” arXiv preprint arXiv:2407.18269 , 2024

    Pith/arXiv arXiv 2024

  3. [3]

    Rtlcoder: Outperforming gpt-3.5 in design rtl generation with our open-source dataset and lightweight solution,

    S. Liu, W. Fang, Y . Lu, Q. Zhang, H. Zhang, and Z. Xie, “Rtlcoder: Outperforming gpt-3.5 in design rtl generation with our open-source dataset and lightweight solution,” arXiv preprint arXiv:2312.08617, 2023

    Pith/arXiv arXiv 2023

  4. [4]

    Towards a comprehensive benchmark for high-level synthesis targeted to fpgas,

    Y . Bai, A. Sohrabizadeh, Z. Qin, Z. Hu, Y . Sun, and J. Cong, “Towards a comprehensive benchmark for high-level synthesis targeted to fpgas,” Advances in Neural Information Processing Systems , vol. 36, pp. 45 288– 45 299, 2023

    work page 2023

  5. [5]

    Masterrtl: A pre-synthesis ppa estimation framework for any rtl design,

    W. Fang, Y . Lu, S. Liu, Q. Zhang, C. Xu, L. W. Wills, H. Zhang, and Z. Xie, “Masterrtl: A pre-synthesis ppa estimation framework for any rtl design,” in Proceedings of 2023 IEEE/ACM International Conference on Computer Aided Design (ICCAD) . IEEE, 2023, pp. 1–9

    work page 2023

  6. [6]

    SNS’s not a synthesizer: a deep- learning-based synthesis predictor,

    C. Xu, C. Kjellqvist, and L. W. Wills, “SNS’s not a synthesizer: a deep- learning-based synthesis predictor,” in Proceedings of the 49th Annual International Symposium on Computer Architecture (ISCA) , 2022, pp. 847–859

    work page 2022

  7. [7]

    Cross-modality program representation learning for electronic design au- tomation with high-level synthesis,

    Z. Qin, Y . Bai, A. Sohrabizadeh, Z. Ding, Z. Hu, Y . Sun, and J. Cong, “Cross-modality program representation learning for electronic design au- tomation with high-level synthesis,” inProceedings of the 2024 ACM/IEEE International Symposium on Machine Learning for CAD , 2024, pp. 1–12

    work page 2024

  8. [8]

    Versatile multi-stage graph neural network for circuit representation,

    S. Yang, Z. Yang, D. Li, Y . Zhang, Z. Zhang, G. Song, and J. Hao, “Versatile multi-stage graph neural network for circuit representation,” Advances in Neural Information Processing Systems , vol. 35, pp. 20 313– 20 324, 2022

    work page 2022

  9. [9]

    Circuitnet: An open-source dataset for machine learning in vlsi cad applications with improved domain-specific evaluation metric and learning strategies,

    Z. Chai, Y . Zhao, W. Liu, Y . Lin, R. Wang, and R. Huang, “Circuitnet: An open-source dataset for machine learning in vlsi cad applications with improved domain-specific evaluation metric and learning strategies,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2023

    work page 2023

  10. [10]

    Circuit as set of points,

    J. Zou, X. Wang, J. Guo, W. Liu, Q. Zhang, and C. Huang, “Circuit as set of points,” Advances in Neural Information Processing Systems , vol. 36, 2024

    work page 2024

  11. [11]

    Archexplorer: Microarchitecture exploration via bottleneck analysis,

    C. Bai, J. Huang, X. Wei, Y . Ma, S. Li, H. Zheng, B. Yu, and Y . Xie, “Archexplorer: Microarchitecture exploration via bottleneck analysis,” in Proceedings of the 56th Annual IEEE/ACM International Symposium on Microarchitecture, 2023, pp. 268–282

    work page 2023

  12. [12]

    Graph Generative Model for Benchmarking Graph Neural Networks

    M. Yoon, Y . Wu, J. Palowitch, B. Perozzi, and R. Salakhutdinov, “Graph generative model for benchmarking graph neural networks,” arXiv preprint arXiv:2207.04396, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  13. [13]

    A survey of circuit foundation model: Foundation ai models for vlsi circuit design and eda,

    W. Fang, J. Wang, Y . Lu, S. Liu, Y . Wu, Y . Ma, and Z. Xie, “A survey of circuit foundation model: Foundation ai models for vlsi circuit design and eda,” arXiv preprint arXiv:2504.03711 , 2025

    Pith/arXiv arXiv 2025

  14. [14]

    Efficient nonmyopic bayesian optimization via one-shot multi-step trees,

    S. Jiang, D. Jiang, M. Balandat, B. Karrer, J. Gardner, and R. Garnett, “Efficient nonmyopic bayesian optimization via one-shot multi-step trees,” Advances in Neural Information Processing Systems , vol. 33, pp. 18 039– 18 049, 2020

    work page 2020

  15. [15]

    D-vae: A variational autoencoder for directed acyclic graphs,

    M. Zhang, S. Jiang, Z. Cui, R. Garnett, and Y . Chen, “D-vae: A variational autoencoder for directed acyclic graphs,” Advances in neural information processing systems, vol. 32, 2019

    work page 2019

  16. [16]

    Brp- nas: Prediction-based nas using gcns,

    L. Dudziak, T. Chau, M. Abdelfattah, R. Lee, H. Kim, and N. Lane, “Brp- nas: Prediction-based nas using gcns,” Advances in Neural Information Processing Systems, vol. 33, pp. 10 480–10 490, 2020

    work page 2020

  17. [17]

    Graphrnn: Gen- erating realistic graphs with deep auto-regressive models,

    J. You, R. Ying, X. Ren, W. Hamilton, and J. Leskovec, “Graphrnn: Gen- erating realistic graphs with deep auto-regressive models,” in International conference on machine learning . PMLR, 2018, pp. 5708–5717

    work page 2018

  18. [18]

    Graphvae: Towards generation of small graphs using variational autoencoders,

    M. Simonovsky and N. Komodakis, “Graphvae: Towards generation of small graphs using variational autoencoders,” in Artificial Neural Networks and Machine Learning–ICANN 2018: 27th International Conference on Artificial Neural Networks, Rhodes, Greece, October 4-7, 2018, Proceed- ings, Part I 27 . Springer, 2018, pp. 412–422

    work page 2018

  19. [19]

    Efficient graph generation with graph recurrent attention networks,

    R. Liao, Y . Li, Y . Song, S. Wang, W. Hamilton, D. K. Duvenaud, R. Urtasun, and R. Zemel, “Efficient graph generation with graph recurrent attention networks,” Advances in neural information processing systems , vol. 32, 2019

    work page 2019

  20. [20]

    Efficient and degree- guided graph generation via discrete diffusion modeling,

    X. Chen, J. He, X. Han, and L.-P. Liu, “Efficient and degree- guided graph generation via discrete diffusion modeling,” arXiv preprint arXiv:2305.04111, 2023

    Pith/arXiv arXiv 2023

  21. [21]

    Spectre: Spectral conditioning helps to overcome the expressivity limits of one- shot graph generators,

    K. Martinkus, A. Loukas, N. Perraudin, and R. Wattenhofer, “Spectre: Spectral conditioning helps to overcome the expressivity limits of one- shot graph generators,” in International Conference on Machine Learning. PMLR, 2022, pp. 15 159–15 179

    work page 2022

  22. [22]

    Digress: Discrete denoising diffusion for graph generation,

    C. Vignac, I. Krawczuk, A. Siraudin, B. Wang, V . Cevher, and P. Frossard, “Digress: Discrete denoising diffusion for graph generation,” arXiv preprint arXiv:2209.14734, 2022

    Pith/arXiv arXiv 2022

  23. [23]

    Layerdag: A layerwise autoregressive diffusion model of directed acyclic graphs for system,

    M. Li, V . Shitole, E. Chien, C. Man, Z. Wang, Y . Zhang, T. Krishna, P. Li et al. , “Layerdag: A layerwise autoregressive diffusion model of directed acyclic graphs for system,” in Machine Learning for Computer Architecture and Systems, 2024

    work page 2024

  24. [24]

    Improved denoising diffusion probabilistic models,

    A. Q. Nichol and P. Dhariwal, “Improved denoising diffusion probabilistic models,” in International conference on machine learning . PMLR, 2021, pp. 8162–8171

    work page 2021

  25. [25]

    Graphmaker: Can diffusion models generate large attributed graphs?

    M. Li, E. Krea ˇci´c, V . K. Potluru, and P. Li, “Graphmaker: Can diffusion models generate large attributed graphs?” arXiv preprint arXiv:2310.13833, 2023

    Pith/arXiv arXiv 2023

  26. [26]

    Translating embeddings for modeling multi-relational data,

    A. Bordes, N. Usunier, A. Garcia-Duran, J. Weston, and O. Yakhnenko, “Translating embeddings for modeling multi-relational data,” Advances in neural information processing systems , vol. 26, 2013

    2013

  27. [27]

    A survey of monte carlo tree search methods,

    C. B. Browne, E. Powley, D. Whitehouse, S. M. Lucas, P. I. Cowling, P. Rohlfshagen, S. Tavener, D. Perez, S. Samothrakis, and S. Colton, “A survey of monte carlo tree search methods,” IEEE Transactions on Computational Intelligence and AI in games , vol. 4, no. 1, pp. 1–43, 2012

    2012

  28. [28]

    Annotating slack directly on your verilog: Fine-grained rtl timing evaluation for early optimization,

    W. Fang, S. Liu, H. Zhang, and Z. Xie, “Annotating slack directly on your verilog: Fine-grained rtl timing evaluation for early optimization,” in Proceedings of 2024 ACM/IEEE Design Automation Conference (DAC) . ACM, 2024, pp. 1–6

    work page 2024

  29. [29]

    Rt-level itc’99 benchmarks and first atpg results,

    F. Corno, M. S. Reorda, and G. Squillero, “Rt-level itc’99 benchmarks and first atpg results,” IEEE Design & Test of computers (ITC) , 2000

    work page 2000

  30. [30]

    Iwls 2005 benchmarks,

    C. Albrecht, “Iwls 2005 benchmarks,” in International Workshop for Logic Synthesis (IWLS): http://www. iwls. org , 2005

    work page 2005

  31. [31]

    Chipyard: Integrated design, simulation, and implementation framework for custom socs,

    A. Amid, D. Biancolin, A. Gonzalez, D. Grubb, S. Karandikar, H. Liew, A. Magyar, H. Mao, A. Ou, N. Pemberton et al. , “Chipyard: Integrated design, simulation, and implementation framework for custom socs,” IEEE Micro, vol. 40, no. 4, 2020

    work page 2020

  32. [32]

    Sparse training of discrete diffusion models for graph generation,

    Y . Qin, C. Vignac, and P. Frossard, “Sparse training of discrete diffusion models for graph generation,” arXiv preprint arXiv:2311.02142 , 2023

    Pith/arXiv arXiv 2023

  33. [33]

    Gravity-inspired graph autoencoders for directed link prediction,

    G. Salha, S. Limnios, R. Hennequin, V .-A. Tran, and M. Vazirgiannis, “Gravity-inspired graph autoencoders for directed link prediction,” in Proceedings of the 28th ACM international conference on information and knowledge management , 2019, pp. 589–598

    work page 2019

  34. [34]

    Large scale learning on non-homophilous graphs: New benchmarks and strong simple methods,

    D. Lim, F. Hohne, X. Li, S. L. Huang, V . Gupta, O. Bhalerao, and S. N. Lim, “Large scale learning on non-homophilous graphs: New benchmarks and strong simple methods,” Advances in Neural Information Processing Systems, vol. 34, pp. 20 887–20 902, 2021

    work page 2021

  35. [35]

    Complex networks: small-world, scale-free and beyond,

    X. F. Wang and G. Chen, “Complex networks: small-world, scale-free and beyond,” IEEE circuits and systems magazine , vol. 3, no. 1, pp. 6–20, 2003

    work page 2003

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.