pith. sign in

arxiv: 2605.08291 · v1 · submitted 2026-05-08 · 💻 cs.LG · cs.AI· cs.AR

Graph Computation Meets Circuit Algebra: A Task-Aligned Analysis of Graph Neural Networks for Electronic Design Automation

Hyunmog Kim This is my paper

Pith reviewed 2026-05-12 02:05 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.AR
keywords graph neural networkselectronic design automationcircuit graphsalgebraic alignmentstatic timing analysisplacementrouting congestionIR drop
0
0 comments X

The pith

Successful GNNs for electronic design automation align their message passing and supervision with the algebraic rules of each specific circuit task.

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

The paper argues that EDA tasks are graph-structured but each one is governed by its own mathematical operations, so a one-size-fits-all GNN rarely works well. Static timing analysis runs on max-plus or min-plus recurrences over a DAG, placement is driven by hypergraph penalties, and other steps like IR drop or activity propagation follow linear systems or probabilistic rules. When a GNN's propagation, aggregation, and training objective match these native operations, the model respects the underlying physics and constraints of the circuit. Generic message-passing networks often fall short precisely because they ignore these differences. A reader would care because the alignment view turns GNN design from trial-and-error into a task-by-task mapping that explains both past successes and current failure modes.

Core claim

The central claim is that successful GNN-for-EDA methods are those whose propagation, aggregation, and supervision align with the native algebra of the target task. Concretely, static timing analysis is a max-plus/min-plus recurrence on a topologically ordered DAG and is structurally aligned with asynchronous DAG-GNNs; placement is governed by hypergraph wirelength and density penalties and is better exploited by differentiable placers; routing congestion is a sparse demand-supply field; switching-activity propagation is a probabilistic recurrence; IR drop is a linear system on the power network; and analog symmetry extraction is a discrete constraint-prediction problem. The paper uses these

What carries the argument

The task-algebra alignment principle, which requires that a GNN's message-passing rules and loss function reproduce the specific recurrence or optimization structure (max-plus, linear solve, hypergraph penalty, etc.) native to each EDA task.

If this is right

  • Asynchronous DAG-GNNs become the natural choice for static timing analysis because they mirror the max-plus/min-plus recurrence on topologically ordered graphs.
  • Placement problems are better addressed by differentiable placers that directly optimize hypergraph wirelength and density than by message-passing GNNs alone.
  • Routing congestion estimation benefits from models that treat the layout grid as a sparse demand-supply field rather than a generic graph.
  • Switching-activity and IR-drop tasks require GNNs whose updates reproduce probabilistic recurrences or linear systems on the netlist or power-delivery network.
  • Failure modes such as stage leakage, proxy-to-signoff gap, and design-distribution shift will dominate future work once basic alignment is achieved.

Where Pith is reading between the lines

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

  • The alignment lens suggests that hybrid architectures combining GNN layers with explicit task solvers (for example, a linear solver block for IR drop) may outperform pure learned models.
  • Circuit graphs differ from generic graphs in being directed, heterogeneous, multi-scale, and containing sequential and clock structure, so new benchmarks should explicitly test these properties.
  • The framework could be extended to decide when a given EDA stage should use a GNN at all versus a conventional algorithm or a learned optimizer.
  • Design-distribution shift may be mitigated by training on synthetic circuits that preserve the algebraic invariants of real designs.

Load-bearing premise

The algebraic structures listed for each EDA task accurately capture the core computation that must be performed, and that mismatches between these structures and GNN components are the main reason existing methods fall short.

What would settle it

A GNN architecture deliberately mismatched to the listed algebraic structure (for example, a standard synchronous message-passing network on a timing-analysis DAG) that nevertheless reaches signoff-level accuracy on a production netlist without additional task-specific solvers or calibration.

Figures

Figures reproduced from arXiv: 2605.08291 by Hyunmog Kim.

Figure 1
Figure 1. Figure 1: A minimal heterogeneous signal graph with typed nodes (cells, nets, pins) and typed edges (driver, load, containment): a driver cell exposes a driver pin onto a net that fans out to load pins of downstream cells. Industrial graphs add clock-tree, PDN, hierarchy, physical coordinates, and timing-arc edges, typically modeled on separate graphs rather than overloaded onto the signal graph. Graph transformers … view at source ↗
Figure 2
Figure 2. Figure 2: Toy AT/RAT/slack propagation on a combinational DAG with [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Synchronous vs. asynchronous DAG message passing on a small DAG. The asynchronous [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Multi-die systems as hierarchical graphs. (a) 2.5D integration: chiplets share a silicon [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The digital design flow with representative GNN methods (and non-GNN baselines) at each [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
read the original abstract

EDA problems are graph-structured, but not all graph-structured problems call for the same GNN computation. We argue that successful GNN-for-EDA methods are those whose propagation, aggregation, and supervision align with the native algebra of the target task. Concretely: static timing analysis is a max-plus/min-plus recurrence on a topologically ordered DAG, structurally aligned with asynchronous DAG-GNNs; placement is governed by hypergraph wirelength and density penalties and is exploited by differentiable placers rather than by message-passing GNNs alone; routing congestion is a sparse demand-supply field over a layout grid; switching-activity propagation is a probabilistic recurrence on a directed netlist; IR drop is a linear system on the power-delivery network; and analog symmetry extraction is a discrete constraint-prediction problem on schematic graphs. Through these task-by-task alignments we (i) review the GNN architectural toolkit relevant to circuits, (ii) formalize how circuit graphs differ from generic graphs (directed, heterogeneous, multi-scale, with sequential and clock structure), (iii) characterize where current methods succeed and where the algebra-architecture mismatch limits them, and (iv) identify failure modes--stage leakage, proxy-to-signoff gap, calibration, and design-distribution shift--that we believe are likely to dominate the next phase of work. We position the paper as a GNN-for-EDA, task-aligned analysis rather than a comprehensive AI-for-chip-design survey. Continuous SE(3)-equivariant geometric GNNs are usually mismatched to Manhattan digital layout, and LLM-for-RTL, HLS, and RL/diffusion-based topology generation are outside our scope.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript argues that GNNs applied to EDA tasks succeed when their propagation, aggregation, and supervision align with each task's native algebraic structure (e.g., max-plus/min-plus recurrences on topologically ordered DAGs for static timing analysis, hypergraph wirelength/density penalties for placement, linear systems on power-delivery networks for IR drop, probabilistic recurrences for switching activity, and discrete constraint prediction for analog symmetry). It reviews circuit-relevant GNN toolkits, formalizes distinguishing properties of circuit graphs (directed, heterogeneous, multi-scale, with sequential/clock structure), characterizes current methods' successes versus algebra-architecture mismatches, and identifies emerging failure modes such as stage leakage, proxy-to-signoff gaps, calibration, and design-distribution shift. The work positions itself as a task-aligned analysis rather than a broad survey and excludes topics such as SE(3)-equivariant geometric GNNs for Manhattan layouts and LLM/RL-based topology generation.

Significance. If the task-algebra mappings are accurate and mismatches are a dominant performance limiter, the framework could guide more principled GNN design for EDA by favoring structures such as asynchronous DAG-GNNs for timing analysis over generic message passing. The explicit enumeration of failure modes supplies a concrete research agenda. The absence of new empirical results or a quantitative alignment metric means the significance is primarily conceptual and depends on whether subsequent work can operationalize the alignments into measurable improvements.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (task-by-task analysis): the central claim that 'architecture mismatches explain why current GNNs fall short' is asserted without a defined quantitative metric of alignment, a proof that the listed algebras are minimal/complete native structures, or controlled experiments comparing aligned versus generic message-passing GNNs on identical EDA benchmarks.
  2. [§3] §3 (formalization of circuit graphs): the characterization of circuit graphs as directed, heterogeneous, multi-scale, and containing sequential/clock structure is presented descriptively but lacks explicit mathematical definitions or comparisons against standard graph properties that would allow readers to verify the claimed distinctions or derive concrete architectural prescriptions.
minor comments (2)
  1. [Abstract] The abstract states that 'continuous SE(3)-equivariant geometric GNNs are usually mismatched to Manhattan digital layout' without a supporting paragraph in the main text explaining the geometric mismatch.
  2. [§4] Several EDA tasks are mapped to algebraic structures (e.g., 'hypergraph penalties for placement') but the paper does not cite the original EDA literature that establishes these structures as the canonical formulations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. We appreciate the acknowledgment of the paper's conceptual framing as a task-aligned analysis. We will revise the manuscript to address the concerns about explicitness in claims and formalization, while preserving the scope as a review of alignments rather than an empirical validation study.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (task-by-task analysis): the central claim that 'architecture mismatches explain why current GNNs fall short' is asserted without a defined quantitative metric of alignment, a proof that the listed algebras are minimal/complete native structures, or controlled experiments comparing aligned versus generic message-passing GNNs on identical EDA benchmarks.

    Authors: We agree that the manuscript does not define a quantitative alignment metric, provide a formal proof of minimality/completeness for the task algebras, or include new controlled experiments. This is consistent with the paper's positioning as a conceptual analysis that reviews existing literature to identify patterns of success and failure, rather than an empirical contribution. The central claim is presented as an organizing hypothesis supported by case studies of published methods. We will revise the abstract and §4 to qualify the language explicitly (e.g., 'we hypothesize that mismatches are a dominant limiter based on observed alignments'), remove any implication of proof, and add a short subsection outlining how future work could operationalize alignment metrics and conduct controlled comparisons on standard EDA benchmarks. revision: yes

  2. Referee: [§3] §3 (formalization of circuit graphs): the characterization of circuit graphs as directed, heterogeneous, multi-scale, and containing sequential/clock structure is presented descriptively but lacks explicit mathematical definitions or comparisons against standard graph properties that would allow readers to verify the claimed distinctions or derive concrete architectural prescriptions.

    Authors: We acknowledge that §3 relies on descriptive characterization. We will revise this section to include explicit mathematical definitions—for instance, defining a circuit graph as a tuple (V, E, τ, σ, κ) where τ assigns node/edge types, σ encodes sequential/clock dependencies as a partial order, and κ captures multi-scale hierarchy—along with direct comparisons to standard graph properties (e.g., undirected homogeneous graphs in generic GNN literature). These definitions will be used to derive concrete architectural implications, such as the need for asynchronous propagation in directed acyclic structures. revision: yes

Circularity Check

0 steps flagged

No circularity detected in derivation chain

full rationale

The paper is a descriptive analysis that maps EDA tasks to algebraic structures (e.g., max-plus recurrences for STA) and reviews GNN alignments without introducing quantitative predictions, fitted parameters, or equations that reduce to their own inputs by construction. No self-citations are load-bearing for central claims, no uniqueness theorems are invoked from prior author work, and no ansatzes or renamings of known results are presented as derivations. The central argument remains an interpretive review rather than a self-referential computation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on domain assumptions about circuit graph structure and GNN operations drawn from existing EDA and graph learning literature; no free parameters, new entities, or ad-hoc axioms are introduced in the abstract.

axioms (1)
  • domain assumption EDA problems are graph-structured and governed by specific algebraic operations (max-plus, hypergraph penalties, linear systems, etc.)
    Invoked throughout the abstract as the basis for requiring architecture-task alignment.

pith-pipeline@v0.9.0 · 5601 in / 1289 out tokens · 62352 ms · 2026-05-12T02:05:47.505580+00:00 · methodology

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

95 extracted references · 95 canonical work pages

  1. [1]

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu. A comprehensive survey on graph neural networks.IEEE Trans. Neural Netw. Learn. Syst., 32(1):4–24, 2021

    work page 2021

  2. [2]

    W. Ju, S. Yi, Y . Wang, Z. Xiao, Z. Mao, H. Li, Y . Gu, Y . Qin, N. Yin, S. Wang, X. Liu, P. S. Yu, and M. Zhang. A survey of graph neural networks in real world: Imbalance, noise, privacy and OOD challenges.arXiv:2403.04468v2, 2025

    work page arXiv 2025

  3. [3]

    Huang, J

    G. Huang, J. Hu, Y . He, J. Liu, M. Ma, Z. Shen, J. Wu, Y . Xu, H. Zhang, K. Zhong, X. Ning, Y . Ma, H. Yang, B. Yu, H. Yang, and Y . Wang. Machine learning for electronic design automation: A survey.ACM TODAES, 26(5):40:1–40:46, 2021

    work page 2021

  4. [4]

    Ren and J

    H. Ren and J. Hu (Eds.).Machine Learning Applications in Electronic Design Automation. Springer, 2023

    work page 2023

  5. [5]

    Bruna, W

    J. Bruna, W. Zaremba, A. Szlam, and Y . LeCun. Spectral networks and locally connected networks on graphs. InProc. ICLR, 2014

    work page 2014

  6. [6]

    Defferrard, X

    M. Defferrard, X. Bresson, and P. Vandergheynst. Convolutional neural networks on graphs with fast localized spectral filtering. InProc. NeurIPS, pp. 3837–3845, 2016

    work page 2016

  7. [7]

    T. N. Kipf and M. Welling. Semi-supervised classification with graph convolutional networks. InProc. ICLR, 2017

    work page 2017

  8. [8]

    Gilmer, S

    J. Gilmer, S. S. Schoenholz, P. F. Riley, O. Vinyals, and G. E. Dahl. Neural message passing for quantum chemistry. InProc. ICML, pp. 1263–1272, 2017

    work page 2017

  9. [9]

    W. L. Hamilton, Z. Ying, and J. Leskovec. Inductive representation learning on large graphs. In Proc. NeurIPS, 2017

    work page 2017

  10. [10]

    Veliˇckovi´c, G

    P. Veliˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y . Bengio. Graph attention networks. InProc. ICLR, 2018

    work page 2018

  11. [11]

    Brody, U

    S. Brody, U. Alon, and E. Yahav. How attentive are graph attention networks? InProc. ICLR, 2022

    work page 2022

  12. [12]

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka. How powerful are graph neural networks? InProc. ICLR, 2019

    work page 2019

  13. [13]

    Schlichtkrull, T

    M. Schlichtkrull, T. N. Kipf, P. Bloem, R. van den Berg, I. Titov, and M. Welling. Modeling relational data with graph convolutional networks. InProc. ESWC, pp. 593–607, 2018

    work page 2018

  14. [14]

    Z. Hu, Y . Dong, K. Wang, and Y . Sun. Heterogeneous graph transformer. InProc. WWW, pp. 2704–2710, 2020

    work page 2020

  15. [15]

    Y . Feng, H. You, Z. Zhang, R. Ji, and Y . Gao. Hypergraph neural networks. InProc. AAAI, 33(01):3558–3565, 2019

    work page 2019

  16. [16]

    R. Ying, J. You, C. Morris, X. Ren, W. L. Hamilton, and J. Leskovec. Hierarchical graph representation learning with differentiable pooling. InProc. NeurIPS, pp. 4800–4810, 2018

    work page 2018

  17. [17]

    J. Lee, I. Lee, and J. Kang. Self-attention graph pooling. InProc. ICML, pp. 3734–3743, 2019

    work page 2019

  18. [18]

    Morris, M

    C. Morris, M. Ritzert, M. Fey, W. L. Hamilton, J. E. Lenssen, G. Rattan, and M. Grohe. Weis- feiler and Leman go neural: Higher-order graph neural networks. InProc. AAAI, 33(01):4602– 4609, 2019

    work page 2019

  19. [19]

    Maron, H

    H. Maron, H. Ben-Hamu, H. Serviansky, and Y . Lipman. Provably powerful graph networks. In Proc. NeurIPS, pp. 2153–2164, 2019

    work page 2019

  20. [20]

    Q. Li, Z. Han, and X.-M. Wu. Deeper insights into graph convolutional networks for semi- supervised learning. InProc. AAAI, pp. 3538–3545, 2018

    work page 2018

  21. [21]

    Oono and T

    K. Oono and T. Suzuki. Graph neural networks exponentially lose expressive power for node classification. InProc. ICLR, 2020. 10

    work page 2020

  22. [22]

    K. Xu, C. Li, Y . Tian, T. Sonobe, K. Kawarabayashi, and S. Jegelka. Representation learning on graphs with jumping knowledge networks. InProc. ICML, 2018

    work page 2018

  23. [23]

    M. Chen, Z. Wei, Z. Huang, B. Ding, and Y . Li. Simple and deep graph convolutional networks. InProc. ICML, pp. 1725–1735, 2020

    work page 2020

  24. [24]

    Zhao and L

    L. Zhao and L. Akoglu. PairNorm: Tackling oversmoothing in GNNs. InProc. ICLR, 2020

    work page 2020

  25. [25]

    Y . Rong, W. Huang, T. Xu, and J. Huang. DropEdge: Towards deep graph convolutional networks on node classification. InProc. ICLR, 2020

    work page 2020

  26. [26]

    Alon and E

    U. Alon and E. Yahav. On the bottleneck of graph neural networks and its practical implications. InProc. ICLR, 2021

    work page 2021

  27. [27]

    Topping, F

    J. Topping, F. Di Giovanni, B. P. Chamberlain, X. Dong, and M. M. Bronstein. Understanding over-squashing and bottlenecks on graphs via curvature. InProc. ICLR, 2022

    work page 2022

  28. [28]

    Zhang, S

    M. Zhang, S. Jiang, Z. Cui, R. Garnett, and Y . Chen. D-V AE: A variational autoencoder for directed acyclic graphs. InProc. NeurIPS, 2019

    work page 2019

  29. [29]

    Thost and J

    V . Thost and J. Chen. Directed acyclic graph neural networks. InProc. ICLR, 2021

    work page 2021

  30. [30]

    C. Ying, T. Cai, S. Luo, S. Zheng, G. Ke, D. He, Y . Shen, and T.-Y . Liu. Do transformers really perform badly for graph representation? InProc. NeurIPS, 2021

    work page 2021

  31. [31]

    Rampášek, M

    L. Rampášek, M. Galkin, V . P. Dwivedi, A. T. Luu, G. Wolf, and D. Beaini. Recipe for a general, powerful, scalable graph transformer. InProc. NeurIPS, 2022

    work page 2022

  32. [32]

    Q. Wu, W. Zhao, Z. Li, D. P. Wipf, and J. Yan. NodeFormer: A scalable graph structure learning transformer for node classification. InProc. NeurIPS, 2022

    work page 2022

  33. [33]

    Shirzad, A

    H. Shirzad, A. Velingker, B. Venkatachalam, D. J. Sutherland, and A. K. Sinop. Exphormer: Sparse transformers for graphs. InProc. ICML, 2023

    work page 2023

  34. [34]

    Q. Wu, W. Zhao, C. Yang, H. Zhang, F. Nie, H. Jiang, Y . Bian, and J. Yan. SGFormer: Simplifying and empowering transformers for large-graph representations. InProc. NeurIPS, 2023

    work page 2023

  35. [35]

    C. Deng, Y . Li, Z. Feng, H. Ren, and Z. Zhang. Polynormer: Polynomial-expressive graph transformer in linear time. InProc. ICLR, 2024

    work page 2024

  36. [36]

    K. T. Schütt, P.-J. Kindermans, H. E. Sauceda, S. Chmiela, A. Tkatchenko, and K.-R. Müller. SchNet: A continuous-filter convolutional neural network for modeling quantum interactions. InProc. NeurIPS, pp. 991–1001, 2017

    work page 2017

  37. [37]

    V . G. Satorras, E. Hoogeboom, and M. Welling. E(n) equivariant graph neural networks. In Proc. ICML, pp. 9323–9332, 2021

    work page 2021

  38. [38]

    Batatia, D

    I. Batatia, D. P. Kovács, G. N. C. Simm, C. Ortner, and G. Csányi. MACE: Higher order equivariant message passing neural networks for fast and accurate force fields. InProc. NeurIPS, pp. 11423–11436, 2022

    work page 2022

  39. [39]

    Y .-L. Liao, B. Wood, A. Das, and T. Smidt. EquiformerV2: Improved equivariant transformer for scaling to higher-degree representations. InProc. ICLR, 2024

    work page 2024

  40. [40]

    M. Li, S. Khan, Z. Shi, N. Wang, H. Yu, and Q. Xu. DeepGate: Learning neural representations of logic gates. InProc. DAC, 2022

    work page 2022

  41. [41]

    Z. Shi, H. Pan, S. Khan, M. Li, Y . Liu, J. Huang, H.-L. Zhen, M. Yuan, Z. Chu, and Q. Xu. DeepGate2: Functionality-aware circuit representation learning. InProc. ICCAD, 2023

    work page 2023

  42. [42]

    Z. Shi, Z. Zheng, S. Khan, J. Zhong, M. Li, and Q. Xu. DeepGate3: Towards scalable circuit representation learning. InProc. ICCAD, 2024

    work page 2024

  43. [43]

    N. Wu, Y . Li, C. Hao, S. Dai, C. Yu, and Y . Xie. Gamora: Graph learning based symbolic reasoning for large-scale Boolean networks. InProc. DAC, 2023. 11

    work page 2023

  44. [44]

    J. Liu, J. Zhai, M. Zhao, Z. Lin, B. Yu, and C. Shi. PolarGate: Breaking the functionality representation bottleneck of And-Inverter graph neural network. InProc. ICCAD, 2024

    work page 2024

  45. [45]

    Y . Lin, S. Dhar, W. Li, H. Ren, B. Khailany, and D. Z. Pan. DREAMPlace: Deep learning toolkit-enabled GPU acceleration for modern VLSI placement. InProc. DAC, 2019

    work page 2019

  46. [46]

    P. Liao, S. Liu, Z. Chen, W. Lv, Y . Lin, and B. Yu. DREAMPlace 4.0: Timing-driven global placement with momentum-based net weighting. InProc. DATE, 2022

    work page 2022

  47. [47]

    J. Lu, P. Chen, C.-C. Chang, L. Sha, D. J.-H. Huang, C.-C. Teng, and C.-K. Cheng. ePlace: Electrostatics-based placement using fast Fourier transform and Nesterov’s method.ACM TODAES, 20(2):17:1–17:34, 2015

    work page 2015

  48. [48]

    Mirhoseini, A

    A. Mirhoseini, A. Goldie, M. Yazgan, J. W. Jiang, E. M. Songhori, S. Wang, Y .-J. Lee, E. John- son, O. Pathak, A. Nova, J. Pak, A. Tong, K. Srinivasa, W. Hang, E. Tuncer, Q. V . Le, J. Laudon, R. Ho, R. Carpenter, and J. Dean. A graph placement methodology for fast chip design.Nature, 594:207–212, 2021

    work page 2021

  49. [49]

    Goldie, A

    A. Goldie, A. Mirhoseini, M. Yazgan, et al. Addendum: A graph placement methodology for fast chip design.Nature, 634:E10–E11, 2024

    work page 2024

  50. [50]

    I. L. Markov, D. Lee, and C.-K. Cheng. Is chip floorplanning a solved problem?Commun. ACM, 67(11):66–74, 2024

    work page 2024

  51. [51]

    Cheng, A

    C.-K. Cheng, A. B. Kahng, S. Kundu, Y . Wang, and Z. Wang. Assessment of reinforcement learning for macro placement. InProc. ISPD, 2023

    work page 2023

  52. [52]

    Y . Lai, J. Liu, Z. Tang, B. Wang, J. Hao, and P. Luo. ChiPFormer: Transferable chip placement via offline decision transformer. InProc. ICML, 2023

    work page 2023

  53. [53]

    Kirby, S

    R. Kirby, S. Godil, R. Roy, and B. Catanzaro. CongestionNet: Routing congestion prediction using deep graph neural networks. InProc. IFIP/IEEE International Conference on Very Large Scale Integration (VLSI-SoC), 2019

    work page 2019

  54. [54]

    Jeong and T

    J. Jeong and T. Kim. Utilizing middle-of-line resource in filler cells for fixing routing failures. InProc. IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2021

    work page 2021

  55. [55]

    B. Wang, G. Shen, D. Li, J. Hao, W. Liu, Y . Huang, H. Wu, Y . Lin, G. Chen, and P. A. Heng. LHNN: Lattice hypergraph neural network for VLSI congestion prediction. InProc. DAC, 2022

    work page 2022

  56. [56]

    S. Liu, P. Liao, P. Sun, Y . Lin, and B. Yu. RouteGNN: Learning to route via graph neural networks. InProc. ICCAD, 2023

    work page 2023

  57. [57]

    Zhong, J

    R. Zhong, J. Ye, Z. Tang, S. Kai, M. Yuan, J. Hao, and J. Yan. PreRoutGNN for timing prediction with order preserving partition. InProc. AAAI, 38(15):17087–17095, 2024

    work page 2024

  58. [58]

    Z. Guo, M. Liu, J. Gu, S. Zhang, D. Z. Pan, and Y . Lin. A timing engine inspired graph neural network model for pre-routing slack prediction. InProc. DAC, 2022

    work page 2022

  59. [59]

    Y . Ye, T. Chen, Y . Gao, H. Yan, B. Yu, and L. Shi. TimingPredict: GNN-based pre-routing path delay prediction. InProc. ICCAD, 2023

    work page 2023

  60. [60]

    Y . Ye, T. Chen, Y . Gao, H. Yan, B. Yu, and L. Shi. Fast and accurate wire timing estimation on tree and non-tree net structures. InProc. DAC, 2023

    work page 2023

  61. [61]

    P. Cao, G. He, and T. Yang. TF-Predictor: Transformer-based prerouting path delay prediction framework.IEEE Trans. Comput.-Aided Design Integr. Circuits Syst., 42(7):2227–2237, 2023

    work page 2023

  62. [62]

    Zhang, H

    Y . Zhang, H. Ren, and B. Khailany. GRANNITE: Graph neural network inference for transfer- able power estimation. InProc. DAC, pp. 1–6, 2020

    work page 2020

  63. [63]

    Y . Zhou, H. Ren, Y . Zhang, B. Keller, B. Khailany, and Z. Zhang. PRIMAL: Power inference using machine learning. InProc. DAC, 2019. 12

    work page 2019

  64. [64]

    Y . C. Pao and M. D. F. Wong. PGNN: Physics-inspired graph neural network for dynamic IR-drop prediction. InProc. DAC, 2023

    work page 2023

  65. [65]

    V . A. Chhabria, V . Ahuja, A. Prabhu, N. Patil, P. Jain, and S. S. Sapatnekar. MA VIREC: ML-aided vectored IR-drop estimation and classification. InProc. DATE, 2021

    work page 2021

  66. [66]

    Jeong and T

    J. Jeong and T. Kim. Machine learning driven early clustering for multi-bit flip-flop allocation. InProc. ACM/IEEE Symposium on Machine Learning for CAD (MLCAD), 2025

    work page 2025

  67. [67]

    Jeong and T

    J. Jeong and T. Kim. Placement legalization for heterogeneous cells of non-integer multiple- heights.Integration, 97:102177, 2024

    work page 2024

  68. [68]

    Jeong and T

    J. Jeong and T. Kim. Binding multi-bit flip-flop cells through design and technology co- optimization. InProc. DAC, 2024

    work page 2024

  69. [69]

    Kunal, M

    K. Kunal, M. Madhusudan, A. K. Sharma, W. Xu, S. M. Burns, R. Harjani, J. Hu, D. A. Kirkpatrick, and S. S. Sapatnekar. ALIGN: Open-source analog layout automation from the ground up. InProc. DAC, 2020

    work page 2020

  70. [70]

    X. Gao, C. Deng, M. Liu, Z. Zhang, D. Z. Pan, and Y . Lin. Layout symmetry annotation for analog circuits with graph neural networks. InProc. ASP-DAC, 2021

    work page 2021

  71. [71]

    H. Chen, K. Zhu, M. Liu, X. Tang, N. Sun, and D. Z. Pan. Universal symmetry constraint extraction for analog and mixed-signal circuits with graph neural networks. InProc. DAC, 2021

    work page 2021

  72. [72]

    Z. Dong, W. Cao, M. Zhang, D. Tao, Y . Chen, and X. Zhang. CktGNN: Circuit graph neural network for electronic design automation. InProc. ICLR, 2023

    work page 2023

  73. [73]

    S. M. P. D. Pentapati, K. Chang, V . Gerousis, R. Sengupta, and S. K. Lim. Pin-3D: A physical synthesis and post-layout optimization flow for heterogeneous 3D ICs. InProc. ICCAD, 2023

    work page 2023

  74. [74]

    Z. Chai, Y . Zhao, Y . Lin, W. Liu, R. Wang, and R. Huang. CircuitNet: An open-source dataset for machine learning applications in electronic design automation.Sci. China Inf. Sci., 65(12):227401, 2022

    work page 2022

  75. [75]

    Jiang, Z

    X. Jiang, Z. Chai, Y . Zhao, Y . Lin, R. Wang, and R. Huang. CircuitNet 2.0: An advanced dataset for promoting machine learning in VLSI EDA. InProc. ICLR, 2024

    work page 2024

  76. [76]

    Liu, T.-D

    M. Liu, T.-D. Ene, R. Kirby, C. Cheng, N. Pinckney, R. Liang, J. Alben, H. Anand, S. Banerjee, et al. ChipNeMo: Domain-adapted LLMs for chip design. arXiv:2311.00176, 2024

    work page arXiv 2024

  77. [77]

    H. Liu, J. Feng, L. Kong, N. Liang, D. Tao, Y . Chen, and M. Zhang. One for all: Towards training one graph model for all classification tasks. InProc. ICLR, 2024

    work page 2024

  78. [78]

    Xia and C

    L. Xia and C. Huang. AnyGraph: Graph foundation model in the wild. arXiv:2408.10700, 2024

    work page arXiv 2024

  79. [79]

    Gasse, D

    M. Gasse, D. Chételat, N. Ferroni, L. Charlin, and A. Lodi. Exact combinatorial optimization with graph convolutional neural networks. InProc. NeurIPS, 2019

    work page 2019

  80. [80]

    R. Ying, R. He, K. Chen, P. Eksombatchai, W. L. Hamilton, and J. Leskovec. Graph convo- lutional neural networks for web-scale recommender systems. InProc. KDD, pp. 974–983, 2018

    work page 2018

Showing first 80 references.