Pith. sign in

REVIEW 5 major objections 5 minor 65 references

TransPlace: Transferable Circuit Global Placement via Graph Neural Network

T0 review · 5 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A GNN trained on a few preplaced circuits can place unseen million-cell chips in continuous space, faster and with less congestion, timing delay, and wirelength than the placer that generated its training data.

desk verdict A genuinely novel GNN placement architecture with real congestion gains, but the quality claims are confounded by per-circuit fine-tuning and the wirelength headline doesn't match the tables. read the letter →

arxiv 2501.05667 v2 pith:IIMABIRT submitted 2025-01-10 cs.LG cs.AIcs.AR

Yunbo Hou , Haoran Ye , Shuwen Yang , Yingxue Zhang , Siyuan Xu , Guojie Song This is my paper · ORCID
classification cs.LGcs.AIcs.AR
keywords EDAcircuitdesignglobalplacementgraphneuralnetworktransferableSE(2)-invariantrepresentationcell-flownetlist
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

TransPlace sets out to prove that global placement—the step in chip design that arranges millions of cells on a die to minimize wire length, congestion, and delay—can be learned once and transferred to new circuits instead of being re-optimized from scratch for every design. The paper trains a graph neural network on a few placements produced by a GPU-accelerated analytical placer, then uses the trained network to produce an initial placement for an unseen netlist in one forward pass, followed by a short analytical fine-tuning pass that adapts to that circuit's terminals and density constraints. On four standard benchmarks the authors report that this two-stage procedure places new circuits about 1.2 times faster while reducing congestion by 30%, timing degradation by 9%, and wirelength by 5%. If those numbers hold, placement stops being a per-design optimization chore and becomes a learnable warm start, which would matter directly for the speed and quality of modern chip design flows.

What carries the argument

Cell-flow: a directed acyclic graph of relative cell positions, built by breadth-first search from fixed terminals through nets, in which each directed edge stores the position of one cell relative to another. This is the object that makes transfer feasible: the encoding $\boldsymbol{\rho}, \Delta\theta$ is invariant to global rotation and translation, so the model learns layout structure rather than absolute coordinates. Around it sit the Netlist Graph, which preserves full cell–net–pin topology and is coarsened hierarchically with hypergraph partitioning for scalability; TPGNN, the paper's message-passing architecture over both graphs; and the two-stage coarse-to-fine strategy that finishes with wirelength and density fine-tuning. The decoding step is linear in the number of pins, which the paper argues is optimal.

What would settle it

Apply the per-circuit hyperparameters from Tables 7–10 to the baseline placer itself, or drop the fine-tuning stage and compare the raw inductive output with the baseline; if the tuned baseline performs as well or better, the claimed transfer gains are not substantiated.

Watch

Extended reading notes

Core claim

TransPlace is presented as the first learning-based framework for global placement at the scale of millions of mixed-size cells in continuous space. Its central claim is that a GNN trained only on a handful of preplaced circuits can inductively place unseen circuits and outperform the analytical placer that generated the training data. The model does not predict coordinates directly; it predicts the SE(2)-invariant relative encoding—distance $\rho$ and deflection $\Delta\theta$—along a cell-flow DAG built from the netlist, then decodes those into absolute positions by averaging paths that start from fixed terminals. A circuit-adaptive fine-tuning stage then refines the solution against wirelength and electrostatic density objectives. The authors report improvements over the baseline placer across routability-oriented and timing-oriented benchmarks, including transfer to timing metrics even though the training labels came from routability-driven placements.

Load-bearing premise

The comparison assumes the baseline placer runs with default settings while each TransPlace circuit receives hand-set learning rates, density weights, iteration limits, and initial offsets; if those per-circuit settings were removed or granted equally to the baseline, the claimed speedup and quality gains could shrink or disappear.

Editorial extensions

If this is right

  • For each new circuit, the inductive stage provides a warm start that already contains transferable placement structure, so optimization begins from a better point rather than from scratch.
  • The reported 1.2x speedup and 30% congestion reduction would translate directly into shorter design cycles and fewer routing failures in a physical-design flow.
  • Because the learning objective is routability-driven but the reported gains include timing, the method implies that a single placement prior can improve multiple design metrics simultaneously.
  • The near-linear decoding complexity means the approach remains usable as circuits grow to millions of cells, a scale where prior learning-based approaches stopped at floorplanning.

Reading between the lines

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

  • An extension the paper does not test is whether training labels from several different placers or technology nodes would make the transferred prior more robust; a natural experiment would compare models trained on one versus multiple label sources.
  • Because the representation is SE(2)-invariant, the model may also tolerate rotated or mirrored layouts in deployment, but the paper does not demonstrate this directly.
  • The contribution is really the combined inductive-plus-fine-tuning system; an editorial reading is that the GNN's role is to supply a high-quality initialization, while the analytically grounded fine-tuning pass performs the final constraint satisfaction.
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

5 major / 5 minor

Summary. TransPlace is a two-stage global placement framework. The first stage uses a graph neural network (TPGNN) applied to a hierarchical 'Netlist Graph' and a 'Cell-flow' DAG to predict relative cell positions in an SE(2)-invariant representation, after training on a few DREAMPlace-generated placements. The second stage, 'Circuit-adaptive Fine-tuning', refines the inductive placement by gradient-based optimization of wirelength and density objectives with Nesterov acceleration. The paper claims that TransPlace can place unseen circuits with 1.2x speedup, 30% less congestion, 9% better timing, and 5% shorter wirelength compared to state-of-the-art analytical placers, and that it is the first learning-based approach for large-scale global placement in continuous space.

Significance. If the claims were supported, TransPlace would be a significant contribution to learning-based VLSI placement: it combines a scalable hierarchical graph representation, a novel cell-flow relative-position encoding, and a complexity-optimal decoding scheme, with an open-source implementation. The algorithmic ideas are well motivated and the authors provide complexity arguments for the graph construction and message passing. However, the central empirical claim—that a GNN trained on a few DREAMPlace placements transfers to unseen circuits and beats the algorithm that generated its training data—is not established by the evidence. The comparison is clouded by per-circuit tuned fine-tuning, the GNN-only ablation fails to route, and the headline wirelength numbers are inconsistent with the tables.

major comments (5)
  1. [Abstract and §3.1, Tables 1–3] The abstract claims a 5% wirelength reduction, but the average RWL ratios in Table 1 and Table 2 are 1.00 (i.e., identical to DREAMPlace), and Table 1 contains multiple circuits where TransPlace is worse (e.g., mgc_des_perf_1: 1.81 vs. 1.70; mgc_des_perf_b: 2.59 vs. 2.38; mgc_fft_1: 0.70 vs. 0.65). Table 3 shows a 12% average improvement on DAC2012, but even there TransPlace is worse on superblue2 and superblue19. The headline 5% figure is therefore not supported by the reported results and should be reconciled.
  2. [§2.5 and Tables 7–10] The comparison with DREAMPlace is unfair. For each test design, TransPlace's fine-tuning uses individually selected learning rates (ranging from 1e-4 to 3e-1), density weights (8e-9 to 8e-3), iteration limits (150 to 2000), and hand-set initial offsets (Delta_x and Delta_y), as shown in Tables 7–10, while DREAMPlace runs with default settings (Appendix E.2). Because the GNN is trained on DREAMPlace placements and the fine-tuning stage minimizes the same wirelength and density objectives that DREAMPlace optimizes, any observed improvement over DREAMPlace could arise entirely from the per-circuit tuning rather than from transferable knowledge. The paper must either apply the same per-circuit tuning to DREAMPlace or provide an ablation showing that TransPlace's advantage persists when both methods receive identical optimization budgets.
  3. [Appendix G, Table 11] The ablation study shows that without fine-tuning, TransPlace's inductive output has overflow up to 1,828,433 (versus 5–41 for DREAMPlace) and often fails to route (entries marked '-'). The text in Appendix G states that 'fine-tuning is crucial to guarantee reasonable placement.' This directly contradicts the abstract's claim that TransPlace 'learns to place millions of mixed-size cells' and that the reported improvements stem from transferable placement knowledge. The GNN-only component is far worse than the baseline, so the reported wins must be attributed to the optimization stage, not to the learned inductive placement.
  4. [§3.1 (training circuits)] The five training circuits are not named. Table 3 evaluates ten DAC2012 circuits named superblue2–superblue19, and ISPD2015 contains superblue11_a, superblue12, superblue14, superblue16_a, and superblue19. Without a disclosed list of which five circuits were used for training, the claim that the evaluation circuits are 'unseen' cannot be verified, and there is a risk of train/test leakage. Please provide the exact training set and confirm that none of the test designs overlap with it.
  5. [Table 4 and §3.2] Table 4 shows that TransPlace improves TNS/WNS on average, but routed wirelength is worse on 4 of 8 circuits and the average rWL ratio is 1.01 (1% worse). The timing improvements are obtained with per-circuit hyperparameters (Table 10) and a timing-driven fine-tuning procedure similar to DREAMPlace 4.0, so the isolation problem identified above applies here as well. The '9% better timing' claim should be reported with the associated wirelength regression and with an explicit statement of the fine-tuning cost.
minor comments (5)
  1. [Abstract and Table 5] The abstract claims a 1.2x speedup, but Table 5 reports an average speedup of 1.49x on ISPD2015. Please reconcile these numbers or state which benchmark the abstract refers to.
  2. [Table 6] The overlap ratio eta is reported as 1.1262, 1.1257, and so on, which is greater than 1. The definition of eta as the ratio of the sum of branch net counts to the original net count should be clarified, since a value above 1 suggests net duplication rather than overlap.
  3. [Algorithm 2] In Algorithm 2, line 12 adds (v, v) to F, which appears to be a self-loop and is inconsistent with Definition 2.2's guarantee that no loop will be found in F. This is likely a typo and should be corrected to (v', v) or similar.
  4. [References] Reference [50] is cited as the DAC2012 benchmark but is titled 'ISPD 2019 Initial Detailed Routing Contest and Benchmark with Advanced Routing Rules.' Please correct the reference or use the proper DAC2012 citation.
  5. [Figure 1] Figure 1 is difficult to read: the fine-tuning loop in part (b) is not clearly separated from the inductive placement stage, and the message-passing arrows are dense. A higher-level flow diagram with clear stage boundaries would improve readability.

Circularity Check

2 steps flagged · score 7.0 of 10

Claimed gains over DREAMPlace are produced by per-circuit-tuned fine-tuning of the same wirelength+density objective, not by the transferable GNN; the GNN-only output is unrouteable.

  1. fitted input called prediction [Section 2.5 (Eq. 31), Appendix E.2, Tables 7-10]
    "Lfine−tune = LW + λDLD. (31) ... We use default baseline settings. Both DREAMPlace and circuit-adaptive fine-tuning use NAG Optimizer [22] for a fair comparison ... Our fine-tuning parameters are given in Table 7, Table 8, Table 9, and Table 4; “Iteration” is the maximum iteration limit, a variable setting up an early stop strategy [22, 31]."

    The final placement is produced by the same NAG optimizer and the same wirelength+density objective as DREAMPlace (Eqs. 24-31), warm-started by the GNN and run with per-circuit hyperparameters (Tables 7-10: learning rates 3e-4 to 3e-1, density weights 8e-9 to 8e-3, iteration limits 150 to 2000, hand-set initial offsets), while DREAMPlace runs with default settings. The claimed 1.2x speedup and OVFL/RWL/TNS improvements are therefore outputs of the per-circuit-tuned fine-tuner, not of transferable GNN knowledge. The GNN-only ablation (Table 11) is unrouteable, so the reported 'prediction' of better placements reduces to the fitted fine-tuning stage by construction.

  2. other [Section 2.4.2 (Eq. 38), Section G / Table 11]
    "TPGNN is trained to imitate preplaced circuits generated by DREAMPlace [31] ... The training loss is determined by the difference between TPGNN outputs and ground truth relative positions using Smooth-L1 loss. ... From the result, we can infer that fine-tuning is crucial to guarantee reasonable placement, and inductive placement can further improve placement quality."

    The GNN's only training signal is DREAMPlace's own placements, and its isolated output has overflow up to 1.8 million and frequently cannot route (Table 11), whereas the fine-tuner optimizes the same wirelength+density objective that DREAMPlace optimizes. Thus the comparison 'TransPlace vs DREAMPlace' does not test learned transfer; it tests whether a per-circuit-tuned DREAMPlace-style optimizer warm-started by a DREAMPlace-trained GNN beats a default DREAMPlace run. The claimed advantage is attributable to the optimization stage whose objective is identical to the baseline, making the 'transferable knowledge' claim unfalsifiable in this setup.

full rationale

The paper's graph construction, SE(2)-invariant encoding/decoding, and GNN message-passing equations are internally consistent and not circular by themselves. The circularity enters at the level of the empirical claim. TransPlace's final output is a DREAMPlace-style NAG optimizer (Section 2.5) minimizing the same wirelength and density objectives used by DREAMPlace, with hyperparameters individually selected per test circuit (Tables 7-10), while DREAMPlace runs with default settings. The GNN is trained to imitate DREAMPlace placements (Eq. 38), and its isolated output is unrouteable (Table 11). Therefore the reported superiority over DREAMPlace is not evidence of transferable placement knowledge; it is by construction a comparison between a per-circuit-tuned optimizer and a default-configuration baseline. No load-bearing self-citation chain or imported uniqueness theorem is present; the issue is that the central 'prediction' of better placements is produced by the fitted fine-tuning stage, not by the learned model. An honest non-circular evaluation would apply the same per-circuit tuning to DREAMPlace or compare GNN-only inductive placements against a fair baseline.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The reported results depend heavily on per-circuit tuned fine-tuning hyperparameters and on the assumption that DREAMPlace placements are good training labels. The GNN itself is trained to imitate DREAMPlace, so the only source of improvement over DREAMPlace is the additional, per-circuit-tuned optimization. The ledger below lists the free parameters and domain assumptions that carry this load.

free parameters (5)
  • Per-circuit fine-tuning learning rate = varies per circuit (e.g., 1e-1, 1e-2, 3e-4, 4e-2)
    Tables 7-10 assign a different learning rate to each test circuit; this tuning directly affects convergence and final quality.
  • Per-circuit density weight = ranges from 5e-9 to 8e-3 across circuits
    Tables 7-10; the density term weight controls spreading and overflow, and is hand-set per circuit.
  • Per-circuit iteration limit = 150 to 2000 depending on circuit
    Tables 7-10; early stopping budget is tuned per circuit, affecting reported runtime and quality.
  • Per-circuit initial offsets (Delta_x, Delta_y) = various signed offsets, e.g., -585.94, 0; -276.80, 201.62
    Tables 7-10; these non-zero offsets shift the initial placement before fine-tuning and are chosen per circuit, which can change the outcome.
  • Rho range hyperparameters alpha, beta in Eq. (22) = alpha=15, beta=-2
    Section 2.4.1 sets these to map outputs to rho in (4e-8, 4e5); they are hand-chosen.
assumptions (4)
  • domain assumption Every connectivity branch in a circuit contains at least one fixed cell (terminal or pseudo-terminal), so the BFS cell-flow covers all movable cells.
    Section 2.2 and Algorithm 2 rely on this to guarantee every cell appears in the DAG and can be decoded via Eq. (9).
  • domain assumption DREAMPlace-generated placements are high-quality enough to serve as supervised training targets.
    Section 2.4.2 trains TPGNN to imitate DREAMPlace; if these labels are suboptimal, the model inherits their bias.
  • domain assumption The weighted wirelength plus electrostatic density objective used in fine-tuning is a valid proxy for the reported congestion and timing metrics.
    Section 2.5 and 3 evaluate against routed overflow and timing slacks, but fine-tuning optimizes only HPWL-like wirelength and density.
  • domain assumption KaHyPar produces sub-netlists with small overlap (eta near 1), so the claimed O(|V|+|U|+|P|) complexity holds.
    Appendix D.1 states the complexity is O(|V|+eta|U|+|P|) and approximates eta->1 using measured values in Table 6.
invented entities (2)
  • Pseudo-cells and pseudo-pins in the hierarchical netlist graph
    purpose: Represent each KaHyPar partition as a single artificial cell in the root graph to coarsen the circuit and enable scalable message passing.
    Appendix C defines pseudo-cells with synthetic width/height sqrt(5*S_i) and pseudo-pins; these are computational abstractions with no physical counterpart.
  • Cell-flow DAG
    purpose: A directed acyclic graph of relative cell positions used for SE(2)-invariant encoding/decoding and message passing.
    Section 2.2 introduces cell-flow as a learned representation; it is not an observed physical quantity, though it is derived from netlist topology.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TransPlace: Transferable Circuit Global Placement via Graph Neural Network." pith.science (2026). https://pith.science/paper/IIMABIRT

@misc{pith2026250105667,
  author       = {Pith},
  title        = {Pith review of: TransPlace: Transferable Circuit Global Placement via Graph Neural Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IIMABIRT}},
  note         = {Machine review of arXiv:2501.05667}
}
read the original abstract

Global placement, a critical step in designing the physical layout of computer chips, is essential to optimize chip performance. Prior global placement methods optimize each circuit design individually from scratch. Their neglect of transferable knowledge limits solution efficiency and chip performance as circuit complexity drastically increases. This study presents TransPlace, a global placement framework that learns to place millions of mixed-size cells in continuous space. TransPlace introduces i) Netlist Graph to efficiently model netlist topology, ii) Cell-flow and relative position encoding to learn SE(2)-invariant representation, iii) a tailored graph neural network architecture for informed parameterization of placement knowledge, and iv) a two-stage strategy for coarse-to-fine placement. Compared to state-of-the-art placement methods, TransPlace-trained on a few high-quality placements-can place unseen circuits with 1.2x speedup while reducing congestion by 30%, timing by 9%, and wirelength by 5%.

Figures

Figures reproduced from arXiv: 2501.05667 by the authors.

Figure 1
Figure 1. A schematic illustration of TransPlace. TransPlace contains two stages: Inductive Placement and Circuit-adaptive Fine-tuning. (a) Inductive Placement efficiently generates relative cell positions in one shot. It constructs the Cell-flow and Netlist Graph, applies message-passing on the two graphs to obtain cell and net representations, fuses and updates the hidden representations, and reads out relative position 𝜌, … view at source ↗
Figure 2
Figure 2. Overview of Netlist Graph, Cell-flow, cell-path, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 1
Figure 1. TPGNN takes a netlist graph and its cell-flow as inputs, then [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figures from the paper (1 more)
Figure 3
Figure 3. Figure 3: Visualizations of comparisons on superblue5. (a) Congestion visualization of the placement generated by DREAMPlace and TransPlace. The grey parts denote the cells, and the red dots indicate that there is congestion. Density of red dots indicates the level of congestion…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 52 canonical work pages

  1. [1]

    Vidal-Obiols Alex, Cortadella Jordi, Petit Jordi, Galceran-Oms Marc, and Martorell Ferran. 2021. Multilevel Dataflow-Driven Macro Placement Guided by RTL Structure and Analytical Methods. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 40, 12 (Dec. 2021), 2542–2555

    work page 2021

  2. [2]

    Alpert, Zhuo Li, Michael D

    Charles J. Alpert, Zhuo Li, Michael D. Moffitt, Gi-Joon Nam, Jarrod A. Roy, and Gustavo Tellez. 2010. What makes a design difficult to route. Paper presented at the 10th International Symposium on Physical Design, San Francisco, California, USA, 14–16 March 2010

    work page 2010

  3. [3]

    Lilas Alrahis, Johann Knechtel, and Ozgur Sinanoglu. 2023. Graph neural net- works: A powerful and versatile tool for advancing design, reliability, and security of ICs. In Proceedings of the 28th Asia and South Pacific Design Automation Con- ference. Association for Computing Machinery, New York, NY, USA, 83–90

    work page 2023

  4. [4]

    Anand, H

    M.B. Anand, H. Shibata, and M Kakumu. 1998. Multiobjective optimization of VLSI interconnect parameters. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 17 (1998), 1252–1261

    work page 1998

  5. [5]

    Mirhoseini Azalia and Anna Goldie. 2021. A graph placement methodology for fast chip design. Nature 594, 7862 (Jun. 2021), 207–212

    work page 2021

  6. [6]

    Federico Berto, Chuanbo Hua, Junyoung Park, Laurin Luttmann, Yining Ma, Fanchen Bu, Jiarui Wang, Haoran Ye, Minsu Kim, Sanghyeok Choi, Nayeli Gast Zepeda, André Hottung, Jianan Zhou, Jieyi Bi, Yu Hu, Fei Liu, Hyeonah Kim, Jiwoo Son, Haeyeon Kim, Davide Angioni, Wouter Kool, Zhiguang Cao, Jie Zhang, Kijung Shin, Cathy Wu, Sungsoo Ahn, Guojie Song, Changhyu...

    arXiv 2024

  7. [7]

    Wang Bowen, Shen Guibao, Li Dong, Hao Jianye, Liu Wulong, Huang Yu, Wu Hongzhong, Lin Yibo, Chen Guangyong, and Heng Pheng Ann. 2022. LHNN: lattice hypergraph neural network for VLSI congestion prediction. In Proceedings of the 59th ACM/IEEE Design Automation Conference (DAC ’22) . Association for Computing Machinery, New York, NY, USA, 1297–1302

    work page 2022

  8. [8]

    Melvin A. Breuer. 1977. A Class of Min-Cut Placement Algorithms. Paper presented at the Proceedings of the 14th Design Automation Conference, January 1977

    work page 1977

Show all 65 references
  1. [9]

    Ismail Bustany, David Chinnery, Joseph Shinnerl, and Vladimir Yutsis. 2015. ISPD 2015 Benchmarks with Fence Regions and Routing Blockages for Detailed- Routing-Driven Placement. In Proceedings of the 2015 Symposium on International Symposium on Physical Design. Association for...

    2015

  2. [10]

    Fu-Chieh Chang, Yu-Wei Tseng, Ya-Wen Yu, and Ssu-Rui Lee. 2022. Flexible Multiple-Objective Reinforcement Learning for Chip Placement. https://doi.org/ 10.48550/ARXIV.2204.06407

    2022 doi arXiv

  3. [11]

    Gengjie Chen, Chak-Wa Pui, Wing-Kai Chow, Ka-Chun Lam, Jian Kuang, Evan- geline F. Y. Young, and Bei Yu. 2018. RippleFPGA: Routability-Driven Simul- taneous Packing and Placement for Modern FPGAs. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 37...

    2018

  4. [12]

    Tung-Chieh Chen, Zhe-Wei Jiang, Tien-Chang Hsu, Hsin-Chen Chen, and Yao- Wen Chang. 2008. NTUplace3: An Analytical Placer for Large-Scale Mixed-Size Designs With Preplaced Blocks and Density Constraints. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Sys...

    2008

  5. [13]

    Kahng, Ilgweon Kang, and Lutong Wang

    Chung-Kuan Cheng, Andrew B. Kahng, Ilgweon Kang, and Lutong Wang. 2019. RePlAce: Advancing Solution Quality and Routability Validation in Global Place- ment. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 38, 9 (2019), 1717–1730. https://doi.org/...

    2019

  6. [14]

    Ruoyu Cheng, Xianglong Lyu, Yang Li, Junjie Ye, Jianye Hao, and Junchi Yan

  7. [15]

    Ruoyu Cheng and Junchi Yan. 2021. On joint learning for solving placement and routing in chip design. Advances in Neural Information Processing Systems 34 (2021), 16508–16519

    2021

  8. [16]

    Ke-Ren Dai, Wen-Hao Liu, and Yih-Lang Li. 2012. NCTU-GR: Efficient Simulated Evolution-Based Rerouting and Congestion-Relaxed Layer Assignment on 3-D Global Routing. IEEE Transactions on Very Large Scale Integration (VLSI) Systems 20 (2012), 459–472. https://doi.org/10.1109/TV...

    2012

  9. [17]

    weitao du, He Zhang, Yuanqi Du, Qi Meng, Wei Chen, Tie-Yan Liu, Nanning Zheng, and Bin Shao. 2022. SE(3) Equivariant Graph Neural Networks with Complete Local Frames. In ICML 2022, Vol. 162. Microsoft, PMLR, Honolulu, Hawaii, USA, 5583–5608

    2022

  10. [18]

    Yunbo Hou, Haoran Ye, Yingxue Zhang, Siyuan Xu, and Guojie Song. 2024. RoutePlacer: An End-to-End Routability-Aware Placer with Graph Neural Net- work. arXiv:2406.02651 [cs.LG] https://arxiv.org/abs/2406.02651

    2024 arXiv

  11. [19]

    Meng-Kai Hsu, Yao-Wen Chang, and Valeriy Balabanov. 2011. TSV-aware analyt- ical placement for 3D IC designs. Paper presented at the 48th ACM/EDAC/IEEE Design Automation Conference, San Diego, CA, USA, 5–9 June 2011

    2011

  12. [20]

    Meng-Kai Hsu, Yi-Fang Chen, Chau-Chin Huang, Sheng Chou, Tzu-Hen Lin, Tung-Chieh Chen, and Yao-Wen Chang. 2014. NTUplace4h: A Novel Routability- Driven Placement Algorithm for Hierarchical Mixed-Size Circuit Designs. IEEE Transactions on Computer-Aided Design of Integrated Cir...

    2014

  13. [21]

    Zixuan Jiang, Ebrahim Songhori, Shen Wang, Anna Goldie, Azalia Mirhoseini, Joe Jiang, Young-Joon Lee, and David Z Pan. 2021. Delving into macro placement with reinforcement learning. In 2021 ACM/IEEE 3rd Workshop on Machine Learning for CAD (MLCAD). IEEE, IEEE, New York, NY, USA, 1–3

    2021

  14. [22]

    Lu Jingwei, Chen Pengwen, Chang Chin-Chih, Sha Lu, Huang Dennis Jen- Hsin, Teng Chin-Chi, and Cheng Chung-Kuan. 2015. ePlace: Electrostatics- Based Placement Using Fast Fourier Transform and Nesterov’s Method. ACM KDD ’25, August 3–7, 2025, Toronto, ON, Canada Trovato et al. T...

    2015 doi

  15. [23]

    Haeyeon Kim, Minsu Kim, Federico Berto, Joungho Kim, and Jinkyoo Park

  16. [24]

    Myung-Chul Kim, Jin Hu, Jiajia Li, and Natarajan Viswanathan. 2015. ICCAD- 2015 CAD Contest in Incremental Timing-driven Placement and Benchmark Suite. In Proceedings of the IEEE/ACM International Conference on Computer-Aided Design (Austin, TX, USA) (ICCAD ’15). IEEE Press, N...

    2015

  17. [25]

    Robert Kirby, Saad Godil, Rajarshi Roy, and Bryan Catanzaro. 2019. Congestion- Net: Routing congestion prediction using deep graph neural networks. In 2019 IFIP/IEEE 27th International Conference on Very Large Scale Integration (VLSI-SoC) . IEEE, IEEE, New York, NY, USA, 217–222

    2019

  18. [26]

    Schütt Kristof, Sauceda Huziel E., Kindermans P.-J, Tkatchenko Alexandre, and Müller Klaus-Robert. 2018. SchNet – A deep learning architecture for molecules and materials. The Journal of Chemical Physics 148, 24 (Jun. 2018), 241722. https: //doi.org/10.1063/1.5019779

    2018 doi

  19. [27]

    Yao Lai, Jinxin Liu, Zhentao Tang, Bin Wang, Jianye Hao, and Ping Luo. 2023. ChiPFormer: transferable chip placement via offline decision transformer. In Proceedings of the 40th International Conference on Machine Learning (ICML’23) . JMLR.org, Honolulu, Hawaii, USA, 18346–183...

    2023 doi

  20. [28]

    Yao Lai, Yao Mu, and Ping Luo. 2022. Maskplace: Fast chip placement via rein- forced visual representation learning. Advances in Neural Information Processing Systems 35 (2022), 24019–24030

    2022

  21. [29]

    Peiyu Liao, Siting Liu, Zhitang Chen, Wenlong Lv, Yibo Lin, and Bei Yu. 2022. DREAMPlace 4.0: Timing-driven Global Placement with Momentum-based Net Weighting. In 2022 Design, Automation & Test in Europe Conference & Exhibition (DATE). European Design and Automation Associatio...

    2022

  22. [30]

    Jai-Ming Lin, You-Lun Deng, Ya-Chu Yang, Jia-Jian Chen, and Po-Chen Lu. 2021. Dataflow-Aware Macro Placement Based on Simulated Evolution Algorithm for Mixed-Size Designs. IEEE Transactions on Very Large Scale Integration (VLSI) Systems 29, 5 (2021), 973–984. https://doi.org/1...

    2021

  23. [31]

    Yibo Lin, Zixuan Jiang, Jiaqi Gu, Wuxi Li, Shounak Dhar, Haoxing Ren, Brucek Khailany, and David Z. Pan. 2021. DREAMPlace: Deep Learning Toolkit-Enabled GPU Acceleration for Modern VLSI Placement. IEEE Transactions on Computer- Aided Design of Integrated Circuits and Systems 4...

    2021

  24. [32]

    Yibo Lin, Wuxi Li, Jiaqi Gu, Haoxing Ren, Brucek Khailany, and David Z. Pan

  25. [33]

    Pan, Haoxing Ren, and Brucek Khailany

    Yibo Lin, David Z. Pan, Haoxing Ren, and Brucek Khailany. 2020. DREAMPlace 2.0: Open-Source GPU-Accelerated Global and Detailed Placement for Large-Scale VLSI Designs. In 2020 China Semiconductor Technology International Conference (CSTIC). IEEE, New York, NY, USA, 1–4

    2020

  26. [34]

    Zhifeng Lin, Yanyue Xie, Gang Qian, Jianli Chen, Sifei Wang, Jun Yu, and Yao- Wen Chang. 2021. Timing-Driven Placement for FPGAs with Heterogeneous Architectures and Clock Constraints. In 2021 Design, Automation & Test in Europe Conference & Exhibition (DATE). IEEE Press, New ...

    2021

  27. [35]

    Wen-Hao Liu, Stefanus Mantik, Wing-Kai Chow, Yixiao Ding, Amin Farshidi, and Gracieli Posser. 2019. ISPD 2019 Initial Detailed Routing Contest and Benchmark with Advanced Routing Rules. Paper presented at the Proceedings of the 2019 Symposium on International Symposium on Phys...

    2019

  28. [36]

    Yiting Liu, Ziyi Ju, Zhengming Li, Mingzhi Dong, Hai Zhou, Jia Wang, Fan Yang, Xuan Zeng, and Li Shang. 2022. Floorplanning with graph attention. In Proceedings of the 59th ACM/IEEE Design Automation Conference . Association for Computing Machinery, San Francisco, California, ...

    2022

  29. [37]

    Yiting Liu, Ziyi Ju, Zhengming Li, Mingzhi Dong, Hai Zhou, Jia Wang, Fan Yang, Xuan Zeng, and Li Shang. 2022. Graphplanner: Floorplanning with graph neural network. ACM Transactions on Design Automation of Electronic Systems 28, 2 (2022), 1–24

    2022

  30. [38]

    Yen-Chun Liu, Tung-Chieh Chen, Yao-Wen Chang, and Sy-Yen Kuo. 2019. MDP- Trees: Multi-Domain Macro Placement for Ultra Large-Scale Mixed-Size Designs. In Proceedings of the 24th Asia and South Pacific Design Automation Conference (Tokyo, Japan) (ASPDAC ’19). Association for Co...

    2019

  31. [39]

    Daniela Sánchez Lopera and Wolfgang Ecker. 2022. Applying gnns to timing estimation at rtl. In Proceedings of the 41st IEEE/ACM International Conference on Computer-Aided Design. Association for Computing Machinery, New York, NY, USA, 1–8

    2022

  32. [40]

    Yi-Chen Lu, Siddhartha Nath, Vishal Khandelwal, and Sung Kyu Lim. 2021. Doomed run prediction in physical design by exploiting sequential flow and graph learning. In 2021 IEEE/ACM International Conference On Computer Aided Design (ICCAD). IEEE, IEEE, New York, NY, USA, 1–9

    2021

  33. [41]

    Yi-Chen Lu, Siddhartha Nath, Vishal Khandelwal, and Sung Kyu Lim. 2021. Rl- sizer: Vlsi gate sizing for timing optimization using deep reinforcement learning. In 2021 58th ACM/IEEE Design Automation Conference (DAC) . IEEE, IEEE, New York, NY, USA, 733–738

    2021

  34. [42]

    Yibai Meng, Wuxi Li, Yibo Lin, and David Z. Pan. 2022. elfPlace: Electrostatics- Based Placement for Large-Scale Heterogeneous FPGAs. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 41, 1 (2022), 155–168. https://doi.org/10.1109/TCAD.2021.3053191

    2022

  35. [43]

    Cheng Ruoyu and Yan Junchi. 2021. On joint learning for solving placement and routing in chip design. In Advances in Neural Information Processing Systems (NeurIPS ’21). Curran Associates, Inc., New York, NY, USA, 16508–16519. https: //doi.org/10.48550/arXiv.2111.00234

    2021 doi arXiv

  36. [44]

    Vıctor Garcia Satorras, Emiel Hoogeboom, and Max Welling. 2021. E (n) equi- variant graph neural networks. In International conference on machine learning . PMLR, Proceedings of Machine Learning Research, Virtual Event, 9323–9332

    2021

  37. [45]

    Sebastian Schlag, Tobias Heuer, Lars Gottesbüren, Yaroslav Akhremtsev, Christian Schulz, and Peter Sanders. 2023. High-Quality Hypergraph Partitioning. ACM J. Exp. Algorithmics 27, Article 1.9 (feb 2023), 39 pages. https://doi.org/10.1145/ 3529090

    2023

  38. [46]

    Sechen and A

    C. Sechen and A. Sangiovanni-Vincentelli. 1985. The TimberWolf placement and routing package. IEEE Journal of Solid-State Circuits 20 (1985), 510–522. https://doi.org/10.1109/JSSC.1985.1052337

    1985

  39. [47]

    Yunqi Shi, Ke Xue, Song Lei, and Chao Qian. 2024. Macro placement by wire- mask-guided black-box optimization. Advances in Neural Information Processing Systems 36 (2024), 19 pages

    2024

  40. [48]

    Yunqi Shi, Ke Xue, Lei Song, and Chao Qian. 2023. Macro Placement by Wire- Mask-Guided Black-Box Optimization. In Advances in Neural Information Process- ing Systems (NeurIPS ’23) . Curran Associates, Inc., New Orleans, USA, 6825–6843. https://doi.org/10.48550/arXiv.2306.16844

    2023 doi arXiv

  41. [49]

    Johannes

    Peter Spindler, Ulf Schlichtmann, and Frank M. Johannes. 2008. Abacus: Fast Legalization of Standard Cell Circuits with Minimal Movement. Paper presented at the Proceedings of the 2008 International Symposium on Physical Design, Portland, Oregon, USA, 13–16 April 2008

    2008

  42. [50]

    Natarajan Viswanathan, Charles Alpert, Cliff Sze, Zhuo Li, and Yaoguang Wei

  43. [51]

    J. Vygen. 2006. Slack in static timing analysis. IEEE Transactions on Computer- Aided Design of Integrated Circuits and Systems 25, 9 (2006), 1876–1885

    2006

  44. [52]

    Nan Wu, Yuan Xie, and Cong Hao. 2021. Ironman: Gnn-assisted design space ex- ploration in high-level synthesis via reinforcement learning. In Proceedings of the 2021 on Great Lakes Symposium on VLSI . Association for Computing Machinery, New York, NY, USA, 39–44

    2021

  45. [53]

    Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. 2020. A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning systems 32, 1 (2020), 4–24

    2020

  46. [54]

    Yue Xu, Yanheng Zhang, and Chris Chu. 2009. FastRoute 4.0: Global router with efficient via minimization. Paper presented at the 2009 Asia and South Pacific Design Automation Conference, Yokohama, 19–22 January 2009

    2009

  47. [55]

    Shuwen Yang, Ziyao Li, Guojie Song, and Lingsheng Cai. 2021. Deep Molecular Representation Learning via Fusing Physical and Chemical Information. Pa- per presented at the Advances in Neural Information Processing Systems, 28 November – 9 December 2022

    2021

  48. [56]

    Shuwen Yang, Zhihao Yang, Dong Li, Yingxue Zhang, Zhanguang Zhang, Guojie Song, and Jianye HAO. 2022. Versatile Multi-stage Graph Neural Network for Circuit Representation. In Advances in Neural Information Processing Systems , Alice H. Oh, Alekh Agarwal, Danielle Belgrave, an...

    2022

  49. [57]

    Haoran Ye, Jiarui Wang, Zhiguang Cao, Federico Berto, Chuanbo Hua, Haeyeon Kim, Jinkyoo Park, and Guojie Song. 2024. ReEvo: Large Language Models as Hyper-Heuristics with Reflective Evolution. In Advances in Neural Information Processing Systems. https://github.com/ai4co/reevo

    2024

  50. [58]

    Haoran Ye, Jiarui Wang, Zhiguang Cao, Helan Liang, and Yong Li. 2023. DeepACO: Neural-enhanced Ant Systems for Combinatorial Optimization. Paper presented at the Advances in Neural Information Processing Systems, New Orleans, USA, 10–16 December 2023

    2023

  51. [59]

    Ricardo Salem Zebulum, Marco Aurélio Pacheco, and Marley Vellasco. 2015. A multi-objective optimisation methodology applied to the synthesis of low-power operational amplifiers. Paper presented at the 13th International Conference in Microelectronics and Packaging, Singapore, ...

    2015

  52. [60]

    Xinyi Zhou, Junjie Ye, Chak-Wa Pui, Kun Shao, Guangliang Zhang, Bin Wang, Jianye Hao, Guangyong Chen, and Pheng Ann Heng. 2022. Heterogeneous graph neural network-based imitation learning for gate sizing acceleration. In Proceedings of the 41st IEEE/ACM International Conferenc...

    2022

  53. [65]

    Iteration

    with default hyperparameters. Timing optimization differs from DREAMPlace [29], evaluating timing metrics and updating net weights every 15 iterations after placement overflow reaches 𝑠𝑡𝑜𝑣𝑒𝑟𝑓𝑙𝑜𝑤 = 0.5 or𝑠𝑡𝑖𝑡𝑒𝑟𝑎𝑡𝑖𝑜𝑛 = 500. For DAC2012, ISPD2015, and ISPD2019, the rudy map guide...

    2025

  54. [2012]

    Paper presented at the Proceedings of the 49th Annual Design Automation Conference, San Francisco, CA, USA, 3–7 June 2012

    ISPD 2019 Initial Detailed Routing Contest and Benchmark with Advanced Routing Rules. Paper presented at the Proceedings of the 49th Annual Design Automation Conference, San Francisco, CA, USA, 3–7 June 2012

    2019

  55. [2020]

    IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 39 (2020), 5083–5096

    ABCDPlace: Accelerated Batch-Based Concurrent Detailed Placement on Multithreaded CPUs and GPUs. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 39 (2020), 5083–5096

    2020

  56. [2022]

    Advances in Neural Information Processing Systems 35 (2022), 26350–26362

    The policy-gradient placement and generative routing neural networks for chip design. Advances in Neural Information Processing Systems 35 (2022), 26350–26362

    2022

  57. [2023]

    In International Conference on Machine Learning

    Devformer: A symmetric transformer for context-aware device placement. In International Conference on Machine Learning . PMLR, JMLR.org, Honolulu, Hawaii, USA, 16541–16566

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.