Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Functional Matching of Logic Subgraphs: Beyond Structural Isomorphism
Authors: Ziyang Zheng, Kezhi Li, Zhengyuan Shi, Qiang Xu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Evaluations on standard benchmarks (ITC99, Open ABCD, Forge EDA) demonstrate significant performance improvements over existing structural methods, with average 93.8% accuracy in functional subgraph detection and a dice score of 91.3% in fuzzy boundary identification. The source code and implementation details can be found at our repository. |
| Researcher Affiliation | Academia | Ziyang Zheng Kezhi Li Zhengyuan Shi Qiang Xu The Chinese University of Hong Kong EMAIL |
| Pseudocode | Yes | Algorithm 1 Random Sample Subgraph |
| Open Source Code | Yes | The source code and implementation details can be found at our repository. We provide the code and data in our supplemental material. |
| Open Datasets | Yes | Our experiments demonstrate the effectiveness of the proposed framework. Evaluations conducted across several widely-used benchmarks, ITC99 [15], Open ABCD [16] and Forge EDA [17], show that our approach significantly surpasses traditional structure-based methods. |
| Dataset Splits | Yes | For dataset split, we first split the training circuits and test circuits in the source dataset, then we cut subgraph for the training circuit and test circuits to generate our small circuit dataset. For ITC99 and Open ABCD, the split follow the previous work [24]. For Forge EDA, we randomly select 10% circuits in the dataset as test circuits. Table 10: Dataset Statistics. We report average and standard error with avg. std. Source Dataset Split #Pair Gsub Gaig Gsyn Gpm #Node Depth #Node Depth #Node Depth #Node Depth ITC99 train 36592 ... test 5917 ... |
| Hardware Specification | Yes | All experiments are run on an NVIDIA A100 GPU with 64 GB of memory. |
| Software Dependencies | No | The paper mentions using the ABC tool [31] and Skywater Open Source PDK [32] and Adam optimizer but does not specify version numbers for these or other key software components like programming languages or libraries. |
| Experiment Setup | Yes | Models are trained using the Adam optimizer with a learning rate of 0.001, a batch size of 1024. We train our model in stage#1 for 100 epochs and finetune it in stage#2 for 10 epochs. Model architectures follow the configurations specified in the original works except that we set the hidden dimension to 128 for all models. |