Wenji Fang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenji Fang has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Zhiyao Xie
The Hong Kong University of Science and Technology
13 shared publications - Shang Liu
Hong Kong University of Science and Technology, Hong Kong, Hong Kong, Hong Kong
10 shared publications - Yao Lu
The Hong Kong University of Science and Technology
9 shared publications - Hongce Zhang
The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, Guangdong, China
6 shared publications - Qijun Zhang
Hong Kong University of Science and Technology, Hong Kong, Hong Kong, Hong Kong
6 shared publications - Mengming Li
HKUST, Hong Kong, Hong Kong, Hong Kong
4 shared publications
Latest Publications
- Top Cited PapersOpen Access
- 31 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 32 (1) , 1-71
A Survey of Circuit Foundation Model: Foundation AI Models for VLSI Circuit Design and EDA
Artificial intelligence (AI)-driven electronic design automation (EDA) techniques have been extensively explored for VLSI circuit design applications. Most recently, foundation AI models for circuits have emerged as a new technology trend. Unlike traditional task-specific AI solutions, these new AI models are developed through two stages: (1) self-supervised pre-training on a large amount of unlabeled data to learn intrinsic circuit properties; and (2) efficient fine-tuning for specific downstream applications, such as early-stage design quality evaluation, circuit-related context generation, and functional verification. This new paradigm brings many advantages: model generalization, less reliance on labeled circuit data, efficient adaptation to new tasks, and unprecedented generative capability. In this article, we propose referring to AI models developed with this new paradigm as circuit foundation models (CFMs) . This article provides a comprehensive survey of the latest progress in circuit foundation models, unprecedentedly covering over 160 relevant works. Over 90% of our introduced works were published in or after 2022, indicating that this emerging research trend has attracted wide attention in a short period. In this survey, we propose to categorize all existing circuit foundation models into two primary types: (1) encoder-based methods performing general circuit representation learning for predictive tasks ; and (2) decoder-based methods leveraging large language models (LLMs) for generative tasks . For our introduced works, we cover their input modalities, model architecture, pre-training strategies, domain adaptation techniques, and downstream design applications. In addition, this article discussed the unique properties of circuits from the data perspective. These circuit properties have motivated many works in this domain and differentiated them from general AI techniques. Finally, we shared our observed challenges and potential future research directions about developing foundation AI models for EDA methodologies.
- 31 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 32 (1) , 1-71
Top Cited PapersOpen AccessA Survey of Circuit Foundation Model: Foundation AI Models for VLSI Circuit Design and EDA
Artificial intelligence (AI)-driven electronic design automation (EDA) techniques have been extensively explored for VLSI circuit design applications. Most recently, foundation AI models for circuits have emerged as a new technology trend. Unlike traditional task-specific AI solutions, these new AI models are developed through two stages: (1) self-supervised pre-training on a large amount of unlabeled data to learn intrinsic circuit properties; and (2) efficient fine-tuning for specific downstream applications, such as early-stage design quality evaluation, circuit-related context generation, and functional verification. This new paradigm brings many advantages: model generalization, less reliance on labeled circuit data, efficient adaptation to new tasks, and unprecedented generative capability. In this article, we propose referring to AI models developed with this new paradigm as circuit foundation models (CFMs) . This article provides a comprehensive survey of the latest progress in circuit foundation models, unprecedentedly covering over 160 relevant works. Over 90% of our introduced works were published in or after 2022, indicating that this emerging research trend has attracted wide attention in a short period. In this survey, we propose to categorize all existing circuit foundation models into two primary types: (1) encoder-based methods performing general circuit representation learning for predictive tasks ; and (2) decoder-based methods leveraging large language models (LLMs) for generative tasks . For our introduced works, we cover their input modalities, model architecture, pre-training strategies, domain adaptation techniques, and downstream design applications. In addition, this article discussed the unique properties of circuits from the data perspective. These circuit properties have motivated many works in this domain and differentiated them from general AI techniques. Finally, we shared our observed challenges and potential future research directions about developing foundation AI models for EDA methodologies.
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
LEAP: A Self-Supervised Per-Cycle Toggle Propagation Model Supports Fast, Transferable, and Early Analysis of Layout Power
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessLEAP: A Self-Supervised Per-Cycle Toggle Propagation Model Supports Fast, Transferable, and Early Analysis of Layout Power
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
RTL-Sequencer: Towards Scalable RTL Timing Prediction with the Sequence-based Paradigm
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessRTL-Sequencer: Towards Scalable RTL Timing Prediction with the Sequence-based Paradigm
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
COOL: A Cooling-Aware Point Transformer Framework for Thermal Prediction in Advanced 3D/3.5D IC Packaging
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessCOOL: A Cooling-Aware Point Transformer Framework for Thermal Prediction in Advanced 3D/3.5D IC Packaging
- 19 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. PP (02780070) , 1
MFSPart: A Generalized Partitioning Framework for Multi-FPGA Systems and Its Ensemble-Based Extension
Multi-FPGA systems (MFSs) are essential for prototyping very large-scale integration (VLSI) designs, necessitating efficient partitioning techniques. Optimizing inter-FPGA communication—particularly minimizing hop count—is critical for improving performance and reducing latency. However, existing topology-aware partitioning methods primarily minimize driver–sink pair cut size under a strict one-hop constraint, which may not always yield feasible solutions. They also typically overlook practical constraints such as FPGA capacity limits and I/O resource optimization, leading to solutions with excessive pin usage and routing complexity. To address these limitations, we propose a unified formulation that jointly optimizes driver–sink pair cut size, connectivity, and mean hop count under maximum-hop and capacity constraints. Unlike prior frameworks that enforce hop-based constraints via propagation before coarsening—imposing significant limitations— we postpone propagation until after coarsening. Based on this, we develop MFSPart, a framework applicable to both fixed-node and non-fixed-node scenarios, featuring improved coarsening, initial partitioning, and refinement strategies. We further propose an enhanced version, MFSPart-Ensemble, which systematically combines high-quality structures from diverse solutions across uncoarsening stages to generate superior partitions. Experiments demonstrate that MFSPart achieves excellent performance across all three objectives with reduced runtime, while MFSPart-Ensemble further improves solution quality, especially for complex FPGA systems.
- 19 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. PP (02780070) , 1
MFSPart: A Generalized Partitioning Framework for Multi-FPGA Systems and Its Ensemble-Based Extension
Multi-FPGA systems (MFSs) are essential for prototyping very large-scale integration (VLSI) designs, necessitating efficient partitioning techniques. Optimizing inter-FPGA communication—particularly minimizing hop count—is critical for improving performance and reducing latency. However, existing topology-aware partitioning methods primarily minimize driver–sink pair cut size under a strict one-hop constraint, which may not always yield feasible solutions. They also typically overlook practical constraints such as FPGA capacity limits and I/O resource optimization, leading to solutions with excessive pin usage and routing complexity. To address these limitations, we propose a unified formulation that jointly optimizes driver–sink pair cut size, connectivity, and mean hop count under maximum-hop and capacity constraints. Unlike prior frameworks that enforce hop-based constraints via propagation before coarsening—imposing significant limitations— we postpone propagation until after coarsening. Based on this, we develop MFSPart, a framework applicable to both fixed-node and non-fixed-node scenarios, featuring improved coarsening, initial partitioning, and refinement strategies. We further propose an enhanced version, MFSPart-Ensemble, which systematically combines high-quality structures from diverse solutions across uncoarsening stages to generate superior partitions. Experiments demonstrate that MFSPart achieves excellent performance across all three objectives with reduced runtime, while MFSPart-Ensemble further improves solution quality, especially for complex FPGA systems.
- Open Access
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 572-585
Profile-Guided Temporal Prefetching
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 572-585
Open AccessProfile-Guided Temporal Prefetching
- Open Access
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 505-512
A Self-Supervised, Pre-Trained, and Cross-Stage-Aligned Circuit Encoder Provides a Foundation for Various Design Tasks
Machine learning (ML) techniques have shown remarkable effectiveness in electronic design automation (EDA). Traditionally, most ML for EDA approaches are task-specific, requiring a tedious development process of a tailored ML model for each individual design task. Recently, circuit representation learning has emerged as a promising trend. This approach converts circuits into embeddings, which can then be adapted to distinct downstream tasks. However, existing methods still fall short of providing a truly general circuit representation that supports highly diverse tasks. In this work, we introduce CircuitEncoder, a self-supervised, pre-trained, and cross-stage-aligned general circuit encoder. It provides a general foundation for diverse ML-based EDA tasks, including both design quality and functional reasoning. CircuitEncoder is pre-trained through multi-stage contrastive learning utilizing unlabeled circuits. It encodes circuits from different design stages into embedding vectors within shared latent space, facilitating fine-tuning for various downstream tasks. CircuitEncoder outperforms the state-of-the-art task-specific supervised solutions for multiple EDA tasks, including design quality tasks for register-transfer level (RTL)-stage timing and area prediction, as well as functional tasks for netlist-stage state register identification.
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 505-512
Open AccessA Self-Supervised, Pre-Trained, and Cross-Stage-Aligned Circuit Encoder Provides a Foundation for Various Design Tasks
Machine learning (ML) techniques have shown remarkable effectiveness in electronic design automation (EDA). Traditionally, most ML for EDA approaches are task-specific, requiring a tedious development process of a tailored ML model for each individual design task. Recently, circuit representation learning has emerged as a promising trend. This approach converts circuits into embeddings, which can then be adapted to distinct downstream tasks. However, existing methods still fall short of providing a truly general circuit representation that supports highly diverse tasks. In this work, we introduce CircuitEncoder, a self-supervised, pre-trained, and cross-stage-aligned general circuit encoder. It provides a general foundation for diverse ML-based EDA tasks, including both design quality and functional reasoning. CircuitEncoder is pre-trained through multi-stage contrastive learning utilizing unlabeled circuits. It encodes circuits from different design stages into embedding vectors within shared latent space, facilitating fine-tuning for various downstream tasks. CircuitEncoder outperforms the state-of-the-art task-specific supervised solutions for multiple EDA tasks, including design quality tasks for register-transfer level (RTL)-stage timing and area prediction, as well as functional tasks for netlist-stage state register identification.
- Open Access
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 527-533
Towards Big Data in AI for EDA Research: Generation of New Pseudo Circuits at RTL Stage
Machine learning (ML) techniques have demonstrated remarkable effectiveness in electronic design automation (EDA). ML models need to be trained on diverse circuit datasets for better accuracy and generalization capabilities. However, the availability of circuit data remains a long-standing severe issue. The strong data privacy concern in the semiconductor industry makes direct sharing of circuit IPs almost impossible. To address the data availability problem, open-source datasets like CircuitNet have been proposed, but they mostly focus on collecting labels of several existing open-source designs, instead of generating any new designs. In this work, we make an innovative exploration to directly generate new pseudo-circuits without human effort. We believe that generating pseudo-circuits is the most promising, if not the only, approach to achieving "big data" in the semiconductor industry in the foreseeable future. We demonstrate that pseudo-circuits can significantly boost the performance of ML models in early design quality predictions, as early as the pre-synthesis RTL stage.
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 527-533
Open AccessTowards Big Data in AI for EDA Research: Generation of New Pseudo Circuits at RTL Stage
Machine learning (ML) techniques have demonstrated remarkable effectiveness in electronic design automation (EDA). ML models need to be trained on diverse circuit datasets for better accuracy and generalization capabilities. However, the availability of circuit data remains a long-standing severe issue. The strong data privacy concern in the semiconductor industry makes direct sharing of circuit IPs almost impossible. To address the data availability problem, open-source datasets like CircuitNet have been proposed, but they mostly focus on collecting labels of several existing open-source designs, instead of generating any new designs. In this work, we make an innovative exploration to directly generate new pseudo-circuits without human effort. We believe that generating pseudo-circuits is the most promising, if not the only, approach to achieving "big data" in the semiconductor industry in the foreseeable future. We demonstrate that pseudo-circuits can significantly boost the performance of ML models in early design quality predictions, as early as the pre-synthesis RTL stage.
- Top Cited PapersOpen Access
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 614-621
AssertLLM: Generating Hardware Verification Assertions from Design Specifications via Multi-LLMs
Assertion-based verification (ABV) is a critical method to ensure logic designs comply with their architectural specifications. ABV requires assertions, which are generally converted from specifications through human interpretation by verification engineers. Existing methods for generating assertions from specification documents are limited to sentences extracted by engineers, discouraging their practical applications. In this work, we present AssertLLM, an automatic assertion generation framework that processes complete specification documents. AssertLLM can generate assertions from both natural language and waveform diagrams in specification files. It first converts unstructured specification sentences and waveforms into structured descriptions using natural language templates. Then, a customized Large Language Model (LLM) generates the final assertions based on these descriptions. Our evaluation demonstrates that AssertLLM can generate more accurate and higher-quality assertions compared to GPT-4o and GPT-3.5.
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 614-621
Top Cited PapersOpen AccessAssertLLM: Generating Hardware Verification Assertions from Design Specifications via Multi-LLMs
Assertion-based verification (ABV) is a critical method to ensure logic designs comply with their architectural specifications. ABV requires assertions, which are generally converted from specifications through human interpretation by verification engineers. Existing methods for generating assertions from specification documents are limited to sentences extracted by engineers, discouraging their practical applications. In this work, we present AssertLLM, an automatic assertion generation framework that processes complete specification documents. AssertLLM can generate assertions from both natural language and waveform diagrams in specification files. It first converts unstructured specification sentences and waveforms into structured descriptions using natural language templates. Then, a customized Large Language Model (LLM) generates the final assertions based on these descriptions. Our evaluation demonstrates that AssertLLM can generate more accurate and higher-quality assertions compared to GPT-4o and GPT-3.5.