Ziyang Yu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ziyang Yu has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Bei Yu
The Chinese University of Hong Kong
11 shared publications - Yuzhe Ma
The Hong Kong University of Science and Technology - Guangzhou Campus
7 shared publications - Su Zheng
Computer Science and Engineering, The Chinese University of Hong Kong
4 shared publications - Guojin Chen
The Chinese University of Hong Kong
3 shared publications Martin D. F. Wong
Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong, SAR, China
3 shared publications- Shuo Yin
The Chinese University of Hong Kong
3 shared publications
Latest Publications
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
CPW-SMO: Generating Mask Sets and Common Source for Maximum Common Process Window
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessCPW-SMO: Generating Mask Sets and Common Source for Maximum Common Process Window
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Coordinated Clip-Wise Gradient Scheduling for Full-Chip ILT via Policy Learning
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessCoordinated Clip-Wise Gradient Scheduling for Full-Chip ILT via Policy Learning
- Open Access
- 14 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (6) , 1-27
FuILT-S: Full Chip ILT With Unified Boundary Healing and SRAF Co-Optimization
Mask optimization in lithography is becoming increasingly important as the technology node size shrinks down. Inverse Lithography Technology (ILT) remains one of the most robust and high-performance solutions adopted in industry, yet it suffers from significant computational overhead and complexity. With the growing scale of transistor integration, industry attention has shifted toward improving efficiency and workload scalability. However, most recent academic efforts remain confined to localized pattern restoration regardless of full-chip manufacturing conditions. In this work, we present FuILT-S , a practical full-chip ILT-based mask optimization framework that incorporates sub-resolution assist features (SRAFs) co-optimization to address real-world manufacturing challenges. To enable full-chip scalability, we build a multi-level partitioning strategy following the divide-and-conquer mindset, paired with a workload distribution framework for efficient multi-GPU parallelization. To fix inter-region boundary stitching artifacts, we propose a gradient-fusion technique alongside a multi-level healing strategy. Furthermore, by leveraging rich information from the fused gradient and mask intensity maps, we integrate an SRAF generation module co-optimized with the main layout patterns. Experimental results across different layers from real-world designs demonstrate that FuILT-S is both effective and generalizable. Our framework is open-sourced at https://github.com/OpenOPC/OpenFuILT.
- 14 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (6) , 1-27
Open AccessFuILT-S: Full Chip ILT With Unified Boundary Healing and SRAF Co-Optimization
Mask optimization in lithography is becoming increasingly important as the technology node size shrinks down. Inverse Lithography Technology (ILT) remains one of the most robust and high-performance solutions adopted in industry, yet it suffers from significant computational overhead and complexity. With the growing scale of transistor integration, industry attention has shifted toward improving efficiency and workload scalability. However, most recent academic efforts remain confined to localized pattern restoration regardless of full-chip manufacturing conditions. In this work, we present FuILT-S , a practical full-chip ILT-based mask optimization framework that incorporates sub-resolution assist features (SRAFs) co-optimization to address real-world manufacturing challenges. To enable full-chip scalability, we build a multi-level partitioning strategy following the divide-and-conquer mindset, paired with a workload distribution framework for efficient multi-GPU parallelization. To fix inter-region boundary stitching artifacts, we propose a gradient-fusion technique alongside a multi-level healing strategy. Furthermore, by leveraging rich information from the fused gradient and mask intensity maps, we integrate an SRAF generation module co-optimized with the main layout patterns. Experimental results across different layers from real-world designs demonstrate that FuILT-S is both effective and generalizable. Our framework is open-sourced at https://github.com/OpenOPC/OpenFuILT.
- Open Access
- 12 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (6) , 1-26
CB-EVO: Contextual Bandit Tuning with Evolutionary Search for Logic Synthesis
In logic synthesis, pre-optimization involves applying a sequence of transformations, referred to as a synthesis flow, to reduce the complexity of a circuit’s Boolean logic graph, such as the And-Inverter Graph (AIG). The primary challenge lies in selecting and ordering these transformations, a task complicated by the exponentially large solution space. In this work, we propose CB-EVO, a novel online learning framework that combines contextual bandit tuning with an evolutionary algorithm to efficiently explore the solution space and generate high-quality synthesis flows. The Syn-LinUCB algorithm functions as the pivotal agent in our bandit tuning process, incorporating circuit-specific characteristics and leveraging long-term payoffs to guide decision-making effectively, thereby circumventing the pitfalls of local optima. To complement the bandit algorithm, we design an evolutionary algorithm that accelerates the decision-making process and enables the efficient expansion of the sequence. Moreover, we explore practical heuristics, such as the “return-back” mechanism and the utilization of a choice network, to further enhance optimization outcomes. Experimental results demonstrate that our framework identifies the optimal synthesis flow with reduced computational time, significantly decreasing the number of AIG nodes and LUT count compared to state-of-the-art (SOTA) methods.
- 12 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (6) , 1-26
Open AccessCB-EVO: Contextual Bandit Tuning with Evolutionary Search for Logic Synthesis
In logic synthesis, pre-optimization involves applying a sequence of transformations, referred to as a synthesis flow, to reduce the complexity of a circuit’s Boolean logic graph, such as the And-Inverter Graph (AIG). The primary challenge lies in selecting and ordering these transformations, a task complicated by the exponentially large solution space. In this work, we propose CB-EVO, a novel online learning framework that combines contextual bandit tuning with an evolutionary algorithm to efficiently explore the solution space and generate high-quality synthesis flows. The Syn-LinUCB algorithm functions as the pivotal agent in our bandit tuning process, incorporating circuit-specific characteristics and leveraging long-term payoffs to guide decision-making effectively, thereby circumventing the pitfalls of local optima. To complement the bandit algorithm, we design an evolutionary algorithm that accelerates the decision-making process and enables the efficient expansion of the sequence. Moreover, we explore practical heuristics, such as the “return-back” mechanism and the utilization of a choice network, to further enhance optimization outcomes. Experimental results demonstrate that our framework identifies the optimal synthesis flow with reduced computational time, significantly decreasing the number of AIG nodes and LUT count compared to state-of-the-art (SOTA) methods.
- Open Access
- 11 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (5) , 1-22
ForgePattern: Flexible Layout Pattern Generation through Reinforcement Learning
Layout pattern generation provides dataset support to numerous Design for Manufacturability (DFM) studies, with different research objectives requiring distinct styles of layout patterns. However, existing layout pattern generation methods only support random generation or limited conditional generation based on image-text pairs annotated for specific tasks, which fails to meet the diverse requirements of downstream tasks. To address this limitation, we propose ForgePattern, a flexible layout pattern generation framework that accommodates diverse task-specific requirements through adaptive selection among customizable reward functions. ForgePattern leverages a customized diffusion model as the generative backbone and incorporates reinforcement learning to directly optimize the diffusion model. By maximizing the expected reward of the reinforcement learning objective, ForgePattern aligns the generated layout patterns with specific requirements without requiring additional data collection or human annotation. Experimental results on challenging layout pattern generation tasks demonstrate that ForgePattern can effectively generate high-quality layout patterns while meeting diverse customization requirements.
- 11 April 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (5) , 1-22
Open AccessForgePattern: Flexible Layout Pattern Generation through Reinforcement Learning
Layout pattern generation provides dataset support to numerous Design for Manufacturability (DFM) studies, with different research objectives requiring distinct styles of layout patterns. However, existing layout pattern generation methods only support random generation or limited conditional generation based on image-text pairs annotated for specific tasks, which fails to meet the diverse requirements of downstream tasks. To address this limitation, we propose ForgePattern, a flexible layout pattern generation framework that accommodates diverse task-specific requirements through adaptive selection among customizable reward functions. ForgePattern leverages a customized diffusion model as the generative backbone and incorporates reinforcement learning to directly optimize the diffusion model. By maximizing the expected reward of the reinforcement learning objective, ForgePattern aligns the generated layout patterns with specific requirements without requiring additional data collection or human annotation. Experimental results on challenging layout pattern generation tasks demonstrate that ForgePattern can effectively generate high-quality layout patterns while meeting diverse customization requirements.
- Open Access
- 3 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
Rank-DSE: Neural Pareto Comparator of Microarchitecture Design Space Exploration
The complexity of microarchitecture design has surged due to the expanding design space and time-intensive verification processes. Existing regression-based machine learning methods struggle with inaccurate estimations because of limited training samples. To address these challenges, we propose Rank-DSE, a novel framework for microarchitecture design space exploration (DSE) that leverages a Neural Pareto Comparator (NPC) to directly model the comparative relationships between different architecture designs. Rank-DSE bypasses the inaccuracies of absolute PPA (performance, power, area) predictions by focusing on relative comparisons. The NPC computes the probability of one architecture dominating another and employs semi-supervised learning to reduce the reliance on labeled data. Additionally, a reinforcement-learning-based sampling scheme with an updating baseline Pareto set accelerates the exploration process. Experimental results on the ICCAD 2021 benchmark demonstrate that Rank-DSE achieves superior search quality and cost-efficiency compared to state-of-the-art methods. Specifically, Rank-DSE improves hypervolume by up to 7% while reducing exploration cost by 53.09% compared to cutting-edge approaches. These results highlight the advantages of Rank-DSE in terms of efficiency and effectiveness for microarchitecture DSE.
- 3 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
Open AccessRank-DSE: Neural Pareto Comparator of Microarchitecture Design Space Exploration
The complexity of microarchitecture design has surged due to the expanding design space and time-intensive verification processes. Existing regression-based machine learning methods struggle with inaccurate estimations because of limited training samples. To address these challenges, we propose Rank-DSE, a novel framework for microarchitecture design space exploration (DSE) that leverages a Neural Pareto Comparator (NPC) to directly model the comparative relationships between different architecture designs. Rank-DSE bypasses the inaccuracies of absolute PPA (performance, power, area) predictions by focusing on relative comparisons. The NPC computes the probability of one architecture dominating another and employs semi-supervised learning to reduce the reliance on labeled data. Additionally, a reinforcement-learning-based sampling scheme with an updating baseline Pareto set accelerates the exploration process. Experimental results on the ICCAD 2021 benchmark demonstrate that Rank-DSE achieves superior search quality and cost-efficiency compared to state-of-the-art methods. Specifically, Rank-DSE improves hypervolume by up to 7% while reducing exploration cost by 53.09% compared to cutting-edge approaches. These results highlight the advantages of Rank-DSE in terms of efficiency and effectiveness for microarchitecture DSE.
- 15 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 44 (5) , 1915-1927
RuleLearner: OPC Rule Extraction From Inverse Lithography Technique Engine
Model-based Optical Proximity Correction (OPC) with Sub-Resolution Assist Feature (SRAF) generation is a critical standard practice for compensating lithography distortions in the fabrication of integrated circuits at advanced technology nodes. Typical model-based OPC and SRAF algorithms involve the selection of user-controlled rule parameters. Conventionally, these rules are heuristically determined and applied globally throughout the correction regions, which can be time-consuming and require expert knowledge of the tool. Additionally, the correlations of rule parameters to the objectives are highly nonlinear. All these factors make designing a high-performance OPC engine for complex metal designs a non-trivial task. This paper proposes RuleLearner, a comprehensive mask optimization system designed for SRAF generation and model-based OPC in real industrial scenarios. The proposed framework learns from the guidance of an information-augmented ILT engine, which, although expressive for complex designs, is expensive to generate refined masks for a whole set of design clips. Considering the non-linearity and the trade-off between local and global performance, the extracted rule value distributions are further optimized with customized natural gradients. The sophisticated SRAF generation, the edge segmentation and movements are then guided by the rule parameter. Experimental results show that RuleLearner can be applied across different complex design patterns and achieve the best lithographic performance and computational efficiency.
- 15 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 44 (5) , 1915-1927
RuleLearner: OPC Rule Extraction From Inverse Lithography Technique Engine
Model-based Optical Proximity Correction (OPC) with Sub-Resolution Assist Feature (SRAF) generation is a critical standard practice for compensating lithography distortions in the fabrication of integrated circuits at advanced technology nodes. Typical model-based OPC and SRAF algorithms involve the selection of user-controlled rule parameters. Conventionally, these rules are heuristically determined and applied globally throughout the correction regions, which can be time-consuming and require expert knowledge of the tool. Additionally, the correlations of rule parameters to the objectives are highly nonlinear. All these factors make designing a high-performance OPC engine for complex metal designs a non-trivial task. This paper proposes RuleLearner, a comprehensive mask optimization system designed for SRAF generation and model-based OPC in real industrial scenarios. The proposed framework learns from the guidance of an information-augmented ILT engine, which, although expressive for complex designs, is expensive to generate refined masks for a whole set of design clips. Considering the non-linearity and the trade-off between local and global performance, the extracted rule value distributions are further optimized with customized natural gradients. The sophisticated SRAF generation, the edge segmentation and movements are then guided by the rule parameter. Experimental results show that RuleLearner can be applied across different complex design patterns and achieve the best lithographic performance and computational efficiency.
- 13 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-12
A Lightweight Reprogramming Framework for Cross-Device Fault Diagnosis in Edge Computing
Fault diagnosis in mechanical equipment is essential for industrial system stability and safety. However, when applying the model to different models of devices, the difference in sample feature distribution seriously affects the diagnosis effect. At the same time, traditional cloud-based deployment faces delays and resource constraints, making it unable to meet real-time requirements. This article introduces a lightweight reprogramming framework for cross-device fault diagnosis in edge computing environments. It mainly includes cloud-based C-model training and edge-based E-model reprogramming and application stages. The model introduces a lightweight feature extraction (LFE) module and a decoupled fully connected (DFC) attention mechanism to enhance feature representation and global information capture. Through lightweight reprogramming, the E-model fits the device data in actual engineering while maintaining the diagnostic capability of the C-model. We used the NVIDIA Jetson Xavier NX kit as an edge computing platform and conducted verification experiments. The results show that the proposed method achieves good diagnostic effects on engineering equipment. At the same time, it achieves excellent lightweight indicators.
- 13 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-12
A Lightweight Reprogramming Framework for Cross-Device Fault Diagnosis in Edge Computing
Fault diagnosis in mechanical equipment is essential for industrial system stability and safety. However, when applying the model to different models of devices, the difference in sample feature distribution seriously affects the diagnosis effect. At the same time, traditional cloud-based deployment faces delays and resource constraints, making it unable to meet real-time requirements. This article introduces a lightweight reprogramming framework for cross-device fault diagnosis in edge computing environments. It mainly includes cloud-based C-model training and edge-based E-model reprogramming and application stages. The model introduces a lightweight feature extraction (LFE) module and a decoupled fully connected (DFC) attention mechanism to enhance feature representation and global information capture. Through lightweight reprogramming, the E-model fits the device data in actual engineering while maintaining the diagnostic capability of the C-model. We used the NVIDIA Jetson Xavier NX kit as an edge computing platform and conducted verification experiments. The results show that the proposed method achieves good diagnostic effects on engineering equipment. At the same time, it achieves excellent lightweight indicators.
- 10 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 43 (3) , 944-955
L2O-ILT: Learning to Optimize Inverse Lithography Techniques
Inverse lithography technique (ILT) is one of the most widely used resolution enhancement techniques (RETs) to compensate for the diffraction effect in the lithography process. However, ILT suffers from runtime overhead issues with the shrinking size of technology nodes. In this article, our proposed L2O-ILT framework unrolls the iterative ILT optimization algorithm into a learnable neural network with high interpretability, which can generate a high-quality initial mask for fast refinement. Experimental results demonstrate that our method achieves better performance on both mask printability and runtime than the previous methods.
- 10 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 43 (3) , 944-955
L2O-ILT: Learning to Optimize Inverse Lithography Techniques
Inverse lithography technique (ILT) is one of the most widely used resolution enhancement techniques (RETs) to compensate for the diffraction effect in the lithography process. However, ILT suffers from runtime overhead issues with the shrinking size of technology nodes. In this article, our proposed L2O-ILT framework unrolls the iterative ILT optimization algorithm into a learnable neural network with high interpretability, which can generate a high-quality initial mask for fast refinement. Experimental results demonstrate that our method achieves better performance on both mask printability and runtime than the previous methods.
- 14 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 42 (12) , 5020-5033
DevelSet: Deep Neural Level Set for Instant Mask Optimization
As one of the key techniques for resolution enhancement technologies (RETs), optical proximity correction (OPC) suffers from prohibitive computational costs as feature sizes continue to shrink. Inverse lithography techniques (ILTs) treat the mask optimization process as an inverse imaging problem, yielding high-quality curvilinear masks. However, ILT methods often fall short of printability and manufacturability due to their time-consuming procedures and excessive computational overhead. In this article, we propose DevelSet, a potent metal layer OPC engine that replaces discrete pixel-based masks with implicit level set-based representations. With a GPU-accelerated lithography simulator, DevelSet achieves end-to-end mask optimization using a neural network to provide quasi-optimized level set initialization and further evolution with a CUDA-based mask optimizer for fast convergence. The backbone of DevelSet-Net is a transformer-based multibranch neural network that offers a parameter selector to eliminate the need for manual parameter initialization. Experimental results demonstrate that the DevelSet framework outperforms state-of-the-art approaches in terms of printability while achieving fast runtime performance (around 1 s). We expect this enhanced level set technique, coupled with a CUDA/DNN accelerated joint optimization paradigm, to have a substantial impact on industrial mask optimization solutions.
- 14 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 42 (12) , 5020-5033
DevelSet: Deep Neural Level Set for Instant Mask Optimization
As one of the key techniques for resolution enhancement technologies (RETs), optical proximity correction (OPC) suffers from prohibitive computational costs as feature sizes continue to shrink. Inverse lithography techniques (ILTs) treat the mask optimization process as an inverse imaging problem, yielding high-quality curvilinear masks. However, ILT methods often fall short of printability and manufacturability due to their time-consuming procedures and excessive computational overhead. In this article, we propose DevelSet, a potent metal layer OPC engine that replaces discrete pixel-based masks with implicit level set-based representations. With a GPU-accelerated lithography simulator, DevelSet achieves end-to-end mask optimization using a neural network to provide quasi-optimized level set initialization and further evolution with a CUDA-based mask optimizer for fast convergence. The backbone of DevelSet-Net is a transformer-based multibranch neural network that offers a parameter selector to eliminate the need for manual parameter initialization. Experimental results demonstrate that the DevelSet framework outperforms state-of-the-art approaches in terms of printability while achieving fast runtime performance (around 1 s). We expect this enhanced level set technique, coupled with a CUDA/DNN accelerated joint optimization paradigm, to have a substantial impact on industrial mask optimization solutions.