Martin D. F. Wong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Martin D. F. Wong has a total of 85 co-authors. Here are the top co-authors by number of shared publications:
- Bei Yu
The Chinese University of Hong Kong
17 shared publications - Evangeline F. Y. Young
Chinese University of Hong Kong, Hong Kong
13 shared publications - Su Zheng
Computer Science and Engineering, The Chinese University of Hong Kong
10 shared publications - Lixin Liu
The Chinese University of Hong Kong, Hong Kong, Hong Kong, China
5 shared publications - Yibo Lin
School of Integrated Circuits, Peking University
5 shared publications - Xinshi Zang
The Chinese University of Hong Kong, Shatin, Hong Kong
4 shared publications
Latest Publications
- Open Access
- 28 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
PointDetector: Large-Scale Lithography Hotspot Detection via Point Cloud Classification
Lithography hotspot detection is essential for ensuring yield and reliability in integrated circuit manufacturing. However, existing approaches typically analyze small layout clips in isolation, resulting in inefficient and fragmented detection workflows. In this work, we propose a novel formulation of hotspot detection as a point cloud classification problem and introduce a tailored neural network architecture based on submanifold sparse convolution to efficiently process large layouts. Our method operates directly on polygon segments used in mask optimization, enabling fine-grained predictions aligned with downstream hotspot fixing. By representing layouts as point clouds and leveraging the sparsity of the data, our approach supports large-scale detection with a single forward pass, improving runtime over the compared clip-based methods. Experimental results on large benchmark datasets show that our method achieves up to 20 × faster inference than existing detection-based techniques while also improving precision, recall, F1 score, and false alarms by 30.0%, 59.6%, 46.0%, and 80.8%, respectively, on our large-scale benchmark. These results highlight the potential of our approach to enable fast, scalable, and fine-grained hotspot detection for advanced IC manufacturing.
- 28 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
Open AccessPointDetector: Large-Scale Lithography Hotspot Detection via Point Cloud Classification
Lithography hotspot detection is essential for ensuring yield and reliability in integrated circuit manufacturing. However, existing approaches typically analyze small layout clips in isolation, resulting in inefficient and fragmented detection workflows. In this work, we propose a novel formulation of hotspot detection as a point cloud classification problem and introduce a tailored neural network architecture based on submanifold sparse convolution to efficiently process large layouts. Our method operates directly on polygon segments used in mask optimization, enabling fine-grained predictions aligned with downstream hotspot fixing. By representing layouts as point clouds and leveraging the sparsity of the data, our approach supports large-scale detection with a single forward pass, improving runtime over the compared clip-based methods. Experimental results on large benchmark datasets show that our method achieves up to 20 × faster inference than existing detection-based techniques while also improving precision, recall, F1 score, and false alarms by 30.0%, 59.6%, 46.0%, and 80.8%, respectively, on our large-scale benchmark. These results highlight the potential of our approach to enable fast, scalable, and fine-grained hotspot detection for advanced IC manufacturing.
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
DR. OPC: Differentiable Rasterization for Advanced Optical Proximity Correction
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessDR. OPC: Differentiable Rasterization for Advanced Optical Proximity Correction
- 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
Accelet: Hardware-software Co-design of Chiplet-based DNN Accelerators
The rapid growth of deep neural network (DNN) applications has brought unprecedented challenges to the functionality and scalability of accelerators. To address these challenges, the multi-chip-module (MCM) approach facilitates the efficient and cost-effective development of accelerator architectures that can scale from mobile devices to data centers. Chiplet-based systems necessitate a robust platform capable of supporting both hardware modeling and software evaluation, since the performance of the entire system is not only determined by the design of functional units and interconnects, but also strongly affected by the scheduling and mapping of the dataflow. In this article, we propose Accelet, a comprehensive framework designed to articulate the hardware architecture of chiplet-based accelerators, as well as to investigate the scheduling and mapping strategies for DNN dataflows. Accelet’s versatility in hardware and software modeling empowers the co-design of critical components, including functional units, interconnect topologies, system architectures, and mapping algorithms. Consequently, Accelet can stand as a pivotal tool for the hardware-software co-design of chiplet-based DNN accelerators. Accelet is available at https://github.com/shelljane/accelet-modeling .
- 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 AccessAccelet: Hardware-software Co-design of Chiplet-based DNN Accelerators
The rapid growth of deep neural network (DNN) applications has brought unprecedented challenges to the functionality and scalability of accelerators. To address these challenges, the multi-chip-module (MCM) approach facilitates the efficient and cost-effective development of accelerator architectures that can scale from mobile devices to data centers. Chiplet-based systems necessitate a robust platform capable of supporting both hardware modeling and software evaluation, since the performance of the entire system is not only determined by the design of functional units and interconnects, but also strongly affected by the scheduling and mapping of the dataflow. In this article, we propose Accelet, a comprehensive framework designed to articulate the hardware architecture of chiplet-based accelerators, as well as to investigate the scheduling and mapping strategies for DNN dataflows. Accelet’s versatility in hardware and software modeling empowers the co-design of critical components, including functional units, interconnect topologies, system architectures, and mapping algorithms. Consequently, Accelet can stand as a pivotal tool for the hardware-software co-design of chiplet-based DNN accelerators. Accelet is available at https://github.com/shelljane/accelet-modeling .
- Open Access
- 19 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31  (4) , 1-15
MALT: ML Assisted Shallow-Light Tree Construction
Timing is a critical issue in electronic design automation (EDA). To reduce the delay of a net, an important strategy is to minimize the path lengths from the source to the sinks. However, minimizing the path lengths will inevitably sacrifice the total wirelength. To balance the two objectives, researchers use shallow-light tree (SLT) to model and optimize the problem. In this article, we introduce MALT, a novel approach that uses a neural network to guide the construction of Steiner shallow-light trees. The constructed trees are further refined by a dynamic programming-based branch merging algorithm, which improves the wirelength without sacrificing the path lengths of any sinks. Our experimental results demonstrate that the proposed framework achieves significant improvements over both state-of-the-art traditional SLT generation algorithms and existing machine learning enhanced methods.
- 19 March 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31  (4) , 1-15
Open AccessMALT: ML Assisted Shallow-Light Tree Construction
Timing is a critical issue in electronic design automation (EDA). To reduce the delay of a net, an important strategy is to minimize the path lengths from the source to the sinks. However, minimizing the path lengths will inevitably sacrifice the total wirelength. To balance the two objectives, researchers use shallow-light tree (SLT) to model and optimize the problem. In this article, we introduce MALT, a novel approach that uses a neural network to guide the construction of Steiner shallow-light trees. The constructed trees are further refined by a dynamic programming-based branch merging algorithm, which improves the wirelength without sacrificing the path lengths of any sinks. Our experimental results demonstrate that the proposed framework achieves significant improvements over both state-of-the-art traditional SLT generation algorithms and existing machine learning enhanced methods.
- Open Access
- 15 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 55-63
Gradient-Guided RC Weighting for Timing-Driven Global Routing
- 15 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 55-63
Open AccessGradient-Guided RC Weighting for Timing-Driven Global Routing
- 6 March 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
Scalable Parallel Logic Resubstitution with Advanced Don’t Care Optimization
Resubstitution is a powerful logic optimization technique that restructures And-Inverter Graphs (AIGs) to achieve significant reductions in circuit area and delay. In this paper, we present a comprehensive, massively parallel algorithm for window-based k-resubstitution that exploits the computational power of GPUs while integrating advanced support for satisfiability and observability don’t cares (SDCs and ODCs). Our approach introduces adaptive, parallelized procedures for divisor collection, candidate evaluation, and network updates. To support don’t cares, we develop efficient, parallel-compatible algorithms for computing SDCs and ODCs, designed specifically for scalability. Experimental results on large benchmarks demonstrate that our GPU-based resubstitution achieves up to 63.7× speedup over state-of-the-art CPU-based tools, with comparable or improved area and delay. When integrating our resubstitution into a GPU-based optimization sequence resyn2rs, we observe a 43.0× speedup over ABC with a 0.9% reduction in area. Notably, our implicit exploitation of SDCs further improves optimization quality with minimal runtime overhead, maintaining a 35.0× speedup. Finally, our fully parallel ODC computation and incremental window expansion techniques significantly enhance scalability and performance for large circuits.
- 6 March 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
Scalable Parallel Logic Resubstitution with Advanced Don’t Care Optimization
Resubstitution is a powerful logic optimization technique that restructures And-Inverter Graphs (AIGs) to achieve significant reductions in circuit area and delay. In this paper, we present a comprehensive, massively parallel algorithm for window-based k-resubstitution that exploits the computational power of GPUs while integrating advanced support for satisfiability and observability don’t cares (SDCs and ODCs). Our approach introduces adaptive, parallelized procedures for divisor collection, candidate evaluation, and network updates. To support don’t cares, we develop efficient, parallel-compatible algorithms for computing SDCs and ODCs, designed specifically for scalability. Experimental results on large benchmarks demonstrate that our GPU-based resubstitution achieves up to 63.7× speedup over state-of-the-art CPU-based tools, with comparable or improved area and delay. When integrating our resubstitution into a GPU-based optimization sequence resyn2rs, we observe a 43.0× speedup over ABC with a 0.9% reduction in area. Notably, our implicit exploitation of SDCs further improves optimization quality with minimal runtime overhead, maintaining a 35.0× speedup. Finally, our fully parallel ODC computation and incremental window expansion techniques significantly enhance scalability and performance for large circuits.
- Open Access
- 11 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31  (3) , 1-20
Routing-aware Legal Hybrid Bonding Terminal Assignment for 3D Face-to-Face Stacked ICs
Face-to-face (F2F) stacked three-dimensional (3D) IC is a promising alternative for scaling beyond Moore’s Law. In F2F 3D ICs, dies are connected through bonding terminals whose positions can significantly impact routing performance. Further, there exists resource competition among all the 3D nets due to the constrained bonding terminal number. In advanced technology nodes, traditional bonding terminal planning may also introduce legality challenges of bonding terminals, as the metal pitches can be much smaller than the sizes of bonding terminals. Previous works attempt to insert bonding terminals automatically using existing 2D commercial P&R tools and then consider interdie connection legality, but they fail to take the legality and routing performance into account simultaneously. In this article, we provide a novel bonding terminal assignment formulation for effective routing-aware bonding terminal planning. We explore the generalized assignment formulation and provide the routability guidance in our hybrid bonding terminal assignment problem. Our framework, BTAssign , offers a strict legality guarantee and an iterative solution. We provide two versions of the BTAssign framework, BTAssign-WL [ 1 ] and BTAssign-R, which BTAssign-R extends BTAssign-WL [ 1 ] by considering routability. The experiments are conducted on 18 open source designs with various 3D net densities and the most advanced bonding scale. The results reveal that all the testing cases with different partitioning and placement strategies could gain benefits from our BTAssign framework.
- 11 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31  (3) , 1-20
Open AccessRouting-aware Legal Hybrid Bonding Terminal Assignment for 3D Face-to-Face Stacked ICs
Face-to-face (F2F) stacked three-dimensional (3D) IC is a promising alternative for scaling beyond Moore’s Law. In F2F 3D ICs, dies are connected through bonding terminals whose positions can significantly impact routing performance. Further, there exists resource competition among all the 3D nets due to the constrained bonding terminal number. In advanced technology nodes, traditional bonding terminal planning may also introduce legality challenges of bonding terminals, as the metal pitches can be much smaller than the sizes of bonding terminals. Previous works attempt to insert bonding terminals automatically using existing 2D commercial P&R tools and then consider interdie connection legality, but they fail to take the legality and routing performance into account simultaneously. In this article, we provide a novel bonding terminal assignment formulation for effective routing-aware bonding terminal planning. We explore the generalized assignment formulation and provide the routability guidance in our hybrid bonding terminal assignment problem. Our framework, BTAssign , offers a strict legality guarantee and an iterative solution. We provide two versions of the BTAssign framework, BTAssign-WL [ 1 ] and BTAssign-R, which BTAssign-R extends BTAssign-WL [ 1 ] by considering routability. The experiments are conducted on 18 open source designs with various 3D net densities and the most advanced bonding scale. The results reveal that all the testing cases with different partitioning and placement strategies could gain benefits from our BTAssign framework.
- 23 December 2025
- 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
KirchhoffNet: End-to-End Analog Circuit Acceleration for ODE-Based Neural Networks
This paper introduces KirchhoffNet, a novel class of neural network models inspired by the principles of analog electronic circuitry, specifically Kirchhoff’s laws. KirchhoffNet operates as an analog circuit, where the network input is represented by initial node voltages, and the output corresponds to the node voltages at a specific time. The dynamics of the node voltages are governed by learnable parameters on the edges, and the evolution of these voltages follows a system of ordinary differential equations (ODEs). Despite the absence of traditional neural network components such as convolutional layers, KirchhoffNet achieves outstanding performance across a wide range of machine learning tasks. We further demonstrate that KirchhoffNet is capable of computing diffusion models, making it a promising candidate for accelerating modern generative AI applications. Most notably, KirchhoffNet can be implemented as a high-speed & low-power analog integrated circuit, which introduces a compelling advantage: irrespective of the number of parameters in the network, its on-chip forward calculation can always be completed within a short time. This property makes KirchhoffNet a highly attractive and scalable paradigm for implementing large-scale neural networks, opening new avenues in the realm of analog neural networks for artificial intelligence.
- 23 December 2025
- 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
KirchhoffNet: End-to-End Analog Circuit Acceleration for ODE-Based Neural Networks
This paper introduces KirchhoffNet, a novel class of neural network models inspired by the principles of analog electronic circuitry, specifically Kirchhoff’s laws. KirchhoffNet operates as an analog circuit, where the network input is represented by initial node voltages, and the output corresponds to the node voltages at a specific time. The dynamics of the node voltages are governed by learnable parameters on the edges, and the evolution of these voltages follows a system of ordinary differential equations (ODEs). Despite the absence of traditional neural network components such as convolutional layers, KirchhoffNet achieves outstanding performance across a wide range of machine learning tasks. We further demonstrate that KirchhoffNet is capable of computing diffusion models, making it a promising candidate for accelerating modern generative AI applications. Most notably, KirchhoffNet can be implemented as a high-speed & low-power analog integrated circuit, which introduces a compelling advantage: irrespective of the number of parameters in the network, its on-chip forward calculation can always be completed within a short time. This property makes KirchhoffNet a highly attractive and scalable paradigm for implementing large-scale neural networks, opening new avenues in the realm of analog neural networks for artificial intelligence.
- 1 December 2025
- 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
Clustering and Binding Flip-Flops in Global Placement with Analytical Optimization
Modern Integrated Circuits (ICs) have obtained remarkable performance due to high integration density. Many challenges arose consequently as the circuit size expanded rapidly, causing severe quality degradation of the circuits. Power optimization has been a bottleneck for circuit design for a long time, and is becoming increasingly expensive for advanced technology nodes. The clock power contributes a large part of the power consumption, which is usually caused by long clock tree wirelength. Thus, the flip-flop clustering becomes an effective method to reduce the clock wirelength. Advanced technology introduced Multi-Bit Flip-Flop to further reduce the clock network size and power consumption. In this paper, we adopt the Multi-Bit Flip-Flop technology and propose an analytical placement framework for flip-flop clustering and binding optimization. A balanced clustering method is proposed to perform the Multi-Bit Flip-Flop generation. Experimental results show that our methods significantly outperform the state-of-the-art OpenROAD flows in terms of timing, power and clock tree construction.
- 1 December 2025
- 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
Clustering and Binding Flip-Flops in Global Placement with Analytical Optimization
Modern Integrated Circuits (ICs) have obtained remarkable performance due to high integration density. Many challenges arose consequently as the circuit size expanded rapidly, causing severe quality degradation of the circuits. Power optimization has been a bottleneck for circuit design for a long time, and is becoming increasingly expensive for advanced technology nodes. The clock power contributes a large part of the power consumption, which is usually caused by long clock tree wirelength. Thus, the flip-flop clustering becomes an effective method to reduce the clock wirelength. Advanced technology introduced Multi-Bit Flip-Flop to further reduce the clock network size and power consumption. In this paper, we adopt the Multi-Bit Flip-Flop technology and propose an analytical placement framework for flip-flop clustering and binding optimization. A balanced clustering method is proposed to perform the Multi-Bit Flip-Flop generation. Experimental results show that our methods significantly outperform the state-of-the-art OpenROAD flows in terms of timing, power and clock tree construction.