Tong Lin
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
Research Subjects
Tong Lin has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Bah-Hwee Gwee
Nanyang Technological University
9 shared publications - Yiqiong Shi
Nanyang Technological University
8 shared publications - Deruo Cheng
Nanyang Technological University
7 shared publications - Xuenong Hong
Nanyang Technological University, Singapore
4 shared publications - Yee-Yang Tee
Nanyang Technological University, Singapore
4 shared publications - Kar-Ann Toh
School of Electrical and Electronic Engineering, Yonsei University, Seoul, South Korea
2 shared publications
Latest Publications
- Open Access
- 26 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 45-49
InvMUX: Cascaded Inverter-MUX-Based Logic Locking Against Machine Learning Attacks
- 26 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 45-49
Open AccessInvMUX: Cascaded Inverter-MUX-Based Logic Locking Against Machine Learning Attacks
- 12 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Intelligent Systems
- Vol. 41 (1) , 37-45
Integrated Circuit Image Synthesis for Unsupervised Circuit Annotation via Shape Consistent Image Translation
Deep learning techniques achieve promising performance on the circuit annotation task for the hardware assurance of integrated circuits, but are reliant on large amounts of costly, labeled training data. Recently, image synthesis techniques have been explored to mitigate this reliance. However, exiting methods often lack precise pixel-level correspondence between the input image and synthetic image. When applied to IC images, this could lead to inconsistencies in the circuit structures, which then cause changes in circuit connectivity. In this paper, we propose shape consistent image translation (SCIT) to synthesize IC images that have a high pixel-level correspondence with the input images. Our experiments show that a segmentation model trained SCIT-generated images reduces the number of circuit connection errors by 56.26% compared to the second best reported technique. Our proposed SCIT also produced synthetic IC images of the highest image quality when evaluated across three IC image datasets.
- 12 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Intelligent Systems
- Vol. 41 (1) , 37-45
Integrated Circuit Image Synthesis for Unsupervised Circuit Annotation via Shape Consistent Image Translation
Deep learning techniques achieve promising performance on the circuit annotation task for the hardware assurance of integrated circuits, but are reliant on large amounts of costly, labeled training data. Recently, image synthesis techniques have been explored to mitigate this reliance. However, exiting methods often lack precise pixel-level correspondence between the input image and synthetic image. When applied to IC images, this could lead to inconsistencies in the circuit structures, which then cause changes in circuit connectivity. In this paper, we propose shape consistent image translation (SCIT) to synthesize IC images that have a high pixel-level correspondence with the input images. Our experiments show that a segmentation model trained SCIT-generated images reduces the number of circuit connection errors by 56.26% compared to the second best reported technique. Our proposed SCIT also produced synthetic IC images of the highest image quality when evaluated across three IC image datasets.
- 23 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Intelligent Systems
- Vol. 40 (3) , 59-67
GNNReveal: A Novel Graph-Neural-Network-Based Attack Method for Integrated Circuit Logic Gate Decamouflaging
Recent advancement in circuit extraction poses new threats to Integrated Circuits (ICs) Intellectual Property (IP) protection. Hardware obfuscation by IC logic gate camouflaging results in logic gates with unknown functionalities in an extracted netlist, protecting manufactured ICs from circuit extraction. Logic gate decamouflaging attacks have been proposed. Conventional attacks are not suitable for large-scale camouflaging due to infeasible computation cost. Existing Graph Neural Network(GNN)-based methods are efficient, but are less effective in differentiating between FanIn and FanOut structures which hampers their effectiveness. In this paper, we propose a novel GNN-based attack method, namely GNNReveal, for logic gate de-camouflaging. Our proposed GNNReveal performs separate FI/FO aggregations to generate unique node embeddings for logic gates with different functionalities, allowing for direct node classification for logic gate de-camouflaging. Our experiments show that our proposed GNNReveal achieved high de-camouflaging accuracy and significantly outperformed competing methods by at most 20%.
- 23 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Intelligent Systems
- Vol. 40 (3) , 59-67
GNNReveal: A Novel Graph-Neural-Network-Based Attack Method for Integrated Circuit Logic Gate Decamouflaging
Recent advancement in circuit extraction poses new threats to Integrated Circuits (ICs) Intellectual Property (IP) protection. Hardware obfuscation by IC logic gate camouflaging results in logic gates with unknown functionalities in an extracted netlist, protecting manufactured ICs from circuit extraction. Logic gate decamouflaging attacks have been proposed. Conventional attacks are not suitable for large-scale camouflaging due to infeasible computation cost. Existing Graph Neural Network(GNN)-based methods are efficient, but are less effective in differentiating between FanIn and FanOut structures which hampers their effectiveness. In this paper, we propose a novel GNN-based attack method, namely GNNReveal, for logic gate de-camouflaging. Our proposed GNNReveal performs separate FI/FO aggregations to generate unique node embeddings for logic gates with different functionalities, allowing for direct node classification for logic gate de-camouflaging. Our experiments show that our proposed GNNReveal achieved high de-camouflaging accuracy and significantly outperformed competing methods by at most 20%.
- 17 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Intelligent Systems
- Vol. 39 (1) , 37-45
Integrated Circuit Mask–Generative Adversarial Network for Circuit Annotation With Targeted Data Augmentation
In recent years, deep learning-based segmentation techniques have been applied to circuit annotation for the hardware assurance of integrated circuits (ICs). However, imperfections in circuit images often cause incorrectly segmented pixels, which result in critical circuit connection errors that are detrimental to subsequent circuit analysis. To mitigate such circuit connection errors, this paper proposes a targeted data augmentation framework for deep learning-based circuit annotation, termed Integrated Circuit Mask-GAN (ICMG), which generates circuit images containing the aforementioned imperfections through GAN-based image translation from synthetic circuit masks. The circuit masks are synthesized by a novel mask generation process that incorporates our domain knowledge of IC layouts and is configurable to emulate various image imperfections such as material residuals in the sample preparation or distortion in the imaging process. In our experiments on a microcontroller IC, our proposed ICMG greatly reduced the circuit connection errors and the required manual effort for data labeling compared to the reported techniques.
- 17 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Intelligent Systems
- Vol. 39 (1) , 37-45
Integrated Circuit Mask–Generative Adversarial Network for Circuit Annotation With Targeted Data Augmentation
In recent years, deep learning-based segmentation techniques have been applied to circuit annotation for the hardware assurance of integrated circuits (ICs). However, imperfections in circuit images often cause incorrectly segmented pixels, which result in critical circuit connection errors that are detrimental to subsequent circuit analysis. To mitigate such circuit connection errors, this paper proposes a targeted data augmentation framework for deep learning-based circuit annotation, termed Integrated Circuit Mask-GAN (ICMG), which generates circuit images containing the aforementioned imperfections through GAN-based image translation from synthetic circuit masks. The circuit masks are synthesized by a novel mask generation process that incorporates our domain knowledge of IC layouts and is configurable to emulate various image imperfections such as material residuals in the sample preparation or distortion in the imaging process. In our experiments on a microcontroller IC, our proposed ICMG greatly reduced the circuit connection errors and the required manual effort for data labeling compared to the reported techniques.
- 18 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-8
Gradient Descent Optimizes Normalization-Free ResNets
Recent empirical studies observe that even without normalization, a deep residual network can be trained reliably. We call such a structure as normalization-free Residual Networks (N-F ResNets), which add a learnable parameter $\alpha$ to control the scale of the residual block instead of normalization. However, the theoretical understanding on N-F ResNets is still limited despite their empirical success. In this paper, we provide the first theoretical understanding of N-F ResNets from two perspectives. Firstly, we prove that the gradient descent (GD) algorithm can find the global minimum of the training loss at a linear rate for over-parameterized N-F ResNets. Secondly, we prove that N-F ResNets can avoid the gradient exploding or vanishing problem, by initializing the key parameter $\alpha$ to be a small constant. Notably, we demonstrate that the gradients of N-F ResNets are more stable than those of ResNets with Kaiming initialization. Moreover, empirical experiments on benchmark datasets verify our theoretical results.- 18 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-8
Gradient Descent Optimizes Normalization-Free ResNets
Recent empirical studies observe that even without normalization, a deep residual network can be trained reliably. We call such a structure as normalization-free Residual Networks (N-F ResNets), which add a learnable parameter $\alpha$ to control the scale of the residual block instead of normalization. However, the theoretical understanding on N-F ResNets is still limited despite their empirical success. In this paper, we provide the first theoretical understanding of N-F ResNets from two perspectives. Firstly, we prove that the gradient descent (GD) algorithm can find the global minimum of the training loss at a linear rate for over-parameterized N-F ResNets. Secondly, we prove that N-F ResNets can avoid the gradient exploding or vanishing problem, by initializing the key parameter $\alpha$ to be a small constant. Notably, we demonstrate that the gradients of N-F ResNets are more stable than those of ResNets with Kaiming initialization. Moreover, empirical experiments on benchmark datasets verify our theoretical results.- 6 April 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 70 (9) , 3694-3698
Patch-Based Adversarial Training for Error-Aware Circuit Annotation of Delayered IC Images
Circuit annotation is one of the most important tasks in the analysis of integrated circuit images for hardware assurance. Recently, deep learning methods have been adopted for circuit annotation due to their promising accuracy. However, the pixel-wise optimization metrics in deep learning methods are insufficient to ensure a good spatial contiguity in the segmentation results. This could result in circuit connection errors that are detrimental to the subsequent circuit analysis. In this paper, a patch-based adversarial training framework is proposed to mitigate such circuit connection errors. Our proposed method consists of a encoder-decoder segmentation network and a patch-based discriminator to provide adversarial supervision for the segmentation network. The adversarial training aligns the distributions of the segmentation results with the ground truth in patches, which we hypothesize to be more effective for mitigating circuit connection errors. In our experiments, we have achieved a 17.4% reduction in circuit connection errors as compared to the second best reported technique. We investigated the explainability of our proposed method through a heat map analysis, and demonstrate that our patch-based discriminator has a higher feature response in regions that are likely to contain circuit connection errors.
- 6 April 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 70 (9) , 3694-3698
Patch-Based Adversarial Training for Error-Aware Circuit Annotation of Delayered IC Images
Circuit annotation is one of the most important tasks in the analysis of integrated circuit images for hardware assurance. Recently, deep learning methods have been adopted for circuit annotation due to their promising accuracy. However, the pixel-wise optimization metrics in deep learning methods are insufficient to ensure a good spatial contiguity in the segmentation results. This could result in circuit connection errors that are detrimental to the subsequent circuit analysis. In this paper, a patch-based adversarial training framework is proposed to mitigate such circuit connection errors. Our proposed method consists of a encoder-decoder segmentation network and a patch-based discriminator to provide adversarial supervision for the segmentation network. The adversarial training aligns the distributions of the segmentation results with the ground truth in patches, which we hypothesize to be more effective for mitigating circuit connection errors. In our experiments, we have achieved a 17.4% reduction in circuit connection errors as compared to the second best reported technique. We investigated the explainability of our proposed method through a heat map analysis, and demonstrate that our patch-based discriminator has a higher feature response in regions that are likely to contain circuit connection errors.
- 7 February 2023
- Published by Springer Nature
- p. 129-151
Domain-Integrated Machine Learning for IC Image Analysis
- 7 February 2023
- Published by Springer Nature
- p. 129-151
Domain-Integrated Machine Learning for IC Image Analysis
- 16 August 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 4 (5) , 1199-1213
GraphClusNet: A Hierarchical Graph Neural Network for Recovered Circuit Netlist Partitioning
Hardware Assurance (HA) is imperative to ensure the integrity of Integrated Circuits (ICs) after manufacturing. To verify the integrity of an IC would require the extraction of relevant functional block of interest through circuit partitioning. This is usually done by converting a recovered circuit netlist into a circuit graph and subsequently performing unsupervised clustering on the graph. However, circuit graphs are difficult to cluster due to the existence of inherent hierarchies and clusters of imbalanced sizes. In this paper, we propose a novel hierarchical Graph Neural Network (GNN), termed as GraphClusNet, to perform circuit graph clustering in a multi-stage process to achieve near-optimal results. We analytically derive a normalized-cut-based loss function, which allows for clusters of imbalanced sizes. We train our GraphClusNet in a multi-stage hierarchical fashion to regularize intermediate node embedding, which greatly alleviates the over-smoothing effect in deep GNNs. We further propose a multi-scale location-based node feature and feature concatenation to provide a good initialization to stabilize the training of our GNN. Based on the experiments of synthetic graphs, we show that our proposed GraphClusNet obtained solutions with n-cut values up to$\mathbf {65\times }$ closer to the global optimum. On FPGA circuit graphs, our proposed GraphClusNet outperformed the reported methods with up to 30% improvement in Normalized Mutual Information for circuit graph clustering. Impact Statement— The security of Integrated Circuits (ICs) is of paramount importance in data sensitive industries. Hardware Assurance (HA) through circuit analysis ensures the highest level of confidence in manufactured ICs to function as intended and are free of known vulnerabilities. The success of HA relies critically on circuit partitioning to accurately segment the IC into smaller functional blocks for design verification. Existing methods have lower accuracy of$\sim$ 56% in partitioning circuits with complex structures. Our proposed method leverages on the latest development of artificial intelligence (AI) on graph and improves the accuracy to$>$ 90% through graph machine learning. This improvement ensures a highly accurate partition of IC which largely increases the success rate of HA. To the best of our knowledge, our proposed method is the first AI method to partition circuit. It opens new possibility of using AI for security of IC.- 16 August 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 4 (5) , 1199-1213
GraphClusNet: A Hierarchical Graph Neural Network for Recovered Circuit Netlist Partitioning
Hardware Assurance (HA) is imperative to ensure the integrity of Integrated Circuits (ICs) after manufacturing. To verify the integrity of an IC would require the extraction of relevant functional block of interest through circuit partitioning. This is usually done by converting a recovered circuit netlist into a circuit graph and subsequently performing unsupervised clustering on the graph. However, circuit graphs are difficult to cluster due to the existence of inherent hierarchies and clusters of imbalanced sizes. In this paper, we propose a novel hierarchical Graph Neural Network (GNN), termed as GraphClusNet, to perform circuit graph clustering in a multi-stage process to achieve near-optimal results. We analytically derive a normalized-cut-based loss function, which allows for clusters of imbalanced sizes. We train our GraphClusNet in a multi-stage hierarchical fashion to regularize intermediate node embedding, which greatly alleviates the over-smoothing effect in deep GNNs. We further propose a multi-scale location-based node feature and feature concatenation to provide a good initialization to stabilize the training of our GNN. Based on the experiments of synthetic graphs, we show that our proposed GraphClusNet obtained solutions with n-cut values up to$\mathbf {65\times }$ closer to the global optimum. On FPGA circuit graphs, our proposed GraphClusNet outperformed the reported methods with up to 30% improvement in Normalized Mutual Information for circuit graph clustering. Impact Statement— The security of Integrated Circuits (ICs) is of paramount importance in data sensitive industries. Hardware Assurance (HA) through circuit analysis ensures the highest level of confidence in manufactured ICs to function as intended and are free of known vulnerabilities. The success of HA relies critically on circuit partitioning to accurately segment the IC into smaller functional blocks for design verification. Existing methods have lower accuracy of$\sim$ 56% in partitioning circuits with complex structures. Our proposed method leverages on the latest development of artificial intelligence (AI) on graph and improves the accuracy to$>$ 90% through graph machine learning. This improvement ensures a highly accurate partition of IC which largely increases the success rate of HA. To the best of our knowledge, our proposed method is the first AI method to partition circuit. It opens new possibility of using AI for security of IC.- Open Access
- 28 March 2022
- Published by American Physical Society (APS) in Physical Review B
- Vol. 105 (10) , L100508
Transient Higgs oscillations and high-order nonlinear light-Higgs coupling in a terahertz wave driven NbN superconductor
We study the nonlinear optical response in a superconducting NbN thin film with a strong terahertz (THz) wave. In. addition to the expected third-harmonic generation, we observe a transient oscillation which softens in frequency with temperature increasing towards the superconducting transition temperature . We identify this feature as transient Higgs oscillation. To verify this proposal, we introduce a time-frequency resolved technique, a spectrogram for visualizing the THz spectrum. The dynamic decaying behavior of the oscillation is observed, which is consistent with the theoretical expectation of intrinsic Higgs oscillation. Moreover, we observe a higher-order nonlinear optics effect, i.e., fifth-harmonic generation, which we assign to the higher-order coupling between the Higgs mode and electromagnetic field.- 28 March 2022
- Published by American Physical Society (APS) in Physical Review B
- Vol. 105 (10) , L100508
Open AccessTransient Higgs oscillations and high-order nonlinear light-Higgs coupling in a terahertz wave driven NbN superconductor
We study the nonlinear optical response in a superconducting NbN thin film with a strong terahertz (THz) wave. In. addition to the expected third-harmonic generation, we observe a transient oscillation which softens in frequency with temperature increasing towards the superconducting transition temperature . We identify this feature as transient Higgs oscillation. To verify this proposal, we introduce a time-frequency resolved technique, a spectrogram for visualizing the THz spectrum. The dynamic decaying behavior of the oscillation is observed, which is consistent with the theoretical expectation of intrinsic Higgs oscillation. Moreover, we observe a higher-order nonlinear optics effect, i.e., fifth-harmonic generation, which we assign to the higher-order coupling between the Higgs mode and electromagnetic field. - 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 552-559
A Supervisory Mask Attentional Network for Person Re-Identification in Uniform Dress Scenes
Person re-identification (Re-ID) aims at retrieving a person's identification across multiple non-overlapping cameras. While recent Re-ID methods have achieved significant success on a number of benchmark datasets, most of them are still insufficient in the scenes with unified or highly similar dressing like construction sites, schools and factories. To address this problem, we propose a supervisory mask attention network (SMA-Net). Our approach combines two key components: (1) ROI mask mapping (RMM) is a supervisory branch to provide ROI mappings that divide a person region into several parts; (2) Partial mask attention (PMA) integrates channel and space attention mechanisms that focus on local features and different accessories in each ROI. Therefore, the network can pay more attention to the local features and different accessories herein. Compared with the state of the art methods, SMA-Net demonstrates excellent performance on our dataset of construction scenes, with improvement of 6.65% in mAP and 3.8% in Rank-1 accuracy.
- 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 552-559
A Supervisory Mask Attentional Network for Person Re-Identification in Uniform Dress Scenes
Person re-identification (Re-ID) aims at retrieving a person's identification across multiple non-overlapping cameras. While recent Re-ID methods have achieved significant success on a number of benchmark datasets, most of them are still insufficient in the scenes with unified or highly similar dressing like construction sites, schools and factories. To address this problem, we propose a supervisory mask attention network (SMA-Net). Our approach combines two key components: (1) ROI mask mapping (RMM) is a supervisory branch to provide ROI mappings that divide a person region into several parts; (2) Partial mask attention (PMA) integrates channel and space attention mechanisms that focus on local features and different accessories in each ROI. Therefore, the network can pay more attention to the local features and different accessories herein. Compared with the state of the art methods, SMA-Net demonstrates excellent performance on our dataset of construction scenes, with improvement of 6.65% in mAP and 3.8% in Rank-1 accuracy.