Pith. sign in

REVIEW 5 major objections 6 minor 64 references

Transformers for Secure Hardware Systems: Applications, Challenges, and Outlook

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This survey argues that Transformer models, through self-attention over temporal and multi-modal hardware data, are becoming a general-purpose foundation for hardware security detection, from hardware Trojans and side-channel attacks to…

desk verdict A useful but overreaching survey: the catalog is solid, the 'significant shift' thesis is not established by the evidence it marshals. read the letter →

arxiv 2505.22605 v1 pith:EMPTTYQB submitted 2025-05-28 cs.CR

Banafsheh Saber Latibari , Najmeh Nazari , Avesta Sasan , Houman Homayoun , Pratik Satam , Soheil Salehi , Hossein Sayadi This is my paper
classification cs.CR
keywords hardwaresecurityTransformerself-attentionside-channelanalysisTrojandetectionmalwaredevicefingerprintingfirmware
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is a survey of recent attempts to apply Transformer models to hardware security. Its thesis is that the same self-attention mechanism that made Transformers dominant in natural language processing also suits the low-level, temporal, and multi-modal data produced by hardware—power traces, netlists, register-transfer-level code, execution telemetry, and RF signals—so attention-based detectors can catch threats that rule-based and conventional machine-learning methods miss. The survey organizes evidence across ten areas, including hardware Trojan detection, side-channel analysis, malware detection, device fingerprinting, firmware vulnerability search, and information-flow tracking, and it reports that Transformer-based systems set new accuracy marks in many of these tasks. The authors also identify five obstacles—computational cost, lack of explainability, adversarial vulnerability, scarce datasets, and adaptation to structured hardware telemetry—that must be solved before the approach becomes deployable. A sympathetic reader would take the paper's central claim to be that Transformers are a genuine shift in hardware-level threat detection, not just a borrowed tool.

What carries the argument

The load-bearing mechanism is the Transformer's multi-head self-attention, computed as $\operatorname{SoftMax}(QK^T/\sqrt{d_k})V$. It lets each element of an input sequence attend to every other element, so long-range and non-local dependencies in traces, netlists, and logs can be modeled without recurrence or convolutions. In the surveyed papers this single mechanism does different jobs: it correlates distant points of interest in desynchronized power traces, relates nodes in graph-transformer Trojan detectors, lets vision transformers see malicious byte patterns when binaries are rendered as images, and fuses heterogeneous side-channel signals into one representation. Efficient variants—relative positional encoding, Gaussian attention, graph message passing, and lightweight architectures—are introduced to keep the quadratic cost of attention manageable.

What would settle it

Run a controlled comparison on fixed hardware-security benchmarks—ASCAD power traces for side-channel analysis, TrustHub netlists for Trojan detection, and hardware performance counter traces for malware detection—matching model size, training data, and compute budget; if Transformers do not show consistent gains over tuned CNNs, LSTMs, and gradient-boosted trees, the survey's central premise fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that Transformer models represent a significant shift in the design of intelligent, context-aware threat detection mechanisms at the hardware level. It argues that because hardware security data—power side-channel traces, gate-level netlists, RTL descriptions, firmware binaries, process resource metrics, and RF emissions—can be encoded as sequences or token-like inputs, multi-head self-attention can capture dependencies that elude traditional methods. Across the surveyed literature, Transformer-based systems report strong results: HTrans reaches 96.7% F1 for hardware Trojan detection and 91.7% localization accuracy; TrojanFormer averages 97.66% F1 on small and medium netlists; a no-fine-tuning LLM study reports perfect Trojan detection in the unperturbed case; EstraNet handles side-channel traces of several thousand points with linear complexity; SecurityBERT reaches 98.2% accuracy on 14 IoT attack types; and masked-autoencoder Transformers outperform CNN-based RF fingerprinting. The survey reads these results as evidence that attention is broadly applicable across the hardware security lifecycle, from design-time vulnerability classification to runtime malware detection, and it frames the open challenges as engineering problems rather than fundamental limitations.

Load-bearing premise

The load-bearing premise is that encoding hardware data as sequences and applying self-attention genuinely outperforms well-tuned non-attention methods across the surveyed tasks—a comparison the cited papers do not make uniformly.

Editorial extensions

If this is right

  • Hardware Trojan detection will shift from hand-crafted netlist features toward tokenized RTL, graph transformers, and pretrained language models, with several reported gains of multiple points in precision and F1 over prior baselines.
  • Side-channel analysis will increasingly process long, desynchronized power traces directly, with shift-invariant attention replacing trace-alignment preprocessing and supporting both masking and jitter countermeasures.
  • Malware and firmware analysis will treat binaries, system calls, and firmware functions as text or images, reusing vision and language transformers rather than designing task-specific feature extractors.
  • Practical deployment on embedded and edge devices will depend on efficient attention variants, pruning, quantization, and offloading to edge servers; without those, the quadratic cost of self-attention blocks real-time hardware-level monitoring.
  • Hardware-assisted malware detection will move toward fusing hardware performance counters and other on-chip telemetry into multi-modal attention models aimed at zero-day and stealthy attacks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the survey's own claims, an apples-to-apples benchmark with identical datasets, compute budgets, and hyperparameter tuning across all five threat areas would be the natural next step; the cited results come from heterogeneous benchmarks and may not preserve their rankings under controlled comparison.
  • If self-attention's advantage is mainly long-range dependency modeling, then tasks with short sequences—small netlists or brief power traces—should show smaller gains over non-attention baselines; this is a testable prediction the paper does not make.
  • The repeated success of pretrained language models on RTL and netlists suggests that domain-specific pretraining corpora, rather than new architectures, may be the highest-leverage research direction for hardware security.
  • The survey's 'significant shift' claim is an interpretation of published positive results; it would be strengthened by negative results or failure cases, which the surveyed literature rarely reports.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. This survey reviews applications of Transformer and attention-based models in hardware security. It covers hardware Trojan detection, packaging/counterfeit detection, side-channel analysis, malware detection, device fingerprinting, remote attestation, insider-threat detection, hardware-software interface attacks, information flow tracking, fault detection, and vulnerability classification. The paper provides background on the Transformer architecture, a taxonomy of attention-based methods in Table 1, qualitative descriptions of recent works (mostly 2021-2025), and a discussion of challenges such as computational cost, explainability, adversarial robustness, data scarcity, and real-time hardware-level detection. The central claim is that Transformers 'represent a significant shift in the design of intelligent, context-aware threat detection mechanisms at the hardware level' and outperform prior methods in many reported cases.

Significance. If the central claim were fully supported, this would be a useful compact entry point to a fast-moving area, bringing together recent work across many hardware security subfields and identifying cross-cutting challenges. The paper's strengths are its breadth, an accessible description of the Transformer mechanism, and attention to practical deployment issues such as efficiency, interpretability, adversarial robustness, and dataset availability. These qualities make the survey a convenient bibliography and framing document. However, the comparative claim of Transformer superiority is not established by the evidence presented, and the scope of what counts as a 'Transformer' is inconsistently applied. The contribution is therefore currently a descriptive catalog rather than a validated assessment of the field.

major comments (5)
  1. [Section 5, Section 1, and Abstract] The central claim that Transformers 'represent a significant shift' in hardware threat detection is not supported by controlled evidence in the manuscript. The survey reports accuracy and F1 numbers from heterogeneous papers that use different datasets, preprocessing, and evaluation protocols, and none of the cited studies isolates the self-attention mechanism from auxiliary components. For example, HTrans uses GCN preprocessing (Section 3.1), TA-MobileViT and SDDM combine convolutional branches (Sections 3.1 and 3.2), TrojanWhisper and NtNDet rely on pretrained LLM/NLP priors (Section 3.1), and SHERLOCK, ViT4Mal, and DDoSViT convert binaries or attack flows into images (Sections 3.4 and 3.8). Reported gains could therefore be driven by these other components rather than by self-attention. I recommend softening the conclusion to describe Transformers as a promising and emerging direction, and explicitly stating that cross-paper comparisons do not establish superiority.
  2. [Section 3 and Table 1] The scope of the survey is internally inconsistent. Although the title and text focus on Transformers, Table 1 and several subsections include attention-augmented LSTMs/GRUs, Graph Attention Networks, and general LLM-based systems under the 'attention-based' or 'Transformer' umbrella. For instance, the insider-threat work in Section 3.7 uses stacked LSTM and GRU attention models, the Smart Monitor in Section 3.10 is not described as a Transformer at all, and Table 1 lists 'Graph Attention Networks (GATs)' as an attention usage. This conflation weakens the causal claim that Transformers specifically drive the reported results. The authors should define explicit inclusion criteria and classify each surveyed method as a Transformer proper, an attention-augmented non-Transformer model, or an LLM-based system.
  3. [Section 3 (general)] The survey is described as 'comprehensive' but does not report a systematic selection methodology, such as search databases, keywords, inclusion/exclusion criteria, or quality assessment. The choice of works appears ad hoc, and several subsections contain only one or two short descriptions. For a survey claiming comprehensiveness, this is a load-bearing limitation because readers cannot assess completeness or potential selection bias. I recommend either adding a short methodology paragraph that explains the selection process, or rephrasing the contribution as a 'representative review' and removing the comprehensiveness claim.
  4. [Sections 3.1-3.8] The quantitative results are presented without comparability constraints, making them misleading if read as evidence of relative performance. For example, HTrans reports 96.7% F1 on TrustHub, NtNDet reports a 3.17% F1 improvement over prior work, and TrojanWhisper reports 100% precision/recall on curated data. These numbers come from different benchmarks, different Trojan types, and different evaluation protocols, so they cannot be compared or aggregated. The survey should either build a comparative table that clearly states dataset, Trojan/attack type, baseline, and metric for each work, or explicitly state that no cross-paper quantitative comparison is possible.
  5. [Section 4.5] The discussion of Transformers for hardware-assisted malware detection (HMD) is forward-looking but repeatedly asserts opportunity without evidence, stating that self-attention 'enables context-aware modeling' and that Transformers are 'well-suited for fusing heterogeneous side-channel signals.' While such possibilities are reasonable, no surveyed HMD study actually demonstrates these advantages on shared benchmarks. This paragraph should be framed as conjecture and future direction rather than as an established benefit, to be consistent with the evidence level of the rest of the survey.
minor comments (6)
  1. [Section 2, Eqs. (1)-(2)] The notation in Eqs. (1) and (2) is confusing: Q_i, K_i, and V_i are not clearly defined as per-head projections, and d_k in Eq. (2) is never defined. Please define d_k = d/h or align the notation with the projections described in the text.
  2. [Section 3.1, paragraph on ref. [34]] The phrase 'Authors in this paper propose' is ambiguous; it should be 'In [34], the authors propose' to avoid confusion with the current manuscript.
  3. [Table 1, Side-channel Attack Detection row] The phrase 'attention-modal on power traces' appears to be a typo for 'attention model on power traces'; please correct it.
  4. [Section 3.5, first sentence] The phrase 'Similar to machine learning fingerprinting [36]' is misleading because ref. [36] (LLM-FIN) is about fingerprinting machine learning models, not device fingerprinting. Please use a citation that directly supports device fingerprinting or rephrase the sentence.
  5. [Section 3.3, Berreby and Sauvage paragraph] The cited work [2] is described as part of 'mitigation of side channel attacks,' but the cited title suggests it studies efficient deep learning architectures for side-channel attacks on AES. Please clarify whether this work uses a Transformer architecture and what 'mitigation' means in this context.
  6. [Section 4.3, He et al. paragraph] The sentence 'He et al. [17] has demonstrated' should be 'He et al. [17] have demonstrated.' Additionally, the DRL-based defense described here is not Transformer-specific; please explain its relevance to the Transformer discussion.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: this survey contains no derivation or fitted prediction of its own, and its central claim rests on externally published primary results; the numerous self-citations are background support and are not load-bearing.

full rationale

This is an aggregative survey, not a derivation chain: it introduces no equations of its own beyond the standard attention formulas (Eqs. 1-2, attributed to Vaswani et al. [55]), fits no parameters, and makes no predictions that could reduce to its inputs by construction. The central thesis, stated in Section 5 as 'Transformers represent a significant shift in the design of intelligent, context-aware threat detection mechanisms at the hardware level,' is supported by a catalog of externally published results with specific numbers on named benchmarks: HTrans (96.7% F1 on TrustHub, [25]), TrojanFormer (97.66% F1, [4]), NtNDet (benchmark gains on Trust-Hub/TRIT, [21]), SDDM (98.6%, [57]), SecurityBERT (98.2% on Edge-IIoTset, [11]), DDoSViT (99.50%, [1]), TrojanWhisper (100% precision/recall, [10]), and TransNet/EstraNet on ASCAD ([16, 15]) - none of these are author works. Self-citations are indeed numerous (refs 14, 20, 22, 23, 26, 31, 36-40, 45-50 are authored or co-authored by the present authors, and Section 1's premise 'Transformers leverage self-attention mechanisms to capture intricate relationships within sequential data [22, 23, 45]' leans on the authors' own prior survey [23]). However, none of these is load-bearing: removing every self-citation leaves the thesis supported by the external primary results enumerated above, and the attention mechanism itself is independently attributed to Vaswani et al. [55]. No uniqueness theorem is imported from the authors' prior work, no alternative approach is dismissed by citation, and no empirical result from an author paper is renamed as a survey-level finding. The skeptical concern - that reported gains do not isolate self-attention from auxiliary components (GCN, CNN branches, LLM priors) or compare against non-attention baselines on shared benchmarks - is an evidential and correctness limitation of the surveyed literature, not a circularity of this paper per rule 5. The score of 2 reflects only the density of self-citation as a flag for readers, not an identified circular step.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The paper introduces no free parameters or invented entities. Its central premise depends on two domain assumptions: that hardware security datasets are well-suited to sequence modeling with attention, and that the reported accuracies in the cited papers are accurate and comparable.

assumptions (2)
  • domain assumption Self-attention can effectively model long-range dependencies and multi-modal patterns in sequential hardware data including power traces, netlists, and execution traces.
    The whole premise of applying Transformers to hardware security rests on this representational assumption. It is asserted in Sections 1 and 2 and never tested against non-attention baselines across the surveyed tasks.
  • domain assumption The accuracy and performance numbers reported in the surveyed papers are reliable and comparable despite using different datasets and metrics.
    The survey's conclusion that Transformers outperform traditional methods is built on headline numbers from the cited papers. These numbers are not independently re-evaluated or placed on a common benchmark.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transformers for Secure Hardware Systems: Applications, Challenges, and Outlook." pith.science (2026). https://pith.science/paper/EMPTTYQB

@misc{pith2026250522605,
  author       = {Pith},
  title        = {Pith review of: Transformers for Secure Hardware Systems: Applications, Challenges, and Outlook},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EMPTTYQB}},
  note         = {Machine review of arXiv:2505.22605}
}
read the original abstract

The rise of hardware-level security threats, such as side-channel attacks, hardware Trojans, and firmware vulnerabilities, demands advanced detection mechanisms that are more intelligent and adaptive. Traditional methods often fall short in addressing the complexity and evasiveness of modern attacks, driving increased interest in machine learning-based solutions. Among these, Transformer models, widely recognized for their success in natural language processing and computer vision, have gained traction in the security domain due to their ability to model complex dependencies, offering enhanced capabilities in identifying vulnerabilities, detecting anomalies, and reinforcing system integrity. This survey provides a comprehensive review of recent advancements on the use of Transformers in hardware security, examining their application across key areas such as side-channel analysis, hardware Trojan detection, vulnerability classification, device fingerprinting, and firmware security. Furthermore, we discuss the practical challenges of applying Transformers to secure hardware systems, and highlight opportunities and future research directions that position them as a foundation for next-generation hardware-assisted security. These insights pave the way for deeper integration of AI-driven techniques into hardware security frameworks, enabling more resilient and intelligent defenses.

Figures

Figures reproduced from arXiv: 2505.22605 by the authors.

Figure 1
Figure 1. Scenarios where Transformers detect, classify, and mitigate vulnerabilities. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Architecture of the Vanilla Transformer, featuring (1) an Encoder-Decoder structure, (2) Multi-Head Self-Attention [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 60 canonical work pages

  1. [1]

    Muhammad Ali, Yasir Saleem, Sadaf Hina, and Ghalib A Shah. 2025. DDoSViT: IoT DDoS attack detection for fortifying firmware Over-The-Air (OTA) updates using vision transformer.Internet of Things(2025), 101527

    work page 2025

  2. [2]

    Yohaï-Eliel Berreby and Laurent Sauvage. 2023. Investigating efficient deep learn- ing architectures for side-channel attacks on AES.arXiv preprint arXiv:2309.13170 (2023)

    arXiv 2023

  3. [3]

    Chaitanya Bhure, Dhruvakumar Aklekar, Wenjie Che, and Fareena Saqib. 2024. Vision Transformers for Counterfeit IC Detection. In2024 IEEE Physical Assurance and Inspection of Electronics (PAINE). 1–7. doi:10.1109/PAINE62042.2024.10792793

    arXiv 2024

  4. [4]

    Menghui Chen, Xiaoyong Kou, and Gongxuan Zhang. 2024. TrojanFormer: Resource-Efficient Hardware Trojan Detection Using Graph Transformer Net- work. In2024 7th International Conference on Electronics Technology (ICET). IEEE, 165–170

    work page 2024

  5. [5]

    Shouhong Chen, Guanxiang Qin, Ying Lu, Tao Wang, and Xingna Hou. 2025. A lightweight Hardware Trojan detection approach in the waveform diagram based on MobileViT and attention mechanism.The Journal of Supercomputing81 (2025), 580

    work page 2025

  6. [6]

    Xing Chen, Yonglei Wu, Xingyou He, and Wuyi Ming. 2023. A comprehensive review of deep learning-based PCB defect detection.IEEE Access11 (2023), 139017–139038

    work page 2023

  7. [7]

    John Demme, Matthew Maycock, Jared Schmitz, Adrian Tang, Adam Waksman, Simha Sethumadhavan, and Salvatore Stolfo. 2013. On the feasibility of on- line malware detection with performance counters.ACM SIGARCH computer architecture news41, 3 (2013), 559–570

    work page 2013

  8. [8]

    Xiaoheng Deng, Zhe Wang, Xinjun Pei, and Kaiping Xue. 2023. TransMalDE: an effective transformer based hierarchical framework for IoT malware detection. IEEE Transactions on Network Science and Engineering11, 1 (2023), 140–151

    work page 2023

Show all 64 references
  1. [9]

    Sai Manoj Pudukotai Dinakarrao et al. 2019. Adversarial attack on microarchitec- tural events based malware detectors. InProceedings of the 56th Annual Design Automation Conference 2019. 1–6

    2019

  2. [10]

    Md Omar Faruque, Peter Jamieson, Ahmad Patooghy, and Abdel-Hameed A Badawy. 2024. TrojanWhisper: Evaluating Pre-trained LLMs to Detect and Localize Hardware Trojans.arXiv preprint arXiv:2412.07636(2024)

    2024 arXiv

  3. [11]

    Mohamed Amine Ferrag et al . 2024. Revolutionizing Cyber Threat Detection With Large Language Models: A Privacy-Preserving BERT-Based Lightweight Model for IoT/IIoT Devices.IEEE Access12 (2024), 23733–23750. doi:10.1109/ ACCESS.2024.3363469

    2024

  4. [12]

    Michael Fu, Van Nguyen, Chakkrit Kla Tantithamthavorn, Trung Le, and Dinh Phung. 2023. Vulexplainer: A transformer-based hierarchical distillation for explaining vulnerability types.IEEE Transactions on Software Engineering49, 10 (2023), 4550–4565

    2023

  5. [13]

    Yifeng Gao, Hosein Mohammadi Makrani, Mehrdad Aliasgari, Amin Rezaei, Jessica Lin, Houman Homayoun, and Hossein Sayadi. 2021. Adaptive-hmd: Accurate and cost-efficient machine learning-driven malware detection using microarchitectural events. In2021 IEEE 27th International Sym...

    2021

  6. [14]

    Kevin Immanuel Gubbi, Banafsheh Saber Latibari, Anirudh Srikanth, Tyler Sheaves, Sayed Arash Beheshti-Shirazi, Sai Manoj PD, Satareh Rafatirad, Avesta Sasan, Houman Homayoun, and Soheil Salehi. 2023. Hardware trojan detection using machine learning: A tutorial.ACM Transactions...

    2023

  7. [15]

    Suvadeep Hajra et al. 2024. Estranet: An efficient shift-invariant transformer network for side-channel analysis.IACR Transactions on Cryptographic Hardware and Embedded Systems2024, 1 (2024), 336–374

    2024

  8. [16]

    Suvadeep Hajra, Sayandeep Saha, Manaar Alam, and Debdeep Mukhopadhyay

  9. [17]

    Zhangying He, Houman Homayoun, and Hossein Sayadi. 2024. Beyond conven- tional defenses: Proactive and adversarial-resilient hardware malware detection using deep reinforcement learning. InProceedings of the 61st ACM/IEEE Design Automation Conference. 1–6

    2024

  10. [18]

    Zhangying He, Houman Homayoun, and Hossein Sayadi. 2024. Guarding against the unknown: Deep transfer learning for hardware image-based malware detec- tion.Journal of Hardware and Systems Security8, 2 (2024), 61–78

    2024

  11. [19]

    Hossein Jafari, Oluwaseyi Omotere, Damilola Adesina, Hsiang-Huang Wu, and Lijun Qian. 2018. IoT Devices Fingerprinting Using Deep Learning. InMILCOM 2018 - 2018 IEEE Military Communications Conference (MILCOM). 1–9. doi:10. 1109/MILCOM.2018.8599826

    2018

  12. [20]

    Ali Karkehabadi, Banafsheh Saber Latibari, Houman Homayoun, and Avesta Sasan. 2024. HLGM: A novel methodology for improving model accuracy us- ing saliency-guided high and low gradient masking. In2024 14th International Conference on Information Science and Technology (ICIST)....

    2024

  13. [21]

    Shijie Kuang, Zhe Quan, Guoqi Xie, Xiaomin Cai, Xiaoqian Chen, and Keqin Li

  14. [22]

    Banafsheh Saber Latibari, Houman Homayoun, and Avesta Sasan. 2025. Optimiz- ing Vision Transformers: Unveiling’Focus and Forget’for Enhanced Computa- tional Efficiency.IEEE Access(2025)

    2025

  15. [23]

    Banafsheh Saber Latibari, Najmeh Nazari, Muhtasim Alam Chowdhury, Kevin Im- manuel Gubbi, Chongzhou Fang, Sujan Ghimire, Elahe Hosseini, Hossein Sayadi, Houman Homayoun, Soheil Salehi, et al. 2024. Transformers: A Security Per- spective.IEEE Access(2024)

    2024

  16. [24]

    Woongsup Lee, Seon Yeob Baek, and Seong Hwan Kim. 2021. Deep-Learning- Aided RF Fingerprinting for NFC Security.IEEE Communications Magazine59, 5 (2021), 96–101. doi:10.1109/MCOM.001.2000912

    2021 doi

  17. [25]

    Yilin Li, Shan Li, and Haihua Shen. 2023. Htrans: Transformer-based method for hardware trojan detection and localization. In2023 IEEE 32nd Asian Test Symposium (ATS). IEEE, 1–6

    2023

  18. [26]

    Yu-Zheng Lin, Muntasir Mamun, Muhtasim Alam Chowdhury, Shuyu Cai, Mingyu Zhu, Banafsheh Saber Latibari, Kevin Immanuel Gubbi, Najmeh Nazari Bavarsad, Arjun Caputo, Avesta Sasan, et al . 2023. Hw-v2w-map: Hardware vulnerability to weakness mapping framework for root cause analy...

    2023 arXiv

  19. [27]

    Fei Liu, Heng Wang, Pingfa Feng, and Long Zeng. 2024. Integrated Circuit Pack- aging Defect Analysis and Deep Learning Detection Method.IEEE Transactions on Components, Packaging and Manufacturing Technology14, 9 (2024), 1707–1719. doi:10.1109/TCPMT.2024.3447040

    2024

  20. [28]

    Hosein Mohammadi Makrani et al. 2021. Security threats in cloud rooted from machine learning-based resource provisioning systems. InSilicon Valley Cyberse- curity Conference. Springer, 22–32

    2021

  21. [29]

    Mounesh Marali et al. 2024. A hybrid transformer-based BERT and LSTM ap- proach for vulnerability classification problems.International Journal of Mathe- matics in Operational Research28, 3 (2024), 275–295

    2024

  22. [30]

    Nowfel Mashnoor, Mohammad Akyash, Hadi Kamali, and Kimia Azar. 2025. LLM-IFT: LLM-Powered Information Flow Tracking for Secure Hardware.arXiv preprint arXiv:2504.07015(2025)

    2025 arXiv

  23. [31]

    Ali Mirzaeian, Zhi Tian, Sai Manoj PD, Banafsheh S Latibari, Ioannis Savidis, Houman Homayoun, and Avesta Sasan. 2022. Adaptive-Gravity: A Defense Against Adversarial Samples. In2022 23rd International Symposium on Quality Electronic Design (ISQED). IEEE, 96–101

    2022

  24. [32]

    2022.Expansion of Conforming to Interface Struc- tured Adaptive Mesh Refinement Algorithm to Higher Order Elements and Crack Propagation

    Mohamad Mohamadsalehi. 2022.Expansion of Conforming to Interface Struc- tured Adaptive Mesh Refinement Algorithm to Higher Order Elements and Crack Propagation. The Ohio State University

    2022

  25. [33]

    Mohamad Mohmadsalehi and Soheil Soghrati. 2022. An automated mesh gen- eration algorithm for simulating complex crack growth problems.Computer Methods in Applied Mechanics and Engineering398 (2022), 115015

    2022

  26. [34]

    Abdurrahman Nasr, Khalil Mohamed, Mohamed Zaki, et al . 2024. Improving Hardware Trojan Detection with Transformer-Based Power Analysis. (2024)

    2024

  27. [35]

    Dimosthenis Natsos and Andreas L Symeonidis. 2025. Transformer-based mal- ware detection using process resource utilization metrics.Results in Engineering (2025), 104250

    2025

  28. [36]

    Najmeh Nazari et al. 2024. LLM-FIN: Large Language Models Fingerprinting At- tack on Edge Devices. In2024 25th International Symposium on Quality Electronic Design (ISQED). IEEE, 1–6

    2024

  29. [37]

    Najmeh Nazari et al. 2024. Securing On-Chip Learning: Navigating Vulnerabilities and Potential Safeguards in Spiking Neural Network Architectures. In2024 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 1–5

    2024

  30. [38]

    Najmeh Nazari et al . 2024. SpecScope: Automating Discovery of Exploitable Spectre Gadgets on Black-Box Microarchitectures. In2024 Design, Automation & Test in Europe Conference & Exhibition (DATE). IEEE, 1–6

    2024

  31. [39]

    Najmeh Nazari, Chongzhou Fang, Hosein Mohammadi Makrani, Behnam Omidi, Mahdi Eslamimehr, Setareh Rafatirad, Avesta Sasan, Hossein Sayadi, Khaled N Khasawneh, and Houman Homayoun. 2024. Architectural Whispers: Robust Ma- chine Learning Models Fingerprinting via Frequency Thrott...

    2024

  32. [40]

    Najmeh Nazari, Hosein Mohammadi Makrani, Chongzhou Fang, Hossein Sayadi, Setareh Rafatirad, Khaled N Khasawneh, and Houman Homayoun. 2024. Forget and Rewire: Enhancing the Resilience of Transformer-based Models against Bit-Flip Attacks. In33rd USENIX Security Symposium. 1349–1366

    2024

  33. [41]

    Preetam Pal, Pratik Chattopadhyay, and Mayank Swarnkar. 2023. Temporal feature aggregation with attention for insider threat detection from activity logs. Expert Systems with Applications224 (2023), 119925

    2023

  34. [42]

    Zhixin Pan, Jennifer Sheldon, and Prabhat Mishra. 2022. Hardware-assisted malware detection and localization using explainable machine learning.IEEE Trans. Comput.71, 12 (2022), 3308–3321

    2022

  35. [43]

    Gavin Parpart, Jonathan H Tu, Bradley Clymer, Jung Lee, and Jasen Babcock. 2024. Transformer Masked Autoencoders for RF Device Fingerprinting. InMILCOM 2024-2024 IEEE Military Communications Conference (MILCOM). IEEE, 859–862

    2024

  36. [44]

    Akshara Ravi, Vivek Chaturvedi, and Muhammad Shafique. 2023. Vit4mal: Light- weight vision transformer for malware detection on edge devices.ACM Transac- tions on Embedded Computing Systems22, 5s (2023), 1–26

    2023

  37. [45]

    Banafsheh Saber Latibari, Soheil Salehi, Houman Homayoun, and Avesta Sasan

  38. [46]

    Hossein Sayadi, Mehrdad Aliasgari, Furkan Aydin, Seetal Potluri, Aydin Aysu, Jack Edmonds, and Sara Tehranipoor. 2022. Towards ai-enabled hardware security: Challenges and opportunities. In2022 IEEE 28th International Symposium on On-Line Testing and Robust System Design (IOLT...

    2022

  39. [47]

    Hossein Sayadi, Yifeng Gao, Hosein Mohammadi Makrani, Tinoosh Mohsenin, Avesta Sasan, Setareh Rafatirad, Jessica Lin, and Houman Homayoun. 2020. Stealthminer: Specialized time series machine learning for run-time stealthy malware detection based on microarchitectural features....

    2020

  40. [48]

    Hossein Sayadi, Zhangying He, Hosein Mohammadi Makrani, and Houman Homayoun. 2024. Intelligent malware detection based on hardware performance counters: A comprehensive survey. In2024 25th International Symposium on Quality Electronic Design (ISQED). IEEE, 1–10

    2024

  41. [49]

    Hossein Sayadi, Hosein Mohammadi Makrani, Sai Manoj Pudukotai Dinakarrao, Tinoosh Mohsenin, Avesta Sasan, Setareh Rafatirad, and Houman Homayoun

  42. [50]

    Hossein Sayadi, Nisarg Patel, Avesta Sasan, Setareh Rafatirad, and Houman Homayoun. 2018. Ensemble learning for effective run-time hardware-based malware detection: A comprehensive analysis and classification. InProceedings of the 55th Annual Design Automation Conference. 1–6

    2018

  43. [51]

    Sachith Seneviratne, Ridwan Shariffdeen, Sanka Rasnayaka, and Nuran Kasthuri- arachchi. 2022. Self-supervised vision transformers for malware detection.IEEE Access10 (2022), 103121–103135

    2022

  44. [52]

    Guanxiong Shen, Junqing Zhang, Alan Marshall, Mikko Valkama, and Joseph Cavallaro. 2021. Radio frequency fingerprint identification for security in low-cost IoT devices. In2021 55th Asilomar conference on signals, systems, and computers. IEEE, 309–313

    2021

  45. [53]

    Ritu-Ranjan Shrivastwa, Sylvain Guilley, and Jean-Luc Danger. 2021. Multi- source fault injection detection using machine learning and sensor fusion. In International Conference on Security and Privacy. Springer, 93–107

    2021

  46. [54]

    Amisha Srivastava et al. 2024. SCAR: Power Side-Channel Analysis at RTL Level. IEEE Transactions on Very Large Scale Integration (VLSI) Systems(2024)

    2024

  47. [55]

    Ashish Vaswani et al . 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

    2017

  48. [56]

    Han Wang, Hossein Sayadi, Sai Manoj Pudukotai Dinakarrao, Avesta Sasan, Setareh Rafatirad, and Houman Homayoun. 2021. Enabling micro AI for securing edge devices at hardware level.IEEE Journal on Emerging and Selected Topics in Circuits and Systems11, 4 (2021), 803–815

    2021

  49. [57]

    Xiaobin Wang, Shuang Gao, Jianlan Guo, Chu Wang, Liping Xiong, and Yuntao Zou. 2024. Deep learning-based integrated circuit surface defect detection: Ad- dressing information density imbalance for industrial application.International Journal of Computational Intelligence Syste...

    2024

  50. [58]

    Qingyang Wu, Carlos Feres, Daniel Kuzmenko, Ding Zhi, Zhou Yu, Xin Liu, and Xiaoguang ‘Leo’Liu. 2018. Deep learning based RF fingerprinting for device identification and wireless security.Electronics Letters54, 24 (2018), 1405–1407

    2018

  51. [59]

    Junjian Ye et al. 2024. Detecting command injection vulnerabilities in Linux- based embedded firmware with LLM-based taint analysis of library functions. Computers & Security144 (2024), 103971

    2024

  52. [60]

    Yingchao Yu, Shuitao Gan, and Xiaojun Qin. 2021. firm VulSeeker: BERT and Siamese based Vulnerability for Embedded Device Firmware Images. In2021 IEEE Symposium on Computers and Communications (ISCC). 1–7

    2021

  53. [2019]

    In2019 Design, Automation & Test in Europe Conference & Exhibition (DATE)

    2smart: A two-stage machine learning-based approach for run-time spe- cialized hardware-assisted malware detection. In2019 Design, Automation & Test in Europe Conference & Exhibition (DATE). IEEE, 728–733

  54. [2022]

    In International Conference on Cryptology in Africa

    Transnet: Shift invariant transformer network for side channel analysis. In International Conference on Cryptology in Africa. Springer, 371–396

  55. [2024]

    InProceedings of the Great Lakes Symposium on VLSI 2024

    IRET: Incremental Resolution Enhancing Transformer. InProceedings of the Great Lakes Symposium on VLSI 2024. 620–625

    2024

  56. [2025]

    Expert Systems with Applications(2025), 126666

    NtNDet: Hardware Trojan detection based on pre-trained language models. Expert Systems with Applications(2025), 126666

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.