pith. sign in

arxiv: 2606.04266 · v1 · pith:6D4V72HVnew · submitted 2026-06-02 · 💻 cs.CR · cs.LG

Long-Term and Short-Term Transistor Aging in Deep Neural Networks: Impact and Mitigation

Alireza Sarmadi , Virinchi Roy Surabhi , Prashanth Krishnamurthy , Hussam Amrouch , Ramesh Karri , Farshad Khorrami This is my paper

Pith reviewed 2026-06-28 09:05 UTC · model grok-4.3

classification 💻 cs.CR cs.LG
keywords transistor agingdeep neural networksinference accuracytiming guardbandsretrainingtiming violationshardware Trojansreliability
0
0 comments X

The pith

Aging-aware retraining produces resilient DNNs that maintain accuracy with reduced timing guardbands despite transistor aging.

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

Transistor aging slows switching speeds in integrated circuits and produces timing violations that degrade DNN inference accuracy. Large guardbands prevent these violations but reduce performance. The paper presents a retraining method that accounts for aging effects during model training, yielding networks that remain accurate even when smaller guardbands are applied. The approach is evaluated on a hardware DNN for image classification. Short-term aging is also noted as a possible excitation method for hardware Trojan detection.

Core claim

An aging-aware retraining methodology generates DNNs resilient to long-term and short-term transistor aging, allowing aggressive guardbands while preserving inference accuracy in the presence of aging-induced degradation.

What carries the argument

Aging-aware retraining, which incorporates modeled aging-induced timing changes into the training process to compensate for slowdowns without hardware changes.

If this is right

  • DNNs can use smaller guardbands and therefore higher clock speeds or throughput while meeting lifetime reliability targets.
  • Inference accuracy remains stable as transistors age without needing larger initial guardbands.
  • No circuit redesign or extra hardware is required to counteract aging effects.
  • Short-term aging stress can be applied as a controlled excitation for hardware Trojan detection.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could be combined with other reliability techniques such as voltage scaling to further reduce power.
  • Results on image classification suggest the approach may apply to other DNN tasks like object detection or speech recognition.
  • Extending the aging models to include process variation or temperature would test broader applicability.

Load-bearing premise

The effects of transistor aging on circuit timing can be sufficiently modeled and compensated via retraining without introducing new failure modes or requiring hardware modifications.

What would settle it

Measure accuracy of an aging-aware retrained DNN versus a standard DNN on the same hardware under accelerated aging with reduced guardbands; the claim fails if accuracy degradation is not reduced.

Figures

Figures reproduced from arXiv: 2606.04266 by Alireza Sarmadi, Farshad Khorrami, Hussam Amrouch, Prashanth Krishnamurthy, Ramesh Karri, Virinchi Roy Surabhi.

Figure 1
Figure 1. Figure 1: Neuron architecture. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Neural network architecture. (b) Layer architecture with [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the framework. The tool flow has three main steps. First, short-term aging [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Accuracy drop of the DNN due to transistor aging for 0.8V and 0.6V. The nominal clock [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Aging has a considerable effect on feature value statistics. The statistics for each layer’s [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Feature errors propagation through different layers for various aging levels. Each bar [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Normalized accuracy drops with increasing aging level for different clock periods. The [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Normalized accuracies for 0.8V and 0.6V when feature noise is applied during training. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Normalized accuracies for 0.8V and 0.6V when gradient noise is applied during training. [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Normalized accuracies for 0.8V and 0.6V when feature noise along with gradient noise [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Normalized accuracies for 0.8V and 0.6V when feature noise during training is applied. [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Propagation of feature errors across multiple layers at different aging levels in a robusti [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Overall pipeline [38] in the short-term aging-based Trojan detection methodology. [PITH_FULL_IMAGE:figures/full_fig_p020_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Overview [38] of short-term aging based Trojan detection approach: First, dynamic [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: (a, b) 0 → 1 (c, d) 1 → 0 bit flips over various clock periods and voltage levels for AES-T100. The color map indicates the number of inputs that have bit flips. 7 Conclusion The effects of long-term aging and short-term aging in circuits (due to transistor aging) were discussed and shown to cause noticeable inference accuracy degradation on DNNs implemented in hardware. It is noted that device aging, whi… view at source ↗
Figure 16
Figure 16. Figure 16: Weighted locations of 0 → 1 (a, b) 1 → 0 (c, d) bit flips over various clock periods and voltage levels for AES-T100. The color map indicates the location of bit flips. nanoscale transistors, can result in a 50% decline in classification accuracy for a representative DNN after 20% of aging once the DNN is deployed. Therefore, it is essential that hardware-implemented DNNs take into consideration the effec… view at source ↗
read the original abstract

Deep neural networks (DNNs) are used in a variety of real-world applications including, for example, image classification and speech recognition. The inference accuracy of DNN implemented on hardware in integrated circuits (ICs) degrades under phenomena such as transistor aging. Aging slows down the switching speed of transistors, resulting in system-level timing violations due to unsustainable clocks. To maintain reliability for the entire projected lifetime, designers add guardbands to prevent timing violations; however, adding large timing guardbands causes losses in performance (speed or throughput). This chapter provides a detailed discussion of the effects of long-term and short-term transistor aging on DNN inference accuracy. Furthermore, to mitigate aging effects on DNN's accuracy and keep them at bay, a methodology for aging-aware retraining is presented in order to generate a resilient DNN even when aggressive (i.e., smaller than required) guardbands are used. This improves the inference accuracy of the DNNs even in the presence of aging-induced degradation. These effects are discussed in this chapter along with mitigation strategies on a hardware implementation of a DNN for image classification on an off-the-shelf image dataset. The application of short-term aging as an excitation mechanism for the detection of hardware Trojans in integrated circuits is also briefly discussed.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper claims that long-term and short-term transistor aging degrades DNN inference accuracy via timing violations in hardware implementations. It proposes an aging-aware retraining methodology to generate resilient DNNs that maintain accuracy even when using aggressive (smaller-than-required) guardbands. The approach is illustrated via a hardware DNN for image classification on an off-the-shelf dataset, with a brief discussion of short-term aging as an excitation mechanism for hardware Trojan detection.

Significance. If validated, the aging-aware retraining technique could enable meaningful reductions in guardband overhead for DNN accelerators, improving performance and energy efficiency while preserving long-term reliability. This hardware-software co-design angle addresses a practical deployment barrier for real-world DNN systems.

major comments (1)
  1. [Abstract] Abstract: the central claim that aging-aware retraining improves inference accuracy under aging-induced degradation is presented without any quantitative results, error bars, baseline comparisons, or experimental details on the hardware test setup, aging models, or retraining procedure. This absence is load-bearing for the primary contribution.
minor comments (1)
  1. The manuscript refers to itself as 'this chapter,' which may indicate it is excerpted from a larger work; ensure the submission is self-contained with all necessary methodological and result details for a journal article.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback. The major comment highlights a valid issue with the abstract. We agree that strengthening the abstract with quantitative support will improve the manuscript and will revise it in the next version. Our point-by-point response follows.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that aging-aware retraining improves inference accuracy under aging-induced degradation is presented without any quantitative results, error bars, baseline comparisons, or experimental details on the hardware test setup, aging models, or retraining procedure. This absence is load-bearing for the primary contribution.

    Authors: We agree that the abstract would be strengthened by including key quantitative results. The full manuscript reports experiments on a hardware DNN implementation for image classification using an off-the-shelf dataset, with aging models for long-term and short-term effects, and details the aging-aware retraining procedure. We will revise the abstract to concisely incorporate representative quantitative findings (e.g., accuracy degradation percentages under aging and recovery achieved by the proposed method relative to baselines), along with brief references to the hardware setup and aging models, while respecting length limits. This addresses the load-bearing nature of the claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper presents an empirical methodology for aging-aware retraining of DNNs to mitigate transistor aging effects on inference accuracy, validated through hardware implementation on an image classification task. No derivation chain, equations, or load-bearing claims reduce to self-definitions, fitted inputs renamed as predictions, or self-citation chains. The approach relies on external aging models and experimental results rather than internal circular reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract contains no mathematical derivations, fitted parameters, axioms, or invented entities; it describes an empirical methodology and experimental discussion without introducing new formal elements.

pith-pipeline@v0.9.1-grok · 5778 in / 1109 out tokens · 29935 ms · 2026-06-28T09:05:52.346214+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references

  1. [1]

    Reliability-aware design to suppress aging

    Hussam Amrouch, Behnam Khaleghi, Andreas Gerstlauer, and J¨ org Henkel. Reliability-aware design to suppress aging. InProceedings of ACM/EDAC/IEEE Design Automation Confer- ence, pages 1–6, Austin,TX, June 2016. IEEE. 23 Table 1: Performance of the one-class SVM classifier for Trojan detection for ICs with 14nm FinFET technology. Each bin has 5 inputs. Bi...

    2016

  2. [2]

    Arechiga and Alan J

    Austin P. Arechiga and Alan J. Michaels. The effect of weight errors on neural networks. In Proceesings of IEEE Annual Computing and Communication Workshop and Conference, pages 190–196, Las Vegas, NV, 2018. IEEE

    2018

  3. [3]

    Arechiga and Alan J

    Austin P. Arechiga and Alan J. Michaels. The robustness of modern deep learning architectures against single event upset errors. InProceedings of IEEE High Performance extreme Computing Conference, pages 1–6, Waltham, MA, 2018. IEEE

    2018

  4. [4]

    Joshi, Matthew M

    Nandhini Chandramoorthy, Karthik Swaminathan, Martin Cochet, Arun Paidimarri, Schuyler Eldridge, Rajiv V. Joshi, Matthew M. Ziegler, Alper Buyuktosunoglu, and Pradip Bose. Re- silient low voltage accelerators for high energy efficiency. InProceedings of IEEE International Symposium on High Performance Computer Architecture, pages 147–158, Washington, DC,

  5. [5]

    Academic Press, Cambridge, MA, 2015

    Yogesh Singh Chauhan, Darsen Duane Lu, Sriramkumar Venugopalan, Sourabh Khandelwal, Juan Pablo Duarte, Navid Paydavosi, Ai Niknejad, and Chenming Hu.FinFET modeling for IC simulation and design: using the BSIM-CMG standard. Academic Press, Cambridge, MA, 2015

    2015

  6. [6]

    Emer, and Vivienne Sze

    Yu-Hsin Chen, Tushar Krishna, Joel S. Emer, and Vivienne Sze. Eyeriss: An energy-efficient reconfigurable accelerator for deep convolutional neural networks.IEEE Journal of Solid-State Circuits, 52(1):127–138, 2017

    2017

  7. [7]

    Sensitivity based error resilient techniques for energy efficient deep neural network accelerators

    Wonseok Choi, Dongyeob Shin, Jongsun Park, and Swaroop Ghosh. Sensitivity based error resilient techniques for energy efficient deep neural network accelerators. InProceedings of ACM/IEEE Design Automation Conference, pages 1–6, Las Vegas, NV, 2019. IEEE

    2019

  8. [8]

    Hardware-oriented approximation of convolutional neural networks, 2016

    Philipp Gysel, Mohammad Motamedi, and Soheil Ghiasi. Hardware-oriented approximation of convolutional neural networks, 2016

    2016

  9. [9]

    Training modern deep neural networks for memory-fault robustness

    Ghouthi Boukli Hacene, Fran¸ cois Leduc-Primeau, Amal Ben Soussia, Vincent Gripon, and Fran¸ cois Gagnon. Training modern deep neural networks for memory-fault robustness. In Proceedings of IEEE International Symposium on Circuits and Systems, pages 1–5, Sapporo, Japan, 2019. IEEE

    2019

  10. [10]

    Horowitz, and William J

    Song Han, Xingyu Liu, Huizi Mao, Jing Pu, Ardavan Pedram, Mark A. Horowitz, and William J. Dally. EIE: Efficient inference engine on compressed deep neural network. InPro- ceedings of IEEE International Symposium on Computer Architecture, page 243–254, Seoul, Republic of Korea, 2016. IEEE

    2016

  11. [11]

    Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding, 2015

    Song Han, Huizi Mao, and William J Dally. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding, 2015

    2015

  12. [12]

    Termi- nal brain damage: Exposing the graceless degradation in deep neural networks under hardware fault attacks

    Sanghyun Hong, Pietro Frigo, Yigitcan Kaya, Cristiano Giuffrida, and Tudor Dumitras. Termi- nal brain damage: Exposing the graceless degradation in deep neural networks under hardware fault attacks. InProceedings of USENIX Security Symposium, pages 497–514, Santa Clara, CA, 2019. USENIX Association

    2019

  13. [13]

    Calibrating process variation at system level with in-situ low-precision transfer learning for analog neural network processors

    Kaige Jia, Zheyu Liu, Qi Wei, Fei Qiao, Xinjun Liu, Yi Yang, Hua Fan, and Huazhong Yang. Calibrating process variation at system level with in-situ low-precision transfer learning for analog neural network processors. InProceedings of ACM/ESDA/IEEE Design Automation Conference, pages 1–6, San Francisco, CA, 2018. IEEE

    2018

  14. [14]

    Xun Jiao, Mulong Luo, Jeng-Hau Lin, and Rajesh K. Gupta. An assessment of vulnerability of hardware neural networks to dynamic voltage and temperature variations. InProceedings of IEEE/ACM International Conference on Computer-Aided Design, pages 945–950, Irvine, CA, 2017. IEEE

    2017

  15. [15]

    MATIC: Learning around errors for efficient low-voltage neural network accelerators

    Sung Kim, Patrick Howe, Thierry Moreau, Armin Alaghi, Luis Ceze, and Visvesh Sathe. MATIC: Learning around errors for efficient low-voltage neural network accelerators. InPro- ceedings of IEEE Design, Automation & Test in Europe Conference & Exhibition, pages 1–6, Dresden, Germany, 2018. IEEE

    2018

  16. [16]

    Understanding error propagation in deep learning neural network (DNN) accelerators and applications

    Guanpeng Li, Siva Kumar Sastry Hari, Michael Sullivan, Timothy Tsai, Karthik Pattabiraman, Joel Emer, and Stephen W Keckler. Understanding error propagation in deep learning neural network (DNN) accelerators and applications. InProceedings of IEEE International Conference 25 for High Performance Computing, Networking, Storage and Analysis, pages 1–12, Denver, CO,

  17. [17]

    Analysis of circuit aging on accuracy degradation of deep neural network accelerator

    Wenye Liu and Chip-Hong Chang. Analysis of circuit aging on accuracy degradation of deep neural network accelerator. InProceedings of IEEE International Symposium on Circuits and Systems, pages 1–5, Sapporo, Japan, 2019. IEEE

    2019

  18. [18]

    Mahapatra, N

    S. Mahapatra, N. Goel, S. Desai, S. Gupta, B. Jose, S. Mukhopadhyay, K. Joshi, A. Jain, A. E. Islam, and M. A. Alam. A comparative study of different physics-based NBTI models. IEEE Transactions on Electron Devices, 60(3):901–916, 2013

    2013

  19. [19]

    Springer, New Delhi, India, 2016

    Souvik Mahapatra.Fundamentals of bias temperature instability in mos transistors. Springer, New Delhi, India, 2016

    2016

  20. [20]

    A simulation study of NBTI impact on 14-nm node FinFET technology for logic applications: Device degradation to circuit-level interaction

    Subrat Mishra, Hussam Amrouch, Jerin Joe, Chetan K Dabhi, Karansingh Thakor, Yogesh S Chauhan, Joerg Henkel, and Souvik Mahapatra. A simulation study of NBTI impact on 14-nm node FinFET technology for logic applications: Device degradation to circuit-level interaction. IEEE Transactions on Electron Devices, 66(1):271–278, 2018

    2018

  21. [21]

    A 0.3–2.6 TOPS/W precision-scalable processor for real- time large-scale ConvNets

    Bert Moons and Marian Verhelst. A 0.3–2.6 TOPS/W precision-scalable processor for real- time large-scale ConvNets. InProceedings of IEEE Symposium on VLSI Circuits, pages 1–2, Honolulu, HI, 2016. IEEE

    2016

  22. [22]

    An energy-efficient precision-scalable convnet processor in 40-nm CMOS.IEEE Journal of Solid-State Circuits, 52(4):903–914, 2017

    Bert Moons and Marian Verhelst. An energy-efficient precision-scalable convnet processor in 40-nm CMOS.IEEE Journal of Solid-State Circuits, 52(4):903–914, 2017

    2017

  23. [23]

    Adding gradient noise improves learning for very deep networks, 2015

    Arvind Neelakantan, Luke Vilnis, Quoc V Le, Ilya Sutskever, Lukasz Kaiser, Karol Kurach, and James Martens. Adding gradient noise improves learning for very deep networks, 2015

    2015

  24. [24]

    Neggaz, Ihsen Alouani, Pablo R

    Mohamed A. Neggaz, Ihsen Alouani, Pablo R. Lorenzo, and Smail Niar. A reliability study on CNNs for critical embedded systems. InProceedings of IEEE International Conference on Computer Design, pages 476–479, Orlando, FL, 2018. IEEE

    2018

  25. [25]

    St-DRC: Stretchable DRAM re- fresh controller with no parity-overhead error correction scheme for energy-efficient dnns

    Duy-Thanh Nguyen, Nhut-Minh Ho, and Ik-Joon Chang. St-DRC: Stretchable DRAM re- fresh controller with no parity-overhead error correction scheme for energy-efficient dnns. In Proceedings of ACM/IEEE Design Automation Conference, pages 1–6, Las Vegas, NV, 2019. IEEE

    2019

  26. [26]

    GreenTPU: Improving timing error resilience of a near-threshold tensor processing unit

    Pramesh Pandey, Prabal Basu, Koushik Chakraborty, and Sanghamitra Roy. GreenTPU: Improving timing error resilience of a near-threshold tensor processing unit. InProceedings of ACM/IEEE Design Automation Conference, pages 1–6, Las Vegas, NV, 2019. IEEE

    2019

  27. [27]

    BTI anal- ysis tool—modeling of NBTI DC, AC stress and recovery time kinetics, nitrogen impact, and EOL estimation.IEEE Transactions on Electron Devices, 65(2):392–403, 2017

    Narendra Parihar, Nilesh Goel, Subhadeep Mukhopadhyay, and Souvik Mahapatra. BTI anal- ysis tool—modeling of NBTI DC, AC stress and recovery time kinetics, nitrogen impact, and EOL estimation.IEEE Transactions on Electron Devices, 65(2):392–403, 2017

    2017

  28. [28]

    Ares: A framework for quantifying the resilience of deep neural networks

    Brandon Reagen, Udit Gupta, Lillian Pentecost, Paul Whatmough, Sae Kyu Lee, Niamh Mulholland, David Brooks, and Gu-Yeon Wei. Ares: A framework for quantifying the resilience of deep neural networks. InProceedings of ACM/ESDA/IEEE Design Automation Conference, pages 1–6, San Francisco, CA, 2018. IEEE. 26

    2018

  29. [29]

    Evaluating fault resiliency of compressed deep neural networks

    Majid Sabbagh, Cheng Gongye, Yunsi Fei, and Yanzhi Wang. Evaluating fault resiliency of compressed deep neural networks. InProceedings of IEEE International Conference on Embedded Software and Systems, pages 1–7, Las Vegas, NV, 2019. IEEE

    2019

  30. [30]

    On the resilience of RTL NN accelerators: Fault characterization and mitigation

    Behzad Salami, Osman S Unsal, and Adrian Cristal Kestelman. On the resilience of RTL NN accelerators: Fault characterization and mitigation. InProceedings of IEEE International Sym- posium on Computer Architecture and High Performance Computing, pages 322–329, Lyon, France, 2018. IEEE

    2018

  31. [31]

    Reliability-aware quantization for anti-aging NPUs

    Sami Salamin, Georgios Zervakis, Ourania Spantidi, Iraklis Anagnostopoulos, J¨ org Henkel, and Hussam Amrouch. Reliability-aware quantization for anti-aging NPUs. InProceedings of IEEE Design, Automation & Test in Europe Conference & Exhibition, pages 1460–1465, Grenoble, France, 2021. IEEE

    2021

  32. [32]

    Salmani, M

    H. Salmani, M. Tehranipoor, and R. Karri. On design vulnerability analysis and trust bench- marks development. InProceedings of the IEEE International Conference on Computer Design (ICCD), pages 471–474, Oct 2013

    2013

  33. [33]

    Accurate neuron resilience prediction for a flexible reliability management in neural network accelerators

    Christoph Schorn, Andre Guntoro, and Gerd Ascheid. Accurate neuron resilience prediction for a flexible reliability management in neural network accelerators. InProceedings of IEEE Design, Automation & Test in Europe Conference & Exhibition, pages 979–984, Dresden, Germany, 2018. IEEE

    2018

  34. [34]

    15nm open-cell library and 45nm FREEPDK, 2021.https://si2.org/ open-cell-library

    Si2. 15nm open-cell library and 45nm FREEPDK, 2021.https://si2.org/ open-cell-library

    2021

  35. [35]

    Thermal effect on performance, power, and BTI aging in FinFET-based designs

    Warin Sootkaneung, Suppachai Howimanporn, and Sasithorn Chookaew. Thermal effect on performance, power, and BTI aging in FinFET-based designs. InProceedings of IEEE Eu- romicro Conference on Digital System Design, pages 345–351, Vienna, Austria, 2017. IEEE

    2017

  36. [36]

    Hardware trojan detection using controlled circuit aging.IEEE Access, 8:77415–77434, 2020

    Virinchi Roy Surabhi, Prashanth Krishnamurthy, Hussam Amrouch, Kanad Basu, J¨ org Henkel, Ramesh Karri, and Farshad Khorrami. Hardware trojan detection using controlled circuit aging.IEEE Access, 8:77415–77434, 2020

    2020

  37. [37]

    Exposing hardware trojans in embedded platforms via short- term aging.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 39(11):3519–3530, 2020

    Virinchi Roy Surabhi, Prashanth Krishnamurthy, Hussam Amrouch, J¨ org Henkel, Ramesh Karri, and Farshad Khorrami. Exposing hardware trojans in embedded platforms via short- term aging.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 39(11):3519–3530, 2020

    2020

  38. [38]

    Trojan detection in embedded systems with FinFET technology

    Virinchi Roy Surabhi, Prashanth Krishnamurthy, Hussam Amrouch, J¨ org Henkel, Ramesh Karri, and Farshad Khorrami. Trojan detection in embedded systems with FinFET technology. IEEE Transactions on Computers, 71(11):3061–3071, 2022

    2022

  39. [39]

    Golden-free robust age estimation to triage recycled ICs.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, pages 1–1, 2023

    Virinchi Roy Surabhi, Prashanth Krishnamurthy, Hussam Amrouch, J¨ org Henkel, Ramesh Karri, and Farshad Khorrami. Golden-free robust age estimation to triage recycled ICs.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, pages 1–1, 2023

    2023

  40. [40]

    Version R-2020.09-SP5

    Synopsys Inc.PrimeTime User Guide, 2020. Version R-2020.09-SP5

    2020

  41. [41]

    Vivienne Sze, Yu-Hsin Chen, Tien-Ju Yang, and Joel S. Emer. Efficient processing of deep neural networks: A tutorial and survey.Proceedings of the IEEE, 105(12):2295–2329, 2017. 27

    2017

  42. [42]

    A defect-tolerant accelerator for emerging high-performance applications

    Olivier Temam. A defect-tolerant accelerator for emerging high-performance applications. In Proceedings of IEEE Annual International Symposium on Computer Architecture, pages 356– 367, Portland, OR, 2012. IEEE

    2012

  43. [43]

    van Santen, Hussam Amrouch, Narendra Parihar, Souvik Mahapatra, and J¨ org Henkel

    Victor M. van Santen, Hussam Amrouch, Narendra Parihar, Souvik Mahapatra, and J¨ org Henkel. Aging-aware voltage scaling. InProceedings of IEEE Design, Automation & Test in Europe Conference & Exhibition, pages 576–581, Dresden, Germany, 2016. IEEE

    2016

  44. [44]

    ZyNet: Automating deep neural network implementation on low-cost re- configurable edge computing platforms

    Kizheppatt Vipin. ZyNet: Automating deep neural network implementation on low-cost re- configurable edge computing platforms. InProceedings of International Conference on Field- Programmable Technology, pages 323–326, Tianjin, China, 2019. IEEE

    2019

  45. [45]

    Krishnan, Rakesh Vattikonda, Srikanth Krishnan, and Yu Cao

    Wenping Wang, Vijay Reddy, Anand T. Krishnan, Rakesh Vattikonda, Srikanth Krishnan, and Yu Cao. Compact modeling and simulation of circuit reliability for 65-nm CMOS technology. IEEE Transactions on Device and Materials Reliability, 7(4):509–517, 2007

    2007

  46. [46]

    Whatmough, Sae Kyu Lee, Hyunkwang Lee, Saketh Rama, David Brooks, and Gu- Yeon Wei

    Paul N. Whatmough, Sae Kyu Lee, Hyunkwang Lee, Saketh Rama, David Brooks, and Gu- Yeon Wei. 14.3 a 28nm SoC with a 1.2GHz 568nJ/prediction sparse deep-neural-network engine with>0.1 timing error rate tolerance for IoT applications. InProceedings of IEEE International Solid-State Circuits Conference, pages 242–243, San Francisco, CA, 2017. IEEE

    2017

  47. [47]

    Analyzing and mitigating the impact of permanent faults on a systolic array based neural network accelerator

    Jeff Jun Zhang, Tianyu Gu, Kanad Basu, and Siddharth Garg. Analyzing and mitigating the impact of permanent faults on a systolic array based neural network accelerator. InProceedings of IEEE VLSI Test Symposium, pages 1–6, San Francisco, CA, 2018. IEEE

    2018

  48. [48]

    AEP: An error-bearing neural network accelerator for energy efficiency and model protection

    Lei Zhao, Youtao Zhang, and Jun Yang. AEP: An error-bearing neural network accelerator for energy efficiency and model protection. InProceedings of IEEE/ACM International Conference on Computer-Aided Design, pages 1047–1053, Irvine, CA, 2017. IEEE. 28

    2017