Quasar—A Process Variability-Aware Radiation Robustness Evaluation Tool

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents Quasar, an open-source tool for evaluating radiation robustness of digital circuits under process variability. It targets the electrical abstraction level, introduces several simulation optimization techniques, and demonstrates the application through a set of case studies. The work has potential value for the design automation and reliability analysis communities, but several critical issues must be addressed before publication.

1,While this work is presented as an extension of a conference paper [17], the manuscript fails to clearly highlight the specific novel contributions beyond the previous version. The authors should explicitly list what new algorithms, features, or validations were added and how they improve upon the initial release.

2,The comparison to prior work [45] is based on a single NAND3 gate and across mismatched technology nodes (32nm vs. 130nm), limiting credibility. The tool should be benchmarked against state-of-the-art commercial or academic tools (e.g., MUSCA SEP3, Synopsys Sentaurus) using consistent PDKs to validate accuracy.

3,All experiments are limited to small-scale examples (e.g., C17, NAND2). There is no analysis of Quasar’s runtime, memory consumption, or convergence behavior for medium-sized benchmarks (e.g., AES core or RISC-V blocks). The scalability and practical applicability of the tool remain uncertain.

4, Algorithm 1 is overly simplified and not presented in standard pseudocode format. Figures such as 10–13 lack detailed annotations or captions that explain their logic flow. These should be revised for clarity and consistency.

5,The LET formula in Eq. (1) relies on a fixed empirical factor (10.8L) without derivation or justification. A more detailed explanation of the physical modeling assumptions (e.g., material properties, device geometry) and sensitivity analysis is needed.

6, Tables 3–5 only provide summary statistics without confidence intervals, cumulative distributions, or error bars. The authors are encouraged to include more detailed plots (e.g., histograms, CDFs) to improve the interpretability of variability effects.

7,Although variability is discussed in terms of Work Function Fluctuation, the manuscript does not cover temperature or voltage variability. Since Quasar aims to support variability-aware design, PVT corner evaluations should be addressed or at least discussed in the conclusion.

8,The tool leverages machine learning (k-NN) for prediction but does not cite recent works in ML-assisted SET analysis or EDA acceleration. The authors should consider citing and comparing with recent DAC/DATE/TCAD literature from 2021–2024.

 

Author Response

Reviewer 1 - Comments and Suggestions for Authors

 

This manuscript presents Quasar, an open-source tool for evaluating radiation robustness of digital circuits under process variability. It targets the electrical abstraction level, introduces several simulation optimization techniques, and demonstrates the application through a set of case studies. The work has potential value for the design automation and reliability analysis communities, but several critical issues must be addressed before publication.

R1.Q1. While this work is presented as an extension of a conference paper [17], the manuscript fails to clearly highlight the specific novel contributions beyond the previous version. The authors should explicitly list what new algorithms, features, or validations were added and how they improve upon the initial release.

R1. A1. Thank you for the comment. We improve the description of the new contributions of this manuscript in the new version of th text. We have introduce in this manuscript expanded sections and new discussions about the Quasar tool. The main  differences between this manuscript and the previous paper published are:  

• An improved Critical LET Search algorithm exploring False Position algorithm that reduces the number of simulations reducing the search space. The paper [17] uses a binary search to find the LETth during the electrical simulations. In this submission, we propose a new strategy based on false position algorithm and provide a comparison between the advantages of the new method. Table 1 compared both strategies on the paper.
• A detailed description of the development of the Logical Simulation step. We improve the discussion about the methodology exploring a graph considering the Electrical Path Groups (EPG) to evaluate multi-level circuits.
• A new feature that significantly speeds up the Variability Evaluation exploring the  k-nearest neighbors (k-NN) regression approach. The k-NN method reduces the number of simulations by about half which implies a speedup of 2 on the full variability evaluation.
• A comprehensive discussion about applications and results obtained with the proposed version of Quasar, demonstrating the tool usability and scalability.

R1.Q2. The comparison to prior work [45] is based on a single NAND3 gate and across mismatched technology nodes (32nm vs. 130nm), limiting credibility. The tool should be benchmarked against state-of-the-art commercial or academic tools (e.g., MUSCA SEP3, Synopsys Sentaurus) using consistent PDKs to validate accuracy.

R1. A2. We recognize this limitation of our current comparison with related work. As the experiments from reference [45] are obtained for the IHP 130 nm technology node, and we found some initial conditions that  limit the easy description of the same circuits and execution of the IHP 130 nm model on HSPICE and NGSPICE, in the current version of the Quasar tool development, we restrict our comparison with a more qualitative approach, showing a rank of the most sensitive input vectors influence on the robustness of a basic cell, the NAND3. This comparative evaluation is to show potential applications of the proposed tool, without intention to demonstrate the correctness of the SET and process variability evaluation provided by Quasar. During the development of the Quasar tool, the results obtained are always compared with equivalent golden reference from pure SPICE simulation data.

R1.Q3. All experiments are limited to small-scale examples (e.g., C17, NAND2). There is no analysis of Quasar’s runtime, memory consumption, or convergence behavior for medium-sized benchmarks (e.g., AES core or RISC-V blocks). The scalability and practical applicability of the tool remain uncertain.

R1.A3. Quasar runtime for small-scale circuits such as the examples is reasonable compared to traditional simulation-based approach. Simple radiation characterization takes at most a few minutes. Full variability analysis however can take up to hours on some cases, according with the fault cases identified and the variability ranges requerided. Runtime is more closely related to the number of fault configurations than the size of the circuit itself. Although, after a first characterization, redundant configurations are discarded, reducing the required number of simulations. Therefore, as the number of input grows, runtime grows exponentially. As the number of nodes, and output in the circuit grows, runtime only grows linearly. A medium-sized benchmark is planned to be executed and the runtime analysis should be divulgate in the Quasar repository.

 R1.Q4. Algorithm 1 is overly simplified and not presented in standard pseudocode format. Figures such as 10–13 lack detailed annotations or captions that explain their logic flow. These should be revised for clarity and consistency.

R1.A4. The standard pseudocode was implemented. In the new version of manuscript we detailed more the explanations of what each function used do, the inputs of the algorithm and other details.

 

R1.Q5. The LET formula in Eq. (1) relies on a fixed empirical factor (10.8L) without derivation or justification. A more detailed explanation of the physical modeling assumptions (e.g., material properties, device geometry) and sensitivity analysis is needed.

R1.A5. Thanks for this observation. We improve the description of the fault models in the new version of the manuscript also including a discussion about other fault models available in the literature. The empirical factor of 10.8 is derived from observations that in a bulk silicon devices, where a typical charge collection depth  is 2 µm for every linear energy tranfer (LET) of 1 MeV-cm2/mg and an ionizing particle deposits about 10.8fC charge along each micron of its track [1][2].

 

[1] Carreno, V.A.; Choi, G.; Iyer, R. Analog-digital simulation of transient-induced logic errors and upset susceptibility of an advanced control system. NASA Technical Memorandum 4241 1990

 

[2] F. Wang and V. D. Agrawal, "Single Event Upset: An Embedded Tutorial," 21st International Conference on VLSI Design (VLSID 2008), Hyderabad, India, 2008, pp. 429-434, doi: 10.1109/VLSI.2008.28.

 

R1.Q6. Tables 3–5 only provide summary statistics without confidence intervals, cumulative distributions, or error bars. The authors are encouraged to include more detailed plots (e.g., histograms, CDFs) to improve the interpretability of variability effects.

R1.A6. We included in the new version of the manuscript a new discussion to show different graphics and analysis that can be obtained by the data generated with Quasar and also to show results considering other technology node. Thus, the new section shows a XOR2 input gate evaluated considering the 32 nm technology node from PTM, and present complementar statistical analysis that can be obtained with the data generated by Quasar. Quasar provides a log based output with all the simulation results indicating the current threshold applied on the circuit to observe a unexpected transition on the output, that is necessary to obtain the LETth or critical charge evaluation. This report identifies all the conditions for the fault, i.e., the node impacted, the output considered, the process variability applied to the NMOS and PMOS devices, and the n-hit or p-hit evaluation. From this data, R or python scripts can be applied (or other statistical tool from the designer preference), to obtain detailed analysis.

 

R1.Q7. Although variability is discussed in terms of Work Function Fluctuation, the manuscript does not cover temperature or voltage variability. Since Quasar aims to support variability-aware design, PVT corner evaluations should be addressed or at least discussed in the conclusion.

R1.A7. To demonstrate the potential of use of Quasar for PVT evaluation, we included a new case study providing a evaluation of a XOR2 gate at 32 nm technology node. With this case study, we shown the versatility of Quasar to deal with different technologies ( planar bulk CMOS, finfet) once that a SPICE compatible model are provided. Also we explore this case study to show that the data generated by the tool can be used for other statistical evaluations, according the user demand. 

 

R1.Q8. The tool leverages machine learning (k-NN) for prediction but does not cite recent works in ML-assisted SET analysis or EDA acceleration. The authors should consider citing and comparing with recent DAC/DATE/TCAD literature from 2021–2024.

R1.A8. Thank you for the suggestion. Indeed, the use of machine learning algorithms to accelerate single event effect analysis is an important topic. There are some works in the literature that propose different solutions. In this way, we introduced some works that use ML-assisted SET analysis in our discussion on section 4.5. Some of the papers also used the k-NN algorithm to classify some faults. It is our intention to include new fault models and functionalities in Quasar open code. We reinforce that this can be done by just  replacing or complementing the current modules.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper looks into how the tension-adjustable lattice structure made of layered zigzag piezoelectric beams behaves mechanically. The authors propose a model that exploits the superposition of deformations induced by piezoelectrics. However, the manuscript presents several critical limitations to which the authors will need to provide appropriate responses to the following observations:

1. How do the authors explain the management of thermal variability during radiation simulation in Quasar? How do the authors explain the management of thermal variability during radiation simulation in Quasar? Integrating and expanding the references with studies that address the monitoring of SAR and temperature variations based on smart electronic devices in Section 4.5 would strengthen the study in the conscious modelling of temperature in human-device interaction scenarios, which aligns with Quasar's goal of accounting for environmental variability (PVT effects).

2. Can Quasar be adapted to simulate radiation-induced failures in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? Can Quasar be adapted to simulate radiation-induced faults in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? Can Quasar be adapted to simulate radiation-induced faults in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? Can Quasar be adapted to simulate radiation-induced faults in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? The authors should integrate this topic into the introduction and references, including the citation of the present study doi: 10.3390/electronics13040790. This study highlights the need for radiation-hardened integrated circuits in patient-centred electronics, which supports a broader use case for Quasar.

3. How is the k-NN prediction model generalised across technological nodes or devices with different doping profiles?

4. Is there support or a plan to address time-dependent dielectric breakdown (TDDB) and ageing mechanisms that affect radiation robustness?

5. What are the limits in the scalability of Quasar to full SoC or RTL abstraction levels?

6. Could the authors elaborate on how their validation of logical masking compares in terms of accuracy or efficiency with traditional SAT-based approaches?

7. What explanation do the authors provide on how circuit-level variability and radiation sensitivity affect the integrity of the electrical signal in detection systems, especially those involving complex spatiotemporal patterns? What explanation do the authors provide on how circuit-level variability and radiation sensitivity affect the integrity of the electrical signal in detection systems, especially those involving complex spatiotemporal patterns? What explanation do the authors provide on how circuit-level variability and radiation sensitivity affect the integrity of the electrical signal in detection systems, especially those involving complex spatiotemporal patterns? The authors must, therefore, integrate the document and the bibliography by citing the present study. DOI:10.1109/TIM.2025.3573363, in which systems that integrate artificial intelligence models such as LSTM and U-Net for monitoring electrical phenomena demonstrate the relevance of signal robustness under variable electronic conditions. The suggested study provides an exemplary case of how soft errors or signal distortion can influence downstream inference in integrated machine learning systems. This quote highlights a possible future direction for extending the applicability of Quasar to AI-enhanced sensor systems, where maintaining signal integrity in the presence of variability and noise is essential for functional reliability.

8. Could the Quasar logical graph model be adapted for non-digital systems, continuous time, or hybrid analogue-digital blocks?

9. Have the authors considered the integration of different approaches, such as tomographic methods, for defect localisation or cross-validation?

10. The study thoroughly addresses radiation robustness at the electrical level but currently lacks a discussion on how the proposed tool can be integrated or validated within physically modelled systems, where structural or biomechanical behaviours also influence circuit performance and fault propagation—especially in embedded applications. Have the authors considered extending the applicability of Quasar to domains where the Finite Element Model (FEM) is used to simulate the physical environment surrounding the circuit, such as in biomedical or wearable systems? I'm not asking to implement a new model, but just to integrate into the document and reference the study doi: 10.3390/electronics14112268, which demonstrates how the integration of FEM with AI-based monitoring enables robust signal processing under variable physical conditions in the real world, which often induce electrical disturbances. This approach could provide a multidisciplinary validation framework for Quasar, linking physical stress, deformation, or electromagnetic interference to circuit-level fault sensitivity.

Author Response

Reviewer 2 - Comments and Suggestions for Authors

 

The paper looks into how the tension-adjustable lattice structure made of layered zigzag piezoelectric beams behaves mechanically. The authors propose a model that exploits the superposition of deformations induced by piezoelectrics. However, the manuscript presents several critical limitations to which the authors will need to provide appropriate responses to the following observations:

R2.Q1. How do the authors explain the management of thermal variability during radiation simulation in Quasar? How do the authors explain the management of thermal variability during radiation simulation in Quasar? Integrating and expanding the references with studies that address the monitoring of SAR and temperature variations based on smart electronic devices in Section 4.5 would strengthen the study in the conscious modelling of temperature in human-device interaction scenarios, which aligns with Quasar's goal of accounting for environmental variability (PVT effects).

R2.A1. Thank you for your suggestion. Currently, Quasar handles temperature variability by enabling users to specify temperature ranges in SPICE netlists, which are then evaluated during radiation simulations (Section 5.2 demonstrates this with a XOR2 logic gate). We have modified Section 4.5 in the revised manuscript to better align with Quasar's focus on environmental variability (PVT). Furthermore, variability analysis is crucial for integrated circuit design, particularly in human-device interaction scenarios. For this reason, we expanded our results to include PVT analysis, demonstrating Quasar's capability to handle reliability-critical applications.

R2.Q2. Can Quasar be adapted to simulate radiation-induced failures in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? Can Quasar be adapted to simulate radiation-induced faults in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? Can Quasar be adapted to simulate radiation-induced faults in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? Can Quasar be adapted to simulate radiation-induced faults in biomedical CMOS circuits (e.g., pacemakers or neural interfaces)? The authors should integrate this topic into the introduction and references, including the citation of the present study doi: 10.3390/electronics13040790. This study highlights the need for radiation-hardened integrated circuits in patient-centred electronics, which supports a broader use case for Quasar.

R2.A2 Thank you for your suggestion to improve the introduction discussing more general applications of the Quasar tool. The current version of Quasar is able to deal with combinational circuits descript in SPICE language for electrical simulation. For gate level simulation, the logic gates must be combinational and the circuit must have a reduced number of reconvergences. This limitation is due the fact that the logic evaluation is currently propagating the signals from the inputs to outputs, sequentially, so, reconvergences and feedbacks loops can affect this stage. However, it could be modified to skip this step in the Quasar code and the tool will proceed with the electrical simulations for SET and variability. Also new modules can be included in the open access code to deal with some particular demand of biomedical CMOS circuits. 

R2.Q3. How is the k-NN prediction model generalised across technological nodes or devices with different doping profiles?

R2.A3. The K-NN prediction model generalised without interference of the technological nodes parameters or devices with different doping profiles. To show this technology independence, we included in the new version of the manuscript a new experiment where we evaluates the process, voltage and temperature influence on the cell robustness of a XOR2 gate described in the bulk CMOS 32 nm technology.

R2.Q4. Is there support or a plan to address time-dependent dielectric breakdown (TDDB) and ageing mechanisms that affect radiation robustness?

R2.A4. Thank you for the comment. In fact, including aging effects like BTI in radiation impact analysis is of interest to us, and some studies already combine these effects [3]. Aging mechanisms such as BTI, HCI, and TDDB could be added as a new feature in the tool. However, further work is needed to model these effects across different transistor arrangements and integrate them with SETs. Since Quasar is under continuous development, this suggestion may be implemented in future versions.

[3] Mounir Mahmoud, Mohamed, et al. "Radiation-Hardened Aging-Resilient Digital Integrated Circuits in Nanoscale CMOS Technology." (2025).

R2.Q5. What are the limits in the scalability of Quasar to full SoC or RTL abstraction levels?

R2.Q5. Quasar seeks to work at an electrical level. It is currently not feasible to scale Quasar to full SoC level as it relies heavily on SPICE simulations. Simple nominal circuit characterization runs in reasonable time. Full variability characterization in circuits such as C17 can take a few hours. Runtime scales mainly with the number fault configuration. As the number of inputs grows runtime scales exponentially, while it only scales linearly with the increase of non input nodes. Improving the scalability is still a work in progress. Most of the specific improvements of this version are done to improve runtime overall but there are still work to be done. A better methodology to determine redundant fault configurations would be a good next step to improve runtime as a whole.

 

R2.Q6. Could the authors elaborate on how their validation of logical masking compares in terms of accuracy or efficiency with traditional SAT-based approaches?

R2.A6. The validation of logical masking implemented was executed with a exhaustive  brute force approach in which every fault configuration, including all input configurations, is tested. At the current version of the Quasar, our approach is not optimized, it was implemented to give a complete answer in reasonable time. As the number of electrical simulations done is the bottleneck of performance optimizing this part of the code is not a main concern right now. In this way, we have an approach that is perfectly accurate but not very efficient, so, at first glance, it is worse than a SAT-based approach. The main difference is that a SAT-based approach does not suffice all cases we need to analyse. Since we operate at an electrical level, not only logical, and we seek to completely map all fault configuration we need to consider different input combinations that would result on mostly the same logical behaviour. If there are two separate paths a fault can propagate to an output both paths will have different electrical characteristics which will influence the minimal LET. A traditional SAT-based approach would underestimate some  single electrical path the fault can take and would only return a single valid path. So, a SAT-based approache although efficient, it could compromise the accuracy of the evaluation.

R2.Q7. What explanation do the authors provide on how circuit-level variability and radiation sensitivity affect the integrity of the electrical signal in detection systems, especially those involving complex spatiotemporal patterns? The authors must, therefore, integrate the document and the bibliography by citing the present study. DOI:10.1109/TIM.2025.3573363, in which systems that integrate artificial intelligence models such as LSTM and U-Net for monitoring electrical phenomena demonstrate the relevance of signal robustness under variable electronic conditions. The suggested study provides an exemplary case of how soft errors or signal distortion can influence downstream inference in integrated machine learning systems. This quote highlights a possible future direction for extending the applicability of Quasar to AI-enhanced sensor systems, where maintaining signal integrity in the presence of variability and noise is essential for functional reliability.

R2.A7 The parts of  electrical signal in detection systems that is composed by logic circuits adopting transistors, mainly with the technology node is at nanometer scale, are susceptible to the effects of process variability, temperature influence and voltage oscillation impacts. Also if the detection system is design to a radiation critical environment, the SET effects can affet these parts of the circuit. We restringe our analysis to show potential application of the Quasar tool and development details. Different applications can use the tool and explore the SET and variability evaluation  for the requirements of each particular design. We included the suggested reference to improve the introduction discussing more general applications of the Quasar tool. Also, we discuss that new modules can be included in the open access code to deal with some particular demand of the target CMOS circuits. 

R2.Q8. Could the Quasar logical graph model be adapted for non-digital systems, continuous time, or hybrid analogue-digital blocks?

R2.A8 The current version of Quasar tool is not able to deal with non-digital systems of hybrid analogue-digital block because some inferences are taken considering well defined low and high signals, as expected in digital systems. Moreover, the logical graph model is implemented to deal with reduced number of reconvergence paths. However, due to the open-acess implementation and modularity, new modules could be inserted to deal with the particularities of non-digital systems, continuous time, or hybrid analogue-digital blocks.

R2.Q9. Have the authors considered the integration of different approaches, such as tomographic methods, for defect localisation or cross-validation?

R2.A9 At this moment, we concentrates our analysis to show potential application of the Quasar tool and development details.New modules can be included in the open access code to deal with some particular demand of the target CMOS circuits. Also, as Quasar input is a electrical level description of a circuit in SPICE language, circuits described in HDL languages must be priorly converted or extracted from the physical synthesis adopting any comercial or open-source tool available for that. The outputs of Quasar are provided as raw data in textual format or some prior configurable graphs and tables. The raw data can be easily converted and adapted for the input of other evaluation tools. New fault models and functionalities can aso be integrated in Quasar open code replacing or complementing the current modules. Thus, we believe that a large number of applications and integrations could be performend providing support for different kinds of applications.

R2.Q10. The study thoroughly addresses radiation robustness at the electrical level but currently lacks a discussion on how the proposed tool can be integrated or validated within physically modelled systems, where structural or biomechanical behaviours also influence circuit performance and fault propagation—especially in embedded applications. Have the authors considered extending the applicability of Quasar to domains where the Finite Element Model (FEM) is used to simulate the physical environment surrounding the circuit, such as in biomedical or wearable systems? I'm not asking to implement a new model, but just to integrate into the document and reference the study doi: 10.3390/electronics14112268, which demonstrates how the integration of FEM with AI-based monitoring enables robust signal processing under variable physical conditions in the real world, which often induce electrical disturbances. This approach could provide a multidisciplinary validation framework for Quasar, linking physical stress, deformation, or electromagnetic interference to circuit-level fault sensitivity.

R2.A10 We are thankful for your suggestion and we have improved the future work discussion to include this comments. The integration with other physically modelled systems can be done with the inclusion of dedicated modules or by the interfaces of the Quasar tool. As Quasar input is a electrical level description of a circuito in SPICE language, circuits described in HDL languages must be priorly converted or extracted from the physical synthesis adopting any comercial or open-source tool available for that. The outputs of Quasar are provided as raw data in textual format or some prior configurable graphs and tables. The raw data can be easily converted and adapted for the input of other evaluation tools. New fault models and functionalities can aso be integrated in Quasar open code replacing or complementing the current modules. Thus, we believe that a large number of applications and integrations could be performend providing support for different kinds of applications.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I appreciate the authors’ thorough revision, detailed point-by-point responses, and substantial improvements to both the manuscript and its presentation. The manuscript presents Quasar, an open-source tool for evaluating radiation robustness under process variability. Several of my major concerns have been partially addressed. However, further clarifications and improvements are required before acceptance。

1， Lack of Cross-Tool Validation (Accuracy and Credibility Concern)

While the authors acknowledge the challenge of comparing with MUSCA SEP3, Sentaurus, or similar tools due to technology node differences, no direct quantitative validation has been presented. Without demonstrating Quasar’s prediction accuracy against an established reference under identical conditions, the tool’s credibility remains questionable.

2. No Evaluation on Medium-Scale Designs (Practical Applicability Concern)

All current experiments remain limited to small logic gates or minimal benchmarks. There is no runtime, memory, or convergence analysis on medium-sized designs (e.g. AES module, RISC-V core, >500 gates). This limits the practical relevance of the tool for real-world design flows.

3. Prediction Model and Algorithm Description Not Reproducible

Although the authors revised Algorithm 1 and explained the k-NN regression model, critical implementation details are missing. For instance, no quantitative prediction error metrics (e.g. RMSE, MAE) are reported, the input feature dimensionality for k-NN is unclear, and the pseudocode lacks formal structure (inputs, outputs, initialization, termination).

 

 

 

 

 

 

 

 

 

 

Author Response

 

Dear Editor and Reviewers,

 

Thank you for reviewing our submission. We sincerely appreciate your comments and constructive feedback on our manuscript. To facilitate clarity in our response, we have organized this letter into sections addressed to each reviewer, following the order of comments as presented in the decision e-mail. Within each section, we reproduce the original content the editorial staff provided, with our responses highlighted in blue for easy reference. New alterations on the manuscript are highlighted in orange. New alterations on the Once again, we are grateful for your valuable input and insights, which have undoubtedly enhanced the quality of our manuscript.

1， Lack of Cross-Tool Validation (Accuracy and Credibility Concern)

While the authors acknowledge the challenge of comparing with MUSCA SEP3, Sentaurus, or similar tools due to technology node differences, no direct quantitative validation has been presented. Without demonstrating Quasar’s prediction accuracy against an established reference under identical conditions, the tool’s credibility remains questionable.

R1.A1 Quasar is basically a SPICE enhancer. Essentially, it ties to variables, inserted current and output voltage, and gives the minimum current needed to force the output voltage into a certain value. In this way, Quasar will give the same answer as if the same simulation was done manually with SPICE.  Thus, we ran exhaustive SPICE simulations for the different circuits evaluated until this moment, checking for the minimal current that provokes an output inversion based on the expected Truth Table behavior.  This demonstrate that Quasar reduces the number o electrical simulations need to identify the most sensitive cases compared to exhaustive electrical simulations and remain the precision reaching the same critical currents. Additionally, we took the XOR circuit and run manual SPICE simulations for all valid cases. We used the same output current provided by Quasar and run a simulation measuring the output voltage. We ignore combinations that are logically masked or combinations that require a P-Hit or a N-Hit on a node in which this is not possible. The results show that the current provided by Quasar does indeed leave the output voltage at 50% the voltage supply (0.7 V).

 

2. No Evaluation on Medium-Scale Designs (Practical Applicability Concern)

All current experiments remain limited to small logic gates or minimal benchmarks. There is no runtime, memory, or convergence analysis on medium-sized designs (e.g. AES module, RISC-V core, >500 gates). This limits the practical relevance of the tool for real-world design flows.

R1.A2 The relevance of Quasar tool is in the electrical level of circuit evaluation and design, for example, in the characterization of standard cells libraries or modules for high abstraction level synthesis. This is relevant and allows a criterial evaluation exploring transistor level mitigation approaches and circuit rearrangements. For larger circuits, considering combinational and sequential elements, digital and analog blocks, high level of evaluation tools could be explored and the results refined at fine grained level for the most sensitive blocks. Our intention and goal with Quasar is provide this environment for electrical level evaluation of combinational blocks or cells at gate or transistor level of design.

(Table attached on the pdf file)

3. Prediction Model and Algorithm Description Not Reproducible

Although the authors revised Algorithm 1 and explained the k-NN regression model, critical implementation details are missing. For instance, no quantitative prediction error metrics (e.g. RMSE, MAE) are reported, the input feature dimensionality for k-NN is unclear, and the pseudocode lacks formal structure (inputs, outputs, initialization, termination).

R1.A3 Thanks for your suggestion. Algorithm 1 was modified to include initialization and termination. Also, we are exploring the K-NN regression algorithm to speed up the Critical LET Search and not as a machine learning model to learn the regression model for the LET behavior. Thus, we have not applied regression evaluation metrics, we measure quality by the reduction of average simulations per fault configuration. The K-NN was chosen due to the smoothness and insignificant noise of the work-function (WF) distribution. More specifically, we are adopting the euclidean distance weighted 3-NN from the Scikit-learn Python module, using as input the physical parameters that are changed in the variability evaluation.. In the test, the WF is taken and its distribution is applied to both NMOS and PMOS devices. We improve this explanation on the text and the description of the algorithms and implementation. 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I thank the authors for their replies to the comments. I have no further comments to ask. 

Author Response

 

Dear Editor and Reviewers,

 

Thank you for reviewing our submission. We sincerely appreciate your comments and constructive feedback on our manuscript. To facilitate clarity in our response, we have organized this letter into sections addressed to each reviewer, following the order of comments as presented in the decision e-mail. Within each section, we reproduce the original content the editorial staff provided, with our responses highlighted in blue for easy reference. New alterations on the manuscript are highlighted in orange. New alterations on the Once again, we are grateful for your valuable input and insights, which have undoubtedly enhanced the quality of our manuscript.

 

Reviewer 2

I thank the authors for their replies to the comments. I have no further comments to ask. 

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have made substantial progress in this revision and have adequately addressed the major concerns raised in previous rounds. They clarified the equivalence between Quasar and SPICE-level simulations, provided targeted validation on critical LET injection, and clearly positioned the tool for electrical-level analysis of standard cells. The use of k-NN for simulation reduction is now well explained, and algorithm descriptions have been improved for clarity and reproducibility.

While the manuscript still requires minor revisions related to figure clarity, such as Low Resolution and Line Clarity, small Fonts and Axis Labels, et al , and language polish, the scientific content is now sound, well-positioned, and ready for acceptance after these final corrections.

 

Author Response

The authors have made substantial progress in this revision and have adequately addressed the major concerns raised in previous rounds. They clarified the equivalence between Quasar and SPICE-level simulations, provided targeted validation on critical LET injection, and clearly positioned the tool for electrical-level analysis of standard cells. The use of k-NN for simulation reduction is now well explained, and algorithm descriptions have been improved for clarity and reproducibility.

While the manuscript still requires minor revisions related to figure clarity, such as Low Resolution and Line Clarity, small Fonts and Axis Labels, et al , and language polish, the scientific content is now sound, well-positioned, and ready for acceptance after these final corrections.

Answer:
Thank you for your kind response. We have reviewed the manuscript and polished the language when we found appropriate. We also remade low resolutions figures to fit in with the style presented on the rest of the manuscript.