Pith. sign in

REVIEW 4 major objections 5 minor 48 references

Template-grounded LLM agents can generate SPICE-validated SAR ADCs

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

ATLAS combines template-constrained LLM agents with Bayesian optimization to produce SAR ADC netlists that meet user specs in simulation.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection A plausible template-constrained LLM-agent pipeline that gets an 8-bit SAR ADC to spec in simulation, but the underspecified human testbench step and missing artifacts keep the central claim conditional. the 4 major comments →

arxiv 2607.14165 v1 pith:HTPULV2R submitted 2026-07-15 cs.SE cs.AIcs.ARcs.LG

Towards Reliable AI-Assisted Analog Design: Template-Constrained LLM Agents for SAR ADC Generation

Dimple Vijay Kochar , Hae-Seung Lee , Anantha P. Chandrakasan This is my paper
classification cs.SE cs.AIcs.ARcs.LG
keywords analog EDASAR ADCLLM agentstemplate-constrained generationcircuit netlist generationSPICE simulationretrieval-augmented generationautomated layout
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Analog circuit design has resisted automation because a single wrong connection can destroy a design, and LLMs asked to produce netlists from scratch hallucinate. This paper argues that the problem becomes tractable if the LLM's search space is constrained by expert-curated templates for each sub-block and grounded in summaries of published ADC designs. It presents ATLAS, a multi-stage agentic pipeline that plans, selects templates, stitches a full Successive Approximation Register (SAR) analog-to-digital converter (ADC) netlist, iterates with simulation feedback, and sizes the final circuit with an optimizer. The headline result is an 8-bit SAR ADC generated on a 45nm CMOS process that passes SPICE and meets its targets (ENOB 7.59, power 6 µW), with additional demos at 4-bit, 10-bit, and on a 65nm process. The pith is that hard physical constraints, encoded as templates, are what turn an untrustworthy text generator into a reliable circuit synthesizer.

Core claim

The paper's central claim is that template-constrained generation—having the LLM select and minimally modify expert-written component templates instead of writing netlists from scratch—makes LLM-based analog synthesis reliable enough to pass SPICE. In their ATLAS flow, a planner first produces a grounded design plan from retrieved ADC literature, a selector picks component templates (dynamic comparators, binary/split-capacitor DAC arrays, SAR logic), an integrator connects them into a full netlist, a bit-modifier adjusts resolution, and a debugger iterates on simulation errors; finally a sampling-based optimizer sizes transistors and capacitors. The authors report that their grounded zero-sh

What carries the argument

The key mechanism is the two-wing generation strategy plus the grounding loop. Wing one is an unconstrained grounded zero-shot generator with rule-based verification; wing two—the one that actually worked—is Template-Constrained Generation, where a curated library of sub-block templates with textual pros/cons is given to a Selector LLM along with retrieval-augmented expert knowledge. The Selector chooses and optionally modifies a template, an Integrator LLM stitches the chosen templates into a complete SAR ADC, a Bit-Modifier adjusts resolution, and a Testbench Creator plus Debugger LLM close the simulation loop. The final sizing stage uses an external multi-objective optimizer to set transi

Load-bearing premise

The reported simulation metrics for the headline 8-bit SAR ADC depend on a testbench that a human expert also rectified to validate correctness; if the human adjusted the testbench or simulation setup until the targets were reached, the measured ENOB, SINAD, and power would not independently confirm the generated ADC's quality.

What would settle it

Take the final generated 8-bit netlist (which the paper does not release) and run it in an unmodified testbench script written by an independent engineer, with no human intervention, and check whether ENOB, SINAD, and power still meet the targets of 7.5, 45 dB, and 8 µW. If they do not, the central claim of simulation-validated autonomous generation collapses.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If this framework generalizes beyond the demonstrated topologies, analog designers could use LLM agents to rapidly explore SAR ADC architecture variants from text-level specs.
  • The success of the template-constrained wing over the zero-shot wing suggests future systems should invest in template libraries rather than hoping raw LLM capability will suffice.
  • The reported failures of direct LLM/multimodal prompting serve as a baseline: any claim that LLMs can do analog design should be compared against this evidence.
  • The port to a 65nm process without node-specific optimization indicates that generated netlists may transfer across process nodes, though sizing must be redone.
  • The 10-bit result, which missed its ENOB target, defines the current capability boundary and motivates work on bit-scaling methods.

Where Pith is reading between the lines

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

  • Because the paper states that a human expert also rectifies the testbench to validate the reported metrics, the headline numbers are not a purely autonomous measurement; an independent testbench pass would strengthen or refute the claim.
  • The framework's dependence on expert-curated templates and retrieved literature means its ceiling is effectively the set of topologies humans already know how to write; it is a tool for automation, not invention.
  • A natural stress test: give ATLAS an unusual spec combination (e.g., ultra-low supply voltage with high speed) and see whether the RAG-based planner selects a template that actually requires novel modification or just a standard one.
  • The paper's own log—zero-shot wing fails, template wing succeeds—suggests that the LLM is performing template matching plus glue logic, so the marginal contribution of the LLM over a conventional scripted compiler is flexibility of choice and stitching, a claim worth measuring directly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes ATLAS, a multi-stage LLM-agent framework for synthesizing SAR ADCs. ATLAS comprises (1) expert-knowledge grounding through retrieval-augmented generation from open-source ADC literature, (2) netlist generation via a grounded zero-shot wing and a template-constrained wing, followed by integration, bit modification, testbench creation, and simulation-debug loops, and (3) netlist sizing through an LLM-driven parameter extractor coupled to multi-objective Bayesian optimization. The authors report that ATLAS generated an 8-bit SAR ADC on Cadence GPDK-45nm that meets target specifications (ENOB 7.59 vs. 7.5 target; Table 1). They also report a TSMC-65nm migration, a 4-bit ADC, and a 10-bit modification attempt, the last of which misses its ENOB target (Table 4). The main claim is that template-constrained LLM agents can produce a simulation-validated SAR ADC from user specifications.

Significance. If fully substantiated, this would be a useful step toward LLM-assisted analog synthesis, particularly the decomposition into grounded planning, template-constrained generation, and optimizer-driven sizing. The paper usefully documents qualitative failures of direct LLM prompting. The central existence proof (the 8-bit ADC) is plausible but not independently verifiable from the manuscript: the testbench was human-rectified (Section 4.2.3), and no netlists or testbenches are released. Given that the reported metrics are produced by a human-adjusted validation setup, the headline claim 'passes simulation validation' cannot be cleanly attributed to ATLAS. The generalization evidence is also weakened by the 65nm result relying on custom migration scripts rather than ATLAS. These issues are fixable with additional disclosure and experiments, so the contribution has merit but needs major revision.

major comments (4)
  1. [Section 4.2.3] The statement that 'a human expert also rectifies the testbench to validate the correctness of the reported metrics' is load-bearing. In SAR ADC simulation, measured ENOB/SINAD/SFDR depend strongly on testbench details: input frequency and coherence, sampling instant, clock timing, ideal-DAC model, settling time, and output loading. If the human rectification involved adjusting any of these to make the target metrics reachable, Table 1 does not independently confirm the generated circuit. The manuscript gives no record of what the human changed, how many changes were made, or whether the changes could bias the metrics. Without releasing the exact netlist, testbench, simulation setup, and edit history, the central claim that the ADC 'passes simulation validation' is not reproducible. Please specify the nature and scope of human rectification, and provide the artifacts or a detailed accoun
  2. [Section 5.2] The paper states that 'the grounded zero-shot netlist generation had some errors and didn’t succeed, but the fallback to template-constrained netlist succeeded.' This means the only successful path in the main demonstration is the template-constrained wing, which relies on human-curated templates and meta-information. The framing 'LLM agentic framework' may overstate the degree of autonomous generation: the LLM selects and lightly modifies known-good templates, rather than synthesizing a topology. This should be acknowledged explicitly, and the contribution of the LLM relative to the template library should be quantified. For instance, what modifications did the LLM make beyond concatenating templates, and how does performance compare to using the templates directly with the optimizer?
  3. [Section 6.1] The TSMC-65nm generalization result in Table 2 is not produced by ATLAS: the netlist was migrated using 'custom scripts developed to automatically map and substitute the technology primitives,' and a 'local optimizer' was used for sizing. This does not demonstrate ATLAS's ability to generate across technology nodes; it demonstrates that the generated netlist can be ported with a separate toolchain. The claim in the abstract and introduction that ATLAS develops SAR ADCs 'across technology nodes' is therefore unsupported. Either run ATLAS end-to-end on a second foundry node or revise the claims to describe this as a technology-transfer experiment.
  4. [Section 6.3] The 10-bit modification (Table 4) fails to meet the target ENOB (8.93 vs. 9.5), SINAD (55.5 vs. 58 dB), SFDR (62.9 vs. 68 dB), and THD (-59.9 vs. -65 dB). This is a negative result, and the text labels it 'capability of bit modification,' but the capability claim is weakened. This is not necessarily a problem if the paper is honest about the failure, but the presentation should clearly distinguish 'successful' demonstrations (8-bit, 4-bit) from the unsuccessful 10-bit attempt, and the contribution list should not imply all four SAR ADCs met their targets. The discussion of human-expert inspection of the bit modifier's 'pattern-matching reasoning' should also be marked as qualitative, not as validation of the circuit.
minor comments (5)
  1. [Section 4.1.2] The retriever is described as using k-NN with k=5, but no details are given on the metric space, normalization, or how the target specs are represented. This is needed for reproducibility of the RAG stage.
  2. [Section 5.1] The power calculation is described as averaging 'the DAC, comparator, and the SAR logic powers,' but it is unclear how these are measured separately in the testbench and whether dynamic power includes clock and digital switching. Please clarify.
  3. [Section 6.2] The 4-bit ADC result is reported as meeting specs, but the text says 'the results show how it meets the specs.' It would be useful to state explicitly which specifications were met and whether the same human-rectified testbench procedure was used.
  4. [General] The paper contains several typographical and formatting issues, e.g., 'ATLASto' and 'ATLAS-AgenticTemplate-constrainedLLM-basedADCSynthesizer' with missing spaces. Figures 2 and 3 are referenced with duplicate numbering in the text. Please proofread.
  5. [Section 6.3] The bit modification experiment uses Gemini 3.1 as the LLM, whereas the main 8-bit experiment uses GPT-4o. This inconsistency should be acknowledged, since it complicates the attribution of performance differences to the framework.

Circularity Check

0 steps flagged

No significant circularity: target specs are design inputs, not predictions; the 10-bit failure shows spec-meeting is not forced.

full rationale

ATLAS is an empirical system paper, not a first-principles derivation, so its claim is that a generated netlist passes SPICE validation. The target specifications are legitimately used as design/optimization inputs in the retriever (Sec. 4.1.2), template selector (Sec. 4.2.2), and sizer (Sec. 4.3), but the paper never calls the resulting Table 1 metrics a prediction, and the 10-bit experiment (Table 4) fails to meet specs, showing that spec-meeting is not guaranteed by construction. The only in-scope limitation is Sec. 4.2.3: 'a human expert also rectifies the testbench to validate the correctness of the reported metrics'; this is a reproducibility/validation caveat, not a circular step, because the text does not state that metrics were tuned to targets. Self-citations ([17] LEDRO, [46] RS-SAR) are prior published methods used as components, not uniqueness theorems or unverified assumptions. No self-definitional, fitted-input-as-prediction, or ansatz-via-citation pattern is present.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 0 invented entities

The framework leans on hand-curated templates, hand-set optimizer ranges, and a human testbench rectification step. No new physical entities are introduced; the central dependencies are the template library and the integrity of the simulation flow.

free parameters (3)
  • k-NN retrieval k = 5
    Section 5.1: retriever uses k=5 nearest neighbors in min-max normalized feature space; chosen by hand.
  • Optimizer parameter ranges = Not specified
    Section 4.3: Sizer supplies parameters and expected ranges 'from prior expert knowledge'; ranges are hand-set and directly affect achievable specs.
  • Number of debugger/simulation iterations = fixed number (unspecified)
    Section 4.2.3: 'This loop is run multiple times until simulation succeeds'; stopping criterion not specified, could influence results.
axioms (3)
  • domain assumption Curated template library contains topologies capable of meeting target specs
    Section 4.2.2: templates for StrongArm, Miyahara, binary-weighted and split-capacitor DACs are curated by human experts; the framework's success depends on these templates being adequate.
  • domain assumption SPICE simulation with human-rectified testbench accurately reflects circuit performance
    Section 4.2.3: 'a human expert also rectifies the testbench to validate the correctness of the reported metrics.' If the testbench was tuned to produce target metrics, results are biased.
  • domain assumption Retrieved paper summaries are free of errors and provide correct grounding
    Section 4.1.1: LLM-based summarization of filtered papers is used as ground truth; errors propagate to planner.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Reliable AI-Assisted Analog Design: Template-Constrained LLM Agents for SAR ADC Generation." pith.science (2026). https://pith.science/paper/HTPULV2R

@misc{pith2026260714165,
  author       = {Pith},
  title        = {Pith review of: Towards Reliable AI-Assisted Analog Design: Template-Constrained LLM Agents for SAR ADC Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HTPULV2R}},
  note         = {Machine review of arXiv:2607.14165}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

While Large Language Models (LLMs) have demonstrated significant capability in software code generation, their application to analog Electronic Design Automation (EDA) is bottlenecked. Owing to limited circuit topology understanding and data, directly prompting LLMs and multimodal models leads to hallucinations and failure to produce schematics capable of passing rigorous SPICE simulations, as we show in our work. Instead, we propose an end-to-end, multi-step LLM agentic framework ATLAS, capable of generating a functional Successive Approximation Register (SAR) Analog-to-Digital Converter (ADC) that successfully passes simulation validation. To adhere to the rigid constraints of analog design, we utilize expert knowledge to ground the LLM in its planning, selection, parameterization, and iterative modification. As part of ATLAS, we introduce Template-Constrained Generation - which unlike other template-based works - builds towards a more generalized SAR ADC generation flow. We demonstrate a strong proof-of-concept of our framework by developing SAR ADCs across technology nodes and input specs. Overall, our expert-knowledge grounded multi-step agentic ATLAS establishes a pragmatic foundation for integrating LLMs into reliable analog design methodologies.

Figures

Figures reproduced from arXiv: 2607.14165 by Anantha P. Chandrakasan, Dimple Vijay Kochar, Hae-Seung Lee.

Figure 1
Figure 1. Figure 1: High-level workflow of our proposed multi-step [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Highlighting several limitations of directly prompting LLMs for netlist generation. On the left (a), we highlight several [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Highlighting the limitations and lack of general [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: Highlighting the limitations of utilizing multi [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: The workflow for this stage comprises two main steps: (1) [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: High-level illustration of stage 3 of our agentic [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Illustrating the choices and the underlying rationale [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Illustration highlighting the bit-modification ca [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗

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 · 11 linked inside Pith

  1. [1]

    The dawn of ai-native eda: Opportunities and challenges of large circuit models,

    L. Chen, Y. Chen, Z. Chu, W. Fang, T.-Y. Ho, R. Huang, Y. Huang, S. Khan, M. Li, X. Liet al., “The dawn of ai-native eda: Opportunities and challenges of large circuit models, ”arXiv preprint arXiv:2403.07257, 2024

    Pith/arXiv arXiv 2024

  2. [2]

    Ai-driven integrated circuit design: A survey of techniques, challenges, and opportunities,

    I. Guven, M. Parlak, D. Lederer, and C. De Vleeschouwer, “Ai-driven integrated circuit design: A survey of techniques, challenges, and opportunities, ”IEEE Access, 2025

    2025

  3. [3]

    Swe- bench: Can language models resolve real-world github issues?

    C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan, “Swe- bench: Can language models resolve real-world github issues?”arXiv preprint arXiv:2310.06770, 2023

    Pith/arXiv arXiv 2023

  4. [4]

    Chipnemo: Domain-adapted llms for chip design,

    M. Liu, T.-D. Ene, R. Kirby, C. Cheng, N. Pinckney, R. Liang, J. Alben, H. Anand, S. Banerjee, I. Bayraktarogluet al., “Chipnemo: Domain-adapted llms for chip design, ”arXiv preprint arXiv:2311.00176, 2023

    Pith/arXiv arXiv 2023

  5. [5]

    Grpo with state mutations: Improving llm-based hardware test plan generation,

    D. V. Kochar, N. Pinckney, G.-T. Liu, C.-T. Ho, C. Deng, H. Ren, and B. Khailany, “Grpo with state mutations: Improving llm-based hardware test plan generation, ” arXiv preprint arXiv:2601.07593, 2026

    arXiv 2026

  6. [6]

    Cmos op-amp sizing using a geometric program- ming formulation,

    P. Mandal and V. Visvanathan, “Cmos op-amp sizing using a geometric program- ming formulation, ”IEEE Transactions on Computer-Aided Design of Integrated circuits and systems, vol. 20, no. 1, pp. 22–38, 2002

    2002

  7. [7]

    An efficient asynchronous batch bayesian optimization approach for analog circuit synthesis,

    S. Zhang, F. Yang, D. Zhou, and X. Zeng, “An efficient asynchronous batch bayesian optimization approach for analog circuit synthesis, ” in2020 57th ACM/IEEE design automation conference (DAC). IEEE, 2020, pp. 1–6

    2020

  8. [8]

    Au- tockt: Deep reinforcement learning of analog circuit designs,

    K. Settaluri, A. Haj-Ali, Q. Huang, K. Hakhamaneshi, and B. Nikolic, “Au- tockt: Deep reinforcement learning of analog circuit designs, ”arXiv preprint arXiv:2001.01808, 2020

    Pith/arXiv arXiv 2001

  9. [9]

    Gcn-rl circuit designer: Transferable transistor sizing with graph neural networks and reinforcement learning,

    H. Wang, K. Wang, J. Yang, L. Shen, N. Sun, H.-S. Lee, and S. Han, “Gcn-rl circuit designer: Transferable transistor sizing with graph neural networks and reinforcement learning, ” in2020 57th ACM/IEEE Design Automation Conference (DAC). IEEE, 2020, pp. 1–6. Towards Reliable AI-Assisted Analog Design

    2020

  10. [10]

    Opensar: An open source automated end-to-end sar adc compiler,

    M. Liu, X. Tang, K. Zhu, H. Chen, N. Sun, and D. Z. Pan, “Opensar: An open source automated end-to-end sar adc compiler, ” in2021 IEEE/ACM International Conference On Computer Aided Design (ICCAD). IEEE, 2021, pp. 1–9

    2021

  11. [11]

    1-and 80-ms/s sar adcs in 40-nm cmos with end-to-end compilation,

    ——, “1-and 80-ms/s sar adcs in 40-nm cmos with end-to-end compilation, ”IEEE Solid-State Circuits Letters, vol. 5, pp. 292–295, 2022

    2022

  12. [12]

    A hybrid design automation tool for sar adcs in iot,

    M. Ding, P. Harpe, G. Chen, B. Busze, Y.-H. Liu, C. Bachmann, K. Philips, and A. Van Roermund, “A hybrid design automation tool for sar adcs in iot, ”IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 26, no. 12, pp. 2853–2862, 2018

    2018

  13. [13]

    A reusable code-based sar adc design with cdac compiler and synthesizable analog building blocks,

    M.-J. Seo, Y.-J. Roh, D.-J. Chang, W. Kim, Y.-D. Kim, and S.-T. Ryu, “A reusable code-based sar adc design with cdac compiler and synthesizable analog building blocks, ”IEEE Transactions on circuits and systems II: Express Briefs, vol. 65, no. 12, pp. 1904–1908, 2018

    1904

  14. [14]

    Multiagent based reinforcement learning (ma-rl): An automated designer for complex analog circuits,

    J. Bao, J. Zhang, Z. Huang, Z. Bi, X. Feng, X. Zeng, and Y. Lu, “Multiagent based reinforcement learning (ma-rl): An automated designer for complex analog circuits, ”IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 43, no. 12, pp. 4398–4411, 2024

    2024

  15. [15]

    A global-local optimization approach for asynchronous sar adc design,

    Y. Hao, K. Li, M. Gandara, S. Li, and B. Liu, “A global-local optimization approach for asynchronous sar adc design, ”arXiv preprint arXiv:2507.19541, 2025

    Pith/arXiv arXiv 2025

  16. [16]

    Automated sar adc sizing using analytical equations,

    Z. Li, Z. Tao, Y. Zhou, Y. Shi, Z. Yu, T.-J. Lin, and L. He, “Automated sar adc sizing using analytical equations, ” in2025 International Symposium of Electronics Design Automation (ISEDA). IEEE, 2025, pp. 214–220

    2025

  17. [17]

    Ledro: Llm-enhanced design space reduction and optimization for analog circuits,

    D. V. Kochar, H. Wang, A. P. Chandrakasan, and X. Zhang, “Ledro: Llm-enhanced design space reduction and optimization for analog circuits, ” in2025 IEEE Inter- national Conference on LLM-Aided Design (ICLAD). IEEE, 2025, pp. 141–148

    2025

  18. [18]

    Heart: A hierarchical circuit reasoning tree-based agentic framework for ams design optimization,

    S. Poddar, C.-T. Ho, Z. Wei, W. Cao, H. Ren, and D. Z. Pan, “Heart: A hierarchical circuit reasoning tree-based agentic framework for ams design optimization, ” arXiv preprint arXiv:2511.19669, 2025

    arXiv 2025

  19. [19]

    Eesizer: Llm-based ai agent for sizing of analog and mixed signal circuit,

    C. Liu and D. Chitnis, “Eesizer: Llm-based ai agent for sizing of analog and mixed signal circuit, ”IEEE Transactions on Circuits and Systems I: Regular Papers, 2025

    2025

  20. [20]

    Lamagic: Language-model-based topology generation for analog integrated circuits,

    C.-C. Chang, Y. Shen, S. Fan, J. Li, S. Zhang, N. Cao, Y. Chen, and X. Zhang, “Lamagic: Language-model-based topology generation for analog integrated circuits, ”arXiv preprint arXiv:2407.18269, 2024

    Pith/arXiv arXiv 2024

  21. [21]

    Atelier: An automated analog circuit design framework via multiple large language model-based agents,

    J. Shen, Z. Chen, J. Zhuang, J. Huang, F. Yang, L. Shang, Z. Bi, C. Yan, D. Zhou, and X. Zeng, “Atelier: An automated analog circuit design framework via multiple large language model-based agents, ”IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2025

    2025

  22. [22]

    Analogxpert: Automating ana- log topology synthesis by incorporating circuit design expertise into large lan- guage models,

    H. Zhang, S. Sun, Y. Lin, R. Wang, and J. Bian, “Analogxpert: Automating ana- log topology synthesis by incorporating circuit design expertise into large lan- guage models, ” in2025 International Symposium of Electronics Design Automation (ISEDA). IEEE, 2025, pp. 772–777

    2025

  23. [23]

    Artisan: Automated operational amplifier design via domain-specific large language model,

    Z. Chen, J. Huang, Y. Liu, F. Yang, L. Shang, D. Zhou, and X. Zeng, “Artisan: Automated operational amplifier design via domain-specific large language model, ” inProceedings of the 61st ACM/IEEE Design Automation Conference, 2024, pp. 1–6

    2024

  24. [24]

    Analogseeker: An open-source foundation language model for analog circuit design,

    Z. Chen, J. Zhuang, J. Shen, X. Ke, X. Yang, M. Zhou, Z. Du, X. Yan, Z. Wu, Z. Xu et al., “Analogseeker: An open-source foundation language model for analog circuit design, ”arXiv preprint arXiv:2508.10409, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    Analogcoder: Analog circuit design via training-free code generation,

    Y. Lai, S. Lee, G. Chen, S. Poddar, M. Hu, D. Z. Pan, and P. Luo, “Analogcoder: Analog circuit design via training-free code generation, ” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 1, 2025, pp. 379–387

    2025

  26. [26]

    Ampagent: An llm-based multi-agent system for multi-stage amplifier schematic design from literature for process and performance porting,

    C. Liu, W. Chen, A. Peng, Y. Du, L. Du, and J. Yang, “Ampagent: An llm-based multi-agent system for multi-stage amplifier schematic design from literature for process and performance porting, ”arXiv preprint arXiv:2409.14739, 2024

    Pith/arXiv arXiv 2024

  27. [27]

    Ladac: Large language model-driven auto- designer for analog circuits,

    C. Liu, Y. Liu, Y. Du, and L. Du, “Ladac: Large language model-driven auto- designer for analog circuits, ”Authorea Preprints, 2024

    2024

  28. [28]

    Gpt-4o system card,

    A. Hurst, A. Lerer, A. P. Goucher, A. Perelman, A. Ramesh, A. Clark, A. Ostrow, A. Welihinda, A. Hayes, A. Radfordet al., “Gpt-4o system card, ”arXiv preprint arXiv:2410.21276, 2024

    Pith/arXiv arXiv 2024

  29. [29]

    A double-tail latch-type voltage sense amplifier with 18ps setup+ hold time,

    D. Schinkel, E. Mensink, E. Klumperink, E. Van Tuijl, and B. Nauta, “A double-tail latch-type voltage sense amplifier with 18ps setup+ hold time, ” in2007 IEEE international solid-state circuits conference. Digest of technical papers. IEEE, 2007, pp. 314–605

    2007

  30. [30]

    Deepseek-v3 technical report,

    A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruanet al., “Deepseek-v3 technical report, ”arXiv preprint arXiv:2412.19437, 2024

    Pith/arXiv arXiv 2024

  31. [31]

    Gemini 3 pro model card,

    Google DeepMind, “Gemini 3 pro model card, ” https://storage.googleapis.com/ deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf, 2026, accessed: 2026-04-12

    2026

  32. [32]

    A current- mode latch sense amplifier and a static power saving input buffer for low-power architecture,

    T. Kobayashi, K. Nogami, T. Shirotori, Y. Fujimoto, and O. Watanabe, “A current- mode latch sense amplifier and a static power saving input buffer for low-power architecture, ” in1992 symposium on VLSI circuits digest of technical papers. IEEE, 1992, pp. 28–29

    1992

  33. [33]

    A survey on multimodal large language models,

    S. Yin, C. Fu, S. Zhao, K. Li, X. Sun, T. Xu, and E. Chen, “A survey on multimodal large language models, ”National Science Review, vol. 11, no. 12, p. nwae403, 2024

    2024

  34. [34]

    Masala-chai: A large-scale spice netlist dataset for analog circuits by harnessing ai,

    J. Bhandari, V. Bhat, Y. He, H. Rahmani, S. Garg, and R. Karri, “Masala-chai: A large-scale spice netlist dataset for analog circuits by harnessing ai, ” 2025. [Online]. Available: https://arxiv.org/abs/2411.14299

    Pith/arXiv arXiv 2025

  35. [35]

    Review of dynamic comparators for adcs,

    Y. Sachdeva, N. Nehra, S. Bansal, and Garima, “Review of dynamic comparators for adcs, ” in2021 Second International Conference on Electronics and Sustainable Communication Systems (ICESC), 2021, pp. 83–90

    2021

  36. [36]

    Ultralytics yolo,

    G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics yolo, ” 2023, computer software. [Online]. Available: https://github.com/ultralytics/ultralytics

    2023

  37. [37]

    Amsnet: Netlist dataset for ams circuits,

    Z. Tao, Y. Shi, Y. Huo, R. Ye, Z. Li, L. Huang, C. Wu, N. Bai, Z. Yu, T.-J. Lin et al., “Amsnet: Netlist dataset for ams circuits, ” in2024 IEEE LLM Aided Design Workshop (LAD). IEEE, 2024, pp. 1–5

    2024

  38. [38]

    Design and comparative analysis of dynamic comparators for sar adc,

    N. A. Ansari, P. Jaiswal, M. Tyagi, and P. Mittal, “Design and comparative analysis of dynamic comparators for sar adc, ” inEmerging Electronics and Automation: Select Proceedings of E2A 2021. Springer, 2022, pp. 391–401

    2021

  39. [39]

    Razavi,Design of Analog CMOS Integrated Circuits

    B. Razavi,Design of Analog CMOS Integrated Circuits. USA: McGraw-Hill, Inc., 2000

    2000

  40. [40]

    Microelectronic circuits pp. 263-267,

    A. S. Sedra and K. C. Smith, “Microelectronic circuits pp. 263-267, ” 1987

    1987

  41. [41]

    ADC Performance Survey 1997-2026,

    B. Murmann, “ADC Performance Survey 1997-2026, ” [Online]. Available: https: //github.com/bmurmann/ADC-survey

    1997

  42. [42]

    A low-noise self-calibrating dynamic comparator for high-speed adcs,

    M. Miyahara, Y. Asada, D. Paik, and A. Matsuzawa, “A low-noise self-calibrating dynamic comparator for high-speed adcs, ” in2008 IEEE Asian Solid-State Circuits Conference. IEEE, 2008, pp. 269–272

    2008

  43. [43]

    All-mos charge redistribution analog-to-digital conversion techniques. i,

    J. L. McCreary and P. R. Gray, “All-mos charge redistribution analog-to-digital conversion techniques. i, ”IEEE Journal of Solid-State Circuits, vol. 10, no. 6, pp. 371–379, 1975

    1975

  44. [44]

    Analog design of asynchronous sar adc,

    M. Aldacher, “Analog design of asynchronous sar adc, ” https://github.com/ muhammadaldacher/Analog-Design-of-Asynchronous-SAR-ADC, 2020

    2020

  45. [45]

    A two-stage weighted capacitor network for d/aa/d conversion,

    Y. Yee, L. Terman, and L. Heller, “A two-stage weighted capacitor network for d/aa/d conversion, ”IEEE Journal of Solid-State Circuits, vol. 14, no. 4, pp. 778–781, 2003

    2003

  46. [46]

    Randomized switching sar (rs-sar) adc for power and em side-channel security,

    M. Ashok, E. V. Levine, and A. P. Chandrakasan, “Randomized switching sar (rs-sar) adc for power and em side-channel security, ”IEEE Solid-State Circuits Letters, vol. 5, pp. 247–250, 2022

    2022

  47. [47]

    Botorch: A framework for efficient monte-carlo bayesian optimization,

    M. Balandat, B. Karrer, D. Jiang, S. Daulton, B. Letham, A. G. Wilson, and E. Bak- shy, “Botorch: A framework for efficient monte-carlo bayesian optimization, ” Advances in neural information processing systems, vol. 33, pp. 21 524–21 538, 2020

    2020

  48. [48]

    Differentiable expected hypervolume improvement for parallel multi-objective bayesian optimization,

    S. Daulton, M. Balandat, and E. Bakshy, “Differentiable expected hypervolume improvement for parallel multi-objective bayesian optimization, ”Advances in neural information processing systems, vol. 33, pp. 9851–9864, 2020

    2020

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.