acceptodds
Withdrawn from ICLR 2027

SMS-Sizing: An Agentic Search Framework for Feasible Analog Circuit Sizing and Archive Expansion

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Analog circuit sizing selects transistor dimensions, bias currents, capacitor values, and other device parameters for a fixed topology so that simulated performance satisfies all specifications simultaneously. Each candidate design requires a circuit simulation to determine whether it meets all specifications. Our goal is to find multiple feasible designs with distinct parameter configurations within a limited simulation budget. Some existing sizing methods simplify the search by restricting parameter ranges or discretizing design variables. These restrictions can exclude feasible designs and limit the diversity of the solutions found. We propose Sensitivity-Manifold Search (\method). Compared with prior work, our study makes three contributions. First, we formulate the task as expanding one verified feasible design into a set of distinct alternatives within the original parameter space, and define metrics for feasible yield and design diversity. Second, we develop a search framework that generates candidates from verified designs, uses surrogate predictions to guide candidate selection and simulation budget allocation, and checks region updates against simulation evidence. Third, we use baseline comparisons and ablation studies to assess performance and examine how different components affect feasible yield and design diversity. All designs added to the archive are verified by circuit simulation. Across six amplifier topologies under eight simultaneous constraints, starting from a single feasible seed and using a budget of 100 simulations, \method reliably expands the design archive. Anchored candidates achieve a conversion rate, while uniform random search finds no feasible designs in our experiments. In head-to-head comparisons, \method maintains up to greater spatial reach and larger worst-case constraint margins, preserving candidate diversity across spatial separation resolutions.

Primary area:
other topics in machine learning (i.e., none of the above)
Keywords:
constrained optimization, surrogate-assisted optimization, active learning, uncertainty quantification, sensitivity analysis, multi-objective optimization, analog circuit design

Reviews

Among:
Shape:
1234this paper, Bottom 10%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.67 over 3 reviews, above 10% of the others.
Reviewer_hwV9OpenReview ↗
Rating
2
Confidence
3
Informativeness
1
Presentation
2
Soundness
2
Read review
Paper summary

The paper presents SMS-Sizing, a method that generates multiple feasible alternatives from a single design that has already been verified through simulation. SMS-Sizing explores alternatives by perturbing verified designs. It uses a pre-trained surrogate model to identify the next designs to simulate and to verify whether the proposed modifications to the search space are supported by available evidence. The experimental validation comprises 6 amplifier topologies assessing SMS-Sizing's ability to find solutions far from the initial design.

Critical strengths

S1. Feasibility is established by circuit simulation.
Surrogate predictions help identify candidate designs, but a design becomes a part of the set of feasible solutions only when circuit simulation confirms it satisfies all 8 criteria. Candidates predicted to be infeasible by surrogates remain eligible for simulation (Section 3.5). Empirical results back this decision. According to Table 24, screening of candidates based on average surrogate predictions would exclude ~25% (1289/5052) of simulator-proven feasible solutions.

S2. SMS-Sizing finds more widely different solutions.
AnchorOnly produces almost the same number of feasible designs as SMS-M (26.6 versus 26.2). Its solutions, however, remain closer to the initial design, and are less separated from one another, and have lower best worst-case constraint margins (Table 14). These results show that the complete method offers benefits beyond finding many nearby feasible solutions.

Critical weaknesses

W1. Contributions of individual elements are not well justified.
In the paper, improved search performance is attributed to surrogate scoring. However, the AnchorOnly model removes a few elements at once, including surrogate predictions, sensitivity scores, predicted feasibility masks, the served controller, gate choices, and box contractions (Appendix D.8). Moreover, selecting coordinates using sensitivity scores gives essentially the same feasible hit rate as uniform selection (19.63% versus 19.73%, Appendix D.10). Additionally, the separate NoContraction and NoGate controls retain similar distances from the initial design (Table 2). These results support the complete method, but leave the importance of its sensitivity analysis and evidence-gating mechanisms unclear. Appendix E also acknowledges that "the reported experiments do not compare against a quality-diversity baseline", leaving an important comparison missing for a contribution centered on producing multiple solutions.

W2. The necessity of multiple solutions is not sufficiently demonstrated.
The motivation for this work is that downstream layout-based decisions would benefit from having a collection of sizing vectors. However, the paper does not identify any specific downstream decisions or tasks that would benefit from an archive of solutions rather than selecting a single carefully optimized solution. The experiments fail to show any reductions in design effort or improvement in the results obtained from subsequent design phases. Furthermore, the computational cost is not clear. The surrogate model training uses 960000 records, and the 100 calls online budget fails to include pre-training, candidate pool inference, and control function calls (Section 4.1, Appendix C). Without a proper comparison of elapsed time and computational cost, it is not possible to prove whether the extra solutions justify the effort.

Questions for authors

Q1. Clarification. Contribution of the individual components.
Are there any experiments where only the surrogate score is varied? AnchorOnly achieves a similar feasible count, while NoContraction achieves a similar distance from the initial design. What is the evidence of further improvement due to the sensitivity analysis and evidence-gated updates?

Q2. Clarification. Comparison with other approaches aiming at finding multiple solutions.
What is the evidence of an advantage over other archive-based approaches when starting with the same initial feasible solution, simulator budget, and separation condition? As no such comparison is provided in Appendix E, how is the contribution evaluated independently of the drawbacks of those baselines tailored to other search purposes?

Q3. Clarification. Need for the diversity measures.
Which downstream analog-design decision benefits from the returned set of solutions rather than one carefully optimized design? On the simple tier, Figure 3 gives EEsizer a larger share of retained solutions when the minimum separation threshold is increased to 0.02 and 0.05. How should this result be weighed against SMS-M’s greater distance from the initial design ("Reach" in Table 1), and which measure better reflects practical usefulness?

Q4. Clarification. Computational effort.
What is the computational effort and time spent on SMS-Sizing (plus training of the offline surrogate model) relative to the baselines? Moreover, the up-front effort might be justified by the ability to reuse the same surrogate across several sizing sessions. At what number of such runs (and under which conditions) does this become worthwhile?

Reviewer_b7j4OpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

This paper proposed SMS-SIZING, which provides a sample-efficient framework for analog circuit sizing: expanding a single simulator-verified seed into an extensive feasible archive under tight simulation budgets.

Critical strengths

No to much is found.

Critical weaknesses

The paper evaluates only six Op-Amp topologies but makes broader claims about analog circuits that are not sufficiently supported by this evaluation.

Furthermore, although the paper describes the method as manifold-based, the authors do not clearly define the manifold structure or explain how the method exploits it.

Also, the paper is difficult to follow because it interleaves the proposed method with extensive descriptions of experimental variants, historical configurations, and audit procedures.

Questions for authors

The paper uses a surrogate model for prediction, but its architecture is not clearly described. Please provide the network architecture, input data format, and circuit topology representation. The paper also mentions 16 training circuit topologies without sufficiently describing them. Please clarify which topologies are included and how the training data are generated.

The surrogate dataset contains 60,000 samples per circuit topology, but it is unclear whether the final discovered designs overlap with this dataset. Please clarify whether these designs are previously unseen parameter configurations.

Greater separation between designs in parameter space does not necessarily make them more useful for circuit design. Please provide a downstream evaluation to substantiate the practical value of the expanded feasible archive. For example, evaluate whether the expanded archive provides more feasible alternatives under additional specs or previously unevaluated PVT conditions, compared with the initial seed and baseline archives.

Reviewer_LqBiOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

This paper proposes SMS-Sizing, a framework for expanding a single simulator-verified feasible analog circuit design into a diverse archive under a limited simulation budget. It combines sparse perturbations of feasible designs with surrogate-guided candidate selection and evidence-gated search-region updates. The method is evaluated on six op-amp topologies under eight simultaneous constraints, with comparisons against LEDRO, EEsizer, and AutoSizer and several ablation studies.

Critical strengths

The paper explicitly formulates feasible-archive expansion as an alternative to conventional single-design sizing. This is a relevant problem when designers need multiple parameter configurations for downstream implementation.

The ablation studies provide useful evidence about the roles of anchor perturbation and surrogate-guided selection.

Critical weaknesses

The method assumes that a simulator-verified feasible seed is already available, but does not address the cost of finding such a seed. More importantly, it searches for additional feasible designs without optimizing their actual performance objectives. The resulting archive may contain many distinct parameter vectors, but it remains unclear how a designer should select the final design or whether these alternatives improve downstream implementation. This limits the practical significance of the proposed formulation.

The contribution of sensitivity analysis is not clearly established. The coordinate-influence score is calculated as the mean absolute Pearson correlation between each design parameter and 11 predicted circuit responses. While this provides a simple measure of association, it does not directly capture local sensitivity or interactions between parameters. More importantly, the ablation study in Appendix D.10 shows that selecting perturbation coordinates based on sensitivity scores achieves a hit rate of 19.63%, nearly identical to the 19.73% obtained with uniform selection. This result does not demonstrate a clear benefit from sensitivity-guided coordinate selection. Although sensitivity scores are also used for search-region updates and candidate screening, their contribution to these components is not sufficiently isolated. It therefore remains unclear whether sensitivity analysis provides a meaningful improvement beyond anchor-based candidate generation and surrogate-guided selection.

The comparison with LEDRO, EEsizer, and AutoSizer does not fully match the stated objective. These methods were developed for analog sizing rather than specifically for expanding a diverse feasible archive. Evaluating their original search strategies primarily through archive size, spatial reach, and constraint margins may disadvantage methods that do not explicitly target these metrics. A comparison against a quality-diversity or constrained multi-solution search method would provide stronger evidence for the claimed advantage.

The normalized constraint-margin formula (m(x)=\min_k[-c_k(x)/|t_k|]), as well as the normalized quantities used in candidate screening, is undefined when (t_k=0). The paper does not specify how zero or near-zero thresholds are handled. A well-defined normalization scheme is needed to make the formulation applicable to constraints for which the threshold is zero.

Questions for authors

How is the initial feasible seed obtained in practice, and how would the method select a final design from the expanded archive? Can the authors provide evidence that the additional designs improve a meaningful downstream outcome?

What evidence shows that sensitivity analysis actually improves the method, rather than the gains coming mainly from surrogate-based candidate selection? Can the authors isolate this component more directly and compare against a multi-solution baseline under the same objective and simulation budget?

How does the implementation handle (t_k=0) or near-zero thresholds in the normalized margin and deficit calculations?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.