AMSchem: Automated Schematic Visualization of AMS Circuits via Verifiable Agentic Generation
Authors: Yichen Shi, Zhiyuan Wang, Jianing Cui, Yao Lai, Jinting Zhang, Zhuofu Tao, Hongyang Wang, Zhiping Yu, Ting Jung Lin, Lei He
Organizations: Shanghai Jiao Tong University, Shanghai, China · Eastern Institute of Technology, Ningbo, China · University of Washington, USA · University of St Andrews, St Andrews, U.K. · University of Cambridge, Cambridge, U.K. · Tianjin University, Tianjin, China · University of California, Los Angeles, CA, USA · School of Computer Science and Technology, Shijiazhuang Tiedao University, Shijiazhuang, China · Tsinghua University, Beijing, China · Ningbo Institute of Digital Twin, Eastern Institute of Technology, Ningbo, China
For decades, the netlist-to-schematic gap has remained a persistent challenge in design automation, especially for analog and mixed-signal (AMS) circuits. Netlists encode connectivity, not visual presentation, making their manual conversion into readable schematics time-consuming and dependent on expert conventions. The same gap leaves LLM-based design tools without structured visual feedback, forcing them to iterate on circuits they cannot see. We present AMSchem, a training-free, verifier-guided agentic framework that automatically generates human-readable schematics from AMS netlists. AMSchem separates circuit-aware visual planning (DrawingIntent) from executable schematic geometry (SchematicIR), and applies localized visual feedback with bounded, revalidated repair patches. Deterministic validators enforce device preservation, pin-net consistency, and geometric correctness throughout. For evaluation, we construct AMSchem-Bench comprising 64 circuits of 12 types. On the 40-circuit schematic subset, AMSchem with gpt-5.5 achieves 85.0% validity (vs. 32.5% direct generation). When integrated as a visual feedback module in LLM-based analog design, AMSchem resolves all 13 previously failed cases from AnalogCoder benchmark across five models, substantially improving the success rate. Our dataset and benchmark are available at https://www.amschem.com/benchmark
Figures & tables
Fig. 1: AMSchem in LLM-driven AMS design flow: AMSchem converts netlists into readable schematics, providing visual feedback for design review and optimization by agents or humans.
Fig. 2: AMSchem overview: (1) Stage 1: the input netlist is enriched with matched building blocks and drawing references to construct DrawingIntent , with constraint-driven refinement (L1) of device preservation, relative placement, and symmetry. (2) Stage 2: DrawingIntent is translated into executable SchematicIR , with structured validation and refinement of connectivity (L2), Manhattan routing, and design-rule compliance. (3) Stage 3: the rendered schematic undergoes bounded visual refinement (L3) that diagnoses readability, overlap, and local connectivity, producing the final output artifacts.
#
Building block
1
Two-transistor NMOS current mirror
2
Two-branch NMOS cascode current mirror
3
Two-transistor PMOS current mirror
4
Two-branch PMOS cascode current mirror
5
Single NMOS common-source device
6
Single PMOS common-source device
TABLE I: Building-block library used by AMSchem.
Fig. 3: Object-level grounding in SchematicIR. Colored callouts link schematic elements to structured placement, orientation, geometry, and net bindings for deterministic validation and localized repair.
Fig. 4: L3 refinement: localized visual feedback over circuit-aware risk regions. Regions proposed by RRP (a) are cropped and diagnosed by an object-grounded visual model (b); actionable findings guide a bounded SchematicIR revision (c) that is revalidated and rerendered.
Fig. 5: AMSchem-Bench statistics.
Method
Valid (%) ↑
Scross↑
Sbend↑
Srel↑
Sorient↑
MOS ↑
E
M
H
E
M
H
E
M
H
E
M
H
E
M
H
Weave [ 19 ]
69.2
60.0
28.6
0.3333
0.3007
0.1905
0.2200
0.1248
0.0347
0.2374
0.1741
0.0611
0.3370
0.2237
0.0986
1.150
gpt-5.5
Direct-IR
84.6
10.0
0.0
0.2051
0.0800
0.0000
0.3968
0.0597
0.0000
0.5067
0.0667
0.0000
0.6316
0.0700
0.0000
–
Direct-IR + Few-shot
69.2
5.0
0.0
0.2308
0.0200
0.0000
0.3475
0.0087
0.0000
0.4106
0.0282
0.0000
0.4974
0.0273
0.0000
–
AMSchem
92.3
95.0
42.9
0.4308
0.3732
0.1111
0.3721
0.3476
0.1911
0.6557
0.6112
0.3031
0.7496
0.6557
0.3070
3.075
TABLE II: Performance across benchmark difficulty tiers. E, M, and H denote Easy, Medium, and Hard, respectively. MOS is pooled across tiers. Shaded rows indicate AMSchem.
Fig. 6: Schematic comparison on a bandgap reference circuit from AMSchem-Bench: (a) Ground truth; (b) AMSchem; (c) Weave [ 19 ] .
DI
IR
VF
Valid (%) ↑
Scross↑
Sbend↑
Srel↑
Roverlap↓
Sorient↑
×
×
×
32.5
0.1067
0.1588
0.1980
5.550
0.2403
✓
×
×
32.5
0.2083
0.1674
0.2533
1.763
0.2488
✓
✓
×
80.0
0.3344
0.3339
0.5305
0.900
0.5916
✓
✓
✓
85.0
0.3461
0.3282
0.5717
0.475
0.6252
TABLE III: Incremental analysis of AMSchem. DI, IR, and VF denote DrawingIntent, SchematicIR validation and refinement, and visual feedback, respectively.
Fig. 7: Mean per-case token consumption by pipeline stage and token type.
BBL Match
BBL Ref.
Refine
Valid (%) ↑
Scross↑
Sbend↑
Srel↑
Sorient↑
×
×
×
72.5
0.3212
0.2924
0.4787
0.5103
✓
×
×
75.0
0.3141
0.3008
0.5099
0.5350
✓
✓
×
77.5
0.3436
0.3030
0.5259
0.5721
✓
✓
✓
85.0
0.3461
0.3282
0.5717
0.6252
TABLE IV: Component ablation of DrawingIntent.
Risk region
Precision (%) ↑
Recall (%) ↑
F1 score (%) ↑
Proposal share (%)
RBBL
69.9
21.7
33.1
25.0
Rpatch
76.0
8.1
14.6
8.6
Rhighdeg
67.0
55.3
60.6
66.4
TABLE V: Comparison of RRP region types. Proposal share denotes the fraction of proposals contributed by each type.
Fig. 8: Effect of AMSchem-aided visual feedback in the AnalogCoder refinement loop across five models.
Fig. 9: Schematic-guided repair of a Gilbert-cell mixer generated by MiniMax-M3 (T19). The task and failed netlist are shown on the left. The schematic on the right identifies the incorrect output connections of the LO switching quad: red dashed segments mark removed connections, and teal segments show the drain reconnections of M4 and M6.
Fig. 10: Gilbert-cell mixer response before (a,c) and after (b,d) schematic-guided repair. (a,b) Differential output waveforms; (c,d) FFT spectra of uniformly resampled 5–20-ms segments (24 kSa/s sampling, NFFT=360 , rectangular window). Excitation: 1.0-kHz RF, 1.2-kHz LO. The plotted traces follow the original evaluator’s automatic LO bias-selection procedure. Additional comparisons at the same LO common-mode voltage show that the original circuit fails the mixing requirement and the repaired circuit satisfies it. Dashed line: 1-mV product threshold.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Fig. 11: Blinded MOS evaluation platform: (a) the landing page and (b) the candidate-scoring interface.
Fig. 12: Qualitative comparison of AMSchem, EESchematic, and Weave on two AMSNet circuits. The left column shows the comparator, and the right column shows the bandgap reference circuit. Within each column, the results are ordered from top to bottom as AMSchem, EESchematic, and Weave. All methods use the same input netlist and rendering settings for each circuit. Red circles in the EESchematic and Weave outputs mark selected topological and visual errors identified during manual inspection.
Analog and mixed-signal (AMS) circuit design remains heavily reliant on expert knowledge. While recent AI-driven automation tools can generate candidate topologies, they critically depend on manually curated datasets with functional and performance annotations -- a requirement that current large language models (LLMs) and vision models cannot automate. Existing approaches still require domain experts to manually interpret circuit functionality. We present AMSnet-q, a fully automated, unsupervised pipeline that eliminates human-in-the-loop annotation by converting schematic images directly into a labeled AMS circuit database. Unlike prior work that stops at netlist extraction, our framework automates the complete verification loop: it performs schematic-to-netlist conversion, topology-aware testbench generation, and simulation-based sizing validation to objectively determine circuit functionality. Validated in 28 nm technology, AMSnet-q processed 739 schematics from the AMSnet 1.0 dataset, automatically constructing a repository of 4 circuit classes, 105 distinct topologies, and 89,789 labeled device configurations. By decoupling human effort from dataset volume and reducing the workload to a one-time testbench template per circuit class, AMSnet-q enables scalable, objective, and fully automated AMS database construction.
Ze Zhang, Junzhuo Zhou, Yichen Shi +6
1Southern University of Science and Technology · University of California Los Angeles · 4Eastern Institute of Technology Ningbo +1
Recent advances in Artificial Intelligence (AI) have revolutionized Electronic Design Automation (EDA), particularly through Large Language Models (LLMs) for circuit design tasks. However, their application to analog and mixed-signal domains remains limited by the lack of machine-readable representations of existing circuit design knowledge. Circuit schematic images found in research manuscripts, textbooks, and websites constitute a vast repository of validated designs; however, these visual representations cannot be directly processed by EDA tools. Converting them into machine-readable netlists is essential for enabling simulation, verification, and building comprehensive databases for AI-based models. Current conversion methods lack generalization across both Integrated Circuit (IC) and Printed Circuit Board (PCB) level schematics. Moreover, they struggle with component recognition and connectivity inference, and fail to distinguish between connected junctions and crossing wires. In this paper, we propose SINA, an open-source circuit schematic image-to-netlist generator. SINA is a fully automated pipeline that integrates deep learning for robust component detection, connected-component labeling for accurate connectivity inference, Optical Character Recognition (OCR) for component reference designator extraction, and a Vision-Language Model (VLM) for reliable reference designator assignment. SINA handles both IC- and PCB-level schematics and incorporates dedicated crossing-wires detection to differentiate wire intersections from connections. We validate the correctness of the generated netlists using graph isomorphism techniques. Our experiments demonstrate an overall netlist generation accuracy of 96.67%, which is 2.72x higher compared to state-of-the-art approaches.
The Department of Electrical and Computer Engineering, University of Utah, Salt Lake City, UT, 48112 USA. · The Department of Electrical, Computer & Energy Engineering, University of Colorado Boulder, Boulder, CO 80309, USA.
Printed circuit board (PCB) schematic design defines nearly all electronic hardware, but it remains manual and expertise-intensive. While generative AI has advanced digital and analog IC design, PCB schematic generation from natural-language intent is largely unexplored. This paper presents SchGen, the first large language model that generates editable PCB schematics from natural-language requests. The key challenge lies in the lack of an LLM-suited representation and a large-scale dataset. Current schematic formats are dominated by verbose, tool-specific syntax and geometry-heavy descriptions, making them difficult to generate reliably. We introduce a semantically grounded code representation that encodes schematic editing primitives with relative placement and pin-name-based wiring, transforming a geometry-driven generation problem into a semantics-driven matching task amenable to LLMs. We further construct a large-scale dataset of PCB schematics paired with user prompts via a human-agent collaborative pipeline that converts open-source hardware designs into our representation. Experiments show that SchGen significantly outperforms alternative representations and even larger general-purpose LLMs on wire connectivity accuracy and functional correctness. Our results highlight the critical role of representation design in enabling generative models for complex hardware design tasks.
Qinpei Luo, Ruichun Ma, Xinyu Zhang +1
University of California, San Diego · Microsoft Research Asia · The University of Texas at Austin