netlist parsing

Machine learning–generated circuit netlists suffer from poor readability and debuggability, hindering their practical adoption in hardware design. To address this, we propose the first large language model (LLM) specialized for circuit design, enabling end-to-end automatic conversion of netlists into LTSpice (.asc) schematics and CircuitTikZ (LaTeX) diagrams. Methodologically, we pioneer the use of LLMs for joint semantic and topological mapping from netlists to schematics; introduce circuit-domain knowledge–enhanced instruction tuning, structure-aware tokenization, and format-constrained decoding; and incorporate compiler feedback via multi-stage supervised fine-tuning. Experiments demonstrate that our model achieves a 93% LaTeX compilation success rate—significantly surpassing state-of-the-art general-purpose and code-specific LLMs (26%)—and attains structural similarity three times higher than human-designed reference schematics, markedly improving topological fidelity and engineering utility.

Existing methods for converting circuit schematic images into EDA-processable netlists suffer from poor generalizability across IC and PCB domains, inaccurate component recognition, ambiguous connectivity inference, and frequent misinterpretation of crossing wires. This work proposes the first fully automated parsing pipeline that integrates deep learning, connected-component labeling, OCR, and vision-language models to precisely extract components, labels, and topological connections. A dedicated crossing-wire detection module is introduced to distinguish genuine electrical junctions from mere graphical overlaps. The approach unifies processing for both IC and PCB schematics and achieves a netlist accuracy of 96.67% on real-world datasets—2.72 times higher than the current state-of-the-art—while ensuring functional equivalence through graph isomorphism verification.

Existing image-to-netlist conversion methods for analog schematics suffer from limited compatibility with diverse schematic styles and insufficient coverage of analog component types. To address these limitations, this paper proposes Image2Net—the first end-to-end framework supporting multi-style, multi-type analog components. Methodologically, it integrates deep learning–driven image recognition and structural parsing, rule-guided topological inference, and semantic consistency verification. We also introduce the first open-source schematic dataset featuring diversity and balanced complexity. Innovatively, we propose Netlist Edit Distance (NED) as a precision metric for quantitative evaluation. Experimental results demonstrate that Image2Net achieves an 80.77% conversion success rate on benchmark tests—surpassing state-of-the-art methods by 34.62–45.19%. Its average NED of 0.116 outperforms existing approaches by 62.1–69.6%, confirming substantial gains in structural and semantic fidelity.

Existing circuit representation learning methods rely heavily on graph models tailored for simple And-Inverter Graphs (AIGs), limiting their capacity to capture complex gate-level semantics; while large language models (LLMs) excel at functional understanding, they lack native awareness of netlist structural topology. This work introduces the first netlist foundation model for integrated circuit design. We propose a novel Text-Annotated Graph (TAG) representation that unifies logical expressions and physical attributes as gate-level textual features. Furthermore, we design an RTL-to-layout co-aligned multi-task self-supervised pretraining paradigm, integrating an LLM-based text encoder with a graph Transformer to jointly optimize semantic comprehension, structural learning, and cross-stage alignment. Evaluated on four functional and physical IC design tasks—including logic optimization, timing prediction, placement, and routing—the model consistently outperforms task-specific baselines and state-of-the-art AIG encoders, demonstrating superior representational generality and cross-task transferability.

This work addresses the challenges of deploying natural language–to–circuit diagram generation, which often suffers from hallucinated details, violations of electrical constraints, and non-machine-readable outputs. The authors propose a multi-agent LLM-assisted design framework that translates natural language instructions into structured, verifiable circuit schematics through a five-stage pipeline: component identification, pin retrieval, expert reasoning, JSON synthesis, and SVG visualization. Integrating an embedded component knowledge base, chain-of-thought reasoning, and the CircuitJSON format, the method introduces a novel dual-metric circuit verification (DMCV) mechanism to guarantee both electrical and topological correctness. Experimental evaluation on 100 embedded-system prompts demonstrates strong performance in microcontroller-related designs, enabling reliable translation from natural language to high-fidelity, deployable hardware specifications.

Existing methods for converting SPICE netlists into human-readable schematics lack guarantees of connectivity correctness and suffer from significantly degraded accuracy as circuit complexity increases. This work proposes a deterministic conversion approach based on Sugiyama’s layered graph layout, integrating an embedded pinout database of 5,093 LTspice symbols and a round-trip connectivity verification mechanism. For the first time, this method enables netlist-to-schematic translation with a binary correctness certificate, ensuring network-by-network equivalence between the generated schematic and the original netlist. Implemented as a client-side, single-file, dependency-free architecture, the approach achieves 100% compilation success and connectivity verification pass rates on the Circuits-LTSpice benchmark and demonstrates an 88.4% verification pass rate on Analog Devices’ official dataset of 3,460 circuits.

This work addresses the longstanding challenge in PCB schematic design, which heavily relies on manual effort and lacks effective methods for automatically generating editable circuit diagrams from natural language. We propose SchGen, the first large language model tailored for this task, introducing a novel semantic-anchored code representation that reframes the geometry-driven generation problem as a semantic matching task. To support training, we construct a large-scale paired dataset linking natural language descriptions to schematics. SchGen integrates pin-name-aware routing with relative component layout to enable end-to-end generation of editable schematics from text. Experimental results demonstrate that SchGen significantly outperforms existing representation schemes and even larger general-purpose language models in terms of wiring accuracy and functional correctness, highlighting the critical role of domain-specific semantic representations in complex hardware generation tasks.

This work addresses the challenges in automated analog circuit design stemming from the scarcity of natural language–to–netlist mapping data and the inability of generic tokenizers to effectively model circuit topology. To overcome these limitations, the authors propose CircuitFormer, which first introduces the largest existing paired dataset of analog circuits and natural language descriptions. They then design a circuit-structure-aware tokenizer, termed Circuit Tokenizer (CKT), that reduces vocabulary complexity from O(n_max) to O(1), substantially compressing input sequence length. Built upon a Transformer encoder–decoder architecture, CircuitFormer enables end-to-end generation of circuit netlists from natural language specifications. Experimental results demonstrate that CircuitFormer achieves 100% syntactic correctness and 83% functional success across all major circuit categories, outperforming open-source large language model baselines by 10%–14% in functionality while using 240× fewer parameters.

This work addresses the ongoing challenge of automatically translating natural language specifications into editable printed circuit board (PCB) schematics for embedded and IoT development. It presents the first end-to-end approach that leverages tool-augmented large language model reasoning, integrating component library retrieval, datasheet knowledge extraction, execution validation, and structural-semantic verification to generate KiCad-compliant schematics. The system supports iterative refinement through an interactive web interface and achieves a pass@1 rate of 0.90 and a pass@5 rate of 1.00 across 20 embedded schematic generation tasks. This method efficiently produces high-quality initial drafts suitable for early-stage prototype review, substantially advancing the state of hardware design automation.

This work addresses the inefficiency and error-proneness of manual symbol and footprint creation in traditional PCB design by proposing SFgen, a novel pipeline that leverages agent-driven multimodal large language models for automated recognition and generation of electronic component symbols and footprints. The approach introduces SFnet, an extensible component library that has already integrated 1,000 components and continues to expand. Experimental results demonstrate that SFgen achieves 86% accuracy in symbol generation and 80% accuracy in footprint generation, significantly enhancing the automation level of PCB design and the efficiency of component library construction.