Research map / Graph learning / Graph self-supervised learning

AnalogToBi: Device-Level Analog Circuit Topology Generation via Bipartite Graph and Grammar Guided Decoding

NeurIPS 2026 · Graph learning › Graph self-supervised learning · approach: Diffusion-based graph generation · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes AnalogToBi, a GPT-style device-level analog topology generator that combines circuit-type conditioning, device renaming augmentation, a bipartite device-net graph representation, and grammar-guided decoding to reach high validity and novelty without HITL training.

Problem

Device-level analog circuit topology generation with transformers suffers from low electrical validity without human-in-the-loop training and memorization of training circuits caused by sequence-based pin-level representations.

Approach

Replace the device-pin token representation with a bipartite device-net graph serialized as a sequence, condition on circuit-type tokens, augment with random device renaming to break index memorization, and enforce pin/net structural rules via a state-machine grammar that masks invalid tokens during decoding.

Key results

• AnalogToBi reaches 97.8% validity, 92.1% novelty, and 89.9% valid-and-novel without HITL, versus 75.6% / 50.4% / 26.0% for reproduced AnalogGenie.
• Exact-match memorization falls from 39.6% (AnalogGenie*) to 0% for AnalogToBi, with GED 226.8 vs 209.5.
• Grammar-guided decoding raises validity of the bipartite representation by 8.2 points (53.6% to 61.8%) before circuit-type tokens are added; circuit-type tokens then lift it to 97.8%.

Surprising findings

• Device renaming augmentation improves novelty for the device-pin representation but drops validity from 86.1% to 55.6%, showing validity and novelty are entangled under memorization.
• Without circuit-type tokens, the bipartite representation has lower validity (53.6%) than the device-pin representation (75.6%).

Limitations

• Validity is measured only as SPICE-convertibility, not functional correctness; simulation case studies are manually selected.
• Novelty evaluation for prior AnalogGenie results was not directly comparable; baselines reproduced by the authors.
• Circuit-type labels rely on a manual 15-category grouping of the dataset.

Experimental setup

Datasets: AnalogGenie dataset (2,165 SPICE netlists)

Benchmarks: AnalogGenie dataset held-out circuits

Metrics: validity novelty valid & novel exact match graph edit distance type accuracy

Baselines: LaMAGIC AnalogGenie AnalogGenie-Lite AnalogGenie*

Builds on: AnalogGenie LaMAGIC grammar-constrained decoding

Closest papers by meaning

• LaMAGIC2: Advanced Circuit Formulations for Language Model-Based Analog Topology Generation (ICML 2025 · similarity 0.91)
• AnalogGenie: A Generative Engine for Automatic Discovery of Analog Circuit Topologies (ICLR 2025 · similarity 0.91)
• AnalogGenie-Lite: Enhancing Scalability and Precision in Circuit Topology Discovery through Lightweight Graph Modeling (ICML 2025 · similarity 0.91)
• LaMAGIC: Language-Model-based Topology Generation for Analog Integrated Circuits (ICML 2024 · similarity 0.91)
• AnalogCoder: Analog Circuit Design via Training-Free Code Generation (AAAI 2025 · similarity 0.88)
• FALCON: An ML Framework for Fully Automated Layout-Constrained Analog Circuit Design (NeurIPS 2025 · similarity 0.87)