Research map / Architectures and efficiency / RNNs and normalization

FALCON: An ML Framework for Fully Automated Layout-Constrained Analog Circuit Design

NeurIPS 2025 · Architectures and efficiency › RNNs and normalization · approach: Differentiable optimization layers · 2 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.

FALCON combines an MLP topology classifier, an edge-centric GNN forward surrogate, and gradient-based layout-aware parameter inference to synthesize mm-wave analog circuits from specifications, trained on a Cadence-simulated dataset of over 1M circuits across 20 topologies.

Problem

Automating analog/RF mm-wave circuit design from performance specifications across topology selection, parameter inference, and layout feasibility, which prior ML methods handle only in isolated sub-tasks or with symbolic simulators.

Approach

Frame inverse design as gradient-based parameter inference through a learned, topology-generalizable edge-centric GNN forward model, with a differentiable analytical area penalty for passive components and a performance-driven MLP for topology selection.

Key results

• Topology classifier reaches 99.57% test accuracy and 99.30% macro F1 across 20 topologies.
• Forward GNN achieves average R² of 0.972 on 16 metrics; fine-tuning only the output head on held-out RVCO reduces relative error from 30.4% zero-shot to 0.9%.
• Gradient-based inverse design achieves a 78.5% success rate (Cadence-simulated within 20% of target) and 17.7% mean relative error on converged designs, under 1 second per instance on CPU.

Measured effects reported in the paper

Surprising findings

• Zero-shot GNN error on the unseen RVCO topology is 30.4%, far above the in-distribution error.

Limitations

• Success rate is 78.5% and mean relative error 17.7% on converged designs, so inverse designs are not always accurate.
• Dataset is limited to a single 30 GHz operating frequency and a fixed topology library of 20 circuits.
• Layout model is analytical and non-learned, excluding active devices and full EM/parasitic effects.

Experimental setup

Datasets: FALCON mm-wave circuit dataset (Cadence Spectre, 45nm PDK)

Benchmarks: FALCON mm-wave circuit dataset test split

Metrics: accuracy balanced accuracy macro F1 R² RMSE MAE mean relative error success rate

Baselines: CktGNN LaMAGIC AnalogCoder GCN-RL BO-SPGP AICircuit AnalogGym AutoCkt

Builds on: Adam t-SNE

Closest papers by meaning

• LaMAGIC: Language-Model-based Topology Generation for Analog Integrated Circuits (ICML 2024 · similarity 0.89)
• LaMAGIC2: Advanced Circuit Formulations for Language Model-Based Analog Topology Generation (ICML 2025 · similarity 0.88)
• AUTOCIRCUIT-RL: Reinforcement Learning-Driven LLM for Automated Circuit Topology Generation (ICML 2025 · similarity 0.88)
• AnalogGenie: A Generative Engine for Automatic Discovery of Analog Circuit Topologies (ICLR 2025 · similarity 0.87)
• AnalogToBi: Device-Level Analog Circuit Topology Generation via Bipartite Graph and Grammar Guided Decoding (NeurIPS 2026 · similarity 0.87)
• AnalogCoder: Analog Circuit Design via Training-Free Code Generation (AAAI 2025 · similarity 0.85)