Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated Physical Design Watermarking Leveraging Graph Neural Networks

Ruisi Zhang, Rachel Selina Rajarathnam, David Z. Pan, Farinaz Koushanfar

TL;DR

AutoMarks addresses IP protection for IC layouts by introducing a graph neural network–driven watermarking framework that automates the region search, watermark insertion, and extraction. By representing layouts as graphs with physical and semantic features, the method learns to predict fidelity loss from watermarking and selects watermark regions that preserve layout quality while enabling ownership verification. The approach demonstrates transferability to unseen designs and resilience against removal and forging attacks, with experimental evidence on ISPD'15 and ISPD'19 benchmarks showing reduced search time and 100% watermark extraction rate. This enables scalable, secure protection of physical design IP in contemporary IC design pipelines.

Abstract

This paper presents AutoMarks, an automated and transferable watermarking framework that leverages graph neural networks to reduce the watermark search overheads during the placement stage. AutoMarks's novel automated watermark search is accomplished by (i) constructing novel graph and node features with physical, semantic, and design constraint-aware representation; (ii) designing a data-efficient sampling strategy for watermarking fidelity label collection; and (iii) leveraging a graph neural network to learn the connectivity between cells and predict the watermarking fidelity on unseen layouts. Extensive evaluations on ISPD'15 and ISPD'19 benchmarks demonstrate that our proposed automated methodology: (i) is capable of finding quality-preserving watermarks in a short time; and (ii) is transferable across various designs, i.e., AutoMarks trained on one layout is generalizable to other benchmark circuits. AutoMarks is also resilient against potential watermark removal and forging attacks

Automated Physical Design Watermarking Leveraging Graph Neural Networks

TL;DR

AutoMarks addresses IP protection for IC layouts by introducing a graph neural network–driven watermarking framework that automates the region search, watermark insertion, and extraction. By representing layouts as graphs with physical and semantic features, the method learns to predict fidelity loss from watermarking and selects watermark regions that preserve layout quality while enabling ownership verification. The approach demonstrates transferability to unseen designs and resilience against removal and forging attacks, with experimental evidence on ISPD'15 and ISPD'19 benchmarks showing reduced search time and 100% watermark extraction rate. This enables scalable, secure protection of physical design IP in contemporary IC design pipelines.

Abstract

This paper presents AutoMarks, an automated and transferable watermarking framework that leverages graph neural networks to reduce the watermark search overheads during the placement stage. AutoMarks's novel automated watermark search is accomplished by (i) constructing novel graph and node features with physical, semantic, and design constraint-aware representation; (ii) designing a data-efficient sampling strategy for watermarking fidelity label collection; and (iii) leveraging a graph neural network to learn the connectivity between cells and predict the watermarking fidelity on unseen layouts. Extensive evaluations on ISPD'15 and ISPD'19 benchmarks demonstrate that our proposed automated methodology: (i) is capable of finding quality-preserving watermarks in a short time; and (ii) is transferable across various designs, i.e., AutoMarks trained on one layout is generalizable to other benchmark circuits. AutoMarks is also resilient against potential watermark removal and forging attacks
Paper Structure (38 sections, 7 equations, 7 figures, 8 tables)

This paper contains 38 sections, 7 equations, 7 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Graph Neural Networks for Physical Design
  4. Physical Design Watermarking
  5. Motivation and Problem Formulation
  6. Methodology
  7. Watermark Search
  8. Graph Construction
  9. Label Collection
  10. Graph Neural Network Training
  11. Inference on Unseen Designs
  12. Watermark Insertion
  13. Watermark Extraction
  14. Experiments
  15. Experiment Setups
  16. ...and 23 more sections

Figures (7)

  • Figure 1: AutoMarks flow: The Watermark Search leverages GNN to identify the region and cells to watermark. Then, the Watermark Insertion encodes the selected watermark region of cells on the layout during the placement stage. The Watermark Extraction decodes the watermark and compares it with the encoded ones for ownership proof.
  • Figure 2: The watermark search time for different designs. The dotted line is the average search time of the large designs ($\geq 500k$ cells).
  • Figure 3: Watermarking performance under different attacks for wirelength-driven placement on the ISPD'2015 bustany2015ispd and ISPD'2019 liu2019ispd benchmarks. The black dotted line in the two left subfigures denotes the quality degradation threshold of 1.005, and the black dotted line in the rightmost subfigure denotes the watermark extraction threshold of 90%.
  • Figure 4: Watermarking performance under different attacks for wirelength-driven placement on the ISPD'2015 bustany2015ispd and ISPD'2019 liu2019ispd benchmarks. The black dotted line in the two left subfigures denotes the quality degradation threshold of 1.005, and the black dotted line in the rightmost subfigure denotes the watermark extraction threshold of 90%.
  • Figure 5: Histogram distribution of AutoMarks's labels.
  • ...and 2 more figures