Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CAMO: Correlation-Aware Mask Optimization with Modulated Reinforcement Learning

Xiaoxiao Liang, Haoyu Yang, Kang Liu, Bei Yu, Yuzhe Ma

TL;DR

OPC remains challenging due to lithography proximity effects; CAMO tackles this by modeling local spatial correlations with a graph representation and employing a modulated reinforcement-learning policy that coordinates neighboring segment movements. The method combines a GraphSAGE-based GNN for local feature fusion with an RNN to capture sequential dependencies, and introduces an OPC-inspired modulator to guide action selection and stabilize training. Across via- and metal-layer experiments, CAMO outperforms state-of-the-art OPC engines and commercial tools in edge-placement accuracy (EPE) and PV band, while offering competitive runtimes, with the modulator notably improving convergence. This approach demonstrates practical gains in mask optimization and scalability to complex layout patterns in modern lithography.

Abstract

Optical proximity correction (OPC) is a vital step to ensure printability in modern VLSI manufacturing. Various OPC approaches based on machine learning have been proposed to pursue performance and efficiency, which are typically data-driven and hardly involve any particular considerations of the OPC problem, leading to potential performance or efficiency bottlenecks. In this paper, we propose CAMO, a reinforcement learning-based OPC system that specifically integrates important principles of the OPC problem. CAMO explicitly involves the spatial correlation among the movements of neighboring segments and an OPC-inspired modulation for movement action selection. Experiments are conducted on both via layer patterns and metal layer patterns. The results demonstrate that CAMO outperforms state-of-the-art OPC engines from both academia and industry.

CAMO: Correlation-Aware Mask Optimization with Modulated Reinforcement Learning

TL;DR

OPC remains challenging due to lithography proximity effects; CAMO tackles this by modeling local spatial correlations with a graph representation and employing a modulated reinforcement-learning policy that coordinates neighboring segment movements. The method combines a GraphSAGE-based GNN for local feature fusion with an RNN to capture sequential dependencies, and introduces an OPC-inspired modulator to guide action selection and stabilize training. Across via- and metal-layer experiments, CAMO outperforms state-of-the-art OPC engines and commercial tools in edge-placement accuracy (EPE) and PV band, while offering competitive runtimes, with the modulator notably improving convergence. This approach demonstrates practical gains in mask optimization and scalability to complex layout patterns in modern lithography.

Abstract

Optical proximity correction (OPC) is a vital step to ensure printability in modern VLSI manufacturing. Various OPC approaches based on machine learning have been proposed to pursue performance and efficiency, which are typically data-driven and hardly involve any particular considerations of the OPC problem, leading to potential performance or efficiency bottlenecks. In this paper, we propose CAMO, a reinforcement learning-based OPC system that specifically integrates important principles of the OPC problem. CAMO explicitly involves the spatial correlation among the movements of neighboring segments and an OPC-inspired modulation for movement action selection. Experiments are conducted on both via layer patterns and metal layer patterns. The results demonstrate that CAMO outperforms state-of-the-art OPC engines from both academia and industry.
Paper Structure (14 sections, 6 equations, 6 figures, 2 tables, 1 algorithm)

This paper contains 14 sections, 6 equations, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Reinforcement Learning
  4. Policy Gradient
  5. Method
  6. RL Environment
  7. CAMO Architecture
  8. Agent Training
  9. Experiments
  10. Experimental Setup
  11. Experiments on Via Layer Patterns
  12. Experiments on Metal Layer Patterns
  13. Effectiveness of the Modulator
  14. Conclusion

Figures (6)

  • Figure 1: The comparison of the inference flow among CAMO and previous regression-based and RL-based OPC.
  • Figure 2: Overall framework of CAMO.
  • Figure 3: Illustration of squish pattern encoding. Given a control point, its neighboring geometries are encoded in $M$ and the distance information is revealed in $\delta_x$ and $\delta_x$.
  • Figure 4: Illustration on the projection function. Given the EPE value of a segment, five points are evenly sampled from region $[0, EPE]$, and are then projected through $f(\cdot)$.
  • Figure 5: The EPE trajectories with / without modulator on M2 and M4 in \ref{['tab:metalcomp']}.
  • ...and 1 more figures