Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Novel End-to-End Production-Ready Machine Learning Flow for Nanolithography Modeling and Correction

Mohamed S. E. Habib, Hossam A. H. Fahmy, Mohamed F. Abu-ElYazeed

TL;DR

The paper analyzes barriers preventing production adoption of ML-RET and introduces TPM-RET, a production-friendly end-to-end ML-RET flow. It advances a true pixel-based correction approach with a compact CNN, an Inverse Intensity Profile ($IIP$) representation, and nonuniform image compression to enable scalable, full-chip lithography corrections. A two-phase data preparation and deployment pipeline trains a CNN to predict $IIP$ classifications, assembling an end-to-end corrected photomask output that aligns with reference tools like pxOPC. The results on a 32nm immersion lithography case demonstrate good correlation with reference corrections, and the authors argue that TPM-RET offers scalable CPU-based deployment, consistent corrections, re-correction capabilities, and broad flexibility for production environments.

Abstract

Optical lithography is the main enabler to semiconductor manufacturing. It requires extensive processing to perform the Resolution Enhancement Techniques (RETs) required to transfer the design data to a working Integrated Circuits (ICs). The processing power and computational runtime for RETs tasks is ever increasing due to the continuous reduction of the feature size and the expansion of the chip area. State-of-the-art research sought Machine Learning (ML) technologies to reduce runtime and computational power, however they are still not used in production yet. In this study, we analyze the reasons holding back ML computational lithography from being production ready and present a novel highly scalable end-to-end flow that enables production ready ML-RET correction.

Novel End-to-End Production-Ready Machine Learning Flow for Nanolithography Modeling and Correction

TL;DR

The paper analyzes barriers preventing production adoption of ML-RET and introduces TPM-RET, a production-friendly end-to-end ML-RET flow. It advances a true pixel-based correction approach with a compact CNN, an Inverse Intensity Profile () representation, and nonuniform image compression to enable scalable, full-chip lithography corrections. A two-phase data preparation and deployment pipeline trains a CNN to predict classifications, assembling an end-to-end corrected photomask output that aligns with reference tools like pxOPC. The results on a 32nm immersion lithography case demonstrate good correlation with reference corrections, and the authors argue that TPM-RET offers scalable CPU-based deployment, consistent corrections, re-correction capabilities, and broad flexibility for production environments.

Abstract

Optical lithography is the main enabler to semiconductor manufacturing. It requires extensive processing to perform the Resolution Enhancement Techniques (RETs) required to transfer the design data to a working Integrated Circuits (ICs). The processing power and computational runtime for RETs tasks is ever increasing due to the continuous reduction of the feature size and the expansion of the chip area. State-of-the-art research sought Machine Learning (ML) technologies to reduce runtime and computational power, however they are still not used in production yet. In this study, we analyze the reasons holding back ML computational lithography from being production ready and present a novel highly scalable end-to-end flow that enables production ready ML-RET correction.
Paper Structure (37 sections, 2 equations, 8 figures, 1 table)

This paper contains 37 sections, 2 equations, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. Traditional RET
  3. ML-RET State-of-the-Art Analysis
  4. State-of-the-Art Synopsis
  5. Image-Based Photomask Correction
  6. Generative ML Techniques
  7. Single RET ML Applications
  8. Model-Based Reference Data
  9. Difficulties Facing Production-Ready ML-RET
  10. Loss of Mask Information
  11. Full-Chip Scale Issues
  12. Correction Consistency
  13. Hardware Requirements
  14. Pattern Interaction Distance
  15. Input resolution
  16. ...and 22 more sections

Figures (8)

  • Figure 1: Example of a simple target wafer pattern $T$ (a), and its corresponding inverse lithography ideal photomask $M^*$ (b)
  • Figure 2: IIP map corresponding to target pattern in fig. \ref{['ilt']}.
  • Figure 3: Flow chart for training data preparation.
  • Figure 4: Flow chart for model deployment in photomask optimization application.
  • Figure 5: Confusion Matrix for TPM-RET IIP predictions versus the IIP reference generated from pxOPC photomask.
  • ...and 3 more figures