Hybrid Overlay Modeling for Field-by-Field Error Correction in the Photolithography Process

Sang Jin Kim, Hyui Geon Yoon, Ki Bum Lee, Chang Ouk Kim, Sung Jae Kim

Research output: Contribution to journalArticle


Overlay modeling, which predicts and corrects the overlay errors of wafers in the photolithography process, is emerging as a key technology for yield improvement. An overlay error is defined as a misalignment between the circuit images of the current and preceding exposure layers. Overlay modeling is classified into wafer-level modeling, which creates a single regression equation to correct the overlay errors of the entire wafer, and field-level modeling, which prepares a regression equation for each field of the wafer to conduct field-by-field error corrections. However, the former does not fully reflect the overlay error signatures of the fields, while the latter has an overfit problem with overlay errors in the fields used in model training. This paper proposes a hybrid modeling that creates field-by-field error prediction models by using an ensemble of ordinary least squares (OLS) regression and ridge regression to consider the wafer-level and field-level overlay error signatures, respectively. The hybrid modeling can appropriately reflect the overlay error signature of each field while avoiding the overfitting issue. The experimental results using field photolithography process data show that the mean squared error of the hybrid model was improved by 15.9% compared to that of the OLS regression-based model.

Original languageEnglish
Article number8922603
Pages (from-to)53-61
Number of pages9
JournalIEEE Transactions on Semiconductor Manufacturing
Issue number1
Publication statusPublished - 2020 Feb

All Science Journal Classification (ASJC) codes

  • Electronic, Optical and Magnetic Materials
  • Condensed Matter Physics
  • Industrial and Manufacturing Engineering
  • Electrical and Electronic Engineering

Fingerprint Dive into the research topics of 'Hybrid Overlay Modeling for Field-by-Field Error Correction in the Photolithography Process'. Together they form a unique fingerprint.

  • Cite this