Photomask Correction Modeling for Accurate Semiconductor Lithography
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Accurate photolithography processes in highly integrated semiconductor devices are challenging due to the complex refraction and diffraction of light, making it difficult to achieve the desired patterns in the photoresist layer.
Innovation Solution
A computing device communicates with semiconductor manufacturing equipment to receive layout and height data, generate levelset data, and train a function model to correct photomask patterns using machine learning, ensuring accurate optical proximity correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional photolithography processes are used for highly integrated semiconductor devices, then manufacturing complexity increases due to light refraction and diffraction, but manufacturing precision deteriorates making it difficult to achieve desired patterns
Solution Approach 1:
The patent replaces traditional optical physics-based photolithography with a machine learning-based computational system. The ML model learns the complex relationship between photomask patterns and resulting semiconductor features, substituting physical optical processes with data-driven predictions to achieve higher precision without increasing process complexity
Solution Approach 2:
The patent transforms the photolithography process from a purely physical optical process to a computational process by introducing machine learning parameters. The system learns optimal photomask patterns by training on historical data, changing the approach from fixed optical parameters to adaptive learned parameters that compensate for refraction and diffraction effects
2Manufacturing precision
If machine learning models are trained with incomplete or inaccurate data, then model training time increases, but manufacturing precision deteriorates due to poor pattern prediction
Solution Approach 1:
The patent performs preliminary data collection and preprocessing before model training by gathering historical photomask and semiconductor data, cleaning and formatting it for ML processing. This preliminary action ensures high-quality training data is ready in advance, improving prediction accuracy while reducing actual training time by avoiding data issues during training
Solution Approach 2:
The patent implements a feedback mechanism where the ML model's predictions are compared against actual semiconductor measurements, and the model is retrained with this feedback data. This continuous feedback loop improves prediction accuracy over time while the system learns from real-world performance, gradually reducing the need for extensive initial training
Data Source
AI summary
Disclosed is a method of operating a computing device configured to communicate with semiconductor manufacturing equipment. The method includes receiving layout data of a photomask to be used to manufacture a semiconductor device from the semiconductor manufacturing equipment, the layout data indicating a pattern of the photomask for each position in a distance direction, receiving height data of the semiconductor device from the semiconductor manufacturing equipment, the height data indicating a height of the semiconductor device for each position in the distance direction, generating levelset data based on the layout data, the levelset data indicating an intensity of light passing through the photomask for each position in the distance direction, and generating function model information by training a relationship between the levelset data and the height data.