ML-Based OPC Model Builder for Semiconductor Pattern Fidelity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical proximity correction (OPC) techniques in semiconductor integrated circuit (IC) manufacturing are inadequate in ensuring precise feature formation, leading to issues like rounded corners, pinching, necking, bridging, and metal line thickness variations, which affect device performance.
Innovation Solution
A machine learning-based model builder is used to optimize the OPC model by generating and classifying random terms into clusters, selecting representative terms based on impact, and adjusting the model until performance meets design requirements, enhancing the fidelity of OPC for improved IC manufacturing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If current OPC techniques are used, then manufacturing process is simple, but manufacturing precision is insufficient leading to pattern defects
Solution Approach 1:
The patent segments the OPC model into multiple components: a base OPC model and additional correction terms. The machine learning model generates multiple terms that are classified into clusters and selectively added to the base model, creating a segmented structure that improves precision while managing complexity through modular addition of correction terms.
Solution Approach 2:
The patent extends the OPC model from traditional two-dimensional pattern correction to three-dimensional feature formation by incorporating depth information and volumetric corrections. The machine learning model generates correction terms that operate in additional dimensions, enabling accurate control of 3D features like fins and trenches while maintaining computational efficiency.
2Manufacturing precision
If OPC correction is applied, then feature accuracy improves, but manufacturing time increases
Solution Approach 1:
The patent performs preliminary classification of correction terms into clusters before actual OPC application. The machine learning model pre-processes and organizes correction terms by their impact and relevance, so that during manufacturing, only the necessary terms are applied. This preliminary organization significantly reduces computation time during production while maintaining high pattern fidelity.
Solution Approach 2:
The patent applies only the necessary subset of correction terms rather than all possible corrections. The machine learning model identifies and selects only the most impactful terms for each specific pattern, avoiding unnecessary computations. This partial action approach maintains high precision where needed while reducing overall processing time across the entire manufacturing workflow.
3Manufacturing precision
If more OPC correction terms are added, then manufacturing precision improves, but model complexity increases
Solution Approach 1:
The patent segments the large set of correction terms into multiple clusters based on their characteristics and impact. Each cluster represents a specific type of correction, allowing the system to manage complexity through organized categorization while still having access to the full range of corrections when needed. This segmentation enables efficient selection and application of only relevant terms.
Solution Approach 2:
The patent dynamically adjusts the number and type of correction terms based on pattern-specific parameters. The machine learning model evaluates each pattern's characteristics and selectively activates only the necessary correction terms, changing the model's effective complexity based on the specific manufacturing task. This parameter-driven approach maintains high precision for complex patterns while keeping the model simple for straightforward cases.
Data Source
AI summary
A system and a method of optimizing an optical proximity correction (OPC) model for a mask pattern of a photo mask is disclosed. A machine learning (ML) based model builder includes an OPC model, measurement data and a random term generator. Random terms are generated in a M-dimensional space by the random term generator. The ML based model builder classifies the random terms to clusters by applying a classifying rule. A representative subset of the random terms is determined among the classified clusters, and the representative subset is added to the OPC model.