Source Mask Optimization Initial Shapes Using Mask Layout Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing source mask optimization processes in semiconductor lithography often rely on arbitrary initial source shapes, leading to longer optimization runtimes and non-optimal solutions due to a lack of consideration for the lithographic mask layout.
Innovation Solution
A machine learning model is used to determine initial source shapes based on the actual mask layout, selecting clips that are challenging to manufacture or sensitive to variations, and inferring corresponding source shapes for the optimization process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If source mask optimization (SMO) is performed to improve manufacturing precision of photomasks, then manufacturing precision is improved, but computation time increases exponentially
Solution Approach 1:
The patent applies preliminary action by using machine learning to pre-select promising initial source shapes before the full SMO computation begins. This pre-selection filters out poor candidates that would waste computational resources, allowing the optimization to start from better initial points and converge faster.
Solution Approach 2:
The patent replaces the traditional exhaustive computational mechanics of SMO with a machine learning-based selection system. The ML model predicts which initial source shapes are likely to lead to successful optimization, substituting brute-force computation with intelligent prediction to reduce overall computation time.
2Manufacturing precision
If multiple initial source shapes are tested to improve manufacturing precision, then manufacturing precision is improved, but computation time increases
Solution Approach 1:
The patent uses machine learning to identify and copy successful patterns from training data. The ML model learns from previously solved SMO problems and copies effective source shape characteristics to new problems, avoiding redundant exploration of unsuccessful shape variations.
Solution Approach 2:
The patent changes the approach from testing many source shapes with fixed parameters to using ML-guided parameter selection. The system dynamically adjusts which source shape parameters to explore based on ML predictions, focusing computational effort on promising parameter ranges rather than exhaustively searching all possibilities.
3Manufacturing precision
If conventional SMO methods are used to improve manufacturing precision, then manufacturing precision is improved, but the process becomes computationally intractable for complex patterns
Solution Approach 1:
The patent segments the SMO process into two distinct stages: (1) ML-based initial source shape selection, and (2) conventional SMO optimization. This segmentation allows each stage to be optimized independently, with the ML stage handling the combinatorial explosion of initial shape selection and the conventional SMO stage handling the precise optimization.
Solution Approach 2:
The machine learning model acts as an intermediary between the problem definition and the SMO optimization process. It translates complex pattern requirements into selected initial source shapes, mediating the transition from problem specification to optimization input and reducing the complexity burden on the SMO algorithm.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Initial source shapes for source mask optimization are determined based on a layout of the lithographic mask. In one approach, a layout (210) of a lithographic mask is received. Different sections of the lithographic mask, referred to as clips (225), are selected. These clips are applied to a machine learning model (230) which infers source shapes from the clips. The inferred source shapes are used as the initial source shapes (245) for source mask optimization.