Feature-Model Process Recipes for Wafer Parameter Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Semiconductor wafer processing complexity increases with shrinking device size, leading to inefficient process optimization due to numerous process control variables and insufficient in-line precision metrology, resulting in sub-optimal recipes and resource wastage.
Innovation Solution
The use of machine learning feature models to create and match process recipes by determining predicted processing parameters based on target properties, optimizing recipe settings through numerical optimization routines and sensor data, and generating offsets to correct for process disruptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional design of experiment (DoE) methods are used to optimize process parameters, then process optimization can be achieved through manual tuning, but it consumes valuable wafer resources and requires extensive manual engineering effort
Solution Approach 1:
The patent creates virtual copies of wafers through simulation models that replicate physical wafer behavior and process responses. These digital twins allow extensive DoE exploration without consuming physical wafer resources, enabling multiple what-if scenarios to be tested virtually before committing to actual process runs.
Solution Approach 2:
The system enables self-service process optimization by automatically generating and evaluating DoE scenarios through computational models. The automated framework performs parameter sweeps, response surface modeling, and optimization algorithms without requiring manual engineer intervention for each experimental design, reducing both wafer consumption and manual effort.
2Measurement precision
If destructive metrology techniques like TEM are used for precision measurement, then high measurement precision can be achieved, but the process is time-consuming and cannot be integrated into production lines
Solution Approach 1:
The patent introduces computational models and simulation frameworks as intermediaries between process parameters and metrology outcomes. These models predict measurement results based on process conditions, allowing virtual metrology that achieves precision comparable to TEM without the time-consuming physical measurement process, enabling integration into production workflows.
Solution Approach 2:
The system replaces physical destructive measurement systems (TEM) with computational prediction systems. By using physics-based models and machine learning algorithms to predict metrology outcomes from process parameters, the system eliminates the need for time-consuming physical sectioning and imaging while maintaining measurement precision.
3Manufacturing precision
If the number of process control variables is increased to handle shrinking device sizes, then processing capability can be maintained, but the complexity of process optimization increases significantly
Solution Approach 1:
The patent segments the complex multidimensional process parameter space into manageable subsets for systematic exploration. By dividing the optimization problem into smaller experimental designs that focus on specific parameter interactions, the system makes the optimization of numerous control variables tractable while maintaining the ability to handle shrinking device dimensions.
Solution Approach 2:
The system adds a computational dimension to the physical process optimization by introducing virtual simulation space. This allows exploration of high-dimensional parameter spaces through computational models rather than physical experimentation alone, making it feasible to optimize numerous process variables without exponential increases in complexity.
Data Source
AI summary
A method includes receiving a set of feature models, each feature model of the set of feature models corresponding to a respective feature associated with processing of a component, receiving a set of target properties for processing the component, where the set of target properties includes, for each feature, a respective target, determining, based on the set of feature models, one or more sets of predicted processing parameters in view of the set of target properties, generating one or more candidate process recipes each corresponding to a respective one of the one or more sets of predicted processing parameters, where the one or more candidate process recipes each correspond to a set of predicted properties including, for each feature, a respective predicted property value resulting from component processing, and selecting, from the one or more candidate process recipes, a process recipe for processing the component.