Neural Network Substrate Geometry Prediction for Semiconductor Process Tuning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current manufacturing processes for semiconductor devices face challenges in real-time monitoring and adjustment, as existing systems are either destructive, time-consuming, or unable to provide detailed information on process variations, leading to inefficiencies and yield losses.
Innovation Solution
A method using machine learning models, specifically neural networks, to receive input signals related to substrate geometry and determine variations in the manufacturing process, allowing for real-time adjustments and simultaneous monitoring of multiple layers, thereby enhancing process stability and yield.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional measurement methods are used to monitor manufacturing process variations, then measurement precision may be adequate, but the process is time-consuming and destructive
Solution Approach 1:
The patent replaces traditional mechanical/physical measurement systems with a computational prediction model (neural network) that processes optical inspection data to predict process variations. This substitution eliminates the need for time-consuming destructive measurements while maintaining detection accuracy through advanced algorithms.
Solution Approach 2:
The patent creates a virtual model (prediction model) that replicates the relationship between process parameters and substrate geometry. This digital copy allows real-time prediction of process variations without physical measurement, enabling fast non-destructive monitoring while preserving measurement precision through trained model predictions.
2Loss of information
If detailed process monitoring is implemented, then information completeness improves, but system complexity increases
Solution Approach 1:
The patent employs a universal prediction model that can simultaneously monitor multiple process parameters (etch rate, deposition thickness, side wall angle) across multiple layers using a single integrated system. This multi-functional approach provides comprehensive process information without proportionally increasing system complexity, as one model handles multiple monitoring tasks.
Solution Approach 2:
The patent introduces an intermediary layer (the neural network prediction model) that processes raw optical inspection data and translates it into meaningful process variation information. This intermediary simplifies the overall system by consolidating complex data processing functions into a single computational component, reducing the complexity burden while delivering detailed process insights.
3Manufacturing precision
If real-time process adjustment is enabled, then manufacturing precision improves, but ease of operation decreases
Solution Approach 1:
The patent implements a feedback mechanism where the prediction model continuously monitors substrate geometry and compares it against target specifications, then automatically generates adjustment recommendations for process parameters. This closed-loop feedback system maintains high manufacturing precision while managing operational complexity through automated decision-support rather than manual intervention.
Solution Approach 2:
The prediction model operates autonomously to identify process variations and suggest parameter adjustments without requiring operator intervention for analysis. This self-service capability improves manufacturing precision through consistent automated monitoring while preserving ease of operation by eliminating complex manual assessment tasks, leaving operators to simply implement straightforward adjustments.
Data Source
AI summary
A method for monitoring performance of a manufacturing process is described. The method includes receiving one or more input signals that convey information related to geometry of a substrate generated by the manufacturing process; and determining, with a prediction model, variation in the manufacturing process based on the one or more input signals. A method for predicting substrate geometry associated with a manufacturing process is also described. The method includes receiving input information including geometry information and manufacturing process information for a substrate; and predicting, using a machine learning prediction model, output substrate geometry based on the input information. The method may further include tuning the predicted output substrate geometry. The tuning includes comparing the output substrate geometry to corresponding physical substrate measurements and/or predictions from a different non-machine learning prediction model, generating a loss function based on the comparison, and optimizing the loss function.