Learning-Based Models for Semiconductor Metrology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face computational challenges in metrology and inspection due to the high costs and complexity of aligning and combining multi-sensor data, particularly for defect detection and simulation tasks, which are intractable for advanced nodes like the 7 nm node and beyond.
Innovation Solution
The implementation of learning-based models, specifically deep neural networks and convolutional neural networks, to acquire and process output data from detectors, simulating outputs for specimens, thereby reducing computational costs and simplifying complex algorithms for metrology and inspection tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional algorithms are used for multi-sensor data alignment and combination in semiconductor metrology and inspection, then measurement precision and defect detection accuracy are maintained, but computational cost and device complexity increase significantly
Solution Approach 1:
The patent replaces conventional physics-based algorithms and numerical methods with machine learning models, substituting traditional computational mechanics with data-driven approaches. This substitution maintains measurement precision while significantly reducing computational complexity and resource requirements for multi-sensor data processing
Solution Approach 2:
The patent creates simplified computational models that replicate the functionality of complex conventional algorithms. By training machine learning models on data generated from conventional methods, the system copies the essential processing capabilities while achieving faster execution and lower computational costs
2Measurement precision
If conventional algorithms are used for processing multi-sensor data in semiconductor inspection, then defect detection accuracy is maintained, but processing time and productivity decrease
Solution Approach 1:
The patent performs preliminary actions by training machine learning models offline on large datasets generated from conventional algorithms. This pre-processing allows the models to learn optimal processing patterns in advance, enabling rapid real-time defect detection without sacrificing accuracy during actual inspection operations
Solution Approach 2:
The patent changes the fundamental parameters of the computational approach by transitioning from deterministic physics-based calculations to probabilistic machine learning predictions. This parameter change enables parallel processing and optimization that significantly improves processing speed while maintaining defect detection accuracy
3Measurement precision
If physics-based solutions are implemented for semiconductor metrology and inspection, then measurement accuracy is maintained, but energy consumption and computational resources increase
Solution Approach 1:
The patent substitutes energy-intensive physics-based computational mechanics with more efficient machine learning inference. Once models are trained, they require significantly less computational energy to process data while maintaining measurement accuracy, making the system more energy-efficient for continuous semiconductor inspection operations
Data Source
AI summary
Methods and systems for performing one or more functions for a specimen using output simulated for the specimen are provided. One system includes one or more computer subsystems configured for acquiring output generated for a specimen by one or more detectors included in a tool configured to perform a process on the specimen. The system also includes one or more components executed by the one or more computer subsystems. The one or more components include a learning based model configured for performing one or more first functions using the acquired output as input to thereby generate simulated output for the specimen. The one or more computer subsystems are also configured for performing one or more second functions for the specimen using the simulated output.