Model-Based Metrology Error Estimation for Semiconductor Structures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor metrology tools face challenges in achieving accurate and precise measurements of shrinking patterned structures due to increasing complexity and tight process limits, with difficulties in error estimation, precision, and throughput, especially in correlating quality metrics with inference model accuracy.
Innovation Solution
A novel system for controlling and verifying measurements by using model-based analysis with machine learning techniques to estimate individual measurement errors for each parameter, providing an upper bound on measurement errors based on specific characteristics, enabling optimal model selection and data analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If model-based measurements are used to extract structure parameters from measured data, then measurement capability for shrinking semiconductor structures is improved, but measurement accuracy and precision deteriorate due to increasing complexity and tight process limits
Solution Approach 1:
The system implements feedback by calculating a distance metric that quantifies the discrepancy between measured data and model predictions, then using this metric to iteratively adjust model parameters until convergence. This feedback loop ensures measurement accuracy by continuously comparing predicted versus actual values and correcting deviations through gradient descent optimization.
Solution Approach 2:
The system performs preliminary action by pre-calculating a distance metric function and its gradient with respect to model parameters before actual measurement optimization. This pre-computation of the objective function and its derivatives enables faster and more accurate parameter extraction by having the optimization landscape ready in advance, rather than computing everything during the measurement process.
2Manufacturing precision
If tight process limits are applied to maintain manufacturing precision, then product quality is improved, but measurement robustness deteriorates due to reduced tolerance for variability
Solution Approach 1:
The system applies parameter changes by transforming the measurement problem into a different parameter space using a distance metric that is insensitive to certain variations. By changing from direct parameter comparison to a metric-based comparison, the system achieves robustness against process variability while maintaining precision within tight limits, as the distance metric captures essential deviations while ignoring noise.
3Ease of manufacture
If general error estimation methods are used, then computational simplicity is improved, but measurement verification capability deteriorates due to lack of individual parameter error bounds
Solution Approach 1:
The system applies segmentation by calculating separate distance metrics and error bounds for each individual structure parameter (e.g., line width, line spacing, layer thickness) rather than providing a single general error estimate. This segmentation enables precise verification of each parameter's measurement accuracy while maintaining computational efficiency through the use of a unified distance metric framework that can be applied to multiple parameters simultaneously.
Data Source
AI summary
A system and method are presented for controlling measurements of various sample's parameters. The system comprises a control unit configured as a computer system comprising data input and output utilities, memory, and a data processor, and being configured to communicate with a measured data provider to receive measured data indicative of measurements on the sample. The data processor is configured to perform model-based processing of the measured data utilizing at least one predetermined model, and determine, for each of one or more measurements of one or more parameters of interest of the sample, an estimated upper bound on an error value for the measurement individually, and generate output data indicative thereof.