Parameter Isolation Model for Semiconductor Metrology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current model-based metrology techniques for semiconductor fabrication face challenges in accurately measuring complex structures due to multiple parameter correlations, low measurement sensitivity, and computational tractability issues, often requiring unsatisfactory compromises between accuracy and computational effort.
Innovation Solution
The development of a parameter isolation model that isolates measurement signal information associated with a parameter of interest from incidental model parameters, allowing for reduced computational effort by fixing incidental parameters and optimizing each parameter of interest individually or in parallel, using trained models to capture variations across a semiconductor wafer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex multiple parameter models are used to represent complex semiconductor structures, then measurement accuracy is improved, but computational tractability deteriorates
Solution Approach 1:
The patent segments the complex measurement model into two distinct components: a parameter isolation model that handles parameter decoupling and a simplified measurement model that performs actual parameter estimation. This segmentation allows the complex structure to be analyzed in manageable parts, improving computational tractability while maintaining measurement accuracy through the coordinated use of both models.
Solution Approach 2:
The patent extracts and isolates the parameter correlation problem from the main measurement model by creating a separate parameter isolation model. This extracted component specifically handles the decoupling of correlated parameters, allowing the main measurement model to focus on accurate parameter estimation without being burdened by computational complexity from parameter correlations.
2Measurement precision
If multiple parameters are floated in the measurement model, then measurement accuracy is improved, but computational effort increases
Solution Approach 1:
The patent applies preliminary action by using the parameter isolation model to pre-process the measurement data and decouple parameter correlations before the actual parameter estimation occurs. This preliminary decoupling reduces the computational burden of the subsequent regression analysis, allowing multiple parameters to be floated without excessive computational time requirements.
Solution Approach 2:
The parameter isolation model serves as an intermediary between the raw measurement data and the final parameter estimation process. It mediates the complex relationship between multiple parameters by providing decoupled, pre-processed data that can be efficiently analyzed, thus enabling accurate multi-parameter measurement without prohibitive computational effort.
3Measurement precision
If parameter correlations are accounted for in complex models, then measurement accuracy is improved, but measurement sensitivity to some parameters deteriorates
Solution Approach 1:
The patent extracts the parameter correlation effects into a separate parameter isolation model, allowing the main measurement model to focus on parameters of interest with full sensitivity. By taking out the correlation handling from the main model, the measurement sensitivity to individual parameters is preserved while still accounting for correlations through the isolated model.
Solution Approach 2:
The parameter isolation model creates a simplified copy or representation of the parameter relationships that captures correlation effects without the full complexity of the original model. This copy allows accurate handling of parameter correlations while maintaining measurement sensitivity in the primary measurement model.
Data Source
AI summary
Methods and systems for building and using a parameter isolation model to isolate measurement signal information associated with a parameter of interest from measurement signal information associated with incidental model parameters are presented herein. The parameter isolation model is trained by mapping measurement signals associated with a first set of instances of a metrology target having known values of a plurality of incidental model parameters and known values of a parameter of interest to measurement signals associated with a second set of instances of the metrology target having nominal values of the plurality of incidental model parameters and the known values of the parameter of interest. The trained parameter isolation model receives raw measurement signals and isolates measurement signal information associated with a specific parameter of interest for model-based parameter estimation. The number of floating parameters of the measurement model is reduced, resulting in a significant reduction of computational effort.