Spectral Endpoint Control for Multi-Step Semiconductor Etching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advanced etch and deposition processes in semiconductor manufacturing introduce variations in electronic device characteristics between wafers, leading to non-uniformities that require time-consuming metrology and adjustments, affecting production efficiency and resource utilization.
Innovation Solution
A method is developed to generate a machine learning model using time-varying spectral data from substrates during multi-step etch processes, allowing for the prediction of substrate parameter values and real-time adjustments of process conditions, such as etch duration and chamber settings, to compensate for wafer-to-wafer variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If metrology is used to identify non-uniformities and modify processing operations, then manufacturing precision is improved, but loss of time increases
Solution Approach 1:
The machine learning model is trained in advance on historical spectral data and metrology measurements to learn the relationship between spectral features and wafer parameter variations. During production, the pre-trained model directly predicts wafer parameters from real-time spectral data without requiring iterative metrology measurements and manual process adjustments, thus eliminating the time delay while maintaining precision.
Solution Approach 2:
The patent replaces the traditional mechanical metrology measurement system with a computational prediction system based on machine learning. Instead of using physical metrology tools to measure wafer parameters after processing, the system uses spectral data combined with a trained ML model to predict parameters in real-time, significantly reducing measurement and correction time while maintaining accuracy.
2Manufacturing precision
If traditional metrology and manual adjustment methods are used, then manufacturing precision can be maintained, but productivity decreases
Solution Approach 1:
The machine learning model automatically predicts wafer parameters and identifies deviations from target values without human intervention. The system self-corrects by providing real-time feedback for process parameter adjustments, eliminating the need for manual metrology analysis and process engineering interventions, thus maintaining precision while significantly improving throughput.
Solution Approach 2:
The system implements real-time feedback by continuously monitoring spectral data during processing, comparing predicted parameters against target values, and immediately suggesting or automatically applying process corrections. This closed-loop control maintains wafer parameter consistency while enabling faster production cycles compared to traditional periodic metrology and manual adjustment approaches.
3Loss of time
If machine learning models are trained on spectral data to predict substrate parameters, then loss of time is reduced, but device complexity increases
Solution Approach 1:
The complex task of model training is performed in advance during a setup phase using historical data. Once trained, the model is deployed as a ready-to-use predictive engine that requires only simple spectral data input during production. This preliminary action transfers the complexity from the operational phase to the setup phase, reducing real-time processing complexity while maintaining fast prediction capability.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables faster identification and correction of non-uniformities, improving production efficiency by reducing the time spent on metrology and resource allocation, and enhancing the consistency of wafer characteristics.
Implementation Method 1
receiving training data including, for each of a plurality of training substrates, (a) time varying spectral data collected in situ from a training substrate over multiple steps of a multi-step etch process
Data Source
AI summary
Methods and systems for using a time-series of spectra to identify endpoint of a multi-step semiconductor fabrication processes such as multi-step deposition and multi-step etch processes. One method includes accessing a virtual carpet (e.g., a machine learning model) that is formed from a time-series of spectra for the multi-step processes collected during a training operation. During production, in-situ time-series of spectra are compared to the virtual carpet as part of end pointing of multi-step fabrication processes.