Dynamic Yield Prediction Model for Semiconductor Wafer Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional semiconductor manufacturing yield prediction methods fail to account for process target drift and changes in process variation, often relying on unrepresentative initial wafers and assuming distributions that may not reflect the larger population.
Innovation Solution
A dynamic yield prediction system that collects sample test information from test-only structures before wafer completion, constructs a yield prediction model using finished test data, and dynamically updates the model to accurately predict die yield, accounting for process characteristics and variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If conventional yield prediction methods use a small number of first wafers for testing, then testing cost and time are reduced, but prediction accuracy deteriorates due to process target drift and unrepresentative sampling
Solution Approach 1:
The system performs preliminary testing on a small subset of test-only structures early in the manufacturing process (before wafer completion) to establish initial yield predictions. This preliminary action allows early detection of yield issues without waiting for full wafer completion, reducing overall testing time while maintaining prediction capability through dynamic updates later in the process.
Solution Approach 2:
The system implements a feedback mechanism where yield predictions are continuously updated as more test data becomes available during wafer manufacturing. The dynamic yield prediction model incorporates new measurements from additional test structures, adjusting predictions to account for process target drift and variation, thereby improving accuracy over time without requiring complete wafer testing upfront.
2Device complexity
If conventional methods assume a normal distribution based on limited sampling, then prediction simplicity is maintained, but prediction accuracy deteriorates due to unrepresentative distribution assumptions
Solution Approach 1:
The system transitions from static distribution assumptions to a dynamic yield prediction model that adapts as manufacturing progresses. The model dynamically adjusts to reflect actual process variations and target drift by incorporating new test data from multiple wafer stages, eliminating the need for fixed normal distribution assumptions while maintaining computational tractability.
Solution Approach 2:
The system changes the parameters of the prediction model from fixed distribution parameters (mean, variance of normal distribution) to dynamic parameters that are continuously updated based on actual test measurements. This allows the model to capture true process variations and target drift without requiring complex non-parametric methods, balancing simplicity with accuracy.
3Measurement precision
If dynamic yield prediction collects data from multiple test structures throughout wafer manufacturing, then prediction accuracy improves by accounting for process drift, but measurement and processing complexity increases
Solution Approach 1:
The system segments the wafer manufacturing process into distinct stages, collecting test data from test-only structures at different manufacturing points. By dividing the continuous manufacturing process into measurable segments with intermediate testing, the system captures process evolution without requiring complete wafer completion before prediction, managing data collection complexity through structured阶段性 sampling.
Solution Approach 2:
The system uses universal test structures that can be tested at multiple stages of wafer manufacturing. These multi-functional test structures serve both as process monitoring points and as the basis for yield prediction, eliminating the need for separate dedicated test structures at each stage and reducing overall measurement complexity while maintaining prediction accuracy.
Data Source
AI summary
Dynamic yield prediction. In accordance with a first method embodiment of the present invention, a computer-implemented method includes collecting sample test information from a plurality of test-only structures prior to completion of the first wafer, gathering finished test data from all die of the first wafer, after completion of the first wafer, constructing a yield prediction model based on the sample test information and on the finished test data, and predicting, using the model, a percentage of die of the first wafer that will meet a particular specification. The method may further include a feedback loop to dynamically update the model.