Semiconductor Process Correction Using Sparse-to-Dense Parameter Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor manufacturing processes face challenges in accurately modeling parameter data due to sparse data sets, leading to sub-optimal corrections and inefficiencies in controlling lithographic processes, particularly in low-k1 lithography, where sparse-to-dense behavior and high-order alignment control impact throughput and precision.
Innovation Solution
A method involving sparse and dense data modeling is employed to determine corrections, adapting models based on sparse-to-dense mismatches using historical data and weighting factors to improve model accuracy and reduce noise, allowing for efficient control of lithographic processes without the need for frequent dense measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If dense measurements are performed frequently to improve model accuracy, then manufacturing precision is improved, but productivity deteriorates due to reduced throughput
Solution Approach 1:
The patent applies partial action by performing dense measurements only occasionally (e.g., once in several lots) rather than frequently, while using sparse measurements for routine monitoring. This reduces the burden of dense measurements while maintaining model accuracy through adaptive updating when dense data becomes available.
Solution Approach 2:
The system uses sparse measurements to maintain operational control and only invokes dense measurements when model accuracy degradation is detected or at scheduled intervals. The model adapts itself by learning from sparse data patterns and automatically requests dense measurements only when necessary, reducing overall measurement burden while maintaining precision.
2Productivity
If sparse measurements are used to maintain throughput, then productivity is improved, but measurement precision deteriorates due to insufficient data density
Solution Approach 1:
The patent introduces an adaptive model as an intermediary that bridges sparse measurements and the need for dense data. The model learns patterns from sparse measurements and occasionally incorporates dense measurements to refine its accuracy, effectively mediating between the limited measurement data and the requirement for precise process control.
Solution Approach 2:
The system dynamically changes the measurement density parameter based on process conditions and model confidence. When the model is confident and process conditions are stable, sparse measurements suffice. When uncertainty increases or anomalies are detected, the system temporarily increases measurement density to improve precision without permanently reducing throughput.
3Manufacturing precision
If dense measurements are performed to reduce sparse-to-dense mismatch, then manufacturing precision is improved, but loss of time increases due to measurement frequency
Solution Approach 1:
The system performs preliminary dense measurements during initial model training or when significant process changes are detected, preparing the model in advance to handle sparse measurements accurately. This preliminary action reduces the need for frequent dense measurements later, minimizing time loss while maintaining precision.
Solution Approach 2:
Dense measurements are performed periodically rather than continuously, at intervals determined by model confidence levels and process stability. This periodic dense measurement schedule reduces the overall time spent on measurements while maintaining model accuracy by updating the model at strategic intervals when dense data can significantly reduce sparse-to-dense mismatch.
Data Source
AI summary
A method for configuring a semiconductor manufacturing process, the method including: obtaining a first value of a first parameter based on measurements associated with a first operation of a process step in the semiconductor manufacturing process and a first sampling scheme; using a recurrent neural network to determine a predicted value of the first parameter based on the first value; and using the predicted value of the first parameter in configuring a subsequent operation of the process step in the semiconductor manufacturing process.