Semiconductor Process Correction via Sparse-to-Dense Fingerprint Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in accurately modeling and controlling Critical Dimension (CD) and overlay performance due to sparse data representation, leading to sub-optimal corrections and inefficient throughput, as sparse-to-dense data mismatches result in poor fingerprint descriptions and erratic behavior across different wafers and lots.
Innovation Solution
A method is introduced that involves obtaining both sparse and dense data sets, applying a model to determine sparse-to-dense mismatches, adapting the model based on these mismatches, and using the adapted model to determine corrections, thereby improving process control and reducing the need for frequent dense data sampling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If sparse data sampling is used for process monitoring, then measurement time and complexity are reduced, but measurement precision and model accuracy deteriorate due to sparse-to-dense mismatches
Solution Approach 1:
The patent introduces a transfer learning framework where a model trained on dense data serves as an intermediary to enable accurate predictions from sparse data. The pre-trained model captures complex patterns from comprehensive datasets, then adapts to sparse measurement scenarios, bridging the gap between sparse sampling efficiency and dense data accuracy.
Solution Approach 2:
The patent dynamically adjusts model parameters through transfer learning, adapting the model's internal representations based on the available data density. By changing parameters learned from dense data and fine-tuning them with sparse data, the system maintains high measurement precision while operating under sparse sampling conditions.
2Measurement precision
If dense data sampling is used for accurate process control, then measurement precision improves, but productivity decreases due to increased measurement frequency requirements
Solution Approach 1:
The patent applies partial action by using sparse sampling (less than full dense sampling) combined with transfer learning to achieve sufficient model accuracy. Instead of requiring complete dense data coverage, the system uses partial data with enhanced processing through pre-trained models, maintaining acceptable precision while improving throughput.
3Productivity
If sparse data is used for process correction, then throughput is improved by reducing measurement frequency, but manufacturing precision deteriorates due to poor fingerprint descriptions
Solution Approach 1:
The patent performs preliminary action by pre-training models on dense data before deployment. This advance preparation creates a robust foundation that enables accurate process corrections even when only sparse data is available during actual operation, ensuring manufacturing precision is maintained despite reduced measurement frequency.
4Manufacturing precision
If model adaptation based on dense data is performed frequently, then manufacturing precision improves, but loss of time increases due to frequent retraining
Solution Approach 1:
The patent implements periodic action by retraining models at optimized intervals rather than continuously. Transfer learning enables the system to maintain accuracy with less frequent retraining compared to traditional approaches, as the pre-trained model provides a robust starting point that requires fewer updates to adapt to process variations.
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.