In the semiconductor manufacturing industry, the need for continuous quality improvement has never been more pronounced. This demand is driven by an unprecedented influx of manufacturing data, with more than 1000 process parameters recorded for a single wafer, and tens of thousands of wafers being produced daily. Traditional statistical methods have proven insufficient to fully exploit these massive volumes of data.
As such, this paper explores the application of hybrid machine learning techniques, specifically Memory Based Reasoning (MBR) and Neural Network (NN) learning, as more powerful tools for managing this complexity and improving yield in semiconductor manufacturing.
Understanding Memory-Based Reasoning (MBR) in Semiconductor Manufacturing
Memory Based Reasoning (MBR) is an instance-based learning method, inspired by the way humans learn from past experiences. In the context of semiconductor manufacturing yield, MBR retrieves previously learned instances from a database (case-base) that most closely matches the current situation or problem. This allows for quick, adaptable solutions based on historical data.
The Implementation of MBR in Yield Management
In practice, MBR uses feature weights to establish the importance of different attributes. These feature weights are calculated from the trained neural network, aiding in the prediction of outcomes by presenting the most similar examples from the case base. This provides a rapid response system to identify potential process deviations and propose appropriate corrective measures.
The Neural Network and Memory-Based Reasoning Framework
An integrated framework is proposed for semiconductor yield management, leveraging the combined power of Neural Network (NN) and Memory Based Reasoning (MBR). This hybrid system is capable of managing large datasets, and high dimensions, and dynamically adapting to different situations.
NNs are a form of machine learning modeled after the human brain, capable of learning complex patterns and relationships in data. The NN part of the system is responsible for learning the non-linear relationships between manufacturing process parameters and yield.
MBR, on the other hand, is a type of instance-based learning which utilizes experience to make predictions about new instances. The MBR component uses the previously learned examples to find the most similar case in the case base when a new instance comes up.
The system operates by calculating a feature weight set from the trained neural network. This feature weight set connects both learning strategies and aids in predicting outcomes by presenting the most similar examples from the case base.
Information Theory-based Model for Strategy Selection
To maximize the efficacy of the proposed hybrid learning system, an information theory-based model for strategy selection is introduced. This model determines optimal strategies for yield management by using metrics such as the knowledge extraction rate per experimentation cycle and per unit of time as benchmarks.
Four key yield analysis tools are examined within this model, namely electrical testing, automatic defect classification, spatial signature analysis, and wafer position analysis. These tools have distinct roles in both R&D and volume production environments, providing critical information that informs strategy selection and execution.
Applying Information Theory to Yield Management
The proposed model examines the utility of four yield analysis tools—electrical testing, automatic defect classification, spatial signature analysis, and wafer position analysis. Each tool generates unique insights critical for both research and development (R&D) and volume production environments. This enables an adaptive strategy that optimizes yield over time.
The Importance of Data Visualization Tools
Data visualization tools, such as yield analysis software, are essential for storing, tracking, and analyzing all data collected during chip manufacturing and testing. These tools enable the conversion of raw data into actionable insights, enhancing the understanding of the manufacturing process, increasing productivity, and improving yield.
Detailed Examination of Visualization Tools
A range of visualization tools such as wafer mapping software, trend charts, correlation charts, histograms, Pareto analysis, fail flip maps, fail trends, fail category maps, and gallery views are discussed in detail. Each of these tools plays a specific role, enabling detailed data analysis, anomaly detection, trend identification, and overall process understanding.
For instance, trend charts analyze parameter behavior over time, whereas correlation charts observe how two different test parameters behave similarly. Fail trends summarize recent yield and failing parameter trends, while histograms visualize data distribution and detect outliers. Pareto analysis identifies the most significant failures and core problems within a production workflow.
Conclusion
The application of machine learning, specifically hybrid systems leveraging Neural Networks and Memory Based Reasoning, presents a powerful approach to handling the immense volume and complexity of data in semiconductor manufacturing. Coupled with an information theory-based model for strategy selection, this approach maximizes knowledge extraction and optimizes yield management.
Furthermore, the use of various data visualization tools facilitates the conversion of raw data into actionable insights, fostering a deeper understanding of the manufacturing process and enabling continuous quality improvement. These innovative approaches provide a more comprehensive understanding of the manufacturing process, paving the way for continuous quality improvement in the semiconductor industry.
References
- Boser, B.E., Guyon, I.M., Vapnik, V.N., 1992. A training algorithm for optimal margin classifiers. In Proceedings of the fifth annual workshop on Computational learning theory (pp. 144-152).
- Zong, B., Song, Q., Min, M.R., Cheng, W., Lumeau, J., Deng, J., Chen, C., 2018. Deep autoencoding Gaussian mixture model for unsupervised anomaly detection. In International Conference on Learning Representations.
- Freund, Y., Schapire, R.E., 1997. A decision-theoretic generalization of online learning and an application to boosting. Journal of Computer and system sciences, 55(1), pp.119-139.
- Han, J., Pei, J., Kamber, M., 2011. Data mining: concepts and techniques. Elsevier.
- Keogh, E., Mueen, A., 2017. Curse of dimensionality. In Encyclopedia of Machine Learning and Data Mining (pp. 314-315). Springer, Boston, MA.
- Ghahramani, Z., 2015. Probabilistic machine learning and artificial intelligence. Nature, 521(7553), pp.452-459.
- Cover, T.M., Thomas, J.A., 2012. Elements of information theory. John Wiley & Sons.
- Few, S., 2012. Show me the numbers: Designing tables and graphs to enlighten. Analytics Press.
