SmartFDC™: Anomaly Detection with Machine Learning

Tyler Christensen – Director Product Development, INFICON

Background

A well-configured FDC system requires intimate knowledge of processes, equipment, and failure modes as well as a concerted effort to setup and maintain the system. Even when it is fully set up, it is only as good as the fault conditions that have already been conceived. While there will always be value to a well-thought-out engineering approach, machine learning algorithms (a subset of Artificial Intelligence) can provide significant advantages for both new and established FDC programs.

Leveraging the power of these advanced algorithms, the Intelligent Manufacturing Systems group at INFICON has designed a new product which will automatically detect anomalies in time-series tool data. This add-on package, called SmartFDC, will assist your FDC team in quickly identifying issues that could lead to yield losses.

Smart FDC Complements Standard FDC

With this system, data collected from each chamber is used to automatically train the algorithm (unsupervised learning) and report any detected anomalies (Figure 1).

Figure 1: The new SmartFDC package complements the standard engineering approach to FDC.

Results of the analysis are displayed on an intuitive web-based user interface (Figure 2). From here you will be able to:

View current and past anomalies including trend charts of process data
Flag anomalies for follow-up
Interact with the algorithm
Filter down to results that are relevant to your job function

Figure 2: The SmartFDC dashboard showing anomaly counts, relative severity, and whether the tool is flagged for follow-up by an engineer.

Easily Refine SmartFDC to Improve Accuracy

To further improve the accuracy of results, the user will then be able to "teach" the algorithm (supervised learning) based on engineering knowledge of the system. For example, known trim violations, noise issues, or test wafer runs could contribute to false anomalies which the system can be taught to ignore for future runs, improving future results of anomaly detection (Figure 3).

Figure 3: Teach the SmartFDC system to ignore certain types of signals and classify their impact to improve anomaly detection.

Integrates with INFICON Software Products

The system will also be shipped with built-in integration capabilities with other INFICON products. For example, our Metrology Sampling Optimizer will allow you to configure a rule to automatically sample material at metrology if an anomaly is detected in SmartFDC (Figure 4).

Figure 4: SmartFDC will alert the Metrology Sampling Optimizer to flag additional lots for measurement when an anomaly is detected.

Summary

With SmartFDC, you will be able to:

Quickly establish baseline detection system for new tool installations or sensor hardware
Reduce overall cost of ownership of FDC due to fewer resources required to maintain system
Quickly adapt to process/technology changes without major effort to add/validate analyses
Detect issues that may have not been conceived by an engineer
Reduce product excursion risk by monitoring for the unexpected

SmartFDC is the next step in the evolution of our products enabling advanced capabilities in your factory.