Detecting defects that reduce breakdown voltage using machine learning and optical profilometry.

Area of Science:

• Materials Science and Engineering: Focus on semiconductor wafer manufacturing and characterization.
• Electrical Engineering: Application in high-voltage and high-frequency power devices.

Background:

• Semiconductor wafer manufacturing requires precise control of performance metrics for integrated circuit (IC) quality.
• Gallium Nitride (GaN) offers advantages for high-voltage/high-frequency power devices, but substrate defects limit performance.
• Optimizing vertical GaN devices is crucial for next-generation power electronics.

Purpose of the Study:

• To apply machine learning (ML) to predict wafer performance metrics, specifically breakdown voltage (Vbk).
• To utilize optical profilometry data as input for ML models.
• To identify wafers with a high probability of meeting critical performance standards.

Main Methods:

• Data acquisition using optical profilometry to capture wafer surface characteristics.
• Implementation of machine learning algorithms to analyze profilometry data.
• Correlation of predicted performance metrics with actual device breakdown voltage (Vbk).

Main Results:

• ML models successfully predicted the probability of wafers meeting performance metrics.
• The approach reliably identified devices prone to premature failure (low breakdown voltage).
• Optical profilometry data proved effective for predicting critical performance parameters.

Conclusions:

• Machine learning, combined with optical profilometry, offers a powerful tool for semiconductor wafer quality control.
• This predictive capability can reduce the number of failing devices and improve manufacturing yields.
• Further investigation with alternative methods may be needed for devices failing at intermediate breakdown voltages.