Improving electronic sensor reliability by robust outlier screening.

Area of Science:

• Semiconductor Manufacturing
• Electronic Sensors
• Quality Control

Background:

• Electronic sensors require extremely low defect rates in critical applications like automotive and medical equipment.
• Increasing sensor reliability is paramount to reduce customer quality incidents (CQIs).
• Outlier detection algorithms are essential for screening latent defects.

Purpose of the Study:

• To introduce and evaluate novel spatial outlier detection algorithms: Good Die in a Bad Cluster with Statistical Bins (GDBC SB) and Bad Bin in a Bad Cluster (BBBC).
• To assess an advanced outlier screening method, Robust Dynamic Part Averaging Testing (RDPAT), and two practical improvements.
• To compare the efficiency and effectiveness of these algorithms in identifying CQIs using production data.

Main Methods:

• Development and application of spatial algorithms GDBC SB and BBBC.
• Implementation of the advanced outlier screening method RDPAT with practical enhancements.
• Analysis of production data from 289,080 dice with 26 identified CQIs.

Main Results:

• The study evaluated the performance of GDBC SB, BBBC, and RDPAT in identifying defects.
• These algorithms have been successfully used in production environments for several years.
• Comparative analysis determined the efficiency and effectiveness of each method in reducing CQIs.

Conclusions:

• The advanced outlier detection algorithms and methods significantly enhance the ability to screen latent defects in electronic sensors.
• Implementation of GDBC SB, BBBC, and RDPAT leads to a decrease in customer quality incidents.
• These techniques are vital for ensuring high reliability in semiconductor manufacturing.