Case Study 2 MIN

Inspection automation with Deep Learning

At a Glance

Client

Private Industrial Manufacturer – Instrumentation & Optics with +$550M Annual Revenue

Industry
Challenge

Develop a deep learning-based pipeline to identify, localize, and segment 20+ defects within high-end wafers and dies. Autonomous system performs all QC in the clients manufacturing workflow.

Solution

End-to-end CNN-based solution classified defects at a rate higher than what humans self-reported at 400x cheaper cost and was able to solve the scaling problem associated with future expert manual review.

Expertise & Technology

Business Challenge

An industrial manufacturer worked with SFL Scientific to develop a deep learning-based pipeline to identify, localize, and segment 20+ defects within high-end wafers and dies. The autonomous system performs the QC functions in the client’s manufacturing workflow.

SFL Scientific Solution

SFL built a CNN model classifying defects, as well as trained to detect the approximate location of the defect in X, Y coordinates based on zone and image coordinates. Six separate classifiers, one for each image source and one for each die type, were trained to detect the seven defects in different zones. Since the images contained multiple zones, the zone will also be included as a CNN model output. In addition, the X and Y coordinates of the defect were also included in the CNN model output.

Defect detection was accomplished by passing inspection images, in previously defined five formats, through the 15 classifiers. A Python program provided to input the five images for a die and produce a results listing. A Python 3.6 application running on a local computer will perform the defect detection operation and reported the results to the user in tabular and graphical format.

Results

End-to-end CNN-based solution classified defects more accurately than what humans self-reported at 400x cheaper cost. Solution is able to solve the scaling problem associated with future expert manual review, creating a true pipeline without bottlenecks. This improved the process, identifying potential critical component failures, and reduce waste and reducing end product rejection rate through upstream QC.

Work with Us

Start a Project

We'd like to help creating your next solution, whether modernizing legacy platforms or developing new AI solutions. Technology moves fast, let's build sustainable solutions.
Get Started