Deep learning was used by CyberOptics to precisely examine the corner fill on integrated circuits (ICs) made by a significant memory supplier. A more sophisticated approach was required since traditional inspection techniques were unable to completely detect the presence or absence of fill. In order to provide a trustworthy, automated solution that met the client’s inspection objectives, CyberOptics drew on its extensive reservoir of algorithmic and neural network knowledge.
Challenge
The memory producer used to fill to encircle an IC’s corners and secure the die to the substrate. To make sure there was neither too much nor too little fill, the customer was required to check its existence and absence. They needed a system that could gauge the corner fill’s length and provide a quality rating.
The majority of the corner fill can be located using traditional corner fill inspection techniques like blob analysis, but a more robust strategy was required to avoid false calls. A few blobs were frequently mistakenly identified as independent entities when they actually may have been joined by some segment by the blob analysis method, which sought to discover a continuous blob within a specific colour intensity or contrast. Knowing whether the computer accurately or mistakenly identified a break in the fill was a hurdle. This is a challenge that many AOI machines’ conventional blob analysis algorithms cannot solve.
Solution
CyberOptics approached the programming as an object detection issue since the customer just required identifying the length of the corner fill. To reduce time, the CyberOptics team applied transfer learning. In order to train another class onto a network architecture that has already been trained for one task, transfer learning is used. For this application, sophisticated deep learning (DL) techniques were applied.
In order to produce high-quality images for machine learning (ML) programming, CyberOptics’ Multi-Reflection SuppressionTM (MRSTM) sensor technology, which combines a distinctive camera architecture and sophisticated algorithms, has an advantage over the competition.
The SQ3000TM Multi-Function system made use of the side cameras on CyberOptics’ high-resolution proprietary sensor to provide a side-view examination of all four component sides without the use of extra mirrors or moving the components. These side cameras take excellent pictures without the use of additional cameras or uptake.
One streamlined model, powered by DL, was able to complete all necessary tasks for the memory supplier’s IC corner fill inspection requirements. The program gives a streamlined user experience and runs without any issues.
Benefits
In order to analyze IC corner fills, CyberOptics developed a system that uses deep learning to enable object classification, identification, and transfer learning.
A feasibility study demonstrates the trained deep learning model’s excellent robustness in identifying and producing corner fills. The object detection program’s bounding box satisfies the customer’s need to measure the corner’s length.
To support customers with ever-more-complex inspection demands, CyberOptics keeps improving its machine learning capabilities. The team is capable of appropriately training models for particular applications because to its extensive realm of algorithmic and neural network skills. The CyberOptics team can train a model on one end and the customer can conduct inference on the other end thanks to SQ software, which makes it simple for them to communicate remotely.
The potential for deep learning-powered enhanced capabilities is limitless, and CyberOptics is consolidating its market leadership with distinctive AI-enabled solutions that successfully handle the complex applications of its SMT and semiconductor customers.
Click on the following link Metrologically Speaking to read more such case studies about the Metrology Industry.