CASE STUDY
PCB Wafer Semiconductor boards
Due to the inherent fragility of PCB Wafer Semiconductor boards, along with their intricate complexity, accentuates their vulnerability to defects, thereby emphasizing the urgent requirement for this global semiconductor supply to have a robut in-house defect detection solution.
The solution required must have the capability to identify defects at a sub-micrometer scale, significantly exceeding the confines of human visual inspection.
Due to this requirement, Mindtrace was selected to create a bespoke AI defect detection solution.
Challenges
- The training dataset contained fewer than 70 images
- The training data lacked proper labeling
- The training dataset did not encompass all possible defects
Step 1 - Model Training
During the training phase, Mindtrace's Brain-Sense™ Platform effectively constructed a model using the limited set of 70 training samples. Given the scarcity of training data, the model employed few-shot learning techniques, enabling it to detect defects that were not explicitly represented in the training datasets.
Step 2 - Model Testing
Once the training phase was complete, the bespoke model was tested to understand it's levels of accuracy. The model achieved a 96% defect detection rate, even detecting defects that weren't available within the original dataset.
Upon completion of the training phase, the bespoke AI brain underwent rigorous testing to assess its accuracy levels. Remarkably, the model achieved an impressive 96% defect detection rate.
Outcome
Mindtrace successfully engineered a tailor-made AI solution, harnessing the power of the Brain-Sense™ Platform, to detect defects at a sub-micrometer scale, boasting an unparalleled industry-leading accuracy rate of 96%.
What sets this solution apart is not only its remarkable accuracy but also its remarkable ability to identify defects that were absent from the initial training dataset, showcasing its exceptional adaptability and robustness.
Learning Samples
Defects Detected
Delivery Time