We consistently hear from our customers that they need new ways to apply the latest technologies, such as AI, to improve efficiency. One area that AI has proven to be particularly beneficial is in helping to automate the visual quality control process for manufacturing customers.
These customers tell us they want AI solutions that help them make quality control and inspections more efficient, to improve overall quality. But, there are many factors that make it difficult to prevent the distribution of damaged products. And the later a defect is caught in the manufacturing process, the more costly it is to fix or replace. Visual inspection helps manufacturing customers identify defects early and at a lower cost, and we’re seeing many innovative ways it’s helping our customers revolutionize their processes.
Chip making made more efficient
One example of a customer using AI to transform their manufacturing process is GlobalFoundries, a leader in the semiconductor manufacturing industry. The company used AutoML Vision to build a visual inspection solution that can detect random defects in wafer map and scanning electron microscope (SEM) images, which are essential pieces for semiconductor manufacturing. A wafer map shows the performance of a semiconductor device, while an SEM’s images, which are created with a focused beam of electrons, can be used to closely examine a wafer.
“Google Cloud AutoML Vision made it easy for our subject matter experts to quickly learn how to navigate and then train the AI,” Dr. DP Prakash, Global Head of AI XR Innovation at GlobalFoundries explained. “In our factory leading the initiative, 40% of the manual inspection workload has already been successfully shifted to the visual inspection solution we built based on AutoML.”
GlobalFoundries’ visual inspection solution integrates AutoML Vision into their in-house content management system, and includes SEM image acquisition, image and sample defect management, defect prediction visualization, and product quality report generation among its features. AutoML Vision reads in the images of wafers and sample defects, and trains customized models to detect these defects. The trained model will be used to detect defects in new incoming product images.
When evaluating technologies, GlobalFoundries was impressed that AutoML Vision could successfully classify 80% of the images based on a limited amount of training data in the initial pass itself. This fast path to high accuracy let GlobalFoundries quickly move to production, start realizing benefits, and scale up.
To capture and control process defects in semiconductor factories, GlobalFoundries deployed hundreds of models in its factories. AutoML Vision’s data and model management features help refresh the data continuously and efficiently, giving the company visibility into all those models.
GlobalFoundries also achieved similar success in their lithography process—where a pattern is transferred onto a chip. In the conventional method, due to the practical constraints of time and cost in high volume manufacturing environments, only a sample of the wafers produced are typically inspected for systematic defect patterns. The new visual inspection solution developed with AutoML, however, increases the validation rate to 95% of wafers, reducing waste, and improving quality and customer satisfaction.
Revolutionizing manufacturing processes
Siemens is another company using AutoML Vision to change the way they manage the inspection process.
“Siemens leveraged Google’s domain expertise in AI technology to create Factory AI service, which revolutionized our manufacturing with automated visual inspections,” said Tigran Bagramyan, Intrapreneur and Data Scientist, Siemens. “We use AutoML Vision to quickly build prototypes and push them to production on the factory floor. AutoML Vision helps us concentrate on use cases and customer value rather than complexity of AI development.”
Meanwhile, LG CNS leverages AutoML Vision Edge to create manufacturing intelligence solutions that detect defects in everything from LCD screens and optical films, to automotive fabrics on the assembly line. AutoML Vision Edge improved defect detection accuracy by 6% and reduced the time to design and train their ML models from seven days to just a few hours.
AutoML Vision lets customers train high-quality defect detection models, deploy models, and run inference on production lines. We look forward to supporting customers as they continue to find innovative new ways to deploy AI.
To learn more about how you can use our vision products for visual inspection and other use cases, check out Google Cloud Vision AI.
Source: Google Cloud Blog