Demystifying ADC

ADC stands for automatic defect classification. It’s a software that classifies defects based on image and metadata such as location, ROI, and other information associated with a defect. ADC is not a mysterious black box that’s impossible to understand. Instead, ADC classifies defects the same way a human operator does, by first being trained by an expert. Then, just like human classification experts, the system starts by examining samples, extracting and “virtually” memorizing the visual characteristics that separate different types of defects into classes and applying that knowledge at runtime. Theoretically, if you can train operators to classify your data, then you can reasonably assume that ADC can do the same.

Why ADC?

Frequently, I get the question, “What makes ADC superior to manual classification?” In truth, there are many different ways an ADC implementation can benefit a manufacturing operation, including:

• Speed: Supported by moderately powered hardware, an ADC classifier can process dozens of samples per second, whereas the best humans max out at ~1 sample per second.
• Accuracy: Trained ADC solutions can easily be configured to perform at least as accurately as highly trained and experienced operators, and frequently these systems have demonstrated superior accuracies when validated against average defect reviewers.
• Consistency: ADC systems do not forget their training. They are not influenced by daily events or tire because of long hours sitting at a review station. ADC systems, when left alone, will classify the same defect today that it saw months or years earlier. Consistency is notably important in large multi-shift manufacturing operations where multiple human reviewers participate in classification activities. In these situations, trained individuals could look at the same image and differ in its classification, especially if the two types of defects have very similar traits.
• Reliability: Automatic defect classification solutions are compiled software being executed on a server. This means the software will operated 24/7/365. It does not take a lunch break; it does not have sick days or require vacations a couple of weeks per year. It’s like the Energizer Bunny; it just keeps going.
• Inline vs Offline: High speed and reliability also allow ADC to be embedded inline, meaning most of the classifications can be done at the same time of, or right after, inspection. This gives engineers the ability to identify problems as early as possible and act accordingly, which translates to better yield and efficiency.

Common misconceptions about ADC

I also find myself educating colleagues and customers alike about misconceptions surrounding the general field of ADC. Here are some classics:

• Automatic v. Automated Defect Classification: People frequently believe the “A” in ADC is for “automatic” and have a perception that an ADC system requires no human interaction whatsoever. The truth is that an ADC solution is no different from any other tool on the manufacturing floor. Just like an etcher or CMP system, ADC executes a recipe and produces a result. Also, like other tools, that recipe needs to be created by a tool owner and from time to time needs to be adjusted as processing changes are implemented.
• ADC is hard to configure: Setting up ADC classifier is like training your operator. Just as you would subject the human trainee to multiple examples of defects, ADC systems need a similar learning session. Again, like a human trainee you’d want to test their ability to learn and based on this test make minor adjustments if needed. Modern ADC solutions are built with an intuitive UI designed to guide you through the natural steps of collecting/managing samples, configuring image detection, setting up classifiers, and verifying the results. The biggest difference verses training a human is that you only need to train a single ADC system, not a small army of human reviewers.
• ADC classifier performance is unpredictable: A well represented set of samples, and clearly defined and visually different classes, is key to both ADC and operator. An ADC classifier is very predictable when that’s the case.
• ADC is perfect: Like a human operator, ADC is not perfect. If an operator is confused on certain samples, then ADC will most likely be, too.

What’s new for ADC with AI and machine learning

Nowadays, you can hardly write a technique blog without mentioning AI, and that’s no exception here. Machine learning has become increasingly accessible with the recent advancement of technology in hardware and software. It opens up a whole slew of opportunities using ADC that otherwise may not be possible.

As a factory yield engineer you may ask yourself the question: How do I ride the wave and make my job easier and advance my career all at the same time?

Even though the entry point for machine learning has never been this low, there is still a learning curve for the average fab to adopt this technology. On the other hand, corporations with internal AI teams may have already invested time on machine learning but struggle to integrate their solution in a complex runtime production environment.

A well-designed enterprise level ADC application should allow users of all different levels to take advantage of machine learning/AI all within the same platform.

• Are you new to machine learning? ADC should allow you to create machine learning models and train/test the model without fiddling with Tensor Flow or python.
• Does your organization already have models trained and want to prove it in production? ADC should allow you to import externally trained models and fit it into runtime production.
• Are you wondering if those well-known models from the public domain such as VGG16, XCEPTIONS, etc. can classify your defects? ADC should allow you to use those transferable models in your fab easily.

Give it a try. In some cases, just two hours can save your operator 80% of the time.