AI in Semiconductor Manufacturing The Impact of AI on Inspection and Metrology in Semiconductor Manufacturing

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Artificial intelligence is increasingly being used in semiconductor inspection and metrology to automate defect detection and increase throughput. Charlie Zhu from Nordson Advanced Technology Solutions explains how AI-based systems address challenges that were previously almost impossible to solve.

The semiconductor industry is under increasing pressure to detect defects and inconsistencies with near-perfect accuracy, while the demand for smaller, faster and more powerful chips is growing. However, the smaller, faster and more powerful these chips become, the more complex the manufacturing processes become.

Traditional manual inspection methods have been reaching their limits for some time now as production volumes and complexity increase. Artificial intelligence (AI) is therefore becoming an important tool in inspection and metrology to automate processes, increase accuracy and throughput and keep pace with the complexity of modern chip designs.

AI-based Defect Detection

Such AI-powered systems use large amounts of data to detect patterns and anomalies that conventional methods can miss. "In most cases, AI can make better decisions than a human operator, with fewer false rejects. It can provide more complex, advanced analysis than traditional algorithms based on simple thresholds and a binary pass/fail system," explains Charlie Zhu, Director, R&D at Nordson Advanced Technology Solutions.

In most cases, AI analyses can also run faster than standard algorithms, which leads to additional time savings. In addition to minimizing energy and material waste, this is a significant factor for profitability. However, according to Zhu, what drives companies the most is the unique ability of AI to solve challenges that were previously unaddressable.

When inspecting microscopic components, traditional methods would have difficulty detecting certain defects or anomalies—for example in corner-fill inspection, where conventional methods such as blob analysis reach their limits. Through deep learning, AI-integrated systems with less reliance on skilled workers could detect and flag problems that would otherwise be misinterpreted or overlooked.

TSV Inspection: From One Hour to Less Than a Minute

One example is the inspection of through-silicon vias (TSVs) at micron level. While conventional methods take around an hour for this complex process, AI can achieve the same level of accuracy in under a minute, according to the manufacturer.

Another advantage is the ability of artificial intelligence to perform real-time inline inspections. While extensive data analysis used to slow down the production line, AI now enables large volumes of data to be processed quickly without significantly slowing down production throughput. Machine learning (ML) models also adapt automatically to new production requirements. This adaptability can be critical on fast-moving production floors to reduce bottlenecks and increase productivity.

Supervised and Unsupervised Learning

Machine learning is central to the further development of AI inspection. Supervised learning is based on pre-labelled data to train AI models to recognize specific defect types. Unsupervised learning, on the other hand, does not require labeled data and analyzes data independently to identify patterns, outliers or anomalies. "This means that it can detect unknown and novel defects that have not been seen before or that customers may not even know exist," says Zhu. Nordson therefore uses both supervised and unsupervised learning in its Nordson Intelligence AI ecosystem.

Challenge: Customer Data and Security

One of the biggest challenges when using AI is the management of customer data to train intelligent systems. Data security and confidentiality are top priorities for customers, and many are understandably reluctant to grant direct access to their data. That's why Nordson says it has developed secure solutions, including private cloud domains and protected remote access. After all, access to usable, real data remains critical for future machine learning.

Outlook: Predictive Maintenance and Generative AI

According to Zhu, Nordson is focusing on emerging growth areas such as predictive maintenance, generative AI and automated ML. At the same time, the current solution portfolio is being further developed, including ongoing system development and supervised learning. "AI development continues to progress and is opening up new fields of application in semiconductor inspection," concludes Zhu. (sb)

Charlie Zhu is Director, R&D at Nordson Advanced Technology Solutions.