PCB factories embrace AI

November 14, 2019

The evolution of PCBs from large and antiquated “printed wiring boards” to today’s fine-line designs on high-density interconnect (HDI) PCBs, IC substrates (ICS) and more, has been matched by manufacturing processes that have evolved from manual assembly to highly automated production. As manufacturing technology further develops, processes become more complex and more sophisticated, including the ability to inspect and then shape defects that would once have resulted in scrapped panels. A significant opportunity is now emerging for the PCB manufacturing industry to capitalize on artificial intelligence (AI) and optimize production processes and, ultimately, the entire PCB manufacturing facility.

PCB manufacturing has typically relied on experts who gained their knowledge over many years. These experts intimately know and understand every step of the manufacturing process. They understand how to leverage their knowledge for production optimization and yield improvements. Human limits, including error and fatigue, hamper what these experts can accomplish. Operator errors or mistaken identification of PCB defects—“false alarms”—can take a toll on the yield and even on the PCBs themselves due to overhandling. By integrating AI into the manufacturing process (Figure 1 ), machines can add value by taking over certain “learned” tasks, while human experts continue to undertake the more complex tasks that require thought and interaction at the same time as optimizing and “training” the AI system. This combination of human and artificial intelligence improves overall efficiency and operations and is the greatest opportunity for AI expert systems.
 

Figure 1 AI could help PCB factories improve quality.


AI and Industry 4.0

The future of PCB manufacturing is a factory with fully integrated Industry 4.0 systems that feature AI at the global and manufacturing system levels. The “global” level includes all systems across the factory, not just the individual manufacturing systems. Industry 4.0 provides the automation and data exchange infrastructure that enables real-time production analysis, bi-directional communication and data sharing, traceability, and on-demand data analysis. Within any particular factory, AI can improve processes by using data acquired from various manufacturing systems and machines, collected through Industry 4.0 mechanisms such as traceability, bi-directional communication. The factory benefits because AI analyzes massive amounts of system-wide data to optimize factory set up parameters and achieve the highest levels of productivity and yield. AI analysis and self-learning is ongoing and occurs through artificial neural networks. Within several years, it will eliminate intervention by human operators, and lead to the creation of fully automated factories.

The requirements for this new PCB manufacturing model will include full connectivity of all factory systems along with AI as a monitoring and decision-making mechanism. Currently, there are proprietary and technical challenges that limit the full automation of PCB factories, and so today AI is being added to individual systems, such as automated optical inspection (AOI) solutions, wherever possible. The advantages of moving production facilities toward a global AI model include much more reliable notification of PCB defects – “true defects” – with the tremendous benefit of a feedback loop that identifies the source of an issue and then automates revision of factory processes to eliminate related defects.

Subsets of AI, including machine learning and deep learning, will move PCB factories toward the goal of full automation. Machine learning uses algorithms that enable computers to improve the performance of a task using data and the examples it has already experienced and learned from, without being explicitly programmed to do so. In the case of PCB manufacturing, machine learning facilitates increased yield, improves fabrication operations and processes, and reduces manual operations while simultaneously helping drive more efficient handling of factory assets, inventory and the supply chain.

Deep learning takes AI to an even more complex level – one that is beneficial at the global factory systems level. Deep learning is inspired by the human brain’s ability to learn, understand and extrapolate using multi-faceted, multi-layer artificial neural networks. In PCB factories, software expert systems effectively learn from sophisticated representations of insights, patterns and context in the data collected. The learning then forms the basis of automated process improvements in PCB manufacturing.

Implementation of machine learning and deep learning provides PCB manufacturers with the ability to go beyond human understanding; that is, the AI system discovers new optimization opportunities by digging deeper and in places where humans don’t even think to explore. AI expert systems are very efficient, reducing the required number of human experts and driving efficiency and best practices by using additional, more complex parameters to monitor factories systems at the global level.

Using Industry 4.0 sensors (those that can send data from equipment) and systems, data is created at the global level throughout the PCB manufacturing process, from simple reading and writing capabilities to advanced tracking of process parameters down to the smallest PCB unit. Process parameters can include etching, resist development and concentration of chemical materials in the manufacturing process. These types of data are analyzed using deep learning to inform optimization of manufacturing methods and parameters, to identify patterns, and to make informed decisions on needed changes in the processes. All of this can take place 24 hours a day, seven days a week in a fully automated fashion.

System-level AI

At the system level, current AI implementations on the PCB manufacturing floor demonstrate measurable impact on productivity and yield during, for example, AOI processes. In this case, it is machine learning that dramatically reduces human error in detecting PCB defects. Examples of PCB defects include shorts and opens, or excess copper even in trace amounts. Automated inspection can detect very small defects that may not be caught with manual inspection or may be missed as a result of human error, a natural consequence of repetitive work.

A classic inspection of 100 panels without the use of AI typically identifies 20 to 30 defects per panel, approximately 75 percent of which may be false alarms. Because policy dictates that all defects must be reviewed manually, the review of false alarms wastes valuable production time, increases handling of the PCBs, which can result in new damage, and opens up possibilities for further false analysis during review by an operator, who may be tired, overworked or distracted.

With machine learning on an AOI system, such false alarms and repairs are dramatically reduced (Figure 2 ). Fewer false alarms means less handling of the panels which in and of itself will increase yield. Additionally, AI provides consistent (and dynamically improving) classification of defects without the limitations inherent in human operators, providing more reliable results and reducing verification time. AI in AOI systems has been found to reduce false alarms by up to 90 percent, based on Orbotech internal studies. AOI is somewhat unique in that the systems collect more data than any other manufacturing solution, which makes it highly appropriate as a first step in AI implementation. At the same time, the AOI room is the most labor-intensive area of a PCB factory and therefore, has the most to gain from the adoption of AI into its processes. For PCB manufacturers, this all means that millions of defects can be more accurately identified and classified, with the potential for producing greater yield and reducing costs.
 

AI-driven AOI

Figure 2 AI-driven AOI can reduce verification and labor on the production floor.

 

Global- and system-level AI working together

An example of AI at work at both the system and global levels is the following:Assume that an AOI system inspects 100 panels. At the system level, AI powered by machine learning filters out the false alarm defects that have been classified by the system as shorts. The AI system generates the smartest possible classification results by evaluating multiple AOI images while leveraging its “panel understanding” (the AOI solution’s understanding of the elements on the panel and how they should look). This information feeds into the global AI system which, powered by deep learning, collects these data from the system-level solutions and determines that the true defects identified are shorts that require additional etching time to remove the excess copper. The AI system uses the data from the system level to make global decisions to adjust the panel parameters in the etch processes, so that all panels manufactured going forward have fewer, if any, defects of the same type. Eventually, communication between system-level solutions will further increase and improve the decision-making capabilities of AI at the global level.


Manufacturing challenges increase

While developments in AI are rapidly moving industry-wide, the challenges in PCB manufacturing increase at least as quickly, if not more so. The two primary areas in which defect detection is increasingly difficult are for flex materials and shrinking trace line geometries. Next-generation complex materials, such as liquid crystal polyamide (LCP) and modified polyamide (MPI), pose new challenges for manufacturers including image acquisition, handling, deformation and finer lines. The more sophisticated materials for flex PCBs, for example, result in more defects being identified, which results in more false alarms. The manufacturer’s objective with such complex materials is to minimize the handling of panels in the process of determining false alarms. Flex PCBs (Figure 3 ), then, are a product type that will likely benefit greatly from AI implementation, as the systems will learn to manufacture within stricter parameters.
 


Figure 3 Flex circuits present additional problems for automated optical inspection.


PCBs for 5G are another application that requires greater manufacturing precision than is currently required and has the potential to benefit greatly from the expertise enabled by artificial intelligence. The HDI PCBs needed for 5G applications demand finer line widths with straight side-wall geometry and highly exacting parameters. This makes defect detection more difficult than ever and will be extremely challenging for human experts to effectively accomplish.

It is with these and other yet-unknown PCB manufacturing challenges in mind that AI-driven factories will become the key to production in the future. Developments in the application of AI at the global level will require a bit more time to come to fruition in PCB manufacturing, but it is clear that system-level AI implementation has arrived, forming the foundation of a future with fully automated PCB factories.

— Meny Gantz is the VP of Marketing in Orbotech’s Printed Circuit Board (PCB) division.