The PCB industry urgently needs AI and machine learning

　　Today, PCB has developed to a new stage, with new technologies such as High-Density Interconnect (HDI) PCBs and IC substrates (ICS) introduced, transforming the entire production process from manual to fully automated. As manufacturing technology advances further, processes become increasingly complex, making defect inspection more important and challenging. These critical defects can lead to the scrapping of entire PCB boards. For the PCB manufacturing industry, opportunities are emerging to leverage Artificial Intelligence (AI) to optimize production processes and ultimately enhance the entire PCB manufacturing workflow.

　　PCB manufacturing typically relies on experts with years of accumulated knowledge who deeply understand every step of the manufacturing process. They know how to use their expertise to optimize production and increase yield. Human limitations, including operational errors and fatigue, hinder efficiency growth. Operator mistakes or misidentification of PCB defects ("false alarms") can affect yield due to overprocessing and may even damage the PCB itself. By integrating AI into the manufacturing process (Figure 1), machines can add value by taking over certain "learned" tasks, while human experts continue to handle more complex tasks that require thinking and interaction with AI systems during optimization and "training." The combination of humans and AI improves overall efficiency and operations, representing the greatest opportunity for AI systems.

　　Figure 1. AI can help PCB factories improve quality.

　　Artificial Intelligence and Industry 4.0

　　The ultimate trend in PCB development is factories with fully integrated Industry 4.0 systems that adopt AI technology at both global and manufacturing system levels. The "global" level includes all systems within the factory, not just individual manufacturing systems. Industry 4.0 provides the infrastructure for automation and data exchange, enabling real-time production analysis, two-way communication and data sharing, traceability, and on-demand data analytics. Within any specific factory, AI can use data obtained from various manufacturing systems and machines collected through Industry 4.0 mechanisms (such as traceability and two-way communication) to improve processes. Factories benefit because AI analyzes large amounts of system-wide data to optimize factory setup parameters and achieve the highest levels of productivity and yield. AI analysis and self-learning are ongoing and conducted through artificial neural networks. Within a few years, this will eliminate human operator intervention and lead to fully automated factories.

　　This new PCB manufacturing model requires all factory systems to be fully connected and AI to serve as the monitoring and decision-making mechanism. Currently, proprietary and technical challenges limit full automation of PCB factories, but AI has been added to individual systems as much as possible, such as automated optical inspection (AOI) solutions. The advantages of moving production facilities toward a global AI model include more reliable notification of PCB defects—"real defects"—and a feedback mechanism that identifies the root cause of problems and then automatically modifies factory processes to eliminate related defect issues.

　　Subsets of AI, including machine learning and deep learning, will drive PCB factories toward full automation. Machine learning uses algorithms that enable computers to improve task performance using data and examples they have experienced and learned from, without explicit programming. In PCB manufacturing, machine learning can increase yield, improve manufacturing operations and processes, reduce manual labor, and help drive more efficient handling of factory assets, inventory, and supply chains.

　　Deep learning elevates AI to a more complex level, which is beneficial at the global factory system level. Inspired by human brain neurons, deep learning uses multilayer artificial neural networks with the ability to learn, understand, and infer. In PCB factories, software systems can effectively collect data and learn from complex representations of patterns and context. This learning then forms the basis for automatic process improvements in PCB manufacturing.

　　The implementation of machine learning and deep learning provides PCB manufacturers with capabilities beyond human understanding; AI systems discover new optimization opportunities by digging deeper where humans are reluctant to explore. AI expert systems are highly efficient, reducing the number of required human experts and improving efficiency and best practices by monitoring factory systems globally using more and more complex parameters.

　　Using Industry 4.0 sensors (sensors that can send data from equipment) and systems, data can be created globally throughout the PCB manufacturing process, from simple read/write functions to advanced tracking of process parameters down to the smallest PCB units. Process parameters can include etching, photoresist development, and even the concentration of chemical materials during manufacturing. Deep learning analyzes these types of data to inform optimized manufacturing methods and parameters, identify patterns, and make informed decisions about necessary process changes. All of this can be done continuously, 24/7.

　　System-Level AI

　　At the system level, such as in the AOI process, AI implementation in PCB manufacturing workshops has a significant impact on productivity and yield. In this case, machine learning greatly reduces human errors when detecting PCB defects. Examples of PCB defects include short circuits and open circuits, and even excess copper. Automated inspection can detect very small defects that manual inspection might miss or overlook due to human error, which is a natural result of repetitive work.

　　Without AI, classic inspections of 100 panels typically find 20 to 30 defects per panel, about 75% of which are false alarms. Because policy requires manual inspection of all defects, reviewing false alarms wastes valuable production time, increases handling of PCBs, which may cause new damage, and may affect operators' further mistakes during the review process.

　　By applying machine learning to AOI systems, such false alarms and rework can be greatly reduced (Figure 2). Fewer false alarms mean less handling of PCB boards and improved efficiency. Additionally, AI provides consistent (dynamically improving) defect classification without the inherent limitations of operators, delivering more reliable results and reducing verification time. According to Orbotech internal research, AI in AOI systems can reduce false alarms by up to 90%. AOI is unique in that it collects more data than any other manufacturing solution, making it very suitable as the first step for AI implementation. Meanwhile, the AOI room is the most labor-intensive area in PCB factories, so adopting AI in its process brings the greatest benefits. For PCB manufacturers, this means more accurate identification and classification of millions of defects, potentially increasing yield and reducing costs.

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

　　The following are examples of system-level and global-level AI working together:

　　Assume the AOI system inspects 100 panels. At the system level, AI supported by machine learning can filter out false defect alarms that have been classified by the system. The AI system generates the smartest classification results by evaluating multiple AOI images while utilizing its "panel understanding" (the AOI solution's understanding of elements on the panel and their appearance). This information is input into the global AI system, powered by deep learning, which collects this data from system-level solutions and determines that the identified real defect is a short circuit requiring additional etching time to remove excess copper. The AI system uses data from the system level to make global decisions to adjust panel parameters during the etching process, resulting in fewer (if any) of the same type of defects in all subsequently manufactured panels. Ultimately, communication between system-level solutions will further increase and improve AI's decision-making capabilities globally.

　　Increasing Manufacturing Challenges

　　Although AI development is rapidly advancing across the entire industry, challenges in PCB manufacturing are growing at the same pace, if not faster. Flexible materials and the geometry of shrinking traces are two difficult areas for defect detection. Next-generation composite materials, such as liquid crystal polymer (LCP) and modified polyimide (MPI), present new challenges for manufacturers, including image acquisition, processing, deformation, and finer lines. For example, the more advanced the materials used for flexible PCBs, the more defects are identified, leading to more false alarms. Manufacturers use these complex materials to minimize panel handling during the process of determining false alarms. Therefore, FlexPCB (Figure 3) is a product type that could greatly benefit from AI implementation, as the system will learn to manufacture within stricter parameter ranges.

　　Figure 3. Flexible circuits bring additional challenges to automated optical inspection.

　　PCBs for 5G are another high-demand type that could greatly benefit from AI-supported expertise. The HDI PCBs required for 5G applications need finer line widths, straight sidewall geometries, and strict parameters. This makes defect detection more difficult than ever, and it will be extremely challenging for human experts to effectively complete defect detection.

　　Considering these and other unknown PCB manufacturing challenges, AI-driven factories will be key to future production. It will take more time to achieve the development of AI applications globally in PCB manufacturing, but it is clear that the realization of system-level AI has already arrived, laying the foundation for the future of fully automated PCB factories.