The Rise of AI in PCB Manufacturing

AI is advancing at a rapid rate

A tremendous opportunity is emerging in the PCB manufacturing industry whereby artificial intelligence (AI) can be utilized to streamline production processes and improve production outcomes in ways never before achievable.

 

On the big screen, we've often seen AI in the form of killer robots or armies of autonomous machines turning against their owners. But in reality, the technology is being used across a range of practical applications — shaping the way we all live and work.

 

For most people, their first encounter with AI will be through smart assistants like Apple's Siri and Amazon's Alexa, which rely upon on

accurate voice recognition to relay data from the internet. But with developments like self-driving cars in the pipeline, there's no doubt AI is advancing at a rapid rate. And it's only a matter of time before it goes one step further.

 

What is AI?

AI describes a machine or software that mimics human cognitive functions, such as problem-solving and learning. Many important innovations are becoming subsets of AI, including machine learning and deep learning. Machine learning is a type of AI that uses various algorithmic techniques to enable computers to improve the performance of a task using data, but without being explicitly programmed to do so. In PCB manufacturing, machine learning enables a host of benefits including improved operations, reduced scrap rates, optimized fab operations, reduction of “non-sophisticated labor,” and the more efficient management of assets, inventory, and the supply chain. All of these relevant features are the very definition of Industry 4.0.

Deep learning is an even more complex form of AI in which computers very effectively learn sophisticated representations of insights, patterns, and the context in data, all of which can be applied to process improvements in manufacturing. Deep learning utilizes multi-faceted, multi-layer artificial neural networks, which mimic the human brain’s ability to learn, understand and extrapolate.

 

Figure 1: AI and the World of Printed Circuit Boards

 

What Can AI Bring to PCB Manufacturers

In printed circuit board (PCB) manufacturing, AI presents tremendous opportunities. Most PCB designers manually route and design their boards, which is a time-consuming and intricate process.

 

The introduction of AI into PCB manufacturing is especially important as the market moves toward Industry 4.0 – or the “Smart Factory” – in which automation systems communicate and cooperate with humans in real-time, decentralizing the decision-making processes and offering numerous benefits.

 

AI will offer manufacturers a higher degree of accuracy when placing components in a compact and densely packed part.

Another area of PCB manufacturing where AI is proving useful is inspections. Based on the common location for a defect, AI can quickly and easily narrow down defects to save both time and money.

 

However, AI cannot be successful without very high-quality, labeled data. In PCB manufacturing, for example, defect classification is a crucial aspect of an automated optical inspection (AOI) solution.

 

In the AOI process, AI machines greatly reduce human error when detecting PCB defects. Typically, AOI machines send images to a remote multi-image verification station where a human operator classifies them as either “true defects” or “false alarms”. However, the human element inevitably opens up the process to error and mistakes can easily be made in classification.

 

For instance, PCB defects include short circuits, open circuits, and excessive copper. Automatic inspection can find small defects. Such defects may not be found by manual inspection, or they may be missed due to human error. These will naturally lead to repetitive work.

 

In traditional inspections that we do not use AI, when inspecting 100 panels, usually 20 to 30 defects are found in each panel. But 75% of the inspecting results are false alarms. As the policy stipulates that all defects must be inspected manually again. Reviewing wastes valuable production time; increases the processing of the PCB, and may cause more damages. When the operator is tired, overworked, or distracted, further error analysis may occur.

 

In contrast, the use of AI in the AOI system can significantly reduce such false alarms, thereby reducing repair work (Figure 2).

 

Figure 2: AOI using AI can reduce verification and labor in the production workshop

 

An AI solution would be able to make classification decisions autonomously and with consistent accuracy once it has learned from the human operators’ decisions. Such an AI system is dependent on accurate data patterns to learn correct algorithmic behavior over time. Even the tiniest change in data could lead to game-changing results, so data sensitivity is crucial.

 

The fewer false, the less processing of the panel. In addition, AI's classification of defects is more consistent and can be continuously improved, without some inherent limitations like human operators, thus providing more reliable results and reducing verification time. According to research, AI in the AOI system can reduce false alarms by up to 90%.

 

AOI plays an important role

AOI is unique in that its system collects more data than any other manufacturing solution, so it is very suitable as the first step in AI implementation. At the same time, the AOI room is the most labor-intensive place in the PCB factory, so AI benefits the most in the application process. For PCB manufacturers, this means that millions of defects can be identified and classified more accurately, which may increase the volume of production and reduce costs.

 

Let's take an AOI system checking 100 panels as an example to understand how AI works. AI-supported by machine learning can filter out false-positive defects that are classified as short circuits by the system.

 

The AI system evaluates multiple AOI images and uses the "understanding of the panel" (the components on the panel and their appearance) to generate the most intelligent classification results. It enters the information into a global AI system driven by deep learning and collects this data from system-level solutions. Then it determines those real short-circuit defects (extra copper is removed by an additional etching process). The AI system uses system-level data to make global decisions to adjust panel parameters in the etching process to reduce similar defects (if any) on all panels manufactured in the future. The intercommunication between system-level solutions will ultimately further enhance and improve AI's global-level decision-making capabilities.